Kleene - Introduction To Metamathematics (1971) PDF

INTRODUCTION TO METAMATHEMATICS
BIBLIOTHECA MATHEMATICA
A Series of Monographs on Pure and
Applied Mathematics
Volume I
Edited with the cooperation of

THE „MATHEMATISCH CENTRUM”
and
THE „WISKUNDIG GENOOTSCHAP”
at Amsterdam
Editors
N. G. de B ruijn
J. de Geoot
A. C. Zaanen
INTRODUCTION TO
METAMATHEMATICS
BY
STEPHEN COLE KLEENE

PROFESSOR OF MATHEMATICS AT THE UNIVERSITY
OF WISCONSIN (MADISON, WIS., U.S.A.)
WOLTERS-NOORDHOFF PUBLISHING - GRONINGEN

NORTH-HOLLAND PUBLISHING COMPANY - AMSTERDAM • OXFORD
AMERICAN ELSEVIER PUBLISHING COMPANY, INC. - NEW YORK
©WOLTERS-NOORDHOFF PUBLISHING
AND NORTH-HOLLAND PUBLISHING COMPANY, 1971
All rights reserved. No part o f this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photocopy
ing, recording or otherwise, without the prior permission o f the Copyright owner.
Library of Congress Catalog Card Number 70-97931
North-Holland ISBN 0 7204 2103 9

American Elsevier ISBN 0 444 10088 1
First published 1952

First reprint 1957
Second reprint 1959
Third reprint 1962
Fourth reprint 1964
Fifth reprint 1967
Sixth reprint 1971
Seventh reprint 1974
Publishers:
WOLTERS-NOORDHOFF PUBLISHING - GRONINGEN
NORTH-HOLLAND PUBLISHING COMPANY - AMSTERDAM • OXFORD
Sole distributors for the U.S.A. and Canada:

AMERICAN ELSEVIER PUBLISHING COMPANY, Inc.
52 Vanderbilt Avenue
New York, N.Y. 10017
Printed in The Netherlands

PREFACE
Two successive eras of investigations of the foundations of mathematics
in the nineteenth century, culminating in the theory of sets and the
arithmetization of analysis, led around 1900 to a new crisis, and a new
era dominated by the programs of Russell and Whitehead, Hilbert and of
Brouwer.
The appearance in 1931 of Godel’s two incompleteness theorems,
in 1933 of Tarski’s work on the concept of truth in formalized languages,
in 1934 of the Herbrand-Godel notion of 'general recursive function’,
and in 1936 of Church’s thesis concerning it, inaugurate a still newer
era in which mathematical tools are being applied both to evaluating the
earlier programs and in unforeseen directions.
The aim of this book is to provide a connected introduction to the
subjects df mathematical logic and recursive functions in particular, and
to the newer foundational investigations in general.
Some selection was necessary. The main choice has been to concentrate
after Part I on the metamathematical investigation of elementary number
theory with the requisite mathematical logic, leaving aside the higher
predicate calculi, analysis, type theory and set theory. This choice was
made because in number theory one finds the first and simplest exempli
fication of the newer methods and concepts, although the extension to
other branches of mathematics is well under way and promises to be
increasingly important in the immediate future.
The book is written to be usable as a text book by first year graduate
students in mathematics (and above) and others at that level of mathe
matical facility, irrespective of their knowledge of any particular mathe
matical subject matter.
In using the book as a text book, it is intended that Part I (Chapters
I — III), which provides the necessary background, should be covered
rapidly (in two or three weeks by a class meeting three times a week). The
intensive study should begin with Part II (Chapter IV), where it is es
sential that the student concentrate upon acquiring a firm grasp of
metamathematical method.
The starred sections can be omitted on a first reading or examined in
VI PREFACE
a cursory manner. Sometimes it will then be necessary later to go back and
study an earlier starred section (e.g. § 37 will have to be studied for § 72).
Godel’s two famous incompleteness theorems are reached in Chapter
VIII, leaving a lemma to be proved in Chapter X. The author has found
it feasible to complete these ten chapters (and sometimes a bit more)
in the semester course which he has given along these lines at the Uni
versity of Wisconsin.
The remaining five chapters can be used to extend such a course to a
year course, or as collateral reading to accompany a seminar.
A semester course on recursive functions for students having some prior
acquaintance with mathematical logic, or under an instructor with such
acquaintance, could start with Part III (Chapter IX). There are other
possibilities for selecting material; e.g. much of Part IV can follow directly
Part II or even Chapter VII for students primarily interested in mathe
matical logic.
The author is indebted to Saunders MacLane for encouraging him to
write this book and for valuable criticism of an early draft of several
chapters. John Addison read the entire first printer’s proof with great care,
independently of the author. Among many others who have been of as
sistance are Evert Beth, Robert Breusch, Arend Heyting, Nancy Kleene,
Leonard Linski, David Nelson, James Renno and Gene Rose. Scientific
indebtedness is acknowledged by references to the Bibliography; especially
extensive use has been made of Hilbert and Bemays’ "Grundlagen der
Mathematik” in two volumes 1934 and 1939.
July 1952 S. C. K leene
Note to the Sixth Reprint (1971). In successive reprints various errors have been
corrected, the principal corrections being those listed in Jour, symbolic logic vol. 19
(1954) p. 216 and vol. 33 (1968) pp. 290-291, and: on p. 505 bottom paragraph
sg((r)0) 'p{(r)-d + Wo '^(Wi) replaced by a function x{P> r) defined by Theorem
XX (c); on p. 506 allowance made in the middle paragraph for x possibly occurring
free in t, and line 5 from below " = ” changed to Moreover, in this sixth
reprint eleven bibliographical references have been updated (cf. end p. 517) and
two short notes have been added (on pp. 65 and 316).
TABLE OF CONTENTS
P art I. THE PROBLEM OF FOUNDATIONS
Chapter I. THE THEORY OF SETS........................................... 3
§ 1- Enumerable sets ................................................. 3
§ 2 . Cantor's diagonal method.............................................. 6
§ 3. Cardinal number.............................................................. 9
*§ 4. The equivalence theorem, finite and infinite sets . . . 11
*§ 5. Higher transfinite cardinals........................................... 14
Chapter II. SOME FUNDAMENTAL CONCEPTS................... 19
§ 6. The natural numbers...................................................... 19
§ 7. Mathematical induction.................................................. 21
§ 8. Systems of objects.......................................................... 24
*§ 9. Number theory vs. analysis........................................... 29
§ 10 . Functions................................................................. 32
Chapter III. A CRITIQUE OF MATHEMATICAL REASONING 36
§ 11. The paradoxes .................................................................. 36
§ 12 . First inferences from the paradoxes........................... 40
§ 13. Intuitionism...................................................................... 46
§ 14. Formalism......................................................................... 53
§ 15. Formalization of a theory.............................................. 59
P art II. MATHEMATICAL LOGIC
Chapter IV. A FORMAL SYSTEM.............................................. 69
§ 16. Formal sym bols.............................................................. 69
§ 17. Formation rules .............................................................. 72
§ 18- Free and bound variables.................................. ... 76
§ 19. Transformation ru les...................................................... 80
VIII TABLE OF CONTENTS
Chapter V. FORMAL D EDUCTION........................................... 86
§ 20 . Formal deduction........................... ... 86

§ 21. The deduction theorem ................................................... 90
§ 22. The deduction theorem (concluded)............................... 94
§23. Introduction and elimination of logical symbols . . . 98
*§24. Dependence and variation....................... ...................... 102
Chapter VI. THE PROPOSITIONAL CALCULUS.................... 108

§25. Proposition letter form ulas................... 108
§26. Equivalence, replacement............................................... 113
§27. Equivalences, duality ...................................................... 118
§28. Valuation, consistency ................................................... 124
§29. Completeness, normal form ........................................... 130
§30. Decision procedure, interpretation............................... 136
Chapter VII. THE PREDICATE CALCULUS........................... 142
§31. Predicate letter form ulas........................... . . . . . 142
§32. Derived rules, free variables. ............................... ... . 146
§33. Replacement..................................................................... 151
*§ 34. Substitution...................................................................... 155
§35. Equivalences, duality, prenex f o r m ........................... 162
§36. Valuation, consistency.................................................. 168
*§37. Set-theoretic predicate logic, ^-transforms . . . . . . 174
Chapter VIII. FORMAL NUMBER THEORY........................... 181
§38. Induction, equality, replacement.................................. 181
§39. Addition, multiplication, order. ............................... . 185
*§40. The further development oh number th eo ry ................ 189
§41. Formal calculation.......................................................... 194
§42. GodeFs theorem .............................................................. 204
P art III. RECURSIVE FUNCTIONS

Chapter IX. PRIMITIVE RECURSIVE FUNCTIONS . . . . 217
§43. Primitive recursive functions....................................... 217
§44. Explicit definition.......................................................... 220
§45. Predicates, prime factor representation....................... 223
TABLE OF CONTENTS IX
§46. Course-of-values recursion................... .......................... 231
*§47. Uniformity . . . . . . . . . . . . . . . . . . . 233
§48. Godel's p-function......................................................... 238
§49. Primitive recursive functions and the number-theoretic
formalism.......................................................................... 241
Chapter X. THE ARITHMETIZATION OF METAMATHE
MATICS .......................................................................... 246
§50. Metamathematics as a generalized arithmetic . . . . 246
§51. Recursive metamathematical definitions.................... 251
§52. Godel numbering.............................................................. 254
*§ 53. Inductive and recursive definitions............................... 258
Chapter XI. GENERAL RECURSIVE FUNCTIONS................ 262
§54. Formal calculation of primitive recursive functions 262
§55. General recursive functions........................................... 270
§56. Arithmetization of the formalism of recursive functions 276
§57. The (i-operator, enumeration, diagonal procedure . . 279
§58. Normal form, Post’s theorem ....................................... 288
*§ 59.General recursive functions and the number-theoretic
formalism.......................................................................... 295
§60. Church's theorem, the generalized Godel theorem. . . 298
§61. A symmetric form of Godel's theorem ....................... 308
Chapter XII. PARTIAL RECURSIVE FUNCTIONS................ 317

§62. Church's t h e s i s .............................................................. 317
§63. Partial recursive fu n ctio n s........................................... 323
§64. The 3-valued logic.......................................................... 332
§65. Godel numbers................................................................. 340
§ 66 . The recursion theorem .................................................. 348
Chapter XIII. COMPUTABLE FUNCTIONS............................... 356

§67. Turing machines.............................................................. 356
§ 68 . Computability of recursive functions........................... 363
§69. Recursiveness of computable functions....................... ... 373
§70. Turing’s thesis.................................................................. 376
*§71. The word problem for semi-groups................................... 382
X TABLE OF CONTENTS
P art IV. MATHEMATICAL LOGIC (ADDITIONAL TOPICS)
Chapter XIV. THE PREDICATE CALCULUS AND AXIOM
SYSTEMS......................................................................... 389
§72. Godel's completeness theorem ................................... ... 389
§73. The predicate calculus with equality. . ........................ 399
*§ 74 Eliminability of descriptive definitions....................... 405
§75. Axiom systems, Skolem’s paradox, the natural number
sequence......................................................................... ... 420
§76. The decision problem . . . . . . . . . . . . . . . 432
Chapter XV. CONSISTENCY, CLASSICAL AND INTUITION-
ISTIC SY STEM S.......................................................... 440
§77. Gentzen’s formal sy ste m .............................................. 440
§78. Gentzen's normal form theorem . ............................... 448
*§ 79. Consistency proofs.......................................................... 460
§80. Decision procedure, intuitionistic unprovability. . . . 479
§81. Reductions of classical to intuitionistic systems . . . 492
§82. Recursive realizability.............................................. . 501
BIBLIOGRAPHY.............................................................................. 517
SYMBOLS AND NOTATIONS....................................................... 538
IN D EX ................................................................................................. 539
PART I
THE PROBLEM OF FOUNDATIONS
Ch a pt e r I
THE THEORY OF SETS
§ 1. Enumerable sets. Before turning to our main subject, it will
be appropriate to notice briefly Cantor's theory of sets.
A flock of four sheep and a grove of four trees are related to each other
in a way in which neither is related to a pile of three stones or a grove of
seven trees. Although the words for numbers have been used to state this
truism on the printed page, the relationship to which we refer underlies
the concept of cardinal number. Without counting the sheep or the trees,
one can pair them with each other, for example by tethering the sheep
to the trees, so that each sheep and each tree belongs to exactly one of the
pairs. Such* a pairing between the members of two collections or 'sets'
of objects is called a one-to-one ( 1- 1) correspondence.
In 1638 Galileo remarked that the squares of the positive integers can
be placed in a 1-1 correspondence with the positive integers them
selves, thus
4, 9, 16, . .., n2, ...
3, 4, . .., n, ...
of
despite the ancient axiom that the whole is greater than any of its parts.
Cantor, between 1874 and 1897, first undertook systematically to compare
infinite sets in terms of the possibility of establishing 1-1 correspond
ences.
The two sets in Galileo's "paradox" and the set of the natural numbers
0, 1, 2, 3, ..., n 1,
are examples of infinite sets which are 'enumerable'. Choosing the last
named as the standard, we define an infinite set to be enumerable (or
denumerable or countable), if it can be placed in a 1-1 correspondence
with the natural numbers.
To show that an infinite set is enumerable, we need merely indicate
how its members can be given (without repetitions) in an 'infinite list';
3
4 THE THEORY OF SETS CH. I
then the first in the list corresponds to 0, the second to 1, and so on.
Although the list itself is infinite, each member occupies a finite position
in the list.
A particular infinite list (without repetitions) of the members of the
set, or 1-1 correspondence between the set and the natural numbers,
is called an enumeration of the set; the number corresponding to a given
member is the index of the member in the enumeration.
The members of a finite set can also be given in a list, i.e. a finite list.
Hence the term enumerable is sometimes applied to sets which are either
infinite and enumerable, i.e. enumerably infinite, or else finite.
The set of the integers can be enumerated, by listing them in the
following order,
0, 1, -1 , 2, - 2 , 3, —3 ,----
The set of the rational numbers is also enumerable, a fact which is
surprising if one first compares them with the integers in the usual
algebraic order. The points on the #-axis with integral abscissas are isolat
ed, while those with rational abscissas are 'everywhere dense', i.e. between
each two no matter how close there are others. The enumeration can be
accomplished by a device which we shall present for the positive rational
numbers, leaving the case of all the rationals to the reader. Let the
fractions of positive integers be arranged in an infinite matrix, thus,
Vi v.-> Vs 7 4 - •••
/ s
7x 7. 7s 7
✓ ^
4
3/l 7a 7s ••.
*
74
71 % 7s 74 ..•
Then let these fractions be enumerated by following the arrows. A rational

number is one which can be expressed as a fraction of integers. Go
through the' enumeration of fractions striking out each one which is
equal in value to one that has preceded it. This leaves the following
enumeration of the positive rational numbers,
1, 2, i/„ V* 3, 4, »/,. •/* 'U.........
§1 ENUMERABLE SETS 5
The device of the matrix constitutes a general one for enumerating the
ordered pairs of members of an enumerable setf e.g. the ordered pairs of
natural numbers, or the ordered pairs of integers. The rows of the matrix
are the enumerations of the pairs with the first member of the pair fixed.
The ordered triples of members of an enumerable set can then be enumer
ated by another application of the matrix, taking as the rows the
enumerations already won of the triples with the first member of the
triple fixed. Successively we can win enumerations of the ordered
n-tuples of members of an enumerable set for each fixed positive integer
n. All of these enumerations, including the enumeration of the original
set, can be taken as the rows of a new matrix to obtain an enum
eration of the ordered n-tuples for variable n, i.e. the finite sequences of
members of an enumerable set.
This result can be applied to obtain an enumeration of the algebraic
equations
a0xn + axxn~x + ... + an_xx + an = 0 (aQ ^ 0)
with integral coefficientsy since each equation can be described by giving
the sequence
(#Q, . . . , &n—1>^n)
of its coefficients. A '(real) algebraic number' is a real root of an equation
of this sort. Since a given equation has at most n different roots, the
algebraic numbers are enumerable.
Another device will illustrate the possibilities for enumerating sets.
In dealing with an enumerable set (finite or infinite), the numbers which
correspond to the members in some specified enumeration can be used
to designate or name the members individually. Now conversely, if a
name or explicit expression can be assigned to every one of the members
of a set individually, in a preassigned and unambiguous system of nota
tion, the set is enumerable (finite or infinite). We stipulate that a name or
expression shall be a finite sequence of symbols chosen from a given
finite alphabet of available symbols. For example, the algebraic equations
with integral coefficients can be written using decimal notation for the
coefficients and exponents. The raised exponents are an inessential feature
of the notation, which can be removed by a suitable convention. Indeed,
so long as we are dealing only with these equations, we may simply write
the exponents on the line. The symbols required are then precisely
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, x9 + ,
The first symbol in an equation is not a 0. Reinterpret these symbols as
the digits (!) in a quattuordecimal number system, i.e. a number system
based on 14 in the same way that the decimal system is based on 10.
Every equation becomes a natural number (distinct equations becoming
distinct numbers). Enumerate the equations in the order of magnitude of
these numbers.
§ 2. Cantor’s diagonal method. That there are infinite sets con
sidered in mathematics which cannot be enumerated was shown by
Cantor's famous 'diagonal method'. The set of the real numbers is
non-enumerable.
Let us first consider the real numbers x in the interval 0 < x <; 1.
Each real number in this interval is represented uniquely by a proper
non-terminating decimal fraction, i.e. a decimal fraction having its first
significant digit to the right of the decimal point, and having infinitely
many digits that are not 0. A number may have a terminating decimal
fraction, i.e. one with repeating 0's, but that fraction is replaceable by a
non-terminating fraction with repeating 9's. For example .483 or
.483000... can be replaced by .482999__ Conversely every proper non
terminating decimal fraction represents a unique real number in the
interval.
Now suppose that
x0, xv x2f x3f ...
is an infinite list or enumeration of some but not necessarily all of the
real numbers belonging to the interval. Write down one below another
their respective non-terminating decimal fractions,
*01 *02 *03
\
• *o o
*10 *11 *12 *13

V
*20 *21 *22 *23
\
•*30 *31 *32 *33
\
Select the diagonal fraction shown by the arrows. In this change

each of the successive digits xnn to a different digit xnni but avoid
producing a terminating fraction. Say, let xnn = 5 if xnn ^ 5, and
x*
*nn = 6 if xnn = 5 .
§2 c a n t o r ’s d ia g o n a l m e t h o d 7
The resulting fraction
/ r t r
.# 0 0 #11 #22 # 33 * * *
represents a real number %which belongs to the interval but not to the
enumeration. For the fraction differs from the first of the given fractions
in the tenths place, from the second in the hundredths place, from the
third in the thousandths place, and so on.
Hence the given enumeration is not an enumeration of all the real
numbers in the interval. An enumeration of all the real numbers in the
interval is non-existent.
To apply the diagonal method to the real numbers without restriction
to the interval 0 < x < 1, it is only necessary to represent the real numbers
in the characteristic-plus-mantissa form, e.g. 37.142... = 37 + .142,
—2,813... = —3 + .186..., and to apply the method to the man
tissas.
It is clear that an essential difference has been revealed between
the set of the rational numbers or the set of the algebraic numbers on the
one hand, and the set of the real numbers on the other.
It is interesting historically to note how Cantor's discoveries in 1874
(see the bibliography) illuminated an earlier discovery of Liouville in 1844.
Liouville had been able to construct by a special method certain tran
scendental (i.e. non-algebraic) real numbers. Cantor's diagonal method
makes the existence of transcendental numbers apparent from only the
very general considerations presented above. In fact, to any given
enumeration #0, xv #2, #3, ... of the algebraic numbers, particular
transcendentals can be obtained by the diagonal method.
The (real) transcendental numbers are not enumerable. For if they
were, like the algebraic numbers, enumerations of the two sets could
be combined to produce an enumeration of all the real numbers. Thus,
in a sense, most real numbers are transcendental.
Another example of a non-enumerable set is the set of the (single
valued) functions for which the independent and dependent variables
each range on an enumerable set. For definiteness, consider the set of the
functions of a natural number taking a natural number as value (or infinite
sequences of natural numbers). Suppose an enumeration is given of some
but not necessarily all of them,
/o(«). /l(«)> /*(»). /s(»)........
Write the sequences of the values of the successive functions one below
another, as the rows of an infinite matrix.
/o(0) /o(l) /o(2) /o(3) ...

\
/i(0) /x(l) A(2) A(3) ...
\
/ 2(0) /,(1) /,(2) /,(3) ...
\
/.(0) / 8(1) /,(2) /,(3) ...
\
Take the sequence of values given by the diagonal. Change every one of
these values to a different value, say by adding 1. The function /(«) with
the resulting sequence of values, which we may write
f(n) = /„(*) + 1,
cannot belong to the enumeration, since it differs from the first of the
enumerated functions in the value taken for 0, from the second in the
value taken for 1, and so on.
To phrase the argument differently, suppose that the function f(n)
were in the enumeration; i.e. suppose that for some natural number q,
m = /.(*)
for every natural number n. Substituting the number q for the variable
n in this and the preceding equation,
m = /.(?)= /.(?) + 1.
This is impossible, since the natural number f q(q) cannot equal itself in
creased by 1.
Still another example of a non-enumerable set is the set of the sets
of natural numbers, (But the set of the finite sets of natural numbers is
enumerable. Why?) We can represent a set of natural numbers by a
representing function, which takes the value 0 for a natural number
belonging to the set and the value 1 for a natural number not belonging
to the set. The sequence of the values of the representing function of a
set of natural numbers is an infinite sequence of 0's and 1's. For example,
the sequence for a set containing 0, 2 and 3 but not 1 and 4 starts out
0 10 0 1 __These sequences are taken as the rows of the infinite matrix.
The alteration performed on the diagonal is the interchange of 0's and 1's.
Can these, several non-enumerable sets be placed in 1-1 corres
pondence with one another, and are there still other types of infinite
sets? The reader may profit by attempting to answer these questions
himself (answers are given in § 5). We shall now look at Cantor's theory in
its general formulation.
§3 CARDINAL NUMBER 9
§ 3. Cardinal number. Cantor’s theory of ‘abstract sets’ deals with

sets in general. (He gave also a theory of 'point sets’.) Cantor describes
his terms set and element as follows. "By a ‘set’ we understand any col
lection M of definite well-distinguished objects m of our perception or our
thought (which are called the 'elements’ of M) into a whole.” (1895 p. 481.)
The sets include the empty or null or vacuous or void set which has no
elements, and the unit sets which have a single element each. We write
the empty set as O; the unit set with sole element a as [a] ; and the set
having a, b, c, ... as its elements as {a, b, c, . ..}.
A set may also be called an aggregate or collection or class or domain or
totality. That a is an element of M may also be expressed by saying that a
is a member of M or belongs to M or is in M f or in symbols, a 8 M. If a is
not an element of M , in symbols, a$M.
We understand two sets M and N to be the same (in symbols, M = N),
if they have the same elements; i.e. if for every object a, a 8 M if and only
if a& N.
Two sets M and N are said to be equivalent (in symbols, M ~ N),
if there exists a 1-1 correspondence (§ 1) between them. (Sometimes
we say "the correspondence M ~ jV” to refer to a particular 1-1
correspondence between M and N, which must exist if M ~ N.)
The relation M ~ N evidently possesses the 'reflexive’, 'symmetric’
and 'transitive’ properties, i.e. for any sets M, N and P : M ~ M. If
M ~ N , then N ~ M . If M ~ N and N ~ P, then M >— P.
The cardinal number of a set M is introduced as an object M which
is associated in common with all and only those sets (including M itself)
which are equivalent to M. By this definition: M = N, if and only if
M ~N.
What cardinal numbers are, further than this, is perhaps immaterial;
but we may notice several interpretations. Cantor describes them thus:
"The general concept which with the aid of our active intelligence results
from a set M, when we abstract from the nature of its various elements
and from the order of their being given, we call the 'power’ or 'cardinal
number’ of M.” This double abstraction suggests his notation "M” for
the cardinal of M. Frege 1884 and Russell 1902 identify the cardinal
number M with the set of the sets equivalent to M ; while von Neumann
1928 chooses from each of these sets of sets ('equivalence classes’) a par
ticular set to serve as the cardinal of any set in the class.
The notion of a 'part’ of a collection is introduced by the following
definition. A set M ± is a subset of a set M (in symbols, M x c M), if each
element of M x is an element of M.
E xample 1. The set {a, b, c) of three elements a, b, c has eight (= 23)
subsets: O, {a}, {b}f {c}t {a, b}f {a, c}, {b, c}, (a, b, c).
Note that the subsets of a set M include the vacuous set O, and the
set M itself. The latter is the improper subset, and the other subsets are
proper. Evidently, if M 2 C M x and M x C M (abbreviated M 2 C M x C M),
then M 2 C M.
The union or sum M + N of two sets M and N is the set of the objects
belonging to at least one of M and N (i.e. belonging to M or to N ); and
their intersection or product M • N is the set of the objects belonging to
both of M and N (i.e. belonging to M and to N). Similarly for more than
two sets. The difference M — N of M and N (when N C M y also called
the complement of N with respect to M) is the set of the objects belonging
to M but not to N.
E xample 2. {a, b, c} + {5, d} = {a, bt c, d}, {af b, c} • {b, d} = {6},
{a, bf c} — {bt d) = {a, b, cj — {bj = {a, c).
Evidently M — M 1 C M; and if Mx C M, then (and only then)
M x -)- (M — Mj) = M. Two sets M and N are disjoint, if they have no
common elements, i.e. if M • N = O. For example, M x and M — M x are
disjoint sets. If M and N are disjoint, either M ^ N or M = N = O.
We turn to the important question of comparing cardinal numbers.
Given two set9 M and N, it may or may not be possible to put M into
1-1 correspondence with some subset N x of N. Vice versa, there may
or may not exist a subset Mx of M which is equivalent to N. Combining
these twp pairs of alternatives gives four cases, exactly one of which must
apply to any given pair of sets M and N :
(la) For some N lt M ~ N 1 C N ; but for no M v N ~ M l C M.
(lb) For no Nlt M ~ N 1 C N ; but for some M v N ~ Mx C M.
(2) For some N v M ~ N 1 c N ; and for some Mi, N ~ M 1 c M .
(3) For no Nv M ~ N 1 c N ; and for no M v N ~ M, C M.
In Case (la), the cardinal of M is said to be less than the cardinal of N
(in symbols, M < N ). To justify considering < as a relation between the
cardinals M and N, and not merely one between the sets M and N t we
must observe that if M' ~ M and N' ~ N, then Case (la) applies to the
pair of sets M 't IV' if and only if it applies to the pair M , N.
The order relation for cardinals is transitive, i.e. for any three cardinals
M, N, P: If M < N a n d < _ P , then M < P. =
We define M to be > N, if N < M. Then M > N exactly in Case (lb).
§4 THE EQUIVALENCE THEOREM 11
The relationship M = ffi, i.e. M ~ N, evidently falls under Case (2) by

taking as N x and M t the improper subsets. Hence, for any two_cardinals
M and N, the three relationships M < N, M = N and M > N are ‘mu
tually exclusive', i.e. not more than one of them can hold.
It does not appear till an advanced stage of the theory (references in
§ 5) whether they are 'exhaustive', i.e. whether at least one of the three
must hold. The situation is partially clarified by the next theorem,
after which the question remains only whether Case (3) can arise.
*§ 4. The equivalence theorem , finite and infinite sets. T heo
rem A. If M r^N x C N and N ~ Mt c M, then M ~ N . In other words:
In Case (2) of § 3, M = F. (F. Bernstein 1898.)
P roof. By the hypotheses, we may suppose given a particular 1-1
correspondence M ~ N x between M and the subset N x of N ; and similarly
N ~ M v Our problem is to find a third 1-1 correspondence M rL N.
Let AQ= M — M t. In the given correspondence M rL N v the elements
of the subset A0oiM will correspond to elements forming a subset B1of N 1
(and hence of N), or in symbols A 0 rL Bv Then in the other given corres
pondence N ~ Mv the elements of the subset Bx of N will correspond to
elements forming a subset A x of M x (and hence of M), or in symbols
Bt rtj A x\ and so on. Thus
A 0 B± ~ A} rL B2ris A 2rL/ B% A%/ i ....
The situation may be grasped by picturing M and N as mirrors by which
the part A 0 of M outside M 1 is reflected back and forth to produce an
infinite succession of images A v A 2, A 3, ... in M and Bv B2, Bz, ... in
N, as shown in the figure. (The sets M , M x and N are represented by
the parts of the horizontal lines to the right of the labels “M”, “M i’
and “N “; the sets A0, Bv A v ... by the intercepted segments.)
Let A = A 0 + A x + A %+ A 3 + . . . ; i.e. A is the subset of M con
taining the elements which fall in A0 or in any of its images A v A 2,
A 3, ... in M. Also let B = Bt + B2 + Bz + ...; i.e. B is the subset
of N containing the elements which fall in any of the images Blt B2,
J53, ... of A0 in N.
To obtain the 1-1 correspondence M r tN , we state a rule which
determines to each element m of M a corresponding element n of N,
and prove that the resulting correspondence is 1-1 between M and N.
R ule . Consider any element m of M . Either m belongs to the subset
A , or m does not belong to A , i.e. m belongs to M —A . If m belongs to A ,
the corresponding element n o iN shall be that which corresponds to m in
the correspondence M rL Nv If m belongs to M —A (in which case m
belongs to Mx), the corresponding element n o iN shall be that to which m
corresponds in the correspondence N rL M v
The resulting correspondence is 1-1 between M and N, for:
(a) To different elements m of M, say m1 and m2) there correspond
different elements nt and n2 of N. This is clear when m1 and m2 both
belong to A or both to M — A. But it is also so when m1 £ A and
m2 £ M — A, since then nx £ B and n2 £ N — B.
(b) Each element of N corresponds to some element m of M. Namely,
the elements of B all correspond to elements in A, and the elements of
N — B all correspond to elements in M — A.
The method of bringing M and N into 1-1 correspondence may be
visualized as a shifting in the above picture of each of the parts A 0,
A v A 2, A3, ... of M one position to the right, so that A0 takes the place
of A v A x of A 2, A 2 of As, __ This changes N rL Mt into N rL M.
Corollary A. If M c N, then M <> N.
(M < N means: M < N or M = N.) For if Af C IV, then either Case
(la) or Case (2) applies with M as the N v
The cardinal number of the empty set O we call 0. (N ote : M' ~ Q
onlyjf M' = O.) The cardinal number of any set N + {a} where a tN we
call N + 1• (Note : For a given set N -f {a} with a t N , a set M' ~
if and only if M' = N f + {a'} where a' g N' and N' — N.)
Regarding the natural numbers 0, 1,2, ..., n, n + 1, ... as a sequence
of objects already known to us, the two definitions just stated correlate
to each natural number n a respective cardinal number which we also
write n. We call these cardinals finite cardinals, and sets which have these
cardinals finite sets. The following two propositions will be proved in
Example 1 § 7.
(1) For each natural number n, the finite cardinal n is the cardinal of the
set of the natural numbers which precede the natural number n in the usual
order of the natural numbers) or in symbols, n = {0, 1, 2, . .. , n — 1}.
§4 THE EQUIVALENCE THEOREM 13
(2) If M = n (for a natural number n) and M ~ c M, then M t = M.

Thus: A finite set is not equivalent to any proper subset of itself.
From these two propositions it is not hard to show that the equality
relation m = n and order relation m < n as determined for finite cardinals
by the definitions of § 3 agree with the familiar equality and order relation
for the natural numbers (in particular, we do have n < n + 1 for finite
cardinals). Thus no confusion will result from identifying the natural
numbers with the finite cardinals when we choose to do so.
A set which is not finite we call infinite, and its cardinal an infinite or
transfinite cardinal. The cardinal number of the set of the natural numbers,
and therefore of every enumerably infinite set (§ 1), we call K0 (read “alef
null”).
Corollary B. If n is a finite cardinal, n < x0.
P roof. Because n is the cardinal of the subset {0, 1,2, ..., n — 1}
of the natural numbers, by Corollary A, n < K0. Assume, contrary to
the corollary, that n = R0. But n + 1 is also a finite cardinal, so similarly
n + 1 < which with n = R0gives n -\~ \ < n, contradicting n < n + 1.
Hence the assumption n = R0is untenable, and the remaining alternative
n < R0 is established.
T heorem B. An infinite set M has an enumerably infinite subset.
P roof. > M is not empty, since otherwise it would have the finite
cardinal 0. Thus M has an element a0. Then M — {a0} is not empty, since
otherwise M would have the finite cardinal 1. Thus M has another element
av Continuing thus, we select distinct elements a0, av a2, a3, ... corres
ponding to the natural numbers 0, 1, 2, 3, ..., which proves the theorem.
If P is the set M — {aQ, av a2, a3, ...} of the elements of M not selected,
M = P + {a0, av a2, a3, ...}.
Corollary A. If M is an infinite cardinal, then K0 < M.
By the theorem with Corollary A Theorem A.
Corollary B. An infinite set M is equivalent to a proper subset of
itself.
For M (expressed as above) is equivalent to its proper subset
M — {a0} = P + {av a2, a3, a4, . .
This with (2) above was proposed by Dedekind 1888 as an alternative
definition of the distinction between finite and infinite sets. (We see that
the property observed in Galileo's “paradox" is characteristic of infinite
sets.)
Corollary C. The cardinal number of an infinite set M is unchanged by
the introduction of a finite or enumerably infinite set of elements.
For new elements bQf blt b2, bz, ... can be introduced thus,
M + {b0, blt b2,bz, ...} = P + {ao>b0, av blt ..,}.
Inversely, the corollary says that the removal of an enumerable set of
elements from a set does not change the cardinal, provided the resulting
set M is infinite. If the original set is non-enumerable, the resulting set
must be infinite, as otherwise there would be an obvious enumeration of
the original set. Thus:
Corollary D. The cardinal number of a non-enumerable set is un
changed by the removal of a finite or enumerably infinite set of elements.
*§ 5. Higher transfinite cardinals. The first of the theorems of
this section is a formulation in general terms of the situation which we
met in the last example of § 2. The reader may also find it instructive to
experiment with the theorem or its lemma for the case that M is a small
finite set. The second of the theorems is a generalization of the situation
encountered in Corollary B of Theorem A.
We take advantage of the equivalence theorem, via its Corollary A,
to simplify the presentation of the proofs. The theorems however can be
proved, with only slight modifications in the argument, without the use of
the equivalence theorem.
Lemma A. If S is a set of subsets of M , and M ~ S, then there is a
subset T of M which does not belong to S.
P roof, by Cantor's diagonal method. A subset of M is defined when it
is determined which of the elements of M belong to the subset. This can
be arranged by stating a general criterion, which, for any element m of M ,
determines whether that element belongs to the subset or does not belong
to the subset. We now give a criterion of this sort to define T.
Criterion . In the 1-1 correspondence given by the hypothesis
M ~ S, any element m of M corresponds to an element S of S. But S
is one of the subsets of M. Therefore either m belongs to S, or m does not
§5 HIGHER TRANSFINITE CARDINALS 15
belong to S. If m belongs to S, then m shall not belong to T. If m does not
belong to S, then m shall belong to T.
Now suppose, contrary to what is to be shown, that T belongs to S.
Select that element of M, call it mlf which corresponds to T in the 1-1
correspondence M ~ S.
Does m1 belong to T ? We apply the criterion, with tn1 as the m. Since
tn1 corresponds to T, the 5 of the criterion is now T. The criterion gives a
contradiction, either if m1 belongs to T, or if nix does not belong to T.
The supposition that T belongs to S thus leads to absurdity. Hence, by
the method of reductio ad absurdum (in which the negation of a proposition
is proved by deducing a contradiction from the proposition), we conclude
that T does not belong to S.
If M is a given set, then the set of the subsets of M, i.e. the set of which
the elements are (all) the subsets of M, is designated as UM (“U” from
the German “Untermenge”).
T heorem C. For any set M, M < UM. (Cantor's theorem.)
P roof. If N 1is the set of the unit subsets of M, then M ~ N ± C UM.
Hence by Corollary A Theorem A, M = Nx < UM. Suppose, contrary
to the theorem, that M ^ UM, i.e. M UM. Then UM would satisfy
the conditions for S in the lemma. By the lemma, there would be a subset
T of M which does not belong to UM. This is absurd, since UM is the set
of all the subsets of M. Therefore the remaining alternative M < UM
must hold.
If we take as the M of the theorem a set with the transfinite cardinal K0,
we discover sets UM, UUM, ... which have greater and greater transfinite
cardinals. These new cardinals are denoted by 22**0, __ (In fact, for
any set M, the cardinal of UM is denoted by UM. Note that this accords
with the usual arithmetic when M is finite.)
L emma B. If Sis a set, and Mis a set of subsets ofS, and to each member
M of M there is another member M f of M such that M < Mf, then M < 5
for every member M of M.
P roof. Since M C S, by Corollary A Theorem A, M ^ 5. Assume,
contrary to the lemma, that M = S. But similarly, M' < 5, which with
M = S gives Mf < M, contradicting M < M'. Hence the assumption
M = 5 is false, and the alternative M < 5 holds.
If M is a set of which the members are sets, then the set of (all) the
objects each of which belongs to some member M of M is called the sum
of the sets belonging to Mand is designated by ©M. The set of the objects
each of which belongs to every member M of Mis called the intersection or
product of the sets belonging to M and is designated by 2)M (“®” from the
German “Durchschnitt”). These notions are the same as were introduced
in § 3, except that now they are expressed as operations on the set M of
the sets M which are added or multiplied. For example, M + N =
©{M, N), M • N = ${M, N}.
T heorem D. If M is a set of sets, and if to each member M of M there
ts another member M' of M such that M < Mf, then M < @M for every
member M of M.
P roof. From the definition of ©M, every element M of M is a subset
of ©M. The theorem now follows from the lemma with ©M as the S of the
lemma.
By this theorem, the sum of the sets M, UM, VLUM, ... which have the
increasing transfinite cardinals K0, 2^°, 22**0, ... is a set having a trans-
finite cardinal greater still than any of those cardinals. This set can be
used by Theorem C to start a new ascending series. This hierarchy extends
indefinitely.
More will be found on Cantor’s theory of abstract sets in Cantor 1895-7 ,
Hausdorff 1914 or 1927, or Fraenkel 1928 or 1953, for example. There is
a cognate branch of the theory dealing with “ordinal numbers”. The
“comparability theorem for cardinal numbers”, which asserts that
M < N, M = N and M > N are exhaustive (end § 3), appears as a cor
ollary of the “well-ordering theorem” of Zermelo 1904 (cf. e.g. Hausdorff
1914 or 1927 p. 61, or Fraenkel 1928 p. 205). For a brief account of the
celebrated “continuum problem”, which deals with the question whether
any cardinal lies between K0 and 2^», see Godel 1947.
We have begun with Cantor’s theory for two quite opposite reasons.
First, some of the ideas and methods which will prove basic later appear
in it in their original and simplest form. Second, the theory, pursued too
far, reveals logical difficulties, which are a point of departure for our
main investigation. This will appear in Chapter III.
E xamples. Sets of cardinal 2*k This is the cardinal assigned to the
set of the subsets of the set of the natural numbers, which we described
in § 2 as the set of the sets of natural numbers. There we represented the
elements of the set by the infinite sequences of 0’s and 1’s. The 0’s and
Ps can be interpreted as the digits in a dual (or dyadic) number system,
§5 HIGHER TRANSFINITE CARDINALS 17
i.e. a number system based on 2 as the decimal system is based on 10, so
that we have the set of the proper dual fractions. Using Theorem B
Corollary D to withdraw the terminating fractions, which are enumerable,
we obtain the proper non-terminating dual fractions. These represent
1-1 the real numbers x in the interval 0 < x < 1. From the proper
non-terminating dual fractions we also obtain 1-1 the infinite se
quences of natural numbers or functions of a natural number taking
a natural number as value, by coordinating to a fraction that func
tion f(n) for which /(0) = the number of 0’s before the first 1 in the
fraction, /( 1) = the number of 0’s between the first 1 and the second 1,
and so on (for example, the function n2 corresponds to the fraction
. 101000010000000001. . . ) .
Now omit from the interval 0 < a < 1 the number x = 1, leaving the
real numbers x in the interval 0 < a < 1. A function y = f(x) is easily
found which, as a ranges over this interval, takes as value y exactly once
each of the real numbers, e.g. the function y = cot izx. Removing the
rational numbers, the (real) irrational numbers are left; or removing
the algebraic numbers, the transcendental numbers. In Cartesian
coordinate geometry, the real numbers are coordinated to the points
of the real Euclidean line. This set is the linear continuum’, and ac
cordingly the cardinal 2^° is the 'power of the continuum’.
Next we can proceed as follows to obtain the set of the ordered
pairs of real numbers, or regarding a pair (x} y) as Cartesian coor
dinates in the plane, the points of the real Euclidean plane. Under the
equivalence already obtained between the real numbers and the infinite
sequences of 0’s and 1’s, any two real numbers x, y correspond respectively
to sequences of 0’s and 1’s
Xq Xi a2 x3 ...,
yo yi y 2
which can be combined into a single sequence
xo ^ xi Vi x2 y% x3 ^3 • • • >
corresponding to a single real number. Conversely, any single sequence

breaks up into a determinate pair of sequences under this method of
combination. Similar procedure gives the n-tuples of real numbers or the
points of real Euclidean n-dimensional space for any fixed positive integer
n, and even the infinite sequences of real numbers or the points of real
Etxclfdean X0-dimensional space. This last example is treated by using the
method of § 1 to combine x0sequences of 0’s and 1’s
*00 X 01 ~^ X q 2 X 03 ~~^
/ * S'
*10 X 11 X 12 X 13 ...
/ /
*20 X 21 X 22 X 23 ...
1 /
*30 X 31 X 32 X 33 ...
.,,
into a single sequence
X 00 X 10 X 01 X 02 X 11 X 20 X 33 X 21 X 12 X 03 * • •
in which each member ot each of the given sequences has a determinate

position.
For any one of the real continuous functions of a real variable,
all the values of the function are determined by the continuity property
as soon as the values are given for the rational values of the independent
variable. These values can be given as an infinite sequence of real numbers,
by following the order of the rational numbers in some fixed enumeration
of the latter. Therefore by Theorem A Corollary A, the set of these func
tions has at most the cardinal But also it must have at least this
cardinal, and hence exactly this cardinal, since the constant functions
constitute a subset with the cardinal.
SETS OF cardinal 22**6. This is the cardinal of the sets of sets of
natural numbers. From the equivalence between the sets of natural
numbers and the real numbers or the points in ^-dimensional or No-
dimensional space, it follows that the sets of real numbers and the point
sets in real Euclidean fi-dimensional or N0-dimensional space have this
cardinal. The real functions of a real variable can be represented by their
graphs, which are point sets in the plane, and hence the set of them has
at most the cardinal 22**0. It has exactly this cardinal, since those of the
functions which take only 0 and 1 as values are the representing func
tions of the sets of real numbers, and so constitute a subset with the
cardinal. If we extend geometric terminology to this example, we have
the set of the points of real Euclidean 2^-dimensional space.
C h a p t e r II
SOME FUNDAMENTAL CONCEPTS
§ 6. The natural numbers. The purpose of this chapter is to bring
together, partly for reference and partly for closer inspection, some of the
ideas and methods of mathematics.
When we write the natural number sequence
0, 1, 2, 3,
we rely on the dots “ . . to suggest the continuation of the sequence

beyond the several members shown.
Kronecker remarked (1886), “God made the integers, all the rest is the
work of man.” We cannot expect that the cognizance of the natural
number sequence can be reduced to that of anything essentially more
primitive than itself.
But by elaborating upon what our conception of it comprises, we may
succeed in making clearer the bases of our reasoning with the natural
numbers.
We begin by describing the natural numbers as the objects which can
be generated by starting with an initial object 0 (zero) and successively
passing from an object n already generated to another object n -f 1 or n'
(the successor of n).
Here we conceive of it as possible, no matter how far we have already
gone to reach n, to go the one step further to reach n'. The use of the accent
notation “nf” instead of the more familiar ftn + 1” emphasizes that ' is
a primitive unary operation or function used in generating the natural
numbers, while + can be defined at a later stage as a binary operation or
function of two natural numbers.
To obtain the natural numbers with the usual notations, it remains
only to explain 0, 1, 2, 3, ... as standing for
0, O', 0", 0' ", . . . ,
respectively. This is a matter of detail concerning decimal notation.

In the foregoing description, we have evoked the conception of a
19
20 SOME FUNDAMENTAL CONCEPTS CH. II
succession of discrete steps. These consist in starting with 0, and proceed
ing repeatedly from a number n to the next nf. The description can be
broken into several clauses, as follows.
1. 0 is a natural number. 2. If n is a natural number, then nf is a natural
number. 3. The only natural numbers are those given by 1 and 2.
In this format, the succession of discrete steps becomes an application
of Clause 1 and a succession of applications of Clause 2. The three clauses
together constitute an example of what we call an inductive definition.
The term (‘natural number’) which is being defined is italicized. The
clauses except the last, which provide instances of the term being defined,
are called direct clauses; the last clause, which says that the only in
stances are those provided by the preceding clauses, is the extremal clause.
Not stated in this inductive definition is the condition for distinctness,
namely that numbers generated by applications of Clauses 1 and 2 in
distinct ways should be distinct objects. This can be separated into two
further propositions.
4. For any natural numbers m and n, m' = «' only if m = n. 5. For any
natural number n, n' ^ 0.
Also it is understood that ' is a univalent operator or single-valued
function, so that conversely to 4: For any natural numbers m and n,
m' == n' if m = n.
To see that Propositions 4 and 5 do require the distinctness of every
two differently generated numbers, we can reason as follows. Suppose
that at a given stage in the generation of the numbers, all the numbers
0, 1, . . n so far generated are distinct. Then the next one generated n'
must be distinct from the successors 1, ..., n among those previously
generated (by 4) and from 0 (by 5). So each successive step in the gener
ation produces a new number.
For example, 0"" 7^ 0", as may also be seen thus. By 4 applied with
O'" as the m and 0' as the n, 0"" = 0" only if O'" = O'. By 4 again,
O'" = 0' only if 0" = 0. But by 5 with 0' as the n, 0" ^ 0.
These five propositions 1—5, with one difference, were taken by
Peano (1889, 1891*) as axioms characterizing the natural number sequence.
Peano stated Proposition 3 instead as the principle of mathematical
induction (§ 7), and placed it fifth on his list, 4 and 5 being moved up to
third and fourth, respectively.
Here we are not considering what the natural numbers are intrinsically,
but only how they form the natural number sequence. A particular nat
ural number is to be recognized as the object occupying a particular
place in the sequence. In other words, a particular number is given when
§7 MATHEMATICAL INDUCTION 21
its generation under the inductive definition is given. For example, the
natural number 4 is given as that object which is obtained by starting
with the initial object 0 and applying the successor operation ' once,
again, again and again; or briefly, 4 is given as 0"". A number such as
872656 in decimal notation could in principle be exhibited by applications
o f' to 0, though in practice we do not do so.
Of course when we deal with propositions such as that a certain equation
has two roots, we further employ the assignment of the natural numbers
as cardinal numbers of finite sets (§ 4).
Order . Under the inductive definition of the natural numbers, they
are generated in a certain order (the familiar one). Thus we define m
to be < n, if m is generated before n in the course of generating n. Dis
secting this, we have the following inductive definition of the relation
m e n (where m, n range over the natural numbers).
01. m < m'. 02. If m < n, then m < n'. 03. m e n only as required
by 01 and 02.
When this definition is read, for a fixed m, as an inductive definition
of the class of the numbers n greater than m, it has the form of the original
inductive definition of the natural numbers, with m' replacing 0.
§ 7. Mathematical induction. Let P be a property of natural
numbers. Suppose that :
(1) 0 has the property P.
(2) If any natural number n has the property P, then its successor
n' has the property P.
Then: Every natural number n has the property P.
This is the principle of mathematical induction. We can state it a little
more briefly using ‘V ’ as a natural number variable and “P(n)" as a
notation for the proposition that n has the property P: If (1) P( 0), and
(2) for all n, if P(n) then P(n')t then, for all n, P(n).
The justification of the induction principle is almost immediate, when
the natural numbers are conceived as the objects generated under the
inductive definition 1—3 of § 6. Suppose we have a property P for which
(1) and (2) hold. Must then every natural number n have the property P?
We interpret an affirmative answer to mean simply that, if any natural
number n were given to us, we could be sure that that n has the property
P. But a natural number n is given precisely when (actually or in principle)
we are given its generation under the inductive definition, by starting
with 0 and applying an exhibited number of times the successor operation '.
Under these circumstances, we can use ( 1) and (2) to conclude that n,
has the property P. For example, P(4) holds because 4 is given as 0"";
by (1), P(0); thence by (2), P{O'); by (2) again, P(0"); by (2) again,
P( 0'"); and by (2) again, P( 0""),
Otherwise expressed, (1) and (2) are tools which enable us, while we
are generating the natural numbers by Clauses 1 and 2 of the inductive
definition, at the same time to verify for each number as we generate it
that it has the property P.
This reasoning of course depends on the extremal clause 3 of the in
ductive definition. Conversely, the principle of induction can be used to
prove Clause 3, by applying it with the following proposition as the P(n):
n is given as a natural number by Clauses 1 and 2, i.e. can be generated by
starting with 0 and applying the successor operation
In connection with a proof by mathematical induction, we use the
following terminology. The proposition P(n) depending on a variable
natural number n we call the induction proposition; and the variable n
the induction variable or induction number or the variable on which the
induction takes place. The part of the proof which consists in establishing
( 1), i.e. the proof that P(0), we call the basis of the induction. The part
which consists in establishing (2), i.e. the proof that if P(n) then P(nf),
we call the induction step. Within the induction step, the assumption
P(n), from which we deduce P(n'), we call the hypothesis of the induction.
Sometimes, in order to carry through the induction step, it is necessary
to assume as hypothesis of the induction, not simply P(n) but that P(m)
for all m <; n. The reader may satisfy himself that the induction principle
is valid in this modification, called a conrse-of-values induction. Induction
may be applied to the proof of a proposition depending on a positive
integer instead of a natural number, in which case the basis consists in
proving P(l).
The student encounters mathematical induction in elementary algebra
courses. Formulas for summing progressions are often given as examples
of propositions to be proved by induction which are not obvious before
the proofs have been given. Many propositions which we commonly take
for granted depend on induction when explicitly proved; and in other
cases an induction step is so simple that it is passed off with the phrase
"and so on” or the like (e.g. Theorems A and B § 4).
E xample 1. Prove the propositions ( 1) and (2) of § 4 by induction
on n. We do so for (2), leaving ( 1) to the reader. The induction proposition
is: For any sets M and Mv if M = n and M ~ Mx c M, then Mx = M.
§7 MATHEMATICAL INDUCTION 23
B a s is : n
== 0. Let M and M 1 be sets such that M — 0, i.e. M = O, and
O ^ M i C O . Then M 1= O. I nduction step . Assume the induction
proposition as stated (as hypothesis of the induction). Now let M and M x
be sets such that M = n + 1, i.e. M ==N + {a} where N = n and a iN ,
and N + {%} ~ M 1 C N + {«}. We must prove that then M x = N + {a}.
In the given 1-1 correspondence iV + W ~ Afi, the element a of
N -j- {a} corresponds to some element b of M v Then N ~ M x — {b} C
(N + {a}) —{6}. Moreover (N + {a}) — {b} ~ N . Hence (.N + {a}) — {b}
= n and (N + {^}) — {b} ~ M x — {6} C (N + {a}) — {6}. By the
hypothesis of the induction, applied using (N + {a}) — {b} as the M and
Mi — {b} as the M v M x — {b} = (N + {a}) — {6}. Hence (since b £ M x
and b 6 IV + {^})» = A7 + {a}.
E x a m p l e 2. In mathematical formulas parentheses are introduced in
pairs to show which way the parts of the formula should be associated.
In complicated cases different species of parentheses such as ( ), { }, [ ]
may be employed; and very complicated cases may be avoided by various
abbreviations. However, the question exists in principle whether, using
one species of parentheses only, the association of a formula is unam
biguously fixed by its parentheses. (The question is equivalent to a ge
ometrical one concerning nesting of intervals.)
To make the question precise, suppose we have 2n parentheses, n of
them being left parentheses (, and n of them right parentheses ), and that
they occur in linear order from left to right. This is the way they would
occur in a mathematical formula, the other symbols of the formula which
we need not notice now being interspersed among them.
We say that two pairs of parentheses separate each other, if they
occur in the order (t- Q ){ )h where the i’s identify either pair and the j’s
the other, and other parentheses may occur interspersed among the four
shown.
We define a 1-1 pairing of the n left parentheses with the n right
parentheses (briefly, a pairing of the 2n parentheses) to be proper, if a left
parenthesis is always paired with a right parenthesis to the right of it,
and if no two of the pairs separate each other.
It is almost immediate that if 2n parentheses are properly paired, on
removing any of the pairs, the remaining parentheses are properly paired.
Also the parentheses included between a given pair in the proper pairing
of the 2n parentheses are properly paired.
The following three lemmas contain the answer to the proposed question
and some related information.
L emma 1. A proper pairing of 2n parentheses (n > 0) contains an
innermost pair , i.e. a pair which includes no other of the parentheses between
them.
Prove by a course-of-values induction on n. One may choose to consider
n either a positive integer or a natural number. If the latter, the basis is
vacuously true, i.e. true because its hypothesis is not satisfied. (H in t :
Under the induction step, the leftmost parenthesis will be a left paren
thesis (,, and this with its mate )t- either themselves constitute an innermost
pair, or else include a set of parentheses lying between them to which
the hypothesis of the induction can be applied.)
L emma 2 . A set of 2 n parentheses admits at most one proper pairing.
Prove by a (simple) induction on n. (H int : Under the induction step,
by Lemma 1 the given parentheses contain an innermost pair. With
drawing this, the hypothesis of the induction applies to the set of the pa
rentheses remaining.)
L emma 3 . If 2n parentheses and a consecutive subset of 2m of them both
admit proper pairings, then the proper pairing in the subset forms a part of
the proper pairing in the whole set, i.e. each parenthesis of the subset has the
same mate in both pairings.
Prove by induction on m.
For illustration, consider the 22 parentheses
/ I /2 /3 /4 /5 \ 6 \ 7 /8 \ 9 \ 1 0 / I I \ 1 2 \ 1 3 \ 1 4 / 1 5 / 1 6 / 1 7 \ 1 8 / 1 9 \ 2 0 \ 2 1 \ 2 2
\ 7 \6 V4 \2 V l / I / 2 V3 / 3 / 4 \ 5 / 5 / 6 / 7 V l l U 0 V8 / 8 \9 / 9 ' 1 0 / I I *
A proper pairing, indicated by the subscripts, is discovered by the fol

lowing ‘algorithm' (suggested by the proof of Lemma 2): at each stage,
proceeding from the left, search out the first innermost pair among those
not already used, and let this pair belong to the pairing. By Lemma 2, it
would be futile to search for any other proper pairing than this. The third
to the twelfth parentheses are a consecutive subset, in which a proper
pairing has already been established in the process of pairing the whole
set. By Lemma 3, it would be futile to search for any consecutive subset
admitting a proper pairing other than one of those already properly paired
in pairing the whole.
§ 8. Systems of objects. By a system S of objects we mean a (non
empty) set or class or domain D (or possibly several such sets) of objects
among which are established certain relationships.
For example, the natural number sequence (§ 6) constitutes a system of
§8 SYSTEMS OF OBJECTS 25
the type (Z), 0,') where Z) is a set, 0 is a member of the set Df and ' a unary
operation on a member of the set D. Another simple type of system is
(D» <) where D is a set and < is a binary relation between members of
the set.
When the objects of the system are known only through the relation
ships of the system, the system is abstract. What is established in this
case is the structure of the system, and what the objects are, in any
respects other than how they fit into the structure, is left unspecified.
Then any further specification of what the objects are gives a rep
resentation (or model) of the abstract system, i.e. a system of objects
which satisfy the relationships of the abstract system and have some
further status as well. These objects are not necessarily more con
crete, as they may be chosen from some other abstract system (or even
from the same one under a reinterpretation of the relationships).
Several representations of the abstract natural number sequence are
(a) the natural numbers as cardinals of finite sets, (b) the positive integers
(1 representing the abstract 0), (c) the even natural numbers (+2 re
presenting the abstract '). (d) Commercial products are sometimes pack
aged in containers which carry advertizing matter including a picture of
the container itself. Physically, the picture must be limited in accuracy.
But if we suppose perfect accuracy, we can represent 0 by the container,
1 by the picture of the container on the container, 2 by the picture of
the container in the picture of the container on the container, and so on.
Two representations of the same abstract system are {simply) isomorphic,
i.e. can be put into a 1-1 correspondence preserving the relationships.
More precisely, two systems {DX) 01# and (D2, 02, '2) of the type (Z), 0,')
are simply isomorphic, if there exists a 1-1 correspondence between
Dx and D2 such that 0Xcorresponds to 02 (in symbols, 01 <—>- 02), and
whenever mx <—► m2 then mx \ <—> m2 '2. Two systems (Dlf <*) and
(D2, < 2) of the type (Z), <) are simply isomorphic, if there exists a 1-1
correspondence between Dx and D2 such that, if m1 <—>■ m2and n± > n2
then: mx < x % if and only if m2 < 2n2.
Conversely, any two simply isomorphic systems constitute represen
tations of the same abstract system, which is obtained by abstracting
from either of them, i.e. by leaving out of account all relationships and
properties except the ones to be considered for the abstract system.
As a second example of an abstract system of the type (Z), 0,'), let
D have just two (distinct) objects 0 and 1, and let O' = 1 and 1' = 0.
We call this system the residues modulo 2. The natural number sequence
becomes this when each number is replaced by its remainder after division
by 2 (i.e. is reduced mod 2), thus,
0, 1, 0, 1, 0 , 1, ....
(Systems of residues were first considered by Gauss, 1801.)
As a third example, let S consist of two sequences
0, 1, 2, 3, ...; to, to -f- 1, to + 2, co -j- 3 , ...,
each by itself of the same structure as the natural numbers, and with
no member of either sequence in the relation of successor to a member
of the other.
We can modify each of these three examples obviously to consider it
as a system of the type (D, <). In the third example, we then take the
elements in the order shown; and we call them the ordinals < 2<o (from
Cantor’s theory of ordinal numbers).
The residues mod 2 (or a representation of them) are not isomorphic
with the natural numbers (or a representation of them), since it is im
possible to establish a 1-1 correspondence. The ordinals < 2co are not
isomorphic with the natural numbers, for it is not possible in establishing
a 1-1 correspondence to preserve the successor operation ' (or the order
relation <).
In this section, we are writing “S” for a system and “D” for its set of
objects, in the case of systems having one set. The notation can often be
simplified without confusion to use one letter for both. It can e.g. in
speaking of the natural numbers N as above. It cannot e.g. in speaking of
the system (N, <) consisting of the natural numbers, with the even
numbers (odd numbers) ordered among themselves as usual, and all the
even numbers preceding all the odd numbers. (This system is a repre
sentation of the ordinals < 2o>.)
Systems of objects are introduced in mathematics under two con
trasting methods or points of view (cf. Hilbert 1900).
The genetic or constructive method is illustrated by the inductive defi
nition of the natural numbers (§ 6). There we conceived of the natural
numbers as being generated or constructed in a certain orderly manner.
(This did not prevent our treating them abstractly.)
In the axiomatic or postulational method, on the other hand, some
propositions, called axioms or postulates, are put down at the outset as
assumptions or conditions on a system 5 of objects. The consequences of
the axioms are then developed as a theory about any existing system S
of objects which satisfies the axioms.
To illustrate, we can take as the axioms Peano's five axioms. To
§8 SYSTEMS OF OBJECTS 27
make the point clear, let us rewrite the Peano axioms substituting
“member of D” for “natural number”.
PI. 0 ED. P2. If nZD , then n' £ D. P3. If m ZD and n £ D, then
m' = n' only if m = n. P4. If n£ D, then n' ^ 0. P5. If P c D and
( 1) OS P and (2) whenever m £ P then n' £ P, then P = D.
We already know that exactly one abstract system S satisfies these
five axioms, namely the natural numbers which we previously introduced
from the genetic standpoint.
But from the axiomatic standpoint, we can equally well consider other
lists of axioms, for example PI—P4. Then S can be the natural numbers,
or the ordinals < 2co, or any one of many other abstractly differing, i.e.
non-isomorphic, systems.
If instead the axioms are PI — P3, P5, then the different abstract
systems which satisfy are precisely the natural numbers and the systems
of residues mod m for each positive integer m.
Next suppose we not merely remove P4, but substitute for it:
P 6. If n £ D, then n' ^ n but n" = n.
Now again exactly one abstract system satisfies, the residues mod 2.
The six axioms PI—P 6 together are satisfied by no system S at all,
since only the natural numbers satisfy PI—P5 and only the residues mod
2 satisfy PI—P3, P5, P 6.
The axioms of an axiomatic theory are sometimes said to constitute
an implicit definition of the system of the objects of the theory; but this
can only mean that the axioms determine to which systems, defined from
outside the theory, the theory applies. Then three cases arise. The axioms
may be satisfied by no system of objects (e.g. PI—P 6); or by exactly
one abstract system, any two systems which satisfy being isomorphic
(e.g. PI—P5, or PI—P3, P5, P 6); or by more than one abstract system,
i.e. non-isomorphic systems exist which satisfy (e.g. PI—P4, or PI—P3,
P5). In the first case we may call the set of axioms vacuous; in the other
two non-vacuous, and furthermore in the second categorical (Veblen 1904)
and in the third ambiguous. (In the genetic method, on the other hand, the
generation process is ordinarily intended to determine the abstract
structure of the system completely, i.e. to constitute a categorical def
inition of the system.)
It may be by no means evident for a given axiomatic theory which
of the three possibilities is the case. This is illustrated historically by the
example of Euclidean geometry without Euclid’s parallel postulate, on
which depends the theorem that through a given point not on a given line
there passes exactly one line parallel to the given line. From Euclid’s
"Elements” (c. 330—320B.C.) until the discovery of a non-Euclidean geome
try by Lobatchevsky (1829) and Bolyai (1833), it was generally supposed
that the axioms are categorical; or at least if the question had been asked
in these terms, it would probably have been so answered.
The Greek’s belief that they were dealing with a unique structure of
space was not formulated in the present terminology. Euclid thought of
his axioms as expressing certain fundamental properties of real space.
The axiomatic method in this older sense, wherein the objects of the system
S are supposed to be known prior to the axioms, may be distinguished as
informal or material axiomatics. In this, the axioms merely express those
properties of the objects which are being taken initially as evident from
their construction, or in the case of theories applying to the empirical
world as abstracted directly from experience or as postulated about
that world.
The axiomatic method as described above, wherein the axioms are
prior to any specification of the system 5 of objects which the axioms
are about (and serve to introduce or "define implicitly” the S), was
first developed systematically in Hilbert’s "Grundlagen der Geometrie
(Foundations of Geometry)” (1899), and may be distinguished as formal
or existential axiomatics. We note that it is only from outside a formal
axiomatic theory (i.e. in some other theory) that one can investigate
whether one, or more than one, or no abstract system S satisfies the
axioms. Within the formal axiomatic theory, the domain D for S plays
the role of a fixed and completed set of objects, assumed as existing all
at once apart from any order of generation, to which the operations,
relations, etc. of S apply.
For a system S of the type (D, 0,'), 0 and ', or D, 0 and ', are then
called the primitive or technical or undefined notions, i.e. they are un
defined prior to the introduction of the axioms. The other terms in the
axioms are ordinary or logical or defined, i.e. their meanings must be pre
viously understood. About D, 0 and ', it has only to be understood in
advance that D is a set, 0 an object belonging to D, and ' an operation on
a member of D; i.e. only the grammatical categories to which " 0”
and " ' ” belong are defined in advance. Similarly for a system of the form
(D, <), the undefined notions are <, or D and <.
In mathematical practice there is often an interplay between the genetic
and axiomatic methods of introducing systems of objects, as when an
example of a system of objects satisfying the axioms is provided genet
ically. At other times an example may be drawn from another formal
axiomatic theory. (In either case, as soon as the S for a given formal
§9 NUMBER THEORY VS. ANALYSIS 29
axiomatic theory is identified with a system of objects provided from
outside the theory, we have an application of the formal axiomatic theory,
in which application it becomes a material axiomatic theory.)
The formal axiomatic method is often used to advantage with am
biguous axiom systems, so as to develop simultaneously a common
portion of theory for many different systems. The example of 'groups'
in algebra is celebrated.
As another example, consider the following axioms for linear order,
which apply to systems of the type (D, <).
LI. If m < n and n < p, then m < p. L2. At most one of m < n,
m—n and m > n holds. L3. At least one oim < n, m=n andm > n holds.
Here m > n means n < m. The variables m, n, p refer to any elements
of D. These axioms are satisfied taking as D the natural numbers, the
ordinals < 2co, the integers, the rational numbers, or the real numbers,
and as < the usual order relation for the same; and by many other
systems. Omitting L3, we have a set of axioms for partial order.
*§ 9. Number theory vs. analysis. Arithmetic or number theory
may be described as the branch of mathematics which deals with the
natural numbers and other (categorically defined) enumerable systems
of objects, such as the integers or the rational numbers. A particular such
system (or the theory of it) may be called an arithmetic. The treatment
is usually abstract (§ 8). The objects are usually treated as individuals
(i.e. they are not analyzed as composed out of other objects), except e.g.
when the fundamental properties of non-negative rational numbers are
being developed by representing them as ordered pairs of natural numbers.
In arithmetic in the narrower sense one is mainly concerned with par
ticular operations called + (addition) and (multiplication), or also a few
other related operations. In arithmetic in the wider sense or number theory
a wider fund of concepts is employed.
These definitions are given to clarify our terminology. Sometimes
"arithmetic" is encountered referring to the theory of + and • for systems
of numbers that are not enumerable (e.g. the 'arithmetic of transfinite
cardinals’).
While the cardinal numbers of the systems studied in arithmetic or
number theory are 8o (or sometimes finite), analysis on the other hand
deals with the real numbers and other systems of objects having the car
dinal 2^0 (or sometimes a higher cardinal). As with number theory, the
systems of objects employed in analysis are usually taken to be cate
gorically determined.
Results of analysis are sometimes applied in number-theoretic in
vestigations, which then constitute analytic number theory. Number
theory without help from analysis is pure or elementary number theory.
We now examine briefly the fundamental system of objects for analysis,
namely the continuum of the real numbers.
The theory of real numbers which is currently used as the basis for
analysis (except by critics of its foundations) is the product of an earlier
critical movement initiated by Gauss (1777—1855), Cauchy (1789—
1857) and Abel (1802—1829).
This led late in the nineteenth century to the arithmetization of analysis,
so called, by Weierstrass (1815—1897), Dedekind (1831—1916), Meray
(1835—1911) and Cantor (1845—1918). Reliance on somewhat vague
geometrical intuitions was replaced by a definition of the real numbers
as certain objects constructed out of natural numbers, integers or rational
numbers. The properties of the real numbers were thereby reduced ul
timately to properties of natural numbers. As Poincar£ said in 1900,
“Today there remain in analysis only integers or finite or infinite systems
of integers, interrelated by a net of relations of equality or inequality.”
The definition of the real numbers from natural numbers, integers
or rationals can be given in several ways. All lead to the same abstract
structure of the real number continuum. In other words, what each of
the definitions accomplishes is to provide a representation (§ 8) of the real
numbers by objects constructed (directly or indirectly) out of natural
numbers.
We have used the representations by the infinite decimal or dual fractions
(§§ 2, 5). In principle any one of the sets proved equivalent to these (§ 5),
e.g. the sets of natural numbers, could be used, but in practice one will
choose a representation which makes it simple to define the properties
of the real numbers.
A representation which makes the ordering of the real numbers es
pecially perspicuous is that by Dedekind cuts (1872). Suppose the rational
numbers R have been separated into two non-empty classes X lt X 2 such
that every rational in X x is < every rational in X 2. Such a separation is
called a Dedekind cut {in R). In case there is neither a greatest rational in
the lower set X x nor a least in the upper X 2, the cut is called open. Dede
kind's insight was that irrationals are called for exactly where the open
cuts occur. A rational goes with either of two closed cuts, one for which it
is the greatest in X v and the other for which it is the least in X 2. In order
to have a unique representative of each real number (rational or irra
tional), we can use the lower sets X x of the cuts for which X 1has no great
§9 NUMBER THEORY VS. ANALYSIS 31
est. This gives us the following definition (writing x in place of X v and
R — x in place of X 2).
A real number is a set x of rationals such that:
(a) Neither x nor R — x is empty, (b) x contains no greatest rational,
(c) Every rational in x is < every rational in R — x.
The set C of reals is the set of all such sets x of rationals.
This definition makes use of the presupposed system R of rationals
to construct the representatives of the reals, without taking R into the
resulting system C as a subsystem. (If the members of R are individuals,
the members of C are sets of those individuals.)
We now define a real number x to be rational, if R — x has a least
member in which case x is said to correspond to the rational x (of the
system R). Otherwise, x is irrational.
The rationals among the reals form a subsystem CR of C which is
isomorphic (§ 8) to the original system R of rationals, as we verify each
time we use the representation to define a notion for the reals which
has previously been defined for the rationals.
E xamples . The real 2 is the set of the rationals < the rational 2,
to which it corresponds. The real \/2 is the set of the rationals which are
either negative or have squares < the rational 2 (among which there is
no greatest). Since the square of no rational = 2 (as Pythagoras discovered
in the sixth century B.C.), R — \/2 consists of the positive rationals
having squares > 2 (among which there is no least), so \/2 is irrational.
The order relation for reals is defined thus: x < y, if there exists a
rational r which is in y but not in x. (Now prove that C is linearly ordered
by < , and that (CR, < ) is isomorphic to (R, <).)
A real number v is an upper bound of a set M of real numbers, if v > x
for every real number x belonging to M.
(A) If a non-vacuous set M of real numbers has an upper bound, it has
a least upper bound u (= l.u .b . M).
P roof. We must construct u as a set of rationals having Properties
(a) — (c). We are given M as a set of such sets of rationals. The definition
of the set u is this: a rational r Cu when and only when, for some real
number x which 8 M, r £ x. In the symbolism of § 5, u = ©M. It is left
to the reader to show now that u = l.u.b. M. (Prove that u is a real
number, u is an upper bound of M, and M has no upper bound v < u.)
Lower bounds are defined similarly. If the real x is rational, let
x == x -f {%}; otherwise, let x = x. Let —x be the set of the rationals
— r forr £ R — x . (If x is ration al, th en — x corresponds to — x.)

Let — M be th e set of th e reals — x for x 8 M. I f w is a low er b o u n d of
M, th e n — w is an u p p er b o u n d of — M, so — M has a l.u .b ., an d
- (l.u.b. - M) = g.l.b. M.
G iv e n reals x an d y, le t x + y b e th e set of th e ratio n als r + s for
r £x an d sZ y ; le t x — y = x + (— y ) ; an d le t [xf = x if x > 0 an d
|x| = — x if x < 0. (D o n o t con fuse + a n d — w ith ad d itio n an d
su b tra ctio n o f sets, w h ich are w ritte n + an d — .)
G iv e n an in fin ite sequence a0, ax, ..., an, ... o f reals an d a real
a, w e s a y th a t lim an = a, e > 0, there is a n a tu ra l
if for e v e r y real
n u m b er ne such th a t, for e v e r y n > ne, |an —a| < e. F o r e x a m p le ,
lim l/ 2 n = 0 (where l/ 2 n is th e real correspon ding to th e ra tio n al l/2 n).
(B) If u = l.u .b . M (as in (A)), there exists a sequence a0, a^ ...,
an, ... of members of M such that lim an = u .
P roof. L e t Mn — th e set o f th e reals w h ich 8 M an d are > u — l/2n.
(P ro ve th a t Mn is n o t e m p ty .) L e t an b e a n y real chosen from Mn.
(P ro ve th a t lim an = u.)
N o tw ith s ta n d in g t h a t in th is th e o r y an alysis is “ a r ith m e tiz e d ” , th e

d istin ctio n b etw ee n arith m e tic an d an alysis rem ains sharp, in th a t
an aly sis find s it n ecessary to e m p lo y in fin ite sets of th e o b je cts of arith m e
tic as its ob jects.
§10. Functions. In th e m ost gen eral sense, a (single-valued) function

f or f(x) or y = f(x) of one variable x is a correspondence b y w h ich , to each
elem ent # of a set X th ere corresponds a single elem ent y of a set Y .
T h e set X is th e range of the independent variable, or th e domain of the
function. T h e fu n ctio n m a y b e ca lled a function from X to Y (or a function
of a m em ber of X taking a m em ber of Y as value, or an operation on a
m em ber of X producing a m em ber of Y, etc.).
T h e range of the dependent variable y or f(x) is th e su b set Y x of Y
com p risin g th e elem ents of Y used in th e correspondence, i.e. th ose w h ich
correspond b y th e fu n ctio n / to som e elem ent of X. T h e n X an d Y x
are in many-one correspondence, since to each elem en t o f X there cor
responds ju s t one elem ent of Y v b u t an elem en t of Yx w ill (in general)
correspond to m a n y elem en ts of X. A n elem ent x of X is an argument of
the function or a value of the independent variable. T h e correspon ding
elem en t y of Y is th e corresponding value of the function or of the dependent
variable, or th e value of the function for that argument. (Som etim es “ ar
g u m e n t” is en co u n tered m ean in g “ in d ep en d en t v a r ia b le ” .)
§10 FUNCTIONS 33
A (single-valued)function f or /(% , . . xn) or y — f(xv . . . , xn) of
n variables xv ..., xn is a correspondence b y w h ich , to each ordered
ft-tuple (xv . . xn) of o b je cts w here xxZ X lt x28 X 2, ..., xnZ X n> there
corresponds a single o b je c t y w here y EY. A fu n c tio n of n v a ria b le s can
be considered as a fu n ctio n of one v a riab le, w ith X as th e class of all th e
ordered n-tu ples (xv . . . , x n). Sim ilar te rm in o lo g y applies. T h u s X x is
th e range of xlf X 2 of x2, ..., X n of xn. H ere X lf X 2, ..., X n m a y all b e
th e sam e set, or there m a y b e several (up to n) differen t ranges. A p a r
ticu lar sequence xlf ..., xn of elem ents from X v ..., X n, re sp e c tiv e ly , is
a set (or n-tuple) of arguments.
In th is p le th o ra o f term in o lo gy, one m a y recognize a m ix tu r e of
term inologies b a sed on tw o id e a s : th e id ea of a fu n ctio n as a m a n y -o n e
correspondence, an d th e id ea of a fu n ctio n as a v a ria b le y w h ich ranges
in relation to an o th er v a ria b le x so th a t th e v a lu e of y is a lw a y s fix e d b y
th a t of x.
T h e first id ea is th e m ore com p reh en sive one, w h ich th e stu d e n t
should keep up p erm o st in his m ind. T h e second id ea h o w e ver g iv e s rise
in a n a tu ral w a y to th e useful n o ta tio n a l co n ven tio n , w h e re b y if "/(# )”
for ih sta n ce stan d s for a ce rta in fu n ctio n of th e in d ep en d e n t v a ria b le
x, an d a, b, etc. are v a lu es of th e in d ep en den t v a ria b le (i.e. argu m en ts),
th en “{(a y ’ stan d s for th e v a lu e of th e fu n ctio n for th e a rgu m e n t a,
"/(£>)” f ° r th e v a lu e w h en x = b, etc.
O n e sh ould b e aw are t h a t th e n “f(x)” m a y h a v e either of tw o m e a n in g s :
1. T h e fu n ctio n itself (i.e. th e m a n y -o n e correspondence b e tw e e n X an d
Y x). x stan d s for an o b je c t from th e d om ain, th e correspon ding
2 . W hen
v a lu e of th e fu n ctio n (i.e. a m em ber y of Y x). W h e n x is un sp ecified , th e
la tte r is ca lled th e ambiguous value of th e fu n ction .
E xample 1. W h e n w e s a y “x + y is s y m m e tric ” w e m ean b y “x + y”

th e fhn ction . W h e n w e s a y " th e sum x + y of a n y tw o n a tu ra l n u m bers
x an d y is > x”, w e m ean b y “x + y” n o t a fu n ctio n b u t a n u m b er (the
am b igu ou s v a lu e of th e function ).
T h is situ a tio n ca n b e a v o id e d b y u sin g in stea d of " / ( * ) ” f ° r th e

fu n ctio n , so lo n g as w e are ta lk in g o n ly a b o u t fu n ctio n s for each of
which a sy m b o l such as “g”, " + ” or ‘ V ’ has been in tro d u ced . B u t
n o tatio n s w h ich sh ow th e in d ep en d en t v a ria b les are v e r y co n ven ien t for
n am in g oth er fu n ctio n s com posed o u t of those fu n ctio n s (and co n sta n ts),
e.g. "** + 3*” or > (2,*)”.
E xample 2. T o consider th is in m ore d eta il, s a y th a t f an d g are
g iv e n number-theoretic functions of one v a ria b le each, i.e. fu n ctio n s from
th e set o f th e n a tu r a l n um bers to th e sam e set. L e t x b e a n y n a tu r a l
num ber. T h e n g(x) is a n a tu ra l num ber, i.e. th e v a lu e o f g for x as argu m en t,
an d f(g(x)) is a n a tu r a l n um ber, i.e. th e v a lu e of / for th e n a tu ra l n u m ber
g(x) as argu m e n t. So to a n y n a tu ra l n u m ber x, an o th er n u m b er f(g(x))
is determ in ed . T h u s “ /(£(*))” stands for th e am b igu o u s v a lu e o f a new
fu n ctio n (M ean ing 2 ); an d it is also convenient to use it as a name for
th e n ew fu n c tio n itse lf (M eaning 1).
T h e re is an o th er n o ta tio n (due to C h u rch 1932) in w h ich th e in d e

p en d en t v a ria b le s app ear, b u t w h ich represents th e fu n ctio n / as d istin c t
from its am b igu o u s v a lu e , n a m e ly :' (\x f(x) ” , or for a fu n ctio n of n va riab les,
“l x 1 . . . x nf{x 1 , . . . , x n)"; e.g. <(\ x f ( g ( x ) y \ “%xx* + 3 x " , “ X*<p(2 , * ) ” . W e
sh all use th is X -notation for em phasis in situ atio n s w here especial care is
required.
E xample 3 . Let 9 b e a fu n ctio n of tw o num bers. U s in g th e X -notation

co n siste n tly , i.e. w h e n e v e r w e m ean th e fu n ctio n in stea d of th e am b igu o u s
y(x, y ), (b) th e fu n ctio n Xx 9 (x, y)
v a lu e , w e ca n d istin gu ish (a) th e n u m b er
o f one v a ria b le x t w ith y as param eter, (c) th e fu n ctio n X*y<p(#, y) o f tw o
va ria b les, w ith x as first a n d y as second va ria b le , (d) th e fu n ctio n
\y x 9 (xf y) , w ith y as first an d x as second v a riab le, (e) th e fu n ctio n X#Xy 9 (x, y)
o f one v a ria b le x, w h ose v a lu e s are fu n ctio n s of a n o th er v a ria b le y , etc.
(Schonfin kel 1924 an d C h u rch id e n tify (c) an d (e), b u t t h a t is n o t n ecessary
for us.)
F o r a n y w -tu ple tv tn o f a rgu m en ts for /,

...,
{ X % . . . x nf{xv . ..,# „ ) } ( * ! ,. . . , * » ) = * •>*»)•

F o r e x a m p le , {k x x 2 + 3 #}(2 ) = 10 , {Xx <p(x, y)}( 0 ) = 9(0, y),
{kyx 9(x, y)}(0 , 3 ) = 9(3, 0 ), {X*y 9(3, y)}{z, x) = 9(2, x).

W e h a v e d escribed a fu n ctio n as a m a n y -o n e correspondence. O n e
m a y go fu rth er in s a y in g w h a t a m a n y -o n e correspondence is to be,
acco rd in g to th e k in d of th e o r y one is w o rk in g in. In se t-th e o retic term s,
th e correspondence ca n b e id en tified w ith th e set of all th e ordered pairs
(x, y) o f correspon din g elem en ts o f X an d Yv O n e m a y sp eak in stea d of
th e la w or rule e sta b lish in g th e correspondence, a t least in d ealin g w ith
su ch fu n ctio n s t h a t a la w or rule in som e u n d erstoo d sense ca n b e g iv e n
for each fu n ctio n . In th e case t h a t X is a fin ite set, a fu n ctio n ca n b e g iv e n
as a tab le.
E xample 4 . Let X an d Y b o th b e th e residues m od u lo 2, i.e.

X = y = {0, 1}. T h e fu n ctio n s x' an d x • y can b e d efined b y th e fo llo w
§ 10 FUNCTIONS 35
in g t a b le s :
%' x •y
y 0 i
X 0 1 * 0 0 0
i 0 i 0 i
T h u s b y the second ta b le, 0 •0 = 0 • 1 = 1-0 = 0 an d 1-1 = 1.

Ch a p t e r III
A C R IT IQ U E O F M A T H E M A T IC A L R E A S O N IN G
§ 11. T h e p a r a d o x e s . T h is ch ap ter is in ten d ed to present th e problem

s itu a tio n o u t of w h ich th e in vestigatio n s to be reported in th e rest of th e
b o o k arose, i.e. th e situ atio n preceding those in v e stig a tio n s (but n o t
h o w it h as since changed).
In th e arith m e tiz a tio n of an alysis (§ 9 ), an in fin ite co llectio n (of ra
t i o n a l fo rm in g th e lower h alf of a D ed ek in d cu t, or of d igits in sequence
fo rm in g a n o n -term in a tin g d ecim al, etc.) is c o n stitu te d an o b je c t, an d th e
set of all such o b jects is considered as a new collection . F ro m th is it is a
n a tu r a l ste p to C an to r's general set th eory.
H a r d ly h ad these theories been con solidated , w hen th e v a lid it y of
th e w h o le con stru ction w as ca st in to d o u b t b y th e d isco v e ry of p ara d o xes
or an tin om ies in th e fringes of th e th e o r y of sets.
(A) The B urali-F orti paradox 18 9 7 *, also kn ow n to C a n to r in 1895 ,

arises in C a n to r ’s th e o r y of tran sfin ite ordinals.
(B) S o m ew h a t sim ilar an tin om ies occur in th e th e o ry of tran sfin ite

card in als, p a rtic u la r ly Cantor's paradox (found b y him in 1899 ). Consider
th e set of all s e t s ; ca ll it M. B y C a n to r’s theorem (Theorem C § 5 ), UM > M.
A lso , since M is th e set of all sets, an d UM is a set of sets (nam ely, th e set
of th e su bsets of M), UM C M. H en ce by C o ro llary A T h eo rem A,
UM < M; an d so b y § 3 , not UM > M- T h u s w e h a v e p ro v ed b o th th a t
UM > M an d t h a t n o t UM > M.
S ta r tin g w ith th e sam e M, w e can also reach a p a ra d o x thus. T o each
m em ber M o f M, i.e. to a n y set M , b y T h eorem C there is an oth er m em ber
M ' of M, n a m e ly U M , such th a t M < M '. H en ce b y T h eo rem D , M < @M
for e v e r y m em ber M of M. B u t M is th e set of all sets, so @M is one of its
m em bers. T a k in g th e M in th e in e q u a lity ju st p ro ved to b e th is m em ber,
w e h a v e @M < @M. B u t b y § 3 , for a n y set M , n ot M < M ; hence in
p articu lar, n o t @M < @M.
T h e p a ra d o x w ith @M results likew ise, if w e sta rt o u t w ith th e set of
a ll card in al num bers, an d choose as M a set co n tain in g, to each card in al
n um ber, a set M h a v in g th a t cardinal.
36
§11 THE PARADOXES 37
I f th e n otion of sets of arb itra ry elem ents used here is th o u g h t to be
to o v a g u e an d hence u n m ath e m a tica l, w e ca n prescribe as adm issible
elem ents of sets (ax) th e n a tu ral num bers 0, 1 , 2 , . . . (or (a2) th e e m p ty set
O) an d (b) arb itra ry sets w hose m em bers are ad m issib le elem ents. W ith
this prescription, th e a b o v e p arado xes an d th e n e x t arise as before (w ith
(ax), G e n tz e n 1936).
(C) The R ussell paradox 19 0 2-3*, d iscovered in d e p e n d e n tly b y Zerm elo,

deals w ith th e set of all sets w h ich are not m em bers of them selves. C a ll
th is set T. Is T a m em ber of itself ?
L e t us assum e, for th e sake of th e argum ent, t h a t T is a m em ber of
itself, i.e. in sy m b o ls T £ T. T h e assum ption says t h a t T is a m em ber of
T f i.e. T is a m em ber of th e set of all sets w h ich are n o t m em bers of
th em selves, i.e. T is a set w h ich is n o t a m em ber of itself, i.e. in sym b o ls
T$T. T h is co n trad icts th e assu m p tion T&T. T h u s far w e h a v e no p ara d o x,
as th e co n trad ictio n b etw ee n T&T an d T t T has arisen o n ly under th e
assum ption T&T. By redu ctio ad absurdum , w e co n clu d e t h a t th e
assu m ption is false. T h u s w e h a v e n ow p ro ved o u trig h t, w ith o u t as
sum ption , th a t T IT .
T t T, w e ca n argue further. T h e result sa y s
F ro m th e estab lish ed result
th a tT is n o t a m em ber of th e set of all sets w h ich are n o t m em bers of
th em selves, i.e. T is n o t a set w h ich is n o t a m em ber of itself, i.e. T is
a set w h ich is a m em ber o f itself, i.e. in sy m b o ls T&T. N o w T t T an d
T& T are b o th established , so w e h a v e a p ara d o x.
T h is p a ra d o x can b e e x tra c te d from C a n to r's thus. I f w e prescribe
(a2) an d (b) as adm issible elem ents, so th a t sets h a v e o n ly sets as m em bers,
th en w hen M is th e set of all sets, UM = M, an d th e set T of th e p a ra d o x
is o b ta in ed b y a p p ly in g th e proof of L e m m a A § 5 to th e id en tica l 1-1
correspondence M ^ UM in w h ich each elem ent of M corresponds to itself
in UM.
A p o p u la riza tio n of th e p a ra d o x (R ussell 1919) concerns th e b arb er in
a certain villa g e , w h o sh ave s all an d o n ly those persons in th e v illa g e w ho
do not sh a ve them selves. D oes he sh a v e h im self ? (Of course here w e ca n
escape th e p a ra d o x s im p ly b y co n clu din g th a t there n ever w as such a
barber.)
E v e r y m u n ic ip a lity in H o lla n d m u st h a v e a m ayo r, an d no tw o m a y
h a v e th e sam e m ayo r. Som etim es it h ap pen s th a t th e m a y o r is a n on
resident of th e m u n icip a lity . Suppose a la w is passed s e ttin g aside a
special area S e x c lu s iv e ly for such non-resident m ayo rs, an d co m p ellin g
all non-resident m ayo rs to reside there.
Suppose fu rth er th a t there are so m a n y non -resid ent m ayo rs th a t S
38 A CRITIQUE OF MATHEMATICAL REASONING CH. Ill
h as to b e c o n stitu te d a m u n ic ip a lity . W h ere sh all th e m a y o r of S reside ?
(M an n ou ry, cf. v a n D a n tz ig 1948.)
S u p p o se th e L ib ra ria n of Congress com piles, for inclusion in th e
L ib r a r y o f Congress, a b ib lio g r a p h y of all those b ib lio grap h ie s in th e L i
b r a r y o f Congress w h ic h d o not list th em selves. (G on seth 1933.)
R u ssell also sh ow ed h o w to recast his p a ra d o x in lo g ica l in stea d of se t-
th eo retic term in o lo gy. A p ro p e r ty is called 'predicable* if it ap p lies to
itself, ‘im predicable* if it does n o t a p p ly to itse lf. F o r exam p le , th e p ro p
e r ty 'ab stract* is a b stra c t, an d hence p red icab le; b u t 'concrete* is also
a b s tr a c t an d n o t con crete, an d hence is im predicable. W h a t a b o u t th e
p ro p e r ty 'im predicable* ?
(D) The R ichard paradox 1905, also su b sta n tia lly g iv e n b y D ix o n 1906,
deals w ith th e n o tio n o f fin ite d e fin a b ility . F o r definiteness, le t th is refer
to a g iv e n lan gu age , s a y th e E n g lis h lan gu age w ith a preassigned a lp h a b et,
d ic tio n a r y an d gram m ar. T h e a lp h a b e t w e m a y ta k e as co n sistin g of
th e b la n k space (to sep arate words), th e 26 L a tin letters, an d th e com m a.
B y an 'expression* in th e la n g u a g e w e m a y u n d ersta n d sim p ly a n y fin ite
seq uen ce of th ese 28 sy m b o ls n o t b egin n in g w ith a b la n k space. T h e
expressions in th e E n g lis h la n g u a g e ca n th en b e en u m erated b y th e d e v ic e
w h ich w e a p p lied a t th e end of § 1 to th e en u m eratio n o f th e a lg e b ra ic
equ ation s.
A n expression m a y d efin e a n u m b er-th eo retic fu n ctio n o f one v a ria b le
(i.e. a fu n ctio n of a n a tu ra l n um ber ta k in g a n a tu ra l n um ber as v a lu e ).
F ro m th e sp ecified en u m eratio n o f all th e expressions in th e E n g lish
la n gu a ge , b y strik in g o u t th ose w h ich d o n o t d efine a n u m b er-th eo retic
fu n ctio n , w e o b ta in an en u m eratio n (say E 0, E lf E 2, . . . ) o f those w h ich
do (say th e fu n ctio n s d efin ed are re sp e c tiv e ly fQ(n), i2{n), ...).
N o w consider th e fo llo w in g expression, " th e fu n ctio n w hose va lu e , for
a n y g iv e n n a tu ral n u m ber as argu m en t, is eq u a l to one m ore th a n th e
v a lu e , for th e g iv e n n a tu ra l n u m ber as argu m en t, of th e fu n ctio n defined
b y th e expression w h ich corresponds to th e g iv e n n a tu ral num ber in th e
last described en u m eratio n ’ *.
In th e q u o te d expression w e refer to th e a b o v e d escribed enu m eration
of th e expressions in th e E n g lis h la n gu a ge d efin in g a nu m b er-th eo retic
fu n ctio n , w ith o u t d efin in g it. B u t w e cou ld easily h a v e w ritte n in th e
d efin ition of th a t en u m eration in full, as p a rt of th e q u o ted expression.
W e sh ould th en h a v e before us a d efin ition of a fu n ctio n (briefly, th e
fu n ction f n(n) + 1), by an expression in th e E n g lish language. T h is
fu n ctio n , b y its defin itio n , m u st differ from e v e ry function d efinable b y
an expression in th e E n glish lan gu age.
§11 THE PARADOXES 39
T h is p a ra d o x is esp e c ia lly in terestin g for its im p lica tio n s con cern in g

lan gu age s such as E n g lish , an d b ecau se it runs so close to C a n to r's proof
of th e n o n -en u m era b ility of th e n u m b er-th eo retic fu n ctio n s (§ 2). R ic h a rd
g a v e th e p a ra d o x in a form re la tin g to th e d efin itio n o f a real num ber,
p arallelin g C a n to r's proof of th e n o n -en u m era b ility of th e real num bers.
Consider th e expression, “ th e least n a tu ra l n u m ber n o t n a m ea b le
in few er th a n tw e n ty -tw o s y lla b le s ". T h is expression nam es in t w e n ty -
one syllab les a n a tu ral n u m b er w h ich b y d efin itio n ca n n o t b e n a m ed in
few er th a n tw e n ty -tw o s y lla b le s ! (B erry 1906.)
(E) T h ese m o d e m p arado xes, w h ich fall m ore or less w ith in th e c o n te x t

of set th e o ry , are related to a v e r y an cien t one.
T h e s ta te m e n t “ C retan s are a lw a y s liars . . . " is a ttr ib u te d to th e
philosoph er E p im en id e s o f C rete (sixth c e n tu r y B .C .). (Th e sta te m e n t w as
q u o te d b y P a u l in “ E p is tle to T it u s " , I, 12, as b y a C re ta n “ p r o p h e t" ,
w h o m e a rly C h ristia n trad itio n , acco rd in g to m ore recent sources, id e n ti
fied w ith E p im en id e s. Cf. W e y l 1949 p. 228 .)
S u ppose w e d istin gu ish tw o kin d s of lia r s : liars of th e first kin d , w h o
te ll th e tr u th som e of th e tim e, an d liars of th e second kin d , w h o te ll o n ly
lies. L e t us in terpret E p im e n id e s' sta te m e n t to m ean t h a t a ll C reta n s are
liars of th e second kin d . Su p p ose his sta te m e n t w ere true. B y w h a t it
sa y s an d th e fa c t he is a C retan , it m u st th e n b e false. T h is is a co n
tra d ictio n ; hence b y red u ctio ad absurdum , th e sta te m e n t m u st b e false.
T h e fa ls ity of th e s ta te m e n t requires th a t there h as been, or w ill e v e n tu a lly
be, a C reta n w h o a t som e tim e tells th e tru th . H a d th e q u o te d s ta te m e n t
been th e o n ly one a n y C re ta n m akes, w e sh ou ld h a v e a p a ra d o x. I t is
lo g ic a lly u n sa tis fa c to ry t h a t w e sh ould escape p a ra d o x o n ly th ro u g h th e h is
to rica l accid en t t h a t som e C retan ex iste d w h o so m etim e to ld th e tru th .
The E pim enides paradox , kn o w n also as the lia r , ap pears in sta rk form ,
if a person sa y s sim p ly , “ T h is sta te m e n t I am n o w m a k in g is a lie ." T h e
q u o te d sta te m e n t ca n n eith er b e true nor false w ith o u t e n ta ilin g a co n
trad ictio n . T h is versio n of th e p a ra d o x is a ttr ib u te d to E u b u lid e s (fourth
c e n tu r y B .C .), an d w as w ell k n o w n in an cien t tim es. (Cf. R iis to w 19 10 .
I f th e sta te m e n t “ C retan s are a lw a y s liars ..." is n o t a u th e n tic a lly
E p im e n id e s', or w as n o t o rig in a lly recogn ized as p a ra d o xica l, th e E u b u li
des version of th e L ia r m a y th e n b e older th a n th e “ ly in g C r e ta n " version.)
In th e an cie n t “ d ilem m a of th e cro co d ile", a crocodile h as stolen a
child. T h e crocodile prom ises th e c h ild ’s fa th er to return th e ch ild , p ro v id
ed th a t th e fa th er guesses w h e th er th e crocodile w ill retu rn th e ch ild
or not. W h a t sh ould th e crocodile do, if th e fa th e r guesses t h a t th e
crocodile w ill n o t return th e ch ild ? (Cf. P r a n tl 18 5 5 p. 493 .)
T h e fo llo w in g riddle also tu rn s u pon th e p ara d o x. A tra v e lle r h as fallen
am o n g cannibals. T h e y offer h im th e o p p o r tu n ity to m a k e a sta te m e n t,
a tta c h in g th e co n ditio n s t h a t if his sta te m e n t b e true, h e w ill b e boiled,
a n d if it b e false, he w ill b e roasted. W h a t sta te m e n t sh ould he m a k e ?
(A form of th is riddle occurs in C e rv a n te s' " D o n Q u ix o te " ( 1605 ), II, 51 .)
§ 12. First inferences from the paradoxes. T h e reader m a y t r y
h is h a n d a t so lv in g th e p arad o xes. In th e h a lf c e n tu r y since th e problem h as
b een open, no so lu tio n h as been fo u n d w h ich is u n iv e rsa lly agreed upon.
T h e sim p lest k in d o f solu tio n w o u ld b e to lo ca te a sp ecific fa lla c y , like
a m ista k e in a stu d e n t's algeb ra exercise or g e o m e try proof, w ith n o th in g
else n eedin g to b e ch an ged .
Id e a s for so lv in g th e p ara d o xe s in th is sense com e to m in d on first
con siderin g th em . O n e m a y propose t h a t th e error in th e p ara d o xe s
(A) — (C) con sists in u sin g to o large sets, such as th e set of all sets or
th e set o f all card in al n u m b ers; or in p e rm ittin g sets to b e considered as
m em bers of th em selves, w h ic h a ga in argues a g a in st th e set of all sets.
T h ese suggestio n s are n o t n ecessarily w ron g, b u t t h e y are n o t a fter all
sim ple. T h e y le a v e us th e problem of refo u n d in g set th e o r y on a d ra s tic a lly
alte red basis, th e d e ta ils of w h ich are n o t fu lly im p licit in th e suggestion s.
F o r ex am p le, if w e b a n th e set of all card in al num bers, w e are u n a b le
to in tro d u ce th e set of th e n a tu ra l num bers, unless w e a lre a d y k n o w t h a t
t h e y are n o t all th e ca rd in al n u m b ers; an d th e sam e d iffic u lty w ill arise
a t h igh er stages. I f w e b a n th e set of all sets, w e fin d ourselves in co n flict
w ith C a n to r's d efin itio n o f set. In order to h a v e set th e o r y a t all, w e m u st
h a v e theorem s a b o u t all sets, an d all sets th e n c o n stitu te a set under
C a n to r's defin ition . I f n o t so, w e m u st s a y w h a t o th er d efin itio n of set w e
sh all use in stead , or w e m u st su p p lem en t C a n to r's d efin itio n w ith som e
fu rth e r criterion to determ in e w h en a collection of o b je c ts as described
in his d efin itio n sh all c o n stitu te a set (Skolem 1929-30).
A xiomatic set theory. R eco n stru ctio n s of set th e o r y ca n b e g iv e n ,

p la c in g around th e n o tio n of set as few restrictions to e x clu d e to o large
sets as a p p ear to b e required to forestall th e k n o w n an tin om ies. S in ce
th e free use o f our co n cep tio n s in co n stru ctin g sets under C a n to r's d e f
in itio n led to disaster, th e notion s o f set th e o r y are go ve rn e d b y axio m s,
lik e those go v e rn in g ‘p o in t' an d ‘line' in E u c lid e a n p lan e g e o m etry . T h e
first sy ste m of axiom atic set theory w as Z erm elo's (1908). R e fin em en ts
in th e a x io m a tic tre a tm e n t o f sets are due to F ra e n k e l (1922, 1925),
S k o lem (1922-3, 1929), v o n N eu m a n n (1925, 1928), B e rn a y s (1937-48),
an d others. A n a ly s is ca n b e fou n d ed on th e basis of a x io m a tic set th e o ry,
§ 12 FIRST INFERENCES FROM THE PARADOXES 41
w h ich perh aps is th e sim plest basis set u p since th e p ara d o xe s for th e
d ed u ctio n of e x istin g m a th em a tics. Som e v e r y in terestin g d iscoveries
h a v e been m ad e in co n n ectio n w ith a x io m a tic set th e o ry, n o ta b ly b y
Skolem (19 2 2-3 ; cf. §75 below) an d G o d el (1938, 1939, 1940).
T he broader problem of foundations. A ssu m in g t h a t th e p ara

do xes are a v o id e d in th e a x io m a tiz a tio n of set th e o ry — an d of th is th e
o n ly assurance w e h a v e is th e n e g a tiv e one th a t so far none h a v e b een
encountered — does it c o n stitu te a fu ll solutio n of th e problem posed
b y th e p ara d o xe s?
I n th e case of g e o m e try , m a th e m a ticia n s h a v e reco gn ized since th e
d isc o v e ry of a n o n -E u clid e a n g e o m e try th a t m ore th a n one k in d of sp ace
is possible. A x io m sy ste m s serve to single o u t one or an oth er k in d of space,
or certain com m on featu res of several spaces, for th e geom eter to stu d y .
A co n tra d ictio n arising in a form al a x io m a tic th e o ry can m ean sim p ly
th a t an u n realizab le co m b in atio n of featu res h as been p o stu lated .
B u t in th e case of arith m e tic an d an alysis, theories cu lm in a tin g in
set th eo ry, m a th e m a ticia n s prior to th e current epoch of criticism gen eral
ly supposed th a t th e y w ere d ealin g w ith sy ste m s of o b je cts, set up
g e n e tic a lly , b y defin ition s p u rp o rtin g to estab lish their stru ctu re c o m
p le te ly . T h e theorem s w ere th o u g h t of as expressin g tru th s a b o u t th ese
sy stem s, rath er th a n as propositions a p p ly in g h y p o th e tic a lly to w h a te v e r
sy ste m s of o b je cts (if an y) s a tis fy th e axiom s. B u t th e n h o w co u ld co n
trad ictio n s h a v e arisen in these su b jects, unless th ere is som e d efect in
th e logic, som e error in th e m eth o d s of co n stru ctin g an d reasoning a b o u t
m a th e m a tic a l o b jects, w h ich w e h a d h ith erto tru ste d ?
T o s a y th a t n o w these su b je cts sh ould in stea d b e estab lish ed on an
a x io m a tic b asis does n o t of itself dispose of th e problem . A fte r axio m atic
za tio n , there m u st still b e som e le v e l a t w h ich w e h a v e tru th an d fa ls ity .
I f th e a x io m a tics is inform al, th e ax io m s m u st b e true. I f th e a x io m a tic s
is form al, a t least w e m u st b e lie v e t h a t th e theorem s do follow from th e
a x io m s; an d also there m u st b e som e relationship b etw ee n th ese results
an d som e a c tu a lity ou tsid e th e a x io m a tic th eo ry, if th e m a th e m a ticia n s'
a c t iv it y is n o t to reduce to nonsense. T h e fo rm a lly a x io m a tiz e d p rop o
sitions of m a th e m a tics ca n n o t c o n stitu te th e w h ole of m a th e m a tic s ;
th ere m u st also b e an in tu itiv e ly u nderstoo d m a th em a tics. I f w e m u st
g iv e u p our form er b elief t h a t it com prises .all of arith m etic, an alysis an d
set th e o ry , w e sh all n o t b e w h o lly satisfied unless w e learn w herein th a t
belief w as m ista k en , an d w here n o w in stea d to d raw a line of separation.
T h e im m e d ia te p roblem of elim in atin g th e p ara d o xes th u s m erges
w ith th e b road er problem o f th e fou n d ation s of m a th e m a tics an d logic.
W h a t is th e n atu re of m a th e m a tic a l tru th ? W h a t m ean in g do m a th e m a t
ica l propositions h a v e , an d on w h a t eviden ce do th e y rest ? T h is b ro ad
problem , or co m p le x o f problem s, ex ists for p h ilo so p h y a p a rt from th e
circu m stan ce t h a t p ara d o xe s h a v e arisen in th e fringes of m a th em a tics.
H isto r ic a lly , th is circu m stan ce has led to a m ore in te n sive s tu d y of th e
p rob lem on th e p art of m a th e m a ticia n s th a n w o u ld otherw ise h a v e
b een l i k e l y ; an d th e p ara d o xe s o b v io u s ly im pose con d ition s on th e
solutio n of th e problem .
I mpredicative definition . W h e n a set M an d a p articu lar o b je c t m

are so defin ed th a t on th e one h a n d m is a m em ber of M, an d on th e
o th er h an d th e defin itio n of m depend s on M , w e s a y th a t th e procedure
(or th e d efin ition of m, or th e d efin ition of M ) is im predicative. S im ilarly,
w h e n a p ro p e r ty P is possessed b y an o b je c t m w hose d efin itio n d ep en d s
on P (here M is th e set of th e o b je cts w h ich possess th e p ro p e r ty P).
A n im p re d ic a tiv e d efin itio n is circular, a t least on its face, as w h a t is
d efin ed p a rticip a te s in its ow n d efinition.
E a c h of th e an tin om ies of § 11 in v o lv e s an im p re d ic a tiv e d efinition.
In (B), th e set M of all sets includ es as m em bers th e sets UM an d @M
d efin ed from M. T h e im p re d ic a tiv e procedure in th e R u ssell p a ra d o x (C)
stan d s o u t w h en th e d efin itio n of T is elab o rated th u s. W e d iv id e th e set
M of all sets in to tw o p arts, th e first co m p risin g th ose m em bers w h ich
co n ta in th em selves, a n d th e second (w hich is T) th ose w h ich do not.
Then we put T (defined b y th is d ivisio n of M in to tw o parts) b a c k in to M,
to a sk in to w h ich p a rt of M it falls. In th e R ic h a rd p a ra d o x (D), th e
t o t a l it y of expressions in th e E n g lish la n g u a g e w h ic h c o n s titu te d ef
in itio n s of a fu n ctio n (real num ber, n a tu ra l num ber) is ta k e n as in c lu d in g
th e q u o te d expression, w h ich refers to th a t to ta lity . In th e E p im e n id e s
p a ra d o x (E), th e t o t a lit y of sta te m e n ts is d iv id e d in to tw o parts, th e tru e
a n d th e false statem e n ts. A sta te m e n t w h ich refers to th is d ivisio n is
reckon ed as of th e original t o ta lit y , w h en w e a sk w h e th e r it is tru e or
false.
Poin car6 (19 0 5-6 ,19 0 8 ) ju d g e d th e cause of th e p ara d o xe s to lie in th ese
im p re d ic a tiv e d efin itio n s; an d R u ssell (1906, 1910) en u n cia ted th e sam e
e x p la n a tio n in his vic io u s circle p rin cip le: No t o t a lit y ca n c o n ta in
m em bers d efin ab le o n ly in term s of th is t o ta lit y , or m em bers in v o lv in g or
p resu pp o sin g th is t o ta lit y . T h u s it m ig h t ap p ear t h a t w e h a v e a su fficien t
so lu tio n an d a d e q u a te in sigh t in to th e p arad o xes, e x c e p t for one cir
cu m sta n c e : p arts of m a th e m a tic s w e w a n t to retain , p a rtic u la r ly an aly sis,
also co n ta in im p re d ica tiv e definitions.
A n e x a m p le is th e d efin itio n of u = l.u .b . M (§ 9 (A)). U n d e r th e
§J2 FIRST INFERENCES FROM THE PARADOXES 43
D e d ek in d c u t d efin itio n of th e real num bers, th e set C of real num bers
is th e set of all sets x of ratio n als h a v in g three properties (a), (b), (c).
N o w th is t o t a lit y has been d iv id e d in to tw o p arts, M an d C — M. W e
define u as ©M, an d th en reckon th is set ©M as a m em ber of C. T h is
defin ition u = ©M depend s on C in th e gen eral case, since in th e gen eral
case M w ill h a v e been d efined from C as th e set of those m em bers o f C
w h ich h a v e a certain p ro p e rty P.
O n e ca n a tte m p t to d efend th is im p re d ica tiv e d efin ition b y in ter
p retin g it, n o t as d efin in g or c rea tin g th e real n um ber u for th e first tim e
(in w h ich in terp retatio n th e d efin ition of th e t o t a lit y C of real num bers
is circular), b u t as o n ly a description w h ich singles o u t th e p a rticu la r
num ber u from an a lre a d y e x istin g t o ta lit y C of real num bers. B u t th e
sam e argu m en t can b e used to u p h o ld th e im p re d ica tiv e d efin ition s in th e
paradoxes.
W eyl 's constructive continuum . T h e im p re d ica tiv e ch aracter of

som e of th e d efinitions in an alysis has been esp ecially em p h asized b y
W e y l, w ho in his b o o k “ D as K o n tin u u m (Th e c o n tin u u m )” (1918)
u n d erto o k to fin d o u t h ow m u ch of an alysis cou ld b e re co n stru cted
w ith o u t im p re d ica tiv e d efinitions. A fu n d of operatio ns ca n b e p ro v id e d
for co n stru ctin g m a n y p a rticu la r categories of irrationals. W e y l w a s th u s
able to o b ta in a fair p a rt of an alysis, b u t n o t th e theorem th a t an ar
bitrary n o n -e m p ty set M of real num bers h a v in g an u p p er b o u n d h as a
least u p p er bound. (Cf. also W e y l 1919.)
T h ere h a v e arisen three m ain schools of th o u g h t on th e fo u n d atio n s of

m a th e m a tic s : (i) th e lo g icistic school (R ussell an d W h ite h e a d , E n g lish ),
(ii) th e in tu itio n istic school (Brouw er, D u tc h ), a n d (iii) th e fo rm alistic or
a x io m a tic school (H ilbert, G erm an). (Som etim es “ lo g istic ” is used in stea d
o f “ lo g ic istic ” ; b u t “ lo g is tic ” also h as an oth er m ean in g § 15 .) T h is b ro ad
classificatio n does n o t in clu d e va rio u s o th er p o in ts of v ie w , w h ich h a v e
n o t been as w id e ly c u ltiv a te d or do n o t com prise to a sim ilar degree b o th
a recon stru ctio n of m a th e m a tic s an d a p h ilo so p h y to su p p o rt it.
Logicism. T h e lo g icistic thesis is th a t m a th e m a tics is a b ran ch of

logic. T h e m a th e m a tic a l notion s are to b e d efin ed in term s of th e lo gical
notions. T h e theorem s of m a th e m a tics are to b e p ro v e d as theorem s of logic.
L eib n iz ( 1666 ) first co n ceiv ed of lo gic as a science co n ta in in g th e id eas
an d principles u n d erly in g all oth er sciences. D e d e k in d (1888) an d F re g e
(1884, 1893, 1903) w ere en g aged in d efin in g m a th e m a tic a l notion s in
term s of lo gica l ones, an d P ea n o (1889, 1894-1908) in exp ressin g m a th e
m atical theorem s in a lo gical sym bolism .
44 A CRITIQUE OF MATHEMATICAL REASONING CH. I ll
To illustrate how mathematical notions can be defined from logical
ones, let us presuppose the Frege-Russell definition of cardinal number
(§ 3), and the definitions of the cardinal number 0 and of the cardinal
number n + 1 for any cardinal number n (§ 4). Then a finite cardinal
(or natural number) can be defined as a cardinal number which possesses
every property P such that ( 1) 0 has the property P and (2) n + 1 has
the property P whenever n has the property P. In brief, a natural number
is defined as a cardinal number for which mathematical induction holds.
The viewpoint here is very different from that of §§ 6 and 7, where we
presupposed an intuitive conception of the natural number sequence,
and elicited from it the principle that, whenever a particular property
P of natural numbers is given such that ( 1) and (2), then any given natural
number must have the property P. Here instead we presuppose the totality
of all properties of cardinal numbers as existing in logic, prior to the def
inition of the natural number sequence. Note that this definition is im-
predicative, because the property of being a natural number, which it
defines, belongs to the totality of properties of cardinal numbers, which
is presupposed in the definition.
To adapt the logicistic construction of mathematics to the situation
arising from the discovery of the paradoxes, Russell excluded impred
icative definitions by his ramified theory of types (1908, 1910). Roughly,
this is as follows. The primary objects or individuals (i.e. the given things
not being subjected to logical analysis) are assigned to one type (say
type 0), properties of individuals to type 1, properties of properties of
individuals to type 2, etc.; and no properties are admitted which do not
fall into one of these logical types (e.g. this puts the properties ‘pred
icate’ and ‘impredicable’ of § 11 outside the pale of logic). A more
detailed account would describe the admitted types for other objects,
such as relations and classes. Then, to exclude impredicative definitions
within a type, the types above type 0 are further separated into orders.
Thus for type 1, properties defined without mentioning any totality
belong to order 0, and properties defined using the totality of properties
of a given order belong to the next higher order. (The logicistic definition
of natural number now becomes predicative, when the P in it is specified
to range only over properties of a given order, in which case the property
of being a natural number is of the next higher order.) But this separation
into orders makes it impossible to construct the familiar analysis, which
as we saw above contains impredicative definitions. To escape this
outcome, Russell postulated his axiom of reducibility, which asserts that to
any property belonging to an order above the lowest, there is a coextensive
§ 12 FIRST INFERENCES FROM THE PARADOXES 45
p ro p e rty (i.e. one possessed b y e x a c tly th e sam e o b jects) o f order 0. I f o n ly
d efin able properties are considered to e x ist, th e n th e a x io m m ean s t h a t
to e v e r y im p re d ica tiv e d efin ition w ith in a g iv e n t y p e th ere is an e q u iv a
len t p re d ic a tiv e one.
T h e d ed u ctio n of m a th e m a tic s as a p ro v in ce o f lo gic w as carried o u t on
this basis, usin g a lo g ica l sym b o lism , in the m o n u m en tal “ P rin cip ia m a th e
matical of W h ite h e a d an d R u ssell (three vo lu m es, 19 10 -13 ). T h is w o rk
h as h a d a gre a t influence on sub seq u en t d e v e lo p m en ts in sy m b o lic logic.
T h is d ed u ctio n of m a th e m a tics from lo gic w as offered as in tu itiv e
ax io m a tics. T h e axio m s w ere in ten d ed to b e b elieved , or a t lea st to b e
a cce p te d as p lau sib le h yp o th e se s con cern in g th e w orld.
T h e d iffic u lty is n o w : on w h a t gro u n ds sh all w e b e lie v e in th e a x io m
of r e d u c ib ility ? I f properties are to b e co n stru cted , th e m a tte r sh o u ld b e
se ttle d on th e basis o f co n stru ctio n s, n o t b y an axio m . A s th e au th o rs
a d m itte d in th e in tro d u c tio n to th eir second ed itio n ( 1925 ), “ T h is a x io m
h as a p u re ly p ra g m a tic ju stific a tio n : it lead s to th e desired results, a n d
to no others [so far as is k n o w n ]. B u t c le a r ly it is n o t th e sort of a x io m
w ith w h ich w e ca n rest c o n te n t/ '
R a m s e y 1926 fo u n d t h a t th e desired results an d no oth ers ca n a p
p a re n tly b e o b ta in e d w ith o u t th e h iera rch y o f orders (i.e. w ith a sim ple
theory of types). H e classified th e k n o w n an tin o m ies in to tw o sorts, n o w
called lo g ic a l' (e.g. th e B u ra li-F o rti, C a n to r an d R ussell) an d 'ep istem o
lo gica l' or 'se m a n tica l' (e.g. th e R ich a rd an d E p im e n id e s ); an d he o b served
th a t th e lo g ica l an tin o m ies are (apparently) sto p p e d b y th e sim p le
h iera rch y of ty p e s , an d th e sem an tical ones are (apparen tly) p re v e n te d
from arising w ith in th e sy m b o lic la n g u a g e b y th e absen ce th erein o f th e
re q u isite m ean s for referring to expressions of th e sam e lan gu age . B u t
R a m s e y 's argu m e n ts to ju s t if y im p re d ica tiv e d efin ition s w ith in a t y p e
en tail a co n cep tio n of th e t o t a lit y of p red icates of th e t y p e as e x istin g
in d e p e n d e n tly of their c o n str u c tib ility or d e fin a b ility . T h is h as b een
called “ th e o lo g ic a l'/ T h u s n eith er W h ite h e a d an d R u ssell nor R a m s e y
succeeded in a tta in in g th e lo g icistic g o a l c o n str u c tiv e ly . (A n in terestin g
proposal for ju s tify in g im p re d ic a tiv e d efin ition s w ith in a ty p e , b y L a n g
ford 19 2 7 an d C arn ap 19 3 1-2 , is also n o t free o f d ifficu lties.)
W eyl 1946 sa y s th a t, in th e sy ste m o f “ P rin cip ia m a th e m a tic a " ,
“ m a th e m a tic s is no longer fo u n d ed on logic, b u t on a sort of lo gicia n 's
paradise a n d he o bserves t h a t one w h o is r e a d y to b e lie v e in th is
“ tra n scen d en ta l w o rld '' co u ld also a c c e p t th e sy ste m of a x io m a tic set
th e o ry (Zerm elo, F raen k el, etc.), w h ich , for th e d ed u ctio n of m a th e m a tics,
h as th e a d v a n ta g e of b ein g sim pler in structure.
L o gicism tre a ts th e ex isten ce of th e n a tu ral num ber series as an

h yp o th e sis a b o u t th e a c tu a l w orld (‘ax io m of in fin ity ’). A q u ite differen t
h an d lin g of th e problem of in fin ity is proposed b y th e in tu itio n ists (§ 13 )
an d th e form alists (§ 14).
F ro m b o th th e in tu itio n istic an d th e form alistic stan d p o in ts, th e
(abstract) n a tu ral n u m ber sequence is m ore e lem en ta ry th a n th e notion s
o f card in al n u m ber an d of all properties of card in al num bers, w h ich are
used in th e lo gicistic ch aracterizatio n of it.
T h e lo g icistic thesis ca n b e question ed fin a lly on th e gro und th a t lo gic
a lre a d y presupposes m a th e m a tic a l ideas in its form ulatio n. In th e in
tu itio n istic v ie w , an essential m a th e m a tic a l kernel is co n ta in ed in th e
id e a o f itera tio n , w h ich m u st b e used e.g. in d escribing th e h iera rch y of
ty p e s or th e n otion of a d ed u ctio n from g iv e n prem ises.
R e c e n t w o rk in th e lo g icistic school is th a t of Q u in e 194 0*. A critical
b u t s y m p a th e tic discussion of th e lo gicistic order of ideas is g iv e n b y G o d el
1944. In tro d u c to ry trea tm en ts are p ro vid ed b y R u ssell 19 19 an d B la c k 1933.
§ 13 . I n t u i t i o n i s m . In th e 1880 ’s, w hen th e m eth o d s o f W eierstrass,

D e d e k in d an d C an to r w ere flourishing, K ro n eck er argu ed v ig o ro u sly
t h a t th eir fu n d a m e n ta l defin ition s w ere o n ly w ords, since t h e y do n o t
en able one in gen eral to decid e w h eth er a g iv e n o b je c t satisfies th e
definition.
Poin care, w hen he defends m a th e m a tic a l in d u ctio n as an irreducible
to o l of in tu itiv e m a th e m a tic a l reasoning (1902, 1905-6), is also a fore
runner of th e m odern in tu itio n istic school.
In 1908 B rouw er, in a pap er e n title d “ T h e un tru stw orth in ess of th e
prin ciples of lo g ic ” , ch allen ged th e b elief th a t th e rules of th e classical
lo gic, w h ich h a v e com e d ow n to us essen tially from A risto tle ( 384— 322
B .C .), h a v e an ab so lu te v a lid ity , in d epen den t of th e su b je ct m a tte r to
w h ic h th e y are applied. Q u o tin g from W e y l 1946, “ A cc o rd in g to his
v ie w an d readin g of h isto ry, classical logic w as a b stra c te d from th e
m a th e m a tics of fin ite sets an d th eir subsets. . . . F o r g e tfu l of th is lim ited
origin, one afterw ards m isto o k th a t lo gic for so m eth in g a b o v e an d prior
to all m a th em a tics, an d fin a lly ap p lied it, w ith o u t ju stifica tio n , to th e
m a th e m a tic s of in fin ite se ts.”
T w o o b vio u s exam p les w ill illu stra te th a t prin ciples v a lid in th in k in g
a b o u t fin ite sets do n o t necessarily c a rry o ver to in fin ite sets. O n e is th e
p rin ciple th a t th e w hole is greater th a n a n y proper p a rt, w hen ap p lied
to 1-1 correspondences b etw ee n sets (§§ 1, 3 , 4 ). A n o th e r is th a t a set of
n a tu ra l num bers co n ta in s a greatest.
§ 13 INTUITIONISM 47
A prin ciple o f classical logic, v a lid in reasoning a b o u t fin ite sets, w h ich
B ro u w er does n o t a c ce p t for in fin ite sets, is th e law of th e exclu d ed m iddle.
for every proposition A , either A or not A,
T h e law , in its general form , sa y s
N o w le t A be th e proposition there exists a member of the set (or domain) D
having the property P . T h e n not A is e q u iv a le n t to every member of D does
not have the property P , or in oth er w ords every member of D has the property
not-P. T h e law , ap p lied to th is A, hence g iv e s either there exists a member
of D having the property P , or every member of D has the property not-P.
F o r definiteness, le t us sp e c ify P to b e a p ro p e r ty such th a t, for a n y
g iv e n m em ber of D, w e ca n determ ine w h e th er t h a t m em ber h as th e
p ro p e rty P or does not.
N o w suppose D is a fin ite set. T h e n w e cou ld exam in e e v e r y m em ber of
D in turn, an d th u s either find a m em ber h a v in g th e p ro p e rty P , or v e r ify
th a t all m em bers h a v e th e p ro p e rty not-P. T h ere m ig h t b e p ra c tic a l d if
ficulties, e.g. w hen D is a v e r y large set h a v in g s a y a m illion m em bers,
or ev en for a sm all D w h en th e d eterm in atio n w h e th er or n o t a g iv e n
m em ber h as th e p ro p e rty P m a y b e tedious. B u t th e p o ssib ility of co m
p le tin g th e search e x ists in principle. I t is th is p o ssib ility w h ich for B ro u w
er m akes th e la w of th e ex clu d e d m id d le a v a lid prin ciple for reasoning
D an d properties P of th e k in d specified.
w ith fin ite sets
D, th e situ a tio n is fu n d a m e n ta lly d ifferen t. I t is no
F o r an in fin ite set
longer possible in prin ciple to search th ro u gh th e entire set D.
M oreover in th is situ a tio n th e la w is n o t sa v e d for B ro u w er b y su b
s titu tin g , for th e im possible search th ro u gh all th e m em bers o f th e
in fin ite set Dt a m a th e m a tic a l solutio n of th e p roblem posed. W e m a y in
som e cases, i.e. for som e sets D an d properties P , succeed in fin d in g a
m em ber of D h a v in g th e p ro p e r ty P ; an d in o th er cases, succeed in sh ow
in g b y m a th e m a tic a l reasonin g th a t e v e r y m em ber of D h as th e p ro p e rty
not-P, e.g. b y d ed u cin g a c o n tra d ictio n from th e assu m p tio n t h a t an
a rb itra ry (i.e. unspecified) m em ber of D has th e p ro p e rty P . (A n e x a m p le
for th e second k in d of solutio n is w hen D is th e set of all th e ordered pairs
(m, n) of p o sitiv e in tegers, an d P is th e p ro p e rty of a p air (m, n) t h a t
m l = 2n 2. T h e result is th e n P y th a g o r a s ' d isco v e ry th a t y/ 2 is irrational.)
B u t w e h a v e no gro u n d for affirm in g th e p o ssib ility of o b ta in in g either
one or th e o th er of th ese kin d s o f solutio ns in e v e r y case.
A n ex a m p le from m odern m a th e m a tic a l h isto ry is afford ed b y F e r m a t's
“ last th e o re m ", w h ich asserts th a t th e eq u a tio n xn + yn = zn has no
solution in p o sitiv e integers x, y, 2, n w ith n > 2. (Forn — 2 , there are
triples of p o s itiv e integers, ca lled P y th a g o r e a n num bers, w h ich sa tisfy ,
e.g. x= 3, y — 4, z = 5 or x= 5, y = 12 , z = 13 .) H ere D is th e set o f
a ll ordered q u adru p les (x, y , z f n) o f p o sitiv e in tegers w ith n > 2, an d

P is th e p ro p e r ty o f a q u a d ru p le (x, y , z, w) t h a t x n + y n — z n. A b o u t 1637
F e r m a t w ro te on th e m argin o f his c o p y of B a c h e t ’s “ D io p h a n tu s ” t h a t
h e h a d d isco v ere d a tr u ly m arvello u s d em o n stratio n o f th is “ th eo rem ”
w h ic h th e m argin w as to o narrow to con tain . D e sp ite an im m ense e x
p en d itu re of effo rt, no one since th en h as su cceeded in p ro v in g or dis
p ro v in g th e alle ged “ th eo rem ” ; an d m oreover w e la c k th e k n o w led ge of
a n y s y s te m a tic m eth o d , th e p u rsu it of w h ich m u st in p rin ciple u ltim a te ly
le a d to a d ete rm in atio n as to its tr u th or fa lsity . (Cf. V a n d iv e r 1946 for
details.)
B ro u w e r’s n o n -a c ce p ta n ce o f th e la w of th e e x clu d e d m id d le for in
fin ite sets D does n o t rest on th e failure of m a th e m a ticia n s th u s far to
h a v e so lv e d th is p a rticu la r problem , or a n y o th er p a rticu la r problem . T o
m e e t his o b jectio n , one w o u ld h a v e to p ro v id e a m e th o d a d e q u a te in
p rin cip le for so lv in g n o t o n ly all th e o u tsta n d in g u n so lve d m a th e m a tic a l
problem s, b u t a n y others th a t m ig h t ever b e proposed in th e fu tu re. H o w
lik e ly it is th a t such a m e th o d w ill b e found , w e le a v e for th e tim e b ein g
to th e reader to sp ecu late. L a te r in th e b o o k w e sh a ll retu rn to th e
q u estio n (§ 60 ).
T h e fam ilia r m a th e m a tics, w ith its m eth o d s a n d logic, as d eve lo p ed
prior to B ro u w e r’s critiq u e or d isregard in g it, w e ca ll classical ; th e
m a th e m a tics, m eth o d s or lo gic w h ich B ro u w er an d his sch ool allow , w e
c a ll intuition istic. T h e classical includ es p a rts w h ich are in tu itio n istic an d
p a rts w h ich are n o n -in tu itio n istic.
T h e n o n -in tu itio n istic m a th e m a tic s w h ich cu lm in a te d in th e theories
o f W eierstrass, D e d e k in d an d C an to r, an d th e in tu itio n istic m a th e m a tic s
o f B ro u w er, differ e sse n tia lly in th eir v ie w of th e in fin ite. In th e form er,
th e in fin ite is tre a te d as actual or completed or extended or existential. An
in fin ite set is regard ed as e x istin g as a c o m p le ted t o ta lit y , prior to or in
d e p e n d e n tly o f a n y h u m a n process of gen eratio n or co n stru ctio n , a n d as
th o u g h it co u ld b e spread o u t c o m p le te ly for our inspection. In th e la tte r ,
th e in fin ite is tre a te d o n ly as potential or becoming or constructive. The
reco gn itio n of th is d istin ctio n , in th e case of in fin ite m agn itu d es, goes
b a c k to G au ss, w h o in 1831 w rote, “ I p ro test . . . a g a in st th e use of an
in fin ite m a g n itu d e as so m eth in g co m p leted , w h ich is n ev er perm issible
in m a th e m a tic s .” (W erke V I I I p. 216 .)
A c c o rd in g to W e y l 1946, “ B ro u w er m ad e it clear, as I th in k b e y o n d
a n y d o u b t, t h a t there is no e v id e n ce su p p o rtin g th e b elief in th e e x is te n tia l
ch a ra cte r of th e t o t a l it y of all n a tu ra l num bers ___T h e sequence of
num bers w h ich grow s b e y o n d a n y sta g e a lr e a d y reach ed b y p assin g to th e
§13 IN T U m 0 N IS M 49
n e x t num ber, is a m an ifo ld of possibilities open to w a rd s in f in it y ; it
rem ains forever in th e sta tu s of creation, b u t is n o t a closed realm of
th in gs e x istin g in th em selves. T h a t w e b lin d ly co n v e rte d one in to th e
oth er is th e tru e source o f our d ifficu lties, in clu d in g th e an tin om ies — a
source of m ore fu n d a m e n ta l n atu re th a n R u sse ll’s v ic io u s circle prin ciple
in d ic a te d . B ro u w er opened our eyes an d m ad e us see h o w far classical
m a th em a tics, nourished b y a b elief in th e 'a b so lu te ' th a t tran scen d s all
h u m a n p o ssibilities of realization , goes b e y o n d such sta te m e n ts as can
claim real m ean in g an d tru th fou n d ed on e v id e n c e .”
B ro u w er’s criticism of th e classical lo gic as ap p lied to an in fin ite set D
(say th e set of th e n a tu ra l num bers) arises from th is sta n d p o in t resp ectin g
in fin ity . W e see th is cle a r ly b y considering th e m ean in gs w h ich th e in -
tu itio n ist a tta c h e s to va rio u s form s of statem en ts.
all natural numbers n have the property P , or
A g e n e ra lity sta te m e n t
for all nyP(n), is u n d erstoo d b y th e in tu itio n ist as an h y p o th e tic a l
b rie fly
assertion to th e e ffect th a t, if a n y p articu lar n a tu ral n u m ber n w ere g iv e n
to us, w e co u ld b e sure t h a t t h a t n um ber n has th e p ro p e rty P. T h is is a
m ean in g w h ich does n o t require us to ta k e in to v ie w th e classical co m
p le ted in fin ity of th e n a tu ra l num bers.
M a th e m a tic a l in d u ctio n is an ex a m p le of an in tu itio n istic m e th o d for
p ro v in g g e n e ra lity propositions a b o u t th e n a tu ra l num bers. A p roof b y
in d u ctio n of th e p rop osition for all n, P(n) show s t h a t a n y g iv e n n w o u ld
h a v e to h a v e th e p ro p e r ty P , b y reasonin g w h ich uses o n ly th e num bers
from 0 u p to n (§ 7 ). O f course, for a p a rticu la r proof b y in d u ctio n to b e
in tu itio n istic, also th e reasonings used w ith in its basis an d in d u ctio n step
m u st b e in tu itio n istic.
there exists a natural number n having the property
A n ex isten ce sta te m e n t
P , or b rie fly there exists an n such that P(n), h as its in tu itio n istic m ean in g
as a p a rtia l co m m u n ica tio n (or a b stract) of a sta te m e n t g iv in g a p a r
ticu lar e x a m p le of a n a tu ra l n u m ber n w h ich has th e p ro p e r ty P , or a t
le a st g iv in g a m e th o d b y w h ich in prin ciple one co u ld fin d su ch an
exam p le.
Th erefo re an in tu itio n istic proof of th e p roposition there exists an n
such that P(n) m u st b e constructive in th e fo llo w in g (strict) sense. T h e
proof a c tu a lly e x h ib its ail Exam ple of an n such th a t P(n), or a t least in
d icates a m e th o d b y w h ich one co u ld in prin ciple fin d such an exam p le .
In classical m a th e m a tic s there occur non-constructive or indirect
existen ce proofs, w h ich th e in tu itio n ists do n o t accep t. F o r exam p le , to
p ro ve there exists an n such that P(n), th e classical m a th e m a tic ia n m a y
deduce a c o n tra d ictio n from th e assu m p tio n for all n, not P(n). U nder
b o th th e classical an d th e in tu itio n istic logic, b y re d u ctio ad ab su rd u m

th is g iv e s not for all nt not P(w). T h e classical lo gic allow s th is result to
b e tran sfo rm ed in to there exists an n such that P (n ), b u t n o t (in general)
th e in tu itio n istic. S u c h a classical ex isten ce proof le a v e s us no nearer
th a n before th e proof w a s g iv e n to h a v in g an e x a m p le of a n u m ber n such
th a t P(n) (though som etim es w e m a y afterw ard s b e able to d isco ver one
b y an o th er m eth od). T h e in tu itio n ist refrains from a c c e p tin g such an
existen ce proof, b eca u se its con clusion there exists an n such that P(n)
ca n h a v e no m ean in g for h im oth er th a n as a reference to an e x a m p le of
a n u m b er n such th a t P(n), an d th is e x a m p le has n o t been produced. T h e
classical m ean in g, th a t som ew here in th e co m p le ted in fin ite t o t a lit y of
th e n a tu ra l num bers there occurs an n such th a t P(n), is n o t a v a ila b le to
h im , sin ce he does n o t co n ceiv e th e n a tu ral n u m b e is as a co m p le ted
t o ta lity .
A s an o th er ex a m p le of a n o n -co n stru ctive existen ce proof, suppose it
has been show n for a ce rta in P , by in tu itio n istic m eth od s, th a t if F e r m a t's
“ la st th e o re m " is true, th e n th e n um ber 5013 has th e p ro p e rty P , an d also
th a t if F e r m a t ’s “ last th eo rem " is false, th en 10 has th e p ro p e rty P .
C la s sic a lly th is suffices to d em o n strate th e e x isten ce of a n um ber n such
th a t P(n). B u t w ith th e problem of th e “ last th e o re m " un so lved, B ro u w er
w o u ld d isallow such an ex isten ce proof, b ecau se no ex a m p le h as been
giv e n . W e do not k n o w t h a t 5013 is an ex am p le, nor do w e kn o w th a t
10 is an ex am p le, nor do w e k n o w a n y procedure w h ich w o u ld in principle
(i.e. a p a rt from p ra c tic a l lim ita tio n s on th e le n g th of procedures w e can
ca rry out) lead us to a p articu lar n u m ber w h ich w e cou ld b e sure is an
exam p le. B ro u w er w o u ld m erely a c ce p t w h a t h as been g iv e n as p ro v in g
th e im p lica tio n (or co n d itio n al statem e n t) if F or not F , then there exists an
n such that P(n), w here F is th e sta te m e n t for all x,y, z > 0 and n > 2,
xn + yn ¥= £n- T h e classical m a th e m a ticia n , b y his la w of th e exclu d ed
m idd le, has th e prem ise F or not F of th is im p licatio n , an d so he can infer
its con clusion there exists an n such that P(n). B u t in th e present s ta te of
k n o w led ge, B ro u w er does n o t a c c e p t th e prem ise F or not F as know n.
As ap pears in th is ex am p le , in tu itio n istic m eth o d s are to b e d is
tin gu ish ed from n o n -in tu itio n istic ones in th e case of d efin ition s as w ell
as in th e case o f proofs. In th e present sta te of our kn o w led ge, B ro u w er
does not a c c e p t the number n which is equal to 5013 if Fy and equal to 10
if not F a 3 a v a lid d efin ition of a n a tu ra l n um ber n.
A d isju n ctio n A or B co n stitu te s for th e in tu itio n ist an in co m p lete
co m m u n ica tio n of a sta te m e n t tellin g us t h a t A h old s or th a t B holds,
or a t least g iv in g a m eth o d b y w h ich w e can choose from A a n d B one
§13 INTUITIONISM 51
w h ich holds. A co n ju n ctio n A and B m eans th a t b o th A an d B hold. A n
im p lica tio n A im plies B (or if A , then B) expresses th a t B follow s from A
by in tu itio n istic reasoning, or m ore e x p lic itly th a t one possesses a
m eth o d w h ich , from a n y proof of A, w o u ld procure a proof of B ; an d a
n egatio n not A (or A is absurd) th a t a co n trad ictio n B and not B follow s
from A b y in tu itio n istic reasoning, or m ore e x p lic itly th a t one possesses
a m eth o d w h ich , from a n y proof of A , w o u ld procure a proof o f a co n tra
B and not B (or of a sta te m e n t a lre a d y kn o w n to b e absurd, such
d ictio n
as 1 = 0 ). A d d itio n a l com m en ts on th ese in tu itio n istic m ean in gs w ill be
given in § 82 . See N o te 1 on p. 65 .
Q u o tin g from H e y tin g 1934, “ A cc o rd in g to B rouw er, m a th e m a tics
is id e n tica l w ith th e e x a c t p a rt of our th in k in g. . . . no science, in p ar
ticu lar n o t p h ilo so p h y or logic, ca n be a p resupposition for m a th em a tics.
I t w o u ld b e circular to a p p ly a n y p hilosoph ical or lo gica l prin ciples as
m eans o f proof, since m a th e m a tic a l con cep tion s are a lre a d y presupposed
in th e fo rm u latio n of such p rin cip le s.” T h ere rem ains for m a th e m a tics
“ no o th er source th a n an in tu itio n , w h ich p laces its co n cep ts an d in
ferences before our eyes as im m e d ia te ly cle a r.” T h is in tu itio n “ is n o th in g
oth er th a n th e fa c u lty of con siderin g se p a r a te ly p articu lar co n cep ts an d
inferences w h ich occur re g u la rly in o rd in ary th in k in g .” T h e id ea of th e
n a tu ral n u m ber series can b e a n a ly z e d as restin g on th e p o ssib ility , first
of con siderin g an o b je c t or experien ce as g iv e n to us se p a ra te ly from th e
rest of th e w orld, second of d istin gu ish in g one such from an oth er, an d
th ird of im agin in g an u n lim ited rep etitio n of th e second process. “ In th e
in tu itio n istic m a th e m a tic s, one does n o t d raw inferences acco rd in g to
fixed norm s, w h ich ca n b e co lle cted in a logic, b u t each single inference is
im m e d ia te ly te s te d on its e v id e n c e .” But also “ T h ere are gen eral
rules, b y w h ich from g iv e n m a th e m a tic a l theorem s n ew theorem s ca n b e
form ed in an in tu itiv e ly clear w a y ; th e th e o r y of these con nection s ca n
be tr e a te d in a ‘m a th e m a tic a l lo g ic ’ , w h ich is th e n a b ran ch of m a th e
m atics an d is n o t sen sib ly ap p lied outsid e of m a th e m a tic s.”
W e tu r n n o w to th e q u estio n : H o w large a p a rt do th e n o n -in tu i-
tio n istic m eth o d s p la y in th e classical m a th e m a tics?
T h e fa c t th a t n o n -in tu itio n istic m eth o d s occur in classical e lem en ta ry
num ber th e o ry is sign ifican t, since it enables e lem en ta ry num ber th e o r y
to serve as th e first an d sim p lest te stin g gro u n d in research on fo u n d a
tions gro w in g o u t of th e in tu itio n istic an d fo rm alistic th in k in g. We
shall b e alm o st w h o lly concerned w ith e le m en ta ry n u m ber th e o r y in
th is b ook.
A c t u a lly , in th e e x istin g b o d y of elem en tary n um ber th eo ry, th e
n o n -in tu itio n istic m eth o d s do n o t p la y a large p art. M ost n o n -co n stru ctive
existen ce proofs ca n b e replaced b y c o n stru c tiv e ones.
O n th e oth er h an d , in an a ly sis (and still m ore tran scen d en ta l b ran ch es
of m ath em atics) th e n o n -in tu itio n istic m eth o d s of d efin itio n a n d proof
p erm ea te th e w hole m e th o d o lo gy . T h e real num bers in th e D e d e k in d c u t
represen tation are in fin ite sets of ratio n als (§ 9 ). T h u s to tre a t th e m as
o b je c ts in th e u su al w a y , w e are a lre a d y u sin g th e co m p le ted infin ite.
In p articu lar, w e do a p p ly th e la w of th e e x c lu d e d m iddle to th ese sets,
in co n n ectio n w ith th e sim plest d efin ition s of th e su b je ct. F o r exam p le,
to sh ow th a t for a n y tw o real num bers x an d y , eith er x < y or x = y
or x > y , w e use it tw ice, th u s: E ith e r there e x ists a r a tio n a l r in y
w h ich does n o t b elo n g to x , or all ration als in y b elo n g to x ; an d sim ilarly
in te rch a n gin g x an d y . In th e im p re d ic a tiv e d efin itio n o f l.u .b . M (§ 9
(A ) , § 12), w e use th e t o t a lit y of th e real num bers in th e sam e w a y .
A n o th e r in stan ce of n o n -co n stru ctive reasoning occurs in th e p roof o f
(B) § 9 , w here w e assum ed th e righ t to choose an elem en t a n from a set
Mn, sim u lta n eo u sly for in fin ite ly m a n y va lu e s of n, w ith o u t g iv in g a n y
p ro p e rty to determ ine w h ic h elem ent is chosen. (Th is is a case of th e
‘ax io m of ch o ice', first n o tice d as an assu m p tion b y Zerm elo 1904. W e
u sed it also for T h eo rem B § 4 .)
A lth o u g h th e c o m p le ted in fin ite h as been b a n n e d for m a g n itu d e s (as
G au ss ca lled u p o n us to do), it reappears in fu ll force for collections.
A s H ilb e rt an d B e m a y s d escribe th e situ a tio n in th eir “ G ru n d la g en der
M a th e m a tik (F o u n d atio n s of m a th e m a tic s )" , v o l. 1 (1934), p. 41 , “ The
. . . a rith m e tiz a tio n o f an a ly sis is n o t w ith o u t a residue le ft over, as certain
sy s te m a tic fu n d a m e n ta l co n cep tio n s are in tro d u ced w h ich d o n o t b elo n g
to th e d o m a in of in tu itiv e a rith m e tic a l th in k in g . T h e in sigh t w h ic h has
g iv e n us th e rigorous fo u n d a tio n of an alysis consists in t h i s : t h a t th ese few
fu n d a m e n ta l assu m ption s do su ffice for b u ild in g u p th e th e o r y o f m a g n i
tu d e s as a th e o r y of sets of in te g e rs."
T h e n e x t qu estion i s : W h a t k in d o f a m a th e m a tics ca n b e b u ilt w ith in
th e in tu itio n istic restriction s ? I f th e e x istin g classical m a th e m a tic s co u ld
b e re b u ilt w ith in th e in tu itio n istic restrictions, w ith o u t to o g re a t increase
in th e lab o r required an d to o g re at sacrifices in th e results ach iev ed , th e
p roblem of its fo u n d atio n s w o u ld ap p ear to b e solved .
T h e in tu itio n ists h a v e created a w h ole new m a th e m a tics, in clu d in g
a th e o r y o f 't h e co n tin u u m an d a set th e o r y (cf. H e y t in g 1934). T h is
m a th e m a tic s em p lo y s co n cep ts an d m akes d istin ctio n s n o t fo u n d in th e
classical m a th e m a tic s; an d it is v e r y a ttr a c tiv e on its ow n acco u n t. A s
a su b stitu te for classical m a th e m a tic s it h as tu rn ed o u t to b e less pow erful,
§14 FORMALISM 53
an d in m a n y w a y s m ore co m p licated to develop. F o r exam p le, in B ro u w er's
th e o ry of th e con tin u u m , w e ca n n o t affirm th a t a n y tw o real num bers
a and b are either equ al or unequal. O u r kn o w led ge a b o u t th e e q u a lity or
in e q u a lity ofa an d b ca n b e m ore or less specific. B y a b, it is m ean t
th a t a = b leads to a co n trad ictio n , w hile a # b is a stronger k in d of
in e q u a lity w h ich m eans th a t one can g iv e an ex am p le of a ra tio n al num ber
w h ich separates a an d b. O f course a # b im plies a ^ b. B u t there are
pairs of real num bers a an d b for w h ich it is not kn ow n th a t either a = b
or a ■=£ b (or a # b). I t is clear th a t such co m p licatio n s replace th e cla s
sical th e o ry of th e co n tin u u m b y som eth in g m u ch less perspicuous in
form.
D e sp ite th is, th e p o ssib ility of an in tu itio n istic recon stru ctio n of
classical m a th e m a tics in a d iffe ren t w a y in v o lv in g rein terp retation (re
c e n tly undertaken) is not to b e ruled o u t (cf. § 81 ).
§ 14. Formalism. B ro u w er has revealed w h a t th e g en etic or co n

stru ctiv e te n d e n c y in v o lv e s in its u ltim a te r e fin e m e n t; H ilb e rt does
th e sam e for th e a x io m a tic or e x iste n tia l (§ 8). T h e a x io m a tic m eth o d
h ad a lre a d y been sharpened from th e m ate rial a x io m a tic s of E u c lid to
th e form al a x io m a tics of H ilb e r t’s “ G ru n d lagen der G eo m etrie” (1899).
Form alism is th e result of a fu rth er step, to m eet th e crisis cau sed b y th e
parad o xes an d th e ch allen ge to classical m a th e m a tics b y B ro u w er an d
W e y l. T h is step w as forecasted b y H ilb e rt in 1904, an d seriously under
ta k e n b y h im an d his co llabo rato rs B ern a ys, A ck erm an n , v o n N eu m a n n
and others since 1920 (cf. B e rn a y s 193 5a, W e y l 1944).
H ilb e rt con ced ed th a t th e propositions of classical m a th e m a tics w h ich
in v o lv e th e c o m p le ted in fin ite go b e y o n d in tu itiv e eviden ce. B u t he re
fused to fo llo w B ro u w er in g iv in g u p classical m a th e m a tics on th is acco u n t.
T o sa lv a g e classical m a th e m a tics in th e face o f th e in tu itio n istic c r iti
cism , he proposed a p rogram w h ic h w e can sta te prelim in arily as fo llo w s :
Classical m a th e m a tics sh all b e fo rm u lated as a form al a x io m a tic th e o ry ,
an d th is th e o r y sh all b e p ro v ed to b e con sistent, i.e. free from co n tra
diction.
Prior to th is proposal of H ilb ert's, th e m eth o d used in co n sisten cy proofs
for a x io m a tic theories, esp ec ia lly in H ilb e rt's earlier a x io m a tic th in k in g,
w as to g iv e a 'm od el'. A model for an a x io m a tic th e o ry is sim p ly a sy ste m
of o b jects, chosen from som e other th e o ry an d sa tisfy in g th e axio m s
(§ 8). T h a t is, to each o b je c t or p rim itive notion of the a x io m a tic th eo ry,
an o b je ct or n o tio n of th e oth er th e o ry is correlated, in such a w a y th a t
th e axio m s beco m e (or correspond to) theorem s of th e oth er th e o ry.
I f th is oth er th e o r y is con sisten t, th e n th e a x io m a tic th e o r y m u st be.

F o r suppose th a t, in th e a x io m a tic th e o ry, a c o n tra d ictio n w ere d ed u cib le
from th e axiom s. T h e n , in th e oth er th eo ry, by correspon ding inferences
a b o u t th e o b je c ts c o n s titu tin g th e m odel, a co n trad ictio n w o u ld b e de
d u cib le from th e correspon ding theorem s. 4
In a fam ou s e a rly ex am p le , B e ltra m i ( 1868 ) sh ow ed th a t th e lines in
th e p lan e n o n -E u clid e a n g e o m e try o f L o b a tc h e v s k y an d B o ly a i (the p lan e
h y p e rb o lic geo m etry) ca n b e represen ted b y th e geodesics on a surface of
co n sta n t n e g a tiv e c u rv a tu re in E u c lid e a n space. T h u s th e p la n e h y p e r
b o lic g e o m e try is con sisten t, if th e E u c lid e a n g e o m e try is con sisten t.
(A n o th er m odel for th e sam e w as g iv e n b y K le in ( 1871 ) in term s of p lan e
p ro je c tiv e g e o m e try w ithCayley's m etric ( 1859 ); th is ca n b e co n stru ed
as a model in th e E u c lid e a n plane. Cf. Young i g n Lectures II an d I I I.)
T h e a n a ly tic g e o m e try o f D esca rtes ( 1619 ), i.e. th e use of coord in ates
to represent g eo m etrical o b jects, co n stitu tes a gen eral m e th o d for
estab lish in g th e c o n sisten cy of geo m etric theories on th e b asis of an alysis,
i.e. th e th e o r y of th e real num bers.
C o n sisten c y proofs b y th e m e th o d of a m o del are re lative . T h e th e o r y
for w h ic h a m odel is set u p is con sisten t, if t h a t from w h ich th e m od el is
ta k e n is con sistent.
O n ly w h en th e la tte r is u n im p each a b le does th e m odel g iv e us an
ab so lu te proof of co n sisten cy. V e b le n an d B u s s e y 1906 ac h ie v e a b so lu te
proofs of c o n sisten cy for certa in ru d im e n ta ry p r o je c tiv e geom etries b y
se ttin g u p m odels u sin g o n ly a fin ite (sic!) class of o b je cts to represent
th e p o in ts (cf. Y o u n g 1 9 1 1 L ectu res I V an d V ).
F o r p ro v in g a b so lu te ly th e co n sisten cy of classical n u m ber th e o ry,
o f an alysis, a n d of set th e o r y (su ita b ly ax io m atized ), th e m eth o d of a
m o d el offers no hope. N o m a th e m a tic a l source is ap p aren t for a m o d el
w h ich w o u ld n o t m e rely ta k e us b a c k to one of th e theories p re v io u sly
redu ced b y th e m e th o d o f a m od el to these.
T h e im p o ssib ility of d raw in g u p o n th e p ercep tu a l or p h y sic a l w orld
for a m o del is argu ed in H ilb e rt an d B e rn a y s 1934 pp. 15— 17 . They
illu stra te it b y con sid erin g Z en o 's first p a ra d o x (fifth c e n tu r y B .C .),
acco rd in g to w h ich a runner ca n n o t run a course in a fin ite tim e. F o r
before he can do so, he m u st run th e first h alf, th e n th e n e x t q uarter,
th e n th e n e x t eigh t, an d so on. B u t th is w o u ld require h im to co m p lete
an in fin ite n u m b er of acts. T h e usual solutio n of th e p a ra d o x consists
in o b servin g th a t th e in fin ite series of th e tim e in terva ls required to run
th e su ccessive segm en ts con verges. “ A c t u a lly there is also a m u ch m ore
ra d ic a l solu tio n o f th e p ara d o x. T h is con sists in th e con sid eration th a t
§14 FORMALISM 55
w e are b y no m eans o b liged to b elieve th a t th e m a th e m a tic a l sp a c e -tim e
represen tation of m o tio n is p h y s ic a lly sign ifican t for a r b itra rily sm a ll
space an d tim e in te r v a ls ; b u t ra th er h a v e e v e r y basis to suppose t h a t th a t
m a th e m a tic a l m odel ex tra p o la te s th e fa cts of a certa in realm of experience,
n a m ely th e m otion s w ith in th e orders of m a g n itu d e h ith erto accessible to
our o b servatio n , in th e sense of a sim ple co n cep t co n stru ctio n , sim ilarly
to th e w a y th e m ech an ics of co n tin u a co m p letes an e x tra p o la tio n in w h ich
a con tin u ou s fillin g of th e space w ith m a tte r is assum ed . . . . T h e situ a tio n
is sim ilar in all cases w here one b elieves it possible to e x h ib it d ir e c tly
an [actual] in fin ity as g iv e n th ro u gh experien ce or p ercep tio n ....
Closer ex a m in a tio n th en show s t h a t an in fin ity is a c tu a lly n o t g iv e n
to us a t all, b u t is first in terp o la ted or e x tra p o la te d th ro u gh an in te l
lectu al p ro c ess/'
Therefore, if c o n sisten cy is to b e p ro v ed for n u m ber th e o r y (includ ing
its n o n -in tu itio n istic portions), for an alysis, etc., it m u st be b y an oth er
m ethod. I t is H ilb e r t’s co n trib u tio n n ow to h a v e c o n ceiv ed a new d irect
approach , an d to h a v e recogn ized w h a t it in v o lv e s for th e a x io m a tiz a tio n .
T h is direct m e th o d is im p lic it in th e m ean in g of c o n sisten cy (at least as
w e n ow th in k of it), n a m e ly t h a t no lo gical c o n tra d ictio n (a p ro p o sitio n
A an d its n eg atio n not A b o th b e in g theorem s) ca n arise in th e th e o r y
ded u ced from th e axiom s. T h u s to p ro v e th e co n siste n cy of a th e o r y
d ire c tly , one sh ould p ro v e a proposition a b o u t th e th e o r y itself, i.e.
sp ecifica lly a b o u t all possible proofs of theorem s in th e th eo ry. The
m a th e m a tic a l th e o r y w hose c o n siste n c y it is h o ped to p ro ve th en b e
com es itself th e o b je c t of a m a th e m a tic a l s tu d y , w h ich H ilb e rt calls
" m e ta m a th e m a tic s ’ ’ or "p ro o f th e o r y ’ ’ . H o w th is is possible, an d w h a t
th e m eth o d s of th e s tu d y m a y be, w e sh all ex am in e in th e n e x t section.
M ean w hile le t us consider fu rth er th e im p o rt of H ilb e r t’s proposal.
H ilb ert (1926, 1928) draw s a d istin ctio n b etw ee n 're a l’ an d 'id e a l’ s ta te
m en ts in classical m a th e m a tic s, in essence as follow s. T h e real statem ents
are those w h ich are b ein g used as h a v in g an in tu itiv e m e a n in g ; the ideal
statements are those w h ich are n o t b ein g so used. T h e sta te m e n ts w hich
correspond to th e tre a tm e n t of th e in fin ite as a c tu a l are ideal. C lassical
m a th e m a tics ad jo in s th e id eal sta te m e n ts to th e real, in order to retain
th e sim ple rules of th e A ris to te lia n lo gic in reasoning a b o u t in fin ite sets.
T h e a d d itio n of 'id ea l elem en ts’ to a sy ste m to c o m p le te its stru ctu re
an d sim p lify th e th e o r y of th e sy ste m is a com m on an d fru itfu l d e v ic e in
m odern m a th em a tics. F o r ex am p le , in E u c lid e a n p lan e g e o m e try tw o
d istin ct lines in tersect in a u n iq u e p o in t, e x c e p t w hen th e lines are parallel.
T o rem o ve this exce p tio n , P o n celet in his p ro je c tiv e g e o m e try ( 1822 )
in tro d u c e d a poin t at in fin ity on each of th e original lines, such th a t p ar

allel lin e s h a v e th e sam e p o in t a t in fin ity an d n o n -p arallel lines h a v e
d ifferen t p o in ts a t in fin ity . T h e t o t a lit y of th ese p o in ts a t in fin ity m a k e
up a lin e at in fin ity. A s a line th ro u gh a fin ite p o in t of th e p r o je c tiv e plane
ro tates, its p o in t a t in fin ity traces o u t th e line a t in fin ity . B y th is d evice,
th e relation sh ips o f in cid en ce b e tw e e n p o in ts an d lines is sim plified. T w o
d istin c t p o in ts d ete rm in e a u n iq u e line (w hich is ‘on' b o th p oin ts, i.e.
th ro u g h b o th of w h ic h th e line p a sse s); a n d tw o d istin c t lines determ ine
a u n iq u e p o in t (w hich is on b o th lines). T h e se tw o propositions are duals
o f each other. T h e re is a gen eral prin ciple, ca lled th e prin ciple of du ality
for p la n e p ro je c tiv e g e o m e try , w h ich s a y s t h a t to each theorem of th e
su b je c t th e s ta te m e n t o b ta in e d from it b y in terch a n gin g th e w ords
*'p o in t" a n d “ lin e ” is also a theorem .
A s o th er e x a m p le s of th e a d d itio n o f elem en ts to a p re v io u sly co n
s titu te d s y s te m o f elem en ts to serve som e th eo retica l purpose, w e m a y
ta k e th e su ccessive en largem en ts o f th e n u m b er sy ste m , sta rtin g s a y
w ith th e n a tu ra l num bers, th e n a d jo in in g th e n e g a tiv e integers, th en th e
fractio n s, th e n th e irration als, a n d fin a lly th e im a g in a r y num bers. T h e
a d ju n c tio n of th e n e g a tiv e in tegers sim plifies th e th e o r y of ad d itio n b y
m a k in g th e in verse o p eratio n (subtraction) a lw a y s po ssib le; etc.
H ilb e r t's p roblem is cr u d e ly an alo go u s to th e problem w h ich existe d
w h e n im a g in a ry n um bers first ca m e in to use. A s t h e y w ere th e n n o t
cle a r ly u n d ersto o d , one m ig h t h a v e proposed to ju s t ify th eir use to
d o u b ters b y p ro v in g th a t, if im agin aries are used acco rd in g to prescribed
rules to d eriv e a result expressed in term s of reals o n ly, th e n t h a t result
m u st b e correct. O f course, th is k in d of ju stific a tio n for im agin aries rel
a tiv e to reals is n o t needed n ow , since th eir in terp retatio n b y p o in ts in
th e p lan e (W essel 1799 ) an d b y pairs of reals (G auss 1831 ) h a v e b ecom e
kn ow n .
T h is a n a lo g y su ggests a s k in g w h eth er, if a proof of co n sisten cy in
H ilb e r t's sense sh ou ld su cceed for a p o rtio n o f classical m a th e m a tic s
co m p risin g b o th real a n d id eal sta tem e n ts, w e co u ld th e n infer th a t th e
real s ta te m e n ts p ro v e d th erein b y an excu rsion th ro u g h th e id eal are tru e
in tu itio n is tic a lly ? T h e e x te n t to w h ich w e co u ld w ill b e d iscussed la te r
(end § 42 , end § 82 ); it w ill d ep en d on w h a t reasonings are co v ere d b y th e
co n siste n c y proof, an d w h a t class o f sta te m e n ts is b ein g ta k e n as real.
T o th is e x te n t, success in H ilb e rt's program w o u ld g iv e to classical m a th e
m a tic s a role as a m e th o d o f proof for th e in tu itio n ists.
A sh arp c o n tro v e rsy arose b e tw e e n B ro u w er an d H ilb e rt in th e ea rly
y ea rs a fter H ilb e r t's p rogram to o k shape. B ro u w er 19 2 3 said, “ A n in correct
§ 14 FORMALISM 57
th e o r y w h ich is n o t sto p p e d b y a co n trad ictio n is none th e less in co rrect,
ju st as a crim inal p o lic y u n ch eck ed b y a reprim an d in g co u rt is none th e
less c rim in a l.” H ilb e rt 1928 reto rted , “ T o ta k e th e la w o f th e e x c lu d e d
m id d le a w a y from th e m a th e m a tic ia n w o u ld b e lik e d e n y in g th e astron o
m er th e telesco p e or th e b o x e r th e use o f his fis ts .”
A c c o rd in g to B ro u w er (1928) an d H e y tin g (19 3 1-2 , 1934), agreem en t
b etw ee n in tu itio n ism an d form alism is possible, p ro v id e d (as in v o n
N eu m a n n 19 3 1-2 ) th e form alist refrains from a ttr ib u tin g to th e n o n -
in tu itio n istic classical m a th e m a tics a m a te ria l m ean in g or co n te n t, in
term s of w h ic h th e co n siste n cy proof ju stifies it. S u c h a ju stific a tio n ,
says B rouw er, “ co n ta in s a v ic io u s circle, b ecau se th is ju stific a tio n d ep en d s
on th e (m aterial (inhaltlichen)) correctness of th e p rop osition th a t from
th e co n siste n cy o f a sta te m e n t th e correctness of th a t sta te m e n t follow s,
i.e. on th e (m aterial) correctness of th e la w of th e ex clu d e d m id d le ” ,
w h ich is p a rt o f th e fo rm alistic m a th e m a tic s th a t is to b e ju stifie d .
T h e d elica te p o in t in th e fo rm alistic p o sitio n is to e x p la in h o w th e n on -
in tu itio n istic classical m a th e m a tics is sign ifican t, a fte r h a v in g in itia lly
agreed w ith th e in tu itio n ists t h a t its theorem s la c k a real m ean in g in
term s of w h ich t h e y are true.
C lassical m a th e m a tic s co n stru cts theories in q u ite a d ifferen t sense
from in tu itio n istic m a th e m a tics. H ilb e rt 1928 say s, “ I t is b y no m ean s
reasonable to set u p in gen eral th e requirem en t th a t each sep arate fo rm u la
should b e in terp reta b le ta k e n b y i t s e l f ___” In th eo retica l p h y sic s “ o n ly
certain co m b in atio n s an d con sequences o f th e p h y sic a l la w s ca n b e
ch eck ed e x p e r im e n ta lly — likew ise in m y proof th e o r y o n ly th e real
sta te m e n ts are im m e d ia te ly c a p a b le of a v e rific a tio n ” .
A th e o r y in classical m a th e m a tic s ca n b e regard ed as a sim ple an d
elegan t s y s te m a tiz in g schem e, b y w h ich a v a r ie t y o f (presum ably) tru e
real sta te m e n ts, p re v io u sly ap p earin g as heterogen eous a n d u n rela ted ,
an d o fte n p re v io u sly u n k n o w n , are com prised as con sequences o f th e
ideal theorem s in th e th e o ry. (Cf. v o n N e u m a n n 19 4 7, E in ste in 194 4
p. 288 .)
T h e e x a m p le of a n a ly tic n u m b er th e o r y illu stra tes th a t th eorem s of
an alysis (lackin g a m ean in g a c c e p ta b le to th e in tu itio n ist) o fte n en ta il
theorem s o f n u m b er th e o ry , w h ic h are m ean in gfu l in tu itio n istic a lly , a n d
for w h ich eith er no n o n -a n a ly tic proofs h a v e been d isco vered or o n ly
m u ch m ore co m p lic a te d ones.
F o r a th e o r y to b e v a lu a b le in th is w a y , th e real sta te m e n ts co m
prised m u st b e true. F o r m e r ly m a th e m a ticia n s su pposed th is to be
gu aran teed b y th e tru th of th e theorem s w h ich w e n ow recogn ize as
id e a l; n o w w e h ope to gu aran tee it in stead b y a c o n sisten cy proof.
B y e a s y stages of tran sitio n , th e th eo rizin g m a y clim b to h igher levels,
from w h ich it is o n ly v e r y in d ire c tly concerned w ith sy s te m a tiz in g th e
real propositions a t th e origin al level, b u t rath er w ith s y ste m a tiz in g ideal
propositions a t in term ed ia te levels. In this co n n ectio n it is of in terest
w h e th er su cce ssiv e ly h igh er th eo retical con stru ction s a c tu a lly ad d to th e
b o d y of real propositions of th e original sort w h ich are com prised, as
w ell as w h e th er th e y do a c tu a lly perm it su b sta n tia l sim p lificatio n s of th e
proofs of those p re v io u sly com prised. (Cf. end § 42 .)
I t is d e b a ta b le h ow h igh a th eo retical stru ctu re is ju stifie d for sy s te m
a tiz in g a g iv e n sort of real tru th s, e.g. w h eth er classical an aly sis is ju s tifie d
as a sy s te m a tiz a tio n of n u m b er-th eo retic tru th s. H isto r ic a lly a n a ly tic
n u m b er th e o r y w as a b y -p r o d u c t, an d th e a c tu a l im p etu s to th e d e
ve lo p m e n t of classical a n a ly sis cam e from th e sciences, in clu d in g
g e o m e try in its p h y sic a l ap p licatio n .
H ilb e rt an d B e rn a y s 1934 em ph asize th a t in th e sciences “ w e h a v e
to do . . . p re d o m in a te ly w ith theories w h ich do n o t reprod u ce th e a c tu a l
sta te of affairs co m p le te ly , b u t represent a sim plifyin g idealization of
th e s ta te of affairs an d h a v e th eir m ean in g th erein ” (pp. 2 — 3 ). A n a ly s is
serves as a “ fo rm atio n of id eas (Ide en b ild u n g)” , in term s of w h ich th ose
theories can b e expressed, or to w h ich th e y ca n b e red u ced b y th e m e th o d
o f m odels. A proof of th e c o n sisten cy of a n aly sis w o u ld assure us of th e
co n siste n cy of th e id ealizatio n s effected in th ose theories (p. 19 ).
W e y l (1926, 1928, 19 3 1) observes th a t in th eo retica l p h y sics it is n o t
the sep arate s ta te m e n ts w h ich are con fron ted w ith experience, b u t th e
th eo retica l s y s te m as a w hole. W h a t is affo rd ed here is n o t a tru e d e
scrip tio n of w h a t is g iv en , b u t th eo retical, p u re ly sy m b o lic co n stru ctio n
of th e w orld, (A lso he argues th a t our th eo retical in terest is n o t e x
c lu s iv e ly or e v e n p rim arily in th e 'real sta te m e n ts’ , e.g. t h a t th is p o in ter
coincides w ith t h a t scale d ivisio n , b u t rath er in th e id eal su p p osition s,
e.g. th e su p p o sitio n of th e electron as a u n iversal electrical q u a n tu m .)
I t is a deep p h ilo so p h ical q u estio n w h a t th e ‘tr u th ’ or o b je c t iv it y is
w h ich p ertain s to th is th eo retica l w orld co n stru ctio n g o in g far b e y o n d
th e g iv e n . T h is is clo sely co n n ected w ith th e question , w h a t m o tiv a te s
us to ta k e as b asis th e p a rticu la r a x io m sy ste m chosen. F o r th is co n
siste n c y is a n ecessary b u t n o t su fficie n t argu m en t. W h e n m a th e m a tics
is ta k e n for itse lf alone, he w o u ld restrict h im self w ith B ro u w er to th e
in tu itiv e tr u th s ; he does n o t fin d a sufficient m o tiv e to go further. B u t
w h en m a th e m a tic s is m erged c o m p le te ly w ith p h y sics in th e process of
th e o r e tic a l w orld co n stru ctio n , he sides w ith H ilbert.
§15 FORMALIZATION OF A THEORY 59
A v e rd ict on th e fo rm alists’ th in k in g w ill d epend p a r tly on th e fru its
of th e program t h e y propose. T h is program calls for a su b je ct ca lled
“ m e ta m a th e m a tic s” , in w h ich t h e y aim in p articu lar to estab lish th e
co n sisten cy of classical m ath em atics.
W e n ote in a d v a n ce th a t m e ta m a th e m a tic s w ill b e fou n d to p ro vid e
a rigorous m a th e m a tic a l tech n iq u e for in v e s tig a tin g a g re a t v a r ie ty of
fo u n d atio n p roblem s for m a th e m a tics an d logic, am o n g w h ich th e co n
siste n c y problem is o n ly one. F o r ex am p le , m e ta m a th e m a tic a l m eth o d s
are ap p lied now in studies of s y ste m a tiz a tio n s of m a th e m a tic s arising
from th e lo gicistic an d in tu itio n istic schools, as w ell as from H ilb e r t’s.
(In versely, m e ta m a th e m a tics owes m uch for its in cep tio n to th e lo gicistic
an d in tu itio n istic in vestigatio n s.) O ur aim in th e rest of th is b o o k is not
to reach a ve rd ict su p p o rtin g or re je ctin g th e fo rm alistic v ie w p o in t in
a n y preassigned v e r s io n ; b u t to see w h a t th e m e ta m a th e m a tic a l m eth o d
consists in, an d to learn som e of th e th in gs th a t h a v e been d isco vered in
pursuing it.
§ 15 . F o r m a l i z a t i o n o f a t h e o r y . W e are n o w a b o u t to u n d erta k e
a program w h ich m akes a m a th e m a tic a l th e o r y itself th e o b je c t of e x a c t
m a th e m a tic a l stu d y . In a m a th e m a tic a l th eo ry, w e s tu d y a sy s te m of
m a th e m a tic a l o b jects. H o w can a m a th e m a tic a l th e o ry itself b e an o b je c t
for m a th e m a tic a l s tu d y ?
T h e result of th e m a th e m a tic ia n ’s a c t iv it y is em b o d ied in propositions,
th e asserted propositions or theorem s of th e g iv e n m a th e m a tic a l th eo ry.
W e ca n n o t hope to s tu d y in e x a c t term s w h a t is in th e m a th e m a tic ia n ’s
m ind, b u t w e can c o n te m p la te th e sy ste m of these propositions.
T h e sy ste m of th ese propositions m u st b e m ad e e n tirely e x p lic it. N o t
all of th e proposition s can b e w ritte n dow n, b u t rath er th e d isciple an d
stu d en t of th e th e o r y sh ould b e to ld all th e co n ditio n s w h ich d eterm in e
w h a t propositions h old in th e th eo ry.
A s th e first step, th e propositions of th e th e o r y sh ould b e arran ged
d e d u c tiv e ly , som e of th em , from w h ich th e others are lo g ic a lly d ed ucible,
b ein g specified as th e a x io m s (or p ostu lates).
T h is step w ill n o t b e finish ed u n til all th e properties of th e u n d efin ed or
tech n ica l term s of th e th e o r y w h ich m a tte r for th e d ed u ctio n of th e
theorem s h a v e been expressed b y axiom s. T h e n it sh ould b e possible to
perform th e d ed u ctio n s tre a tin g th e tech n ica l term s as w ords in th em selves
w ith o u t m eaning. F o r to s a y th a t th e y h a v e m ean in gs n ecessary to th e
d ed u ctio n of th e theorem s, other th a n w h a t th e y d erive from th e axio m s
w hich g o ve rn th em , am o u n ts to sa y in g th a t n o t all of th eir p ro p erties
60 A CRITIQUE OF MATHEMATICAL REASONING CH. I ll
w h ich m a tte r for th e d ed u ctio n s h a v e been expressed b y axiom s. W h e n

th e m ean in gs o f th e te ch n ica l term s are th u s left o u t of acco u n t, w e h a v e
a rriv ed a t th e sta n d p o in t of form al a x io m a tics (§8).
T h e te ch n ica l term s still h a v e g ra m m a tic a l a ttr ib u te s , b e in g nouns,
a d je c tiv e s, ve rb s, etc. A ls o th ere rem ain o rd in ary or lo gica l term s, w hose
m ean in gs are em p lo y e d in th e d eductions. In d eed th e p o in t a t w h ich
fo rm al a x io m a tiz a tio n sto p s is a rb itra ry , in so far as no a b so lu te b asis
e x ists for th e d istin ctio n b e tw e e n th e tech n ica l an d th e o rd in ary term s.
A t a n y rate, w e are still sh ort o f our g o a l o f m a k in g e x p lic it all th e
co n d itio n s w h ich d eterm in e w h a t proposition s h o ld in th e th eo ry. F o r
w e h a v e n o t specified th e lo g ica l prin ciples to b e used in th e d ed u ctio n s.
T h ese prin ciples are n o t th e sam e for all theories, as w e are n o w w ell
aw are (§ 13 ).
In order to m a k e th ese e x p lic it, a second step is required, w h ic h co m
p le te s th e step p re v io u sly carried o u t for th e so -called te ch n ica l term s in
resp ect to th e n o n -g ra m m a tica l p a rt o f th eir m eanings. A l l th e m ean in gs
o f all th e w ords are le ft o u t of a c co u n t, an d all th e co n d itio n s w h ich g o v e rn
th eir use in th e th e o r y are sta te d e x p lic itly . T h e lo gica l prin ciples w h ic h
fo rm erly entered im p lic itly th ro u gh th e m ean in gs o f th e o rd in ary term s
w ill n o w b e g iv e n e ffect in p a rt perh aps b y n ew axiom s, an d in som e
p a rt a t le a s t b y rules p e rm ittin g th e inference of one sen ten ce from
a n o th e r or others. S in ce w e h a v e a b stra c te d en tire ly from th e co n te n t
or m a tte r , le a v in g o n ly th e form , w e s a y th a t th e original th e o r y h as
b een form alized . In its stru ctu re, th e th e o r y is no longer a sy ste m of
m e a n in g fu l propositions, b u t one of sentences as sequences of w ord s,
w h ic h in tu rn are sequences o f letters. W e s a y b y reference to th e form
alone w h ic h co m b in atio n s of w ords are sentences, w h ich sen ten ces are
axio m s, an d w h ich sen ten ces follow as im m e d ia te consequences from
others.
Is su ch fo rm alizatio n possible? T o w h a t e x te n t a g iv e n th e o r y ca n b e
fo rm alized w e sh all learn o n ly a fter a tte m p tin g it an d s tu d y in g th e
results (e.g. §§ 29 , 42 , 60 , 72 ).
T h a t a t lea st a v e r y con siderable m easure of fo rm alizatio n is possible
for m a th e m a tic a l theories is a d isco v e ry w h ich has been spread o ve r a
lo n g stre tch o f m a n 's in te lle c tu a l h isto ry.
The d is c o v e ry o f th e a x io m a tic -d e d u c tiv e m eth o d in m a th e m a tic s
is a ttr ib u te d b y an cien t G reek trad itio n to P y th a g o r a s (sixth c e n tu r y
B .C .), an d com es to us from E u c lid ( 3 6 5 ?— 2 7 5 ? B .C .), w hose “ E le m e n ts "
is said to h a v e h ad th e gre a te st circu latio n of a n y b o o k e x c e p t th e B ib le.
E u c lid failed to m a k e e x p lic it all of th e p o stu lates required in th e d e-
§ 15 FORMALIZATION OF A THEORY 61
d u ctio n of his theorem s. O th ers h a v e b een b ro u g h t in to th e lig h t in
m odern tim es, e.g. th ose g o v e rn in g th e order o f th e p o in ts on a line b y
P asch 1882.
T h e d isco v e ry of th e form al tre a tm e n t o f lo gic, i.e. o f th e p o ssib ility
of describing d e d u c tiv e reasoning w ith sentences in term s o f th eir form ,
appears w ith A r is to tle ( 384— 322 B .C .). A g a in there h a v e been m odern
refinem ents.
W e use b o th discoveries w hen w e form alize a m a th e m a tic a l th eo ry.
T o do so fu lly rigorously, it is p r a c tic a lly n ecessary to re co n stru ct th e
th eo ry in a special sy m b o lic lan gu age, i.e. to sym bolize it. In ste a d of
ca rry in g o u t th e steps described a b o v e on th e th e o r y as w e fin d it in som e
n atu ral w ord lan gu age, such as G reek or E n glish , w e b u ild a n ew sy m b o lic
lan gu age sp e cia lly for th e purpose of expressing th e th eo ry. T h e n a tu r a l
w ord lan gu age s are to o cum bersom e, to o irregular in co n stru ctio n a n d
too am bigu ou s to b e suitable. (Th e sym b o ls in a sy m b o lic la n g u a g e w ill
u su a lly correspond to w hole w ord s in stea d of to le tte rs; an d sequences
of sym b o ls w h ich correspond to sen ten ces w ill b e ca lled “ fo rm u las” .)
T h is n ew la n g u a g e w ill b e of th e gen eral ch aracter o f th e sy m b o lism
w h ich w e fin d in m ath em atics. In algeb ra w e perform d ed u ctio n s as form al
m an ip u lation s w ith equ ation s, w h ich w o u ld be e x c e e d in g ly ted io u s to
perform in ord in ary lan gu age , as som e of th em w ere before th e in v e n tio n
b y V ie ta ( 1591 ) an d others of th e m odern algeb raic n o tatio n s. T h e d is
co v e ry of sim ple sy m b o lic n o tatio n s w h ich lend th em selves to m a n ip
u latio n b y form al rules has been one of th e w a y s b y w h ich m odern m a th e
m atics h ad a d v a n c e d in power. H o w e v e r th e o rd in ary p ractice in m a th e
m atics illu strates o n ly a p a rtia l sy m b o liza tio n an d fo rm alizatio n , since
p art of th e sta te m e n ts rem ain expressed in w ords, an d p a rt o f th e de
d u ction s are perform ed in term s of th e m eanings o f th e w ord s ra th er
th an b y form al rules.
Since L e ib n iz ( 1666 ) co n ceiv ed his id ea of a u n iversal ch aracteristic,
form al logic also has been receiv in g a sy m b o lic tre a tm e n t, w ith th e aid
of m a th e m a tica l tech n iqu es, under D e M organ (1847, 1864), B o o le (1847,
1854), Peirce (1867, 1880), Schroder (18 77, 1890-1905) an d others.
T h ese concurrent d eve lo p m en ts h a v e fin a lly led to fo rm alizatio n s o f
portions of m a th em a tics, in th e strict sense, b y F rege (1893, 1903),
P eano (1894-1908) an d W h ite h e a d an d R u ssell (19 10 -13 ). (Th e m e th o d
of m ak in g a th e o ry e x p lic it w h ich w e h a v e been describing is o fte n ca lled
th e logistic m ethod.)
T o H ilb e rt is due now , first, th e em phasis th a t strict fo rm alizatio n of
a th e o r y in v o lv e s th e to ta l a b stractio n from th e m ean in g, th e result
b e in g ca lled a formal system or form alism (or som etim es a form al theory or
form al m athem atics ) ; an d second, his m eth o d of m a k in g th e fo rm al
sy ste m as a w h o le th e o b je c t of a m a th e m a tic a l s tu d y ca lled meta
mathematics or proof theory.
M e ta m a th e m a tic s in clu d es th e d escription or d efin itio n of form al
sy ste m s as w ell as th e in v e stig a tio n of properties of fo rm al sy stem s.
In d ealin g w ith a p a rticu la r form al sy ste m , w e m a y ca ll th e sy ste m th e
object theory , and th e m e ta m a th e m a tic s re latin g to it its metatheory.
F ro m th e sta n d p o in t o f th e m e ta th e o ry , th e o b je c t th e o r y is n o t
p ro p erly a th e o r y a t all as w e fo rm erly u n d erstoo d th e term , b u t a sy ste m
of m eaningless o b je c ts lik e th e positions in a ga m e of chess, su b je c t to
m ech an ical m an ip u latio n s like th e m o ves in chess. T h e o b je c t th e o r y
is described an d stu d ied as a sy ste m of sy m b o ls an d of o b je c ts b u ilt u p
o u t of sym b o ls. T h e sy m b o ls are regard ed sim p ly as va rio u s kin d s of rec
o gn iza b le o b jects. T o fix our ideas w e m a y th in k of th em co n cre te ly as
m ark s on p ap er; or m ore a c c u ra te ly as a b stra c te d from our exp erien ce
w ith sy m b o ls as m ark s on paper. (Proof th e o r y m u st be to som e e x te n t
a b stra c t, since it supposes arb itra rily lo n g sequences of sy m b o ls to b e
co n stru ctib le, a lth o u g h th e q u a n tity of p aper an d in k in th e w o rld is
finite.) T h e o th er o b je c ts of th e sy ste m are a n a ly z e d o n ly w ith regard
to th e m ann er of th eir co m p o sitio n o u t of th e sym b o ls. B y defin ition ,
th is is all th a t a form al sy ste m sh all b e as an o b je c t of s tu d y for m e ta -
m a th e m a tics.
T h e m e ta th e o ry b elo n gs to in tu itiv e an d inform al m a th e m a tics (unless
th e m e ta th e o r y is itself form alized from a m e ta m e ta th e o ry , w h ich here
w e le a v e o u t o f acco u n t). T h e m e ta th e o ry w ill b e expressed in o rd in ary
lan gu age , w ith m a th e m a tic a l sym b o ls, su ch as m e ta m a th e m a tic a l
va riab les, in tro d u ced acco rd in g to need. T h e assertions of th e m e ta -
th e o r y m u st b e understood. T h e d ed u ctio n s m u st ca rry co n victio n . T h e y
m u st proceed b y in tu itiv e inferences, an d n o t, as th e d ed u ctio n s in th e
form al th e o ry , b y a p p lica tio n s of s ta te d rules. R u le s h a v e been s ta te d to
form alize th e o b je c t th e o ry, b u t n ow w e m u st u n d ersta n d w ith o u t rules
h o w th o se rules w ork. A n in tu itiv e m a th e m a tic s is n ecessary ev en to
define th e form al m a th e m a tics.
(W e sh all u n d ersta n d th is to m ean t h a t th e u ltim a te a p p ea l to ju s t if y
a m e ta m a th e m a tic a l inference m u st b e to th e m ean in g an d e v id e n ce
ra th er th a n to a n y set of co n v e n tio n a l rules. I t w ill n o t p re v e n t us in
p ra c tic e from sy ste m a tiz in g our m e ta m a th e m a tic a l results in theorem s
or rules, w h ich can th en b e ap p lied q u a si-fo rm a lly to a b b re v ia te th e in
t u it iv e reasoning. T h is is a fam iliar procedure in inform al m a th e m a tics.
We sh all som etim es ev e n refer to principles of (intuitionistic) lo g ic
sta te d fo rm a lly , w h en th e form al d eriv a tio n of th o se prin ciples in d ic a te s
th e m e th o d b y w h ich th e reasonin g can be carried o u t in form ally.)
T h e m eth o d s used in th e m e ta th e o r y sh all b e re stricted to m eth o d s,
called fin ita ry b y th e form alists, w h ich e m p lo y o n ly in tu itiv e ly co n
ce iv a b le o b je c ts an d p erform able processes. (W e tra n sla te th e G erm a n
“ fin it” as “ fin ita r y ” , since th e E n g lish “ fin ite ” is used for th e G erm a n
“ en d lich ” .) N o in fin ite class m a y b e regard ed as a c o m p le ted w hole.
Proofs of existen ce sh all g iv e , a t least im p lic itly , a m e th o d for co n
stru ctin g th e o b je c t w h ich is b ein g p ro v ed to exist. (Cf. § 13 .)
T h is restriction is requisite for th e purpose for w h ich H ilb e rt in tro d u ces
m e ta m a th e m a tics. Pro po sitio n s of a g iv e n m a th e m a tic a l th e o r y m a y fa il
to h a v e a clear m ean in g, an d inferences in it m a y n o t ca rry in d u b ita b le
evidence. B y fo rm alizin g th e th e o ry, th e d eve lo p m en t o f th e th e o r y
is reduced to form an d rule. T h e re is no longer a m b ig u ity a b o u t w h a t
co n stitu te s a sta te m e n t of th e th eo ry, or w h a t co n stitu te s a proof in th e
th eo ry. T h e n th e q u estio n w h e th er th e m eth o d s w h ich h a v e been fo rm al
ized in it lead to co n trad ictio n , an d o th er qu estio n s a b o u t th e e ffe ct of
those m eth od s, are to b e in v e s tig a te d in th e m e ta th e o ry , b y m eth o d s n o t
su b je ct to th e sam e d o u b ts as th e m eth o d s of th e original th eo ry.
T h e fin ita r y m eth o d s are of sorts used in in tu itio n istic ele m e n ta ry
num ber th eo ry. Som e form alists a tte m p t to circum scribe th e m still
m ore n a rro w ly (H ilb ert an d B e rn a y s 193 4 p. 43 , an d B e r n a y s 19 3 5,
1938).
W e sh all le a v e th e discussion of th is u n til la te r ( § 81 ). F o r th e purpose
of defen d in g classical m a th e m a tic s a g a in st th e in tu itio n ists, th ere is no
need to use less th a n th e in tu itio n ists w o u ld allow . H o w e v e r it is n a tu ra l
to proceed on th e b asis of s tr ic tly ele m e n ta ry m eth o d s so lo n g as t h e y w ill
suffice. A ll th e exam p les of in tu itio n istic n u m b er-th eo retic reasonin g
g iv e n in § 13 w e sh all ta k e to b e fin ita ry . W e sh all fin d t h a t u p to a la te
stage in our m e ta m a th e m a tic a l in v estig a tio n s, in tu itio n istic m eth o d s of
an e n tirely e le m e n ta ty sort w ill suffice. T h e u ltim a te te st w h e th e r a
m eth o d is ad m issible in m e ta m a th e m a tic s m u st of course b e w h e th er it
is in tu itiv e ly co n vin cin g.
(Som e au th o rs use “ m e ta -” to id e n tify a la n g u a g e or th e o r y in w h ich
an oth er la n gu a ge or th e o r y is m ad e th e o b je c t of a s tu d y not restricted
to fin ita r y m eth ods. A lso “ s y n t a x la n g u a g e ” v s . “ o b je c t la n g u a g e ” is
used in th is con nection . C f. C arn ap 19 3 4 ; also cf. § 37 . In th is b o o k , w e
o n ly use “ m e ta -” w hen th e m eth o d s are fin itary.)
T h e form al sy ste m s w h ich are stu d ied in m e ta m a th e m a tic s are (usually)
so chosen th a t th e y serve as m odels for p arts of inform al m a th e m a tic s

an d lo gic w ith w h ich w e are a lre a d y m ore or less fam iliar, an d from w h ich
t h e y arose b y fo rm alizatio n . T h e m ean in gs w h ich are in ten d ed to b e
a tta c h e d to th e sym b o ls, form ulas, etc. of a g iv e n form al sy ste m , in co n
sidering th e syste m as a fo rm alizatio n of an inform al th eo ry, w e ca ll the
{intended) interpretation of th e sy ste m (or of its sym b o ls, form ulas, etc.).
I n oth er w ords, th e in terp retatio n s of th e sym b o ls, form ulas, etc. are th e
o b je cts, propositions, etc. o f th e inform al th e o ry w h ich are correlated
u n der th e m eth o d b y w h ich th e sy ste m co n stitu tes a m odel for th e in
fo rm al th eo ry.
In th e case of a form u la w h ich represents an id eal sta te m e n t of cla s
sical m a th e m a tic s (§ 14 ), th e in terp retatio n ca n n o t c o n stitu te a w h o lly
in tu itiv e (or fin itary) m ean in g, b u t m u st consist in w h a te v e r else it is th e
classical m a th e m a tic ia n th in k s in term s of in th e in form al (or n o t s tr ic tly
form alized) d eve lo p m en t of classical m a th em a tics, i.e. in th e d e v e lo p m e n t
w h ich has ta k e n p lace h isto r ic a lly an d ta k e s p la ce cu rren tly, w h en th e
procedure is n o t b ein g co n scio u sly form alized in th e stric t sense of proof
th e o ry.
T h e in terp reta tio n m o tiv a te s th e m e ta m a th e m a tic ia n in his choice of
th e p a rtic u la r form al sy ste m w h ich he in trod uces b y his definitions.
I t gu id es h im in ch oosin g th e problem s re la tin g to th e sy ste m w h ic h he
in v e stig a te s. I t m a y ev en p ro v id e h im w ith essential clues to w a rd a c h ie v
in g th e solutio n of th ose problem s. O n ly in th e fin al sta te m e n t an d proof
o f his results is he p ro h ib ite d (as a m eta m a th em a ticia n ) from usin g th e
in terp retatio n .
H o w re strictiv e is th is p ro h ib itio n ? M e ta m a th e m a tic s m u st s tu d y th e
form al sy ste m as a sy ste m of sym b o ls, etc. w h ich are considered w h o lly
o b je c tiv e ly . T h is m eans sim p ly th a t those sym b o ls, etc. are th em selves
th e u ltim a te o b je cts, an d are n o t b ein g used to refer to so m eth in g other
th a n th em selves. T h e m e ta m a th e m a tic ia n looks a t th em , n o t th ro u gh
an d beyond th e m ; th u s th ey are o b je cts w ith o u t in terp reta tio n or
m eaning.
N o w in s tu d y in g th ose o b jects, m e ta m a th e m a tic s m u st b rin g to b ear
its o w n m eth o d s an d tools. T h ese m a y b e a n y th a t are fin ita ry . F o r
exam p le , m e ta m a th e m a tic s m ay e m p lo y th e n a tu ra l num bers in a
fin ita r y w a y . In th e case of form ulas a d m ittin g (outside of m e ta m a th e
m atics) a fin ita r y in terp retatio n , it m a y b e possible w ith in m e ta m a th e
m a tic s to define properties of those form al o b je c ts w h ich (from ou tsid e
th e m e ta m a th e m a tics) are e q u iv a le n t to th eir in terpretatio n s. T h u s the
fin ita r y in terp retatio n s m a y b e b ro u g h t in th ro u gh th e b a c k door. B u t
m e ta m a th e m a tic s ca n n o t in a n y w a y d eal w ith th e n o n -fin ita r y in ter
p retatio n s of th e ideal proposition s of classical m a th em a tics.
In order to m a k e it clear all alo n g w h y w e are in terested in th e form al
syste m s w h ic h w e are considering, an d h o w t h e y c o n stitu te fo rm alizatio n s
of portions of lo gic an d m a th e m a tic s th a t w e are a lre a d y fam ila r w ith
in fo rm a lly, w e Shall in th is b o o k in d icate th e possibilities of in terp retatio n ,
an d use s u g g e stiv e term in o lo gy, such as " p r o o f” for fo rm al d eriv atio n s,
an d " a n d ” for th e n am e of th e sy m b o l & . T h is is n ecessary to our fu ll
purpose, ev en th o u g h th e in terp reta tio n is extran eo u s to th e m e ta
m a th e m a tics itself.
L e t us b rie fly re cap itu la te. In th e full p ictu re, there w ill b e th ree sep
arate an d d istin ct " th e o r ie s ” : (a) th e inform al th e o r y of w h ich th e fo rm al
sy ste m co n stitu te s a fo rm alizatio n , (b) th e form al sy ste m or o b je c t th e o ry,
an d (c) th e m e ta th e o ry , in w h ich th e form al sy ste m is d escribed an d
studied.
H ere (b), w h ich is form al, is n o t a th e o r y in th e com m on sense, b u t a
syste m of sy m b o ls an d o f o b je c ts b u ilt from sy m b o ls (described from (c)),
w h ich h o w e ver form s a k in d of co n ven tio n a liz ed im age or m o d el for (a).
O n th e oth er h an d , (a) an d (c), w h ich are inform al, do n o t h a v e an
e x a c tly determ in ed stru ctu re, as does (b).
T h e n (c) is a th e o r y w ith (b) as its su b je ct m a tte r, w h ich m u st a p p ly
to (b) w ith o u t lo o k in g a t (a), or rpore p recisely w ith o u t lo o k in g a t th e
in te rp re ta tio n of (b) in term s of (a).
F u rth erm o re (c) is restricted to th e use of fin ita r y m eth o d s, w h ile in
general (a) w ill n o t be.
Note 1: At the top of p. 51, the seeming circularity th at not B is used in explaining
not A is to be avoided thus. Sameness and distinctness of two natural numbers (or
of two finite sequences of symbols) are basic concepts (cf. p. 51 lines 20-24). For any
B of the form m = n where m and n are natural numbers, not B shall mean th at m
and n are distinct. The explanation of not A in lines 5-8 then serves for any A other
than of th at form, by taking the B in it to be of that form. Equivalently, since the
distinctness of 1 from 0 is given by intuition (so not 1 = 0 holds), not A means that
one possesses a method which, from any proof of A , would procure a proof of 1 = 0
(cf. lines 8-9).
PART II
MATHEMATICAL LOGIC
Chapter IV
A FO R M AL SYSTEM
§ 16 . F o r m a l s y m b o l s . W e sh all n o w in tro d u ce a p a rticu la r fo rm al

syste m . T h e s y ste m d escribed in th is ch a p te r w ill b e su b je c t m a tte r for
th e four fo llo w in g ch ap ters a n d p arts o f la te r ch ap ters. T h e sy ste m
co n stitu te s a fo rm alizatio n o f a p ortion o f classical e le m e n ta ry n u m b er
th e o r y in c lu d in g th e lo gic required for it.
In s e ttin g u p th e sy ste m , w e h a v e m ad e use o f H ilb e rt a n d A c k e rm a n n
1928, H ilb e rt an d B e rn a y s 1934, 1939, G e n tz e n 193 4-5, B e rn a y s 1936,
an d less im m e d ia te sources.
O u r u n d e rta k in g h as tw o d istin ct aspects. F irst, th e form al sy ste m itself
m u st b e d escrib ed a n d in v e stig a te d , b y fin ita r y m eth o d s a n d w ith o u t
m a k in g use o f an in terp reta tio n of th e sy ste m . T h is is th e m e ta m a th e
m atics. Second, an in terp reta tio n of th e sy ste m m u st b e recognized, u nd er
w h ich th e sy ste m does c o n stitu te a fo rm alizatio n of n u m ber th eo ry.
O n e ap p ro ach w o u ld stress th e second asp ect. W e could a n a ly z e e x is tin g
inform al m a th em a tics, selectin g an d ste re o ty p in g fu n d a m e n ta l co n cep ts,
presupposition s an d d e d u c tiv e con nection s, an d th u s e v e n tu a lly arrive
a t a form al system .
H ere in stead, w e sh all p la ce th e in itial em phasis on th e first asp ect.
T h e form al sy ste m w ill b e in tro d u ced a t once in its fu ll-fled ged co m
p le x ity , an d th e m e ta m a th e m a tic a l in v e stig a tio n s w ill b e pursued w ith
only in c id e n ta l attention to th e in terp retatio n . T h e reader is a sk ed to
co n cen tra te on learn in g p recisely w h a t th e fo rm al sy ste m is, an d h o w it
is in v e stig a te d . T h e in terp reta tio n an d th e reasons for th e ch oices m a d e
in se ttin g u p th is p a rtic u la r sy ste m w ill th e n g r a d u a lly u n fo ld as we
proceed.
T h e first step in se ttin g u p th e fo rm al sy ste m is to list th e form al
symbols. T h e list o f form al sy m b o ls is an alogo u s s tru c tu r a lly to th e a lp h a
b e t of a lan gu age , a lth o u g h und er th e in terp reta tio n m a n y of th e fo rm al
sym b o ls correspond to entire w ords an d phrases ra th er th a n to single
letters. T h e list of the form al sy m b o ls follows.
Logical sym bols : D (im plies), & (and), V (or), -1 (not), V (for all),
69
70 A FORMAL SYSTEM CH. IV
3 Predicate symbols: = (equals). Function symbols: +
(there e x ists).
(plus), • (tim es), ' (successor). Individual symbols: 0 (zero). Variables: a,
b, cy .... Parentheses: (, ).
T h e w ord s sh ow n p a re n th e tic a lly m a y b e used in read in g th e sym b o ls, an d
are in ten d ed to su ggest th e in terpretatio n s in a p relim in ary w a y , e.g. th e
in terp retatio n s of th e lo gical sym b o ls as 'lo gical co n sta n ts’ . T h e va riab les are
in terp reted as ra n gin g o ve r th e n a tu ral num bers. A n in fin ite list or enu m era
tio n of th e v a ria b les is supposed to b e a t h an d (p o te n tially, cf. § 13).
W e reiterate th a t th e in terpretatio n s are extran eou s to th e d escription
of th e fo rm al s y ste m as such. I t m u st b e possible to proceed regard in g
th e form al sy m b o ls as m ere m arks, an d n o t as sy m b o ls in th e sense of
sym b o ls for so m eth in g w h ich th e y sy m b o lize or sign ify. I t is supposed
o n ly t h a t w e are able to recognize each form al sy m b o l as th e sam e in
each of its recurrences, an d as d istin ct from th e o th er form al sym bols.
In th e case o f th e v a ria b les th is m u st in clude our b ein g able to recognize
a sy m b o l w h ic h is a v a ria b le to b e such.
T h e form al sy m b o ls c o n stitu te th e first c a te g o r y of form al o b jects.
W e d eriv e from th is a second c a te g o r y b y co n stru ctin g fin ite sequences
(of occurrences) of form al sym b o ls, w h ich w e c a ll formal expressions.
T h e w o rd "o ccu rren ce” is used here to refer to th e m em bers of th e se
quen ce in th eir sta tu s as m em bers, an d to em phasize t h a t d ifferen t
m em bers m a y b e th e sam e sy m b o l (w hich agrees w ith our p revio u s use
o f th e term 'seq u en ce’, e.g. §§ 1, 2). T h e form al expressions in clu d e those
co n sistin g of a single (occurrence of a) form al sym bol. E x c e p t w h en sta te d ,
th e e m p ty sequence (w ith no m em ber) w ill n o t b e includ ed . F o r e x a m p le ,
0, (#) + (£), 0
(tf) = ( ) an d 0 00=
(( V are form al expressions. The la st
con sists of seven (occurrences of) sym b ols, i.e. it h as seven m em bers;
th e th ird, fifth an d six th (occurrences of) sy m b o ls in it are each an
(occurrence of) 0; an d th e (distinct) sym b o ls w h ich occur in it are (, 0,
V , = . T h e form al expressions are an alogous stru c tu r a lly to th e w ord s of
a la n g u a g e ; b u t under th e in terp retatio n som e of th e m correspon d to
entire sentences, e.g. (^) = (0), an d others are w ith o u t sign ifican ce, e.g.
(( V0 00= . A g a in our term in o lo gy belies th e fa c t th a t, for th e fo rm al
sy ste m as such, th e expressions express n oth in g, b u t are o n ly ce rta in
recognizable an d d istin gu ish ab le objects.
W e shall also use, as a th ird c a te g o r y of form al o b je cts, th e fin ite
sequences df (occurrences of) form al expressions.
In discussing th e form al o b je cts w e shall often, in s te a d of e x h ib itin g
th em , represent (i.e. denote) th em b y letters in tro d u ced for th e purpose,
or b y expressions in v o lv in g letters alrea d y so in trod u ced . F o r e x a m p le ,
§16 FORMAL SYMBOLS 71
th e le tte r “ s ” m ig h t b e used to represent th e fo rm al expression (a) + (b),
an d “ A ” to represent ( a ) = 0 . F u r th e r illu stra tio n s w ill ap p ear p resen tly.
L e tte r s an d expressions so u sed are n o t form al sy m b o ls a n d expressions,
b u t in form al or m e ta m a th e m a tic a l sy m b o ls an d expressions, w h ich sta n d
as nam es for form al o b jects. In th is, as co m p ared w ith o rd in ary inform al
uses of sym b o lism , th ere is th e new featu re t h a t th e o b je c ts n a m ed are
th em selves sy m b o ls or o b je c ts c o n stru cted from sym b o ls. W e h a v e th u s
a d istin ctio n to preserve b etw ee n tw o kin d s of sym b o lism , th e form al
sym b o lism a b o u t w h ich w e are sp eak in g, an d th e in tu itiv e or m e ta m a th e
m a tic a l sy m b o lism in w h ich w e are sp eak in g a b o u t th e other. D ifferen ces
in th e kin d s of t y p e w h ich w ill b e used for th e tw o purposes (a, b, t,x,
3
< , S vs. a, b , t, x , A , B) w ill assist in keep in g th e m a tte r straig h t.
T h e use of sy m b o ls an d expressions to n am e th e o b je c ts w e are ta lk in g
a b o u t sh ould n o t b e considered as n o v e l; our e v e r y d a y m e th o d for
co n stru ctin g a sen ten ce a b o u t an o b je c t requires this. W h a t is n o v el,
rath er, is th e oth er procedure, w h ich w e use so m ew h a t in our m e ta
m a th e m a tics, of in co rp o ra tin g th e o b je c t itself, i.e. a sp ecim en of th e
o b je c t, d ir e c tly in to th e sentence. A lth o u g h th is v io la te s th e u su al canons
of g ra m m a tic a l p ro p riety , it is u n a m b igu o u s w h en w e are en g a g e d in
m e ta m a th e m a tics. F o r in m e ta m a th e m a tic s w e m u st tr e a t th e fo rm al
sy m b o ls as m eaningless, a n d therefore th e fo rm al o b je c ts ca n n o t serve as
nam es for oth er o b je cts, an d a sen ten ce co n ta in in g a sp ecim en o f a fo rm al
o b je c t ca n o n ly b e a b o u t th e form al o b je c t itself.
T h e se rem arks a p p ly to our m eta m a th e m a tics. In an o ccasio n al p a s
sage, con cerned w ith th e in terp reta tio n an d so la b eled for th e reader, w e
m a y g iv e th e fo rm al sy m b o ls an in form al sta tu s, tre a tin g th e m th e n as
m ean in gfu l.
In our m e ta m a th e m a tic a l s tu d y of th e fo rm al expressions, w e sh all
m a k e use o f th e o p eratio n o f juxtaposition (or concatenation ), in w h ich
tw o or m ore sequences of fo rm al sy m b o ls are co m b in ed c o n se c u tiv e ly
to prod u ce a n ew sequence. F o r ex am p le, th e ju x ta p o s itio n o f th e tw o
form al expressions ((0 V 0 0 = a n d (a) -{-(b) in t h a t order prod uces th e n ew
form al expression (( V 0 00 = (# ) + (£); an d th e ju x ta p o s itio n of th e sev en
form al expressions (> ( a) + (b)f ), *, (, (c)', ) in th e g iv e n order p roduces
th e n ew form al expression ((tf) + (iOM (c)').
W h e n som e o f th e fo rm al expressions to b e ju x ta p o s e d are b e in g rep
resented b y m e ta m a th e m a tic a l le tters or expressions, th ese la tte r m a y
a p p ear in p la ce o f th e fo rm al expressions w h ic h t h e y represent in w ritin g
th e result o f th e ju x ta p o sitio n . F o r ex am p le , if th e le tte r “ s ” represents
som e fo rm al expression, th e resu lt of th e ju x ta p o s itio n of th e sev en fo rm al
expressions (, s , ), *, (, (c )\ ) is w ritte n “ (s)-((c)')” H ere “ (s)-((c)')'' is a

m e ta m a th e m a tic a l expression represen tin g a form al expression, th e form al
expression represen ted d ep en d in g on w h a t form al expression th e le tte r
"s” represents. In p articu lar, if s is (a) + (b)} th e n (s)*((c)') is
((d) + (i>))-((c)').
§ 17 . Formation rules. W e sh all n o w d efine certa in su b ca tego ries

o f th e form al expressions, b y d efin ition s an alogo u s to th e rules of s y n t a x
in gram m ar.
F ir s t w e define ‘term ', w h ic h is an alogo u s to noun in gram m ar. T h e
term s of th is s y ste m all represent n a tu ra l num bers, fix e d or va ria b le .
T h e d efin itio n is fo rm u la te d w ith th e aid of m e ta m a th e m a tic a l v a ria b le s
“ s ” an d “ t ” , a n d th e o p eratio n of ju x ta p o sitio n , as e x p la in e d a b o ve . I t
h as th e form o f an in d u c tiv e d efin ition , w h ich enables us to proceed from
kn o w n ex am p le s o f term s to fu rth er ones.
1. 0 is a term. 2. A v a ria b le is a term. 3— 5 . I f s an d t are terms, th e n
( s ) + ( t) , (s)*(t) an d (s)' are terms. 6. T h e o n ly terms are th ose g iv e n
b y 1— 5.
E xample 1. 2 , 0 , a, h an d c are term s. T h e n b y 5 , (0)' an d
B y 1 and
(e)' are term s. A p p ly in g 5 again , ((0 )')' is a te rm ; an d a p p ly in g 3 , ((c)f) + (a)
is a term .
W e n o w g iv e a d efin itio n o f ‘fo rm u la', an alo go u s to (d eclarative)

sen ten ce in gram m ar.
1. I f s a n d t are term s, th e n (s) = (t) is a formula. 2— 5 . I f A a n d B
are formulas, th e n (A) D (B), (A) & (B), (A) V (B) an d - i ( A ) are for
mulas. 6— 7 . I f x is a v a ria b le a n d A is a form ula, th e n V x (A ) a n d 3 x (A )
are formulas. 8. T h e o n ly formulas are th ose g iv e n b y 1— 7 .
E xample 2 . U sin g 1 an d th e ex am p le s o f term s a lre a d y o b ta in ed ,

(a) = (b) an d (((<:)') + (a)) = (b) are form ulas. Then u sin g 5 an d 7,
- i ((a) = (b)) an d 3 c((((c)') + (tf))=(i>)) are form ulas. F in a lly by an
a p p lic a tio n of 2, th e fo llo w in g is a fo rm u la:
(A) 3
( c((((c)') + (a)) = (*))) D h ( ( a ) = (b))).
T h e in d u c tiv e d efin ition s of term an d form ula h a v e th e con sequence
t h a t each term or fo rm u la ca n b e b u ilt u p from 0 a n d v a ria b les b y a series
o f steps, each of w h ich step s corresponds to a d irect clause of one of th o se
d efin itio n s (§ 6), an d m a y b e ca lled an application of th a t clause.
E a c h step , e x c e p t an a p p lica tio n of 1 or 2 of th e d efin itio n of term , is
§17 FORMATION RULES 73
of th e fo llo w in g kin d . A t th e sta rt w e h a v e g iv e n an expression or p air
of expressions p re v io u sly o b ta in ed . W e enclose th e g iv e n expression or
each o f th e g iv e n expressions in parentheses, an d in tro d u ce an ex
pression o f one of th e te n form s
(B) D , & , V, - i , V x , l x , = , +, *, ',
w here x is a va riab le. L e t us ca ll an expression of one of th ese te n form s

an operator. propositional connectives, an d
In p articu lar, D , & , V, - i are
operators of th e form s V x a n d l x are quantifiers, V x b ein g a universal
an d l x an existential quantifier) th ese six are logical operators.
T h e g iv e n expression or p air of expressions w e c a ll th e scope of th e
o perator in th e resu ltin g expression. B y fo llo w in g th ro u gh th e entire
co n stru ctio n of a term or form ula, estab lish in g in th e o b vio u s w a y a
correspondence b etw ee n th e p arts of th e g iv e n expression or p air o f
expressions an d p a rts of th e resu ltin g expression a t each step, w e are led
to an assign m en t of a scope, n o t m erely to th e operator la st in tro d u ced in
th e co m p le te d term or form ula, b u t to e v e r y o perator in t h a t term or
form ula.
E xample 3 . In th e form u la (A) th e scope of th e first occurrence o f

= con sists of th e p a rt ((c)') + (tf) an d th e first occurrence of b, an d th e
scope of th e 1 c is th e p a rt (((c)f)-i-(a)) = (b).
W e n o w sta te th e fo llo w in g fa c t, th e rigorous proof of w h ich w e sh all
consider in a m om en t. In a g iv e n term or form ula, th e scopes o f th e
operators can b e recognized w ith o u t a m b ig u ity from th e arran gem en t of
th e parentheses. In oth er w ords, th e parentheses m a k e it possible, g iv e n
th e term or form u la as a fin ite sequence of form al sym b o ls, to reco ver
all essential d eta ils of its co n stru ctio n und er th e in d u c tiv e d efin itio n s of
term an d form ula.
T h e rigorous proof of th is fa c t is afford ed b y L e m m a 2 of § 7 E x a m p le 2,
to g e th e r w ith th e fo llo w in g lem m a w h ich ca n b e p ro v e d b y in d u ctio n
from th e in d u c tiv e d efin ition s of term an d form ula.
Lemma 4 . In a given term or form ula, there exists a proper pairin g of

the parentheses (which are 2 n in number, n being left parentheses and n being
right parentheses) such that the scope of each operator occurs as follows.
(a) F or operators having one expression as scope, the scope is im m ediately
enclosed w ithin paired parentheses, and the operator stands im m ediately
outside this p a ir of parentheses , i.e. im m ediately to the left of the left paren~
thesis (in the case of - i , V x , 3 x) or im m ediately to the right of the right
parenthesis (in the case of ').
(b) F or operators having two expressions as scope {nam ely D , & , V,

= , *)> °/ expressions is im m ediately enclosed w ithin paired
parentheses, and the operator stands im m ediately between the right parenthesis
of the p a ir enclosing the left expression and the left parenthesis of the p a ir
enclosing the right expression.
E xample 3 (concluded). T h e d isp la y e d e x a m p le (A) of a fo rm u la
co n ta in s 22 parentheses. By Lem m a 4, th e 22 p aren th eses a d m it a
proper pairing, w h ich is d iscovered in th e process of c o n stru ctin g th e
form ula u n der th e d efin ition s of term an d form ula, an d w h ich in d icate s
th e scopes of th e operators. K n o w in g t h a t th ere ex ists a proper p airin g,
b y Lem m a 2 t h a t p airin g is un iq ue, a n d ca n therefore b e d isco vered b y
th e algo rith m of § 7 , w ith o u t prior kn o w led ge of th e c o n stru ctio n o f th e
form u la under th e d efin itio n s of term an d form ula. W e a c tu a lly d id so
a t th e end of § 7, w here w e e x a m in e d th e sam e 22 parentheses w ith o u t
lo o k in g a t th e in te rv e n in g sym b ols. U s in g th e resu ltin g p airin g o f th e
22 parentheses as t h e y occu r w ith in th e co m p le te form ula, w e ca n see
t h a t th e scope of th e first occurrence of = con sists of th e expression
enclosed b y th e parentheses (3 )*° an d th e expression enclosed b y th e
parentheses (J1 )*2. T h is agrees w ith our p revio u s id e n tifica tio n o f th a t
scope. S im ilarly, th e sco pe of 3c is enclosed b y th e parentheses )*3.
Lem m a 3 of § 7, w h ile n o t n ecessary to th e p roof th a t th e scopes c a n

b e d isco vered from th e arran gem en t o f th e parentheses, is u sefu l in
reasonin g a b o u t th e scopes in p a rts a n d th e w h ole o f a term or form ula.
F o r ex am p le , if M , N a n d A are form ulas, an d A occurs as a (con secutive)
p a rt, n o t th e w hole, o f (M) D (N), w e ca n infer t h a t th is p a rt (or each
such part) is eith er a p a r t of M or a p a rt o f N .
In ch oosin g our defin itio n s o f term a n d form ula, w e of course p ro v id e d
th e p aren th eses for th e a b o v e d escribed purpose o f in d ic a tin g th e scopes
u n a m b ig u o u sly . N ow e v id e n tly m ore paren theses w ill u su a lly b e in
tro d u ce d u n d er th e d efin itio n s th a n are s tr ic tly n ecessary for th e purpose.
L e a v in g th e d efin itio n s as t h e y stan d , w e ca n agree to o m it superfluous
p aren th eses as an a b b re v ia tio n in th e w ritin g d o w n o f term s an d form ulas,
or o f m e ta m a th e m a tic a l expressions represen tin g th em .
T h e possib ilities in th is d irectio n are e x te n d e d b y e m p lo y in g c o n v e n
tio n s o f a sort fam iliar from algeb ra, w here “a - b + c ” is u n d ersto o d to
m ean (a*b)+ c. W e s a y here t h a t + ranks ah ead o f •, an d ra n k our
o perators in th e order in w h ich w e h a v e liste d th e m a t (B) a b o ve . T o restore
a n y p aren th eses w h ich are le ft o u t in a b b r e v ia tin g a term or form ula,
one m a y p roceed step b y step, each tim e selectin g an operato r w h ich o f
§17 FORMATION RULES 75
those present com es earliest in th e list, i.e. an operato r of h igh est ra n k,
an d g iv in g it th e gre ate st scope co m p a tib le w ith th e requirem en t t h a t
th e w hole b e a term or form ula.
W e sh all n o t a lw a y s o m it th e m a x im u m n um ber of parentheses w h ic h
our co n ven tio n w o u ld allow , b u t aim a t securing m a x im u m re a d a b ility .
(W ith th is aim , w e also som etim es alter parentheses to square b ra c k e ts
or c u r ly brackets.)
E xample 4. R esto rin g parentheses to “A 3 B V C & D ” g iv e s suc

ce ssiv e ly “A 3 ( B V C & D )” , “A D ( ( B V C ) & D ) ” , “ (A ) 3 (((B)V(C))
& (D ))” . W e a b b re v ia te th e d isp la ye d ex am p le (A) of a form u la as fo llo w s :
(A ') 3c ( c '+ a = b ) 3 -i a=b.

A n o th e r k in d of a b b re v ia tio n is afford ed b y in tro d u cin g a new sy m b o l,
w ith a m eth o d for tra n sla tin g an expression co n ta in in g th e n ew sy m b o l
b a c k in to one w ith o u t it. F o r ex am p le, w e a b b re v ia te th e term s ( )', 0
1” , " 2 ” , “ 3 ” , . . . ,
((0)')', (((0)')')', . . . as " r e s p e c tiv e ly ; an d w e a b b re v ia te
the form ula -ia= b as “a ^ b ” , an d th e form ula 3 c ( c '+ a = b ) as
“a < b ”. T h d d isp la y e d form u la (A) can th en b e w r it te n :
(A") a< b 3 a^b.

T h e gen eral rule for th e a b b re v ia tio n allow s us to w rite “ s ^ t ”
as a b b re v ia tio n for - i s = t w h en ever s an d t are term s. T h e gen eral rule
for th e a b b re v ia tio n “ < ” allow s us to w rite “ s e t ” as a b b re v ia tio n for
3 x ( x '+ s = t ) w h e n eve r x is a v a ria b le an d s an d t are term s n o t co n
ta in in g x . In u n a b b re v ia tin g , w hen th e in tro d u ctio n o f th e a b b re v ia tio n
h as suppressed a va ria b le , as in th e case of “ < ” , th ere is an a m b ig u ity
resp ectin g th e v a ria b le to b e su pplied. T h u s in u n a b b re v ia tin g “ s e t ” ,
w e m a y choose as th e x a n y v a ria b le w h ich s an d t do n o t co n ta in . T h is
a m b ig u ity is of m inor con sequence, since th e sta te m e n ts w e sh all w ish to
m ak e a b o u t th e fo rm u la a b b r e v ia te d w ill h o ld regardless of w h a t a d
m issible v a ria b le is chosen.
W e sh all regard all th is a b b re v ia tio n as m erely in th e e xp o sitio n of th e
m e ta m a th e m a tics. T h is is a d e q u a te for our purposes, a n d th e r e b y w e
keep th e fu n d a m e n ta l d efinitions, w h ich estab lish th e fo rm al sy ste m ,
th e o r e tic a lly sim pler. M e ta m a th e m a tic a l sta te m e n ts a b o u t term s an d
form ulas of th e sy ste m are h ence to b e u n d erstoo d to refer to th e u n
ab b re v ia te d expressions in th e literal sense of th e d efinitions, w h a te v e r
sh o rth an d w e m a y e m p lo y in w ritin g th e statem e n ts.
76 A FORMAL SYSTEM CH. IY
§ 18. Free and bound v a r i a b l e s . A n occurrence o f a v a ria b le x

in a fo rm u la A is said to b e bound (or as a bound variable ), if th e occurrence
is in a q u a n tifie r V x or 3 x or in th e scope o f a q u a n tifie r V x or 3 x (w ith
th e sam e x ) ; otherw ise, free (or as a free variable).
E xample 1. In 3 c ( c '+ a = b ) 3 - i a = b , b o th occurrences o f a an d

b o th occurrences o f b are free, an d b o th occurrences of c are b ou n d . In
3c{cf+ a = b ) 3 - i a = b + c , th e first tw o occurrences of c are b o u n d
an d th e th ird is free. In 3c(3c(c'+^=^) 3 - i a = b + c ) all occurrences
o f c are boun d.
W e also s a y t h a t a n y occurrence of a v a ria b le x in a term t is free, as

w ill fo llo w from th e a b o v e d efin itio n if ap p lied read in g " te r m t ” in stea d
o f " fo rm u la A " . T h e d istin ctio n b e tw e e n a free an d a b o u n d occurrence
o f a v a ria b le is a lw a y s r e la tiv e to th e term or form u la in w h ich it is (at
th e m om en t) b e in g con sidered as an occurrence.
E xample 2. T h e th ird occurrence of c in lc ( 3 c(c,+ a = b ) 3 -i a = b + c )

is free w h e n considered as an occurrence in th e p a rt c ta k e n b y itself or c '
b y itself or c '+ a b y itself or c ' + a = b b y itself, an d b o u n d as an o c
currence in 3c{c,Jr a —b) b y itself or 3c ( c '+ a = b ) 3 ~ ia = b -\-c b y itse lf
or in th e w hole form ula.
A v a ria b le x w h ich occurs as a free v a ria b le (briefly, occurs free) in A

is ca lled a free variable of A , a n d A is th e n said to contain x as a free
variable (briefly, to contain x free) ; a n d likew ise for b o u n d variab les.
E xample 3. T h e free v a ria b le s o f 3 c ( c ' + a = £ ) 3 - i a = £ + c are a, b
a n d c , a n d th e o n ly b o u n d v a ria b le is c.
A b o u n d occurrence o f a v a ria b le x in a fo rm u la A is b o u n d by t h a t
p a rticu la r one, o f th e q u a n tifiers V x or 3 x (w ith th e sam e x) in th e
scope o f w h ic h it lies, w h ic h h as th e le a st scope (briefly, b y th e in n erm o st
q u a n tifie r in w h ose scope it lies), or in case it is an occurrence in a q u a n
tifier V x or 3x, b y t h a t q u a n tifie r itself (or th e la tte r binds th e form er).
E xample 4 . In 3 c( 3 c ( c '+ t f = ^ ) 3 - i a = b + c ) th e first an d fo u rth

occurrences o f c are b o u n d b y th e first q u a n tifie r 3 c, an d th e secon d
a n d th ird occurrences o f c b y th e second q u a n tifie r 3 c.
I n b u ild in g u p a fo rm u la u n der th e d efinitions of term an d form ula,
a g iv e n b o u n d occurrence o f a v a ria b le in th e resu ltin g form u la is b o u n d b y
t h a t one o f th e q u an tifiers w hose in tro d u ctio n first co n v e r te d it from a
free to a b o u n d occurrence (or if it is a v a ria b le in a q u an tifier, b y th e
q u a n tifie r in w h ich it is introduced).
| 18 FREE AND BOUND VARIABLES 77
E xample 5. C om p are E x a m p le 4 w ith E x a m p le 2 .
A few p relim in ary rem arks are offered n o w on th e in terp reta tio n o f
free an d b o u n d v a ria b les (som etim es called 'real' an d 'a p p a re n t' variab les).
T h e rem arks are o f course n o t p a rt o f th e m eta m a th e m a tics, b u t sh ou ld
h elp to e x p la in th e ad o p tio n o f th e m e ta m a th e m a tic a l d iscrim in ation s.
A n expression c o n ta in in g a free v a ria b le represents a q u a n tity or p ro p
osition d ep en d in g on th e v a lu e of th e va riab le. A n expression co n ta in in g
a b ou n d v a ria b le represents th e result of an operatio n perform ed o v e r th e
range of th e va riab le. O u r b o u n d v a riab les are asso ciated w ith th e lo g ic a l
operations o f q u a n tific a tio n , b u t ex am p le s occu r w ith o th er sorts o f
operations fam iliar to m ath em aticia n s. In th e fo llo w in g n an d y are free,
i an d x are b o u n d :
n ry
(A) 2 ait lim f(x, y), / f(x, y) dx.
i —1 x —>0 ■ '—y
In th e fo llo w in g th e occurrence o f t as u pper lim it of th e in te g r a l is free

an d th e occurrences in th e in tegra n d are b o u n d :
(B) f f (t) it.
T o go a little fu rth er w ith th e in terp retatio n , w e m a y n o te som e

ch aracteristic differences w h ich it im poses on th e w a y w e m a y use th e
tw o kind s of v a ria b les in in form al m ath em atics. A b o u n d v a ria b le form s
p art of a circu m lo cu tio n for expressing th e result of an op eratio n carried
o u t o ve r th e range of th e v a ria b le, an d one can h en ce ch an ge th e v a r ia b le
to a n y oth er h a v in g th e sam e ran ge w ith o u t alte rin g th e m ean in g (su b ject
to certain precautions). F o r ex am p le,
(C) lim f(z ,y ), ( f ( t ,y ) d t

j = 1 z—>0 ■ '-y
w ould (ordinarily) m ean th e sam e as th e re sp ective expressions (A)

ab o v e (but lim f(y, y) is n o t (usually) th e sam e as lim f{x ,y )). I f in an
2/—>0 £ -> 0
expression w e s u b s titu te for a free v a r ia b le an expression represen tin g
a co n sta n t or v a ria b le o b je c t from its range, w e (ordinarily) o b ta in a
m ean in gfu l result, w h ile such a su b stitu tio n w o u ld resu lt in nonsense if
applied to a b o u n d va ria b le. F o r ex am p le (su b stitu tin g in (A)),
(D) Z a it
i =l
lim
x -+ Q
f(x, 2), f
J~ z
f(x ,z) dx
78 A FORM AL SYSTEM CH . IV
are (ordinarily) sign ifica n t expressions, b u t n o t
W h e n th e sam e v a ria b le occurs b o th free an d b o u n d in an expression,

th e q u a n tity represen ted b y th e expression d epen ds o n ly on th e v a lu e
o f t h a t v a ria b le in its free occurrences. T h u s th e in tegra l (B) is a fu n ctio n
of t , w hose v a lu e for t— 3 is
not
S ubstitution . In s ta tin g th e m e ta m a th e m a tic a l d efin itio n s of th e

n e x t section , w e sh all use an op eratio n of su b stitu tio n , w h ich w e d efine
as follow s. T h e substitution of a term t for a va ria b le x in (or sy n o n y m o u sly ,
throughout) a term or fo rm u la A sh all con sist in rep lacin g sim u lta n eo u sly
each free occurrence of x in A b y an occurrence of t. T o describe th is in
ju x ta p o s itio n n o ta tio n , le t n b e th e num ber of free occurrences of x in
A (w ;> 0 ); an d w rite A as “A qxA ^ . . .A ^ x A * ” sh o w in g th ese o c
currences (A 0, Av ..., A n_v A n b ein g p arts p o ssib ly e m p ty co n ta in in g
no occurrence of x free r e la tiv e to A as a w hole, an d all th e n occurrences
o f x sh ow n b ein g free). T h e n th e result of th e su b stitu tio n of t for x in
A is A o t A ^ . . , A w_ !tA n.
A co m p a c t m e ta m a th e m a tic a l n o ta tio n w ill b e u sefu l in represen tin g
th e result o f a su b stitu tio n . I f su b stitu tio n is to b e perform ed for x , w e
first in tro d u ce a co m p o site n o ta tio n such as <#A ( x ) ” for th e su b stitu e n d ,
sh o w in g its d ependence on x a fte r th e m ann er o f n o ta tio n for fu n ctio n s in
m a th e m a tic s (§ 10). T h e resu lt of su b stitu tin g t for x in A (x ) is th en
w ritte n “ A ( t ) '\
E xample 6. L e t x b e c, an d
A (x ) or A (c) b e 3c ( c '+ tf = i> ) 3 ~ ia = b -\-c .
Then A(0) is 3 c(c#4-4==/0 3 - i a —b + 0 ,
an d A(tf) is 3c ( c '^ a ~ b ) 3 mm\a = b + a .
E xample 7. L e t x b e a , an d A (x ) b e a+c=a. T h e n A(0) is 0 + c ^ = 0 ,
an d A (b) is b + c = b .
T h e su b stitu tio n w h ich g iv e s A (t) m u st a lw a y s b e perform ed for th e

origin al v a ria b le x in th e original form u la A (x ), i.e. for th e v a ria b le
an d in th e form u la for w h ich th e n o ta tio n “ A ( x ) ” is first in trod u ced.
E xample 7 (concluded). F o r th e a b o v e x an d A (x ), A (c) is c+ c=c.

I f w e s u b s titu te b for c in A (c), w e o b ta in b+b=b. T h is is n o t th e sam e
§ 18 FREE AND BOUND VARIABLES 79
as A (b), w h ich w e o b ta in e d a b o v e co rre c tly b y s u b stitu tin g b for a in A (a),

i.e. for th e origin al x in th e o rigin al A (x ). (Th e sam e d iffic u lty can occur
b y th e m isuse of th e n o ta tio n for a fu n ctio n in inform al m ath em atics.)
W e h a v e n o t required t h a t th e v a ria b le x a c tu a lly occur as a free

v a ria b le in A (x ). W h e n x is n o t a free v a ria b le of A (x ), th e result A (t)
of th e su b stitu tio n is th e original expression A (x ) itself.
S im ila rly, we define s u b stitu tio n perform ed sim u lta n eo u sly for a
n u m b er o f d istin c t v a ria b le s ; an d w e sh all e m p lo y like n o tatio n s, such as
" A ( x x>. . . , x n)” for th e su b stitu e n d , an d " A f t j , . . . , t n) ” for th e result.
H e n cefo rth w e sh all o fte n in tro d u ce these co m p o site n o tatio n s, such
as “ A ( x ) " or " A ( x x, . . . , x n)” in stea d of " A " , w h en w e are in terested in th e
d epen den ce of A on a v a ria b le x or v a ria b les xv . . . , x n, w h e th er or n o t
w e are a b o u t to m a k e a su b stitu tio n . F o r ex am p le , w e u su a lly d esign ate
a fo rm u la b y " A ( x ) ” in stea d of " A " , w h en w e w a n t to use it in V x A ( x )
(read 'T o r all x , A o f x ” , or b rie fly " a ll x , A of x ” ) or 3 x A (x ) (read " th e re
ex ists an x such t h a t A of x " , or b rie fly " e x is ts x , A of x ” ). W e re p e at
th a t b y u sin g " A ( x ) ” (or " A ( x x, . . . , x n)”) w e do n o t im p ly t h a t x (or
each of xv . . . , x n) n ecessarily occurs free in th e fo rm u la d esign ated .
T h e p relim in ary rem arks on th e in terp reta tio n shed lig h t on w h y w e
h a v e elected to define our m e ta m a th e m a tic a l su b stitu tio n o p era tio n as
a p p ly in g o n ly to th e free occurrences of th e variab les.
W e n o w s a y t h a t a term t is free at the free occurrences of a v a ria b le x
in a fo rm u la A (x ) (or t is free at the substitution positions for x in A (x ), or
b rie fly t is free for x in A (x )), if no free occurrence o f x in A (x ) is in th e
scope o f a q u a n tifie r V y or 3y w here y is a v a ria b le o f t (i.e. occurs in t).
E xample 8. T h e term s d, rf+ 0' an d a*d are free for a in th e first

b u t n o t in th e second of th e fo llo w in g form u las:
(I) 3c { c '+ a = b ) & - w / = 0, 3d ( d '+ a = b ) &

U n d e r th is d efin ition , w h en t is free for x in A (x ) an d o n ly th en , th e
su b stitu tio n o f t for x in A (x ) w ill n o t in tro d u ce t in to A (x ) a t a n y p la ce
w here a (free) v a ria b le y of t b ecom es a b o u n d occurrence of y in th e
result A (t).
E xample 8 (concluded). d + 0 ' for a in (I) g iv e s

S u b s titu tin g
(II) 3 c { c '+ ( d + 0 ')= b ) & - u / = 0 , l d { d ' + ( d + 0 ')=l>) & - u / = 0 ,

re sp e ctiv e ly . In th e first of these, th e d of th e occurrence of d-\- 0 ' in
tro d u ced b y th e s u b s titu tio n rem ains free in th e w h ole form ula, b u t n o t
in t h e second.
W e s a y th a t th e s u b stitu tio n o f t for x in A (x ) is free, w h en t is free for
x in A (x ). W ith o n ly th e sm a tte r in g of in terp reta tio n in d ic a te d a b o v e ,
it sh o u ld b e clear t h a t a s u b s titu tio n is in a pp ro p riate w h e n it is n o t free.
T h e tw o form ulas in (I) m ean th e sam e; b u t th e tw o in (II) do n ot.
F o r an in form al exam p le , consider th e second expression o f (A) or (C).
T h is stan d s for a fu n ctio n o f y , ca ll it
(G) f{y) = lim f{x, y) = lim f(z, y).

2 — >0 Z —>0
T h e v a lu e o f /(y) for y = z is th e n g iv e n p ro p erly b y
(H) f(z) = lim /(*, z), not b y f{z) = lim f(z, z).
x— >0 z— > 0
E xample 9. To illu stra te th e h a n d lin g of th e te rm in o lo g y an d

n o ta tio n s e x p la in ed in th is section, s a y th a t x is (i.e. “ x ” denotes) a
v a ria b le , A (x ) is (i.e. " A ( x ) ” denotes) a form ula, an d b is (i.e. “ b ” denotes)
a v a ria b le such th a t (i) b is free for x in A (x ) an d (ii) b does n o t occu r free
in A (x ) (unless b is x). U n d e r our su b stitu tio n n o ta tio n , since “ x ” an d
“ A (x )” are in tro d u ced first, (iii) A (b ) is (b y definition) th e resu lt o f
s u b s titu tin g b for (the free occurrences of) x in A (x ). B y (i), th e occur
rences of b in A (b) w h ic h are in tro d u ced b y th is s u b stitu tio n are free. B y
(ii), there are no o th er free occurrences of b in A (b ). T h u s th e free o c
currences o f b in A (b ) are e x a c t ly th e occurrences in tro d u ced b y th e
su b stitu tio n . H en ce (in versely to (i) — (iii))* (iv) x is free for b in A (b ),
(v) x does n o t o ccu r free in A (b ) (unless x is b), an d (vi) A (x ) is (in fact)
th e result of s u b stitu tin g x for (the free occurrences of) b in A (b ). T o m a k e
th is e x a m p le p articu lar,
x, A (x ), b, A (b )
m a y b e re sp e c tiv e ly ,
c, 3c (c ' + a = b) D -\ a= b+ c, df 3c (c ' + a — b) ^ ~ ia= b + d.
§ 19. Transform ation rules. In th is sectio n we sh all in tro d u ce

fu rth er m e ta m a th e m a tic a l d efin itio n s (called deductive rules or trans
form ation rules) w h ich g iv e th e form al sy ste m th e stru ctu re o f a d e d u c tiv e
th e o ry . T o em ph asize th e a n a lo g y to an in form al d e d u c tiv e th e o ry , w e
sh all s ta rt w ith a list of 'p o stu la te s*; h o w ever, for th e m e ta m a th e m a tics,
th ese are n o t p o stu la tes in th e sense of assum ptions, as ind eed t h e y ca n n o t
b e w h e n o ffic ia lly t h e y h a v e no m ean in g, b u t o n ly form ulas a n d form s
(or sch em ata) to w h ich w e sh all refer w h en w e g iv e th e definitions.
§19 TRANSFORMATION RULES 81
B efo re g iv in g th e p o s tu la te list, le t us illu stra te th e ty p e s of p o stu la te s
w h ic h w ill ap p ear in th e list. T h e sim plest is an "axiom ', o f w h ic h - i a ' = 0
is an ex am p le. T h is is a fo rm u la of th e fo rm al sy ste m . T h e n w e m a y h a v e
an "axiom form ' or "axiom sch em a', of w h ic h ""B 3 A V B" is an ex a m p le .
T h is is a m e ta m a th e m a tic a l expression, w h ic h g iv e s a p a rticu la r a x io m
each tim e form ulas are sp ecified as represen ted b y th e m e ta m a th e m a tic a l
le tte rs ""A" a n d ""B". F o r ex am p le , w h en A is a!= 0 an d B is -ia'= 0,
we o b ta in th e a x io m - ia '= 0 D a '= 0 V The a x io m sch em a
is th u s a m e ta m a th e m a tic a l d e v ic e for sp e c ify in g an in fin ite class of
a x io m s h a v in g a com m on form .
W e m u st also h a v e an o th er k in d of p o stu la tes, w h ic h form alize th e
op eratio n s of d e d u cin g fu rth e r theorem s from th e axio m s. T h e se are th e
"rules o f inference', o f w h ich th e fo llo w in g is an e x a m p le :
A, A d B
B.
This is a sch em a co n ta in in g th r e e m e ta m a th e m a tic a l expressions ""A",

"A D B " an d ""B", w h ich represen t form ulas w h e n eve r form u las are
specified as represen ted b y th e m e ta m a th e m a tic a l le tters ""A" an d ""B".
T h e sense of th e rule is t h a t th e form u la represen ted b y th e expression
w ritte n b e lo w th e line m a y b e "inferred' from th e pair of form u las re
p resen ted b y th e tw o expressions w ritte n a b o v e th e line. F o r e x a m p le ,
b y ta k in g as A th e fo rm u la n a = 0 an d as B th e fo rm u la a’= 0 V
th e rule allow s th e inference from - i ^ '= 0 an d
—1^'==0 D a '= 0 V —ia '= 0 to a ’= 0 V S in ce - i a '= 0 an d
- i a '= 0 D a '= Q V are a x io m s (as w e ju s t saw ), a '= 0 V -i< z '= 0
is a fu rth e r "formal th eo rem '. (Our term in o lo g y w ill in clu d e th e a xio m s
as theorem s.)
W e sh all n o w d isp la y th e fu ll p o stu la te list, an d th e n g iv e th e d ef
in ition s e stab lish in g th e d e d u c tiv e stru ctu re of th e form al sy s te m b y
referring to th e list. T h e reader m a y v e r ify t h a t th e c u m u la tiv e e ffe ct
o f th e series o f defin itio n s w ill b e to define a subclass of th e class o f for
m u las ca lled "provable fo rm u las’ or "formal theorem s'.
P ostulates for the formal system
D ramatis personae . F o r P o stu la te s 1— 8 , A , B an d C are form ulas.

F o r P o s tu la te s 9— 13, x is a va ria b le , A (x ) is a form ula, C is a form u la
w h ich does n o t co n ta in x free, a n d t is a term w h ic h is free for x in A (x ).
G roup A. P o stu la te s for th e p red icate calculus.

G roup A l. P o stu la te s for th e proposition al calculus.
la. A D (B D A ). A, A D B
2.
lb . (A 3 B) 3 ((A D ( B 3 C)) 3 (A 3 C)). B.
3. A 3 (B 3 A & B ). 4a. A & B 3 A.
4b. A & B 3 B.
5a. A 3A V B. 6. (A 3 C) 3 ( (B 3 C)
5b. BDAVB. 3 (A V B 3 C)).
7. (A 3 B) 3 ((A 3 -.B ) 3 -.A ). 8 °. - i - i A 3 A.
G roup A2. (A dditional) P ostu lates for the predicate calculus.

C 3 A(x)
9. 10 . V xA (x) 3 A (t).
C 3 V x A (x ).
A (x) 3 C
11. A (t) 3 BxA(x). 12.
3xA (x) 3 C.
G roup B. (Additional) P ostulates for num ber theory.
13. A(0) & V x(A (x) 3 A (x ')) 3 A (x).
■ na'=0.
5s-
if
II
u
&
14. 15.
a = b 3 (a = c 3 b = c ).
if
II
u
&
16. 17.
18. d -\-0 = <2. 19. a + b '= (a+ by.

20 . a- 0 = 0 . 21, a-b'=a-b-{~a.
(The reason for w ritin g ''° '' on P o stu la te 8 w ill b e g iv e n in § 23.)
O n e m a y v e r ify th a t 14— 21 are form u las; an d th a t 1— 13 (or in th e
case of 2, 9 an d 12, th e expression (s) a b o ve , an d th e expression belo w , th e
line) are form ulas, for each ch oice of th e A , B , C, or x , A (x ), C, t, su b je ct
to th e stip u latio n s g iv e n a t th e h ead of th e p o stu la te list.
T h e class of 'a x io m s' is d efin ed thus. A form ula is an axiom , if it has
one of th e form s la , lb , 3— 8 , 10, 11, 13 or if it is one of th e form ulas
14— 2 1.
T h e relatio n of 'im m e d ia te con sequence' is d efined thus. A form ula is
an im m ediate consequence of one or tw o other form ulas, if it has th e form
show n b elo w th e line, w h ile th e other(s) h a v e th e form(s) show n a b o v e
th e line, in 2, 9 or 12.
T h is is th e b asic m e ta m a th e m a tic a l d efin ition corresponding to
P o stu la te s 2, 9 and 12, b u t w e shall re sta te it w ith ad d itio n al term in o lo gy
§19 TRANSFORMATION RULES 83
w h ich draw s a tte n tio n to th e process of a p p ly in g th e d e fin e ion. P o stu la te s
2, 9 a n d 12 w e ca ll th e rules of inference. F o r a n y (fixed) choice of th e A
an d B , or th e x, A (x) an d C, su b je ct to th e stipu lation s, th e form ula(s)
show n a b o v e th e line is the prem ise (are th e first and second prem ise,
re sp ectively), an d th e form ula show n b elo w the line is th e conclusion,
for th eapplication of th e rule (or th e (formal) inference b y th e ru le). T h e
conclusion is an im m ediate consequence of the premise(s) (by th e ru le).
C arn ap 1934 brings th e tw o kinds of p o stu lates under th e com m on
term Tran sform ation rules’ , b y considering th e axio m s as th e result of
tran sform ation from zero premises.
T h e definition of a ‘ (form ally) p ro v ab le fo rm u la ’ or ‘ (formal) th eo rem ’
can now be g iv e n in d u c tiv e ly as follows.
1. I f D is an axiom , then D is provable. 2. If E is provable, an d D is an
im m edia te consequence of E , th en D is provable. 3. If E and F are provable,
an d D is an im m ed iate consequence of E an d F , th en D is provable. 4. A
form ula is provable o n ly as required b y 1— 3.
T h e notion can also be reached b y using th e in term ed ia te co n cep t of
a '(form al) p ro o f’ , thus. A (formal) proof is a fin ite sequence of one or m ore
(occurrences^ of) form ulas such th a t each form ula of th e sequence is
either an axio m or an im m ed ia te consequence of preced ing form ulas of
th e sequence. A proof is said to b e a proof of its last form ula, an d this
form ula is said to be (formally’) provable or to be a (formal) theorem.
E x a m pl e 1 . T h e follow in g sequence of 17 form ulas is a proof of th e
form ula a = a . F o rm u la 1 is A x io m 16. F o rm u la 2 is an axiom , b y an
ap p licatio n of A x io m S ch em a la in w h ich th e A an d th e B of th e sch em a
are b o th 0 = 0 ; an d F o rm u la 3 b y an a p p licatio n in w h ich th e A is
a = b 3 ( a = c 3 b —c) and th e B is 0 = 0 3 ( 0 = 0 3 0 = 0 ) . F o rm u la 4 is
an im m ediate consequence of F o rm u las 1 an d 3, as first an d second
prem ise resp ectively, by an ap p lica tio n of R u le 2 in w h ich th e A
of th e rule is a = b 3 (a —b 3 b = c ) an d th e B is [0 = 0 3 (0 = 0 3 0 = 0)] 3
[ a = b 3 (a = c 3 b = c )]. F o rm u la 5 is an im m ed ia te con sequence of
F o rm u la 4, b y an a p p lica tio n of R u le 9 in w h ich th e x is c, the A (x) is
a = b 3 ( a = c 3 b = c ), an d th e C is 0 = 0 3 ( 0 = 0 3 0 = 0 ) (which, note,
does not co n ta in th e x free). F o rm u la 9 is an axio m b y an a p p lic a tio n
of A x io m S ch em a 10 , in w h ich th e x is a, th e A (x) is V bV c[a = b 3
( ^ = c 3 /?= < :)], an d th e t is a+0 (which, note, is free for the x in the
A (x)). T h e A (t), b y our su b stitu tio n n o ta tio n (§ 18), is th e result of su b
s titu tin g th e t for (the free occurrences of) th e x in the A (x ), i.e. here
th e A (t) is V A V c [ t f + 0 = i 3 ( * + 0 = c 3 £ = c)].
84 A FORMAL SYSTEM CH . IV
1. a=b 3 (a=c 3 b=c) — Axiom 16.

2. 0=0 3 (0 = 0 3 0 = 0 )— Axiom Schema la.
3. {a=b 3 (a=c 3 b=c)} 3 {[0=0 3 (0=0 3 0=0)] 3
[<z=/> 3 (d=c 3 i>=c)]} — Axiom Schema la.
4. [0=0 3 (0=0 3 0=0)] 3 [a=b 3 (<z=c 3 £=c)] — Rule 2, 1,3.
5. [0=0 3 (0=0 3 0=0)] 3 Vc[a=b 3 (<j=c 3 b=c)] — Rule 9, 4.
6. [0=0 3 (0 = 0 3 0=0)] 3 V^Vc[a=^ 3 (a=c 3 b=c)] — Rule 9, 5.
7. [0=0 3 (0=0 3 0=0)] 3 VaVbVc[a=b 3 (a= c 3 £ = c)] — Rule 9, 6.
8. 'i d i b'i c[a—b 3 (a=c 3 £=c)] — Rule 2, 2, 7.
9. VflV£Vc[a=£ 3 (<*=C 3 b=c)] 3 V^Vc [a+0=£ 3 (a+0=c 3 £=c)]
— Axiom Schema 10.
10. 'lb'4c[a+0=b'D{a+0=c'Db=c)} — Rule 2, 8, 9.
11 . VbVc[a+0=b 3 (a + 0 = c 3 b=c)] 3 Vc [j + 0 = a 3 (<j+0=c3<i=c)]
— Axiom Schema 10.
12. Vc[<J-f-0=tf 3 (<j+0=c 3 a = c ) ] — Rule 2, 10, 11.
13. V c[«+ 0= j3(tf+ 0= c3d = c)] 3[<7+0=tf3(<j+0=d3<7=d)] —
Axiom Schema 10.
14. a+0=aZ>{a+0=aZ>a=a) — Rule 2, 12, 13.
15. a+0=a — Axiom 18.
16. d + 0= a 3a = tf — Rule 2, 15, 14.
17. a=a — Rule 2, 15, 16.
E xample 2. Let A be any formula. Then the following sequence of

five formulas is a proof of the formula A 3 A. (In other words, what we
exhibit below is a 'proof schema’, which becomes a particular proof on
substituting any particular formula, such as 0=0, for the metamathe-
matical letter “A” ; and its last expression "A 3 A” is accordingly a
‘theorem schema’.) Formula 1 is an axiom, by an application of Axiom
Schema la in which the A and the B of the schema are the A of this
example. Formula 2 is an axiom, by an application of Axiom Schema lb
in which the A and the C of the schema are the A of this example, and
the B of the schema is the A 3 A of this example. Formula 3 is an imme
diate consequence of Formulas 1 and 2, as first and second premise,
respectively, by an application of Rule 2 in which the A of the rule
is the A 3 (A 3 A) of this example, and the B of the rule is the
[A 3 ((A 3 A) 3 A)] 3 [A 3 A] of this example.
§ 19 TRANSFORMATION RULES 85
1. A d (A d A) — A x io m Sch em a 1 a.
2. {A 3 (A 3 A ) } D {[A D ((A 3 A ) D A)] D [ A 3 A ]} — A x io m S ch e-
m a lb .
[ } 3. [A D ( ( A D A ) D A ) ] D [A D A ] — R u le 2, 1, 2.
4. A D ((A D A ) D A ) — A x io m S ch em a la .
5. A D A — R u le 2, 4, 3.
T h e term s proof, theorem , etc. as d efin ed for th e form al sy ste m (i.e.

form al proof, form al theorem , etc.) m u st b e sh a rp ly d istin gu ish ed from
these term s in th eir ord in ary inform al senses, w h ich w e em p lo y in pre
sen tin g th e m eta m a th em a tics. A form al theorem is a form ula (i.e. a cer
ta in k in d o f fin ite sequence of m arks), an d its form al proof is a certain
kin d o f fin ite sequence of form ulas. A m e ta m a th e m a tic a l theorem is a
m ean in gfu l sta te m e n t a b o u t th e form al o b jects, an d its proof is an
in tu itiv e d em o n stratio n of th e tr u th of th a t statem e n t.
W e m en tio n ed three categories of form al o b je cts (§ 16), b u t w e shall
b e free to in tro d u ce others in th e s tu d y of th em , so lo n g as th e tre a tm e n t
is fin ita ry . B esid es this, a so m ew h at different exten sio n of our su b je c t
m a tte r occurs w h en w e discuss th e form of our m e ta m a th e m a tic a l d e f
inition s a n d theorem s in turn. I f w e chose to b e m eticu lo u s in our w a y
o f d o in g this, it w o u ld c o n stitu te a m eta m e ta m a th e m a tic s. H o w ev er, th e
sam e p ra c tic e is com m on in (other bran ch es of) inform al m a th e m a tic s ;
an d w e sh all regard su ch discussions as in cid en tal exp la n atio n s, in ten d ed
som etim es to m a k e it easier to grasp q u ic k ly w h a t is b ein g done in th e
m e ta m a th e m a tics, an d som etim es to enable us to condense th e sta te m e n t
of m e ta m a th e m a tic a l theorem s w h ich cou ld b e s ta te d w ith o u t th em .
Chapter V
F O R M A L D E D U C T IO N
§ 20. F o r m a l d e d u c t io n . F o r m a l proofs of even q u ite ele m e n ta ry

theorem s te n d to b e long. A s a price for h a v in g a n a ly z e d lo gica l d e
d u ctio n in to sim ple steps, m ore of those steps h a v e to b e used.
T h e purpose o f fo rm alizin g a th e o ry is to g e t an e x p lic it d efin itio n of
w h a t co n stitu te s proof in th e th eo ry. H a v in g ac h ie v e d this, there is no
need a lw a y s to a p p e a l d ir e c tly to th e d efinition. T h e lab o r required to
estab lish th e form al p r o v a b ility of form ulas can b e g r e a tly lessened b y
u sin g m e ta m a th e m a tic a l theorem s con cerning th e e x iste n c e of form al
proofs. I f th e dem o n stratio n s of those theorem s do h a v e th e fin ita r y
ch aracter w h ich m e ta m a th e m a tic s is su pposed to h a v e , th e d em o n stratio n s
w ill in d icate, a t least im p lic itly , m eth o d s for o b ta in in g th e form al proofs.
T h e use of th e m e ta m a th e m a tic a l theorem s th e n am o u n ts to a b b r e v ia
tio n , o fte n of v e r y gre a t e x te n t, in th e p resen tatio n of th e form al proofs.
T h e sim pler of such m e ta m a th e m a tic a l theorem s w e sh all c a ll derived
rules , since t h e y express prin ciples w h ich can b e said to b e d eriv ed from
th e p o stu la te d rules b y sh ow in g th a t th e use of th e m as a d d itio n a l
m eth o d s of inference does n o t increase th e class o f p ro v a b le form ulas.
W e sh a ll seek b y m ean s o f d eriv ed rules to b rin g th e m eth o d s for es
ta b lish in g th e fa c ts o f form al p r o v a b ility as close as possible to th e in
form al m eth o d s o f th e th e o r y w h ic h is b ein g form alized.
In s e ttin g u p th e fo rm al sy ste m , proof w as g iv e n th e sim plest possible
stru ctu re, co n sistin g o f a sin gle sequence of form ulas. Som e of our d erived
rules, ca lled 'd irect ru les,, w ill serve to a b b re v ia te for us w h ole segm en ts
of such a seq u en ce; w e can th en , so to speak, use these segm en ts as
p re fa b rica te d u n its in b u ild in g proofs.
B u t also, in m a th e m a tic a l p ractice, proofs are com m on wTh ich h a v e
a m ore co m p lic a te d stru ctu re, em p lo y in g 'su b sid ia ry d ed u ctio n ', i.e.
d e d u ctio n u n der assu m p tio n s for th e sake of th e a rgu m en t, w h ich as
su m p tio n s are su b se q u e n tly d isch arged. F o r exam p le, su b sid iary d e
d u ctio n is used in a p roof b y red u ctio ad ab surdum , an d less o b tr u siv e ly
w hen w e place th e h y p o th e sis of a theorem on a p ar w ith p ro v ed propo-
86
§20 FORMAL DEDUCTION 87
sitions to d educe th e conclusion. O th e r d erived rules, ca lled 'su b sid ia ry
d ed u ctio n ru les’ , w ill g iv e us th is k in d of procedure.
W e n ow in trodu ce, b y a m e ta m a th e m a tic a l d efin ition , th e n o tio n
of 'fo rm al d e d u c ib ility u n der assu m p tio n s’ . G iv e n a list D 1, . . . , D J
(l ^ 0) of (occurrences of) form ulas, a fin ite sequence of one or m ore
(occurrences of) form ulas is ca lled a (formal) deduction from th e as
sum ption form ulas D v . . . , D z, if each form ula of th e seq uence is eith er
one of th e form ulas D v . . . , D ,, or an ax io m , or an im m e d ia te con sequen ce
of p reced in g form ulas of th e sequence. A d ed u ctio n is said to b e a d e
d u ctio n of its la st form u la E ; an d th is form u la is said to b e deducible
from th e assu m p tion form u las (in sym b o ls, D 1, . . . , D z \- E ), a n d is c a lle d
th e conclusion (or endformula) of th e d ed u ctio n . (T h e sy m b o l " h” m a y
b e read " y ie ld s ” .)
T h e definitions of d ed u ctio n an d of d e d u c ib ility are gen eraliza tio n s o f
those of proof an d of p r o v a b ility (w hich t h e y in clu d e as th e case for
l= 0) to p erm it th e use o f a n y form ulas D v . .., D , w e please, ca lled as
su m p tio n form ulas for th e d ed u ctio n , as pro tem pore on a par w ith th e
axiom s.
E xample 1. L e t A , B an d C b e form ulas. T h e n th e fo llo w in g se

quen ce of fiv e form ulas is a d ed u ctio n of C from th e three assu m p tio n
form ulas A D (B D C), B an d A . (W e e x h ib it a 'd e d u c tio n sch em a'.)
1. B — second assu m p tio n form ula.
2. A — th ird assu m p tio n form ula.
(2) 3. A D (B D C ) — first assu m p tio n form ula.
4. B D C — R u le 2, 2, 3.
5. C — R u le 2, 1 , 4.
E xample 2. Let th e reader c o n s tr u c t: (3) a d e d u ctio n of A & B

from A an d B ; (4) a d e d u ctio n o f C from A & B D C, A , B .
B y an analysis o f a d ed u ctio n or proof A v , . . , A *, w e m ean a sp ec

ifica tio n , for each j (j — !,...,&), either th a t A j is one of th e assu m p tio n
form ulas an d w h ich one in th e list D x, . . . , D z, or th a t A s is an a x io m a n d
b y w h ich a x io m sch em a or p a rticu la r a x io m of th e p o stu la te list, or th a t
Aj is an im m e d ia te con sequence of p reced in g form ulas an d b y w h ic h rule
of inference an d of w h ich p reced in g form ulas as th e re sp e ctive prem ises
of th a t rule. In brief, an a n a ly sis o f a d e d u c tio n con sists o f th e e x p la
n atio n s e m p lo y e d to ju s t if y each occurrence of a fo rm u la in it (i.e. in our
exam p le s, th e e x p la n a tio n s g iv e n a t th e rig h t of th e form ulas).
I t m a y o cc a sio n a lly h ap p e n th a t an occurrence of a form u la in a d e
d u ctio n (or proof) can b e ju stifie d in m ore th a n one w a y , e.g. th e form ulas
88 FORMAL DEDUCTION CH. V
A , B an d C m ig h t b e such th a t one of th e fiv e form ulas in (2) is an axiom .
C o n se q u e n tly , for som e of th e discussions belo w , th e procedure to be
a p p lie d to a g iv e n d e d u ctio n is o n ly d eterm ined u n iq u ely, w hen along
w ith th e d ed u ctio n itself there is g iv e n a p articu lar an alysis of it.
I t is to be em p h asized th a t th e expression .. , D l b E ” , w h ich
w e use to sta te b rie fly t h a t E is d ed u cib le from Dv ..., D lt is n o t a form ula
of th e sy ste m , b u t a b rief w a y of w ritin g a m e ta m a th e m a tic a l sta te m e n t
about th e form ulas D v . . ., D lf E , n a m e ly th e sta te m e n t th a t there
ex ists a certain k in d of a fin ite sequence of form ulas. W h e n 1 = 0, th e
n o ta tio n b ecom es “ b E ” , m ean in g th a t E is p ro v a b le . The sy m b o l
“ b” goes b a c k to F re g e 18 7 9 ; th e present use of it to R osser 19 3 5 * and
K le en e 19 3 4 *.
E xample 3 . T h e fo llo w in g tw o sta te m e n ts (!') an d (2') h a v e been

ju stifie d b y e x h ib itin g a b o v e th e tw o d ed u ctio n s (1) an d (2), r e s p e c tiv e ly ;
3 4
an d ( ') an d ( ') b y E x a m p le 2.
1
( ') b A D A . (2') A D (B D C), B , A b C.
3
( ') A, B b A & B. (4') A & B D C, A , B b C.
N o tic e th a t th e s y m b o l “ b” appears in c o n te x t p reced ed b y a fin ite

sequence of zero or m ore form ulas an d follow ed b y a single form u la (or
in stea d of form ulas, m e ta m a th e m a tic a l le tters or expressions represen tin g
form ulas). T h is m akes u n a m b igu o u s th e scope of an occurrence of th e
sy m b o l “ b” in a m e ta m a th e m a tic a l sentence. In p articu lar, th e scopes
of th e form al operators are necessarily con fined w ith in form ulas of th e
sy ste m , w h ile “ b” is a m e ta m a th e m a tic a l v e rb ly in g ou tsid e any
fo rm u la o f th e sy stem .
T h e defin itio n of ‘d ed u cib le from D x, . . . , D / can also b e s ta te d w ith o u t
u sin g th e in term ed ia te co n cep t of a d ed u ctio n (cf. th e first d efin itio n of
‘p r o v a b le ’ in § 19 ). W e le a v e it to th e reader to s ta te th e fiv e clauses
required. B r ie fly , “ D 1, . . D z b E ” th en m eans t h a t it is possible to g e t
from (zero or m ore of) th e form ulas Dv ..., D t an d (zero or more) axio m s
b y th e rules of inference to th e form ula E . T h e tw o versions o f th e d e f
in itio n are b ro u g h t in to agreem en t b y o b servin g th a t, w hen th e form ulas
considered in th e process of g e ttin g from D lf.. ., D* an d axio m s to E are
p u t d o w n in order o f first con sid eration , w e h a v e a d ed u ctio n of E from
D x, . . . , Di*
W e sh all use G reek c a p ita l letters, such as ‘ T ” , “ A ” , “ 0 ” , etc., to sta n d
for fin ite sequences of zero or m ore (occurrences of) form ulas, w h en w e
w ish to in d ic a te sets of assu m p tio n form ulas w ith o u t n a m in g th e form ulas
20 FORMAL DEDUCTION tv
ndividually (or sometimes seV(x)’\“ A(.\i f . . , :,yhy etv , whcn we wish
'..Iso to emphasize certain variables which mac -mr hi teem).
From the definition of the deducibility lelmm there follow se\ eraL
:neral properties of b which can be seen to be - c e u Edmo* icteiwice to
vhat particular postulates are in the pcwtuiatv led wr the formal ^yuem.
; F r- E when E is in the list lb (ii) If F w -hen A, I' f E for any A.
A. particular, we can regard any piovabh- fominE as dedu< ihie iiom
nv assumption formulas we please. (iiit :C b b, i he n A j- E where A
ones from E by permuting the formula.- F w »11•?tin^ any wliieh are
midicates of others remaining. (i\) li E • b. do-r A ■ E when* A eumt s
wen F by omitting any of the formulas E w inch arc pi«»v able or <J<.d?u tide
••in tliose remaining. For, given a deduction ot if iu.nu V v.e cm: obtain
■ is from A by inserting into the giw-n d-Nuo; urn. in -hire of each no
•' nrence of an assumption formula which we wbb m suppress, a do-
diction of the same from the remaining as mm Aion tumn-iis, Idiese four
: coral properties can be analyzed info tire w n p E r ciiw of the following
wmma. However (while the inferences we wade by general properties of
can actually all be made from (i) — (iv) 01 (i; (\ }). the reader is en
couraged to reason flexibly with b on the basis of its meauiiig
L emma 5 . (I) E h E. (II) I f F b E, t h e n Ch V | E. bl11) I f C, C, P f- M.
then C, F b E. (IV) If A, D, C, F h E, then A, k\ h F r Mb *(V) li A j- C
anl C, F b E, then A, T b E. (After (jentzen 1^34-5.)
Example 4. If A b B and A, B, C h J> and B ,l) ; f., then A, E b h -
ihe reader m ay convince himself of this directly bom die meaning of
- (under both versions of its definition), and aim ctriiy that it follows
e -in (i) - - (iv) and from (I) --- (V).
The definition which w e have given for h i> reia b\ t to a paitinnyr
loimal system as determined by a postulate list. Speiifwady, it is relative
noth to the part of the postulate list which determines +he axioms, and
to the rales of inference. Thus far we have been wmet rued only with the
one formal system, but we shall make use of o in Jikt seme in ewnnecuon
with other formal systems, e.g. subsystems of the t one obtained by con-
sideling only part of the postulate list to be m force foi determining the
olass of axioms and the relation of immediate _c?iscquenee We shall
always understand b to be relative to the formal so stem we are studying
.• ' :' • gi\ ci Erne.
Notice that A, F b E for a given system is equivalent to A -r t lot
. stem resulting from the gown one \>y adding thr formulas 1' to the
cat oi axioms.
§ 21. T h e d e d u c tio n th e o r e m . We sh all consider th e fo llo w in g
th eo rem first for th e p ro p o sitio n al calculus, i.e. w ith o n ly th e p o stu la tes
o f G ro u p A 1 in force.
T heorem 1. F or the propositional calculus, if T , A h B , then T b A D B.

(T h e d e d u ctio n theorem .)
P roof. The hypothesis of th e theorem sa y s t h a t th ere is a fin ite

seq u en ce of form ulas su ch t h a t : each form u la of th e sequence is eith er
(a) one o f th e form ulas F , (b) th e fo rm u la A , (c) an axio m , or (d) an im
m e d ia te con sequence b y R u le 2 of tw o p reced in g form ulas (since R u le 2
is th e o n ly rule of inference h e re ); an d th e la st fo rm u la of th e sequence
is th e fo rm u la B . T h is seq uence w e sh all ca ll th e 'g iv e n d e d u c tio n 1 o f B
from T , A .
T h e con clusion o f th e theorem sa y s t h a t th ere is a fin ite sequence of
fo rm u las such t h a t : each form u la o f th e sequence is eith er (a) one o f th e
fo rm u las T , (c) an axio m , or (d) an im m e d ia te con sequence b y R u le 2 o f
tw o p reced in g form u las; an d th e la st form u la o f th e seq uence is th e
r.
fo rm u la A D B . T h is sequence w e sh all c a ll th e 'resu ltin g deduction* of
A D B from
T h e th eorem w ill b e p ro v e d b y a co u rse-o f-va lu es in d u ctio n on th e
le n g th k of th e g iv e n d e d u ctio n (§ 7), ta k in g th e B of th e th eorem to b e
v a ria b le , b u t th e T , A fix e d for th e in d u ction .
P (k) o r P ( F , A , k) is: F or every form ula B ,
T h e in d u ctio n proposition
if there is given a deduction of B from T, A of length k, then there can be found
a deduction o / A D B from F.
B asis (to p ro v e th e p roposition for k = 1, i.e. to p ro v e P ( F , A , 1)).
S u p p o se g iv e n a fo rm u la B an d a d ed u ctio n of B from F , A of le n g th 1.
W e d istin gu ish th ree cases, acco rd in g to w h ich o f th e possibilities (a)— (c)
ap p lies to th e la st (and since k = 1, only) form u la B of th e g iv e n d e d u c
tio n . T h e p o ssib ility (d) is e x clu d e d here, since B is th e o n ly form ula.
F o r each case, w e sh ow h o w to co n stru ct th e resu ltin g d ed u ctio n ,
le a v in g it to t h e reader to v e r ify th a t th e sequence o f form ulas w h ich
w e su b m it as such does h a v e th e required features.
Case (a ): B is one o f th e form ulas F . T h e n th e fo llo w in g seq uen ce o f

form u las is th e re su ltin g d ed u ctio n .
1. B — one o f th e form ulas T.

2. BD(ADB) — A x io m S ch em a la .
3. A D B — R u le 2, 1, 2.
§21 THE DEDUCTION THEOREM 91
Case (b): B is A. The resulting deduction is the sequence of formulas

(1) which was given in § 19 Example 2 as a proof of A DA. Since B is A,
the formula ADA is A D B.
Case (c) : B is an axiom. The resulting deduction is the same as in
Case (a), except that now the first step is justified on the ground that
B is an axiom.
I nduction step . Assume (as hypothesis of the induction) that, for
eviry l ^ k, P(F, A, l ); i.e. that for every l < k and every B, if there is
given a deduction of B from T, A of length l , then there can be found a
deduction of A D B from F. Now (to prove P(r, A, k + 1)) suppose
given a'formula B and a deduction of B from T, A of length k + 1. We
distinguish four cases, according to which of the possibilities (a) — (d)
applies to the last formula B of the given deduction. The treatment of
Cases (a) — (c) is the same as under the basis.
Case (d): B is an immediate consequence by Rule 2 of two preceding
formulas. By the statement of Rule 2, we may call these two formulas
P and P D B. (We use the letter P now, instead of A as in the statement
o f the rulet since A is reserved here to designate the last assumption
formula for the given deduction.) If we discard the part of the given
deduction below the formula P, the part remaining will be a deduction
of P from T, A of length l <> k. By the hypothesis of the induction
(with P as its B), we can hence find a deduction of A D P from T. Likewise,
applying the hypothesis of the induction to the part of the given de
duction down to PD B inclusive, we obtain a deduction of A D(PDB)
from r. We use these two deductions (say they are of lengths p and qt
respectively) in constructing the resulting deduction, as follows.
deduction of A 3 P from V, given by the
A DP hypothesis of the induction.
p-
deduction of A D (PD B) from P, given by
P+q- A D (PD B ) the hypothesis of the induction.
p + q + \. (A d P )3 ((A 3 (P 3 B ))3 (A 3 B )) — Axiom Schema lb.
P +q+2. (A D (P D B)) D (A D B) — Rule 2, p, P + q + 1.
p+ q+ 3. A IDB — Rule 2, P+q$ p+ q + 2 .
This completes the proof of the theorem by mathematical induction.
The theorem includes the case that T is empty: For the propositional
, if A h B, then |- ADB.
c a lc u lu s
E x a m ple. W e w ere able to sta te, as (2') a b o ve , th a t A d (B d C), B ,

A h C. B y T h eorem 1, w e ca n th en ce infer th a t A D (B D C ) , B h A D C .
T o ex am in e th is ex a m p le m ore closely, le t us ta k e th e d ed u ctio n (2),
w h ich w e e x h ib ite d in ju stific a tio n of (2'), as th e g iv e n d ed u ctio n of C
from A D (B D C ) , B , A . B y referring to th e proof of T h eorem 1, w e
sh ou ld b e able to fin d th e resu ltin g d ed u ctio n of A D C from A D (B d C ) ,B .
Sin ce th e g iv e n d ed u ctio n is of len gth > 1, an d th e la st form u la com es
from p reced in g form ulas b y an a p p lica tio n of R u le 2, th e case w h ich
ap p lies is C ase (d) und er th e in d u ctio n step. T h ere w e fin d som e d eta ils,
an d in stru ctio n s to find th e rest b y a p p ly in g T h eorem 1 to th e d ed u ctio n s
1 an d 1 — 4 occurring as p arts of (2). C o n tin u in g in th is m anner, w e
e v e n tu a lly o b ta in th e follow in g as th e resulting d ed u ction .
1. B — second assu m ption form ula.

2. B D (A D B) — A x io m S ch em a la .
3. A D B — R u le 2, 1, 2.
4. A D (A D A ) — - A x io m S ch em a la .
5. {A D (A D A)} D {[A D ((A D A) D A )] D [A D A ]} — A x io m S ch e
ma lb .
6. [A D ((A D A ) D A ) ] D [A D A ] — R u le 2, 4, 5.
7. A D ((A D A ) D A ) — A x io m S ch em a la .
8 . A D A — Rule 2, 7, 6.
(5) 9. A D (B D C) — first assu m p tion form ula.
10. {A D (B D C)} D {A D (A D (B D C))} — A x io m S ch em a la .
11. AD(AD(BDC)) — R u le 2, 9, 10.
12. {A D A } D {[A D (A D (B D C))] D [A D (B D C)]} — A x io m S c h e
m a lb .
13. [A D (A D (B D C))] D [A D (B D C)] — R u le 2, 8, 12.
14. AD(BDC) — R u le 2, 11 , 13.
15. (A D B ) D ((A D (B D C)) D (A D C)) — A x io m S ch em a lb.
16. (A D (B D O ) D (A D C) — R u le 2, 3, 15.
17. A DC — R u le 2, 14, 16.
T h e d ed u ctio n (5) is n o t th e o n ly d ed u ctio n o f A D C f r o m A D (B D C), B .

I t h ap p e n s th a t there is a shorter one, w h ich w e o b ta in from (5) b y o m it
tin g F o rm u la s 4— 8 an d 10— 14, an d citin g 9 (instead o f 14) as first
prem ise for th e inference b y R u le 2 a t S te p 17.
B u t (5) is th e p a rticu la r one w h ich results b y th e m e th o d u sed in
p ro v in g T h eo rem 1 w hen (2) is ta k e n as g iv e n d ed u ction . W e h a v e
carried through th e exercise of fin d in g (5) to em phasize th e fin ita r y
ch aracter of th e reasoning used in proof of T h eo rem 1, an d in p a rticu la r
§21 THE DEDUCTION THEOREM 93
to show w h a t is in v o lv e d in th e use of m a th e m a tic a l in d u ctio n . H e n ce
fo rth w e sh all be satisfied to k n o w th a t th e re su ltin g d ed u ctio n s e x ist an d
co u ld b e found.
T h e proof of T h eo rem 1 w ill serve as a m odel for m e ta m a th e m a tic a l
proofs of certain ty p e s. In th e fu tu re w e sh all o fte n g iv e such proofs
in a m ore a b b re v ia te d w a y , w h en th e reader co u ld ca st th e a rgu m e n t in to
e x p lic it a p p licatio n s of in d u ctio n . A few proofs w ill b e set u p fu lly e x
p lic itly as m odels.
T h e a b o v e proof of T h eo rem 1 can b e g iv e n in a m ore a b b r e v ia te d
w a y as follows. T o each form u la of th e g iv e n d ed u ctio n of B from T , A ,
let A D be p refixed. (In th e ex am p le , from F o rm u las 1, 2, 3, 4, 5 o f (2),
w e th u s o b ta in F o rm u la s 3, 8, 11, 14, 17 of (5).) T h e resu ltin g seq uence
of form ulas (w ith A D B as end-form ula) is n o t (in general) a d ed u ctio n
from r, b u t ca n b e m ad e one b y in sertin g ad d itio n a l form ulas in th e
m anner in d ic a te d in th e tre a tm e n t of th e cases. (Th is sim ple p la n of p roof
w ill b e m odified slig h tly , w h en w e com e to e x te n d th e th eorem to th e
p red icate ca lcu lu s in § 22.)
F ro m A d (B d C), B b A D C w e infer b y a second a p p lica tio n o f

T h eorem 1 th a t A D ( B D C ) hBD(ADC). A co n ven ien t arran gem en t
of these inferences is th e follow ing.
1. AD(BDC), B, A b C — (2).
(S') 2. A D (B D C ) , B b A DC — T h eo rem 1, 1.
3. AD(BDC) bBD (AD C) - T h eo rem 1, 2.
In this presen tation , w e h a v e a sequence of expressions an alo go u s to, b u t

on a differen t le v e l from , th e sequence of form ulas w h ich c o n stitu te s a
form al proof or d ed u ction . T h e expressions in th is seq uence are m e ta
m a th e m a tica l sta te m e n ts a b o u t th e form al sy ste m , w h ile in a fo rm al proof
or d ed u ctio n th e y are form ulas of th e sy stem .
A n o th e r exam p le of a d ed u ctio n an d of a series of m e ta m a th e m a tic a l
sta tem e n ts follow s.
1. A & B — second assu m p tio n form ula.

2. A & B DA — A x io m S c h em a 4a.
3. A — R u le 2, 1, 2.
( 6 4.
) A D (B D C) — first assu m p tion form ula.
5. B D C — R u le 2, 3, 4.
6. A & B D B — A x io m S ch em a 4b.
7. B — R u le 2, 1, 6.
8. C — R u le 2, 7, 5.
1. A D ( B D C ) , A & B b C — (6 ).
' 2. AD(BDC) h A & B d C — T h eorem 1, 1.
A s fu rth er exam p les, th e reader m a y e s ta b lis h :
(7') A & B DC b A 3 (B D C ) — cf. (4') § 20.

1. ADB, B DC h A DC.
^ 2. A D B b (B D C ) D (A D C ) — T h eorem 1, 1.
§ 22. T h e d e d u c t i o n t h e o r e m ( c o n c lu d e d ) . T h eo rem 1 is a d e riv ed

rule of th e subsidiary deduction t y p e (cf. § 20). F o r an a p p lica tio n of th e
rule, th e g iv e n d ed u ctio n of B from T, A is th e subsidiary deduction ; an d
th e d ed u ctio n of A D B from V o b ta in ed from th e g iv e n d ed u ctio n b y
th e m eth o d in d ic a te d in th e proof of th e theorem w e h a v e called th e
resulting deduction. W h e n w e are sta tin g th e existen ce of d ed u ctio n s
w ith o u t a c tu a lly e x h ib itin g th em , w e m a y a d o p t an ellip tical p h raseology,
sp eak in g for ex a m p le of ‘ T h e d ed u ctio n T, A (- B ” , w hen w e m ean th e
d ed u ctio n w h ich th e s ta te m e n t ‘ T , A f- B ” asserts to exist.
In T h eorem 1 , th e last assu m p tion form ula A of th e su b sid iary de
d u ctio n r, A (- B is not used in m a k in g u p th e list of assu m p tion form ulas
for th e resu ltin g d ed u ctio n F A D B ; a c co rd in g ly w e s a y th is (occurrence
of A as) assu m p tio n form ula of th e su b sid iary d ed u ctio n is discharged.
(There m ig h t also b e occurrences of A in th e list T, w h ich w ould n o t b e
discharged.)
In general, a subsidiary deduction rule is a m e ta m a th e m a tic a l theorem
w h ich has one or m ore h y p o th ese s of th e form A,- (- E* called th e sub
sidiary deductions , an d a conclusion of th e form A (- E called th e resulting
deduction. F ro m each of th e su b sid iary d eductions, one or m ore as
su m p tio n form ulas m a y b e discharged.
E x a m p l e 1. T h e rule “ / / T, A h C and F , B f- C, then T, A V B h C” ,
w h ich w ill b e estab lish ed in th e n e x t section, has tw o su b sid iary d e
d uctions, F , A j- C an d T , B |- C, from th e first of w h ich th e last as
su m p tio n form ula A is d isch arged, an d from th e second th e B .
A m e ta m a th e m a tic a l theorem of th e sim ple form A h E is a d erived

rule of the direct ty p e . I t sa y s th a t it is possible to proceed from th e
form ulas A an d th e ax io m s d ire c tly to E b y ap p licatio n s of th e rules
of inference.
T h ere is th e follow in g im p o rta n t difference b etw een these tw o kin d s
of d erived rules. A direct rule necessarily rem ains true w hen th e form al
sy ste m is enlarged b y ad d in g new axiom s an d rules of inference, since
th e rule sta te s sim p ly th a t certain d eduction s can be co n stru cted , and
§22 THE DEDUCTION THEOREM (CONCLUDED) 95
th e n ew p o stu la tes o n ly ch an ge th e situ a tio n b y p ro v id in g a d d itio n a l
m eans of co n stru ctin g those sam e d edu ction s. B u t a su b sid iary d ed u ctio n
rule does n o t n ecessarily rem ain tru e w h en new p o stu la te s are ad d ed ,
since th e en largem en t of th e sy ste m ten d s to create n ew in stan ces of th e
su b sid iary d ed u ctio n s, an d it beco m es a q u estio n w h e th er resu ltin g d e
d u ctio n s e x ist to correspond to these new su b sid iary d ed u ction s. M ost
of th e su b sid iary d ed u ctio n rules w h ich w e sh all s ta te (in p articu lar, all
those of th e present ch apter) have an am b igu o u s set of assu m p tion
form ulas T before th e sy m b o l “ b ” th ro u gh o u t, so th a t th e a d d itio n of
new a x io m s can cause no trouble. B u t th e a d d itio n of new rules of in
ference w ill create new cases to b e considered in th e proof of th e rule.
We sh all n e x t tre a t T h eo rem 1 under th e co n d itio n th a t all th e
p o stu lates of G ro u p A are in force, either e x a c tly these, or also th e ones of
G roup B (w hich are o n ly an ax io m sch em a an d axiom s). A ce rta in
restriction w ill b e required in order to h an d le th e new cases in th e proof.
I t seem s easier to g iv e th is tre a tm e n t now , w h ile th e proof of T h eo rem
1 for th e p rop osition al ca lcu lu s is fre s h ; b u t som e readers m a y prefer to
p ostpo n e th e rem ainder of th is ch ap ter, e x c e p tin g th e p a rts of § 23
referring to th e p rop osition al calcu lu s, u n til after C h a p te r V I .
W e b egin b y s ta tin g som e d efin ition s w h ich are u sefu l in fo rm u la tin g
th e restriction. G iv e n a d ed u ctio n Av . . . , A fc from assu m p tio n form ulas
D Xl. . . , D i an d a p a rticu la r an a ly sis o f th e d ed u ctio n (§ 20), w e d efine
w hen (an occurrence of) a form u la A t in th e d ed u ctio n 'd epen d s' on a g iv e n
one D 5 of th e (occurrences of) assu m p tio n form ulas D x, . . .,D * , as follow s.
1. I f A i in th e g iv e n an alysis is D i# th e n A t- depends on D ,. 2. I f A {±
depends on D 3, an d A t in th e g iv e n a n a ly sis is an im m e d ia te con sequence
of A ix (or of A ix an d som e A tjs, in either order), th e n A * depends on D ,.
3. A t- depends on D § o n ly as required b y 1 an d 2 .
I t is e a s ily seen th a t A { d epends on D jf if a n d o n ly if there ex ists no
subsequence o f th e d ed u ctio n (not necessarily con secutive) w h ich under
the g iv e n an aly sis co n stitu te s a d ed u ctio n of A* from th e rem ain in g
assum ption form ulas D x, . . . , D ^ , D m , . . D {.
E xample 2. In th e d ed u ctio n (6 ), F o rm u las 4, 5 an d 8 d epen d on

th e first assu m ption form ula A D (B D C ) , an d th e other form ulas do not.
F orm u las 1 , 6 , 7 (w ith th e g iv e n analysis) co n stitu te a d ed u ctio n of 7
from th e oth er assu m p tion form u la A & B .
W e n ow s a y th a t a v a ria b le y is varied in a g iv e n d ed u ctio n (w ith a

giv en analysis) for a g iv e n assu m p tion form ula D ,, if (A) y occurs free in
D „ an d (B) th e d ed u ctio n co n ta in s an a p p lica tio n of R u le 9 or R u le 12
96 FORMAL DEDUCTION CH y
w ith respect to y (as th e x for th e a p p lica tio n of th e rule) to a fo rm u la

d ep en d in g on D , (as th e prem ise for th e a p p lica tio n o f th e rule). O th erw ise,
w e s a y th a t y is held constant in th e d ed u ctio n for th e a ssu m p tio n for
m u la D 3.
E xample 3. L e t x b e a va ria b le, A (x ) a form ula, an d b a v a ria b le ,

such th a t (i) b is free for x in A (x ) an d (ii) b does n o t occur free in A (x )
(unless b is x ) ; an d le t C b e a form u la n o t co n ta in in g b free. T h e n th e
fo llo w in g is a d ed u ctio n of C D V x A ( x ) from C D A (b ). In v e r ify in g t h a t
th e stip u la tio n s for P o stu la te s 9 an d 10 are m et, w e use th e fa c ts (iv) — (vi)
w h ich w ere w o rk ed o u t in E x a m p le 9 § 18.
1 . V b A (b ) D A ( x ) — A x io m S ch em a 10 (noting (iv) an d (vi)).

2. V b A (b ) D V x A (x ) — R u le 9, 1 (noting th a t, b y (v), x does n o t
occur free in V b A (b )).
3. CDA(b) — assu m p tio n form ula.
4. CDVbA(b) — R u le 9, 3.
6 . C 3 (V bA (b ) D V x A (x )) — from 2 as in C ase (a) for T h eo rem 1 § 2 1.
9. CDVxA(x) — from 4, 6 as in C ase (d) for T h eo rem 1 .
I f A (x ) co n tain s x free, th en in th is d ed u ctio n b is va ried , since (using

(i) an d (iii)) th e assu m p tio n form u la C D A (b) co n tain s b free, an d R u le
9 is ap p lied a t S te p 4 w ith respect to b to th e prem ise 3, w h ich d epends
on th e assu m p tio n form ula. B u t x is n o t va ried , since th e prem ise 1 for
th e a p p lica tio n of R u le 9 w ith respect to x a t S te p 2 does n o t d epen d on
th e assu m p tio n form ula.
In a g iv e n d ed u ctio n (w ith a g iv e n an alysis), a g iv e n v a ria b le y is

a lw a y s h eld co n sta n t for each assu m p tio n form ula in w h ich it does n o t
occu r free, w h ile it m a y b e v a rie d for som e of th e assu m p tio n form ulas
in w h ic h it occurs free an d h eld co n sta n t for others.
The a b o v e te rm in o lo g y su ggests itself, since R u le s 9 an d 12 (the
“ V -ru le ” an d th e “ 3 -ru le” ) are th e o n ly tw o p o stu la tes of G ro u p A in
w h ich a free v a ria b le p a rticip a te s as such. A x io m S ch em a 10, for exam p le,
can be a p p lied u sin g a free v a ria b le as th e t, b u t in th a t case th e va ria b le is
used in a w a y th a t a term n o t a v a ria b le (such as 0 ) can e q u a lly w ell b e
used. (T h e e m p lo y m e n t of free v a ria b les in s ta tin g th e p o stu la tes of
G ro u p B is inessential.)
T h e restriction on T h eo rem 1 for th e p red icate calcu lu s is th a t in th e
su b sid iary d ed u ctio n th e free v a riab les sh ould b e h eld co n sta n t for th e
assu m p tio n form u la to b e disch arged. (This w ill b e exp la in ed in term s
of th e in terp reta tio n in § 32.)
§22 THE DEDUCTION THEOREM (CONCLUDED) 97
T heorem 1 (concluded). F or the predicate calculus (or the full num ber-
theoretic form al system ), if T, A f- B w ith the free variables held constant
for the last assum ption form ula A , then T b A D B .
P ro o f is o b ta in e d from t h a t g iv e n in § 21 b y su p p ly in g th e tre a tm e n t of
th e tw o a d d itio n a l cases w h ic h can n ow arise u n d er th e in d u ctio n
step.
Case (e): B is an im m e d ia te con sequence of a p reced in g fo rm u la

b y an a p p lica tio n of R u le 9. B y th e sta te m e n t o f R u le 9, t h a t p reced in g
form u la is o f th e form C D A (x ), w here x is a v a ria b le , A (x ) a form ula,
an d C a form u la n o t co n ta in in g x free. T h e n B is C D V x A ( x ) . W e d is
tin gu ish tw o subcases, acco rd in g as in th e g iv e n d ed u ctio n (for a g iv e n
analysis) th a t preced in g form u la CDA(x) d epen d s on th e la st as
su m p tio n form ula A or not.
S ubcase ( e l ) : C D A (x ) d epend s on A . T h e n A does n o t co n ta in x free,

since otherw ise th e h yp o th e sis t h a t th e free v a ria b le s are h eld c o n sta n t
for A in th e g iv e n d ed u ctio n w o u ld b e co n trad icted . S in ce n o w n eith er
A nor C co n ta in s x free, th e form u la A & C does n o t co n ta in x free. T h is
fa c t is used b elo w in ju s tify in g th e new ap p lica tio n of R u le 9 a t S te p
p j r q jr \ t A p p ly in g th e h yp o th e sis of th e in d u ctio n to th e segm en t of
th e g iv e n d ed u ctio n en d in g w ith th e form ula C D A ( x ) , w e o b ta in a
d ed u ctio n of A D (C D A (x)) from T. T h is d ed u ctio n is in co rp o rated in
co n stru ctin g th e resu ltin g d ed u ctio n , as follow s.
d ed u ctio n of A D (C D A (x)) from T,

g iv e n b y th e h y p o th e sis of th e in
du ction .
p. AD(CDA(x)) d ed u ctio n of A & C D A (x ) from
A 3 (C D A (x )), g iv e n b y (6 '): 2 (end
p+q. A & C D A (x ) of § 2 1 ).
p + q + l. A & C D VxA(x) — R u le 9, p+q.
d ed u ctio n of A D (C D V x A (x )) from
A&CDVxA(x), g iv e n b y (7').
^ p + q + r + l. AD (CD VxA(x))
S ubcase (e2): C d A(x) (and hence CDVxA(x)) does not d ep en d
on A . T h e n som e subseq uence of th e g iv e n d ed u ctio n co n stitu te s a d e
d u ction of C D V x A (x ) from th e rem ain in g a ssu m p tio n form ulas I\
W e use th is in co n stru ctin g th e resu ltin g d ed u ctio n , as follow s.
d ed u ctio n of C 3 V x A ( x ) from T , g iv e n b y th e
h yp o th e sis o f independence.
p. C D V xA (x)
p - \- 1. ( C D V x A (x )) D (A D (C D V x A (x ))) — A x io m S c h em a la .
p+2. A D (C D V x A ( x ) ) — R u le 2, p, p + 1.
Ca s e (f): B is an im m e d ia te con sequence of a p reced in g form u la
b y an a p p lica tio n o f R u le 12. T h e tre a tm e n t of th is case is sim ilar, usin g
(S') :3 tw ic e in th e first subcase.
T h e d ed u ctio n th eo rem w as first p ro v ed as a d eriv ed rule b y H erb ran d

1930. (Cf. also H erb ran d 19 2 8 , T a rsk i 1930 , C h u rch 19 3 2 , H ilb e rt-
B e r n a y s 1934 p. 155, J a s k o w s k i 19 3 4 .)
§ 23. Introduction and elimination of logical symbols. The

fo llo w in g theorem co n ta in s a co llectio n of d erived rules, w ith row an d
colu m n d esign atio n s a tta c h e d to p ro vid e co n ven ien t d e scrip tiv e nam es
for th e rules. F o r exam p le , " V x A ( x ) f- A ( t ) " is th e rule of " g e n e r a lity
e lim in a tio n " or b rie fly " V -e lim in a tio n " .
T h e v a ria b le " x " w ritte n as su p erscript on th e sy m b o l " b ” in tw o
of th e rules is to m ark th e a p p lica tio n of R u le 9 or 12 w ith respect to x
in co n stru ctin g th e resu ltin g d ed u ction .
T heorem 2.F or the following rules , A , B and C , or x , A (x ), C and t,

are subject to the sam e stipulations as for the corresponding postulates (§ 19),
and T or T(x) is any list of form ulas.
F or the propositional calculus , the rules hold from " Im p lic a tio n " to
" N e g a tio n " , inclusive.
F or the predicate calculus {or the full number-theoretic system ), all the
rules hold , provided that in each subsidiary deduction the free variables are
held constant for the assum ption form ula to be discharged.
(Introd uction) (E lim in ation )
(Im plication ) I f T, A b B , A, A D B b B.
then T h A D B . (M odus ponens.)
(C onjunction) A , B b A & B. A & B b A.

A & B b B.
(D isjun ction ) A b A V B. I f T , A b C and T, B b C,

B b A V B. then T , A V B b C.
(Proof b y cases.)
§23 INTRODUCTION AND ELIMINATION RULES 99
(N egation) If T, A b B and T, A b " iB , - 1 - 1A b A .

r h -iA . (D isch arge of
(R ed u c tio ad absurdum .) d ou ble n e g a tio n .)0
(G en erality) A (x ) bx V x A ( x ) . V x A ( x ) b A (t).
(E xisten ce) A (t) b 3 x A (x ). If T (x), A (x ) b C,

then T (x), 3 x A (x ) bx C.
P roofs. T h e rule o f D -in tr o d u c tio n is T h eo rem 1 . T h e re rem ain

ten direct an d three o th er su b sid iary d ed u ctio n rules. T h e d irect rules
m a y be estab lish ed b y e x h ib itin g th e required d ed u ction s. T h e proofs of
th e su b sid iary d ed u ctio n rules are c o n v e n ie n tly p resen ted as sequences of
m e ta m a th e m a tic a l sta te m e n ts (certain of w h ich sta te m e n ts are to b e
su b sta n tia te d b y e x h ib itin g a d ed u ctio n as in th e p roof o f a d irect rule,
an d others of w h ich follow from p reced in g of th e sta te m e n ts b y T h eorem
1 or b y gen eral properties of b)- In b o th cases_ ap p eal is m ad e a t som e
p o in t to a correspon ding one of th e po stu lates. T h ese proofs are g iv e n
b elo w for several of th e rules of each ty p e , th e others b ein g le ft to th e
reader. H o w e v e r here an d in sim ilar situ ation s, th e reader is u rged first
to a tte m p t him self ev e n those w h ich w e giv e.
D irect rules , d -elimination.

1. A — first assu m p tion form ula.
2. A D B — second assu m p tion form ula.
3. B — R u le 2, 1, 2.
T h is rule is sim p ly R u le 2 of th e p o stu la te list (the " D - r u l e ” , or

"m o d u s pon en s” of tra d itio n a l logic) resta ted as a d erived rule.
& -in tro d u ctio n . W e a lre a d y h a v e th is as (3') § 20.
-l-e lim in a tio n , or d isch arge of d o u b le negation.
1. - i - i A — assu m ption form ula.

2. - i - i A D A — A x io m S ch em a 8 .
3. A — R u le 2, 1, 2.
V -in tro d u ctio n . L e t C b e som e ax io m n o t c o n ta in in g x free.
1. A (x ) — assumption formula.
2. A (x ) D (C D A (x)) — A x io m S ch em a la.
3. CDA(x) — R u le 2, 1 , 2.
4. CDVxA(x) — R u le 9, 3.
5. C— an axiom .
6 . V x A (x ) — R u le 2, 5, 4.
S ubsidiary deduction rules . V -elim ination.
1. T, A [- C — hypothesis.
2. r h A DC — T h eo rem 1, 1.
3. T, B h C — hypothesis.
4. T h B d C — T h eo rem 1 , 3.
5. ADC, B C h A V B C — usin g A x io m S ch em a 6 an d
d d R u le 2.
6. A V B , A V B D C > C — D -e lim in a tio n (or u sin g R u le 2).
7. T, AVB b C — 2, 4, 5, 6.
3-elim in ation .
1 . T (x ), A (x ) b C — h yp o th esis.
2 . T (x ) b A ( x ) D C — T h eo rem 1 , 1 .
3. A (x ) D C bx 3 x A (x ) D C — usin g R u le 12.
4. 3 x A (x ), 3 x A (x ) D C b C — D -elim in atio n .
5. T (x ), 3 x A (x ) bx C — 2 , 3, 4.
D iscussion. T h ese rules g iv e a classification ol lo gical operatio ns as

in tro d u c tio n s an d elim in ation s o f th e lo gical sy m b o ls, a d a p te d from
G e n tz e n 1934 - 5 .
T h e rule called “ V -elim in atio n ” does serve to elim in ate a d isju n ctio n
sy m b o l, w h en it is used as follows.
1. b A V B — suppose given .
2. b C — suppose g iv e n , w ith th e free v a ria b les held co n sta n t for A .
A
3. B b C — suppose g iv e n , w ith th e free v a riab les held co n sta n t for B .
4. A V B b C — V -e lim in a tio n (w ith V e m p ty ), 2, 3.
5. b C — 1,4.
T h is process corresponds to th e fam iliar inform al m eth o d of p roof b y
cases: E ith e r A or B . Case 1 : A . T h e n C. Case 2: B . T h e n C. H en ce C.
S im ilarly, 3-elim in ation , used as follow s, elim in ates an existen ce
sym b o l.
1. b 3 x A (x ) — suppose given .
2 . A (x ) bC — suppose g iv en , where C does n o t co n ta in x as a free
va riab le, an d w ith th e free va riab les h eld co n sta n t for A (x ).
3. - 3 x A (x ) b C — 3-elim ination , 2 .
4. b C — 1, 3.
T h is corresponds to th e fam iliar argu m e n t: T h ere e x ists an x such
th a t A ( x ) ; consider such an x. T h en C, w h ich does not d epend on x.
H en ce C.
§23 INTRODUCTION AND ELIMINATION RULES 101
S im ila rly , -» -in tro d u ctio n corresponds to th e m eth o d of red u ctio ad

absurdum .
U sin g th e T p ro v id e d for in th e theorem , a n y of these procedures ca n

b e carried o u t in th e presence of a n y list o f a d d itio n a l assu m p tion form ulas.
T h e fo llo w in g show s th a t A , - » A b B (in w o r d s : from a co n trad ictio n
A an d - i A , a n y form u la B is deducible). T h is w e sh all cite as th e rule
of weak -elim ination.
1. A , -iA , - iB b A.
2. A, i A , -iB b -iA .
3. A, -iA b t i B — -i-in tr o d u c tio n , 1 , 2 .
4. -i-iB b B — -i-e lim in a tio n .
(9') 5. A, -i A bB — 3, 4, as w as to be proved .
S te p 3 am o u n ts to b la m in g th e form u la - i B for th e co n trad ictio n

A an d - i A of 1 an d 2 . C o n tin u in g, w e h a v e :
6. -iA b A d B — D -in tro d u c tio n , 5.

7. b “i A D ( A D B ) — D -in tro d u c tio n , 6 .
O u r form al sy ste m w a s in ten d ed as a fo rm alizatio n of n um ber th eo ry,

in clu d in g m eth o d s o n ly a c ce p te d u nder th e classical v ie w p o in t (cf. § 13).
H o w ever, if A x io m S ch em a 8 (-i - i A D A ) is replaced b y th e fo llo w in g
(cf. (9'): 7), all th e p o stu la tes express principles also a c ce p te d b y th e
in tu itio n ists (cf. end § 30):
8 1. -iAD (AD B).
In term s of th e d erived rules of T h eorem 2, th is m eans re p lacin g

-i-e lim in a tio n b y w e a k - i -elim in ation . W h e n w e w ish to consider th is
sy ste m also, w e ca ll th e origin al sy ste m w ith P o stu la te 8 th e classical
system , an d th e sy ste m w ith P o stu la te 81 in stea d th e (corresponding)
in tu ition istic system . O u r results are m ark ed w ith th e sy m b o l in e v e r y
case w h en th e d em o n stratio n w e g iv e is n o t v a lid for b o th syste m s, b u t
o n ly for th e classical (and no d em o n stratio n w h ich th e reader is e x p e c te d
to d iscover for h im self is a v a ila b le for th e in tu itio n istic system ).
U se of V -in tro d u c tio n follow ed b y V -elim in a tio n g iv e s us th e follow in g

rule.
S ubstitution for an individual variable. I f x is a variable , A (x)

is a form ula , and t is a term which is free for x in A ( x ) : A (x) bx A (t).
W e sh all a b b re v ia te th e p resen tatio n of a p p licatio n s of our d erived

rules, u sin g t a c it ly gen eral properties of h
102 FORMAL DEDUCTION CH.
E xample 1. Consider th e follow in g argum ent.
1 . A , B [- C — suppose th is given .
2. A & B h A — & -elim in atio n .
3. A & B h B — & -elim in ation .
4. A & B b C — 1, 2, 3.
W e condense th is as follow s.
1. A , B h C — suppose th is given .
2. A & B h C — & -e lim ., 1.
G iv e n ‘T h P, P b Q” (w hich m eans: T b P an d P b Q), we

condense to ‘ T b P b Q ” ; an d sim ilarly w ith longer ch ains of ded uctions,
each of w h ich after th e first has as its o n ly assu m ption form ula th e
con clusion of th e preceding. ( B u t ‘ T h P , h Q ” m e a n s : T h P an d b Q-)
E xample 2.
1. A D B , A b B — D -elim in atio n .
2. B b B V C — V -in troduction .
3. ADB, A b B VC — 1 , 2.
W e condense th is to :
1. A D B , A b B b B V C — D -e lim ., V -introd.
* § 24. D e p e n d e n c e and v a r i a t i o n . F o r th e p red icate calculus, in

order to use a d e d u ctio n o b ta in e d (i.e. p ro v ed to exist) b y one of th e
d eriv ed rules of T h eo rem 2 as a su b sid iary d ed u ctio n for a n ew a p p licatio n
o f one of th e rules, w e sh all need (so far as our in form ation goes) to kn o w
n o t o n ly th a t th e d ed u ctio n exists, b u t also th a t th e free va riab les are
h eld co n sta n t for t h e a ssu m p tio n form ula to b e discharged.
In order to h a v e su ch in fo rm a tio n on h a n d w h en it is needed, w e shall
m a k e it a p ra c tic e in a p p ly in g th e rules to keep tr a c k of all cases w hen
a v a ria b le m a y b e v a rie d in th e resu ltin g d edu ction . I t is co n ven ien t to
do th is b y w ritin g a n y v a ria b les w h ich m a y b e v a rie d as superscripts
on th e s y m b o l “ b ” « T h is n o ta tio n is n o t fu lly e x p licit, as it does n o t
sh ow for w h ich of th e assu m p tio n form ulas a g iv e n superscript v a ria b le
m a y b e varied . W e m a y th e n sim p ly associate th e superscript w ith th e
assu m p tio n form ulas in w h ic h th e v a ria b le occurs free. (W hen there is
occasion to be m ore e x p licit, th e fa c ts m a y be sta ted v e rb a lly , e.g. as in
L e m m a 8 a 'b e lo w .)
W e recall th a t, u nder th e defin ition of va riatio n (§ 22), a v a ria b le y
ca n b e v a rie d o n ly for an assu m p tio n form ula D ; in w h ich it occurs free.
I t is ea s ily seen th a t, g iv e n a n y d eduction D 1#. . D, b E , assu m p tio n
§24 DEPENDENCE AND VARIATION 103
form u la D 3, an d v a ria b le y th a t occurs free in D i# one ca n fin d an o th er
d ed u ctio n of E from Dv ..., in w h ich y is v a rie d for D 5. (H i n t : in tro -
d u ce in to th e d ed u ctio n som e superfluous steps.) Th erefo re our in terest
w ill a lw a y s b e in w h e th er th ere is som e d ed u ctio n D 1#. . . , D t b E in
w h ich y is n o t va rie d for D ,; an d our sta te m e n ts to th e effect th a t a
va ria b le y is va rie d o n ly for such an d such a ssu m p tio n form u las w ill
m ean t h a t there is som e d ed u ctio n (w ith th e g iv e n assu m p tio n form ulas
an d conclusion) in w h ich th is is th e case.
S im ilarly, g iv e n a n y d ed u ctio n Dv . . Dj b E , it is a lw a y s possible
to fin d an o th er d ed u ctio n of E from D v . .. D l in w h ich th e con clusion
E depend s on D ;.
T h e procedure for keep in g tr a c k o f v a ria tio n is en tire ly stra ig h tfo rw a rd
(likew ise for dependence). T h u s far our o n ly d erived rules w h ich ca ll for
in tro d u cin g a superscript are V -in tro d u c tio n , 1 -e lim in a tio n an d su b
stitu tio n . (E v e n th en th e superscript is n o t a lw a y s necessary, e.g. w hen
th e A (x ) for an V -in tro d u c tio n or su b stitu tio n does n o t co n ta in x free.
A lso cf. L e m m a 7 b below .) M oreover, once superscripts h a v e been in
trod u ced , w e m u st ca rry th em forw ard in th e o b v io u s w a y from g iv e n d e
d u ctio n s to resu ltin g d ed u ctio n (unless som e reason to th e c o n tr a ry ca n
b e giv en ), b o th in a p p ly in g th e su b sid iary d e d u ctio n rules of th is section ,
an d in co m b in in g d ed u ctio n s b y gen eral properties of b ( § 20 ).
T h e situ atio n s w h ich arise in p ra c tic e are sim ple en o u gh so t h a t w e
h a v e little tro u b le in seeing w h a t is h ap pen in g. V a ria b le s w h ich are b ein g
va rie d h a v e u s u a lly ju s t p re v io u sly b een in tro d u ced in t h a t role for som e
im m e d ia te purpose, so t h a t t h e y are n o t lik e ly to b e o verlooked . H o w
ever, to m a k e th e th e o r y of our d e riv ed rules co m p le te, th e fa c ts are s ta te d
in m ore d e ta il in th e fo llo w in g lem m as.
L emma 6. In Theorem 1 , A D B depends on a given one of the form ulas

T in the resulting deduction T h A d B , only if B depends on the sam e one
in the given deduction T , A b B . S im ila rly , in the other subsidiary deduction
rules of Theorem 2 , the conclusion depends on a given one of the F ’s in the
resulting deduction , only if the conclusion depends on the sam e one in the
given deduction {or in one at least of the two given deductions ).
F o r otherw ise th a t assu m p tio n form u la co u ld b e o m itte d from th e
F s in a p p ly in g th e rule, an d afterw ard s in tro d u ced b y (II) (and (IV )) of
L e m m a 5.
In V -elim in ation , if th e C does n o t d epen d b o th on th e A in T , A b C
an d on th e B in T , B b C, th e V -elim in atio n ca n b e a v o id e d a lto g e th e r.
S im ila rly in 3-elim in atio n , if th e C does n o t d ep en d on th e A (x ).
Lemma 7a. In Theorem 1 , a variable is varied for a given one of the
r s in the resulting deduction T \- A D B , only if it is varied for the sam e one
in the given deduction F , A f- B . S im ilarly in the other subsidiary deduction
rules of Theorem 2 , except that for the variable x of ^-elim ination the situation
is as stated in Lem m a 7b .
L emma In 3 - elim in ation , the x is varied in the resulting deduction
7b .
r(x), 3 x A (x )\-x C only for those of the T ( x ) ,s which contain the x free and
on which the C depends in the given deduction T (x ), A (x ) |- C. (In 3 - elim i
n ation , no variable is varied for the 3 x A ( x ) ; and likew ise in ^ -elim in ation
for the A V B.)
T h e se tw o lem m as m a y b e v e rifie d b y e x a m in in g th e proofs of T h eo rem s
1 an d 2 . F o r 3 -elim in ation , if th e x is v a ried in th e g iv e n d ed u ctio n for a n y
one of th e r ( x ) 's on w h ic h th e C does n o t d epend, t h a t one of th e T ( x ) ,s
m a y b e o m itte d for th e 3-elim ination .
T h e discussions of d epen d en ce an d v a ria tio n under step s perform ed
b y gen eral properties of b (using th e list of such properties p ro v id e d in
L e m m a 5) are le ft to th e reader, e x c e p tin g t h a t of v a ria tio n for th e A ’s
of (V). B efo re tre a tin g th is (in L e m m a 9), w e p ro v e th e fo llo w in g b a sic
lem m as.
L emma 8 a. If
(I) D x, D 2, . . . , D j |- E ,
where , for j = 1 only the distinct variables y ?1, . . . , y .p , are varied for
D ?. (but y n > . . . , y .p , need not be distinct from the variables y kl , . . . , y kPk for
] 7^ k), then
(II) h V y u . . . V y lp D , D ( V y 21. . . V y 2p2D 2 D . . . ( V y a . . . V y ,p D p E ) . . . ) ;
and conversely.
For (II) follow s from (I) b y V -elim in atio n s an d D -in tro d u c tio n s.
C o n ve rsely, (I) follow s from (II) b y V -in tro d u c tio n s a n d D -e lim in a tio n s .
Lemma Given a deduction of E from D 1?. . . , D*, another deduction

8b.
of E from D lt . . . , D t can be found in which , for j = 1 , . . . , I, R ule 9 is
app lied , to prem ises dependent on D js only with respect to variables which are
varied for D 5- in the given deduction , and R ule 12 is applied to no prem ise
dependent" on an assum ption formula.
W e ta k e th e g iv e n d ed u ctio n as (I) for L e m m a 8 a, p ass to (II), an d
th en ce c o n versely b a c k to an o th er d ed u ctio n (I). In th is th e a p p lica tio n s
of R u le s 9 an d 12 w h ich co m e from th e proof (II) are to prem ises d ep en d en t
on no assu m p tio n form ulas. T h o se w h ich com e from th e V -in tro d u c tio n s
used to o b ta in (I) from (II) are e x a c tly as d escribed in th e present lem m a.
L e m m a 9. In (V) of Lem m a 5, a variable y is varied for a given one of the

is!s in the resulting deduction A , F h E , only (a) if y is varied for the same
one of the A ’s in the first given deduction A b C, or (b) if y is varied for C
in the second given deduction C, F b E , and C depends on that one of the
A ’s in the first given deduction A h C, and that one of the A 's contains y free.
L e m m a 9 w o u ld be im m e d ia te, e x c e p t for th e fo llo w in g co n tin g e n cy .

In th e second g iv e n d ed u ctio n C, F b E there m ig h t b e an a p p lica tio n of
R u le 9 or 12 w ith respect to y to a prem ise d ep en d en t on C, an d y e t y
b e h eld co n sta n t for C b ecau se C does n o t co n ta in y free. I f th is d ed u ctio n
w ere co m b in ed w ith a d ed u ctio n A f- C in w h ich C d epends on one of th e
A 's co n ta in in g y free, th en y w o u ld b e v a ried for th a t one of th e A ’s in th e
resu ltin g d ed u ctio n A , F h E . B u t w e can use L e m m a 8 b to replace th e
g iv e n d ed u ctio n C, T b E b y an other, after w h ich th e c o n tin g e n c y d e
scribed ca n n o t arise for a n y v a ria b le y.
H ereafter, for n ew d eriv ed rules th e fa c ts resp ectin g d epen d en ce a n d
v a ria tio n w ill b e as one w o u ld e x p e ct, w ith a n y e x ce p tio n s n o ted , an d all
cases w hen v a ria tio n m a y b e in tro d u ced in d ic a te d b y su p erscripts on “ b ” .
In gen eral, in a su b sid iary d ed u ctio n rule h a v in g assu m p tio n form ulas
for g iv e n an d resu ltin g d ed u ctio n s in o b v io u s correspondence: T h e co n
clusion depends on (A g iv e n v a ria b le is va rie d for) a g iv e n one of th e as
su m p tio n form ulas of th e resu ltin g d ed u ctio n , o n ly if it does on (is for) th e
corresponding assu m p tio n form u la of th e g iv e n d ed u ctio n or of either
g iv e n d ed u ction . E x a m p le s are T h eorem s 3 an d 4 § 25 (for d epend ence),
15 an d 16 § 3 4 , th e form al in d u ctio n rule § 3 8 , 41 (b) an d (c) § 7 3 , 42
( III)— (V) an d 43 ( V i l a ) — ( V U I b ) § 7 4 , 59 an d 60 (b 2 )— (d) § 8 1 .
Strong V-introduction and 3-elimination. O c c a sio n a lly it is u sefu l

to em p lo y V -in tr o d u c tio n an d 3 -elim in atio n in a s lig h tly stren gth en ed
version, w h ic h p erm its a ch an ge in the va riab le.
Lemma 10 . L et x be a variable , A (x ) a form ula, and b a variable , such

that (i) b is free for x in A (x ) and (ii) b does not occur free in A (x ) {unless
b is x). Furtherm ore, for the ^-elim ination rule, let C be a form ula not con
taining b free, and let the free variables be held constant for A (b) in the sub
sidiary deduction. Then:
A (b ) bb V x A (x ). I f T(b), A (b) b C, then T(b), 3 x A (x ) hb C.
(Stron g V -in tro d u c tio n .) (Stron g 3-elim ination.)
P roofs. In E x a m p le 3 § 22 we d erived th e rule C 3 A (b) bb C 3 V x A (x ).

U sin g th is in stea d of th e p o stu la te d rule 9 in our form er proof for V -

in tro d u c tio n (§ 23), w e o b ta in th e stro n g version.
D erived rules and respective postulates. W e ca ll P o s tu la te s

la , l b an d 2 th e postulates for 3 , P o stu la te s 3, 4a an d 4b th e Sc-postulates ,
P o stu la te s 7 an d 8 (or for th e in tu itio n istic sy ste m 7 an d 8 1) th e - i -
postulates , etc. P o s tu la te s for 3 are used in estab lish in g all th e d e riv ed
rules of T h eo rem 2 .
L emma 11.F or each selection of one or more of the logical sym bols
3, & , V, - , Vand 3 : The rules of Theorem 2 for 3 and those sym bols
1
{including for V and 3 the strong versions , and for - i in the intuition istic
system the weak - i -elim ination rule in place of the other) hold good in the
form al system which has as postulates only the "D-postulates and the postu
lates for the sym bols in question , provided that in case the sym bols include
V but not & the V -postulates include an additional axiom schema as follows ,
where x , A (x ) and C are subject to the sam e stipulations as for R ule 9:
9a. V x (C 3 A (x)) 3 ( C D V x A ( x ) ) .
P roof. T h is m a y b e v e rified b y a perusal of th e a b o v e proofs o f th e

rules, w ith an e x ce p tio n in th e case th e sy m b o ls in clu d e V b u t n o t & ,
since th e tre a tm e n t of C ase (e) S u b ca se (el) of T h eo rem 1 (§ 22 ) e n tails
use of th e & -p o stu la te s. W ith th e a d d itio n a l V -sch em a, h o w ever, t h a t
ca n b e rep laced b y th e follow ing.
p. A 3 (C 3 A (x)) — as before.
p + l. A 3 V x ( C 3 A (x)) — R u le 9, p.
p + 2. V x (C 3 A (x)) 3 (C 3 V x A (x )) — A x io m S ch em a 9a.
d ed u ctio n from p+1 an d p+2
g iv e n b y (8 '):1 (end § 2 1 ).
p+q+2. A 3 (C 3 V x A (x )).
D eductions in tree form. W e h a v e been ta k in g a d e d u ctio n to
b e a linear sequence of (occurrences of) form ulas. S o m etim es it is u sefu l
in ste a d to consider th e (occurrences of) form ulas in a p a r tia l ordering
w h ic h represents th e lo g ic a l stru ctu re d ire c tly . In th is ordering, th e
prem ises for each inference are w ritte n im m e d ia te ly o ve r th e conclusion,
as in th e sta te m e n t of th e rules o f inference; an d no (occurrence of a)
fo rm u la serves as prem ise fo r m ore th a n one inference. A d e d u ctio n (or
proof) in th e form er arran gem en t w e s a y is in sequence fo rm ; in th is, in
tree form.
T h e m e th o d o f c o n v e r tin g a d e d u ctio n of E from T g iv e n in seq uence
form , w ith a g iv e n an alysis, in to one in tree form (called “ resolution in to
p roof th re a d s” b y H ilb e rt an d B e rn a y s 1934 p. 2 2 1 ), an d in versely , w ill
b e clear from an exam p le.
E xample 1. Consider th e d ed u ctio n (6 ) of § 2 1 . By th e a n a ly sis,

th e b o tto m form u la 8 is an im m e d ia te con sequence of 7 an d 5. L e t us
w rite 7 an d 5 im m e d ia te ly a b o v e 8 . T h e n 7 is an im m e d ia te con sequence
o f 1 an d 6 ; so w e w rite 1 an d 6 ju st o ve r 7 ; etc. L o o k in g ju st a t th e
num bers 1 — 8 , w e o b ta in th e fo llo w in g figure.
(a) 1 2
1 6 3 4
7 5
8
W r itin g in th e form ulas th em selves (w ith new num bers 1 ' — 9' an d th e
an alysis), w e h a v e th e d ed u ctio n in tree form .
(b) second
assu m p tio n A x io m
second form u la S ch em a 4a first
assu m p tio n A x io m 4'. A & B 5 '. A & B D A assu m p tio n
form ula , S ch em a 4b -------------------------------------------- 2 form u la
1 '. A & B 2'. A & B D B 6 '. A 7' . A d (B d C)
----------- ------------------------------------------------ 2 ---------------------------------------------------------------------------------- 2
3'. B 8 '. B d C
9'. C
In v e rse ly , from th is d ed u ctio n in tree form of C from A d (B d C)

and A & B , b y arran gin g th e (occurrences of) form ulas as a linear se
quen ce s a y in th e order of th e n um bers 1'— 9', w e o b ta in one in sequence
form (not th e original one).
B r ie fly described, a branch of a d ed u ctio n in tree form con sists of th e

(occurrences of) form ulas in linear sequence, p assin g d o w n w ard w ith in
th e tree stru ctu re, b e gin n in g w ith a form u la o ccurring as an a x io m or
assu m p tio n form ula, an d te rm in a tin g in th e con clusion (or endform ula)
o f th e d ed u ctio n . T h e height of a d ed u ctio n in tree form is th e le n g th of a
longest b ran ch (or in o th er w ords, th e n u m b er of levels). A n (occurrence of)
a form ula is said to b e above an o th er (or th e la tte r to b e below th e form er),
if th e form er is a b o v e th e la tte r in th e sam e bran ch.
E xample 1 (concluded). T h e d ed u ctio n (b) has 5 bran ch es, n a m e ly :

r , 3', 9 '; 2 ', 3 ', 9 '; 4', 6 ', 8 ', 9 '; 5 ', 6 ', 8 ', 9 '; 7 ', 8 ', 9'. T h e h e ig h t is 4.
T h e (occurrence of a) form ula 4' is a b o v e 8 ' b u t n o t a b o v e 3'.
C hapter VI
THE PROPOSITIONAL CALCULUS
§ 25. P r o p o s i t i o n l e t t e r f o r m u l a s . In th is ch a p te r w e single o u t
for in te n s ive s tu d y th a t p a rt of th e form al sy ste m w h ich is o b ta in e d b y
u sin g o n ly th e p o stu la tes of G ro u p Al. T h e m eanings of ‘provable*,
‘deducible* an d ‘ K are to b e u n d erstoo d acco rd in gly.
U n d er th e d efin ition of ‘form ula* w h ich w as g iv e n for th e fu ll sy ste m
in § 17 our form ulas are all b u ilt u p in term s of th e n u m b er-th eo retic
sym b o lism . B u t so lo n g as w e are usin g o n ly th e p o stu la tes of G ro u p A l ,
m a n y d eta ils of th is sym b o lism are irrelevan t.
I t is undesirable th a t w e sh ould restrict th e g e n e ra lity o f our tre a tm e n t
of th e p rop osition al ca lcu lu s becau se w e in ten d a p p ly in g it in th e n u m b er-
th eo retic syste m . O n th e oth er h and , w e m u st prepare th e gro u n d for
th a t ap p licatio n .
W e n o w g iv e , for use in th e p roposition al calcu lu s, an a lte r n a tiv e
d efin itio n of ‘form ula*, w h ich elim inates th e irrelevan t d eta ils of th e
n u m b er-th eo retic definition.
W e sta rt b y in tro d u cin g form al sy m b o ls of a new kind,
c3, B, C, ...,
ca lled proposition letters , o f w h ich w e suppose a (p o ten tially) in fin ite

list to b e a v a ila b le . T h e n ew d efin itio n o f ‘form ula* follow s.
1 . A p roposition le tte r is a form ula. 2 — 5. I f A an d B are form ulas ,
th en (A) D (B), (A) & (B), (A) V (B) an d - i ( A ) are form ulas . 6 . T h e o n ly
form ulas are those g iv e n b y 1 — 5.
C o m p arin g th is w ith th e d efin itio n in § 17, C lause 1 o f th a t d efin itio n
is replaced b y th e n ew C lau se 1 , an d C lauses 6 — 7 are suppressed. W h e n
w e w ish to d istin gu ish b etw ee n th e tw o notions of form ula, w e sh all
ca ll t h a t o f § 17 number-theoretic form ula , an d th e present one proposition
letter form ula.
E xample L <3V ( i <3 & S ) is a proposition letter formula (pa
rentheses being omitted in continuation of practices established in § 17).
W e h en cefo rth agree, for th is ch a p ter, t h a t w h en w e s a y “ formula**
w ith o u t sp e c ify in g a p a rticu la r sense, th e w ord m a y b e read eith er in
108
§25 PROPOSITION LETTER FORMULAS 109
th e sense of proposition le tte r form ula, or in th e sense of p re d ic a te le tte r
form u la to b e d efin ed in th e n e x t ch ap ter, or in th e n u m b er-th eo retic
sense. (It m ig h t b e read in still o th er su ita b le senses. B u t for definiteness
w e restrict it here to these three, le a v in g th e qu estion w h e th er an o th er
sense is su itab le to b e considered w h en one h as an o th er sense in m ind.)
R e su lts s ta te d in th is c h a p te r u sin g sim p ly “ fo rm u la ” w ill th u s a p p ly ,
a t no e x tr a expense, to a n y one of three form al syste m s, h a v in g in co m
m on G ro u p A 1 as p o stu la te list, b u t d ifferin g in th e sense of form ula.
pure propo
T h ese three sy ste m s w e m a y d istin gu ish r e sp e c tiv e ly as th e
sitional calculus , th e predicate letter propositional calculus an d th e
number-theoretic propositional calculus.
Som e of our results, h ow ever, w ill b e s ta te d u sin g “ p roposition le tte r
form u la” . T h ese w ill also a p p ly gen erally. T h e o n ly differen ce in th e case
of these is th a t it is easier to e x p la in th em in term s of p roposition letters,
lea vin g it to th e reader to tra n sla te th em to oth er senses o f form ula, w h e n
he needs to , b y m eans of tw o gen eral rules for tra n sla tio n w h ich w e sh all
n e x t p ro v id e (Theorem s 3 an d 4).
L e t P 1? . . . , P w b e a list of d istin ct proposition letters. (H ere “ P x” , . . . ,
“ P w” are m et% m ath em atical letters, used as nam es for proposition le t
ters w hen w e do n o t w ish to lim it our discussion b y u sin g p a rticu la r
proposition letters.)
A proposition le tte r form u la A is said to b e a proposition le tte r form u la
in P 1# . . . , P w, if no proposition letters oth er th a n P lf . . . , P m occur in A .
E xample 2. H V (-i<£? & B) is a proposition le tte r form u la in

a %S, C.
Substitution for a proposition le tte r (or sim u lta n eo u sly for severa l
d istin ct proposition letters) is d efin ed as for a v a ria b le in § 18, e x c e p t
th a t it applies to all occurrences w ith o u t e x ce p tio n (there b e in g here no
'bo un d occurrences'). A lso , la te r in th e section, w e use an operatio n called
replacement in all occurrences of a form u la (or sim u ltan eo u sly of several
d istin ct form ulas), defined sim ilarly (the x in th e d efin itio n of § 18
b ecom in g a fo rm u la); th is op eratio n is u n a m b igu o u s becau se th e o c
currences w ill b e n o n -o verlap p in g.
T heorem 3. S ubstitution for proposition letters. L ei T be

proposition letter form ulas, and E a proposition letter form ula, in the distinct
proposition letters P x, . . . , P m. L et A v . . . , A w be form ulas. Let T * and E *
result from V and E , respectively, by substituting sim ultaneously A x, . . . , A m
for P 1# . . . , P w, respectively. If P b E , then T * b E * . (For th e case t h a t T
is e m p ty : If b E , then b E * .)
110 THE PROPOSITIONAL CALCULUS CH. VI
P roof. F o r th e p o stu la tes of G ro u p A l , n o th in g is required o f th e

A , B , C w h ich ap p ear in those p o stu lates e x c e p t th a t t h e y b e form ulas
fix e d th ro u g h o u t a g iv e n a p p lica tio n of a p o stu late. N o w consider th e
g iv e n d ed u ctio n of E from T in th e pure propositional calculus. T h e for
m u las T an d th e form u la E are proposition le tte r form ulas in th e d istin ct
p roposition letters P x, . . . , P m, an d besides these som e m ore P m+1,
. . P m+r m a y occur in o th er form ulas of th e d ed u ction . L e t A m+1, . .
A m+r b e a n y form ulas. F o r each of th e p rop osition letters P x, . . P m+r,
le t th e re sp e ctive form ulas A lf ..., A m+r b e su b stitu te d th ro u gh o u t e v e r y
form u la of th e g iv e n ded u ction . F o r each a p p lica tio n o f a p o stu la te in th e
g iv e n d ed u ctio n , th e A , B , C of th e a p p licatio n w ill b e tran sform ed b y
th e su b stitu tio n in to expressions A * , B * , C * (every occurrence of A
b eco m in g an occurrence of A * , etc.). T h ese expressions A * , B * , C * w ill b e
form ulas, since to each a p p lica tio n of one of Clauses 2 — 5 of th e d efin i
tio n of proposition le tte r form u la used in b u ild in g u p A , B , C from th e
proposition letters Pv . . . , P w+r there w ill correspond an a p p lica tio n of
th e sam e-n u m bered clause of th e d efin ition of form ula used in b u ild in g
u p A * , B * , C * from th e form ulas Av . . . , A m+r. H en ce w e sh all h a v e aga in
an a p p lica tio n of th e sam e p o stu late. T h u s th e sequence of form ulas
in to w h ich th e g iv e n d ed u ctio n is tran sform ed is aga in a d ed u ctio n w ith
th e sam e an alysis. I t is a d ed u ctio n of E * from T * .
E xample 3. T o illu stra te th e proof of th e rule, consider th e follow in g

d ed u ctio n of c ? D B from B.
1. B — assu m p tio n form ula.

(a) 2. B D (c2f D B) — A x io m S ch em a la.
3. - R u le 2 , 1 , 2 .
O n s u b s titu tin g B, -ic2f & C (or lc (a = c '), -i a —0) for <3f, B, w e o b ta in

th e fo llo w in g d ed u ctio n (b) (or (c)) w ith th e sam e an alysis as (a).
1. -i<C? & C — assu m p tio n form ula.

(b) 2. -ic2f & C D ( B & C) — A x io m S ch em a la .
3. B D ic Z&C — R u le 2, 1, 2 .
1. -ia = 0 — assu m p tion form ula.

(c) 2. ~ ia = 0 D (3 c(a = c') D —\a = 0 ) — A x io m S ch em a la.
3. 3c(<z=c') D -i< z = 0 — R u le 2 , 1 , 2 .
T o illu stra te th e a p p lica tio n of th e rule w ith th e sam e tw o exam p les,

w e k n o w (b y reference to (a)) th a t
(a') B h ^ D B .
§25 PROPOSITION LETTER FORMULAS 111
T h e rule allow s us to infer, b y th e su b stitu tio n s described,
(b') -t e 2 & C b ®
(c') b 3c(<z=c') 3 —\a = Q .

In one case ((a') to (b')), w e are u sin g th e rule w ith in th e pure proposi
tio n al calculus, for w h ich it is a d erived rule of th e su b sid iary d ed u ctio n
ty p e . In th e other case, b y su b stitu tin g form ulas in an o th er sense, w e
h a v e used th e rule to infer (c') in th e n u m b er-th eo retic p roposition al
calculus from (a') in th e pure calculus.
E v id e n t ly (a'), (b'), (c'), etc. can all b e in clud ed in th e s t a t e m e n t :
If A an d B are form ulas, th en
(10') B b ADB.
T h e sense of th e rule appears as sim p ly th a t, h a v in g estab lish ed a d e

d u c ib ility relationship in term s of p articu lar proposition letters cC?,
S, C, . . . , w e can assert th e sam e relationship in th e form of a sch em a w ith
m e ta m a th e m a tic a l letters " A " , “ B ” , “ C ” , . . . represen tin g a n y form ulas.
C o m b in in g th is rem ark w ith our earlier o b servatio n th a t a rule of th e
direct fornj T b E a lw a y s rem ains v a lid in th e presence of a d d itio n a l
p o stu lates (§ 22 ), w e see th a t all results of th e form T b E o b ta in e d in
this ch a p ter (w hether or n o t sta te d in term s of proposition letters) w ill
h old go o d for la te r ch ap ters w here w e ta k e in to ac co u n t m ore o f th e
structure o f th e form ulas an d a larger p a rt of th e p o stu la te list of th e
original form al syste m .
R emark 1. W ith in th e pure p rop osition al calculus, su b stitu tio n can

b e perform ed for a single v a ria b le P 5 a t a tim e, b y u sin g th e rule w ith
P j , . . . , Py_i, Py_j_j, • • •, P m $ A m.
th e A^, • • • A A y _ | _ j , . . . ,
A form u la w ill b e said to b e prim e (for the propositional calculus ), if

it does n o t h a v e a n y one of th e form s A D B , A & B , A V B , - i A w here
A an d B are form ulas.
E xample 4. a = 0, 3 c (a = c ') a n d \fc(a —c' V a = b ) are prim e, b u t

~ ia = 0 an d - i a = 0 & 3 c (a = c ') are not. A proposition le tte r form ula
is prim e, o n ly w hen it consists sim p ly o f a proposition letter.
F ro m th e fa c t th a t th e scopes of th e operators 3 , & , V, - i in a form ula

can be recognized w ith o u t a m b ig u ity (§ 17), it follow s t h a t a n y g iv e n for
m u la is co n stru cted in a u n iq u e ly d eterm ined m ann er o u t of prim e
form ulas b y ap p licatio n s of C lauses 2 — 5 of th e d efin ition of form ula.
W e ca ll th e d istin ct prim e form ulas ou t of w h ich a form ula or several for
m u las are th u s co n stru cted th e distinct prim e components (for the propo
sitional calculus) of th a t form ula or set of form ulas.
E xample 5. The d istin ct prim e com pon en ts of a = 0 V (—\a = 0

& 3 c (a = c ')) 3 V c (a = c ' V a —b) are a = 0 , 3 c (a = c ')f yic ( a = c ' V a = b ) .
T heorem 4. Converse of substitution for proposition letters .
U nder the same stipulations as in Theorem 3, and provided in addition that
A lf . . ., A m are distinct prim e form u las: I f T * |- E * , then T b E .
P roof . Consider a g iv e n d ed u ctio n of E * from F*. T h e d istin ct prim e
com p on en ts of th e form ulas T * an d E * are th e form ulas A v . . ., A m.
O th e r form ulas of th e d ed u ctio n m ay co n trib u te a d d itio n a l d istin ct
prim e com pon en ts A m+1> ..., A m+r. L e t P m+1, . . . . P m+r be proposition
letters. T h ro u g h o u t th e form ulas of th e g iv e n d ed u ction , let Av ...,
A m+r b e replaced sim u ltan eo u sly in e v e ry occurrence b y P x, P w+r,
re sp e ctiv e ly . Consider a n y p o stu la te ap p licatio n of th e g iv e n d ed u ctio n
of E * from T * . I t is re a d ily show n (using L e m m a 3 as illu stra ted in § 17)
th a t th e replacem en ts w ill ta k e p lace w ith in th e A , B , C of th e a p p licatio n ,
p ro d u cin g proposition le tte r form ulas A ', B ', C ', e v e r y occu rren ce of A
b e co m in g an occurrence of A ', etc. T h u s th e sequence of proposition le tte r
form ulas in to w h ich th e g iv e n d ed u ctio n of E * from r * is tran sform ed b y
th e replacem en ts is a d e d u ctio n of E from F w ith th e sam e an alysis. A s
th e m eth o d of th is proof shows, th e con verse rule ca n also b e fo rm u lated
th u s:
T heorem 4 (second version). Let F* be form ulas and E * a form ula

having as their distinct prim e components A lt . . . , A m. Let Fv . . . , P m
be proposition letters , not necessarily distinct. Let T, E result from F*, E * ,
respectively , by replacing , sim ultaneously in all occurrences , A v . . . , A w by
P x, . . ., P m, respectively. Then F* b E * only if T b E .
E x c e p t w h en “ form ula* * is read in an oth er sense th a n proposition

le tte r form ula, T h eo rem 4 is in clu d ed in T h eorem 3 .
E xample 6. To illu stra te th e proof, le t —\a = 0 D (3 c (a = c r) 3

- 10 = 0) occur in th e g iv e n d ed u ctio n of E * from F* as an axio m by
S ch em a l a (as a t E x a m p le 3 (c) S te p 2 ). R e p la c in g th e d istin ct prim e
co m p o n en ts a = 0 , 3 c (a = c ') b y th e proposition letters Ui, B, re sp e c tiv e ly
(or b o th b y d ) , g iv e s -»c3f 3 (B D -i< 3 ) (or -i< 3 3 (U1 3 - i <£?)), w h ich is
an a x io m b y S ch em a la in th e pure propositional calculus. B u t replacin g
a — 0, 3 c ( 0 = c ') 3 -u z = 0 (the la tte r not b ein g prime) b y c2f, B, or re
p la c in g th e three prim e p arts a = 0 , 3 c (a = c '), a = 0 (the first an d th ird
n o t b ein g d istinct) b y TT, B, C, w o u ld not g iv e an axio m b y S ch em a la.
§26 EQUIVALENCE, REPLACEMENT 113
E xample 7. T o illu stra te th e a p p lica tio n c f th e con verse rule, ta k e
th e fa c t (w hich w e w ill estab lish in § 28 E x a m p le 3) t h a t <C3 V (-i & $)
is u n p ro v a b le in th e pure p rop osition al calculus. It follow s, b y th e
con verse rule, s u b stitu tin g a = 0 , 3 c (a = c ') for <3, ®, re sp e c tiv e ly , th a t
£ = 0 V (-1 0 = 0 & 3 c (a = c ')) is u n p ro va b le in th e n u m b er-th eo retic
p ro p o sitio n al calcu lu s, i.e. th is form ula of our origin al sy ste m ca n n o t b e
p ro v e d on th e basis of th e p o stu la tes of G ro u p A 1 o n ly (alth o u gh in fa c t
it is p ro v a b le u sin g th e co m p le te list, §39). B u t from th e u n p r o v a b ility
of ( 5 f V ( n c^7 & S) in th e p ure ca lcu lu s, it does not follow th a t
a —0 V (~ ia = 0 & - ia = 0 ) is u n p ro va b le in th e n u m b er-th eo retic p ro p o
sitio n a l calculus. W h y ?
§ 26. E q u i v a l e n c e , r e p l a c e m e n t . L e t A an d B be form ulas. W e use

“A ~ B " as a b b re v ia tio n for (A d B )& (B d A). The sy m b o l
m a y b e read “ e q u iv a le n t” . I t fu n ctio n s as a form al operator, w h ic h
p la ced b etw ee n tw o form ulas of th e sy ste m g iv e s an o th er form u la of th e
syste m . In o m ittin g parentheses, it is th en ran ked ah ead of th e oth er
form al operators (§ 17).
W e s a y th a t A is equivalent to B in th e p ro p o sitio n al calcu lu s or o th er
form al sy ste m , if in th a t form al sy ste m b A ~ B . H ere th e w o rd “ e q u iv
a le n t" fu n ctio n s as a m e ta m a th e m a tic a l ve rb , w h ich p la ced b e tw e e n
tw o form ulas of th e sy ste m g iv e s a sta te m e n t a b o u t those form ulas.
T heorem 5. If A, B and C are form u las :

*1. h A D A . *2. A d B, B DC b A DC.
*3. AD(BDC) b B D (A D C ) .
*4. AD(BDC) h A & B D C . *5. A & B DC F-A d (B d C).
(P rinciple of id e n tity , ch ain inference, in terch a n ge of prem ises,
im p o rtatio n , ex p o rtatio n .)
* 6. A DB h(B d C)d (A d C). *7. A DB h (C d A ) d ( C d B).

* 8 a. A DB bA & C d B&C. * 8b. A D B ( - C & A d C&B.
*9a. A D B hA V C D B V C . *9 b . A D B h C V A D C V B .
(In tro d u ctio n of a con clusion, prem ise, c o n ju n c tiv e m em ber, or
d is ju n c tiv e m em ber, in to an im plicatio n .)
*10 a . -iA h ADB. *10 b . A h ^ A D B . * 11. B h A d B.

(D em o n stratio n of an im p lica tio n b y re fu tin g th e prem ise, or b y p ro v in g
th e conclusion.)
*12 . A D B b “iB D -iA . * 1 3. A D -iB b B D nA.

* 1 4 °. n A D B b “iBD A. *15°. -i A D n B b B D A .
(C ontraposition, an d co n trap o sitio n w ith double negatio n s suppressed.)
*16 . A D B , B =5A h A ~ B .
*17a. A ~ B h AD B. *17b . A ~ B h B 3 A .
* 18 a . A ~ B, A |- B . *18 b . A ~ B, B b A.
(From th e d efin itio n o f interm s of 3 and&.)

*19 . |- A ~ A . * 20 . A ~ B h B ~ A . *21. A ~ B, B ~ C |- A ~ C.
(R e fle x iv e , sy m m e tric an d tra n sitiv e properties of equivalence.)
*22. A3(B3C), -i-iA, —

i—
iB b —
i—
iC.
*23 . —i - i ( A D B ) b t i A D t i B.
*24. _i - i (A 13 B ), - 1 —i (B 13 C) b - i - i (A D C).
*25 . b —<—i ( A & B ) ~ - i —t A & i —i B ; in particu lar,
b - i “ i (A ~ B) ~ - i - i (A 3 B) & —i —i (B 13 A ).
(A d d itio n a l results of in terest for th e in tu itio n istic system .)
P roofs. E ig h t of these h a v e a lre a d y been establish ed , as fo llo w s:

*1 at § 2 0 ( 1 '); *2 at § 2 1 (8 ' ) : 1 ; *3 a t ( 5 ' ) : 3; * 4 at (6 ' ) : 2 ; 5 * a t (7');
*6 a t (8 '): 2 ; * 10a at § 23 (9'): 6 ; an d * 11 a t § 25 ( 1 O'). T h e reader m a y
estab lish th e others, u sin g th e d eriv ed rules of T h eo rem 2 for th e
proposition al calcu lu s (§23 ). F o r ex a m p le :
*9 a. 1. A3B, A b B b BVC — 3 - e l i m ., V-introd.

2. A 3 B , C bB V C — V -introd.
3. A 3 B , A V C b B V C — V -elim „ 1 , 2.
4. A3B b AVC3BVC —3-introd., 3.
*12. 1. A DB, -iB , A b B —3-elim.

2. A D B, i B , A b “ 'B .
3. A 3 B , —iB b ">A — - i-in tro d ., 1, 2.
4. A 3 B b “> B 3 - i A — 3 -in tr o d ., 3.
* 1 4. S im ila rly to * 12 , b u t now an a p p lica tio n o f -i-e lim . is ad d ed a t

S te p 3.
* 22 . 1. A 3 ( B 3 C ) , B b A 3 C — 3 - e li m „ *3 (or § 2 1 (5'): 2).

2. A 3 (B 3 C), B b “ i “ iA 3 - i - i C — *12 tw ice, 1.
3. A 3 (B3 C), -i- i A b B 3-i—
iC — 3-elim.,3-introd., 2.
- 4. A 3 (B 3 C), - t - i A h -in B 3 -i-iC — * 1 3 , * 12 , 3.
*23. T a k in g A 3 B, A , B as th e A , B , C, re sp e ctiv e ly , in *2 2 :
1. (A 3 B ) 3 (A 3 B ), - i - i (A 3 B ), - i - i A bnnB.But:
2. b (A 3 B) 3 (A 3 B) — *1. H en ce:
3. —i - i ( A 3 B ), —i —i A b 'B — 1, 2.
§2 6 EQUIVALENCE, REPLACEMENT 115
*24. Taking A D B, B D C, A D C for A, B, C, respectively, in *22:
1. ( A DB ) D ( ( BD C ) D ( A D C ) ) , - i - i (A d B), n n ( B D C ) b
“i (A D C). But:
2. h (ADB)D ((B D C ) D ( A D C)) — D-introd., *6.
*25. Taking A, B, A & B for A, B, C, respectively, in *22, and using
Axiom Schema 3 (§ 19), —\ - i A, nB b "‘(A & B). Applying *12
twice to Axiom Schema 4a, b —ii (A & B) D n n A ; etc.
R eplacement. Let A be a formal expression. Consider another
formal expression C. It may happen that A occurs as a (consecutive)
part of C; indeed, this may happen in more than one way. Suppose that
it does happen, and that, if it happens in more than one way, a particular
occurrence of A in C has been specified. We now denote C, with a par
ticular occurrence of A in C specified, by “CA”. In juxtaposition notation,
CA is EAF, where E and F are the parts (possibly empty) which precede
and follow the specified part A. Now let B be a formal expression. The
result of replacing the specified part A of C by B is the expression EBF.
This we denote by “CB”.
Contrast this definition of replacement with the definition of sub
stitution given in § 18. Replacement takes place for a specified occurrence
of an expression consisting of one or more symbols. Substitution takes
place for all occurrences of a single symbol, unless there is a distinction
between ‘free’ and ‘bound' occurrences, in which case it takes place for all
free occurrences. (In § 25 we used replacement in all occurrences. That
is equivalent to replacement, as now defined, applied successively to
each of the original non-overlapping occurrences of an expression in an
expression.)
E xample 1. If A is <3 D B, CA is (<b? D B) & -i((eT "DB) V icC7)
and B is -ic2f V B} then CB is («TDS)&n ((-i V B) V -ic3).
The foregoing definition of replacement is stated for formal expressions
in general. For the case that A, CA and B are proposition letter formulas
(as in Example 1), we have the following situation (as can be demonstrated
rigorously by applying the analysis of the scopes of operators (§ 17) to
the definition of proposition letter formula (§ 25)): The formula C.v can
be built up from the specified part A by applications of Clauses 2—5 of
the definition of proposition letter formula, and Cr> can be constructed
from B by parallel steps. The number of steps in this construction of C v
from A, after A is given and exclusive of the steps required to build up
the parts not containing the specified occurrence of A, we call the depth
o f th a t occurrence o f A in C A. In oth er w ords, th e d e p th of th e p a rt A in

C A is th e n u m ber o f o perators w ith in th e scopes of w h ich it lies.
E xample 1 (continued). T h e parallel co n stru ctio n s of C A from A

a n d o f C B from B are as follow s, an d th e d e p th is 3.
dfDS nc3VS
3® ) V-I<3
( a (-|J? V V-i<3 B )
-i((c3 3 B) V-i«3) -i((-i«£¥VS)V -i«3)

(<33 8) &-i((<3 3 S) V-i<3) (<33 8) V8) V-i<C
?)
T heorem 6 . / / A , B , C A and C B are proposition letter form ulas related
as in the foregoing definition of replacement, then A ~ B b C A ~ C B.
(R ep la c em e n t theorem .)
P roof, b y in d u ctio n on th e d ep th of A in C A, ta k in g th e A an d B fix ed

for th e in d u ctio n . T h e in d u ctio n proposition is th a t w h a t is sta te d in th e
th eorem is true, w ith th e fix e d A an d B , for e v e r y C A in w h ich th e specified
occurrence of A is a t d e p th d. B asis : A is a t d e p th 0 in C A'. T h e n C A is A ,
C B is B , an d th e co n clu sio n of th e theorem is sim p ly A ~ B |- A ~ B,
w h ich h olds as a gen eral p ro p e r ty of h I nduction step : A is a t d e p th
d -\-1 in C A. A s h yp o th esis of th e in d u ctio n , A ^ B h MA ~ M B for
a n y proposition le tte r form u la M A in w h ich th e specified occurrence
of A lies a t d e p th d. N ow m u st h a v e one of th e seven form s M A 3 N,
N 3 Ma, Ma & N , N & M a , M a V N , N V M a , - i M a , w here M A an d N are
p roposition le tte r form ulas, an d A is a t d e p th d in M A. B y th e h yp o th e sis
of th e in d u ctio n , A ~ B b MA ~ M B. Furth erm ore, b y th e ap p ro p riate
one of th e fo llo w in g lem m as (ta k in g M A as th e A , M B as th e B , an d N as
th e C of th e lem m a), M A ~ M B b C A ~ C B. Therefore A ~ B b CA ~ C B.
Lemmas for replacement. I f A, B and C are form u las :

*26. A ~ B b A D C ^ B D C , *2 7 . A ~ B b C D A ^ C D B .
*28a. A ~ B b A & C ^ B & C . *28b. A ~ B b C & A ^ C & B .
*29 a. A ~ B b A V C ^ B V C . *29 b. A ~ B b C V A ^ C V B .
*30. A ~ B b ^ A ^ -iB .
P roofs.
*26. * 1 . A ^ B bB 3 A — &>elim. (*17 b ).
2. A ~ B b (A 3 C) 3
(B D C) — *6, 1.
3. A ~ B 3 (A 3
b (B 3 C) C) — sim ilarly, usin g * 1 7 a an d * 6 .
4. A ~ B b A 3 C ~ B 3 C — & -in trod. (*16), 2, 3.
*2 7 . S im ilarly, u sin g * 1 7 a an d * 7 , th en * 17b an d *7.

§26 EQUIVALENCE, REPLACEMENT 117
E xample 1 (concluded). Let ~ be w ritte n b etw ee n each of th e
four pairs of form ulas in th e parallel co n stru ctio n s of C A an d C B. T h e
second of th e resu ltin g four form ulas is d ed u cib le from th e first b y *29 a,
th e th ird from th e second b y *30, an d th e fo u rth from th e th ird b y *28b.
C o m b in in g these d ed u ctio n s co n se cu tiv e ly , w e h a v e th e d ed u ctio n of
CA ~ C B from A ^ - B w h ich is g iv e n b y th e m eth o d of proof of T h eo rem 6 .
T h eorem 6 has b een sta te d in term s of proposition le tte r form ulas.

B u t b y th e su b stitu tio n rule (Theorem 3 § 25 ), w e can a p p ly it for oth er
senses of form ula, p ro v id e d t h a t th e A , B , C A o f th e ap p lica tio n ca n b e
o b ta in ed from proposition le tte r form u las by s u b stitu tin g form ulas
sim u ltan eo u sly for th e proposition letters. H en ce, or d ir e c tly b y th e
m eth o d of th e a b o v e proof of T h eorem 6 :
T heorem 6 (second version). I f A and B are form ulas , C A is a

form ula constructed from a specified occurrence of A using only the operators
D , & , V, - i , and C B results from C A by replacing this occurrence of A by B ,
then A ^ B b C A ~ C B.
E xample 2. L e t x b e a va riab le, an d A , B an d C(x) b e form ulas.
B y th e theorem , A ~ B b A V V x ( A D C(x)) ~ B V V x ( A D C(x)). (This
can b e considered as co m in g from <C? ~ B f-<3fV C^2V C b y su b
s titu tin g A , B , V x ( A D C(x)) for e^f, B, C, re sp ectively.) B u t our present
m eans are in a d e q u a te for d e d u c in g A V V x ( A d C(x)) ~ A V V x ( B D C(x))
from A ~ B . In th is th e occurrence of A to b e replaced is w ith in th e
p art V x ( A D C (x)), an d so th e C A ca n n o t b e b u ilt u p from th e occurrence
of A b y u sin g o n ly th e operators D , & , V, - i .
Corollary. U nder the conditions of the theorem (in either version),

A ~ B, CA b C B. (R ep lacem e n t p ro p e rty of equivalence.)
F ro m th e theorem b y * 18 a . (C onversely, th e theorem is o b ta in e d from

th e co ro llary an d * 1 9 b y ta k in g C a ~ C a as th e C a of th e corollary. T h e
theorem in clu d es th e lem m as, as th e cases w hen th e d e p th is 1 .)
O u r results h a v e been d e v e lo p e d to g iv e replacem en t of a single o c
currence o f A a t a tim e. B y ite ra te d ap p licatio n s, w e ca n th e n replace
a n y set of occurrences.
Chains of equivalences. W e can n o w present d em on stration s of

eq u ivalen ce b etw ee n proposition le tte r form ulas in th e fo llo w in g a b
b re v ia te d w a y . L e t us w rite
b C0 ~ Cx ~ ~ Cw ~ C n,
w here for each i [i = 1, . . . , n) e ith e r :

(a) C* is th e sam e form u la as Q _ x; or
(bj) b Q_i ~ C< or (b2) b C< ~ C , . , ; or
(c) C i com es from b y replacin g one or m ore occurrences of A t
b y B f w here (1) h B j or (2) b Bj ~ A t.
T h e n w e can regard " C 0 ~ Cx ~ ^ Cn_x ~ C n” as an a b b re v ia tio n
for ( . . . ((C0 ~ C x) & ( Q ~ C 2)) & . . . & (C n_2 ~ C n-1)) & (C n ^ ^ C M) ;
an d w e can u n d ersta n d t h a t b C s ~ Ck for all pairs j, k (j, k — 0, ...,#).
F o r w e h a v e for each i either b ~ C j or b C f ~ Q . ^ b y * 1 9 in C a s e ( a ) ,
im m e d ia te ly in Case (b), an d b y a p p licatio n s of T h eorem 6 in C ase (c).
T h en w e h a v e b C, ~ Ck for all oth er pairs j, k b y * 19, *20 an d *2 1 . T h e
m eth o d ap p lies likew ise w hen w e h a v e a n y list T of assu m p tio n form ulas
w ritte n before th e sy m b o l " b ” th ro u gh o u t. W e m a y insert e x p la n a to r y
rem arks in b ra c k e ts b etw ee n lin k s of th e chain. (For an ex a m p le see th e
proof of * 5 7 in § 27.)
T h e ch ain m eth o d w ill a p p ly also w hen w e h a v e in stea d of ~ some-
other relation al sy m b o l for w h ich w e h a v e estab lish ed th e corresponding
re flex ive, sy m m etric, tra n sitiv e an d replacem ent properties. Fu rth erm ore
in th e absen ce of som e of th ese properties (except tr a n s itiv ity ) it ca n b e
m odified to a p p ly , as follow s. W h e n s y m m e tr y is ab sen t, o m it (b2) an d
(c) (2), an d require th a t j < k. W h en reflexiven ess is also ab sen t, om it
also (a), an d require th a t j < k. W h en re p la c e a b ility is ab sen t, o m it (c).
(E x a m p le s from C h a p te r V I I I : w ith all properties present, = ; la c k in g
s y m m e tr y an d re p la c e a b ility , < ; la c k in g also reflexiveness, < .)
§ 27. Equivalences, duality. T heorem 7. If A, B and C are

form ulas:
*31. b (A & B )& C ~ A & (B & C ). *32. h (A V B) V C ~ A V (B V C).
*33. b A & B ~ B & A. *34. b A V B ~ B V A .
*3 5 . h A & (B V C) ~ *36. b A V (B C) & ~
(A & B) V (A & C). (A V B) (A V C). &
*3 7 . r A & A —' A . *38. b A VA ~ A.
*39. b A & (A V B ) ~ A . *40. b A V (A & B) ~ A.
(Associative, commutative, distributive, idempotent and
elimination laws.)
*4 1. A r A DB ~ B. *42. B b A o B ~ B .
~i A ~>A. - tB D B ~ —iA .
•&
4b.
CO
r .A D B ~ *44. b A
*45. B l A & B ~ A. *46. B b A V B ~ B .
*47. -i B r A & B ~ B. *48. -iB b A V B ~ A.
^Special cams of implication, conjunction and disjunction.)
§27 EQUIVALENCES, DUALITY 119
*4 9 °. b -1 - i A ~ A.
♦ 50. b - i ( A & -iA ). *5 1°. b AV-iA.
(L a w of d ou ble n egatio n , d en ial o f co n tra d ictio n , la w of

th e ex clu d e d m iddle.)
*5 2 °. b A & ( B V - i B ) ~ A. *5 3 . b A V ( B & iB ) ~ A.
*54. b A & B & -iB ~ B & -iB . ♦ 55°. b A V B V - i B ~ B V - i B.
(For sim p lify in g a d isju n ctio n o f co n ju n ctio n s, or a co n ju n ctio n
of disjunction s.)
*5 6 °. b A V B ^ n ( i A & n B ). * 5 7 °. b A & B ~ n ( - i A V -iB ).

♦ 58°. b A D B ^ -i(A & n B ). *5 9 °. b A D B ^ v n A V B .
*60°. b A & B ~ -» (A D - i B ) . *6 1°. b A V B ^ i A d B.

(E a c h tw o of D , & , V in term s of th e o th er a n d - i .)
*62°. b i ( A & B) ~ —i A V - i B . ♦ 63. b -i(AVB) ~ - i A & - i B.
(Transfer of - i across & an d V (D e M o rgan 's law s, 18 4 7 ).)
♦ 49a. b A 3 - i - i A. *49b. b “ i -i A ~ -» A .
*49c. b A V - 1A 3 (-1 - 1A 3 A ) ; he:nee b A V - 1A 3 (-1 - i A ~ A ).
*50a. b ~ i(A ~ -iA ). * 5 la . b -i-i(AV-iA).
*56a. b A V B D - i ( - i A & - i B). *5 1b . b -1 —1(—1 - i A 3 A ).
*56b. b -iA V B 3 - i( A & - i B). *57a. b A & B 3 - i ( - 1A V - i B ) .
*58a. b (A 3 B) 3 - . ( A & - i B ) . *5 7 b . b A & - 1B 3 - i ( - i A V B ).
*58b-dL b A D —1B ^ - i (A & B ) ~ n n A D nB ^ n n ( n A V —iB ).
*58e, f.. n n B D B b n n A D B ~ A D B ~ n (A & n B ).
*58 g. b ( - 1 - 1A D B ) 3 - i (A & - i B ) . *5 9 a . b - i A V B 3 ( A 3 B).

*60a. b A & B 3 -1 (A 3 - i B ) . *5 9 b . b (A 3 B) 3 - 1 —i ( - i A V B ) .
*60b. b A & - 1B 3 - i ( A 3 B ). *59c. b ( - i A 3 B) 3 - i - i (A V B ) .
*60c. b n -iA & B D n (A D n B ). *6 1a . b A V B 3 ( - i A 3 B ).

*6 0 d -f . b n n A & iB ^ n (A DB) ~ n(nAVB) ~ n n (A & "iB ).
*60g-i., b 1 n ( A D B ) ^ n ( A & n B ) o*' A D 1 nB ^ n n A D n “iB .
*62a. b 1 A V - 1B 3 - i ( A & B ) . *6 1 b . b i ( A V B ) ------ - (—1A 3 B).
P roofs for the classical system , e x c e p tin g *32, * 3 4 , *36, *38, *40,
*5 3 , *5 5 . A little w o rk m a y be sa v e d if one chooses b y p o stp o n in g pro o f
of th ese sev en u n til d u a lity (C orollary T h eo rem 8 ) is a v a ila b le.
*35. 1. A, B |- A & B b (A & B) V (A & C) — & -in tr o d „ V -introd.

2. A, C |- A & C |- (A & B ) V (A & C) — & -in tro d ., V -introd.
3. A, B V C b (A & B) V (A & C) — V -elim „ 1, 2.
4. A & (B V C) b (A & B ) V (A & C) — & -e lim „ 3.
5. b A & (B V C) Z> (A & B) V (A & C) — 3 -in tr o d ., 4.
6. A & B b A — & -elim .
7. A & B — & -elim ., V -introd.
b B b B V C
8. A & B & (B V C) — & -in tr o d „ 6, 7.
b A
9. A & C b A & (B V C) — sim ilarly.
10. (A & B) V (A & C) b A & (B V C) — V -e lim „ 8, 9.
11. b (A & B) V (A & C) 3 A & (B V C) — D -in tr o d ., 10.
12. b A & (B V C) ~ (A & B) V (A & C) — & -in tr o d . (*16 ), 5, 11.
*49. 1. b i i A D A — - i - e l i m ., D -in tro d . (or A x io m S ch em a 8).

2. A , “ 1A b A.
3. A, -< A b “IA .
4. A b ~ iA — -i-in tr o d ., 2, 3 ; etc.
*51. 1. -i(AV-iA), A b AV-iA — V -introd.

2. -i(AV-iA), A b-«(AV-iA).
3. -i(AV-iA) — -i-in tr o d ., 1,
b “iA 2.
4. - i ( A V " i A ) b —i—i A — sim ilarly.
5. b “ i “ i (A V - i A ) — -i-in tr o d ., 3, 4.
6. b A V - i A — —i-elim ., 5.
R emark 1. T h u s in th e form al sy ste m w ith o u t A x io m S ch em a 8,

-h B d B h A V -*i A w here B is A V A . C o n versely, in th e in tu itio n istic
sy ste m , A V - i A b n n A DA, th u s:
1. A h n A D A — -*11.
2. -iA b “n A D A - *10 b .
3 ^ y b n nA D A — V-elim ., 1, 2.
T h u s eith er of i “i A D A or A V n A can b e chosen as th e one non-

in tu itio n istic p o stu la te of th e classical system .
*5 2 . By *4 5 an d * 5 1 .
*54. S im ilarly, b y * 4 7 an d *50. P aren th eses h a v e been o m itte d in th e

result, since b y *3 1 it is im m a terial w h ich w a y th e association is tak en .
*56. 1. A, n A & —iR b A.

2. A, n A & -iB b nA — & -elim .
3. A b n (“ i A & ~ iB ) — -i-in tr o d ., 1, 2.
4. B b n ( - i A & -iB ) — sim ilarly.
5. b A V B D -i(-iA& -iB) — V -elim ., 3, 4, 3 -in tr o d .
6. (A V B ), A b A V B — V -introd.
7. -i(A VB) , A b ~ i(A V B).
8. - i ( A V B ) b ~ *A — -i-in tr o d ., 6 , 7.
9. i(A V B ) b ~ iB — sim ilarly.
10. b i(A V B ) 3 -1 A & - 1 B — & -in tro d . 8 , 9, 3 -in tr o d .
11. b “ i (“i A & - i B ) D A V B — co n trap o sitio n (*14 ), 10 .
*5 7. P resen ted as a ch ain of eq u ivalen ces: b -i(-iA V - i B ) ^

- i -i( -i - i A & -i -iB ) [*56] ~ - i n A & n - i B [*49] ~ A & B [*4 9 ].
*59-63. S et th ese u p b y th e ch ain m eth od.
P roofs for the intuitionistic system . *49c. B y R e m a rk 1; b y *49a,

* 1 6 and * 1 7 a , b - i “ i A D A ~ (-i - i A ~ A ). * 5 1 a . B y th e p roof of *5 1
o m ittin g S te p 6 . * 5 lb. A p p ly * 1 2 tw ic e to *49c, an d use * 5 la. *58 d .
U s e *63. *60f. U se *25.
To interchange tw o expressions A an d B th ro u gh o u t a th ird expression

C is to replace in C, sim u ltan eo u sly, all occurrences of A b y B an d all
occurrences of B b y A (exam ples w ill follow ).
T heorem Let D be a proposition letter form ula constructed from the

8 °.
distinct proposition letters P x, . . . , P w and their negations ~ iP x, . . . , ~ » P m
using only the operators & , V. Then a form ula D t equivalent to the negation
- i D o/D is obtained by the interchange throughout D of & with V and of
each letter with its negation.
In other words , if D be such a proposition letter form ula , and D t be the
result of the described interchange perform ed on D : b ^ D t.
E xample 1°. Taking -n<3f & (-iB V B) as the D, -iD is equivalent

t o e 3 V ( S & - i 8 ).
P roof am o u n ts esse n tia lly to th is: th e -i of -iD ca n b e tran sferred

p ro gressively to th e interior b y a p p licatio n s of *6 2 an d *63, an d a n y re
su ltin g d o u b le n egatio n s th en disch arged b y a p p lica tio n s of *49, in d o in g
w h ich D is tran sform ed in to D t (details to follow).
E xample 1 (concluded).
b -i( ( -iBV B)) ~
n n c3 V n ( nS V B) ~
n V (—i “ i B& —\B) ~
<C?V (B & - 1 8 ).
T o g iv e th e proof m ore e x p lic itly , w e ta k e as th e in d u ctio n n u m b er

(for a co u rse-o f-valu es induction ) th e num ber of occurrences of & an d

V in D ; ca ll it th e grade of D .
B a sis : D is of gra d e 0. T h e n for som e proposition le tte r P , D is P or D is
n P . C a se 1: D is P . T h e n b - i D ~ - i P [case h yp o th esis] ~ P t [def
in itio n of t] ~ D t [case h yp o th esis]. Case 2: D is - i P. S im ilarly, usin g
*49. I nduction step : D is of grade g + 1 . T h e n for som e proposition le tte r
form u las A an d B of th e t y p e under con sid eration w ith grades < g, either
D is A & B or D is A V B . C a se 1 : D is A & B . T h e n 1- -n D ~ “ i (A & B)
[case h yp o th esis] ~ - i A V - i B [*62] ~ A t V B t [h yp o th esis of th e in
d u ctio n ] ~ (A & B ) t [definition of t] ^ D t [ca seh y p o th esis]. Case 2: D is
A V B . S im ilarly.
E xample 2°. B y th e su b stitu tio n rule (Theorem 3 § 25), su b stitu tin g

B, - i e 3 V C for e2, B in th e result of E x a m p le 1,
b n(nS&(n[nc3VqV[nc2VC]))-SV([n^Vq&n[K3Vq).
In fa ct, for a n y form u las A an d B ,
b n(nA & (nB V B)) -A V (B & iB ).

A s th is e x a m p le illu strates, th e theorem ca n also b e sta te d usin g a n y
form ulas Av . , . , A m in p la ce of th e proposition le tters P x, . . . , P W,
p ro v id e d th e A lt . . . , A w retain th eir id e n tity th ro u g h o u t th e co n stru ctio n
o f D , an d are h eld in ta c t in th e in terch a n ge operation. (Second versio n of
T h eo rem 8 .)
Corollary0. A n equivalence between two letter form ulas and F of E

the type described in the theorem is preserved under the interchange throughout
E and F of & w ith V.
In other words , if E and F be two such proposition letter form ulas , and E '
and F ' be the results of the described interchange perform ed on E and F ,
respectively : I f b E ~ F , then b E ' ~ F '. (Principle of d u a lity .)
E xample 3°. B y *5 2 , h & (B V -i B ) ~ £1. H en ce (ta k in g
c 3 & ( B V - i B ) as th e E a n d as th e F ) b ^ V (B & - i B ) ~ <3.
P roof. B y h yp o th esis, b E ~ F . L e t us su b s titu te for each p ro p o

sitio n le tte r P w ith in E an d F th e n eg atio n - i P o f th a t letter, in d ic a tin g
th is s u b s titu tio n op eratio n b y B y th e su b stitu tio n rule (Theorem 3
§ 25), th e n b E * ~ F * . N e x t let us replace w ith in E * an d F * each d o u b ly
n e g a te d le tte r —i —i P b y th e sim ple le tte r P , in d ic a tin g th is o p eratio n b y
. B y th e la w of d o u b le n e g a tio n (*49) an d th e re placem en t p ro p e r ty of
e q u iv a le n c e (C orollary T h eo rem 6 ), b E *t ~ F * t . T h e e ffect of these tw o
op eratio n s is to in terch a n ge th e proposition letters w ith th eir n egatio n s
in th e g iv e n eq u ivalen ce. N o w b y T h eo rem 6 (or *30), b - i E * t ~ -iF *t.
E v a lu a t in g th e n egatio n s b y T h eo rem 8, b E *tt ^ F * t t . T h ese la st
tw o steps effect th e in terch a n ge of & w ith V, an d in terch a n ge th e
p roposition letters w ith th eir n egatio n s a second tim e to restore th e m
to th eir original con dition . W h a t w e now h a v e is therefore b E' ~ F\
b E ~ F. b <3&( 2 V -i 2 ) ~ c3.
b E* ~ F *. b -ic3&(-i2V -ii2 ) ~ -1 <3.
1- E * t ~ F * t . b ( —1 2 V 2) ~ —1 a .
b -iE * t~ -iF * t. b -i ( - i < v ? & ( - i 2 V 2)) ~ m e X
b E*tt ~ F * t t , i.e. b E ' ~ F ' . b c3V( 2 & -.2) ~ a.
E xam ple 4°. S u b s titu tin g a n y form ulas A , B for <C2, $ in th e result
o f E x a m p le 3 (b y T h eorem 3 § 25), bA V (B & - i B) ~ A . T h is is *53.
S im ila rly *32, *34, *36, *38, *40, *5 5 follow from * 31 , * 3 3 , *3 5 , *3 7 ,

*3 9 , *54, re sp e ctiv e ly , b y d u a lity (C orollary T h eo rem 8 ), w h en A , B , C
are sim ple proposition le tt e r s ; an d th en ce b y th e su b stitu tio n rule (T h eo
rem 3), w hen A , B , C are a n y form ulas.
E xam plI 5°. B y d u a lity (Corollary T h eo rem 8 ):

(a) If b VB ~ <3, th en b^ ® ~
B u t w e ca n n o t infer th e fo llo w in g :
(b) ‘ T o r a n y form ulas A an d B , if bA V B ~ A , th en bA & B ~ A."
(Indeed, t a k in g <3, B & -iS for th e A , B in (b), “ bA V B ~ A ” b eco m es
“ b V (2 & - . 2 ) ~ e T \ w h ich is tru e b y * 5 3 ; w h ile “ bA & B ~ A”
b ecom es “b w h ich is false as w ill b e sh ow n in
E x a m p le 4 § 28.) E x p la in . (H ow does th e present situ a tio n differ from
E x a m p le s 2 an d 4?) — T o s ta te a second version of th e corollary, w e m u st
require th a t th e Av . . . , A m b e d istin ct prim e form u las (cf. T h eo rem s 3
an d 4). — Since (b) is false, b y T h eorem 3 so is:
(C) “e 2 V £ ~ e 3 b
T h u s d u a lity holds o n ly as a su b sid iary d ed u ctio n rule, n o t as a d irect rule.
T h e recogn ition of d u a lity in lo gic goes b a c k to Schroder 1 8 7 7 .

In a p p ly in g d u a lity to a form u la th a t has been a b b re v ia te d b y o m ittin g
parentheses under th e co n v e n tio n of § 17, w h ich ranks & ah ead of V, care
m ust be ta k e n to sh ow th e scopes of th e operators w ith o u t ch an ge in th e
result. (H ence w e u su a lly prefer in th is co n n ection n o t to a p p ly th e co n
ve n tio n b etw ee n & an d V.)
A s an exercise, th e reader m a y v e r ify th e fo llo w in g ad d en d u m to th e
corollary, b y re ex a m in in g th e proof.
Corollary (second p a r t)0. A lso: I f (- E D F, then h F' D E '. (D u a l

con verse relationship.)
E xample 6 °. T h e a x io m < 3 & ® D c 3 h as as d u al-co n verse th e a x io m

«£? D < 3 V S . T h e p ro v a b le form ula d & S D J J V S is its ow n d u a l
converse.
§ 28. V a l u a t i o n , c o n s i s t e n c y . Sin ce se ttin g u p th e form al sy ste m ,

or in th is ch a p te r a su b sy ste m , our m e ta m a th e m a tic a l in v e stig a tio n s
h a v e been d e v o te d c h ie fly to estab lish in g th e p r o v a b ility o f ce rta in for
m u las, an d th e d e d u c ib ility o f certa in form ulas from oth er form ulas, i.e.
to d e v e lo p in g lo gic an d m a th e m a tics w ith in th e form al sy ste m . T h is is
a n ecessary p a rt of our program , an d co n trib u tes to sh o w in g t h a t th e
fo rm al sy ste m does c o n s titu te a fo rm alizatio n of a certa in p a rt of m a th e
m atics.
B u t also in m e ta m a th e m a tic s question s are a sk ed w h ic h relate to a
fo rm al sy ste m as a w hole. O n e o f these is th e q u estio n of th e 'co n siste n cy '
o f th e sy ste m , w h ic h is fu n d a m en ta l in H ilb e rt's program (§ 14).
T h e p ro p o sitio n al ca lcu lu s (and gen erally, a n y form al sy ste m h a v in g
th e s y m b o l - i for n eg atio n ) is said to b e [sim ply) consistent, if for n o
fo rm u la A are b o th A an d - i A p ro v a b le in th e s y s t e m ; an d to b e [sim ply)
inconsistent in th e c o n tr a ry case t h a t for som e form ula A , b o th h A a n d
b -iA .
T h is is a s tr ic tly m e ta m a th e m a tic a l d efinition. I t refers o n ly to th e
fo rm al s y m b o l - i , an d to th e d efin ition s of fo rm u la an d p ro v a b le form ula.
I t th u s b ecom es an e x a c t m a th e m a tic a l problem , w h ich w e ca n consider
in m e ta m a th e m a tics, to p ro v e th e c o n sisten cy of a g iv e n fo rm al sy ste m .
T h e d efin itio n an d p rob lem of co n sisten cy ta k e on sign ifican ce from
o u tsid e th e m e ta m a th e m a tic s, u nder th e in terp retatio n of th e form al
sy ste m as a fo rm alizatio n o f an inform al th e o ry, w ith th e sy m b o l - i
exp ressin g n egatio n . T h e propositions expressed b y tw o n u m b er-th eo retic
form u las A an d - i A , if A does n o t co n ta in free v a riab les (or th e p rop o
sitions expressed for each p a rticu la r set o f va lu e s of th e va riab les, if A does
co n ta in free va riab les), ta k e n to g e th e r co n stitu te a co n trad ictio n . L ik e
w ise, in th e case o f p roposition le tte r form ulas, on in terp retin g th e propo
sition letters b y a n y p articu lar propositions. A m e ta m a th e m a tic a l proof
o f th e c o n sisten cy of th e form al sy ste m w o u ld h en ce affo rd se c u rity
a g a in st a c o n tra d ictio n 's arising in th e inform al th eo ry.
F o r th e p rop osition al calcu lu s (and gen erally, for a n y form al sy s te m
w h ich h as & -e lim in a tio n an d w e a k -elim in ation as p o stu la te d or d e riv e d
rules), th e a b o v e defin ition is eq u iv a le n t to th e fo llo w in g. T h e sy ste m is
§28 VALUATION, CONSISTENCY 125
(sim p ly) consistent , if there is som e im p ro va b le fo rm u la; (sim ply) in -
consistent , if e v e r y fo rm u la is p ro vab le. F o r if b o th bA an d b ” iA ,
th en b y use of w e a k -elim in atio n (§ 23), b B for e v e r y form u la B . In th e
case o f co n sisten cy, A & - i A is an ex am p le o f an u n p ro va b le form ula,
since otherw ise b y & -e lim in atio n b o th bA an d b ""»A.
T h e d efin ition of co n sisten cy (in th e first form) an d th e d efin itio n of
p ro v a b le form ula (say m th e first form , § 19) su ggest a p lan of a t ta c k on
th e problem of p ro v in g co n sisten cy (not th e o n ly possible plan). Suppose
w e ca n fin d a m e ta m a th e m a tic a l p ro p e rty of form ulas such th a t (a) th e
a xio m s h a v e th e p ro p erty , (b) if th e prem ises for an a p p lica tio n of a rule
of inference h a v e th e p ro p erty , so does th e conclusion, an d (c) tw o fo rm u
las of th e form s A an d - i A ca n n o t b o th h a v e th e p ro p erty. T h e n usin g
(a) an d (b) e v e ry p ro v a b le form ula w o u ld h a v e th e p ro p erty , an d b y (c) th e
sy ste m w o u ld be con sistent. In th is section w e sh all g iv e a m e ta m a th e
m a tica l proof o f co n sisten cy for th e propositional calcu lu s follow in g th is
plan.
T h e p ro p e rty of form ulas w h ich w ill b e used is su ggested b y th e lo gical
in terp retatio n of th e propositional calculus. W e co n ceive of each propo
sition le tte r as a v a ria b le w hose va lu es are propositions, an d w e co n ceiv e
of these propositions as b ein g each either true or false. T h e operators of
th e calcu lu s 3 , & , V, -t form from these propositions oth er propositions
w hose tru th or fa ls ity w ill d epen d o n ly on th e tru th or fa ls ity of th e co m
ponen t propositions, accord in g to ta b les to b e g iv e n presen tly. (H ence th e
operators of th e calcu lu s are som etim es called T r u th -v a lu e fu n ctio n s of
p ro p o sitio n s’ .) T h e n it w ill ap p ear th a t th e p ro v a b le proposition le tte r
form ulas all h a v e th e p ro p e rty th a t th e y are id e n tic a lly true, in th e sense
th a t th e y represent true propositions for all possible p erm u tatio n s
of true an d false propositions as va lu es of th e proposition letters co n ta in ed
in them .
T o use this id ea for th e purpose of a m e ta m a th e m a tic a l co n siste n cy
proof for th e calculus, it is n ecessary to a v o id th e reference to 'p ro p o
sitio n s’ , 'tr u th ’ an d 'f a ls ity ’ , w h ich h a v e co n n o tatio n s th a t are extran eo u s
to th e m eta m a th em a tics. T h is w e can do, since n o th in g essential in th e
argu m en t o u tlin ed a b o v e d epends on th e v a lu es of th e proposition letters
b ein g propositions, or on th e n atu re of tru th an d fa lsity , e x c e p t t h a t true
an d false propositions are d istin ct from each other.
T h e p u rely m a th e m a tica l ch aracter of w h a t w e sh all do m a y b e em
phasized b y an a n a lo g y to th e elem en tary school a rith m etic o f p o sitiv e
integers. W h ile th e num bers 1 , 2 , 3, . . . in that, arith m etic w ere in ten d ed
to h a v e a m ean in g for co u n tin g an d m easurem ent, so far as th e ad d itio n
an d m u ltip lic a tio n ta b le s w ere concerned th e y cou ld be a n y en u m eratio n

of d istin c t o b jects. From th is stan d p o in t, th e arith m e tic deals w ith
operations, i.e. fu n ctio n s, + an d *, o ver a d om ain o f o b je c ts {1, 2, 3, . . . } ,
an d d ep en d s o n ly on th e p o ssib ility of recogn izin g an d d istin gu ish in g
b e tw e e n th ose o b je cts, an d n o t on their intrin sic nature.
W e sh all n o w set u p an arith m e tic in like sense for a d om ain of o n ly
tw o o b je cts, w ith four fu n ctio n s D, & , V, - i . T h is describes su c c in c tly
w h a t w e sh all do. Sin ce for m e ta m a th e m a tics, D , & , V, - i are m eaningless
g iv e n o b je c ts, a m ore precise sta te m e n t of w h at w e sh all do is th e
follow in g. W e in tro d u ce a m e ta m a th e m a tic a l co m p u ta tio n process (called
a valuation procedure), b y w h ic h a fu n ctio n in th e arith m e tic (or a ta b le
for such a fu n ctio n , called a truth table) is correlated to each of th e sy m b o ls
3 , & , V, - i , an d th en ce to each proposition le tte r form ula. Then we
s tu d y m e ta m a th e m a tic a l properties of proposition le tte r form ulas d efin ed
in term s of th e correlated fu n ctio n s (or tables).
Sin ce o n ly th e d istin ctn ess of th e tw o o b je c ts {truth values) is required for
th e a b s tr a c t a rith m etic , it is im m a terial w h a t t h e y are called. W e m ig h t
d e sig n a te th em as " 0 ” a n d ' T \ or ” an d ” , or “ t ” a n d " I ” , or
“ t” an d “ f ” , etc. W e choose th e la st pair of sym b o ls, w h ich su ggest re
s p e c tiv e ly th e notions 'tru e ' an d 'false' of th e lo gical in terpretatio n .
W e b egin b y con siderin g th e proposition le tters as va riab les ran gin g
o ve r th e d om ain {t, f}.
W e th e n consider th e operators of th e calcu lu s as fu n ctio n s o ve r th is
d o m ain , d efin ed b y th e follow in g tab les, an alogo u s to th e ad d itio n an d
m u ltip lic a tio n ta b les in th e a rith m etic of p o sitiv e integers. F ro m th e tab le,
to a n y g iv e n v a lu e (s) of th e in d epen den t v a ria b le (s), th e corresponding
v a lu e of th e fu n ctio n ca n b e read.
<v? D S ^vs n <£?

s t f s t f s t f
t t f a t t f a t t t a t f
f t t f f f f t f f t
(Th e ta b le s for - i an d V are th e sam e as were g iv e n in § 10 E x a m p le 4
for ' an d *, w hen th e tw o o b je c ts are w ritte n " 0 ” an d " 1 ” .)
T h e n each p rop osition le tte r form u la A in a g iv e n list P x, . . . , P w of
d istin ct proposition letters represents a fu n ctio n of those letters regard ed
as in d ep en den t v a riab les o ve r th e dom ain {t, f}. T o each w -tu p le of va lu es
of th e letters, th e correspon ding v a lu e of the fu n ctio n can be co m p u te d
b y a series of a p p licatio n s of th e fu n d am en tal tables.
§28 V A L U A T IO N , C O N S IS T E N C Y 127
E xample 1. The le tte r form ula cT ~ B , i.e. 3 8 ) & (8 D <3)
(§ 26), represents th e fu n ctio n w h ich has th e follow in g tab le.
e2~2
2 t f
a t t f
f f t
T h e co m p u ta tio n of th e v a lu e entered in th e u pper righ t square is as

follows.
(<3f 3 2) & (2 3 <C?)
( t 3 f ) &( f 3 t )
f & t
T h e square arran gem en t of th e tab les is special to tw o va riab les (where

it helps to su ggest properties of th e functions). F o r m v a ria b les P 1# . . Pm
gen erally, w e ca n arrange th e 2 m possible m -tu p les of argu m en ts v e r tic a lly
one b elo w an oth er in som e fix e d order, an d w rite th e correspon ding fu n c
tion v a lu es in a value column op p o site these.
E xample 2.
a 2 c -1 [<3 V 2 3 (2 & C) V -i<CT]
t t t f
t t f t
t f t t
t f f t
f t t f
f t f f
f f t f
f f f f
A proposition le tte r form ula E in th e d istin ct proposition letters
P3, . . . , P m is said to be identically true, if th e v a lu e colu m n o f its ta b le
con tains o n ly t's ; identically false, if o n ly f s . T w o proposition le tte r
form ulas E an d F in Flt . . . , P m are said to b e identically equal, if their
tab les h a v e th e sam e v a lu e colum n. (In oth er w ords, an id e n tic a lly tru e
E represents th e co n sta n t fu n ctio n t, an id e n tic a lly false E th e co n sta n t
fu n ction f, an d id e n tic a lly eq u a l E an d F th e sam e function .)
Th ese d efinitions so sta te d a p p ly to E (to E an d F ) con sidered as
proposition le tte r form ula(s) in a specified list P lf . . . , P W of d istin c t
p rop osition letters, n o t all of w h ich (according to § 25) need o ccur in th e

form ula(s). W e m a y suppose first th a t P x, . . P m is th e m in im al eligible
list, i.e. th e list com p risin g e x a c tly those d istin ct proposition letters
w h ich o ccur in E (in E or in F ). T h e n if w e a d d to th e list o th er letters,
each row in th e table(s) for th e form ula(s) w ill sim p ly b e sp lit in to a
n u m b er o f row s ( 2k o f th e m if k letters are added) sh o w in g th e possible
a ssign m en ts of va lu e s to th ese a d d itio n a l letters, w ith o u t a lte rin g th e
c o m p u ta tio n process or v a lu e en try . H en ce if w e h a v e id en tica l tr u th or
id e n tic a l fa ls ity (id en tical eq u a lity ) w ith respect to th e m in im al list, w e
sh all h a v e it also w ith respect to a n y o th er list, a n d co n versely. T h erefo re
th e reference to th e list m a y b e o m itted .
T heorem A necessary condition that a proposition letter form ula E

9.
be provable (or deducible from identically true form ulas T) in the propositional
calculus is that it be identically tru e ; i.e. if j- E , then E is identically true .
P ro o f is b y co u rse-o f-va lu es in d u ctio n on th e le n g th of th e g iv e n p roof
o f E , u sin g th e tw o fo llo w in g lem m as.
L emma 12 a. A proposition letter form ula which is an axiom is iden

tically true.
P roof . F o r each of th e ten a x io m sch em ata of th e p ro p o sitio n al
ca lcu lu s (P o stu late s la , lb , 3 — 8 § 19, or in tu itio n istic a lly 81 § 23 in stea d
o f 8 ), w e ca n e a s ily v e r ify th e fo llo w in g fa c t b y co m p u ta tio n : T h e ta b le
w h ich is o b ta in e d b y assign in g th e v a lu e s t an d f in all possible w a y s
d ir e c tly to th e p arts A , B , C w h ich ap p ear in th e sch em a co n ta in s o n ly
t's in th e v a lu e colum n. T h is am o u n ts to v e r ify in g th e lem m a tre a tin g
th e A , B , C of th e sch e m a ta as sim ple proposition letters $ , C.
T h e tr u th of th e lem m a follow s from this. F o r consider a n y a x io m
w h ich is a p roposition le tte r form ula. T h is com es from one of th e sch em a ta
by ta k in g as th e A , B , C certain proposition le tte r form ulas in s a y
Pv . . . P w as jo in t list of d istin ct proposition letters. N o w no m a tte r
w h a t m -tu p le o f t ’s an d f ’s is assigned as va lu es to P 1# . . . P m, th e trip le
o f v a lu e s for A , B , C to w h ich it leads m u st in tu rn (as a lre a d y verified)
lead to th e v a lu e t for th e w h o le axiom .
L emma 12b. I f the prem ises for an application of the rule of inference
are identically true proposition letter form ulas , so is the conclusion .
P roof . ' B y in spectio n of th e fo llo w in g ta b le, w e see th a t th e o n ly
pair of va lu e s of A , B w h ich g iv e s th e v a lu e t to b o th prem ises A an d A D B
for th e rule o f inference of th e calcu lu s (P o stu la te 2) is th e pair t, t ; an d
th is pair d o es g iv e th e v a lu e t to th e conclusion B .
A B A A D B B
t t t t t
t f t f f
f t f i t
f f f t f
H en ce, if A an d B are p rop osition le tte r form ulas in P 2, . . P m, a n y
m -tu p le of va lu e s of P x, . . . , P m w h ich g iv e s b o th prem ises th e v a lu e t
m ust g iv e th is v a lu e also to th e conclusion. B y h y p o th esis, th e prem ises
h a v e th e v a lu e t for all w -tu p le s of va lu e s of P 2, . . . , P m. T h erefo re th e
conclusion h as also.
E xample 3. T h e form u la - i [<3 V B D (S & C ) V -i<£3f] is n o t p r o v

able in th e p roposition al calcu lu s, b ecau se there are f ’s in th e v a lu e co lu m n
of its ta b le (E x a m p le 2). L ikew ise, <3 V ( n c 3 & S) is n o t p ro v a b le , b e
cause it ta k e s th e v a lu e f w h en B ta k e th e v a lu es f, f.
Corollary 1 . A necessary condition that two proposition letter form ulas

E and F be equivalent is that E and F be identically equal ; i.e. if b E ~ F ,
then E and F are identically equal.
P roof . I f E an d F are e q u iv a le n t, th en by d efin itio n of e q u iva len c e,
E ~ F is p ro vab le. H en ce b y th e theorem , E ~ F is id e n tic a lly true. R e
ferring to th e ta b le for ~ (E x a m p le 1 ), w e see t h a t E ^ F c a n o n ly
receive th e v a lu e t w h en E an d F receive either b o th th e v a lu e t or b o th
th e v a lu e f, i.e. w hen t h e y receive th e sam e va lu e. Sin ce E ~ F is id en
tic a lly true, i.e. does a lw a y s receive th e v a lu e t, E an d F do a lw a y s receive
th e sam e v a lu e , i.e. th e y are id e n tic a lly equal.
E xample 4. & B & —iB an d are n o t e q u iv a le n t, b eca u se t h e y

ta k e d ifferen t v a lu e s (nam ely, f an d t, re sp ectively) w h en e.g. <3, B are
g iv en th e va lu e s t, t
Corollary 2 . The propositional calculus is {sim ply) consistent ; i.e.

for no form ula A, both b A and b “ »A.
P roof . U sin g th e second versio n of th e d efin itio n of sim ple c o n

sisten cy (above), th is is a lre a d y p ro v e d b y E x a m p le 3.
T o g iv e th e proof d ir e c tly from th e original d efin ition , suppose one of
A an d - i A is p ro v ab le. T h e n b y th e theorem it is id e n tic a lly tru e ; th e n
using the ta b le for - i , th e o th er is id e n tic a lly false, h ence n o t id e n tic a lly
true, an d h en ce b y th e theorem u n p ro vab le.
T h is estab lish es th e c o n siste n cy for th e p rop osition al ca lcu lu s in term s

o f proposition le tte r form ulas. I f A * an d - i A * b e p ro v a b le form ulas in
th e ca lcu lu s for an o th er n o tio n o f form ula, th en b y th e co n verse o f th e
s u b s titu tio n rule (Theorem 4 § 25), there w o u ld b e p ro v a b le p ro p o sitio n
le tte r form ulas A an d - i A . T h u s th e co n sisten cy e x te n d s to th e o th e r
senses o f form ula.
T h is co n siste n cy proof, o f course, does not h old go o d for th e a d d itio n o f

an o th er gro u p of p o stu lates, ev e n should th a t gro u p b y itself b e co n
sisten t.
T h e co n siste n cy proof for th e pro p o sitio n al ca lcu lu s w as first g iv e n b y
P o s t 1 9 2 1 . (Cf. also L u k a s ie w ic z 1 9 2 5 , H ilb e rt-A c k e rm a n n 19 2 8 .)
T h e a r ith m e tic of a d o m ain of tw o o b jects, used in th is section to g iv e
th e co n siste n c y proof, m a y be th o u g h t of as an in terp reta tio n o f th e
ca lcu lu s in th e fo llo w in g sense. T h e v a lu a tio n ta b le s p ro v id e a rith m e tic a l
sign ifican ces for th e operators of th e calculus. T h e p rop osition le tte rs are
con sidered as in d ep en d en t va ria b le s ra n gin g o ve r th e d o m ain {t, f} of th e
a rith m etic . E a c h p rop osition le tte r form u la is in terp reted as exp ressin g
th e p rop osition th a t its v a lu e is t for all choices o f th e va lu e s o f its in
d ep en d en t va riab les. B y T h eo rem 9, o n ly form ulas w h ich are tru e u nd er
th is in terp reta tio n are p ro v a b le . F o r th e proposition expressed b y a
fo rm u la u nder th is in te rp re ta tio n is e q u iv a le n t to th e fo rm u la’s h a v in g
th e m e ta m a th e m a tic a l p ro p e r ty of id en tica l tru th , as d efin ed a b o v e
in term s o f a c o m p u ta tio n procedure u sin g t an d f.
O n th e o th er h an d , in the {usual) logical interpretation , th e proposition
le tte rs are con sidered as in d ep en d en t v a ria b les ra n gin g o ver som e d om ain
o f propositions. A p ro p o sitio n le tte r form u la th e n expresses th e gen eral
p rop osition th a t all th e p a rticu la r propositions, expressed b y it for d if
feren t ch oices o f proposition s from t h a t d om ain as va lu e s of its in d ep en d
en t va ria b les, are true. T h e lo g ica l in terp reta tio n is re lated to th e
a r ith m e tic a l in terp reta tio n b y p u ttin g th e p a rticu la r propositions (from
th e d o m a in considered) in to m a n y -o n e correspondence w ith th e tw o
o b je c ts t, f, th o se propositions w h ich correspond to t b ein g tru e an d those
w h ic h correspond to f b e in g false. T h e proposition expressed b y a form u la
u n d er th is in terp reta tio n is n o t e q u iv a le n t to a m e ta m a th e m a tic a lly
d efin ab le p ro p e r ty o f th e form ula, e x c e p t for sp e cia lly restricted d om ain s
o f th e propositions allo w ed as va lu e s of th e proposition letters.
§ 29. C o m p le te n e s s , n o r m a l fo r m . A n o th e r problem w h ic h w e
m a y b e able to tre a t in m e ta m a th e m a tic s is th a t of th e ‘co m p leten ess’ o f a
§29 COMPLETENESS, NORMAL FORM 131
g iv e n form al system . F o r ex am p le , w e h a v e liste d eleven p o stu la tes for
th e p roposition al calcu lu s (§ 19). C an w e g iv e a reason w h y w e sto p w ith
ju st th ese ? M ig h t w e w ith a d v a n ta g e a tte m p t to d iscover others w h ich
could b e a d d ed to th e list to g iv e m ore p ro v a b le form ulas ? T o be able to
answer these question s, w e m u st first p ro v id e som e criterion as to w h a t
w e w a n t to b e able to p ro ve in th e system . D ifferen t n otions of c o m p le te
ness w ill result acco rd in g to th e criterion chosen.
W e m a y g iv e th e criterion a p o s itiv e form , an d s a y th a t th e sy ste m is
com p lete, if its p o stu la te list a lre a d y p rovid es all w e need for som e p u r
pose. F o r exam p le, suppose t h a t som e p ro p e rty has been d efined for for
m ulas of th e s y s t e m ; or a lte rn a tiv e ly , th a t an in terp retatio n has been
g iv e n to th e form ulas of th e sy ste m , in w h ich case th e p ro p e rty is th a t th e
form ulas express true propositions under th e in terpretatio n . R e la tiv e to
such a p ro p e r ty or in terp retatio n , d efinitions of b o th co n siste n cy an d
com pleten ess ca n b e g iv e n as follow s.
T h e sy ste m is consistent w ith respect to th e p ro p e rty (or in terpretatio n ),
if o n ly form ulas w h ich h a v e th e p ro p e rty (or express true propositions
under th e in terpretatio n ) are p ro v a b le. T h e sy ste m is complete w ith respect
to th e p ro p e rty (or in terpretatio n ), if all form ulas w h ich h a v e th e p ro p e rty
(or express tru e propositions under th e in terpretatio n ) are p ro v ab le.
U n lik e th e n o tio n of sim ple co n sisten cy g iv e n in th e p reced in g section ,
th ese notion s of co n siste n cy an d com pleten ess re la tiv e to a p ro p e rty or
in terp retatio n m a y n o t a lw a y s b elo n g to m e ta m a th e m a tics. W h eth e r or
not t h e y do w e sh all h a v e to consider from case to case, acco rd in g to
w h eth er th e p ro p e rty (or in terpretatio n ) is one w h ic h ca n b e fo rm u lated
w ith in m eta m a th em a tics.
F o r th e p rop osition al calcu lu s w e h a v e th e p ro p e r ty o f id en tica l tru th
(or if w e prefer, th e in terp reta tio n of th e ca lcu lu s as an a rith m etic of a
dom ain of tw o o b jects), w h ic h can b e fo rm u lated in m e ta m a th e m a tics.
Th eorem 9 is th u s a m e ta m a th e m a tic a l co n siste n cy theorem for a certain
p ro p erty of le tte r form ulas (or in terp retatio n of th e calculus).
T o re ca p itu la te th e id ea of com pleten ess re la tiv e to an in terp retatio n :
A sy ste m is co m p le te und er a g iv e n in terpretatio n , if th e d e d u c tiv e
p o stu lates (or tran sfo rm atio n rules) enable us to p ro ve in th e sy ste m all
the true propositions w h ich its form ation rules enable us to express in th e
system .
W e are led to other form u latio n s of com pleten ess, if w e g iv e th e criterion
for w h a t form ulas should b e p ro v a b le a n e g a tiv e form , an d s a y th a t th e
syste m is co m p lete, if th e p o stu la tes p ro vide all t h a t w e can afford to
h a v e lest som e undesirable effect ensue. A n effect w h ich com es to m ind is
sim ple in co n sisten cy. T h e com pleten ess n o tio n s o b ta in ed in th is w a y w ill

a lw a y s b e m e ta m a th e m a tic a l, if th e e ffect to b e av o id e d is m e ta m a th e -
m a tic a lly describ ab le (as in p a rticu la r sim ple in co n sisten cy is). E x a c t
fo rm u latio n s w ill b e g iv e n for p articu lar form al sy ste m s as w e com e to
th em .
N o tic e t h a t a co n siste n cy theorem w ill a lw a y s b e a theorem to th e
e ffe ct t h a t a t m o st such an d such form ulas are p r o v a b le ; an d a co m
pleten ess theorem one to th e e ffe ct th a t a t least such an d su ch form ulas
are p ro v ab le.
F r o m th is gen eral in tro d u c tio n to th e problem of com pleten ess, w e
tu rn n o w to consider it for th e p roposition al calculus. W e sh all sh ow first
(Th eorem 10) t h a t th e calcu lu s is c o m p le te w ith respect to th e p ro p e r ty
o f id e n tica l tru th .
T heorem 10° and Corollary 1°. The conditions of Theorem 9 and

Corollary 1 are sufficient (as w ell as n e c e ssa ry ); i.e. if E is identically true ,
then b E , and if E and F are identically equal, then b E ~ F .
T h e proof w ill b e b ased on tw o lem m as. L e t P x, . . . , P m b e d istin ct
p ro p o sitio n letters. G iv e n an w -tu p le of t's an d f's as va lu e s of P 1# . . . , P m,
b y th e corresponding letter m -tuple w e sh all m ean th e sequence Q lr . . . , Q m
o f letters an d n e g a te d le tte rs w here, for each j (j — 1 , . . . , m ), Q , is P ,
or - i P , acco rd in g as th e g iv e n v a lu e of P ? is t or f.
E xample 1. Let 3 , S, C ta k e th e va lu e s t, f, t, re sp e ctiv e ly . T h e

correspon ding le tte r m -tu p le is 3 , -iB ,.C .
Lemma 13. L et E be a proposition letter form ula in the distinct propo

sition letters P x, . . . , P w ; let an m -tuple of t's and f s be given as values of
P x, . . P m; and let Q lf . . . , Q m be the corresponding letter m -tuple. Then
Qi , . . . , Q m b E or Q lt . . . , Q m b “iE, according as for the given m -tuple
of values E takes the value t or the value f.
E xample 2. C orrespond ing to th e ta b le in E x a m p le 2 § 28 for
n [c3V S D (B & C) V - i 3 ] , w e n ow h a v e eigh t d e d u c tio n s :
37, 2 , \- —i—i[ 3 vs D (S & Q

C V —i 3 ] .
a, 2, b —i[ 3 vs D (S& C)
C V —13].
<3, -iS, c —i[ 3 vs D (S& Q V —13].
<3, -iS, b -i [ 3 vs D (S& Q
b
—i C V —i3 ] .
“13, 2, c b —i1 [ 3 vs D (B Q & V -i 3 ] .
—1(3, 2, “i c b —i—i[ 3 vs D (B& Q V —i3].
—1<3, -i 2, c b —i—i[ 3 vs D (B 0 & V —i3j.
—13, ->2, “i c b —i—i[ 3 vs D (8& Q V -» 3 J.
P roof of L emma 13, b y co u rse-o f-va lu es in d u ctio n on th e n u m b er of
(occurrences of) lo gical sy m b o ls in E ; ca ll th is n u m ber th e degree.
B a sis : E is of degree 0. T h e n E is for som e j. In th e g iv e n m -tu p le of
valu es, th e v a lu e of P , is t or is f . Case 1: th e v a lu e of P 3 is t. T h e n Q , is P 5,
i,e. Qj is E ; an d so b y general properties of h Qi, • • •, Qm b E , as w as to
b e sh ow n sin ce in th is case E ta k e s th e v a lu e t. Case 2: th e v a lu e of P , is f.
T h en sim ila rly Q ; is - i E , an d so Q x, . . . , Qm b “ *E , as w as to b e show n.
I nduction st e p : E is of degree d+ 1. T h e n for som e p roposition

letter, form ulas A an d B in P x, . . . , P w of degrees < d, E is A D B or
A& B or A V B or - i A . Case 1: E is A D B . F o u r subcases arise, acco rd in g
as for th e g iv e n m -tu p le of v a lu es of P x, . . P m, th e form ulas A , B ta k e
the resp ective va lu es t, t or t, f or f, t or f, f. S ubcase 2 : A , B ta k e th e va lu e s
t, f . T h e n (by th e u pper righ t e n try in th e ta b le for D ) E ta k e s th e v a lu e f,
so w e are to show th a t Q lt ..., Qm b - i E . B u t b y th e h yp o th e sis of th e
in du ction , Qv . . . , Q w b A and Q x, . . . , Q w b -» B . A lso , u sin g *4 1 or *44,
A, -iB b -i(A D B ) ; i.e. A , - i B b “ «E. H en ce Q lt ..., Qm b “ *E , as
w as to be shown. T h e tre a tm e n t of th e o th er cases an d subcases is
sim ilar, usin g * 4 1 —-*48, *49a.
L emma 14. Let E be a 'proposition letter form ula in the distinct propo
sition letters P x, . . . , P m. I f for each of the 2 m m -tuples of t's and f's,
Q lf . . . , Q m |- E where Q v . . . , Q m is the corresponding letter m -tuple ,
then P ^ - i P i , . . . , P w V - i P m b E .
P roof of L emma 14. B y 2 m- 1 + 2 m~2 + ... + 1 a p p lica tio n s of
V -elim in ation . F o r ex a m p le w hen m = 2, b y h y p o th e s is :
E i, P2 b E.
(a) P j,
-1 P i,
“i P 2 b E .
P2 b E.
- 1 P 1, n P 2 b E.
B y tw o ap p licatio n s of V -elim in atio n :
P i, P 2 V - i P 2 b E .
(b)
- i P j, P 2 V - i P 2 b E.
B y one a d d itio n a l ap p licatio n ,
(c) P iV -iP !, P 2 V - i P 2 b E,
as w as to b e shown.
P roof of T heorem 10. Say E is a proposition le tte r form u la in

P j, . . . , P m. B y h yp o th esis E is id e n tic a lly true, i.e. ta k e s th e v a lu e t
for each m -tu p le of t's an d f ’s as v a lu e s o f P lf . . . , P m. T h e n b y L e m m a

13, th e h y p o th e sis o f L e m m a 14 is satisfied ; so b y L e m m a 14,
P x V -iP v . . P TOV - i P m b E . T h e n c e u sin g * 5 1 , b E.
R emark 1. A ll b u t th e fin al ste p o f th is proof is go o d for th e in -

tu itio n istic s y ste m also. H en ce in th a t, for a n y le tte r fo rm u la E in
P i, . . . , P m‘
(a) P j V n iP j, . . P m V n P m b E , if E is id e n tic a lly true.
(b) P ^ -iP * . . . , P mV - i P w b E V - i E .
T h e theorem expresses th e com pleten ess of th e d e d u c tiv e rules o f th e

p ro p o sitio n al ca lcu lu s for th e purpose of estab lish in g le tte r form ulas
w h ic h are id e n tic a lly tru e under th e arith m e tic in te rp re ta tio n ; an d
C o ro lla ry 1 sa y s th a t th e d e d u c tiv e th e o r y g iv e s a co m p le te a c co u n t of
th e e q u a lity of fu n ctio n s d efin ab le in th e arith m etic.
Corollary 2°. The addition to the postulate list for the propositional
calculus of an unprovable letter form ula for use as an axiom schema would
destroy the sim ple consistency .
P roof. B y T h eo rem 10, th is le tte r form u la m u st receive th e v a lu e f
for som e set of v a lu e s of th e proposition letters w h ich it con tains. S e lect
such a set of va lu es, an d use th e n ew a x io m sch em a b y su b s titu tin g
V for th e le tters w h ich h a v e th e v a lu e t, an d «£¥ & - i <C2 for th ose
w h ic h h a v e th e v a lu e f. T h e resu ltin g new a x io m w o u ld b e id e n tic a lly
false. H en ce b y C o ro llary 1 it w o u ld b e e q u iv a le n t to <£? & i w h ich is
also id e n tic a lly false. So (using *18 a) <£? w o u ld also b e p ro v a b le,
a n d th e sy ste m w o u ld th u s b e incon sisten t (§ 28).
C o ro llary 2 sa y s t h a t th e p rop osition al ca lcu lu s is in cap ab le of b ein g

en larged b y p o stu la te s o f th e sam e ch aracter as those a lre a d y listed w ith o u t
d e stro y in g th e sim ple co n sisten cy . T h is is a com pleten ess p ro p e rty of th e
second ty p e described in th e in tro d u c to ry rem arks.
G iv e n a n m -tu p le o f t's an d f s as v a lu es of P lt . . . , P m, an d le ttin g
Q i, . . Q m b e th e correspon ding le tte r m -tu ple, w e call Qx& ... & Q m
(Q i V . . .V Q ^ , §2 7 ) th e corresponding elem entary conjunction {disjunction).
E xample 1 (concluded). T h e corresponding ele m e n ta ry co n ju n ctio n
(disjunction) is & C ( - i c 3 V S V - i C).
T heorem 11°. A proposition letter form ula E in the distinct propo

sition letters P x, . . . , P m is equivalent to a form ula F (called a prin cipal
disjunctive norm al form of E ) having one of two form s , as follows. I f E takes
the value t for some m -tuples of Vs and f\s as values of P j, . . ., P w, then F
is the disjunction (in some order) of the corresponding elem entary con
junctions. I f E is identically false, then F is P x & (D ually, E has a
prin cipal conjunctive norm al form G , in which the roles of V and & and of
t and f are interchanged) h E ~ G).
E xample 2 (con tin ued)0, —i [<£? V S D (B & C) V - t <3] has as a p .d .n .f.

(<3 & S & - i C ) V (<£? & - i 2 & C) V (<3 & - i S & - i C ) , an d as a p .c.n .f.
c^V-tSVC& e3VSV-.C& d V S V C
P roof. B y T h eo rem 10 C o ro llary 1 , since (as is e a s ily seen from the

specified form of a p .d .n .f., an d th e v a lu a tio n ta b les for - i , & , V) E
an d F are id e n tic a lly equal.
T h e p rop osition al calcu lu s, besides b ein g d e d u c tiv e ly c o m p le te in tw o

senses (Theorem 10 an d C o ro lla ry 2 ), is also n o ta tio n a lly co m p le te in
th e sense th a t each of th e 2%m possible tr u th -v a lu e fu n ctio n s of m v a ria b le s
P j, . . . , P m ca n b e represen ted b y a le tte r form u la in th ose letters. F o r,
g iv e n th e ta b le for th e fu n c tio n , w e ca n b u ild a p .d .n .f. (u n iq u e ly d e
term in ed to w ith in th e order of its d isju n c tiv e m em bers) to represent it.
T h ese three co m p leten ess results ap p eared first in P o st 1 9 2 1 ; th e
present m eth o d o f proof of T h eo rem 10 is due to K a lm a r 19 34 - 5 . (Th e
w riter also used it, as an ap p lica tio n of V -elim in atio n in a n o th er c o n te x t,
in a first d ra ft o f 19 3 4 .)
H ilb e rt an d A c k e rm a n n 1928 g a v e a ve rsio n o f P o s t's proof, w h ich
consists in estab lish in g th e norm al form th eorem first, b y re d u ctio n
tech n iq u es g o in g b a c k in p a rt to th e n in ete en th c e n tu r y w orkers on
sy m b o lic logic. B rie fly , a le tte r fo rm u la E ca n b e redu ced to a p .d .n .f.
(or p .c.n .f.), u sin g th e ch ain m e th o d § 26, as fo llo w s. F ir st, *5 8 or *5 9 is
used to rem o ve th e occurrences of D . Second , th e occurrences of -1 are
tran sferred to th e interior b y rep e ate d a p p lica tio n s o f T h eo rem 8 . T h ird ,
th e resu ltin g expression is “ m u ltip lie d o u t" u sin g th e d is trib u tiv e la w
*3 5 (w ith *33) as in o rd in ary algeb ra w ith V, & in th e role of + , * (or
v ic e versa, u sin g *3 6 w ith *3 4 , for p .c.n .f.). F o u r th , sim p lificatio n s are
perform ed, b ased (besides on *3 1 — *34) on th e id e m p o te n tla w s * 3 7 , *38 a n d
on * 5 2 — *5 5 , so th a t th e re su ltin g d isju n c tiv e m em bers (for p .c.n .f.,
co n ju n c tiv e m em bers) or “ te rm s" c o n ta in each a t m o st one occurrence
of each le tter, e x c e p t if th e p .d .n .f. P x (the p .c.n .f. P x V - 1 P x) is
reached. F ifth , in th e n o n -e x c e p tio n a l case, m issin g letters are in tro
d u ced in to th e term s u sin g * 5 2 an d *3 5 (*53 and *36). S ix th , u sin g
*3 1 — *34 , *3 7 , *38, th e letters a n d n e g a te d le tters are b ro u g h t to th e norm al
order w ith in term s, so t h a t th e term s b eco m e ele m e n ta ry co n ju n ctio n s
(elem en tary d isju n ction s), an d d u p licatio n s of term s are suppressed.
E xample 2 (con clu ded)0. A lth o u g h our p roof of T h eo rem 11 es

tab lish es th a t V S D ( S & C ) V ic 3 ] is e q u iv a le n t to
{a & b & -»c) v (a &-is &c) v (<sr&-i» &-iC),

th e reader m a y fin d it o f in terest to reduce th e one to th e o th er b y th e
proced ure ju st described. T h e fo u rth op eratio n lead s to
(©3f & - i B ) V ( < £ 7 & -iC ) V ( B & - 1 C & «3). — F ro m th is b y *40 w e g e t as
a *'d is ju n c tiv e norm al fo rm ” (not " p r in c ip a l” ) (<£7 & - i S ) V (<C3 & n C ) , an d
th e n ce u sin g *3 5 th e still shorter eq u iv a le n t (not a d.n.f.) <£?& ( - i S V - i C ) .
§ 30. Decision procedure, interpretation. We know e x a m p le s

in m a th e m a tic s of gen eral question s, such th a t a n y p a rticu la r in stan ce
o f th e q u estio n ca n b e answ ered b y a preassigned uniform m eth o d .
M ore p recisely, in such an ex am p le , there is an in fin ite class of p a r
ticu la r question s, an d a procedure in relation to t h a t class, b o th b ein g
d escribed in a d v a n ce , su ch th a t if w e th ereafter select a n y p a rticu la r
q u estio n of th e class, th e procedure w ill su rely a p p ly an d lead us to a
d efin ite answ er, eith er " y e s ” or " n o ” , to th e p a rticu la r q u estio n selected .
E xample 1. Let f(x) an d g(x) b e p o ly n o m ials w ith g iv e n in te g ra l

coefficien ts. Is f(x) a fa c to r of g(x)? W e ca n d iv id e g(x) b y f(x). T h is d i
v isio n is perform able step b y step b y a preassigned m eth o d . I t w ill
te rm in a te in a fin ite n u m b er of step s (how m a n y w ill d epen d on h o w w e
co u n t th e steps, b u t th is co u ld b e m ad e precise so t h a t th e n u m ber w ill
depend on the degrees of the polynomials and the size of the coefficients).
We shall then have the remainder before us. This remainder we can rec
ognize to b e eith er 0 or d ifferen t from 0 . I f it is 0, th e answ er to th e
q u estio n is " y e s ” . I f it is d ifferen t from 0, th e an sw er is " n o ” .
E xample 2. D o es th e eq u a tio n ax + by — c , w here a y b, c are

g iv e n integers, h a v e a solu tio n in integers for x an d y ? T h ere is a w e ll-
k n o w n m e th o d for an sw ering th e question , usin g E u c lid 's algorith m .
A m eth o d of th is sort, w h ic h suffices to answer, either b y " y e s ” or b y

"n o” , any p a rticu la r in stan ce o f a gen eral question , w e ca ll a decision
procedure
; or decision method or algorithm for th e question. T h e problem of
decision problem for th e question .
fin d in g such a m e th o d w e sh all c a ll th e
T h e problem appears in m odern logic w ith Schroder 18 9 5 , L o w e n h e im 1 9 1 5
an d H ilb e rt l g i 8 . T h e present a cco u n t is o n ly in tro d u c to ry , an d w e sh a ll
a tte m p t a m ore precise d efin itio n o f w h a t co n stitu te s a decision m eth o d
la te r (§§ 60, 61). F o r th e present, it w ill b e enough th a t w e sh ould b e
able to recognize p articu lar exam p les of decision procedures.
§30 DECISION PROCEDURE, INTERPRETATION 137
S im ila rly , w e m a y h a v e a calculation procedure or algorithm (and hence
a calculation problem) in relatio n to a gen eral q u estio n w h ich requires for
an answ er, n o t " y e s ” or " n o ” , b u t th e e x h ib itin g of som e o b ject.
N o w , in co n n ectio n w ith a g iv e n form al sy ste m , such as th e one w e
h a v e been s tu d y in g , th ere are som e gen eral question s, such as "Is a g iv e n
fo rm al expression a fo rm u la ? ' an d 'Is a g iv e n fin ite sequence of form al
expressions a p ro o f? ', for w h ich a decision m e th o d is p ro v id e d d ir e c tly
b y th e defin ition s estab lish in g th e syste m . In fa c t, th is m u st b e th e case if
th e fo rm alizatio n b y m eans of th e sy ste m is to acco m plish w h a t w as
in ten d ed . O f a differen t n a tu re is th e q u estio n 'Is a g iv e n form u la p ro v
a b le ? ' T o see th is difference, let us com pare th e d efin ition s of th e th ree
n otion s: 'fo rm u la ', 'proof', 'p ro v a b le fo rm u la'. F o r each o f th e three
definitions, in a p p ly in g it to a p a rticu la r g iv e n o b je c t, w e h a v e to rec
ognize th a t th e g iv e n o b je c t belo ngs to th e class d efin ed (if it does
belong) th ro u gh th e con sid eration o f a sequence o f o b je cts, n a m e ly re
s p e c tiv e ly : th e form ulas o b ta in e d on th e w a y in th e co n stru ctio n of th e
g iv e n form ula, th e segm en ts o f th e g iv e n proof, th e form ulas in a proof
o f th e g iv e n p ro v a b le form ula. In th e cases o f form u la an d proof, th is
seq u en ce o f o b je c ts is co n ta in ed w ith in th e g iv e n o b je c t, from w h ich it
can b e regain ed for our con sideration . B u t in th e case o f p ro v a b le form ula,
th is sequence of o b je c ts is n o t co n ta in ed w ith in th e g iv e n o b je c t. H en ce,
for th e la st question , if a decision m e th o d exists, it m u st con sist in som e
th in g else th a n a direct or n e a rly direct a p p lic a tio n of th e defin ition , an d
th e decision p roblem for th is q u estio n is n o t tr iv ia l. I t is o fte n ca lled
the decision problem for th e form al sy ste m . T h is problem is so lve d for th e
pure p ro p o sitio n al ca lcu lu s b y T h eo rem s 9 an d 10 (§§ 28, 29):
T heorem 12°. A decision procedure (or algorithm) for determ ining

whether or not a proposition letter form ula E is provable in the propositional
calculus is afforded by the process of calculating the table for the function of
t and f represented by E . A ccording as only t's occur in the value column or
not , E is provable or not.
F u rth erm o re, th e decision procedure ex te n d s to th e o th er senses o f
form ula, as w e m a y first pass to a correspon ding le tte r form u la b y si
m u lta n e o u sly rep lacin g th e d istin ct prim e co m p o n en ts b y re sp e ctiv e
d istin c t proposition letters (Theorem s 3 an d 4 § 25). A decision p roced u re
for eq u iv a le n c e is in clu d ed in th a t for p r o v a b ility , b y th e d efin itio n of
e q u ivalen ce (§26 ), or ca n b e b ased sim ilarly on th e Corollaries 1. T h e
decision p roblem for d e d u c ib ility is reduced to th a t for p r o v a b ility b y
n o tin g, from th e d eriv ed rules for D an d & (Theorem 2 § 23) t h a t , in
th e p ro p o sitio n al calcu lu s, D x, . . D t 1- E if an d o n ly if
b D j & . . . & D* 3 E . (A second set of decision procedures is afford ed
b y th e process o f re d u ctio n to p .d .n .f., end § 29.)
I n t e r p r e t a t io n . O u r m e ta m a th e m a tic a l result t h a t th e p ro p o sitio n al

ca lcu lu s ad m its an a rith m e tic a l in terpretatio n , w ith tw o o b je c ts t, f in th e
arith m e tic , illu m in a tes th e u su al lo gical in terp retatio n (cf. end of § 28).
W e see th a t our pro p o sitio n al ca lcu lu s is a su itab le lo gical in stru m en t ( 1 )
w h en th e p a rticu la r propositions are such th a t each is d e fin ite ly eith er
tru e or false, or (2 ) w h en w e w ish to m ake it an assu m p tio n of a th e o r y
w h ic h w e are d e v e lo p in g th a t each is either tru e or false. F o r an in -
tu itio n ist, S itu a tio n ( 1 ) is represen ted in th e m a th e m a tics of a fin ite
d o m a in of o b jects, an d also in reasoning w ith propositions a b o u t th e
o b je c ts of an in fin ite d o m ain w hen th ose propositions are of a ty p e for
w h ic h there are decision procedures (cf. R e m a rk 1 § 29). T h e use of th e
p ro p o sitio n a l ca lcu lu s in classical m a th e m a tics ca n be ta k e n as an
e x a m p le of S itu a tio n (2).
T h e tr u th ta b le s n o w m a k e it d efin ite e x a c tly h o w th e operators
o f th e ca lcu lu s are to b e in terp reted as tr u th -v a lu e fu n ctio n s of p ro p o
sitions. For exam p le , we see t h a t V is th e inclusive ‘o r*: A V B is
tru e, if either A is true or B is tru e or b o th . (Th e exclusive ‘or* is exp re s
sible th u s : ( A V B ) & n ( A & B).)
T h e im p lica tio n A 3 B m eans th e sam e as i A V B (*59 § 27), an d is
ca lle d m aterial im plication. T h e h o ld in g of A 3 B does n o t require a
n ecessa ry co n n ectio n of id eas b e tw e e n A an d B . F o r exam p le , th e m oon
is m ad e of green cheese m a te ria lly im plies 2 + 2 = 5 (because th e p rem
ise is false). F e r m a t’s “ la st th eo rem ” m a te ria lly im plies 2 + 2 = 4
(because th e con clusion is true). T h is is considered p a ra d o x ic a l b y som e
w riters (Lew is 1 9 1 2 , 1 9 1 7 ). W ith o u t a tte m p tin g to en ter fu lly in to a
co n tro versial question , w e offer th e fo llo w in g brief rem arks. T h e role of
m a te ria l im p lica tio n is b e st und erstoo d, w hen it is considered in a w id er
c o n te x t, su ch as t h a t p ro v id e d b y th e fu ll n u m b er-th eo retic sy ste m .
V x (A (x ) 3 B (x)) expresses a relatio n sh ip b e tw e e n A (x ) an d B (x ) as
v a ria b le propositions (or 'p ro p o sitio n al fu n ctio n s of x ’), ca lled form al
im plication. T h e p ro p e r ty of m ate rial im p lica tio n of h o ld in g w h e n eve r th e
first m em ber is false, w h en th e m ate rial im p lica tio n is used in co m b in a tio n
w ith g e n e ra lity to b u ild a form al im p lica tio n , allow s a th eorem
V x (A (x ) 3 B (x)) to h old vacuously for certain va lu e s of x. T h is is a d evice,
o f a piece w ith th e ad m ission of 0 in to th e n u m ber sy ste m an d of th e
v a c u o u s set in to th e th e o r y of sets, w h ich con duces to sim pler an d m ore
co m p reh en sive form ulatio n s of theorem s. T h e sam e is true of th e p ro p e rty
o f m a te ria l im p lica tio n of h o ld in g w h en ever th e second m em ber is true.
F ro m th e sta n d p o in t of la y E n g lish , 3 is p ro b a b ly b e tte r rendered as
“ if . . . , th en . . . ” or “ o n ly i f ” . N o tw ith sta n d in g , “ im p lic a tio n ” is a h a n d y
n am e for 3. In usin g it, w e follow th e p ra c tic e com m on in m a th e m a tics
o f e m p lo y in g th e sam e d esign atio n for an alogous n otions arising in re la ted
tech n ica l theories. (A n ex a m p le is th e m a n y d ifferen t kind s of “ a d d itio n ”
an d “ m u ltip lic a tio n ” in m ath em atics.) T h e operator 3 does h a v e th e
ch aracter of an im p lica tio n in our form al sy ste m , in con sequence of th e
tw o properties of it expressed b y th e d ed u ctio n theorem an d R u le 2
(i.e. th e tw o 3 -rules of T h eorem 2 ). So it does represent lo gical conse
quen ce, n o t in som e a priori sense, b u t in th e sense d efin ed for th e form al
sy ste m b y th e d e d u c tiv e p o stu la te s of th e system .
O t h e r fo r m s o f t h e c a l c u l u s . In v ie w of *5 6 — * 6 1, th e n o ta tio n a l
com pleten ess of th e p rop osition al calcu lu s (end of § 29) co u ld h a v e been
o b ta in e d b y ta k in g -i an d o n ly one of th e three oth er operators 3, &, V
as p rim itive operators for th e calcu lu s (form al sym bols), an d d efin in g th e
o th er tw o of 3, & , V from those as sym b o ls of a b b re v ia tio n (as ~ w as
defin ed in § 26). T h e still fu rth er red u ctio n ca n b e m ad e to a single
p rim itiv e operator | (called 'a lte r n a tiv e d en ial', or th e 'S h effe r stro k e',
1 9 1 3 *), w ith th e t a b l e :
<3 12
s t f
a t f t
f t t
T h e n c e one defines -1 as | <3f, an d < 3 V B as (“ i<3?) I (~1®)*

T h e p rop osition al ca lcu lu s m a y b e set u p ta k in g th e su b stitu tio n rule,
w h ich w e d erived as a su b sid ia ry d ed u ctio n rule in T h eo rem 3 § 25,
in stea d as a d irect p o s tu la te d rule:
E *.
I n th is case, th e proposition letters are called proposition variables , an d

p a rticu la r axio m s u sin g th e form al v a ria b les J?, C ca n b e used in p la ce
of th e ax io m sch em a ta u sin g th e m e ta m a th e m a tic a l va ria b le s “A ”,
“ B ” , “ C ” . T h e direct su b stitu tio n rule is u s u a lly co n stru ed to a p p ly to
a sin gle v a ria b le a t a tim e (cf. R e m a rk 1 § 25).
E xam ple 3. W e th en h a v e D (B D Z%) as A x io m la . T o c o n stru ct

th e n ce a proof of A D (B D A ), for a n y g iv e n form ulas A an d B , w e
p roceed thus. I f A does n o t co n ta in B, first su b s titu te A for <3, th en B
fo r B. I f A co n ta in s B, let P be a proposition le tte r d istin ct from and
n o t occu rrin g in A , an d s u b stitu te su c cessiv ely P for B, A for <3, B for P .
T h is is th e m ore usu al m eth o d of s e ttin g u p th e p rop osition al c a lc u lu s ;

th e m e th o d w e h a v e chosen u sin g ax io m sch em ata is due to v o n N e u m a n n
1 9 2 7 . In eith er case, th e rules of inference m u st h a v e th e ch aracter o f
sch em ata, i.e. th e y m u st e m p lo y m e ta m a th e m a tic a l va riab les, sin ce
in fin ite ly m a n y a p p licatio n s h a v e to be p ro v id e d for. F o r th e ca lcu lu s
w it h th e d irect su b stitu tio n rule, w h ile th e p r o v a b ility n otion is th e sam e
(as th e reader m a y dem on strate), th e d e d u c ib ility relation is m ore e x
ten siv e. T h e d ed u ctio n theorem , an d other su b sid iary d ed u ctio n rules
d ep en d in g on it, m u st th e n ca rry a restriction on th e use of proposition
v a ria b le s in th e su b sid iary d eduction s, s ta te d w ith reference to ap
p licatio n s o f th e new su b stitu tio n rule e x a c tly like th e one w e h a v e s ta te d
w ith reference to R u les 9 an d 12 for in d iv id u a l v a riab les in th e p re d ic a te
calculus.
W ith either of these v a ria tio n s in th e w a y of se ttin g u p th e prop o
sitio n a l ca lcu lu s as a form al sy ste m (and there e x ist others), w e still h a v e
e ss e n tia lly th e sam e calcu lu s. A llo w in g for differences in th e selection o f
th e operators, th e class of p ro v a b le form ulas an d th e in terp retatio n as
an a rith m e tic o f tw o o b je c ts rem ain th e sam e. A ll th ese syste m s, an d th e
co m m o n th e o r y if w e a b stra c t from th e p a rticu la r form u latio n , m a y b e
ca lled classical or two-valued propositional calculus.
O t h e r p r o p o s it io n a l c a l c u l i . T h ere are oth er sy ste m s of proposi
tio n a l calcu lu s, w h ich are p rop osition al ca lcu li in th a t t h e y a n a ly z e
proposition s o n ly w ith resp ect to h o w th e y are form ed from oth er p rop o
sitions ta k e n as w holes, an d w h ich h a v e therefore a d efin ition of fo rm u la
o f th e sam e ch aracter (to w ith in th e p articu lar ch oice of operators), b u t
w h ich differ e ss e n tia lly from th e sy ste m stu d ied here.
O n e class of exam p le s is o b ta in ed b y gen eralizin g from th e tw o -v a lu e d
pro p o sitio n al ca lcu lu s to th e n-valued propositional calculi for a n y
p o sitiv e in teger n > 2 . T h e se for n > 2 were discussed b y L u k a sie w ic z
1920 (n = 3) an d P o st 19 2 1 (an y n). T h e y are ca lcu li w h ich can b e tre a te d
on th e b asis o f tru th ta b le s in an arith m etic of n o b jects, as th e
classical sy ste m is tre a te d on th e basis of tw o. (Cf. e.g. L u k a sie w icz an d
T a rsk i 1930 , R osser an d T u r q u e tte 19 4 5 , 194 9 , 19 5 2 .)
A n o th e r e x a m p le is th e intuition istic propositional calculus (H e y tin g
1 9 3 0 ), which is intended as a formalization of intuitionistic mathematical
reasoning with propositions. As noted in § 23, we obtain a postulate list
for it (for the intuitionistic predicate calculus) from our postulate group
A1 (group A) simply by replacing Axiom Schema 8 by Axiom Schema 8 r.
This is not Heyting's original postulate list, but one suggested by Gentzen's
1 9 3 4 -5 . (Heyting's postulates for the predicate calculus are in 1 9 3 0 a.)
Whether our numbered results and italicized theorems marked with the
symbol ''OMas not being established intuitionistically, so far as our dis
cussion has gone, actually do not hold for the intuitionistic system is a
question which in each case requires further consideration. We shall return
to this later in the book (§§ 80, 82). The result of this section, that a
decision procedure exists for the propositional calculus, is one such result
which does hold for the intuitionistic system (Theorem 56 (d) § 80).
It is known that in the intuitionistic propositional calculus none of the
four operators can be expressed in terms of the remaining ones (Wajsberg
1 9 3 8 , McKinsey 1 9 3 9 ); and that the calculus cannot be treated on the
basis of truth tables for any finite (Godel 1 9 3 2 ), but can be for n
n = N0 (Jaskowski 1 9 3 6 *).
Still further apart from the standpoint of the classical calculus studied
here are thfe p r o p o s itio n a l c a lc u li (Lewis 1 9 1 2 ), and
o f s tr ic t im p lic a tio n
the m o d a l which deal with 'possibility', 'necessity',

p r o p o s itio n a l c a lc u li
etc. (Cf. Lewis and Langford 1 9 3 2 , Feys 1 9 3 7 -8 , McKinsey and Tarski

1 9 4 8 , Feys 1 9 6 5 *.)
Chapter VII
TH E P R E D IC A T E CALCULUS
§31. Predicate letter formulas. In th is ch ap ter, w e s tu d y th e

p a rt o f th e form al s y ste m o b ta in e d b y u sin g e x a c t ly th e p o stu la te s o f
G ro u p A .
The p rop osition al calcu lu s, stu d ied in th e p reced in g ch ap ter, is a
fo rm alizatio n of lo gical relation sh ips w h ich d ep en d o n ly on th e a n a ly sis
o f th e w a y ce rta in propositions are com posed o u t o f sim pler propositions,
u sin g o peratio n s o f co m p o sitio n in w h ich th e sim pler propositions en ter as
u n a n a ly z e d w holes.
In th e p red ic a te calcu lu s, th e an alysis goes a ste p further, an d w e are
allo w ed to consider also w h a t m a y b e ca lled th e 'su b je c t-p re d ic a te '
stru ctu re o f th e sim pler propositions, an d to use operatio ns of co m
p osition w h ic h d ep en d on th a t structure.
T h is an alysis still does n o t ta k e in to a cco u n t all features of th e stru ctu re
of n u m b er-th eo retic propositions. T h is w e can em phasize, ju s t as in th e
p reced in g ch a p ter, b y in tro d u c in g an a lte rn a tiv e n o tio n of form u la,
w h ich elim in ates th e irreleva n t d eta ils o f th e n u m b er-th eo retic d e f
in itio n o f form ula, an d le a v e s th e w a y open to oth er a p p licatio n s as w ell.
W e s ta r t b y in tro d u c in g form al expressions of a new species, co n
s titu tin g a gen eraliza tio n of th e proposition le tters in tro d u ced in § 25,
as follow s,
c2, £t(a, b)f ..., 2, B(a), B(a,b), ..., C,C(a), C(a,b),----

T h ese expressions w e c a ll predicate letters (with attached or nam e form
variables ). E a c h of th e sy m b o ls fo rm erly used as a proposition le tte r form s
a differen t p red icate le tte r w ith each d ifferen t num ber n ^ 0 o f a tta c h e d
va riab les, an d for n= 0 a p red icate le tte r is a proposition letter. T h e n
a tta c h e d va ria b le s m a y b e a n y n d istin ct variab les. D ifferen t choices of
th e n a tta c h e d v a ria b le s w ith a g iv e n p red ic a te le tte r are said to g iv e
differen t nam e form s of th e sam e p red icate letter, e.g. <ZZ(a, b ), e1 (b , a)
an d (c, d ) are three n am e form s of th e p red icate le tte r form ed b y u sin g
w ith tw o a tta c h e d va riab les, b u t £l(a) an d £2(a, b,c) are oth er p red -
142
§31 PREDICATE LETTER FORMULAS 143
ica te letters, an d c% ( a t a ) is not a p red icate letter. I t o rd in arily su ffices
th ro u g h o u t a discussion to consider each of th e d istin ct p red icate le tters
em p lo y e d in th e discussion as represented b y ju st one nam e form , i.e.
as ta k e n w ith a fix e d sequence of n a tta c h e d va ria b le s th ro u g h o u t th e
d iscu s sio n ; an d u su a lly w e ta k e these n a tta c h e d v a ria b les to b e th e first
n v a riab les in order from a g iv e n in fin ite list a x, a 2, a 3, . . . of v a ria b les
(u su ally here th e list a, b , c, . . . ) .
F o r th e d efin itio n of predicate letter form ula , g iv e n in d u c tiv e ly as fol
low s, th e o n ly term s sh all b e th e variab les. H o w e v e r w e elect to s a y
“ te rm ” a t som e places an d “ v a r ia b le ” a t others, so th a t it w ill b e clear
h ow th e discussion generalizes, if w e la te r w ish to allow a w id er class of
term s th a n sim p ly th e variab les.
n) is a p red icate le tte r w ith a tta c h e d va riab les, a n d
1. I f P (a x, . . . , a
t n are term s, th en P ( t x, . . . , t n) is a form ula. 2 — 5. I f A an d B
t x, . . . ,
are form ulas, th e n (A) D (B), (A) & (B), (A) V (B) a n d - i (A) are form ulas.
6 — 7. I f x is a v a riab le, an d A (x ) is a form ula , th en V x (A (x )) an d 3 x (A (x ))
are form ulas. 8 . T h e o n ly form ulas are those g iv e n b y 1 — 7.
E xample 1. By 1, i% ( b , a )> B , <%(a , b ) an d 3

< (a , a ) are p red icate
le tte r form ulas. H ere it suffices to sta rt from th e tw o nam e form s <3f(a , b )
an d B . B y su ccessive a p p licatio n s of 3 an d 2 (o m ittin g parentheses under
th e usu al co n ven tio n s § 17), c% ( b , a ) & B an d ^ ( b t a ) & B O ^ { a , b ) are
p red icate le tte r form ulas. F in a lly b y 6 , V b (<£l (b , a ) & B D <C?(<2, b )) is a
p red icate le tte r form ula.
F o r th is ch ap ter, w h en w e s a y “ te rm ” an d “ fo rm u la” w ith o u t sp e cify in g

th e senses, th e w ords m a y b e u n derstoo d eith er in th e re sp e ctive senses
of free v a ria b le an d p red icate le tte r form ula, or else in th eir re sp e ctive
n u m b er-th eo retic senses (§ 17). T h e tw o form al syste m s, h a v in g in co m
m on G ro u p A (§ 19) as p o stu la te list, b u t d iffe rin g th u s in th eir fo rm atio n
rules, w e distin gu ish as th e pure predicate calculus an d th e num ber-
theoretic predicate calculus.
B efo re g o in g ah ead w ith th e m eta m a th e m a tics, let us see h o w th e
form alism is in terp reted as a calcu lu s of pred icates.
In w ord lan gu ages, a proposition is expressed b y a sentence. T h e n
a ‘p red icate ' is expressed b y an in co m p lete sen ten ce or sen ten ce sk eleto n
co n ta in in g an open place. F o r ex am p le , “ __ is a m an” expresses a
p red icate. W h e n w e fill th e open p lace w ith th e nam e of a su b je ct such as
“ S o c ra te s” , a sen ten ce such as “ So crates is a m a n ” is o b tain ed . T h e
situ a tio n is co n v e n ie n tly d escribed b y u sin g th e m odern m a th e m a tic a l
n o tio n of a ‘fu n ctio n ' (§ 10 ). T h e p red icate is th e n a fu n ctio n of one v a r i
144 THE PREDICATE CALCULUS CH, VII
able. T h is va ria b le ranges o ve r som e d om ain , in clu d in g as m em bers

Socrates, Chiron, etc. T o each m em ber of th is dom ain, th e fu n ctio n co r
relates a p r o p o sitio n ; i.e. w h en th e in d ep en den t v a ria b le ta k e s a m em ber of
th e d om ain as va lu e, th e p red icate ta k e s a proposition as correspon ding
v a lu e . T h u s th e p red ic a te is a propositional function of one variable .
P re d ica te s are o ften ca lled 'properties', e.g. in C h ap ters I I , I I I . In th is
te rm in o lo g y " __ is a m a n " expresses th e p ro p e rty P of b ein g a m an ,
an d " S o c r a te s is a m a n " expresses th e proposition t h a t S o crates has th e
p ro p e r ty P. A n o th e r term used in th is co n n ection is 'c la ss'; " S o c r a te s
is a m a n " expresses t h a t So crates belo n gs to th e class C of m en. ('P re d
ic a te ' in its stric t g ra m m a tic a l m ean in g of t h a t w h ich a sen ten ce sa y s
a b o u t its su b je c t is narrow er th a n 'p ro p o sitio n al fu n ctio n of one v a r i
a b le ' or 'p r o p e r ty ', since for a p red icate th e o m itte d noun in th e sen ten ce
sk eleto n m u st b e th e s u b je c t of th e sentence.)
A s a second illu stratio n , consider th e sen ten ce sk eleto n __l o v e s ___ ".
I n g r a m m a tic a l te rm in o lo gy, th is consists of a tra n sitiv e v e rb an d tw o
open places, one to b e filled b y th e nam e of a su b je ct such as " J a n e " ,
an d th e o th er of an o b je c t such as " J o h n " . W h e th e r th e resu ltin g sen ten ce
expresses a tru e or a false proposition w e are n o t sa y in g . T h e sen ten ce
sk eleto n in th is illu stra tio n expresses a binary relation , i.e. a relation
a m o n g tw o m em bers, or in o th er w ords a propositional function of two
variables.
In o th er exam p le s, th ere m a y be several correlated open spaces to b e
filled w ith th e sam e nam e. F o r ex am p le, " __ x is th e fa th er o f ___2, or
__ i is th e m o th er o f __ 2" expresses a b in a r y relation, also expressed b y
" __ i is a p are n t o f ___2".
T h e reader m a y e a sily m a k e u p exam p les of sen ten ce sk eleton s co n
ta in in g a n y greater n u m ber n of open places or of sets of correlated open
places. S u ch a sen ten ce sk eleto n expresses an n-ary relation , or propo
sition al function of n variables .
F ro m th e sta n d p o in t of fu n ctio n s, th e d istin ctio n b etw ee n a 'p red ica te'
in th e tra d itio n a l sense of t h e first illu stra tio n an d a 're la tio n ' is of m inor
sig n ifica n c e ; an d likew ise th e d istin ctio n b etw ee n 's u b je c t' an d 'o b je c t'
in th e first tw o exam p les. I t w ill b e m ore co n ven ien t h en cefo rth to s a y
s im p ly “ p re d ic a te " an d " o b je c t " in all cases. B y a predicate (of n variables)
w e sh all a c co rd in g ly m ean a p rop osition al fu n ctio n of n va riab les, w here n
m a y b e 0 g iv in g a proposition, or 1 g iv in g a p red icate in th e tra d itio n a l
sense or a p ro p erty , or > 1 g iv in g an n -a ry relation. W e call th e va lu es
o f th e in d ep en den t v a ria b le s (when n > 0) objects , an d th e in d ep en d en t
v a riab les object variables.
§31 PREDICATE LETTER FORMULAS 145
T h e p red icate calculus w ill tre a t of th e lo gic of pred icates in th is gen eral
sense of 'p re d ic a te ’ , i.e. as p rop osition al fu n ctio n . Som e w riters hence s a y
" fu n c tio n a l c a lc u lu s” in stea d of "p r e d ic a te c a lcu lu s” .
W e shall consider here o n ly th e case of p red icate calcu lu s w h ich has
one d om ain of o b je cts for all its o b je c t va riab les, in w h ich case th e o b je cts
m ay be called also in d ivid u a ls , an d th e o b je c t v a riab les in dividu al
variables. T h is case suffices for th e in ten d ed a p p lica tio n to our n u m b er-
th eo retic sy ste m , for w h ich th e d o m ain is th e set of th e n a tu ra l num bers.
T h e trea tm en t of th e p red icate ca lcu lu s w ill n o t d epen d on a n y su p
position a b o u t th e o b je c t d om ain, e x c e p t th a t it b e n o n -e m p ty , i.e.
co n ta in a t least one elem ent. F o r th e pure form of th e calculus, no p ro
visio n is m ad e for referring to p a rticu la r o b je c ts of th e dom ain, i.e. there
are in d iv id u a l v a ria b le s b u t no in d iv id u a l co n stan ts.
N o w le t us see h o w w e com e to choose th e sym b o lism w h ich w e use
to represent predicates. T h e b la n k s e m p lo y e d a b o v e to sh ow th e open
places in a sen ten ce sk eleto n w e replace b y th e d evic e c u s to m a ry in
m a th e m a tic s of letters called " v a r ia b le s ” . T h u s in stea d of " __ •, is
fa th er of __ 2> or - __ 1 m o th er o f __ %\ w e m ore c o n v e n ie n tly w rite
( 1 ) " a is fa th er of b , or a is m oth er of V \ Som e m a th e m a tic a l exam p le s
are (2) “a is e v e n ” , (3) "a equals 6 ” , an d (4) " a is less th a n b” .
F u rth erm o re, since a p red icate is a k in d of a fu n ction , n a m e ly one
w hose v a lu e s are propositions, w e em p lo y fu n ctio n a l n o ta tio n (§ 10 ) in
n a m in g pred icates, e x c e p t in cases w hen som e oth er n o ta tio n is in c o m
m on usage. T h u s w e m ig h t d esign ate th e p re d ic a te of ( 1 ) as “P (a, &)”
(for "a is a p aren t of 6” ) an d of (2) as “E (a )” , u sin g th e fu n ctio n al n o ta tio n
w ith th e p red icate s y m b o l ( " P ” or " £ ” ) w ritte n ah ead of th e in d ep en den t
variab les. (W e h a v e a lre a d y done so in § 7 in u sin g " P ( n ) ” to express th a t
n h as th e p ro p e r ty P .) F o r (3) an d (4), w e use th e cu s to m a ry re la tio n a l
n o ta tio n s "a— b” an d " # < & ” , w ith th e p red icate sy m b o l (" = ” or
" < ” ) w ritte n b e tw e e n th e in d ep en den t variab les.
F o r th e pure p red icate calcu lu s, th e p red icate letters such as b),
etc. are to b e in terp reted as sta n d in g for u n sp ecified p red icates, i.e.
<^(at b) for a p red icate of tw o variab les, $ for a p red icate of zero va ria b le s
(i.e. a proposition), etc. T h e n a n y p red icate le tte r form u la can b e in te r
p reted as sta n d in g fd f a p red icate w h ich is d eterm in ed b y th e p red icates
represen ted b y th e d istin ct p red icate letters from w h ich it is co n stru cted ,
e.g. V/>(e2f(£, a) & B D b)) represents a p red icate of one v a ria b le
(corresponding to th e free a) d eterm in ed b y th e p red icate of tw o va ria b le s
represen ted b y c2f(<z, b) an d th e proposition represen ted b y #.
N o te t h a t w hen w e are usin g <^(a, b) as th e nam e form for th e p red ic a te
146 THE PREDICATE CALCULUS CH. VL
le tte r <3 w ith tw o a tta c h e d va riab les, th e n for th e in terp retatio n , a fte r
ch oosin g in d e p e n d e n tly w h a t p red icate <£t(a, b) sh all sta n d for, th e
m ean in gs of b), <Fl(a, a), «3f (b, a), etc. d epen d upon th a t, b y th e
sta n d a rd co n v e n tio n for fu n ctio n a l n o ta tio n (§ 10 ).
S im ilarly, a n y fo rm u la in th e n u m b er-th eo retic sy ste m ca n b e in ter
p re te d as exp ressin g a p red icate, under th e usual n u m b er-th eo retic
m ean in gs o f th e sym b o ls. For ex am p le , 3 c ( # = 0 " -c ) expresses E(a)
or a is even, a ~ b expresses a = 6, an d lc ( c '+ a = b ) (a b b rev ia ted a<b
in § 17) expresses a<b.
L e t x v . . . , x n b e d istin ct v a ria b les, an d A ( x x, . . . , x n) a form u la (under
eith er n o tio n of form ula). W h e n w e are in terp retin g A ( x x, . . . , x n) b y
a p red icate , or perform in g form al operatio ns w ith it w h ich are in keep in g
w ith an in terp reta tio n b y a p re d ic a te (even th o u g h th e in terp retatio n is
n o t in v o lv e d in th e form al operations), w e call A ( x lf . . . , x n) a nam e
form in x lt . , . , x n as th e nam e form variables, an d s a y t h a t x v . . , , x n
have th e nam e form interpretation or th e predicate interpretation . T h e
n a m e form A ( x 1# . . . , x n) is th e form u la of th e sy ste m ; “ A ( x x, . . . , x n)”
is our m e ta m a th e m a tic a l n am e for t h a t form ula (under our su b stitu tio n
n o ta tio n § 18); an d w e m a y on occasion in tro d u ce ttA (x 1, . . x n)” as a
n am e for th e p red icate A (x v ..., x n) w h ich th e form u la A ( x x, . . . , x n)
expresses u n der th e in terpretatio n .
I t is n a tu ra l to in terp ret a form u la w ith free v a ria b les b y a p red icate,
e'.g., w hen w e are con cerned w ith th e form ation rules o f th e sy ste m , a n d
th e fo rm u la in q u estio n is b ein g considered as a co n stitu e n t of oth er for
m ulas. A discussion of th e in terp retatio n s of a form u la b y a proposition
w ill b e g iv e n a t th e end of § 32.
§ 32. Derived rules, free variables. In u sin g th e d erived rules

of T h eo rem 2 (§ 23) for th e p red icate calculus, w e m u st o bserve c a re fu lly
th e restriction s on th e h a n d lin g of v a r ia b le s : ( 1 ) T h e t for an V -e lim in a tio n
or 3 -in tro d u ctio n m u st b e free for th e x in th e A (x) (cf. § 18). (2 ) T h e x for
an 3 -elim in a tio n m u st n o t o ccu r free in th e C. (3) In a su b sid iary d e
d u ctio n th e free v a ria b le s m u st b e held co n sta n t for th e assu m p tio n
fo rm u la to b e d isch arged (cf. § 22 ). A t first p a re n th etica l rem arks w ill
call a tte n tio n to th e p recau tio n s tak en . L a te r it w ill b e le ft in crea sin gly
to th e reader to ob serve them .
T heorem F or *6 4 — *68, let x v . . . , x n be distinct variables,

13.
be a form ula, and t x, . . . , t n be term s (not necessarily distinct)
A( x j , . . . , x n)
which are free for x v . . . , x n, respectively, in A (x lf . . . , x w). F or *6 7 ,
§32 DERIVED RULES, FREE VARIABLES 147
also let C b e a form ula not containing any of x v . . . , x n free, and r ( x 1 , . . . , x n)
be a fin ite sequence of zero or more form ulas, and suppose that the free
variables are held constant for the last assum ption form ula A ( x x, . . . , x n)
in the subsidiary deduction . T h en :
*64. A(xif . . x w) bXl"*Xw V x x . . . V x wA ( x 1, . . . , x n).
Vxt .... VxnA(x1, . . . , xn) b

*6 5 . A ( tji . . . , t n).
* 66 . A(xlf . . . , xtt) bXl**‘Xw A ( tj, . . . , t n).
*68. A{tv . . . , t n) b 3x! ... 3xnA(x1, ..., xw).

*6 7. If T ( x 1, . . . , x w), A ( x p . . . , xn) b C, then
T(xv . . . , x n), 3 x x . . . 3xnA(xj_, . . . , x„) bXl" ,x* C.
(w-fold V -in tro d u c tio n , V -eh m in atio n , su b stitu tio n , 1 -in tro d u ctio n
an d 3-elim ination.)
I f x is a variable, and A (x ) and B (x) are form u las :

*69. A(x) 3 B (x) bx V x A (x ) D V x B (x ).
*70 . A (x ) D B (x) bx 3 x A (x ) D Ix B (x ).
P roofs . *64. By n successive ap p licatio n s of th e sim p le V -in

tro d u ctio n rule, (§ 23).
F o r *6 5 an d *66 (and la te r * 68 ), w e m ak e tw o cases. Case 1: tv ..., tw

do n o t co n ta in x x, . . . , x n. Case 2 : otherw ise.
*65 Case 1. By n-successive sim ple V -elim in atio n s,
V x x . . . V x nA ( x 1, . . . , x n) b V x 2 . . . V x nA ( t!, x 2, . . . , x n) b
V x 3 . . . V x wA ( t lf t 2, x 3, . . . , x n) b .. - b V x nA ( tj, . . . , t n_v x n) b
A ( t x, . . . , t n).
T h e case h yp o th esis an d th e h yp o th esis of th e theorem th a t tv ..., tn

are free for xv . . . , x w, re sp e c tiv e ly , in A ( x x, . . . , x w), to g e th e r insure
th a t tx is free for xx in V x 2 . . . V x nA ( x 1, x 2, . . . , x w), t 2 for x 2 in
V x 3 . . . V x nA ( tj, x 2, x 3, . . . , x n), etc. (Case 2 w ill follow *66 Case 1 .)
*66 Case 1. B y co m b in in g *64 an d *6 5 Case 1 (or b y n successive
ap p licatio n s of th e sim ple su b stitu tio n rule § 23).
*6 5 Case 2. Let wv . . . , w n b e va riab les d istin ct from each oth er

an d from xv ..., x n, an d n o t occurring in A ( x x, . . . , x n) or tv . . . , t n.
T h en , b y Cases 1 of *65 an d * 66 ,
V x x . . . V x „ A ( x 1, . . x„) 1- A ( w 1, . . w„) A ( t lt . . t„).
*68 Case 2. A (t„ . . t„) \- 3 w x . . . 3 w „ A (w 1, . . w„) . A lso

A ( w 1( . . w„) |- 3 x j . . . 3 x nA ( x 1( . . x„), w h en ce b y *6 7 ,
3 w x . . . 3 w „ A (w 1, . . w„) I- ... 3 x nA ( x 1, . . x„).

148 THE PREDICATE CALCULUS CH. VII
*6 9 . 1. A (x ) D B (x ), A (x ) h B (x ) bx V x B (x ) — D -e lim ., V -in tro d .

2. A (x ) D B (x ), V x A ( x ) hx V x B (x ) — V -elim ., 1 [the term x is
free for x in A (x ) b y use of th e defin ition § 18].
3. A (x ) D B (x) bx V x A ( x ) D V x B ( x ) — D-introd., 2 [the variable
x is held constant in the subsidiary deduction 2 for the as
sumption formula V x A (x ) which is being discharged, since x
does not occur free in V x A ( x ) ] .
*70. 1. A (x ) D B (x ), A (x ) b B (x ) h 3 x B (x ) — D-elim., 3-in trod . [cf.
*69 S te p 2].
2. A (x ) D B (x ), 3 x A (x ) bx 3 x B (x ) — 3-elim ., 1 [3 x B (x ) does n o t
c o n ta in x free, an d x is h eld c o n sta n t in th e su b sid iary d ed u ctio n 1 ].
D B (x) bx 3 x A (x ) D 3 x B (x )
3. A (x ) — D-introd., 2 [x does not
occur free in 3 x A ( x ) ] .
T h e 3 -elim in a tio n rule required a t S te p 2 for *7 0 is a su b sid iary d ed u ctio n
rule, w h ile th e V -elim in a tio n rule used a t S te p 2 for *6 9 w as estab lish ed in
th e stronger direct form (alth ou gh under an a b b re v ia tio n in tro d u ced in
§ 23 E x a m p le 1 th e step is presented in th e sam e form at). H en ce th e
g re ate r care required in ju s tify in g S te p 2 for *70.
E xample 1. V -in tro d u c tio n can b e ap p lied to th e form u la <31(a, a)

in o n ly one w a y , w hile 3 -in tro d u ctio n (sim ple or 2 -fold) ca n b e ap p lied in
severa l w a y s, as follows.
<3(a,a) K V a£l(a,a) — V -in tro d ., le ttin g x b e a ; A (x ) b e <^(a,a).

C l(afa) b 3 a 3 l{a ,a ) — 3 -in tro d ., le ttin g x b e j ; A ( x ) b e <3l(a,a);t be a.
<3l(a,a) b 3 b 3 l{ a ,b ) — 3 -in tro d ., le ttin g x b e b \ A (x ) b e 3 t(a }b ) ; t be a.
<3l(a,a) b 3b<3l(b,a) — 3 -in tro d ., le ttin g x b e b; A (x ) b e <3I(b,a); t be a.
3L{ata) b 3a3b<31(a,b) — 2 -fo ld 3-in trod ., le ttin g x v x 2 b e a, b; A (x lf x 2)
b e <3l(atb ) ; t v t 2 b e a , a .
cH(a,a) b 3alb<3(b,a) — 2 -fo ld 3 -in tro d ., le ttin g x 1# x 2 b e a, b \ A ( x 1# x 2)
b e <3l(b,a ) ; t 1# t 2 b e a, a.
I n t e r p r e t a t io n o f f o r m u l a s w it h f r e e v a r ia b l e s . T h e restriction
for our su b sid ia r y d ed u ctio n rules of T h eorem 2 , th a t th e free v a ria b les
sh ould b e h eld c o n sta n t for each assu m ption form u la to b e disch arged,
ca n b e illu m in a ted b y som e rem arks on th e in terpretatio n , w h ich of
course are n o t p a rt of th e m e ta m a th e m a tics.
In E x a m p le 2 follow in g, w e h a v e in S te p 1 a d ed u ctio n from th e as
su m p tio n form ula bÔ w ith b h eld co n sta n t, so th a t D -in tr o d u c tio n is
a p p lica b le a t once. In E x a m p le 3, D -in tr o d u c tio n is not ap p licab le a t
on ce becau se b is va ried , a lth o u g h it can b e ap p lied a fter an V -elim in atio n .
§32 DERIVED RULES, FREE VARIABLES 149
In E x a m p le 4, w e see h ow a false result (for th e n u m b er-th eo retic in ter
pretation) is o b ta in ab le, if w e v io la te th e restriction on su b sid iary d e
duction.
E xample 2. 1. bÔ b a+bâ — can be established , h old in g b

co n sta n t, in th e n u m b er-th eo retic sy ste m (§39 ; “ s ^ t ” a b b re v ia te s
-is = t, § 17).
2. b b ^ O D-introd., 1.
a + b ^ a —
3. b V b (b ^ 0 D a + b ^ a ) — V -in tro d ., 2.
E xample 3. 1. b=£ 0 b^ 0 ^ 0 — su b stitu tio n (§23 ), b b ein g varied .

2. V b (b j£ 0) b 0 ^ 0 — V -elim ., 1 .
3. b 0) D 0^0 —D-introd., 2.
V b {b = £
Example 4. 1. b=£0 b^0^0 — same as E x a m p le 3 S te p 1.

2? b b=£0 D0^0—D-introd., misapplied to 1.
3? b V b { b ^ 0 D 0 ^ 0 ) — V -in tro d ,, 2.
4? b 0V0 D 0^0 — V -elim ., 3.
5. b O'yÔ — b y su b stitu tio n in A x io m 15.
6? b 0^0 — R u le 2, 5, 4.
T h e o n ly v io la tio n of th e form al rules in E x a m p le 4 is a t S te p 2 . I t is

su ggested th a t th e reader u n d erta k e h im self to e x p la in th e fa lla c y in
term s of th e in terpretatio n , before read ing th e fu rth er discussion belo w .
E s p e c ia lly he should n o te th e difference b etw ee n th e form ulas a t S te p 3
of E x a m p le s 3 an d 4 . T h e reader m a y also s u p p ly th e form al d eta ils
for th e follow in g tw o exam p les, an d com pare th e results.
E xample 5. G iv e n A (x ) bB an d A (x ) b ~ iB w ith x h eld co n sta n t,

th en b ~»A(x) an d b V x -iA ( x ).
E xample 6. G iv e n A (x ) bx B an d A (x ) bx ~ iB w ith x n o t neces
sarily h eld co n sta n t, th en b “ iV x A (x ).
T h e rules of V -elim in ation an d 3 -elim in ation m a y be discussed in lik e
m anner.
In inform al m ath em atics, w e kn o w tw o d ifferen t w a y s of usin g free
va riab les in s ta tin g propositions, as illu stra ted in algeb ra b y an identical
equation (x + y )2 = x 2 + 2 xy + y 2 an d a conditional equation x 2 + 2 = 3x.
T h e first of these in terp retatio n s is th e one w h ich applies to th e free
va riab les in th e axio m s an d form al theorem s of our syste m , an d w e
call it th e generality interpretation. F o r ex am p le, A x io m 14, w h ich is
a '= b ' D a = b , m eans th a t, for e v e r y pair a an d b of n a tu ral num bers,
if a ' = b'y th en a = 6 ; an d th e form al theorem a=a, p ro v ed in E x a m p le

1 § 19, expresses th a t e v e r y n a tu ral n um ber equals itself.
B u t w h en a form u la A (x ) w ith a free va ria b le x is ta k e n as an assu m p
tio n fo rm u la for a form al d ed u ctio n , w e h a v e a choice. W e m a y in ten d th e
a ssu m p tio n in th e sense “ Su p p ose th a t, for all A (x )’’, so th a t x has th e
g e n e r a lity in terp retatio n . O r w e m a y in ten d th e assu m p tion in th e sense
“ L e t x b e a n u m b er such th a t A {x )” , in w h ich case w e sa y x h as th e con
ditional interpretation.
In th e second case, th e use in th e d ed u ctio n of operations w h ich depend,
considered in term s of th e m eaning, on th e p o s sib ility of allow in g x to
range o ver th e o b je ct d om ain w ill b e out of k e e p in g w ith th e in terp reta
tion. F o r th e g e n e ra lity in terpretatio n , there is no such lim itatio n . S te p 1
E x a m p le 2 is a d ed u ctio n co n stru cted in keep in g w ith th e con dition al
in te rp re ta tio n ; an d S te p 1 E x a m p le s 3 an d 4 o n ly w ith th e g e n e ra lity
in terp retatio n . T h a t th e assu m ption form ula bÔ is false under th e
la tte r in terp retatio n is beside th e p o in t here, an d th e co n clu din g form ula
of E x a m p le 3 is q u ite correct (alth ough n ot v e r y interesting).
T h e stu d e n t of ele m e n ta ry m a th e m a tics is a c q u a in te d w ith th e dis
tin ctio n b etw ee n sy m b o ls classified as co n sta n ts an d sy m b o ls classified as
va riab les. Close in spectio n show s th a t th e d istin ctio n in th e use of th e
sym b o ls is a lw a y s r e la tiv e to a c o n te x t. A g iv e n sy m b o l is in tro d u ced as
nam e for an o b je c t, an d th ro u gh o u t a certa in c o n te x t e v e ry occurrence
of th e s y m b o l is as nam e for th e sam e o b je ct. F ro m outside th e c o n te x t,
it is in d ic a te d t h a t th e o b je c t m a y b e a n y one (some one, etc.) o f th e
m em bers of som e set. F u n d a m e n ta lly th en , th e sy m b o l is co n sta n t, i.e.
its m ean in g ca n n o t b e ch an ged , w ith in th e c o n te x t, w hile from ou tsid e
th e c o n te x t it is va ria b le.
(T h e term in o lo g y a c tu a lly e m p lo y ed in a g iv e n th e o r y is g e n era lly th a t
su ita b le to th e c o n te x t c o n stitu te d b y th e th e o r y as a w hole. Som etim es
sy m b o ls w h ich are co n sta n t th ro u gh o u t an im p o rta n t su b c o n te x t, b u t
v a ria b le for th e th e o r y as a w hole, are called “ p ara m eters” or “ a rb itra ry
c o n sta n ts” .)
F o r th e g e n e ra lity in terp reta tio n of a va ria b le x in a form u la A (x ), th e
c o n te x t w ith in w h ich all (free) occurrences of x m u st represent th e sam e
o b je c t is e x a c t ly th e w h ole form ula A (x ). T h e form u la A (x ) th en m eans
th e sam e as V x A ( x ) , an d in a n a lo g y to th e scope of th e q u a n tifier V x , w e
also ca ll A (x ) th e scope of th e g e n e ra lity expressed b y th e free v a ria b le x.
F o r th e co n d itio n al in terp retatio n , th e c o n te x t w ith in w h ich all (free)
occurrences of x halve th e sam e m ean in g is not ju st A (x ) b u t th e w hole
d ed u ctio n from A (x) (or th e part of it d epen d en t on A (x)).
§33 REPLACEMENT 151
In Example 4 Step 1, under the generality interpretation the scope

of the generality expressed by b is exactly the assumption formula b ^0.
Were the formula of Step 2 provable, the generality interpretation would
then apply to it as a whole, and the scope would become b=£Q 3 0#0,
not just the part b ^ 0.
The universal quantifier Vx functions in our logical symbolism as
a device to restrict the scope of generality to a part of a formula. The
formulas of Steps 2 and 3 Example 4 (or of the two conclusions of Example
5) are synonymous under the generality interpretation of the free variable
in the first. No formula without a quantifier can be written which is
synonymous with the formula of Step 3 Example 3 (or of the conclusion
of Example 6).
Let A be a formula containing free exactly the distinct variables
xv ..., xn in order of first occurrence. According as n > 0 or n = 0,
we call A open or closed. The closed formula Vxx ... VxnA (sometimes
abbreviated “VA”) we call the closure of A. By *64 and *65, A and VA are
interdeducible (i.e. each can be deduced from the other), with xv ..., xn
varied in the deduction of VA from A. Under the generality interpretation
A and VA are synonymous.§*
§ 33. Replacement. Let CAbe a formula in which there is a specified
occurrence of a formula A, and let CB be the result of replacing this
occurrence by B (§ 26). There will then exist parallel constructions of CA
from A and of CB from B by Clauses 2 — 7 of the definition of formula
(§ 17 or § 31). The number of steps in this construction of CA (or of oper
ators in the scopes of which the part A lies) we call the depth of the part
A in CA.
E xample 1. Let A be 3l(b,a)f CA be Vb(£l(b, a) & S 3 £Z(a, b))
(the specified occurrence here being the only occurrence), and B be
3d^l(d, a, c). Then CB is Vb(3d<iX(d, a,c) &B 3 <3(a, b)). The parallel
constructions of CA from A and of CB from B are as follows, and the
depth is 3.
«Vl(b, a) ld<^l(d, a, c)
Qi(b, a) 8c 3 3d a {d , a,c)8c 3
csl(b, a) & 3 3 b) 3 d ^ { d , a , c ) 8 c 3 D <3T(a , b)
Vbiplib, a) 8c 3 d ( a , b))
d 'ib$d£Z{d, a, c) 8c 3 D <3((a, £))
T heorem 14. If A, B, CAand CB are formulas related as in the foregoing
discussion of replacement, then
A ~ B bXl**’x* CA ~ CB
where x v . . . , x n are the free variables of A or B which belong to a quantifier

of C A having the specified occurrence of A w ithin its scope. (R ep lacem e n t
theorem .)
In oth er w ords, x lf . . . , x n are th e free v a ria b les of A ~ B w h ich

are q u a n tifie d in th e co n stru ctio n of C A from th e specified p a rt A . P roof
of th e theorem is b y in d u ctio n on th e d e p th as before (Theorem 6 § 26),
usin g n ow tw o a d d itio n a l lem m as.
A dditional lemmas for replacement. I f x is a variable , and A (x)

and B (x ) are form u las :
*71. A(x)~B(x) bx V x A (x ) ~ V x B (x ).
*7 2 . A(x)~B(x) bx 3 x A (x ) ~ 3 x B (x ).
P roofs. F ro m *6 9 an d *70 , re sp e ctive ly.
E xample 1 (concluded). Let ~ b e w ritte n b etw ee n th e form ulas

of each pair in th e p arallel colum ns. T h e resu ltin g form ulas are ded ucible,
each from th e p reced in g, u sin g su c c essiv ely *28a, *2 6 an d * 7 1 (v a ry in g b ).
Corollary 1. U nder the conditions of the theorem :

If f- A ^ B , then h CA ^ C B.
E xample 2°. By *49 *20), b A (x ) ^ n - i A ( x ) .

(and H en ce
b - n V x A ( x ) ~ -i - i Vx - i- i A(x).
E xample 3. N o w (cf. E x a m p le 2 § 26), if A an d B do n o t co n ta in
x free: A ~ B bA V V x (AD C(x)) ~ A V V x ( B D C(x)). I f A an d B
m a y co n ta in x free: A ~ bx A V V x ( A D C(x)) ~ A V V x ( B D C(x)).
B
If b A ~ B , th e n b A V V x ( A D C(x)) — A V V x ( B D C(x)).
Corollary 2. U nder the conditions of the theorem :
A ~ B , C A bXl"'Xw C B, with x lf . . . , x w varied only for the first as
sum ption form ula. I f b A ^ B , then CA b C B. (R ep lacem e n t p ro p e rty of
equ ivalen ce.)
A rep lacem en t m a y be p receded b y a s u b stitu tio n for in d iv id u a l

v a ria b les (* 66 ).
E xample 4. b+ 0= a ~ a=b \-b b ' + 0= a ~ a = b ' \-b

l b (b f -\ - 0= a) ~ 3 b (a — b '). B u t in § 38 w e w ill h a v e b b -\ - = a0 ^ a= b.
H en ce th en b 3 ^ ( ^ '+ 0 = ^ ) ~ 3 b {a = b ').
T h e in terp reta tio n illu m in a tes th e tr e a tm e n t of th e v a ria b les in re

p lacem en t. In inform al m a th e m a tics, k n o w in g t h a t sin x is th e sam e
fu n ctio n as cos (n/2 — x), i.e. h a v in g sin x = cos (tc/2 — x) as an id e n tica l
§33 REPLACEMENT 153
x ta k in g th e g e n e ra lity in terp retatio n § 32), w e are ju stifie d
e q u a tio n (w ith
in rep lacin g “ sinx ” b y “ cos (n/2 — x)'* in ‘ To s*n x d x ’\ a n d in re p lacin g
“ sin 2x” b y “ cos (n/2 — 2x)” . B u t assu m in g th e co n d itio n a l eq u a tio n
s i n x = 1 — x g iv e s us no rig h t to replace “ sin x ” b y “ 1 — x ” in
“To sin x dx” , or to replace “ sin 2x ” b y “ 1 — 2 x '\
Change of bound variables. T wo form ulas A an d B w ill b e said

to b e congruent, if A an d B h a v e th e sam e n u m ber k o f sy m b o ls, an d
for each i (i = 1, . . . , k) : (I) I f th e i- th sy m b o l o f A is n o t a v a ria b le ,
th e n th e i- t h sy m b o l of B is th e sam e sy m b o l. (II) I f th e i -t h sym bol of
A is a free occurrence o f a v a ria b le, th en th e i-th sy m b o l o f B is a free
occurrence of th e sam e va ria b le . ( I l l ) I f th e i-th sy m b o l o f A is a n o c
currence of a v a ria b le b o u n d b y th e /-th q u a n tifie r o f A , th e n th e i- t h
sy m b o l of B is an occurrence of a va ria b le (not n ecessarily o f th e sam e
variab le) b o u n d b y th e /-th q u a n tifie r of B .
In brief, tw o form ulas are con gru en t, if t h e y differ o n ly in th eir b o u n d
va riab les, an d correspon ding b o u n d va ria b le s are bound by corre
sp o n d in g quantifiers.
E xample The two following form ulas are co n gru en t:

V a ^ i a , c) V 1 aS(a) 3 3 bC{a, b ))t Vb[C2(b, c) V 3 cS(c) 3 3 aC[b, a)).
T h is ca n b e m ad e a p p a ren t b y in tro d u cin g indices to sh ow w h ic h o c
currences of va ria b le s are b o u n d b y th e sam e q u a n tifie r:
'ia l { a { a 1,c) V 3 a 2S(tf2) 3 3bzC(av b s)), V b ^ a ^ c ) V 3 c 2B (c2) D l a zC(bv a 3)).
I f w e erase o u t th e b o u n d v a ria b le s th em selves, le a v in g b la n k s n u m b ered
w ith th e indices, id e n tica l expressions w ill b e o b tain ed .
Lemma 15a. If x is a variable, A (x ) is a formula, and b is a variable

such that (i) b is free for x in A (x ), and (ii) b does not occur free in A (x )
[unless b is x), then:
*7 3 . b V x A (x ) ~ V b A (b ). *7 4 . b 3 x A (x ) ~ 3 b A (b ).
P roofs. *73. B y E x a m p le 3 § 22 (Steps 1— 2), b V b A (b ) D V x A ( x ) ,

an d sim ila rly b V x A ( x ) D V b A (b ). — N o te b y E x a m p le 9 § 18 th a t
(i) an d (ii) are n ecessary a n d su fficien t in order t h a t V b A (b ) b e co n gru en t
to V x A ( x ) (or 3 b A (b ) to 3 x A (x )).
Lemma 15b. Congruent formulas are equivalent; i.e. if A is congruent to

B, then b A ^ B .
P roof. L e t A co n ta in in order e x a c tly r q u an tifiers w ith re sp e c tiv e

v a ria b le s nv . . . , u r (not necessarily d istin ct). L e t w x, . . . , w r b e d istin c t
v a ria b les occurring neith er in A nor in B . L e t C com e from A b y ch a n g in g
each u* (/ = 1 , i . . , r) to w* in e x a c tly th e occurrences w h ich th e /-th

q u a n tifie r binds. B y r su ccessive replacem ents, u sin g * 7 3 or * 7 4 an d
C o ro lla ry 1 T h eo rem 14, f- A ~ C. S im ilarly, b B ~ C. H en ce (*20, *21),
b A ~ B.
E x a m pl e 5 (concluded), b ^ a { H { a , c) V l a B ( a ) D 3bC(a, b)) ~

V d {^ l{d ,c) V Z < a{a)-D *bC {d9b)) [*73] — V d(£l{dt c) V 3 ^ ) D 3 bC {d,b))
[*74] ~ V d {Z (d , c) V 3e3(e) D 3fC(</, /')) [*74]. S im ilarly
b V b(£l(b, c) V 3 c B (c )D 3 *C (£ , *)) — V ^ ( ^ , c) V 3 e % ) D 3/t(</, f ) ) .
R emark 1. (a) S im ila rly to T h eo rem 14, by u sin g *6 — *9b, *12, *69,
*7 0 (instead of *2 6 — *30, * 7 1 , *72) as th e le m m a s : Let the part A stand in CA
w ith in the scopes of only certain of the sym bols D , & , V, - i , V , 3.
A d B bXl*'*Xn C a d C b or B D A bXl'"x* C A D C B m system having
as postulates only the D -postulates and the postulates for the sym bols in
question, provided that in case the sym bols include V but not & the ^-postulates
include A xiom Schema 9a of Lem m a 11 § 24. (Cf. H erb ran d 1930 § 3.2,
M a c L a n e 19 34 pp. 28 ff., C u rry 1939 pp. 290— 291.)
(b) b and b V b A (b ) D V x A ( x ) {sim ilarly with 3)
V x A ( x ) D V b A (b )
using only Postulates 9 and 10 (11 and 12). (c) T h e r e fo r e : I f A is congruent
to B , then b A d B and b B D A {and hence A and B are interdeducible)
using only the D - postulates and the postulates for {at most) the logical
sym bols which A contains, provided as in (a).
P ermanent abbreviations. O u r use of p erm an en t a b b rev ia tio n s,
su ch as “ a < b ” (discussed a t th e end of § 17) will d iffer from t h a t of
te m p o r a ry ab b re v ia tio n s, such as “ A ( x ) ” , “ A ( x 1, . . . , x n)” , etc. (§ 18),
in tw o respects. F ir st, t h e y sh all n o t co n ta in free va ria b les n o t sh ow n in
th e a b b re v ia tio n (‘an o n ym o u s free v a ria b le s’). Second , in stea d of a v o id in g
th e s u b stitu tio n of term s n o t free a t th e su b stitu tio n positions, w e p erm it
th e b o u n d va ria b le s suppressed b y th e a b b re v ia tio n (‘an o n ym o u s b o u n d
v a r ia b le s ’) to b e chosen a t w ill to m a k e w h a te v e r term s w e w ish to su b
s titu te free a t th e s u b s titu tio n positions. A ll le g itim a te u n a b b re v ia tio n s
o f a g iv e n a b b re v ia tio n are co n gru en t, an d hence b y L e m m a 15b e q u iv a
len t. T h u s it is im m a terial in con sid erin g question s of d e d u c ib ility an d
p r o v a b ility w h ich le g itim a te u n a b b re v ia tio n is used.
F o r th e q u estio n w h e th er a p o stu la te applies, th e m ann er o f u n
a b b r e v ia tin g m a y m ak e a d ifferen ce, e.g. s e t D (B D s e t ) is an a x io m
b y S c h em a l a o n ly if b o th occurrences of “ s e t ” are u n a b b re v ia te d alike.
H e rea fter in our s ta te m e n ts t h a t a p o stu la te applies w e sh all b e t a c it ly
su p p o sin g t h a t like ap p earin g a b b rev ia tio n s are u n a b b re v ia te d alike.
§34 SUBSTITUTION 155
*§ 34. Substitution. The use of a form al su b stitu tio n rule for
p re d ic a te letters ca n b e la rg e ly a v o id e d b y s ta tin g results in th e sch em a tic
form , w ith m e ta m a th e m a tic a l letters in stea d o f p a rticu la r p red icate
letters. W e th e n s u b stitu te in fo rm a lly in a p p ly in g th e results w ith a
ch a n ge in th e sign ificatio n of th e m e ta m a th e m a tic a l letters, b u t th is
su b s titu tio n does not c o n s titu te a p p lica tio n of a form al su b stitu tio n rule.
W e h a v e b een d o in g th is c o n tin u a lly , from th e v e r y b eg in n in g of our s tu d y
o f th e form al sy ste m . O n e n ew e x a m p le is g iv e n to sh ow w h a t is m ean t.
E xample 1 °. As *83 (§ 35) we will estab lish that, if x is a n y v a ria b le

an d A (x ) a n y form ula, b 3 x A (x ) ~ -iV x -tA (x ). N o w le t x be any
v a ria b le an d A (x) a n y form ula. B y ta k in g th e n eg a tio n - i A ( x ) of th is
A(x) as th e A (x ) of *83, w e h a v e b 3x-iA(x) ~ -iVx-i-iA(x). T h is
an d E x a m p le 2 § 33 e x p la in th e second an d th ird step s o f th e follow in g
c h a in : b V xA (x) ~ —i- iVxA(x) [*49] ~ n n V x -i-iA (x ) [*49]
------- i 3 x - i A ( x ) [*8 3 ].
T h e o n ly essential use w e sh all m a k e of th e form al su b stitu tio n rule

for th e p red icate ca lcu lu s (Theorem 15) is in e stab lish in g d u a lity (Corol
la r y T h eorem 18 § 35), w here w e su b s titu te n egatio n s o f th e p red ic a te
le tte rs for th e letters. T h a t su b stitu tio n ca n b e ju stifie d b y reasonin g
a lr e a d y used in p ro v in g th e s u b s titu tio n rule for th e p rop osition al ca lcu lu s
(Theorem 3 § 25), w ith no n ew co m p licatio n s. A fu rth er a p p lic a tio n of
form al su b stitu tio n occurs in p assin g from a n u m b er o f results, first p ro v ed
b y d u a lity in term s of p a rticu la r p red icate letters, to th e gen eral results of
th e sam e form w ith m e ta m a th e m a tic a l letters. T h a t a p p lic a tio n co u ld
b e a v o id e d b y u sin g th e su b s titu tio n rule o n ly h eu ristica lly, to d isco v er
proofs th a t w e can afterw ard s v a lid a te w ith o u t use o f it. T h e reader m a y
therefore, if he w ishes, o m it th e d eta ile d tre a tm e n t of s u b stitu tio n g iv e n
in th e rem ainder of th is section.
A gen eralized n o tio n of 'occurren ce' is ap p ro p riate to th e n am e form
in terp retatio n . In in form al m a th e m a tics, " s in x " as an expression for a
fu n ctio n occurs in " 3 sin x + cos x " , in " / 0* sin x d x ” a n d in " c o s x sin 2 x ” ,
a lth o u g h as an expression for a n u m ber it occurs o n ly in th e first.
T o sim p lify th e n o ta tio n in th is section, w e sh all a n a ly z e each su b
s titu tio n ta k in g th e a tta c h e d va ria b le s for each o f th e d istin c t p red ic a te
le tte rs to b e th e first n va ria b le s from an in fin ite list a x, a 2> a 3, . . . of
va ria b le s (cf. § 3 1). H o w e v e r th is list m a y h a v e to b e chosen d iffe re n tly
for d ifferen t su b stitu tio n s (see below ).
By an occurrence of a p red icate le tte r P ( a x, . . . , a n) w ith a tta c h e d
v a ria b les in a p red icate le tte r form ula E w e sh all m ean a (consecutive)
p art of E of the form P ( t 3, . . . , t w) where tv . . . , t n are term s. A p red ic a te
le tte r form ula E is said to b e a p red icate le tte r form ula in th e d istin c t
p red icate letters
( 1) P ^ a ,, . . a^) , . . P m(a1( . . a„J K , ..., nm > 0; m > 1 ).

if no p red icate letters oth er th a n ( 1 ) occur in E .
E xample 2. 31(a, b) occurs tw ic e in yi b (3 1 (b , a)

T h e p red icate le tte r
& B D (a, b)), first as th e p art c31(b, a) an d second as th e p a rt <31(a, b ).
T h e form ula yib(31(b, a) & B D 3t(a, b)) is a p red icate le tte r fo rm u la in
31(a, b), B, C(a, b, c).
The substitution of form ulas (in either sense § 31)
(2) A1(a1, . . an Aw(a1, . • •> anw),

considered as nam e form s in th e resp ective v a riab les show n, for th e p re
d ica te letters ( 1 ) in E (w ith result E * ) sh all consist in replacin g, sim u l
ta n e o u s ly for each j (j = 1,..., m), each occurrence P i (t1, . . . , t».) of
P ^ a j, . . . , a^) in E by A,^, . . . , t n/).
T h e nam e form v a riab les (i.e. th e a's) do n o t ap p ear as such in E an d
E * . In ask in g w h eth er a g iv e n form ula E * does com e from an oth er g iv e n
form u la E b y su b stitu tio n for certain p red icate letters, it suffices to a sk
th e qu estio n ta k in g as th e nam e form va riab les ones n o t occurring in
E an d E * , th o u gh w e are n o t restrictin g ourselves to such a choice of th e
n am e form v a riab les w hen others w ill do also.
The su b stitu tio n is said to be free , if for each j (j = 1, ...,m ) ,
A i (a1, . . . , a Uj) is T ree’ for P^(a1, . . . , an;) in E , in th e fo llo w in g sense:
A ( a x, . . . , a n) is free for P ( a x, . . . , a n) in E , if, for each occurrence
P ( t x, . . . , t n) of P (a 1, . . . , a n) in E , ( Al ) t 1# . . . , t n are free for a x, . . . , a n,
re sp e c tiv e ly , in A ( a v . . . , a n), an d (A 2 ) P ( t x, . . t w) does n o t sta n d in
E w ith in th e scope of a q u a n tifier V y or 3 y w here y is a free va ria b le o f
A ( a x, . . . , a n) oth er th a n one of th e nam e form va riab les a 1# . . a n.
E xample 3. Let m=
1 ; n = nx = 2 ; a 2, a 2 b e c, d P ( a x, a 2) b e \
31 (c, d ) ; A ( a x, a 2) yi b <B {a, b, c, d V 31 d c ) ; an d E b e 3c31(c, a ).

be ) ( ,
T h e n E * is 3 c{VbB(a, b, c, a) V 31(a, c)). T h e su b stitu tio n is free.
It is co n ven ien t to refer to th e free occurrences of a 3, . . a n in

explicit occurrences; an d to other occurrences of va ria b le s
A ( a x, . . . , a n) as
as anonymous occurrences V aria bles occurring e x
in A ( a 1, . . . , a n) .
p lic itly [an onym ously] free (bound) in A ( a x, . . . , a n) are explicit anony [
mous free bound variables of A ( a 3, . . . , a n). T h e term in o lo gy e x te n d s

] ( )
ge n e ra lly to situ atio n s in w h ich form ulas or p arts of form ulas are b ein g
represen ted b y m e ta m a th e m a tic a l letters.
E xam ple 4. In considering <C?(a, b) & âS(a} h, c) as a name form
in a, b} and in using “A(a, b),y (or “A(a1} a2)” where , <<a2Mstand for
a, b) to stand for it, the first occurrence of a and both occurrences of b
are explicit, the second and third occurrences of a and the occurrence of
c are anonymous. So a and b are explicit free variables, c is an anonymous
free variable, and a is an anonymous bound variable. In using ("iak{af b)ff
to stand for "ia(c^(af b) & 3aB(a, b, c)), the first two occurrences of a are
explicit, the other two anonymous; so a is both an explicit bound variable
and an anonymous bound variable. The Va is an explicit quantifier, and
the 3a an anonymous quantifier.
In our further examples of substitution, the variables ap a2, a3, ...
will be ciy by c, .... (This choice can always be made, except when as
in Example 3 it would interfere with the anonymous variables for the
substitution.)
Failure of (Al) or (A2) is always due to the presence of anonymous
variables.
E x a m p l e 5. The formula 3ce^?(c, a, b) is not free for c1(a) in c2f(c) D B,
because (Al) is violated (after substitution, with result 3cc2f(c, c, b) D B,
the c of c) would become bound by the anonymous quantifier 3c of
3ce2f(c, a, b))y nor in Wb(^3(a) 3 S(i>)), because (A2) is violated (after
substitution, with result V/>(3cc2f(c, a, b) D B(b)), the anonymous free b of
3ce2f(c, a, b) would become bound by the Vb of yib(^(a) D B(^)))-
The conditions (Al) and (A2) can be regarded as conditions that each
part A(tx, ..., tn) ofE*resulting by the substitution of A(ax, ..., aw) for
P(a1} .. ., an) should constitute an occurrence of A(ax, ..., an) as a name
form in at, ..., an.
E x a m p l e 6 . The a’s in this example shall be a, bt c. But we supply
indices to assist in referring to different occurrences of the variables.
Let E be
(i) V ^ (c ^ 2, az) & B D <Jt(aAybn)).
Let A(a,b)y B, C(a,b}c) (to be substituted for <H(a, b)t B, C(a,b,c),
respectively) be
(ii) 3c6C(c7, a, b, b), n B ^ ), ^(a, b)
(the indexed occurrences of variables being anonymous). Then E* (the
result of the substitution performed on E) is
(iii) Viq(3c6C(c7, b2) a3>a3) & -iB(a8) D 3ctiC(c7, a4, b5, b5)).
The substitution is free.
T h e m ean in g of th e n e x t lem m a w ill b e m ad e clear b y th e e x a m p le

fo llo w in g it.
Lemma 16a.I f the substitution of (2 ) for ( 1) in E is free , then in the

result E * each free occurrence of a variable originates as a free occurrence of
the variable either in E or anonym ously in some A ^ a ^ . . an .), and each
bound occurrence of a variable and the quantifier binding it originate together
in the sam e relationship either in E or anonym ously in some A ^ a ^ . . . , an;.).
E x a m p l e 6 (co n clu d ed ). T h e tw o free a z s in (iii) o rigin ate as th e
free a z in (i). T h e free a 8 o rigin ates as th e a n o n ym o u s free a 8 in (ii). T h e
tw o b b’$ b o u n d b y o rigin ate to g e th e r as th e hb b o u n d b y yi b 1 in (i).
T h e c 7 b o u n d b y 3c 6 (either such pair) origin ate as th e an o n ym o u s c1
b o u n d b y 3 ce in (ii).
Outline of proof. C onsider a g iv e n occurrence of a v a ria b le in E*.

T h is is either (Case 1 ) n o t in a n y of th e p arts A ^ , . . t nj) (e.g. b j, or
(Case 2 ) in one o f th e p arts A ^ t ^ . . t»;) b u t n o t in a n y t t- (e.g. c 6, c 7, a 8),
or (Case 3) in one of th e occurrences of t t w h ic h is in tro d u ced in to
A ^ tj, . . . , t nj) b y s u b stitu tio n for a free occurrence o f a t in A ^ a ^ . . . , an.)
(e.g. b 2) a z, a 4, bb). T h e lem m a follow s, usin g in C ase 2 th e co n d itio n (A 2 )
for freedom , an d in C ase 3 th e co n d itio n (A l) .
Lemma- 16b. Let

(2) î(ai> • • •» • • • >X.m(a1, . . . , anm)
be form ulas respectively congruent to the form ulas (2 ), and let P be a form ula
congruent to F . I f the substitution of (2 ) for ( 1) in F with result F * , and the
substitution of (2 ) for ( 1) in P w ith result p t, are both free , then F 1- is con
gruent to F * .
B y L e m m a 16a, n o tin g th a t, if th e ^>-th sy m b o l o f F * o rigin ates as
th e y -th o f F (of A i (a1> . . . , an/)), th e ^ -th sy m b o l of f t origin ates as th e
? - th of P (of A ^ a ^ . . a«,•)).
L emma 17. Given a proof of F, and a list of variables z lt . . . , z ff, we

can fin d a form ula P congruent to F and containing none of the variables
Zj, . . z q bound, and a proof of P containing no applications of R ule 9
or 12 w ith respect to any of z lf . . z a.
P roof. L e t th e d is tin c t free v a ria b les of F b e bv . . . , b s, an d ca ll
F also “ F^bj, . . . , b ,) ” . L e t uv . . . , u r be all th e d istin ct va riab les
occu rrin g free or b o u n d in a n y form u la of th e g iv e n proof of F (including
bv . . . , b*). L e t uv . . . , u r be new variab les, d istin ct from each other
an d from nv . . . , u r, zv ..., z q. F o r th e d efin ition of w h a t co n stitu te s
a proof in th e p red icate ca lcu lu s (P o stu la te G ro u p A § 19), all v a ria b le s
are on a par in itia lly . H en ce if, th ro u g h o u t th e g iv e n p roof of F , w e ch an ge
u 1# . . u r sim u lta n eo u sly in all occurrences free an d b o u n d to nlf . . . , u r,
re sp e c tiv e ly , th e resu ltin g figu re m u st also b e a proof. S a y it is a p roof of
P(bx,. . . , b s); so (a) b b 8). N o w b y w -fold s u b stitu tio n (* 66 ),
(b) F ( b x, . . . , b s) bî ••• *9 P (b lf . . . , b 8). R eferrin g to th e proof of * 66 ,
bv . . . , bg are d istin ct from b x, . . . , b„ th e d ed u ctio n (b) requires
of R u le 9 o n ly with respect to b1# ..., bg (and R u le 12 n o t a t all).
since
th e use
Let P b e P (b v . . . , b 8). C o m b in in g (a) an d (b), w e o b ta in a p roof of P in
w h ich R u le s 9 a n d 12 are used o n ly w ith resp ect to th e n ew v a ria b le s
u lf . . . , u r (including b lf . . . , b g), therefore n o t w ith respect to a n y of
Zj, ..., zff.
T heorem 15. S ubstitution for predicate letters. L e tD l } .. . , D ,, E
be predicate letter form ulas in the distinct predicate letters ( 1 ). L et D * , . . . ,D *,
E * result by the substitution of (2 ) (as name form s in the variables shown) for
( 1 ) throughout D v . , . , D f, E , respectively. Then, provided that
(A) the substitution is free, and
(B) for j = 1, . . . , m, the anonym ous free variables of A *(a1# . . . , an/)
are held constant in the given deduction for each assum ption form ula which
contains the corresponding predicate letter P i (a1, . . . , a Uj) :
I f D 1( . . . , D , f- E , then D f , . . . , D * |- E * .
(P roviso (B) is of course satisfied , if all th e an o n ym o u s free v a ria b le s
of (2) are h eld con stant in th e g iv e n d ed u ctio n . In th e case 1 = 0, P ro viso
(B) disappears, an d w e h a v e s im p ly : If b E , then b E * , provided (A)
the substitution is free.)
P roof . We ta k e th e g iv e n d ed u ctio n D x, . . . , D ? bE as (I) for
L e m m a 8 a (§ 24), an d pass to a proof (II). T h e lo n g fo rm u la in (II) w e
now w rite as " F ” . T h e su b stitu tio n of (2) for ( 1 ) in F is free, as w e see b y
using P ro viso (A), an d also P ro v iso (B) to insure th a t th e co n d itio n (A2)
is m et resp ectin g th e y ’s of (II). I f w e can sh ow n ow t h a t b F * , th e n b y
using L e m m a 8 a in th e con verse d irection , it w ill follow th a t D * , . . . , D |
b E * , as is to b e p roved .
A c c o rd in g ly , consider a g iv e n proof of F . In form ulas of th is proof
there m a y occur som e o th er p re d ic a te letters th a n ( 1 ). L e t (V) b e th e list
( 1 ) increased to in clu d e th ese ; an d le t (2 ') b e (2 ) increased co rrespo n d in gly,
using as th e a d d itio n a l n am e form s form ulas w h ich co n ta in no an o n ym o u s
variables.
Suppose w e w ere to su b stitu te (2 ') for ( 1 ') th ro u g h o u t th e g iv e n p roof
of F . T h e n w e could reason, e x a c tly as in th e proof of th e s u b stitu tio n
rule for proposition letters (Theorem 3 § 25), th a t th e resu ltin g figure is a

p roof of F * , e x c e p t for tw o con tin gencies.
F irst, th e C for an a p p lica tio n of R u le 9 or 12 m a y b e tran sform ed b y
th e s u b stitu tio n in to a form u la C * w h ich co n tain s th e x of th e ap p lica tio n
free, so th a t th e rule no longer applies. T h is can h ap p e n o n ly if th e x of
th e a p p lic a tio n is one of th e a n o n ym o u s free va ria b le s zv ..., zq of (2 ).
U sin g L e m m a 17, w e can replace th e g iv e n proof of F b y a proof of a
fo rm u la F co n gru en t to F an d c o n ta in in g none of zv ..., z q bound, so
t h a t in th e new proof there are no ap p licatio n s of R u le s 9 an d 12 w ith
resp ect to zv . . . , z 5.
Second, th e t for an a p p lica tio n of A x io m S ch em a 10 or 11 m a y a fte r
th e su b stitu tio n no longer b e free for th e x in th e A (x ), due to th e in
tro d u ctio n o f an o n ym o u s q u an tifiers in th e form ulas (2 ) w ith va ria b le s
occurring in th e t. L e t us choose form ulas ( 2 ') con gru en t to (2 ') an d
co n ta in in g b o u n d no v a ria b les occurring either free or b o u n d in a n y for
m u la of th e proof of F.
N o w if w e su b stitu te (2') for (V) in th e proof of F , neith er co n tin g e n c y
ca n arise; an d so, d en o tin g th e su b stitu tio n of ( 2 ') b y “ t ” , th e resu ltin g
seq uence o f form ulas w ill b e a proof of F L T h u s b F t.
B y th e ch oice of th e b o u n d va ria b le s in (2')> co n d itio n ( Al ) for freedom
is satisfied in th e s u b stitu tio n of (2) for (1) in F. B eca u se F does n o t
co n ta in b o u n d a n y of th e a n o n ym o u s free v a ria b les zv . . .,z q of (2 ),
h ence of ( 2 ), co n ditio n (A 2 ) is satisfied also. T h u s th e su b stitu tio n of (2 )
for ( 1 ) in P is free. So is th a t of (2) for ( 1 ) in F (as rem arked ab o ve). H en ce
b y L e m m a 16b, F t is co n gru en t to F * ; an d b y L e m m a 15b, b F t ~ F *.
T h is w ith b P* (and *18 a) g iv e s b F * , as rem ained to b e show n.
E xam ple 7 °. W e sh all p ro v e (E x a m p le 2 § 35) th a t

(a) b <3 V VflS(tf) ~ V#(<3f V B(a)). Thence b y L em m a 15b,
(b) b V V x S (x ) ~ V x ( J 7 V S (x )), where x is a n y v a riab le. T h e n c e
b y T h eo rem 15, (c) b A V V x B ( x ) ^ V x ( A V B ( x ) ) , w here B (x ) is a n y
form ula, an d A a n y fo rm u la n o t co n ta in in g x free (as otherw ise (A 2 ) for
P ro viso (A) w o u ld b e v io la ted ). T h is is *9 2 w ith th e sam e stip u la tio n s as
a p p ear in T h eo rem 17. F ro m (b) w e ea s ily infer (d) c C2 V V x S (x ) b ^ V S (x)
an d (e) <3 V S (x) bx ^ I V V x S ( x ) . S u b stitu tio n fo r of a form u la A
co n ta in in g x free is perm issible in (d), b u t P ro viso (B) p re v e n ts it in (e).
T h e situ atio n s w h ich th e provisos of th e su b stitu tio n rule, or th e

stip u la tio n s on our m e ta m a th e m a tic a l letters, p rev en t a lw a y s in v o lv e a
v a ria b le o ccurring b o th a n o n y m o u sly an d e x p lic itly . A b la n k e t rule th a t
co u ld b e used, in p la ce of th e m ore d eta iled con dition s s ta te d from case
to case, is s im p ly th a t th e a n o n ym o u s v a ria b les b e d istin c t from th e
e x p lic it va riab les. T h is is of course a little m ore re strictiv e th a n is n ec
essary, e.g. an o n ym o u s b o u n d x 's in th e A of *9 2 are cle a rly innocuous.
R emark I f P, E are predicate letter form ulas in the predicate letters

1.
( 1 ), and T (- E , then there is a deduction of E from T in (each form ula of)
which no predicate letters other than ( 1) occur. F o r in th e g iv e n d ed u ctio n
w e ca n s u b s titu te (2 ') for (T ), w here (2 ) is th e sam e as ( 1 ), an d th e a d
d itio n al n am e form s in (2 ') co n ta in o n ly ( 1 ), e .g . each of th em ca n b e
V x x . . . V x „ iP 1 (x1, . . . , x Wj).
Converse of substitution . A fo rm u la A ( a x, . . . , a w) w ill b e ca lled a

prim e nam e form (in th e d istin ct va ria b le s a 1# . . . , a n), p ro v id e d (i) it h as
none of th e form s A D B , A & B , A V B , n A , V x A ( x ) or 3 x A ( x ) , w here
A an d B are form ulas, x is a va ria b le , an d A (x ) is a form ula, a n d (ii) it
con tain s e x a c tly th e va ria b le s av ..., a n. T w o prim e n am e form s
A ^ a j, . . . , a^) a nd A ^ a ^ ..., a nJ w ill b e said to b e distinct (as p rim e
nam e form s), if A ^ , ..., t„.) an d A ^ u ^ ..., u n^) are n o t th e sam e
form ula for a n y term s t lt . . . , t n<, u v . . . , u nj. F o r exam p le , a + a = b and
0 = a -b are d istin ct, b u t a + a = b an d a = b -c are n o t d istin ct.
T heorem" 16. Converse of substitution for predicate letters.
Under the sam e stipu lation s as in Theorem 15, without P rovisos (A) and (B),
but provided instead that (2) be distin ct prim e nam e fo rm s : I f D * , . . . , D *
b E * , then D v . . . , D* h E .
T h is can b e p ro v e d u sin g th e sam e id eas as T h eo rem 4 (§ 25), an d lik e

w ise a d m its a second version.
N ame form replacement. U sin g th e n o tio n of n am e form oc

currence in d ic a te d a b o v e (follow ing E x a m p le 5), th e rep lacem e n t th e o r y
(Theorem 14 w ith * 66 , cf. § 33 fo llo w in g C o ro llary 2) can b e fo rm u la te d
th u s: A (x lt . . . , x n) ~ B ( X l, . . . , x n) bXl*"XM C a ^ .....tn) ~ Cb^ ,
p ro v id e d th e p a rts A ( t lf . . . , t w) an d B ( t 1, . . . , t n) c o n stitu te occurrences
of A ( x x, . . . , x n) an d B (x v . . . , x n), re sp e ctiv e ly , as n am e form s in
x x, . . . , x n.
R emark 2. R e su lts sim ilar to T h eo rem s 15 a n d 16 a n d R e m a r k 1

h old for in d iv id u a l va riab les. W e shall sta te o n ly th e follow ing, (a) I f
zv ..., z q are distinct variables not occurring bound in T>(zv . . . , z q) and
E ( z j, . . . , z q), and D (zlt . . . , z q) b E ( z lf . . . , z q) w ith z v . . . , z q held
constant, then there is a deduction of E ( z x, . . . , z a) from D ( z x, . . . , z a)
in which z v . . . , z q do not occur bound. P roof, s ta te d for q = 1 an d w ith
ju s t one va ria b le y varied . B y L e m m a 8 a, V -in tro d . (on z), an d ch a n g e
to a new variable z, Vz[VyD(z) D E(z)] is provable. This formula does not
contain z at all. For the definition of what constitutes a proof in the predicate
calculus of a given formula, all variables not in the formula are on a par.
Therefore there is a proof of Vz[VyD(z) D E(z)] not containing z. From this
formula and D(z) we can deduce E(z) by V-elim. (on z), V-introd. (on y)
and D-elim. (b) Let zlt .. ., zq be distinct variables not occurring bound in
D(zx, ..., zQ) and E(zx, . . ., z Q) ; and let tx, .. ., t Qbe distinct prime terms
(i.e. individual symbols or variables) none of which occurs in D(z1, ..., zQ)
or E(zx, . . ., zQ) unless it is one of zv . . ., zQ. Then D(zx, . .., zQ)
b E(zx, . . ., zQ) with zv . . ., zq held constant, if and only if D(tx, . . t Q)
b E(tx, ... , t Q) with (ithe variables among) tlf . . ., t q held constant. For by
(a), zx, .. ., z5, tlf . . t q can be eliminated as bound variables from the
given deduction, after which every inference will remain valid on sub
stituting throughout tv . . ., t Qfor zx, . . ., zQ, or vice versa.
§ 35. E q u i v a l e n c e s , d u a l i t y , p r e n e x f o r m . T h eo r e m 17. If x and y
are distinct variables, A, B, A(x), B(x) and A(x, y) are formulas, A and B
do not contain x free, and for *79 and *80 if x is free for y in A(x, y), then:
*75. b VxA ~ A. *76. h 3xA ~ A.
*77. b VxVyA(x,y) ~ VyVxA(x,y). *78. b 3x3yA(x, y) ~ 3y3xA(x, y)
*79. b VxVyA(x, y) D VxA(x, x). *80. b 3xA(x, x) 3 3x3yA(x, y).
*81. b VxA(x) 3 3xA(x).
*82. f- 3xVyA(x, y) 3 Vy3xA(x, y).
(Alterations of quantifiers.)
*83°. b 3xA(x) ~ - i V x - i A ( x ) . *84°. b VxA(x) ~ -i3x-iA(x).
(Each of 3 and V in terms of the other and - i.)
*85°. b “iVxA(x) ~ 3x-iA(x). *86. b ~i3xA(x) ~ Vx-iA(x).
*87. b VxA(x) & VxB(x) ~ *88. b 3xA(x) V 3xB(x) ~
Vx(A(x) & B(x)). 3x(A(x) V B(x)).
*89. b A & VxB(x) ^ *90. b A V 3xB(x) ~
Vx(A & B(x)). 3x(A V B(x)).
*91. b A & 3xB(x) ~ *92°. b A V VxB(x) ~
3x(A & B(x)). Vx(A V B(x)).
*93. b 3x(A(x) & B(x)) D *94. b VxA(x) VVxB(x) 3
3xA(x) & 3xB(x). Vx(A(x) V B(x)).
(Transfer of -i, & and V across quantifiers).
§35 EQUIVALENCES, DUALITY, PRENEX FORM 163
*83a. b 3 x A (x ) D - i V x - i A ( x ) . *84a. b V x A ( x ) D -n i3 x -iA (x ).
*85a. b 3x n A (x ) D -iV x A (x ). *92a. b A V V x B ( x ) D V x (A V B (x )).
*95. b V x (A D B (x)) ~ A D V x B (x ).
*96. b V x (A (x ) D B) ~ 3 x A (x ) D B .
*9 7 °. b 3 x ( A D B (x)) ~ A D 3 x B (x ).
*98°. b 3 x (A (x ) D B) ~ V x A (x ) D B.
*99°. b 3x(A(x) D B (x)) ^ V x A ( x ) D 3 x B (x ).
*97a. b 3x(A D B(x)) D (A D 3 x B (x )).

*98a. b 3 x (A (x ) D B) D (V x A (x ) D B).
*99a. b 3 x (A (x ) D B (x)) D (V x A (x ) D 3 x B (x )).

(Transfer of q uantifiers across D, w ith com parison of classical an d
in tu itio n istic results.)
P roofs, for th e classical sy ste m , of * 7 5 — *94, e x c e p tin g *7 6 , *78 ,

*80, * 88 , *90, *92, *94. W o rk is sa v ed b y p o stpo n in g these seven u n til w e
h a v e d u a lity (or for *80 an d *94, th e d u al-co n verse relationship). T h en ,
classically, *9 5 — *99 w ill follow b y usin g *5 9 (§27) w ith * 88 , *90, *9 2 an d
*85, * 86 .
*75 I f w e redesign ate as “ A ( x ) ” , th en since A does not co n ta in
x free, A (t) is also A (x ) (§ 18). T h e result follow s, u sin g V -elim . an d
D -in tr o d . (or A x io m S ch em a 10), an d V -in tro d . an d D -in tro d . [x is
not va rie d in the V -in tro d ., since A (x ) does not co n ta in it free], an d *1 6 .
*79 . 1. V x V y A ( x , y) b A ( x , x) bx V x A ( x , x) — d ou ble V -elim .

(*65) [x, x is a pair of term s free for x , y in A (x , y )], V -in tro d .
2. b V x V y A ( x , y) D V x A ( x , x) — D -in tro d ., 1 [x is n o t va ried
in 1 , since V x V y A ( x , y) does n o t co n ta in x free].
*82. 1. A (x , y) b 3 x A (x , y) by V y 3 x A ( x , y ) — 3-in trod ., V -in tro d .
2. V y A ( x , y) b V y 3 x A ( x , y) — V -elim ., 1.
3. 3 x V y A ( x , y) b V y 3 x A ( x , y) — 3-elim ., 2 [ V y 3 x A (x , y) does not
co n ta in x free, an d no va ria b le is va ried in 2 , since y (cf. 1 )
does not occur free in V v A ( x , y)].
4. b 3 x V y A ( x , y ) D V y 3 x A ( x , y) — D-introd., 3 [no v a ria b le
is v a ried in 3 (cf. L e m m a 7 b § 24)].
N o te h ow th e a tte m p t to g iv e a corresponding d em o n stratio n of th e c o n

verse of *82 (i.e. o f *82 w ith th e d irection of D reversed) is d efeated b y
th e restriction on the use of su b sid iary d ed u ctio n in th e p red icate c a lc u lu s :
1. A (x , y) by V y A ( x , y) b 3 x V yA (x , y) — V -in tro d ., 3-introd.
2? 3xA(x, y) [- 3xVyA(x, y) — 3-elim., 1. But this is illegitimate,

since the rule of 3-elimination (in contrast to the V-elimination rule used
at Step 2 for *82) is a subsidiary deduction rule, and is inapplicable here
because in the subsidiary deduction 1 the variable y is varied for the
assumption formula A(x, y) to be discharged (except if A(x, y) does not
contain y free).
There is no way around this difficulty, and the converse of *82 should
not be provable for arbitrary A(x, y). In terms of the interpretation, the
formula 3xVyA(x, y) says that there is one x such that for every y,
A(x, y); and Vy3xA(x, y) says merely that for every y there is some x,
not necessarily the same x for different y's, such that A(x, y). The dis
tinction is familiar to mathematicians from the example of uniform con
vergence vs. ordinary convergence of a sequence of functions a n(x) to a
limit function a(x) on an interval or other range X of x. Using the present
logical symbolism, and variables p, n and N ranging over natural numbers,
and x over X , the properties of uniform convergence and ordinary con
vergence are expressed respectively by
(i) Vp3N\fxVn(n > N D | a n(x) — a(x) | < 1/2P),
(ii) VpVxlN\fn(n > iV D | a n(x) — a(x) | < \/2 p).
Then *82 says that uniform convergence implies ordinary convergence;
but the converse is not generally true. (Similarly for uniform and ordinary
continuity.)
A metamathematical demonstration that Vbla<^l(a, b) D la yib^X (ai b)
is unprovable in the predicate calculus will be given in Example 2 § 36.
*83. 1. A(x), Vx-i A(x) h “iA(x)— V-elim.
2. A(x), Vx-iA(x) (- A(x).
3. A(x) |- -i Vx-iA(x) — -i-introd., 1,2 [no variable of Vx-iA(x)
is varied in 1 or 2].
4. 3xA(x) b n VxnA(x) — 3-elim., 3 [nVxnA(x) does not con
tain x free, and no variable is varied in 3].
5. b 3xA(x) D nV x -i A(x) — D-introd., 4 [no variable is varied
in 4].
6. i3xA(x), A(x) b 3xA(x) — 3-introd.
7. -i3xA(x), A(x) b “i3xA(x).
8. -i3xA(x) b "~iA(x) bx Vx-iA(x) — “i-in trod., 6, 7 [no var
iables are varied in 6, 7], V-introd.
9. b -i3xA(x) D Vx-iA(x) — D-introd., 8 [x does not occur free
in the assumption formula -i3xA(x) of 8 wThich is being dis
charged] .
10. b - i V x m A ( x ) D 3 x A (x ) — co n trap o sitio n (*14 ), 9.
11. b 3 x A (x ) ~ - i V x - i A ( x ) — & -in tro d . (*16 ), 5, 10 .
*84. See E x a m p le 1 § 34.
*8 7. 1 . A (x ), B (x ) b A(x) & B (x ) bx Vx(A(x) & B(x)) — & -in tro d .,

V -in tro d .
2. V x A (x ), V x B (x ) b V x (A (x ) & B (x)) — V -elim . tw ice, 1.
3. V x A (x ) & V x B ( x ) b Vx(A(x) & B (x)) — & -e lim ., 2.
4. b V x A ( x ) & V x B (x ) D V x (A (x ) & B (x)) — D-introd., 3 [no v a r
iable is varied in 3, since x (cf. 1) does not occur free in
V x A (x ) & V x B (x )].
5. A (x ) & B (x) b A(x) bx V x A (x ) — & -e lim ., V -in tro d .
6 . A (x ) & B (x ) b B(x) b x V x B (x ) — & -e lim ., V -in tro d .
7. A (x ) & B (x) bx V x A (x ) & V x B (x ) — & -in tro d ., 5, 6 .
8 . V x (A (x ) & B (x)) b V x A (x ) & V x B (x ) — V -elim ., 7.
9. b V x (A (x ) & B (x )) D V x A (x ) & V x B (x ) — D-introd., 8 [no var
iable is varied in 8 , since x (cf. 7) does not occur free in
V x (A (x ) & B (x ))].
10 . b V x A (x ) & V x B (x ) ~ Vx(A(x> & B (x)) — & -in tro d ., 4, 9.
*89, * 9 f , *93. I f w e read “ A ” for " A ( x ) ” an d “ V x A ( x ) ” in th e pre

cedin g, o m ittin g one V -elim in atio n a t S te p 2 an d th e V -in tro d u c tio n a t
S te p 5, it reads as a d em o n stratio n of *89. T h e n s u b s titu tin g 3 for V
th ro u gh o u t, w e g e t a d em o n stratio n of *9 1 . T h e reader m a y w rite th is o u t,
an d v e r ify th a t th e con d ition s for th e 3 -elim in a tio n s are satisfied. B u t
su b s titu tio n of 3 for V in th e proof of *8 7 does n o t w ork. W h y ? W e th u s
o b ta in o n ly *9 3 b u t n o t th e converse.
A s a fu rth er exercise, th e reader m a y a tte m p t to g iv e a correspon ding
proof of *92, an d see h o w th is is d e fe a te d b y th e restriction on su b sid ia ry
ded u ction s. T h e result *92, w h ich w e sh all infer from *9 1 a fte r w e h a v e
d u a lity , is in terestin g as an ex a m p le of a form u la w h ich does n o t co n ta in
- i b u t w h ich w e do n o t succeed in p ro v in g w ith o u t P o s tu la te 8 .
In v ie w of * 49 b § 2 7 , there can be a t m ost 18 ( = 3*2*3) in tu itio n istic a lly
n o n -eq u iva len t form ulas form ed from A (x ) b y q u a n tify in g x an d p o ssib ly
a p p ly in g negation. (W e use either 0 ,1 or 2 - V s first, th en either V x or 3 x ,
th en aga in eith er 0, 1 or 2 -V s .)
Corollary. Each of the four tables I — I V com prises form ulas equiv
alent to one another in the classical predicate calculus . F or each table , in
the intuition istic system : Each two form ulas not separated by a line are equiv
alent. Each form ula im plies any form ula below it, i.e. the im plication from
the one to the other is provable . The double negation of the im plication
from each form ula to any form ula not separated from it by a double line is
provable {and hence , u sin g *4 9 a an d *2 5 , of the equivalence). (H e y tin g 19 4 6 .)
I II
a. V x A (x ) a. 3 x A (x )
b. - i - i V x A (x ) b. 3 x - i - i A (x)
Cl- Vx n n A (x ) Cl- -i-i3 x A (x )

c2. -i-i Vx n n A (x ) c2. - i - i 3 x —i—i A (x)
c3. - i 3 x - i A (x ) c 3. -i V x -iA (x )
III IV
a. 3 x - i A (x) a i- V x -iA (x )
bi- - l —i 3 x —i A (x ) a 2. - i —i V x —i A (x)
b 2. —i V x —i—i A (x) a3. -i 3x -i-iA (x )
c. - i V x A (x ) a 4. —i3 x A (x )
P roofs for T a b le I I , in th e in tu itio n istic syste m , b H a 3 l l b [*49a, *70 ].

b lib 3 IIc3 [*8 5 a ]. b IIc3 ^ lie ! [* 86 ]. b H c3 ^ —1 V x n n n A ( x )
[*49b] — I I c 2 [* 86 ]. b -i-i(I I c 2 3 l l b ) [*5 1b ]. L ikew ise, b -v n (H e p
H a ). H e n ce b “ i—1( l i b 3 H a ), b y *2 4 w ith b H b 3 I I c x an d *49a.
T heorem 18°. Let D be a predicate letter form ula constructed from the
distinct predicate letters P 1 (a1, . . a n ), . . . , P rw(a1, . . a Um) and their
negations - i P 1 (a1, . . . , ani), . . . , - i P m(a1, . . . , a»m) using only the opera
tors Sc, V, V x and 3 x (for any variable x). Then a form ula D t equivalent to
the negation n D o / D is obtained by the interchange throughout D of Sc with
V, of V with 3, and of each letter with its negation.
In other words, if D be such a predicate letter form ula, and D t be the
result of the described interchange perform ed on D : b “i D ^ Dt .
E xample 1 °. b -1 la {\/b-y<3l{b) & (-iSV3c C(a, c, b))) ~
{3b <3(ift) V ( B 8 c V c ~ iC {a ,c ,b ))).
P ro o f is b y th e sam e m e th o d as T h eorem 8 (§ 27), usin g *8 5 an d *86
to h an d le th e tw o new cases w h ich n ow arise under th e in d u ctio n step.
T h e theorem as before a d m its a second version.
Corollary 0. A n equivalence between two letter form ulas and F of the E

type described in the theorem is preserved under the interchange throughout
E and F of Sc w ith V and of V with 3.
In other words, if E and F be two such predicate letter formulas, and E '
and F ' be the results of the described interchange performed on E and F ,
respectively: If h E ~ F , then b E ' ~ F '. (P rinciple of d u a lity .) Also, if
b E D F , then b F ' D E '. (D u al-co n verse relationship.)
T h e co ro llary follow s from th e theorem as before.
E xam ple 2°. By *9 1, b c3 & 3aB(a) ~ la(G I & B(a)). T h e n c e b y

d u a lity , b V VaB(a) ~ V a(G l V B(a)). T h e n c e *92, as in E x a m p le 7 § 34.
S im ilarly, w e o b ta in *7 6 , *78 , * 88 , *90 as d uals of *7 5 , * 7 7 , *8 7, *89
an d *80, *9 4 as d u al-co n verses of *7 9 , *93, re sp e ctiv e ly . (N o te th a t
each of *81 an d *8 2 is self-dual-con verse.)
T heorem Given any formida C , there can be found a formula D

19°.
(called a prenex form of C) with the following two properties. The formula
C is equivalent to D , i.e. b C ~ D . In D , all the quantifiers (if any) stand
at the front, i.e. all the other logical symbols 3, &, V, -i (if any) stand
within the scope of every quantifier (such a form ula w e s a y is prenex).
P roof . T o reduce a form u la C to p ren ex form , w e can, step b y step,
m o ve all th e q u an tifiers o u tsid e th e scopes of th e lo gica l sy m b o ls 3, &,
V ,- i b y ap p licatio n s of *85, * 86 , *89— *92, *9 5 — *98, n o tin g th e follow in g.
In case th e form u la for a p p lica tio n of *89— *92, *9 5 — *98 fails to m eet th e
co n d itio n t h a t th e A or th e B n o t co n ta in th e x free, a ch an ge of b o u n d
va ria b le s ca n b e m ade b y * 7 3 or *74 . In case th e A for *8 9— *9 2 stan d s on
th e w ro n g side of th e & or V, *33 or *3 4 ca n b e used. (This procedure for
re d u ctio n to p ren ex form does n o t require *8 7, *88 or *9 9 ; b u t a t a n y
p o in t w here one of th em ca n b e ap p lied, th e use of it w ill sa v e step s an d
lead to a shorter p ren ex form.)
T o p ro v e th a t th e procedure term in ates, w e ca n use as in d u ctio n n u m b er
th e n u m ber of in stan ces o f a q u a n tifie r sta n d in g inside th e scope o f an
3, &, V or - i , i.e. th e to ta l n u m ber of pairs, one m em ber of w h ich is an
D , & , V or n , an d th e o th er m em ber is a q u a n tifie r sta n d in g inside
th e scope of th a t m em ber. I f th is n um ber is n o t 0, there m u st b e som e
in s ta n ce w here there is no lo gical sy m b o l of in term ed ia te scope. A step
is th en carried o u t w h ich rem o ves th is in stan ce, le a v in g th e others u n
ch an ged , so t h a t th e in d u ctio n n u m ber is redu ced b y one. (If *8 7, *88 or
*99 is used, th e in d u ctio n n u m b er is re d u c e d b y tw o or more.)
E x a m p l e 3°. T h e n u m b ered form ulas w h ich follow are su cce ssiv e ly

e q u iv a le n t, each to th e n e x t, b y replacem en ts b a sed on th e eq u ivalen ces
cited a t th e righ t. T h e la st fo rm u la is a p ren ex form of th e first, or in d eed
of a n y form u la in th e list.
1. [-1 3a &(a) V VaB(a)] & [C D V a© (a)].

2. \Va - i ^ (a ) V VaB{a)] & V a [ C D T>(a)] — * 86 , *95.
3 . V a [V * - n ^ { a )V S(a)} & V a [C D ©(<*)] — *92.
4. V4 V H ^ ) V % ) ] & V a [ C D © ( j ) ] — *73 .
5. V a V * [ - n 3 ( A ) V B(tf)] & V ^ [C D ©(a)] — *34, *92.
6. Va{Vb[-i£l(b)VB(a)]&[C3T>(a)]} — *8 7.
7. V ^ { h ^ ) V S W ] & [ C D ® ^ ] } — *33, *89.
§ 36. V a l u a t i o n , c o n s i s t e n c y . T h e p red icate calcu lu s is in ten d ed

to b e a fo rm alizatio n of principles of p red icate lo gic w h ich h old good
in d e p e n d e n tly of th e n u m b er of elem ents in th e o b je c t d om ain, p ro v id e d
th ere is a t least one elem ent ( §31) . H en ce th e p ro v a b le form ulas sh o u ld
a ll b e true, if w e s p e c ify th e n um ber to b e k , w here k is a n y in teger ^ 1.
W e m a y com bine th is id ea w ith th e one used in § 28, w here w e a b stra c te d
from tru e an d false propositions to o b ta in tw o arith m e tic o b je cts t an d f.
T h is su ggests a fin ita r y v a lu a tio n procedure, w h ich w ill en able us to
e sta b lish m e ta m a th e m a tic a lly th e c o n sisten cy of th e p red icate calculus.
In th e v a lu a tio n procedure, a p red icate le tte r form u la is considered
as represen tin g a fu n ctio n of th e free in d iv id u a l va ria b le s an d of th e
p re d ic a te le tte rs (including proposition letters) w h ich it con tain s, or pos
s ib ly of th ese an d oth er free v a ria b les an d p red icate letters as w ell. T h e n
w h en w e define th e n o tio n w h ich w ill correspond to th a t of id e n tica l
tr u th for th e p ro p o sitio n al calculus, w e sh all require th a t th e form ulas
p r o v a b le in th e ca lcu lu s sh ould b e true w h a te v e r p red icate each p red icate
le tte r represents, an d also, in v ie w of th e g e n e ra lity in terp retatio n of th e
free in d iv id u a l v a ria b les (§ 32), w h a te v e r o b je c t from th e o b je c t d om ain
ea c h free in d iv id u a l va ria b le represents.
A fte r ch oosin g a fix e d p o sitiv e in teger k for th e n um ber of d istin ct
o b je c ts in th e o b je c t dom ain, it does not m a tte r for th e v a lu a tio n proce
dure w h a t th e o b je c ts th em selves are. I t is co n ven ien t to ta k e th em to b e
(or to ca ll them ) th e num bers 1 , . . . , k. T h ese num bers are to be th e v a lu es
w h ic h th e in d iv id u a l v a ria b les take.
A s before (§ 28), th e proposition letters (i.e. letters for p red icates of 0
va riab les) ta k e th e v a lu e s t, f.
N o w consider th e p re d ic a te letters. F o r a n y g iv e n in teger n > 0, a
p re d ic a te le tte r w ith n a tta c h e d v a ria b le s differs under th e lo gical in ter
p re ta tio n df th e sy ste m from a proposition le tte r in th a t it represents n o t
a p rop osition b u t ra th er a p rop osition al fu n ctio n of n va riab les, i.e. a
fu n c tio n w h ich ta k e s a proposition as v a lu e for each set of va lu es o f th e
a tta c h e d va riab les (§3 1). T h e va lu es w h ich w e g iv e a p red icate letter w ith n
a tta c h e d v a riab les in our v a lu a tio n procedure, w h en th e propositions
are replaced b y t an d f, w ill ac co rd in g ly b e n o t t, f, b u t rath er fu n ctio n s
of n v a riab les each o ver th e d om ain { 1 , . . . , & } ta k in g v a lu e s in th e dom ain
{t, f}. T h ere are e x a c tly 2kn d iffe re n t such function s. W e ca ll th em th e
logical functions of n v a riab les over th e dom ain of k o b jects. T h e tru th
v a lu e s t, f, w h ich w e use for th e case n— 0 , can be con sid ered as b e in g
th e 2 ( = 2 fc0) lo g ica l fu n ctio n s of 0 variables.
A s before, w e in terpret 3 , &, V, -i as fix e d fu n ction s o ve r th e d om ain
of tw o o b je c ts {t, f} ta k in g va lu e s in th e sam e dom ain, d efin ed b y th e
ta b le s g iv e n p re v io u sly (§ 28). W e n ow in terpret V an d 3 as fix e d fu n ctio n s
o ve r th e lo g ic a l fu n ctio n s of one v a ria b le ta k in g v a lu e s from th e d o m ain
{t, f}, w here th e v a ria b le x in V x or 3 x in d icate s of w h a t in d iv id u a l
va ria b le th e operand sh all be con sidered as lo gical fu n ction . T h ese tw o
fix e d lo g ica l fu n ctio n s are d efined as follow s. F o r a g iv e n lo gical fu n ctio n
A (x ), th e v a lu e of V x A ( x ) is t, if A (x ) has th e v a lu e t for e v e r y v a lu e of x
in th e dom ain { 1 , . . otherw ise th e v a lu e is f. T h e v a lu e of 3 x A (x )
is t, if A (x ) has th e v a lu e t for som e v a lu e of x in th e d o m ain { 1 , . . . , k};
an d otherw ise th e v a lu e is f.
G iv e n a p re d ic a te le tte r form ula, w e are n o w in a p osition to co m p u te
a ta b le exp ressin g th e v a lu e s of th e fu n ctio n , of th e d istin ct free in
d iv id u a l v a ria b les an d p red icate letters occurring in it (or of th ese an d
oth er va ria b le s an d letters), w h ich th e form u la represents. F o r th e t a b
u latio n , it w ill b e co n ven ien t first to list in som e fix e d order th e lo gica l
fu n ctio n s w h ich w ill b e required as va lu es of th e p red icate letters, an d to
in tro d u ce sy m b o ls to sta n d for them .
E xam ple 1. For k = 2, let us co n stru ct th e ta b le for th e p red icate

le tte r form u la Va{B 3 C3(a)) V (-i £l(b) B & ), for e x a c tly th e free v a r i
ables an d p red icate letters co n ta in ed in it. T h is form u la th e n represents
a fu n ctio n of th ree va riab les, n a m e ly b, B an d <^(a), w here a is th e a tta c h e d
v a ria b le of th e nam e form used for as a p red icate le tte r w ith one
v a ria b le (cf. § 3 1). B efo re c o n stru ctin g th e ta b le for th e form ula, w e sh all
in tro d u ce n o ta tio n s for th e lo g ica l fu n ction s of one va ria b le, w h ic h w e
sh all b e e m p lo y in g as va lu e s o f th e v a ria b le (a). S in ce k = 2, th ere
are 4 ( = 2%1) of th em , 1 ±(a)f t2(a), \z{a), 14(a), defin ed b y th e fo llo w in g
ta b le of th eir valu es.
170 THE P R E D IC A T E CALCULUS CH . V II
L ogical functions of one variable in a domain of two objects

V a lu e o f th e in d e C orrespond ing v a lu e of th e re sp e c tiv e
p en d en t v a ria b le fu n ctio n
a liM U4 U(a)
i t t f f
2 t f t f
T h e ta b le for th e g iv e n form u la is n o w as show n (the c o m p u ta tio n o f a

sa m p le e n tr y w ill follow ).
V a lu e of th e re sp ective C orresponding v a lu e of
in d ep en d en t v a ria b le th e fu n ctio n
b s c 1(a) V a (B 3 3 {a )) V ( -i< 3 ( £ ) & B )
1 t h i* ) t
1 t h(a) f
1 t W ) t
1 t U {*) t
1 f k(a) t
1 f t
1 f m t
1 f I4(a) t
2 t h (a) t
2 t k (* ) t
2 t (a)
I3 f
2 t U(a) t
2 f li(a ) t
2 f m t
2 f m t
2 f Ua) t
W e n o w g iv e th e c o m p u ta tio n for th e second v a lu e (line 2 o f th e tab le).

F o r th is purpose, w h en w e co m e to th e V -o p era tio n , w e sh all need th e
ta b le for S 3 ^ f ( j) w ith B h a v in g th e v a lu e t an d <£7 (a) th e v a lu e l2(<z).
W e g iv e th is ta b le first (its c o m p u ta tio n to follow).
V a lu e of th e in d e C orresponding va lu e
p en d en t va ria b le of th e fu n ctio n
a t 31 2(a)
i t
2 f
T h e co m p u ta tio n s of th e tw o v a lu es in th is su b sid iary ta b le, u sin g th e

ta b le for 12(<j) g iv e n at th e o u tset an d th e ta b le for 3 from § 28, are th ese:
t ^ I,(a) t 3 1 2(a)
t =3 1.(1) t3lg(2)
t 3 t t 3 f
t f
W e n ow return to th e original q u estio n of co m p u tin g th e e n try for th e

second line of th e ta b le for Va(B 3 <d(a)) V (-1 £%(b) & B), i.e. th e v a lu e
of th is form ula w hen b h as th e v a lu e 1, B th e v a lu e t an d (a) th e v a lu e
l2(<z). W e sta rt o u t thus.
Va(B 3 a ( a ) ) V (-ic 2{b) & B)

V a ( t D 12(a)) V ( - i I a(l) & t)
S in ce th e v a lu e colum n of th e ta b le for t 3 I2(#) does not h a v e a ll i ’s,

w e can replace V<z(t 3 I2(j)) b y f und er th e in terp reta tio n g iv e n for V ,
an d also w e can replace I2(l) b y its v a lu e t from th e ta b le for l2(a), thus.
f V ( - it & t)
f v (f & t)
fv f
f
T h e reader m a y v e r ify t h a t he u n d erstan d s th e procedure b y c a lc u la tin g

oth er entries in our ta b le for Va(B 3 V (~ic ^(b) & B ). O f course
there are u su a lly short cu ts th a t can b e used in ascertain in g th e ta b le
for a form ula, so t h a t it is n o t n ecessary to go th ro u gh th e w h ole c a lc u la tio n
procedure se p a ra te ly for e v e r y e n tr y of th e ta b le (e.g. here w e ca n re c
ogn ize a t once th a t all eig h t entries for w h ich B is f are t). B u t th is c a l
cu la tio n illu stra tes th e u n d e rly in g d efin itio n of th e fu n ctio n represen ted
b y a p red icate le tte r fo rm u la for a g iv e n p o sitiv e in teger k.
If, for a g iv e n k, th e v a lu e colu m n of th e ta b le for a p red icate le tte r
fo rm u la, as a fu n ctio n of certa in free v a riab les an d p red ica te le tte rs
in c lu d in g all co n ta in ed in it, co n ta in s o n ly t ’s, w e s a y t h a t th e fo rm u la
is k-identical , or valid in a dom ain of k objects. I f it co n ta in s som e t's ,

th e form u la is said to b e satisfiable in a dom ain of k objects.
A s before (§ 28), if it is ^ -id en tical (or satisfiab le in a d om ain of k
ob jects) for th e m in im al list of free va riab les an d p red icate letters, th e n
it is for a n y oth er l i s t ; an d co n versely. H en ce aga in th e reference to th e
list ca n b e o m itted .
I f th e v a lu e colum ns of th e ta b le s for tw o p red icate le tte r form u las
con sidered as fu n ctio n s of ju s t those free v a riab les an d p red icate le tte rs
co n ta in ed in either (or of a n y others as well) are th e sam e, th e tw o
form ulas are said to be k-equal.
T heorem 20. F or each fixed integer k (k ;> 1): A necessary condition
that a predicate letter form ula E be provable {or deducible from k-identical
form ulas T) in the predicate calculus is that E be k-identical.
P roof. The th eorem follow s from tw o lem m as, corresponding to
th o se for T h eo rem 9, b u t n o w referring to th e p o stu la te list of th e pred
ica te calcu lu s, p red ic a te le tte r form ulas an d ^ -id e n tity . T h e reason in g
a lre a d y g iv e n for th e p o stu la tes of G ro u p A 1 carries o ve r in its essen tials
to th e present situ atio n , an d for th e four p o stu la tes of G ro u p A 2 w e sh o w
th e tre a tm e n t of tw o.
A xiom S chema 10. An a x io m by th is sch em a is V x A ( x ) D A ( t) ,

w here th e term t for th e pure p red icate calcu lu s is sim p ly a v a ria b le ,
an d th is v a ria b le t is free for x in A (x ).
T h e v a ria b le t m a y b e th e sam e or d istin ct from x, an d x m a y or m a y
n o t occur free in A (x ).
W e m u st sh ow t h a t V x A ( x ) D A (t) ta k e s th e v a lu e t, for e v e r y as
sign m en t o f lo gica l fu n ctio n s as v a lu e s of th e p red icate letters, an d o f
th e o b je c ts 1, as v a lu e s of th e free va riab les, co n ta in ed in
V x A (x ) D A (t).
Consider a p a rticu la r such assignm ent. I f th e assign ed va lu e s are g iv e n
to th e p red icate letters a n d free v a ria b les o f A (x ), e x c e p t (if x occurs free
in A (x )) to th e v a ria b le x , th e n A (x ) represents a lo g ica l fu n ctio n o f th e
v a ria b le x (w hether or n o t x occurs free in A (x )).
Sin ce t is free for x in A (x ), in th e sequence of sy m b o ls A (t) th e free
occurrences o f t o c c u p y th e p osition s w h ich are o ccu p ied in A (x ) b y free
occurrences, of either t or x. T h erefo re th e v a lu e of th e lo gical fu n ctio n
represen ted b y A (x ), w hen x ta k e s as v a lu e th e assign ed v a lu e of t, is th e
v a lu e o f A (t) for th e p articu lar assignm ent under consideration.
N o w th ere are tw o cases.
C a se 1 : th e lo g ica l fu n ctio n represented b y A (x ) has o n ly t's as valu es.
T h e n th e v a lu e of A ( th w h ich is one of these va lu es, is t. H en ce, b y th e
v a lu a tio n ta b le for D (§ 28), V x A ( x ) D A (t) h as th e v a lu e t.
C a se 2: th e lo gical fu n ctio n represented b y A (x ) has som e f ’s in its

v a lu e colum n. T h e n , b y th e d efin ition of th e v a lu a tio n process for V ,
th e form ula V x A ( x ) h as th e v a lu e f; an d again V x A ( x ) D A (t) has th e
v a lu e t.
R ule 12. T h e prem ise is A (x ) D C, an d th e conclusion is 3 x A (x ) D C,

w here C does not co n tain x free.
W e h a v e as h yp o th esis t h a t A (x ) D C ta k e s th e v a lu e t for e v e r y
assign m en t of lo gical fu n ctio n s as va lu e s of th e p red ica te letters, an d of
th e o b je cts 1 , . . . , k as va lu e s of th e free variab les, co n ta in ed in A (x ) D C.
W e m u st show th e like for 3 x A (x ) D C.
Consider a p articu lar assign m en t for 3 x A (x ) D C, for ju s t th e p red icate
letters an d free va riab les w h ich it con tains. Sin ce 3 x A (x ) D C does n o t
co n ta in x free, th is does n o t in clu d e an assign m en t for x.
N o w if th e assigned va lu es are g iv e n to th e p red icate letters an d free
va riab les o f\A (x), e x c e p t (if x occurs free in A (x)) to x , th en A (x ) repre
sents a lo gical fu n ctio n of th e v a ria b le x.
C a se 1 : th e lo gical fu n ctio n represented b y A (x) h as som e t's in its

v a lu e colum n. Choose a v a lu e of x corresponding to one of these t's ; an d
consider th e g iv e n assignm ent for 3 x A (x ) D C, to g e th e r w ith th is v a lu e
of x , as an assign m en t for A (x ) D C. T h e n A (x) has th e v a lu e t; by
h yp o th esis, A (x ) D C has th e v a lu e t; an d hence b y th e v a lu a tio n ta b le
for D , C has th e v a lu e t. Sin ce C does not co n ta in x free, th is v a lu e of
C is on th e basis of th e g iv e n assignm ent for 3 x A (x ) D C, w ith o u t regard to
th e v a lu e of x. H en ce, b y th e v a lu a tio n ta b le for D , 3 x A (x ) D C ta k e s
th e v a lu e t for th e g iv e n assignm ent.
C a se 2 : th e lo gical fu n ctio n represented b y A (x ) has o n ly f ’s as valu es.

T h e n 3 x A (x ) has th e v a lu e f ; an d hence 3 x A (x ) D C has th e v a lu e t.
E x a m pl e 1 (concluded). V ^ (S D (a)) V (-i<3(/>) &B) is not k-

id en tical, since its v a lu e colu m n con tain s f s, an d hence it is u n p ro va b le
in th e p red icate calculus.
E x a m pl e 2 . F o r k = 2, w hen d^(a, b) ta k e s as v a lu e th e lo gical

fu n ctio n 1 (a, b) such th a t 1( 1 , 1 ) = 1(2 , 2 ) = t, 1( 1 , 2 ) = 1(2 , 1 ) = f, th e
form u la yi b la Z i{ a ) b) D 3aVb<^l(a, b) ta k es th e v a lu e f, so it is un p ro vab le.
Corollary 1. For each integer k > 1 , a necessary condition that two

predicate letter form ulas E and F be equivalent is that they be k-equal , i.e.
if \- E ~ F , then E and F are k-equal.
E xample 3. F o r k — 2, w ith Iâ), f2(<z), l 3(tf), h ia) as in E x a m p le 1 ,
w e h a v e th e fo llo w in g ta b les for th e to p form ulas of I — I V C o ro llary
T h eo rem 17 w ritte n w ith a p a rticu la r v a ria b le an d p red icate letter.
H en ce no tw o of th e four are eq u ivalen t.
a 2(d) Va£t{a) 3 a ^ ( a ) 3 a -\£ l{a )
h i* ) t t f f
k (a ) f t t f
U(a) f t t f
k (a ) f f t t
Corollary 2. The predicate calculus is {sim ply) consistent , i.e. for no
form ula A, both (- A and 1- —iA .
P roof. I f A is a p red icate le tte r form ula, th en , for each k, A an d - i A

ca n n o t b o th b e ^ -iden tical. (H ere it w o u ld suffice to h a v e p ro ved th e
theorem for a n y one fix e d k. In th e origin al proof b y H ilb e rt an d A c k e r -
m an n 19 28 , k= 1.) F o r th e oth er sense of form ula, th e c o n sisten cy follow s
b y th e con verse of th e s u b stitu tio n rule (Theorem 16 § 3 4 ).
*§ 37. Set-theoretic predicate logic, ^-transforms. T h e lo g ic a l

fu n ctio n s used in th e v a lu a tio n procedure for a g iv e n fin ite k are fin ite
o b jects, in th e sense th a t each is represented b y a ta b le h a v in g a fin ite
n u m ber of entries.
T h e n o tio n of a lo gical fu n ctio n of n v a riab les can b e sta te d sim ila rly
for th e case t h a t th e o b je c t d o m ain h as an in fin ite n um ber of elem ents,
but w ith th e difference t h a t th e fu n ction s ca n n o t be d escribed by
fin ite tab les. W e can th en d efine th e notions of va lid ity , an d of sat
isfia b ility , in a g iv e n n o n -e m p ty o b je c t dom ain. C le a rly o n ly th e card in al
n um ber of th e o b je c t dom ain m a tters for th is purpose, an d not w h a t th e
elem ents are them selves.
T h e n b y reasoning as in th e proof of T h eo rem 20 , w e can show th a t
e v e r y p ro v a b le p red icate le tte r form u la is v a lid in e v e r y n o n -e m p ty
o b j e c t d om ain. In d em o n stratin g th is for th e case of an in fin ite d om ain,
th e reasonin g e m p lo y ed is no longer fin ita ry. A n o n -fin ita ry step appears,
for ex am p le , in th e tre a tm e n t of A x io m S ch em a 1 0, w here w e d istin gu ish
tw o cases acco rd in g as all of th e v a lu e s of a certain fu n ctio n are C s or
som e are f s , there b ein g n ow in fin ite ly m a n y va lu es under consideration.
§37 SET-THEORETIC PREDICATE LOGIC 175
T h is co n stitu te s an a p p lica tio n of th e la w of th e e xclu d ed m id d le for an
in fin ite set (§ 13). In fa c t th e n otion itself of v a lid it y , for th e case of
an in fin ite dom ain an d a form u la co n ta in in g a p red icate le tte r w ith n > 0
a tta c h e d va riab les, is n o t fin ita r y . F o r it requires th a t th e v a lu e of a
fu n ctio n b e t for all lo gical fu n ctio n s of n v a ria b le s as va lu e s of t h a t
p red icate l e t t e r ; an d th e class of th ose lo gical fu n ctio n s is n o n -en u m erab le,
an d so o n ly co n ce iv a b le (as w e u su a lly thin k) in term s of th e co m p le ted
infin ite.
T h e result th a t e v e r y p ro v a b le le tte r form u la of th e p red icate calcu lu s
is v a lid in e v e r y n o n -e m p ty o b je c t d om ain therefore does n o t b elo n g to
m e ta m a th e m a tic s (cf. § 15). I t belo ngs rath er to w h a t m a y b e c a lle d
set-theoretic predicate logic (H ilb e rt-B e rn a y s 1934 p. 125), w h ich h as in
com m on w ith m e ta m a th e m a tic s t h a t it m akes th e lo g ica l form alism an
o b je c t of s tu d y , b u t differs from it in n o t b e in g restricted in th e s tu d y to
fin ita r y m eth ods. W h ile our m ain business is m e ta m a th e m a tics, th e e x tr a -
m e ta m a th e m a tic a l co n cep tio n s an d results o f se t-th e o retic p red icate lo gic
m a y h a v e h eu ristic v a lu e , i.e. th e y m a y su ggest to us w h a t w e m a y hope
to d isco ver in th e m e ta m a th e m a tics.
Our success in p ro v in g th e co n siste n cy of th e p red icate ca lcu lu s
(Theorem 20 C o ro llary 2 ) stem s from th e fa c t th a t w e w ere ab le to fit
to th e form ulas of th e ca lcu lu s a fin ita r y in terp retatio n , n a m e ly th e
in te rp re ta tio n by v a lid it y in a fix ed fin ite d om ain (i.e. ^ -id e n tity ).
B u t th a t does n o t correspond to th e usual in terp reta tio n of th e calcu lu s,
to w h ich corresponds rath er v a lid it y in an a r b itra ry n o n -e m p ty
dom ain.
O n th e b asis o f th e se t-th e o re tic result th a t e v e r y p ro v a b le fo rm u la
is v a lid in e v e r y n o n -e m p ty d om ain , w e sh all n o w see h e u ristica lly th a t
th e t o t a lit y of th e n ecessary co n d itio n s g iv e n b y T h eo rem 20 , i.e. th e
p ro p e r ty of b e in g ^ -id en tical for e v e r y fin ite k , is in su fficien t for p ro v a
b ility .
T h is w ill b e a con sequence o f th e fa c t t h a t a fo rm al a x io m sy ste m
(cf. § 8 ) a b o u t a set D of elem en ts w h ich requires D to b e in fin ite can
b e expressed b y a p red icate le tte r form ula. T o c o n v e y th e id ea, w e sh all
first g iv e th e axio m s as three form ulas in our lo gica l sy m b o lism , b u t u sin g th e
p red ic a te sy m b o l < to su ggest th a t th e y are a x io m s for an order r e la tio n :
-~ ia < a , a < b & h < c Z) a < c , 3 b (a < b ).

T h e se are order properties w h ich are satisfied (in th e old in tu itiv e sense,
§ 8 ) w h en th e d om ain o f elem en ts is th e set of th e n a tu ra l num bers,
an d < is th e usu al order relation for th em . I t is easily seen th a t th e y
ca n n o t b e satisfied in a n y fin ite n o n -e m p ty dom ain of elem ents. (Th e

d eta ils are left to th e reader.)
N o w let us express th e a x io m sy ste m as one p red icate le tte r form ula,
u sin g <£?(*, b) in place of a < b t an d also usin g o n ly b o u n d v a r ia b le s :
V * - i <£?(*, a) & V aV bV c[£l(a9 b) & & (b 9 c) 3 & {a 9 c)] & \fd 3 b £ i{a 9 b).
T h is form ula, ca ll it “ F ” , is not satisfiable (in th e new set-th eo retic
sense) in a dom ain of k o b je cts, for a n y fin ite k> 0 , b u t is satisfiab le in
th e enum erable dom ain of th e natural num bers. Its n egatio n i F is th en
v a lid in a dom ain of k o b jects for e v e r y fin ite k, i.e. is ^ -id en tical for
e v e r y fin ite k, b u t is not v a lid in th e enum erable dom ain. T h u s i F
m eets th e t o ta lit y of n ecessary con dition s of T h eorem 20 , b u t b y th e
set-th e o retic result th a t e v e r y p ro v ab le form ula is v a lid in e v e r y non
e m p ty dom ain, it ca n n o t be p ro v ab le in th e p red icate calculus. T h is e x
am ple - i F an d others are g iv e n in H ilb e rt-B e r n a y s 1934 pp. 123— 124.
In term s of th e set-th e o retic in terpretatio n , th e com pleten ess of th e
p red icate calculus should m ean th a t every p red icate le tte r form ula
w h ich is v a lid in e v e r y n o n -e m p ty d om ain sh ould be p ro vab le. T h is
in terp retatio n is not fin ita ry , u n like th e corresponding in terp retatio n
for th e propositional calcu lu s (§§28, 29), an d so th e corresponding co m
pleten ess problem does n ot belo n g to m eta m a th em a tics. T h ese rem arks
su ggest th a t th e situ a tio n is not as sim ple as for the propositional calculus,
w hen w e com e to th e question of com pleten ess an d to th e d ecision problem .
W e sh all return to these problem s in a later c h a p ter on th e p red icate
calculus, w here several results w ill be presented, p a r tly m e ta m a th e m a tic a l
an d p a r tly of th e se t-th e o retic v a r ie ty (C hapter X I V ) .
F o r later reference, w e sum m arize our present conclusions in a theorem
an d corollary, w h ich w e m a rk w ith th e le tte r “ c ” to show th a t th e y do
n o t b elo n g to our sequence of m e ta m a th e m a tic a l theorem s, b u t are o n ly
estab lish ed here b y use of n o n -fin ita ry classical m eth ods. A lth o u g h w e
are usin g “ s e t-th e o re tic ” for th e present kin d of p red icate logic, som e of
th e results are o n ly on th e lev e l of classical n u m b e r -th e o r y ; e.g. w hile
T h eo rem 21 in its fu ll g e n e ra lity in v o lv e s sets of arb itra rily h igh cardin al
num ber, th e co ro llary can be inferred also from a sp ecializatio n of
T h eo rem 21 re latin g to th e enum erable dom ain of th e n a tu ral num bers an d
th e enum erable class of th e lo gical fu n ction s w h ich are expressible b y th e
use of th e lo gical sym b o lism ap p lied to < (where a < b is t or f , acco rd in g
as a is a lesser n a tu ra l num ber th a n b or not).
T heorem In each non-em pty object dom ain : E very predicate letter
2 1°.
form ula which is provable (or deducible from valid form ulas) in the predicate
calculus is valid.
C o r o l l a r y 0 . In order that a predicate letter form ula be provable in the
predicate calculus , it is not (in general) sufficient that the form ula be k-
identical for every positive integer k.
A nalogy betw een V, 3 and & , V. W h e n in terp reted in a fin ite d o m ain
of k o b je c ts 1, th e form ula V x A (x ) is sy n o n y m o u s w ith
A ( l ) & . . . & A (ft), an d 3 x A (x ) w ith A ( l ) V . . . V A(k), w here 1 , . . . , k
are nam es in th e form al sy ste m for th e o b jects. T h is su ggests a s lig h tly
differen t ap p ro a ch to th e results of th e preced in g section, from w h ich
w e w ill o b ta in an in term ed ia te m e ta m a th e m a tic a l result of som e in terest
in itself. (For th e case k = 2, cf. H ilb e rt-A c k e rm a n n 1928 pp. 66 — 68 .)'
W e sh all ta k e 1 , 2 , . . . , k to b e th e form al expressions O', 0 " , . . . , 0' *“ '
(the la st h a v in g k num erals from 1 to k. (H o w ever
accents) w h ich w e c a ll
it w o u ld serve our purpose e q u a lly w ell to use k in d iv id u a l sym bols.)
W e define predicate letter form ula with k in dividu als , or b rie fly k-
predicate letter form ula , b y allo w in g th e term s for C lau se 1 of th e d efin itio n
of p red icate le tte r form u la in § 31 to in clude n ow th e num erals from 1 to k
as w ell as th e v a ria b le s; a n d th e p red icate ca lcu lu s w ith th is n o tio n
of form ula w e ca ll th e predicate calculus with k in dividu als , or b rie fly th e
k-predicate calculus.
T h e n w e define k-proposition letter form ula either as a ^ -p red icate le tte r
form u la co n ta in in g no va ria b le s free or b o u n d (and therefore no q u a n ti
fiers) ; or e q u iv a le n tly b y allo w in g th e form ulas for C lause 1 of th e d ef
in itio n of proposition le tte r fo rm u la (§ 25) to in clu d e not o n ly th e p ro p o
sition letters b u t now also th e expressions resu ltin g from th e p red icate
letters b y s u b stitu tin g num erals from 1 to k for each of their a tta c h e d
va riab les, e.g. for k = 2 , <3(1), 3 ( 2 ) , S ( l) , <3(1, 1 ), <3(1, 2 ), <3(2, 1 ).
A co n ven ien t a b b re v ia tio n is to w rite th e la tte r as “ 3 / ' , “ 3 2” , “Bx” ,
“*^ 11 *> “ *^ 12” * “ <321” , re sp e ctiv e ly , so th a t in effect w e m erely a u g m e n t
th e form er list of proposition le tters b y th e sam e a lp h a b e tic a l le tters w ith
fin ite num bers of p o sitiv e in te g ra l subscripts <; k. O u r form er th e o r y for
th e pure p roposition al ca lcu lu s o b v io u s ly w ill a p p ly u n ch an ged , if n o w
each tw o letters differin g either a lp h a b e tic a lly or in th eir su b scrip ts are
tre a te d as d istin ct proposition letters.
G iv e n a n y closed ^ -p red icate le tte r form ula, w e define its k-transform
to b e th e ^ -proposition le tte r form u la w h ich results from it b y replacin g,
su c cessiv ely, each p a rt of th e form V x A (x ) w here A (x ) is a ^ -p red ica te
le tte r form u la b y A ( l ) & . . . & A (ft), an d likew ise each p art 3 x A (x ) b y
A ( l ) V . . . V A(fe), u n til all of th e qu an tifiers are elim inated. I t is ea sily
seen th a t th e order of th e replacem en ts does n o t affe ct th e result.
E xam ple 1. The 2-tran sfo rm of Vbla<£t(a, b) 3 3aVb£l(a, b) is

(<3(1, 1) V < 3(2 , 1 )) & (<3(1, 2) V<3(2, 2 )) 3 ( 3 ( 1 , 1) & < 3 (1, 2)) V (<3(2, 1 )
& <3(2 , 2 )), or b rie fly
( 3 n V <3al) & (<3ia V 3 22) 3 (<3n & <312) V (<321 & <322).
G iv e n a ^ -p red icate le tte r form ula A ( x v . . . , x n) w ith e x a c tly th e

d istin c t free v a riab les . ..,
x v th e set of its
x n, k -tr a n s fo r m s shall be th e
tran sform s of th e k n closed ^ -p red icate le tte r form ulas o b ta in e d from
A ( x v .. x n) by s u b s titu tin g for x lf . . . , x n each of th e k n d ifferen t
w -tuples of n um erals from 1 to k.
E xam ple 2. T h e 2-tran sfo rm s of Va3(d, c) 3 <3(£, c) are
&c^2j -D , eT^2&3 22 3 12, <3u & 32j,
c <3j2^ c^22Dc^22.
T heorem 22. For A > 1: If a fo r m u la E is p r o v a b le { d e d u c ib le
fr o m fo r m id a s T) in the p tir e or k - p r e d ic a te c a lc u lu s , th e n a ll o f th e k -
tr a n sfo r m s o f E p r o v a b le { d e d u c ib le fr o m th e k - tr a n s fo r m s o f th e fo r m u la s
T) in ^ p r o p o s itio n a l c a lc u lu s . (H ilb e rt-B e rn a y s 19 3 4 , pp. 119 ff.)
P roof is left to th e reader.

W e easily see th a t a p red ic a te le tte r form u la is ^ -id en tical, if an d o n ly
if all of its ^-transform s are id e n tic a lly true (§ 28). T h u s, u sin g T h eo rem
9, T h eo rem 20 becom es a co ro llary of T h eo rem 22.
E xample 1 (concluded). The form ula

( 3 U V <321) & (<312 V 3 22) 3 (<3n & <312) V (<321 & <322) is n o t p ro v a b le
in th e p ro p o sitio n al calcu lu s, since it ta k e s th e v a lu e f w hen <3n , <312,
3 21, 3 22 ta k e th e re sp ective v a lu es t, f, f, t. H en ce yib3aIH{a, b) 3
3 aiib€l{a, b) is u n p ro v a b le in th e p red icate calculus. C o m p arin g these
v a lu e s for 3 n , 3 12, 3 21, 3 22 w ith th e ta b le for I(j, b) in E x a m p le 2
§ 36, w e see th a t th is is b a s ic a lly th e sam e re fu ta tio n as there.
E xam ple 3. Is -vn3<z<3(tf) 3 3 a n n < 3 (a ) (w hich is of th e form

lie ! 3 l i b of C o ro llary T h eorem 17) p ro v a b le in th e in tu itio n istic p red
ic a te ca lcu lu s? B y T h eo rem 22 (w ith k = 2), it is o n ly if - t - i ( 3 1 V 3 2)
3 m c 3 1V - i n 3 2 is p ro v a b le in th e in tu itio n istic propositional
ca lcu lu s (a result to be used in § 80).
I t is o ften useful h e u ristica lly in s tu d y in g th e p red icate calcu lu s to

th in k of V x A ( x ) as a co n ju n ctio n e x te n d e d over all th e m em bers of th e
o b je c t dom ain, an d of 3 x A (x ) sim ilarly as a d isju n ctio n , even th o u g h
o n ly in th e case of a fin ite dom ain w ith a g iv e n num ber k of m em bers
are w e able to co n stru ct sy n o n y m o u s form al expressions on th is b asis.
From this analogy between V and & and between 3 and V, and the notion
of & as analogous to • and V to + (end § 29), some authors write VxA(x)
as "IIxA(x)” and 3xA(x) as “ E xA ( x ) ” . (Compare also the definitions of the
product and sum of a set of sets, § 5.)
In retrospect, we can see how the postulates 3 — 6 for & and V suggest
those for V and 3. The actual postulates of Group A2 are what the analogy
calls for, allowing for certain differences in detail. The analogy is clear
in the corresponding derived rules of Theorem 2 . Of the numbered
results, e.g., *75, *76, *83, *84, *91, *92 are analogous to *37, *38, *56,
*57, *35, *36, respectively.
P r e d icate calcu lu s w ith a po stu lated su bstitu tio n r u l e . A s in
the case of the propositional calculus (§30), the predicate calculus is
usually formulated with a postulated instead of a derived substitution
rule, namely, in the notation of Theorem 15 § 34 and under Proviso (A):
E
E*.
The predicate letters are then called predicate variables; the rule is usually
construed to apply to a single variable at a time; and as before these
variables must be held constant in subsidiary deductions for assumptions
formulas to be discharged. (The first accurate statement of the substi
tution rule seems to be that of Hilbert-Bernays 1 9 3 9 pp. 377—378, or
1 9 3 4 p. 98 understanding the restriction on “Einsetzung” to apply also
to <‘Umbenennung,\)
w-v a l u e d pr ed ica te c a l c u l i. Cf. R osser and T u r q u e tte 1948 - 5 1 ,
1952.
S e v e r a l - sorted pr ed ica te c a l c u l i . A s was implied in § 31, the
predicate calculus may be studied with several object domains, some of
the variables being specified as ranging over one of these domains,
others over another, etc. If these several object domains are regarded as
simply several different primary categories of objects, the only new feature
involved is that no operation should be performed which substitutes a
term or variable of one sort, i.e. referring to one of the domains, for one
of another. For example, *79 would hold only when x and y are variables
of the same sort. Each of the attached variables for a name form should
be of a specified sort. (Cf. Herbrand 1 9 3 0 , Schmidt 1 9 3 8 , Wang 1 9 5 2 and
Example 13 § 74.)
H igher pred ica te c a l c u l i . However, one may also obtain predicate
calculi with several types of variables by starting with predicate calculus
first w ith a single p rim a ry d o m ain of o b je cts called in d ivid u a ls ; th e n

w ith an a d d itio n a l d o m ain o f o b je c ts co n sistin g of th e p red icates o ve r th e
first d o m ain of o b jects, th u s a d m ittin g qu an tifiers V P an d 3 P w here
P ( a x, . . . , a n) is a p red icate v a ria b le o f th e first s y s te m ; an d so on. W h e n
a h iera rch y of p red icate ca lc u li b u ilt on th is plan is b e in g considered, th e
first is called th e restricted predicate calculus or predicate calculus of first
order , an d th e others predicate calculi of second order , third order , e tc.,
or g e n e ra lly higher predicate calculi. M a n y d ifficu lt question s arise in
con siderin g hierarchies o f sy ste m s of th is sort, w h ich are in v e s tig a te d b y
th e lo gicistic school (§ 12 ). A b rief in tro d u ctio n is p ro v id e d b y C h a p te r
IV of th e 2n d (1938) or 3rd (1949) ed ition o f H ilb e rt-A c k e rm a n n 19 2 8 .
Predicate calculi of second order are treated in Church 1956 Chapter V,
and Church planned to treat predicate calculi of higher order in Chapter
VI of his projected vol. II.
Chapter VIII
FORM AL NUM BER TH EO R Y
§ 38. I n d u c t i o n , e q u a l i t y , r e p l a c e m e n t . In th is ch a p te r w e return
to th e s tu d y of th e fu ll form al sy ste m of C h a p te r I V .
W e sh all n ow s ta te our results m a in ly w ith p a rticu la r form al va riab les,
as th e p o stu la tes of G ro u p B (after 13) were sta te d , so t h a t th e p ro v a b le
form ulas w ill read as p a rticu la r theorem s of num ber th e o r y form alized
in th e sym b o lism of th e syste m . R e su lts of th e form b A sta te d w ith
p a rticu la r free v a riab les ca n b e ap p lied w ith term s s u b s titu te d for those
va riab les, in v ie w of th e su b stitu tio n rule for in d iv id u a l v a riab les (§ 23
an d *66 § 32).
F ro m P o stu la te 13 (using D -in tro d ., V -in tro d ., & -in tro d . an d D -elim .),
w e h a v e th e fo llo w in g form al rule of m a th e m a tic a l in d u ctio n . F o r m a l in
d u ctio n s b y th is rule are o f course alto g e th e r sep arate from inform al
m a th e m a tic a l in d u ctio n s used in p ro v in g m e ta m a th e m a tic a l theorem s.
I nduction rule . L et x be a variable , A (x ) be a form ula , and T be a

list of form ulas not containing x free . I f T b A ( 0), and V, A (x ) b A (x ')
w ith the free variables held constant for A (x ), then V b A (x ).
B e g in n in g w ith th e proofs of th e n e x t theorem , w e sh all fre q u e n tly
a d o p t a m ore inform al p resen tatio n o f d em o n stratio n s of form al p r o v a
b ilit y or d e d u cib ility . T h e use a b o v e of th e ch ain m e th o d in h a n d lin g
eq u ivalen ces (§ 26) w as a step in th is d irection . W e n ow go further, o m it
tin g th e s y m b o l “ b” in m a n y situ ation s. T h u s w e sh all s a y “ assum e A ” ,
m ean in g t h a t w e w ish to ta k e A as an a ssu m p tio n form u la in co n stru ctin g
a d ed u ction . O f course, p ro p erly w e are n o t assum in g a n y th in g , b u t
in d ic a tin g th a t th e form ulas to follow are to b e fo rm a lly d ed u cib le from A
(and a n y oth er assu m p tio n form ulas w h ic h w e h a v e in trod uced), u n til
th e disch arge o f th e assu m p tio n fo rm u la A is in d ic a te d or im p lied b y
th e c o n te x t. A t each sta g e in such in form al presen tation , th e form ulas
g iv e n are to b e und erstoo d as d ed u cib le from all th e assu m p tion form ulas
n o t y e t discharged.
E xample 1. T h e follow ing is a d em on stration in th is presentation
181
182 FORMAL NUMBER THEORY CH. VIII
that b 3xA(x) 3 Vx -i A(x) (cf. the first half of the proof of *83 § 35).
In preparation for 3-introd., assume 3 xA(x). Preparatory to 3-elim.
from this, assume A(x). We shall deduce n VxnA(x) by reductio ad
absurdum (i.e. —i-introd.). Assume for this purpose Vx-iA(x). Then by
V-elim. -iA(x), contradicting A(x). [Then -i Vx -iA(x) by the -i-introd.,
which discharges the assumption Vx - i A(x). Since -iVx-iA(x) does not
contain x free, the 3-elim. can be completed now, discharging A(x).
Finally we have 3xA(x) 3 -iVxnA(x) by the 3-introd., discharging
3xA(x).] The bracketed steps will often be tacit.
To analyze this, we list the formulas, showing by an arrow how long
each assumption remains in force.
1 . 3xA(x) — assumed,
j 2. A(x) — assumed,
j 3. Vx-nA(x) — assumed,
j 4. -iA(x) — V-elim., 3.
v 5. n Vx-iA(x) — -i-introd., 2, 4.
v. 6. “i VxiA(x) — 3-elim., 5.
7. 3xA(x) 3 -1 Vx “i A(x) — 3-introd., 6.
Each formula is deducible from the assumption formulas whose arrows
appear opposite it, e.g. Line 5 means that nVxnA(x) is deducible
from 3xA(x) (Line 1) and A(x) (Line 2). Stating these facts in the
b -notation, our demonstration appears in the former style.
1. 3xA(x) b 3xA(x).
2. A(x), 3xA(x) b A(x).
3. Vx-iA(x), A(x), 3xA(x) b Vx-iA(x).
4. Vx -iA(x), A(x), 3xA(x) b ~«A(x) — V-elim., 3.
5. A(x), 3xA(x) b “iVxnA(x) — -i-introd., 2, 4.
6. 3xA(x) b “i Vx 1 A(x) — 3-elim., 5.
7. b 3xA(x) 3 —iVx“iA(x) — 3-introd., 6.
Note that the columns of assumption formulas take the place of the
arrows. For Step 5, we can supply Vx -i A(x) as an additional assumption
formula in 2 by general properties of h At Step 6, 3-elim. gives 3xA(x),
3xA(x) b “i Vx -i A(x), and the extra 3xA(x) is omitted by general prop
erties of h Actually it is immaterial whether 3xA(x) is considered as
an assumption formula for 2—5 or not.
Applications of the rule of V-elimination will now often be described
in the language of “cases” (cf. § 23). Applications of the formal induction
rule will often be presented in the following manner, using the same ter
§38 INDUCTION, EQUALITY, REPLACEMENT 183
m in o lo g y as in in form al in d u ctio n s (§ 7 ): B asis. . . . A(0). In d u c tio n ste p .
A ssu m e A (x ) (the h yp o th e sis o f th e in d u ction ). T h e n . . . A ( x '). [ A t th is
p o in t th e in d u ctio n step is co m p le ted , an d A (x ) ceases to b e an a ssu m p tio n
form ula.] H en ce A (x ). [O th er form ulas T m a y h a v e been in use as as
su m p tio n form ulas th ro u gh o u t. T h is proof t h a t b A (x ), or t h a t T b A (x ),
w e s a y is “ b y (formal) in d u ctio n on x ” ; A (x ) is th e “ in d u c tio n fo rm u la ” .]
T h is in form al p resen tatio n is co n ven ien t in situ a tio n s w here th e fo rm al
d eve lo p m en t clo se ly parallels in tu itiv e reasonings. I t sa v e s sp ace, a n d
brin gs our procedures for d e m o n stra tin g th e fa c ts a b o u t fo rm al p r o v a b ility
an d d e d u c ib ility still closer in ap p earan ce to th e m eth o d s of in fo rm al
m a th e m a tic s (cf. § 20 ).
T h e reader sh ould th e n u n d ersta n d a t a ll tim es h o w th e procedures
ca n b e m ad e rigorous as a p p lica tio n s o f our d eriv ed rules s ta te d in term s
of th e s y m b o l “ b” . T h e b -n o ta tio n h as been u sed to g iv e con cise a n d
accu ra te sta te m e n ts of our d eriv ed rules, sh o w in g c le a r ly th eir stru ctu re.
W e co n tin u e to use it w h en w e h a v e new rules to sta te , a n d in o th er
passages w here it helps to em p h asize th e form of th e d e d u c ib ility re latio n
ships or th e fa c t th a t w e are ta lk in g a b o u t th e form u las of th e sy s te m
(and n o t in them ).
W e sh all h ereafter c o m m o n ly a b b re v ia te a-b to “a b ” .
T heorem 23. (Properties o f e q u a lity.)
*100 . b a=a. *10 1. b a=b D b=a.

*10 2 . b a=b & b —c D a=c.
(R e fle x iv e , sy m m e tric an d tra n s itiv e properties.)
*10 3 (A x io m 17). b a = b 'D a '= b '.

*10 4 . bd —b D a -\-c = b -\-c . *10 5 . b & =b D c -\-a = c + b .
*10 6 . ba = b D ac=bc. * 10 7 . b a = b D ca=cb.

(Special re p lacem e n t'p ro p erties o f th e fu n c tio n sy m b o ls ', + > *•)
*10 8 (A x io m 16). b a=b D (a = c D b —c).

*10 9 . b a=b D {c = a D c= b ).
(Special rep lacem en t properties of th e p red icate s y m b o l = . )
P roofs. *100. W e e x h ib ite d a proof o f th is fo rm u la d ir e c tly from the

p o stu la te s as E x a m p le 1 § 19. T a k in g a d v a n ta g e o f th e d e riv e d rules
e sta b lish e d m ean w h ile, w e ca n n o w sum m arize it thus. F ro m A x io m 16
b y s u b s titu tio n (*66 § 3 2 ), a+0=a 3 {aJ\-0 = a mD a = a ) . T h e n c e b y A x io m
18 (using D -elim . tw ice), a=a.
* 10 1 . U sin g th e inform al presen tatio n , assum e a=b. B y A x io m 16,
a —c D b —c. S u b s titu tin g a for c, a=a D b=a. T h e n c e b y *100 , b —a.
*10 4 . P roof w ill b e b y th e form al in d u ctio n rule. In b o th th e basis
an d th e in d u ctio n step , th e m eth o d of a ch ain o f e q u a lities w ill b e used
(cf. end of § 26). O u r use of th is m eth o d d epends on h a v in g a lre a d y
estab lish ed * 100— * 102. Sin ce w e do n o t y e t h a v e th e th e gen eral replace
m en t p ro p e rty (Theorem 24 (a)), a n y step w h ich en tails replacin g a term
b y a n o th er term g iv e n as eq u a l to it w ill h a v e to be ju stifie d b y a sp ecial
re p lacem e n t result fittin g th e situ atio n . H ere w e shall use A x io m 17
(*103) for th e purpose. S u b stitu tio n in an ax io m or p re v io u sly estab lish ed
form u la w ill n o t b e e x p lic itly m entioned. A ssu m e a= b. We d ed u ce
a Jr c = b - \- c b y in d u ctio n on c, th u s. B asis . a + 0 = a [A x io m 18] = b
[assum ption] = 0 [A x io m 18]. I nduction step . A ssu m e, as h y
p o th esis of th e in d u ctio n , a+c=b+c. Then a+cf= ( t f+ c ) ' [A x io m 19]
= (b + c )' [h yp o th esis of th e in d u ctio n , A x io m 17] —b + c 9 [A x io m 19].
*10 5 . L e t a —b D c - \- a = c + b b e a b b re v ia te d “A ( a , b y \ W e shall p ro v e
V bA (a, b) b y in d u ctio n on a. W e do th e in d u ctio n step, le a v in g th e
b asis to b e done sim ila rly b y th e reader. I n d . step . A ssu m e yib A (a t b ).
B y V -elim ., A (a, b). W e n o w d ed u ce A (af, b) b y in d u ctio n on b. B asis .
A ssu m e a!— 0. B u t b y A x io m 15, - i <z'=0. H en ce b y w e a k -l-e lim .
(§2 3 ), c + a ' = c - \ - 0. I n d . step . (W e do n o t need to use th e h yp o th e sis
of th e in d u ctio n on b.) A ssu m e a '= b '. T h e n b y A x . 14, a = b ; a n d u sin g
A {a ,b ), c + a —c + b . N o w c + a ' = (c + a )' [A x. 19] = (c + b )' [using
c -\-a = c -\-b w ith A x . 17] = c-\-b' [A x . 19]. (W h y d id w e ta k e yi b A (a i b)
in s te a d of A (a, b) as th e in d u ctio n form u la for th e in d u ctio n on a? Cf.
th e s ta te m e n t of th e in d u ctio n rule. F o r th e in d u ctio n on b w ith in th e
in d u ctio n step of th e in d u c tio n on a, V bA (a, b) is th e T.)
A n o th e r w a y to h an d le *1 0 5 is to p ro v e * 1 1 8 an d * 1 1 9 first, a fter
w h ich *1 0 5 ca n b e inferred from *10 4 .
*10 6 an d *10 7 . S im ila rly to *10 4 an d *10 5 . T o ju s t if y rep lacin g ac

by be in ac-\-a after assu m in g ac=bc, w e m u st use *1 0 4 ; etc.
R eplacement. T heorem 24. (a) I f u r is a term containing a specified

occurrence of a term r, and us is the result of replacing this occurrence by a
term, s, then
r==s h u r= u s.
(b) I f Cr is a form ula containing a specified occurrence of a term r (not

as the variable of a quantifier ), and C s is the result of replacing this occur
rence by a term s, then
r= s hXl''*,Xw C r ~ C8
§39 ADDITION, MULTIPLICATION, ORDER 185
where x lf . . . , x n are the variables of r or s which belong to a quantifier of
C r having the specified occurrence of r w ithin its scope. (R ep la c em e n t
theorem .)
E x a m p le to follow . P ro o f is b y th e sam e m eth o d as before (§§ 26, 33),

u sin g sev en a d d itio n a l lem m as.
A dditional lemmas for replacement. I f r and s are term s:

*110. r= s b r '= s '.
♦ 111. r= s b r + t= s + t. *112. r= s bt+ r= t+ s.
*113. r= s b r t= s t. *114. r= s b tr = ts .
*115. r= s b r = t^ s = t. *116. r=s bt= r ~ t= s.
P roofs. The first fiv e o f these lem m as follow from *10 3 — *10 7,
re sp e ctive ly, by su b stitu tio n (*66) an d D -elim . The la st tw o fo llo w
from * 1 08 a n d * 1 0 9 w ith *66, * 1 0 1 , D -e lim . a n d * 1 6 .
E xample 2. L et r be b, s be a} C r be 3 d ( d '+ b = c ) . T h e p arallel

co n stru ctio n s of C r from r an d of C s from s are as follow s, a n d th e d e p th
is 3.
b a
d '+ b d '+ a
3 d {d '+ b z= c ) 3 d ( d '+ a = c )
Let = b e w ritte n b etw ee n th e expressions in each of th e to p tw o lines,
an d ~ in each of th e b o tto m tw o . T h e resu ltin g form ulas are d ed u cib le,
each from th e preceding, u sin g su cce ssiv e ly * 1 1 2 , * 1 1 5 , * 7 2 (v a ry in g d).
U sin g th e a b b re v ia tio n (§ 17), th e re su lt ca n b e w ritte n b = a b
b<c ~ a < c. d n o t v a ried in th e r e s u lt?)
(W h y is
Corollary 1. U nder the conditions of the theorem: I f b r= s, then

b u r= u 8 and b C r ~ C s.
Corollary 2. U nder the conditions of the theorem: r = s , C r bXl‘“x* C s

w ith x v . . . , x n varied only for the first assum ption form ula. I f b r = s ,
then C r b C s. (R ep lacem e n t p ro p e r ty of eq u a lity .)
E xample 2 (concluded). U sin g also * 1 0 1 : a = b f b < c \ a < c .
A s before, a replacem en t m a y b e p reced ed b y a su b stitu tio n for in
d iv id u a l v a ria b les (cf. § 33 E x a m p le 4, an d § 34 ju s t a b o v e R e m a r k 2).
§ 39. A d d i t i o n , m u l t i p l i c a t i o n , o r d e r . P o stu la te G ro u p B (§19)

m a y b e described as follow s. P o stu la te s 14, 15 an d 13 express fo rm a lly
th e la st three of P ea n o 's ax io m s (§ 6). (Th e first tw o en ter in th e present

sy ste m , w h ich h as o n ly n a tu ra l n u m ber variab les, th ro u gh Clauses 1 an d 5
in th e d efin itio n of term , § 17.) A x io m s 16 an d 17 g iv e properties of
e q u a lity (includ ing as 17 th e u n iv a len ce of th e successor fu n ctio n ',
w h ich w as im p licit in P e a n o 's form ulation). A x io m s 18 an d 19 are w h a t
m a y b e ca lled "recursion eq u a tio n s' d efin in g th e fu n ctio n + , an d A x io m s
20 an d 21 are th e lik e for th e fu n ctio n *.
F o r our a b b re v ia tio n s “a ^ b " an d ""1", “ 2” , ""3", . . . , see § 17.
T h e o r e m 25. (A rith m e tic law s.)
*117. b (a + b )+ c = a + (b + c ). *12 1. b (ab)c=a(bc).

*118 . b a ’+ b = (a + b y . *12 2 . b a’b = a b + b .
*119 . b a+ b= b+ a. *12 3 . b ab= ba.
*12 0 . b a (b Jrc )= a b -{-a c .
(A sso ciative, c o m m u ta tiv e an d d istrib u tiv e law s for + an d *, w ith lem m as
used in p r o v in g th e c o m m u ta tiv e law s.)
*12 4 (A x io m 18). b a-\-0= a. *12 5 (A x io m 20). b a -0= 0.

*12 6 . b a -b 1 . *12 7. b a -\= a .
(D irect law s for 0 an d 1.)
* 12 8 . V a + b = 0 Z> a = 0 & b = 0 . *12 9 . b ab— 0 3 a= 0 V b= 0.

*13 0 . b a + b = 1 3 a= l V b = 1. *131. b ab= \ 3 a = 1 & b = \.
(In verse law s for 0 an d 1.)
*13 2 . b a + c —b + c 3 a= b. *13 3 . b C tÔ D (i a c— bc 3 a = b ).

(In verse law s for + a n d *.)
P roofs. In th ese proofs, since w e n ow h a v e T h eo rem 24 (a), w e d o

n o t need to in v o k e th e sp ecial cases of it * 1 0 3 — * 1 0 7 , as w e d id in th e
proofs for T h eo rem 23.
* 1 1 7 and *118 . B y in d u c tio n on c an d b, re sp e ctiv e ly .
* 119 . a , u sin g in d u ctio n on b in th e basis.

B y in d u ctio n on
* 1 2 2 . B y in d u ctio n on b. I n d . s t e p . A ssu m e a 'b = a b + b . T h e n

a 'b ' = a 'b + a ' [A x . 21] = ( a b + b ) + a f [hyp. ind.] = ( ( a b + b ) + a ) f [A x .
19] = (ab+(b+a)Y [ * 1 1 7 ] = (ab+(a+b))' [*1 1 9 ] = ( ( a b + a ) + b ) f [ * 1 1 7 ]
= (d b '+ b y [A x . 21] = a b '+ b ' [A x . 19].
*12 8 . I f h A(0) and b A (x’)t then b A (x ). F o r from b A ( x ') b y gen eral

prop erties of b, A (x ) b A ( x ') , an d so in d u ctio n on x applies. T h is rule
w e c a ll induction cases (on x). F o r * 1 2 8 , ca ll th e fo rm u la to b e p ro v ed
" A (a, b )” . T o p ro v e A (a, b) b y in d u ctio n cases on a , it w ill suffice to p ro v e
§39 ADDITION, MULTIPLICATION, ORDER 187
A ( 0, b) an d A(a', b). T o cases on b, it w ill
p ro v e these tw o , b y in d u ctio n
suffice to p ro v e th e four form ulas A (0, 0 ), A ( 0 , b'), A(a', 0 ), A (a', b'), i.e.
th e form ulas
0 + 0 = 0 D 0=0 & 0=0, 0 + ^ = 0 D 0=0 & b'= 0,
a'+0=0 D af= 0 & 0 = 0 , a'+ b'= 0 D a ' = 0 & b'= 0.
T h e proofs of these four are easy, as in each case w e ca n either refu te
th e prem ise of th e im p lica tio n (using A x io m 15) or p ro v e th e con clusion
(cf. * 10 a an d * 1 1 § 2 6 ).
* 130. T h is is estab lish ed sim ila rly , b u t w ith an itera tio n o f th e argu m e n t
b y cases, so th a t to p ro ve A ( x '), w e pro ve A(O') an d A ( x " ) . A lto g e th e r, to
p ro v e A (a, b)t it th u s suffices to p ro ve th e nine form ulas
A ( 0 , 0), A ( 0 , 1 ), A M "), A ( l , 0), A ( l , 1 ),
A (U "), A ( a " , 0 ), A ( ^ " ,l) , A{ani b").

The tre a tm e n t of each of th e nine is rou tine (using A x io m s 14 an d 15).
*132 . In d u ctio n on c, u sin g A x io m 14.
*1 3 3 . L e t " A (a, b y a b b re v ia te ac=bc D a—b. A ssu m e c + 0 . W e

d ed u ce VaA(a, b) b y in d u ctio n on bf as follow s. (Cf. *95.) B asis . A ssu m e
ac=0c. I3y *12 5 (and * 1 2 3 ), a c = 0 . But c+ 0 . H en ce b y * 1 2 9 an d
p ro p o sitio n al calculus, a—0. I nd . step . A ssu m e VaA(a,b). By V -elim .,
A (a,b). W e sh all ded u ce A(a, b') b y in d u ctio n on a. B asis . A ssu m e
0 c=b'c. B y * 1 2 5 (and * 1 2 3 , * 1 0 1 ) , b'c= 0; an d b y *12 9 , b'=0V c=0.
B u t b* + 0 b y A x . 15, an d c + 0 b y h y p o th e sis; h ence - i ( i ' = 0 V c = 0 ) .
F r o m th is co n trad ictio n , b y w e a k -i-e lim . (§2 3 ), 0 =b'. I nd . step .
A ssu m e a'c=b'c. B y * 122 , ac+c—bc+c. B y 13 2 *, ac=bc. T h e n c e b y
A(a,b), a~b\ an d b y *10 3 , a'=b'.
A n o th e r w a y to h an d le *13 3 is to w a it u n til * 1 3 9 is estab lish ed ,
a fte r w h ich it can b e p ro v e d sim ila rly to *14 6 b .
We now use th e a b b re v ia tio n “a b ” as a b b re v ia tio n
for b c a ; “a < ,b ” for a b ” for b < ,a ; “a a<c. *1 3 4 d . b ^ k c D ^ c .
(T ra n sitiv e law s.)
*135a. b a<a'. *135b . b 0 <a’. *136 . b 0 ^ .
*137 ( = * 1 3 7 0). b a=0V3b(a=b'). * 1 3 7 j. b a= 0 V a = 1V U{a=b").

* 13 7 ,. b a = 0 V a = I V a = 2 V 3 b {a = b ' " ) .
*13 8 a . b a b ~ a ^ b '.

(Order properties for 0 an d '.)
♦ 139. b a b.

*14 0 . b ~i a < a . *14 1. b a b.
(C o n n e x ity , irreflexiven ess, asy m m etry .)
*14 2 a. b a+ b > a. *14 3 a . b b ^ O 'D a b > . a ,

*14 2 b . \- b ^ O a+b>a. *14 3 b . b aÔ & b > 1 3 ab> a.
b b 3
*1 4 3 c . a ' b > a ; hence b £^0 3 lc ( c ^ > ^ ) .
* 144a. b ~ *145a. b c # 0 3 ~ a c c b c ).
*144b. b &<b ~ e. * 145b. b 3 (a< b ~ ac< bc).
(In eq u alities u n der a d d itio n an d m u ltip licatio n .)
*14 6 a . b b =£0 3 3 ^ 3 r (a = h q + r & r < b ).

*14 6 b . b a = b q 1-\ - r l & r x< b & a = bq 2+ r 2 & r 2< b 3 q \ — q % & r x= r 2.
(E xiste n ce an d uniq ueness of q u o tien t an d rem ainder.)
P roofs. *1 3 4 a . A ssu m e a < b < c , i.e. a < b & b < c, i.e.

3 d (d '- \ - a = b ) & 3 e ( e ' + b = c ). P r e p a ra to r y to 3- an d & -e lim ., assum e
d ' + a = b an d e ' + b = c . T h e n e ' + ( d f -{- a ) = c , w h ich ca n b e reassociated
as (e ' + d ) f + a = c . B y 3 -in tro d ., l f ( f ' + a = c ), i.e. a < c .
*13 4 b . F ro m * 1 3 4 a , w ith th e h elp of proof b y cases (§ 23). (Cf.

E x a m p le 2 § 38.)
*13 6 . B y in d u ctio n cases on a , u sin g * 1 3 5 b .
*137, *13 7*. By rep e ate d use of in d u ctio n cases; or th u s: The

fo rm u la o f * 1 3 6 is e q u iv a le n t to th a t o f * 1 3 7 , u sin g A x . 18 an d properties
of == (cf. E x a m p le 4 § 3 3 ); an d * 1 3 7 !, * 1 3 7 2, . . . follow su ccessively.
*1 3 8 a . U sin g 3- an d V -elim . an d in tro d . w ith * 1 3 7 ,

b a < b ' ~ Q' - \ - a = b ' V 3c (c " + a = b '). U s in g * 1 1 9 an d A x s . 19, 17, 14
an d 18, b 0 ’ -\- a = b ' ~ a = b . U sin g * 1 1 9 an d A x s . 19, 17 an d 14 w ith
* 7 2 (or 3-elim . a n d introd.), b 3 c (c " + a = b ') ~ a < b .
*13 9 . B y in d u ctio n on b , u sin g * 1 3 6 in th e basis, an d * 1 3 8 a ,b in th e

in d u ctio n step.
*14 0 . A ssu m e a < a , i.e. 3 b (b f + a = a ). F o r 3-elim ., assum e b ' + a = a .

Then by *132 w ith *12 4 and *119 , b' = 0, co n tra d ic tin g Ax. 15.
R emark. B e ca u se th e co n tr a d ic to ry form ulas b' = 0 an d b '^ 0
co n ta in b free, w e c a n n o t c a r r y out th e 3-elim . im m e d ia te ly . But by
§40 THE FURTH ER DEVELOPM ENT 189
w e a k -i-e lim . (§ 23), w e ca n first d ed u ce a p air o f c o n tr a d ic to r y fo rm u las
n o t co n ta in in g b free, e.g. 0 = 0 a n d 0=£0.
B y *14 0 , 1- a —b 3 -i a<b.
*14 3b. A ssu m e a ^ 0 & / > > 1. B y <*=£0 w ith * 1 3 7 , l c ( a = c ' ) ; a n d b y ^ > 1
w ith *14 0 , * 1 4 1 a n d * 135a, b ^ 0 & b ^ \, a n d th en ce w ith * 13 7 x, 3 d (b = d " ).
A ssu m e (prep aratory to 3-elim inations) a —c' an d b=d".
* 144a. F ro m * 104, * 132, * 1 1 7 an d *7 2 .
*1 4 5 a . A ssu m e c ^ 0 . P art 1: to d educe a<b a c < b c . (L e ft to th e

3
reader.) P art 2: to ded u ce a c < b c 3 a < b . A ssu m e a c < b c . T o d ed u ce
a < b , it w ill suffice b y cases (V-elim.) from * 1 3 9 to d ed u ce a < b u n d er each
o f th ree case h yp o th eses. C ase 1: a b , i.e. b < a . T h e n b y th e result of
P a r t 1, b c < a c , i.e. a c > b c . T h e n c e b y * 1 4 1 , - i a c < b c .
*14 6 a . U se in d u ctio n on a (after assu m in g b ^ t 0). (Proof ca n also

b e b ased on * 143c an d * 149.)
*14 6 b . A ssu m e a=bqx+ r 1 & rx q 2- A ssu m e qx< q 2. F o r
a co n trad ictio n from
3-elim . from this, assum e e'+ qx= q 2. N o w bqx-\-rx = a = bq2+ r 2 =
b(e'+q1)+ r 2 = bq.+ibe'+r^. So r i = be'+r2 [*13 2 ] ^ be’ [*14 2a]
;> b [* 14 3 a]. T h is co n tra d icts r ^ b , by *14 0 a n d * 1 4 1 . The o th er
case is sim ilar. H a v in g d ed u ced qi=q% , w e âve a = ^ ^ i+ ri =
w h en ce r x= r 2 b y *13 2 .
* § 40. The further development of number theory. O u r fo rm a l

sy ste m o f n u m ber th e o r y differs from th e in form al th e o r y in t h a t th e
lo gic is m ad e e x p licit. W e h a v e b ro u gh t our a c q u a in ta n ce w ith th e lo g ic
to a sta g e w here th e fu rth er d eve lo p m en t of n u m ber th e o r y in th e fo rm al
sy ste m w ill proceed m u ch alo n g lines a lre a d y fam iliar to us from th e
in form al th eo ry. W e sh all n o t co n tin u e s y s te m a tic a lly w ith th is d e v e lo p
m en t, b u t w ill o n ly n o te several asp ects of it, before tu rn in g to gen eral
m e ta m a th e m a tic a l q u estion s a b o u t th e system ,
F o r th is section,let x be a variable, A (x ) be a form ula , and y and z be
variables distinct from x and each other which are free for x in A (x ) and do
not occur free in A (x ).
T h e least num ber prin ciple (or well-orderedness of th e n a tu ra l num bers)
sa y s th a t, if there e x ists a n a tu ra l n u m ber x such th a t A (x), th e n th ere
ex ists a least such x , c a ll it y . T h e p ro p e rty o f y ca n b e expressed in th e
form al sy m b o lism by A (y ) & V z ( z < y 3 -iA ( z ) ) , or in an e q u iv a le n t

form u sin g:
*14 7. b z< y 3 -iA (z ) ~ A (z) 3 y < z .
( B y * 1 3 , *13 9 .) W e estab lish first:
* 1 4 8 °. b 3 y [ y < x & A (y ) & V z ( z < y 3 -iA ( z ) ) ] V Vy[y<x 3 -iA (y )].

(L a w of th e exclu d e d m iddle, an d least n u m ber principle,
for an in itial segm en t o f th e n a tu ral num bers.)
*14 8 a . A (x )V -iA (x ) bx 3 y [ y < x & A (y ) & V z ( z < y 3 - i A ( z ) ) ] V

V y[y< x o -iA (y )].
*14 8 b . |- - i “ i { 3 y [ y < x & A (y ) & V z ( z < y D - i A ( z ) ) ] V
V y [ y < x D iA ( y )] } .
(In tu itio n istic versions of th e sam e.)
P roofs . *14 8 . By in d u ctio n on x , thus. L e t th e form u la of * 1 4 8 b e

a b b r e v ia te d “ P (x )V Q (x )” . B a s is . F ro m *13 6 , *14 0 , * 1 4 1 an d *10 a ,
b Q(0), w h en ce b P(0) V Q(0). I n d . s t e p . A ssu m e P(x) V Q (x). W e d ed u ce
P (x ') V Q (x ') th en ce b y cases (V-elim.). F o r Case 1, P (x) b P (x ') b P (x ')
V Q (x '), u sin g * 1 3 5 a , *13 4 a . F o r Case 2, w e use subcases from A (x ) V
- i A ( x ) (*5 1). F o r S u b ca se 2a, Q (x), A (x ) b P (x ') b P (x ') V Q (x '), u sin g
* 1 3 5 a . F o r S u b case 2b, Q (x ), - i A ( x ) b Q (x ') b P ( x ') V Q ( x ') , usin g * 138a.
*14 8 a . S in ce *5 1 is n o t a v a ila b le in th e in tu itio n istic sy ste m , w e n ow

use V x (A (x ) V - i A ( x ) ) as an assu m p tion form ula T for th e induction .
*14 8 b . F ro m A x io m S ch em a 6, b y 3 -e lim . an d con trap osition tw ic e

(*13 , * 1 2 ), A 3 -i-iC , B 3 ~i —»C b A V B 3 - n C b nn(AVB)
3 -i C. T h e n c e b y th e 3 - r u l e s : / / T, A b i i C and T, B b "i C
(with the free variables held constant for A and B, respectively ), then F,
—i (A V B ) b “ i “i C. T h u s w e ju s tify a m o d ificatio n of proof b y cases,
in w h ich th e case fo rm u la A V B an d th e conclusion C are d o u b ly n egated.
H en ce, if w e replace th e in d u ctio n form ula in th e proof of * 1 4 8 b y
- i - i ( P ( x ) V Q (x)), an d th e oth er case form ula b y -i-i( A ( x ) V -iA (x ))
(w hich is p ro v a b le in tu itio n is tic a lly b y * 5 la), th e in d u ctio n aga in w orks
(using *49a), an d g iv e s us *14 8 b .
N o w w e can infer th e least n um ber principle.
* 1 4 9 °. b 3 x A (x ) 3 3 y [A (y ) & V z ( z < y 3 -iA ( z ) ) ] .

(L east n um ber principle.)
* 14 9 a . A (x )V -iA (x ) p 3 x A (x ) 3 3 y [A (y ) & V z ( z < y 3 -iA ( z ) ) ] .
*14 9 b . b " i “ i{ 3 x A ( x ) 3 3 y [A (y ) & V z ( z < y 3 iA (z ))]} .

(In tu itio n istic versions of th e least num ber principle.)
§40 THE FURTHER DEVELOPMENT 191
P roofs . *14 9 (or *14 9 a ). A ssu m e 3 x A (x ); an d for 3-elim ., A (x ).

S u b s titu te x ' for x in * 1 4 8 (or in th e con clusion of *14 8 a) to o b ta in
P ( x ') V Q ( x ') . Case 1 : P (x '). Thence 3 y [A (y ) & V z ( z < y 3 -iA ( z ) ) ] .
Case 2 : Q (x '). T h e n c e w ith * 1 3 5 a , - i A ( x ) , co n tra d ic tin g A (x ). B y w e a k
-i-e lim ., 3 y [A (y ) & V z ( z < y D -iA ( z ) ) ] .
*14 9 b . U sin g in stea d *1 4 8 b an d th e m o d ific a tio n of p roof b y cases,

3 x A (x ) D - i - i 3 y [ A ( y ) & V z ( z < y D -i A ( z ) ) ] . N o w use * 6 0 h ,g .
O th e r consequences o f * 148a a r e :
*15 0 . A ( x ) V - i A ( x ) |-x 3 y [ y < x & A (y )] V - i 3 y [ y < x & A (y )].
*151. A ( x ) V - i A ( x ) hx V y [ y < x Z) A (y)] V - i V y [ y < x D A (y )].

(Of in terest for th e in tu itio n istic system .)
P roofs . *1 5 0 . F ro m *14 8 a , A ( x ) V - i A ( x ) hx 3 y [ y < x & A (y)] V

V y[y< x D -iA (y )]. But b V y[y< x D -iA (y )] ~ V y - i [ y < x & A (y )|
[*58b] ~ —i3 y [ y < x & A (y )] [ * 86 ].
*15 1. S im ilarly, a p p ly in g *1 4 8 a to i A ( x ) , an d replacin g n n A ( x )

b y A (x ) (since b y *49c, A (x ) V - i A (x ) b -v -iA ( x ) ~ A (x)).
A s an e x a m p le of a n u m b er-th eo retic theorem requirin g som e fu rth er

con cepts, w e sh all tre a t E u c lid 's theorem t h a t there e x is t in fin ite ly m a n y
prim e num bers. T h is m a y b e expressed b y sa y in g th a t to a n y n u m b er a,
there is a prim e greater th a n a. In fa c t, there m u st b e a prim e b e tw e e n
a +1 an d a ! -f 1 , in clu sive, b y th e fo llo w in g reasoning. E v e r y p o sitiv e
in teger n < a d iv id e s a!. H en ce none of th e m e x c e p t 1 d iv id e s a\ + 1 .
B u t a! + 1 > 1 ; so it is, or h as as facto r, a prim e. T h is prim e is b e tw e e n
a +1 an d a !+ 1 , in clu sive.
T h e tw o cases m a y b e co m b in ed b y n o tin g th a t th e le a st d iviso r of
a !+ 1 greater th a n 1 is a prim e greater th a n a. A lso th e reasonin g h o ld s
go o d u sin g in p la ce of a ! any com m on m u ltip le of 1 , . . . , a.
P r e p a ra to r y to th e fo rm al tre a tm e n t of E u c lid 's theorem , w e n ow
in tro d u ce “ a \b " “a d iv id e s b ” or “a is a fa cto r (divisor) of b ")
(read
as a b b re v ia tio n for 3 c (a c = b ). W e ca n sh ow :
*152. b a\ab. *153. V a\a. *154. b a\b & b\c 3 a\c.
*155. b a > \ D -*(a\b & a \b '). *156. b b ^ O D (a\b 3 0 < a < b ) .
(Properties of |.)
(H ints : F o r * 1 5 5 , use * 1 3 7 ! , * 1 4 5 a (w ith * 135a), * 1 3 2 , A x . 15. F o r * 1 5 6 ,

use *14 3 a .) N e x t w e in tro d u ce “ Pr(tf)" (“a is prim e") as a b b re v ia tio n for
a > 1 & -i3 c (l <c<a & c\a). T h e n E u c lid 's theorem is expressed in th e
form al s y ste m b y th e form u la lb (P r(b ) & b > a ).
A d iffic u lty for fo rm alizin g th e foregoing p roof is th a t th e sy ste m
has no term to express th e fu n ctio n ai W e a v o id th is b y e s ta b lish in g :
♦ 157. b 3 d [d > 0 & V b {0 < b < a 3 b\d)}.

(E x iste n c e of com m on m u ltip les of 1 , . . . , a.)
(P roof b y in d u ctio n on a. C a ll th e form u la “ 3dA (a, d ) ” . F o r th e basis,
h A(0, 1). For th e ind. step (prep aratory to 3-elim .), A (a ,d ) b
A(a', da').) N o w assum e in p rep aratio n for 3-elim .,
(1) d > 0 & V b (0 < b < a 3 b\d).

T h e v a ria b le d go ve rn e d b y this form ula then has th e role of a !; m ore
precisely, it m a y represent und er th e in terp retatio n a n y com m on m u ltip le
of 1, . . . , a.
B y ( 1 ), * 1 4 4 a an d * 1 5 3 , d '> 1 & d '\d '. H en ce 3 e(e > l & c\d '). By th e
least n um ber prin ciple (*14 9),
lb [ b > \& b \d '& V c ( c l & c\d'))]. A ssu m e, for 3-elim .,
(2 ) b > 1 & b\d' & V c (c l & c\d')).
A ssu m e \ < c l & c\d ' ) ; b u t from
l < c , c\b, b\d' (from (2)) an d *1 5 4 , c > l & c\d'. B y - i - an d V -in tro d .,
V c - \ ( \ < c & c\b). U sin g also
b > 1 from (2 ), Vv(b).
B y (2), b > \ & b \d '. H en ce b y * 1 5 5 , ~ib\d. H en ce b y (1), b > a .
B y & - an d 3 -in tro d ., 3b(Pr(b) & b > a ) . S in ce th is form u la does n o t
co n ta in b or d free, b y 3-elim . th e assu m p tion s (2) an d (1) are disch arged.
T h is co m p letes th e p roof of E u c lid ’s theorem in th e classical syste m .
T o p ro v e it in th e in tu itio n istic sy ste m , usin g * 1 4 9 a in stea d of *14 9 ,
it rem ains to estab lish ( e > l & e\d') V -i(e > l & e\d'). F o r th is purpose,
w e first e s ta b lis h :
*15 8 . b &—b V “ ia = b . *1 5 9 . b a<b V -i a<b.

*16 0 . b a\b 3c ( c < b & a c = b ).
(F o r use in th e in tu itio n istic tre a tm e n t of E u c lid ’s theorem .)
(P ro ve * 1 5 8 an d * 1 5 9 b y * 1 3 9 — * 1 4 1 ; *16 0 sim ilarly to *15 6 .) T h e n w e
p ro v e su c cessiv ely e>\ V - i e > l (b y *15 9 ), e\d' V -ie \d ' (b y *16 0 , *13 8 a ,
*15 0 , *15 8 ), an d (e> l & e\d') V -i(e > l & e\d r) (thence by R e m a rk
1 (b) § 29 an d T h eo rem 3 § 25). T h u s, in tu itio n is tic a lly as w ell as classi
cally-:
*16 1. b 3b'(Pr(b) & b > a ). (E u c lid ’s theorem .)
T h e process o f reco gn izin g t h a t th e proofs in an inform al th e o r y ca n

b e fo rm alized in a g iv e n fo rm al sy ste m is one of co n tin u al a n a ly sis an d
s te r e o ty p in g of a rg u m e n ts w h ic h recur in th e inform al th eo ry, to keep
§40 THE FURTHER DEVELOPMENT 193
p a ce as th e th e o r y d evelo p s w ith th e increasing co n den satio n of th e in
form al reasoning. W e a tte m p t to recognize su c c essiv ely these ty p e s of
inform al argu m en ts as b ein g form alizable, an d w e m a y record th e results
w e th u s o b ta in as d eriv ed rules for th e form al sy ste m . T h is is v e r y sim ilar
to th e sem i-form al process of d e v e lo p in g an in form al th e o r y itself from
e x p lic itly sta te d p o s tu la te s ; b u t here w e h a v e go n e fu rth er b a c k to
p o stu la te th e lo g ica l as w ell as th e (in th e o rd in ary sense) m a th e m a tic a l
principles.
In fo rm a lly , w e h a v e used m a th e m a tic a l in d u ctio n n o t o n ly in th e sim ple
(or ordinary) form , b u t also in a m o d ific a tio n ca lled ‘co u rse-o f-va lu es
in d u ctio n ' (cf. § 7 in clu d in g E x a m p le 2, th e proof of T h eorem 1 § 2 1, etc.).
I t is of in terest now to recognize th a t in th e form al sy ste m th is m o d
ifica tio n ca n b e d erived from th e sim ple form of in d u ctio n w h ic h is
p o stu la te d for th e system . W e sh all sta te it as a theorem sch em a; th e
m eth od , of fo rm u latin g a rule on th e b asis of th e sch em a has b een su f
fic ie n tly illu stra ted on sim ple in d u ctio n (cf. A x io m S c h em a 13 in § 19
w ith th e in d u ctio n rule in § 38).
In th e first sch em a *1 6 2 a , th e expressions A(0) an d
V x [V y (y< x D A (y)) 3 A (x ')] form alize th e b asis an d in d u ctio n step,
r e s p e c tiv e ly ; in th e m ore co m p a ct form *16 2 b , th e tw o are b ro u g h t
to g e th e r in th e sin gle expression V x [ V y ( y < x 3 A (y)) D A (x )].
*16 2 a . b A(0) & V x [ V y ( y < x 3 A (y)) 3 A (x ')] 3 A (x).
*16 2 b . b V x [V y (y< x 3 A (y)) 3 A (x)] 3 A (x ).

(C ou rse-of-valu es ind uction .)
P roofs . *16 2 a . A ssu m e A(0) & V x [ V y ( y < x D A (y)) 3 A ( x ')] , d ed u ce

V y(y < x 3 A (y)) b y sim ple in d u ctio n on x , an d infer A (x ) b y V -elim .
S o m etim es in d u ctio n s require a double basis; i.e. w e estab lish as th e

basis A(0) an d A( 1), an d th e n for th e in d u ctio n step infer A(x") from th e
tw o p reced in g cases A(x) an d A(x'). T h is can be tre a te d fo rm a lly as a
co u rse-o f-va lu es in d u ctio n , u sin g cases acco rd in g as x ' — 1 or x '> l
und er th e in d u ctio n s t e p ; or w e can m ak e it in to a sim ple in d u ctio n b y
usin g A (x ) & A (x ') as th e in d u ctio n form ula. T h is d e v ic e of u sin g a co n
ju n c tio n as th e in d u ctio n proposition applies sim ilarly to induction from a
k-fold basis for a n y fix e d k>2, an d also to th e inductive proof of several
propositions simultaneously.
In d u c tiv e argu m en ts are som etim es presented in th e guise of a de
scending induction or proof b y th e method of infinite descent. T h is consists
in e stab lish in g t h a t A(x) is false for e v e r y #, b y sh o w in g th a t if A (*) is
true for a n y x, there is a lesser n u m b er for w h ich it is also true.
♦ 163. b V x [A (x ) 3 3 y ( y < x & A (y))] 3 - i A ( x ) .
(M ethod of in fin ite descent.)
♦ 163a. b V x [A (x ) 3 - i in l y ( y < x & A (y))] 3 -iA (x ).

(A d d itio n a l version of in terest for th e in tu itio n istic system .)
P roofs . A s su m in g th e prem ise of eith er *16 3 or * 16 3 a , n A (x )

fo llo w s b y a co u rse-o f-va lu es in d u ctio n .
A nother method . T a k in g - i A ( x ) as th e A (x ) o f *1 6 2 b , th e result is
e q u iv a le n t to * 1 6 3 a b y th e fo llo w in g steps.
b V y(y< x 3 -iA (y )) 3 -iA (x ) ~ ^ [ V y ^ i ( y < x & A (y )) & A (x)] [*58 b
tw ice] ~ A (x ) 3 —i V y —i (y < x & A (y )) [*33, *58b] ~
A (x ) 3 - i - i 3 y ( y < x & A (y )) [♦ 86 ]. T h e n * 1 6 3 follow s from *16 3 a .
P ro o fs o f * 1 6 3 a an d * 1 6 3 ca n also be g iv e n b y redu ctio ad absu rdu m
from *14 9 or from *14 9 b ; an d c o n v e rse ly * 1 4 9 b can be p ro v e d by
re d u ctio ad ab su rd u m from * 1 6 3 a (using *60h, g). (In each case, th e
fo rm u la of th e one is d ed u cib le in th e in tu itio n istic p red icate ca lcu lu s
from t h a t o f th e other.)
T h e se e x a m p le s su g g e s t th a t th e form s of a rg u m e n ta tio n o rd in arily

en co u n tered in in form al e le m e n ta ry n u m b er th e o r y w ill tu rn o u t to be
fo rm alizab le in our fo rm al sy ste m . T h e la c k of su ch fu n ctio n s as al
rem ains a cause for d o u b t (alth o u gh w e d id g e t aroun d it in p ro v in g
E u c lid 's theorem ). A tte n tio n w ill b e g iv e n to th is q u estio n con cern in g
fu n ctio n s in §§ 4 1, 49, 59, 74, 82.
In § 42 w e sh all ta k e u p th e q u estio n o f th e com pleten ess o f th e form al
sy ste m . F ro m th e sta n d p o in t of th e in terp retatio n , th is in clud es w h e th er
all th e possible reasonin gs o f e le m e n ta ry n u m b er th e o r y (not m erely
th e co m m o n ly en co u n tered ones) are form alizab le in th e sy ste m , a t least
in so far as t h e y c o n tr ib u te to th e p roof o f propositions expressible in th e
sy ste m . W e sh all also con sider a m ore specific, s tr ic tly m e ta m a th e m a tic a l
n o tio n o f com pleten ess.
§ 4 1. F o r m a l c a lc u la tio n . A form u la A is s a id to b e {formally)

refutable , if —iA is p ro v ab le.
A closed form u la A (end § 32) is {formally) decidable, if A is either
p ro v a b le or refu tab le, i.e. if eith er b A or b "»A .
The fo rm al n u m b er-th eo retic sy ste m (or a sy ste m w ith form ation
rules of a like sort) is said to b e {sim ply) complete, if e v e r y closed form ula
A is fo rm a lly d e cid a b le ; {sim ply) incom plete in th e co n tr a ry case th a t
there is a fo rm a lly u n d ecid a b le closed form ula.
§41 FORMAL CALCULATION 195
T h e restriction th a t A b e closed is essential here, in order t h a t th e
m e ta m a th e m a tic a l n o tio n of sim ple com pleten ess sh ou ld h a v e th e in
ten d ed sign ificance. O th erw ise, und er th e g e n e ra lity in terp reta tio n of
th e free v a ria b le s in b o th A an d - * A , th e second form u la w o u ld n o t e x
press th e n eg atio n of th e proposition expressed b y th e first (§ 32).
E x a m pl e 1. T h e form u la 2 |a (i.e. 3c( 0 " -c = tf))e x p r e s s e s : e v e r y n u m b er

a is e v e n ; an d - i 2 | a expresses: e v e r y num ber a is n o t even , i.e. e v e r y
n u m b er a is odd. N e ith e r p ro p o sitio n is tru e ; and, w e hope, n eith er
form u la is p ro vab le. B u t V a 2 \a e x p re sse s: e v e r y n u m b er a is e v e n ; an d
-yV a2\a e x p re sse s: n o t e v e r y n um ber a is even. B y th e classical la w o f
th e e x clu d e d m iddle, one of th e tw o propositions sh ould b e true. In fa c t
th e second is; an d th e form u la -» V a 2 |a is p ro v a b le.
W e d id n o t a p p ly th e n o tio n of sim ple com pleten ess to th e p ro p o

sitio n a l an d p red icate calculi, b ecau se th e proposition an d p red ic a te
letters h a d th e role of free v a ria b les for th e in terp reta tio n (§§ 28, 29, 36,
37). S im p le com pleten ess is an oth er e x a m p le of a n o tio n of com pleten ess
w ith a p o sitiv e criterion (§ 29).
T h e term s 0, O', 0 ", ..., w h ich represent th e p a rticu la r n a tu ra l
num bers under th e in te rp re ta tio n of th e sy ste m , w e ca ll num erals , an d w e
a b b re v ia te th em b y th e sam e sy m b o ls “ 0 ” , “ 1 ” , “ 2 " , . . . , re sp e ctiv e ly ,
as w e use for th e n a tu ra l num bers in tu itiv e ly (as in §§ 17, 37). M oreover,
w h e n eve r w e h a v e in tro d u ced an ita lic letter, such as ‘ V \ to d esign ate
an in tu itiv e n a tu ra l num ber, th en th e correspon ding b o ld ita lic le tter*
" x ” sh all d e sig n a te th e corresponding num eral 0 (x), i.e. O'” *' w ith x
a ccen ts (x > 0) (as in § 37). In th is con n ection , w e can also use “x — 1 "
to d esign ate th e num eral w ith x —1 a ccen ts (for x> 0); th ere is no
a m b ig u ity , since w e h a v e no form al But “x + 1 ” d esign ates
0 te,+ 0 '.
Let P (x v . . . , x n) b e an in tu itiv e n u m b er-th eo retic p red icate . W e s a y
th a t P (x v . . . , x n) is num eralwise expressible in th e form al sy ste m , if there
is a form ula P ( x x, . . . , x n) w ith no free va ria b le s oth er th a n th e d istin c t
v a ria b les xv . . . , x n such t h a t , for each p a rticu la r n -tu p le o f n a tu ra l
num bers x nt
(i) if P (x v . . . , x n) is true, th en P ( x 2, . . . , x n), an d
(ii) if P (x l t ..., x n) is false, th en f- -i P ( x l f ..., x „).

In th is case, th e form ula P ( x p . . . , x n) num eralwise expresses th e p red icate
P (x v ..., x n) (w ith th e form al va ria b le s x v . . . , x n correspon ding to th e
re sp ective in tu itiv e va ria b le s xv . .., x n).
O u r m e ta m a th e m a tic a l use o f th is n otion w ill be con fined to cases
w h en there is a decision procedure for th e p red icate P (x v . . x n) (§ 30),
so th a t for e a c h ^ -tu p le x v . . x n,
(iii) P (x v ..., x n) is true, or P (x v ..., x n) is false.
U sin g (iii) w ith (i) an d (ii),
xn)
(iv) b P ( x l f . . . , x w) or b P ( X !,. . . , x n) ;
th u s P ( x 1}. . . , is d ecid ab le for each xv . . ., x n, or as w e shall s a y
P ( x 1,. . . , x n) is n u m e r a lw is e d e c id a b le . T h e form ula - i P ( x 1, . . . , x n) th en
n um eralw ise expresses th e p red icate n o t -P (x v . . . , x n).
T h e notion of num eralw ise e x p re ssib ility giv e s o n ly one of th e senses in
w h ic h a form ula P ( x 1, . . . , x n) m a y express a p red icate P (x l t . . . , x n).
I t requires m ore of th e d e d u c tiv e ap p aratu s of th e sy ste m th a n m erely
th a t P ( x 1?. . . , x n) should express P (x v ..., x n) under th e in terp retatio n of
th e sym b o lism (w ith th e n am e form in terp retatio n for x v . . x n, § 3 1),
w h ich after all requires n o th in g d e d u c tiv e ly . I t does not require th a t
form ulas expressin g vario u s general properties of th e p red icate should
be fo rm a lly p ro vab le.
E x am ple 2. T h e form ulas 3 c (c '- \ - a = b ) an d 3 c (a + c ' — b ) each ex

press a < b (w ith a , b corresponding to a ,b ) under th e m ean in gs of
th e form al sym b o ls, as w e cou ld see b a c k in § 17 before w e kn ew a n y th in g
a b o u t th e d e d u c tiv e rules.
T h e first form ula 3 c (c '- \ - a = b ) (w hich is th e one w e p ick ed for our
p erm an en t a b b re v ia tio n §§ 1 7 ,3 9 ) num eralw ise expresses a<b
in th e fo rm al n u m b er-th eo retic sy ste m , an d ev en in th e sy ste m w ith o u t
th e in d u ctio n sch em a (or A x io m s 20 an d 2 1 ), as w e n ow establish.
F o r (i), w e m u st sh ow th a t, if a an d b are a n y tw o n a tu ral num bers
such th a t a < b, th e n b 3 c ( e '+ a = 6 ) . For illu stratio n , let a = 3,
6 = 5. N ow b 0 " + 0 '" - (0 "+ 0 ")' [A x. 19] = ( 0 " + 0 ') " [A x. 19,
Ax. 17] = (0 "+ 0 )"' [A x . 19, A x. 17 tw ice] = 0 '" " [A x. 18, A x . 17
thrice]. Thus (ta c itly u sin g *10 2), b 0 " + 0 " ,= 0 " " '. B y 3-in tro d .,
b 3 c ( c '+ 0 /" = 0 " " /), i.e. b 3 c ( c '+ 3 = 5 ) . A sim ilar series of steps w ill
g iv e us b 3 c ( c '+ a = 6 ) for a n y an d 6 such th a t a < 6 . T o p ro ve
a
th is in general, w e m a y first establish as a lem m a b y inform al in d u ctio n

on k th a t for a n y term t, usin g A x s . 1 7 — 19 (and *10 2), b t + 0 (fe)= t (fc).
F o r (ii) w e m u st show th a t, if a an d 6 are a n y tw o n a tu ral num bers
such “th a t a is n ot < 6, th en b “i 3 c ( c ' + a = 6 ). If not a < 6 , th en
a > 6 . F o r illu stratio n , le t a = 3, 6 = 2. B y ap p licatio n s of A x s . 17— 19,
e x a c tly as a b o v e e x c e p t th a t c ' replaces 0 " (or ta k in g c ' in stead of 0 " as
th e t of th e lem m a), b c ' + 0 " ' = c " " . T h en ce c ' + 0 ' " = 0 " b c""= 0 "
b c" = 0 [using Ax. 14 tw ice]. But by Ax. 15, b “ i ^ ,=r0- By
redu ctio ad ab surdum (-i-in tro d .), b c ' + 0 " /= 0 " ; w hence b y V -in tro d .
an d *86 § 3 5 , b - i 3 c ( c , + 0 /" — 0"), i.e. b “i 3 c ( c ' + 3 = 2 ) . S im ila rly for
a n y other a an d b such th a t a > bt b c(c'-\-a=b).
Thus 3c(c'+a=b) num eralw ise expresses a <b in th e form al sy ste m
w ith o u t A x io m S ch em a 13. B u t w e can n o t e x p e c t th a t in th is sy ste m
form ulas expressing gen eral properties of < , such as those of * 1 3 4 a —
146b (w ith “a < b ” as a b b re v ia tio n for 3c(c'+a=b)), w ill be p ro v ab le,
e x ce p t in a few cases (e.g. *13 5 b ).
The other form ula 3 c(aJr c,=b) (w hich b y *119 is eq u iv a le n t to
3c(c'+a=b) in th e fu ll sy ste m w ith A x io m S ch em a 13) seem in gly
does not num eralw ise express a<b in the sy ste m w ith o u t A x io m S ch em a
13. O f course it does in th e fu ll sy ste m (or e v en w ith o u t A x io m S ch em a 13,
p ro v id e d * 118 or * 119 is su pplied as an a x io m ).
T h e num bered results of th is section (beginning w ith *(164)) refer pri

m a rily to th e fu ll n u m b er-th eo retic sy ste m (as th ro u gh o u t this c h a p te r).
B u t in fa c t for th em w e need no new ap p licatio n s of th e form al in d u ctio n
rule (or A x io m S ch em a 13), p ro vid ed certain p a rticu la r form ulas p rev io u s
l y p ro v ed b y m eans of it are av ailab le. More precisely, w e can g e t alon g
here w ith th e p red icate calculus, th e p articu lar n u m b er-th eo retic axio m s
14 — 2 1 , th e replacem en t p ro p e rty of e q u a lity w h ich depends on h a v in g
in ad d itio n o n ly *10 4 — * 1 0 7 § 38, an d * 1 3 7 (or *13 6 ) § 39, e x c e p t in a
few cases w h ich w e w ill keep tra c k of an d list a t th e end of th e section.
(This su b system of th e full sy ste m w as singled o u t b y R a p h a e l R obin son,
1950 a b stra c t*, in a con nection to b e discussed in §76 .)
The predicates
*(164) a=b, *(16 5) a<b
are numeralwise expressed by the respective formulas a= b and a <b,
i.e. lc(c'+a=b).
P roofs . *(16 5 ). By E x a m p le 2; or (in th e full n u m b er-th eo retic
system ) usin g * 1 3 5 a , *13 4 a , *14 0 , * 1 41 .
If x is a variable, A (x ) and B (x) are formulas, k is a natural number,

y is a variable distinct from x and free for x in A (x ) and not occurring free in
A (x ), and t is a term not containing x and free for x in A ( x ) :
*16 6 . A(0), A( l ) , ...,A(fc-l) b V x (x < * D A (x)).
*16 6 a . A(0), A( l ) , ..., A (k) b V x (x < ft D A (x)).
*16 7. V x (x < fc D A (x)) b A (i) for i = 0, 1, . . . , k—\.

*1 6 7 a . V x (x < fe D A (x)) b A(i) for i = 0, 1, . . k.
*16 8 . A (t) b V x [x > t D 3y ( y < x & A (y))].
*16 9 . V x [x < t D A (x )], ¥ x[x ^ t 3 B (x)] b V x [A ( x ) V B ( x ) ] .
W h en k = 0, th e list A(0), A ( 1 ) , . . . , A (fe — 1) of assu m p tio n form ulas

for * 1 6 6 is e m p ty , an d there are no A (i)'s for * 1 6 7 .
P roofs. *16 6 . I f w e ca n show th a t A(0), A( l ) , . . . , K {k — 1 ), x < k

b A ( x ) , th e n * 166 w ill follow b y D - an d V -in trod . B y * 137* (or v ia its proof,
f r o m * 1 3 7 o r *13 6 ), b x=0Vx=l V . . . V x = f c - 1 V x = * V 3 y ( x = y (* +1)).
A c c o r d in g ly b y V -elim . w ith w e a k -i-e lim . an d 3-elim ., it w ill suffice to
deduce either A (x ) or a co n trad ictio n from A ( 0), A( l ) , ..., A (ft— 1 ),
x<k w ith each of x = 0 , x = l , ..., x = f e — 1 , x = f e , x = y (/H1) in turn.
B u t (using n ow th e inform al presen tation , b egin n in g § 38) from each of
x = 0, x = l , ..., x=k—1 w ith th e corresponding one of A ( 0), A( l ) , ...,
A(fe— 1 ), w e o b ta in A (x ) b y re placem en t (C orollary 2 T h eo rem 24 §3 8 ).
F ro m x=k w ith x<k b y replacem en t, k<k, co n tra d ictin g - i f e c f e w hich
is p ro v a b le b y *(16 5 ). F ro m x = y (A:+1) an d x < f e b y replacem en t, y ik+1)< k ,
i.e. 3 z ( z ' + y (A;+1)= fe ). A ssu m e (for 3-elim .) z fJr y {k+l)= k . T h en ce b y k-\-\
a p p licatio n s of A x . 19 (w ith som e ap p licatio n s of A x . 17), (z'-\-y){k+1)= k ;
w h en ce b y k ap p licatio n s of A x . 14, ( z ' + y ) ' = 0 , co n tra d ictin g A x . 15. (Cf.
th e rem ark in th e proof of * 140 § 39.)
*16 7. Sin ce i < k, b y *(16 5 ) b i<k.

*16 9 . U sin g cases from *13 9 . R emark . W e require * 1 3 9 o n ly w ith t
su b stitu te d for b. W hen t is a num eral fc, this form ula a<kV a=k
V a>k can b e p ro v e d from * 1 3 7 or * 1 3 6 sim ila rly to *16 6 . (Use *(16 5 )
in th e first k cases. F o r th e & + 2 -n d case, a — b {k+1) = (b'){k) = (b’+ 0 ) (k)
[A xs. 18, 17] = b '+ k [A xs. 19, 17].)
A lth o u g h th u s each of th e fo rm u las a<0V a=0V a>0, a< 1 V #= 1
V # > 1, a< 2V a= 2V a>2, . . . is p ro v ab le in th e sy ste m la c k in g A x io m
S ch em a 13 (and ev en A x s . 14, 15, 20 , 2 1 ) b u t h a v in g th e form ula of * 1 3 7 or
* 136 as a d d itio n a l axio m , w e h a v e no ground for b e lie v in g th a t th e fo rm u la
ab of * 1 3 9 itself is p ro vab le in th a t system .
T h e rem ainder of th is section m a y be postponed, if th e reader prefers,

u n til ju s t before § 49.
U n d e r th e in terp retatio n of th e form al sym bolism , a n u m b er-th eo retic
fu n c tio n y {x 1}. . . , #n) is expressed b y a term t ^ , . . . , x w).
E xample ^ . The fu n ctio n (a + 1)2 is expressed under th e in ter

p re ta tio n b y th e term s (a')-(a'), a a + ( 2 a + \ ) t etc.
T h e o n ly n u m b er-th eo retic fu n ctio n s w h ich can be th u s d ire c tly e x

pressed are th e polyn om ials. H o w ever, w e sh all find th a t it is possible
to paraph rase m a n y propositions in w h ich oth er n u m b er-th eo retic fu n c
tion s occur, so th a t th ose propositions beco m e exp ressible in th e fo rm al
sym b o lism d esp ite th e la c k of term s expressin g th e fu n ctio n s th em selves.
F o r let f ( % , . . . , x n) b e a g iv e n n u m b er-th eo retic fu n ctio n , a n d let
P (x l f . . x nt w ) b e th e p red icate y(x l t . . x n) = w , w h ich w e c a ll th e
representing predicate of th e fu n ctio n <?(xv . . . , #n). I f th e p red icate
P (x v . . %n ) w ) is expressed in th e sy ste m b y a fo rm u la P ( x x, . . x n, w) ,
an d C(x) is a p red icate expressed b y C (x), th en C(cp(x v . . . , x n)) is expressed
b y 3 w (P (x 1, . . . , x n, w) & C ( w ) ) (and also b y V w ( P ( x x, . . . , x n, w) D C(w))).
T h is su ggests th a t, if th e represen tin g p red icate of a fu n c tio n is exp re s
sible in th e sy ste m , w e can h ope to be ab le to express an d d e v e lo p th e
th e o r y of th e fu n ctio n in th e sy ste m m u ch as th o u g h a term for th e
fu n ctio n itself w ere a v a ila b le. A m e ta m a th e m a tic a l in v e stig a tio n to
confirm th is co n jectu re w ill b e u n d erta k en la te r (§ 74).
W e seek som e in fo rm atio n n o w (and in § 49) on th e q u estio n for w h a t
fu n ctio n s th e represen tin g p red icates are expressible, or b rie fly , w h a t
fu n ctio n s are 'rep resen tab le'. T h u s far w e h a v e been ta lk in g o n ly a b o u t
th e in terp retatio n of th e sym b o lism , b u t p re se n tly w e w ill in tro d u ce a
n otion for th e represen tation of fu n ctio n s an alogo u s to 'n u m eralw ise
e x p re ssib ility ' for th e expression of predicates.
T h e n ecessary an d su fficien t co n d itio n th a t a p red icateP (x v . . . , x n> w)
be th e represen tin g p red icate of som e (single-valued) fu n ctio n cp(#x, . . . ,xn)
is t h a t for each w -tuple x v . . . , x n there e x ists a u n iq u e w such th a t
P ( % , . . . , x n, w). W h e n th is co n d itio n holds, th e fu n ctio n <p(xv . . . , x n)
represen ted ca n b e defin ed 'd e s c r ip tiv e ly ' from th e p red icate P {x l f . . . ,
x nf w) as the w such that P {x l f . . . , x ni w).
W e n ow in trod u ce “ 3 !x A ( x ) '' under th e u su al stip u la tio n s on th e le tters
(beginn ing § 40, an d end § 33) as a b b re v ia tio n for 3 x [A (x ) & V y (A (y ) D
x= y)] (read " th e re e x ists a u n iq u e x such th a t A ( x ) ” ). Then if
P (x v . . . , x n, w) is expressed b y th e form ula P ( x x, . . . , x n, w ), th e co n
d itio n t h a t P (x v . . . , x n, w) b e a represen tin g p red icate is expressed b y
th e fo rm u la V x x. . . V x n3 !w P ( x x, . . . , x w, w), or sim p ly b y 3 !w P ( x x, . . . ,
x n, w) w h en x v . . . , x n have th e g e n e ra lity in terp reta tio n (§ 32).
I f x, y and z are distinct variables , A (x ) is a form ula , t, r and s are term s ,

T are form ulas not containing x free, y , z, r, s and t are free for x in A (x ),
z and x do not occur in t, and y and z do not occur free in A ( x ) :
*170 . I f T , A (t), A (x ) b t = x with the free variables held constant for

A (x ), then T, A (t) (- 3 !x A (x ).
*171. b 3 !x (t= x ).
*172 . A (r), A (s), 3 !x A (x ) b r= s.

*173 . r ^ s , A(r), 3 !x A (x ) b n A ( s ) .
(Properties of 3!.)
*174 a. A (t) & V z ( z < t D -iA ( z ) ) b 3 !y [A (y ) & V z ( z < y D -iA ( z ) ) ] .

*174 b . b 3 y [A (y ) & V z ( z < y D n A ( z ) ) ] ~ 3 !y [A (y ) & V z ( z < y D ~ iA (z))].
(U niqueness of th e least x such th a t A (x).)
P roofs. *174 a. Using * 1 7 0 , *13 9 . Cf. the re m a rk in th e proof of *16 9 .

An in tu itiv e n u m b er-th eo retic fu n ctio n <p(xv . . x n) is said to be
numeralwise representable in th e form al syste m , if there is a form ula
P ( x x, . . . , x n, w) w ith no free v a riab les other th a n th e d istin ct v a riab les
xv . . . , x w, w such th a t, for each p articu lar w -tuple of n a tu ra l num bers
x^}..., x n,
(v) if <?(xlt . . . , x n) = w, th en b P(Xi,.. x tt, w ), an d
(vi) b 3 !w P ( x 1, . . xni w).
In th is case, th e fo rm u la P ( x 1, . . . , x n, w) numeralwise represents th e

fu n ctio n <?(xl f . . . , x n) (w ith th e o b vio u s correspondence of variables).
O u r fin ita r y (i.e. in tu itio n istic) use of th is n o tio n w ill b e con fin ed to
cases w h en there is a c a lcu la tio n procedure for th e fu n ctio n <p(xv . . x n)
(§ 30), so th a t th e w of (v) ca n b e fo u n d for a n y g iv e n x v . , . , x n (or w ill
m a k e th is t a c it ly an h yp o th esis).
If (p(xv . . . , x n) is num eralw ise represented b y P ( x j , . . . , x n, w ), th e
la tte r n um eralw ise expresses th e represen tin g p red icate P (x l t . . . , x n> w)
of 9 . F o r from (v) an d (vi), w e can infer th a t for e v e r y w,
(vii) if <p(xv . . . , x n) ^ w , th en b1 P ^ , . . . , x n, 10),
as follow s. T a k e th e w of (v) as th e r an d th e w of (vii) as th e s for * 1 7 3 .

U se *(164 ) to g e t th e p r o v a b ility of th e r ^ s .
W e h a v e no gro u n d to b e lie ve th a t, co n versely, (vii) for e v e r y w to g e th e r
w ith (v) n ecessarily im plies (vi).
W e h a v e used th e 3 !-n o ta tio n to sta te (vi) co m p a c tly . 3 !x A (x ) is
e q u iv a le n t to 3 x A (x ) & V x V y ( A ( x ) & A (y ) D x = y ) , in w h ich th e first
p a rt expresses existen ce a n d th e second uniqueness. T h e existen ce p a rt
for (vi) follow s a lre a d y from ( v ) ; so w h a t (vi) add s is th e uniqueness.
In num eralw ise re p re se n ta b ility of fu n ctio n s (just as in num eralw ise
e x p re s sib ility of pred icates), w e are lim itin g ourselves to th e con sid eration
o f th e va lu e s for p a rticu la r argu m en ts, in co n trast to gen eral properties.
T h e q u estion s considered are in th is w a y an alogo us to co m p u ta tio n a l
q u estion s in inform al arith m etic.
F o r ex am p le, w e h a v e n o t required in d efinin g 'num eralw ise repre-
s e n ta b ility ’ th a t th e fo rm u la 3 !w P (x 1, . . . , x n, w) should be p ro v a b le
w ith x 2, . . . , x n as form al variab les. T h a t w o u ld b e stronger th a n our
d em an d t h a t (vi) h o ld for e v e r y x l f . . . , xn w ith % , . . . , xn as in tu itiv e
va riab les, w h ich w o u ld follow from it b y su b stitu tio n (*66 § 32 w ith
xlf . . xn as th e t v . . t n). T h a t stronger co n d itio n (w ith ou t (v)) w ill
b e th e prerequisite for th e th e o r y to b e d eve lo p ed in § 74. F o r th e six
fu n ctio n s w e consider now , w e re a d ily o b ta in th a t also.
The functions
*(17 5 ) a', *(17 6 ) a + b , * (1 7 7 ) ab
are num eralwise represented by the respective form ulas a '= b , a + b —c, a b ~ c .
P roofs. *(17 6 ) B y * 1 7 1 an d th e form of th e represen tin g form ula
a -\-b = c , (vi) is im m e d ia te (and ev en b 3\c(a-\-b—c)). F o r (v), w e m u st
sh ow th a t for each p air a, b of n a tu ra l num bers, if c = a-\-b , th e n
b a -\~ b = c . F o r exam p le , if a = 2 an d b = 3 (then c = 5), w e h a v e
b a + b = c (i.e. b 0 " + 0 " ' = 0 ' " " ) as for (i) in E x a m p le 2.
*(177 ) S im ilarly. T h e proof of (v) ca n b e c o n v e n ie n tly arran ged as an

in tu itiv e in d u ctio n on b. I n d . step . Say c = ab, d = ab' (= ab-\-a =
c + a ). Then b obf = ab-\-a [A x . 21] = c+a [hyp. ind., *104] = d
[b y (v) for *(17 6 )].
Som e gen eral principles w ill illu m in a te our tre a tm e n t o f th e n e x t

exam p les. A prim e form u la is one co n ta in in g no lo gical sym b o ls, i.e.
here it is s = t for som e term s s an d t.
(A) Each closed prim e form ula s = t is form ally decidable [and s = t
is provable or refutable according as the terms s and t express the sam e or
different numbers under the usual interpretation of 0, ', + , •). Each prim e
form ula is num eralwise decidable.
P roof. U se *(17 6 ), * ( 1 7 7 ) , T h eorem 24 § 38, an d *(164 ).
E xample 4. Let s = t b e 0 " /- 0 " " + 0 ' = ( 0 " ' * 0 " ) " , i.e. a b b re v ia te d
3*44 -1 = (3-2)". N o w h 3 - 4 + l = ( 3 - 2 ) " ~ 1 2 + 1 = (3-2)" [since b y * ( 1 7 7 ) ,
h 3*4 = 12 ] ~ 1 3 = (3*2)" [since b y *(17 6 ), b 12 4 -1 = 13] ~ 1 3 = 8 [since
b y * ( 1 7 7 ) , b 3 * 2 = 6 ; a n d n o tin g th a t 6 " , i.e. ( 0 " " " ) " , i s 8]. B u t b y *(16 4 ),
b —i 13 = 8 . H en ce b -i3 * 4 4- l = (3*2)". W e h a v e used t a c it ly T h eo rem
24 (b) or its C o ro lla ry 1 , an d *21 § 26, in co n clu d in g b y th e ch ain th a t
b 3 * 4 4 -1 = ( 3 * 2 ) " ~ 1 3 = 8 ; an d *30 *1 8 b , or *20 an d C o ro llary T h eo rem
6 , in co m b in in g th is w ith b 13= 8 to infer t h a t b ~ 3*44 -1 =
i (3*2)".
(B) Let P ( x 1#. . . , x n) be a form ula containing free only the distinct
variables x l f . . . , x n, and suppose P ( x 1, . . . , x n) is num eralwise decidable
(<and num eralwise expresses P {x v . . . , x n)). Then: I f t v . . . , t n are term s

containing no variables {and therefore expressing numbers tv . . . , tn),
p ( t i ........ t n) is decidable {and P ( t 1?. . . , t n) is provable or refutable according
as P {tl t . . , , t n) is true or false) . t v . . t n are terms free for x v . . x n
I f
in P ( x 1, . . . , x n), P ( t j , . . . , t n) is num eralwise decidable.

P r o v e d lik e (A), w h ich is th e special case of (B) for w h ich P(xli.. ., xn)
is x 1 = x 2.
(C) A form ula composed out of closed decidable form ulas using only the
operators D , & , V, of the propositional calculus is decidable {and whether
it is provable or refutable can be determ ined by use of the classical 2-valued
truth tables § 28 taking t and f as 'provable' and ‘refutable', respectively).
B y L e m m a 13 § 29 w ith T h eorem 3 § 25.
(D) H e n c e : Each form ula without variables is decidable. Each form ula
without quantifiers is num eralwise decidable.
(E) Let A ( x x, <. . , x n,y ) be a num eralwise decidable form ula containing
free only the distinct variables x 1#. . . , x n, y ; and let z be a variable
distin ct from x v . . . , x n, y . Then V y ( y < z D A ( x 1, . . x n, y)) and
3 y ( y < z & A ( x l l . . . , x n, y)) are num eralwise decidable {and V y ( y < zD
A ( X 1#. . x n, y)) is provable or refutable, according as all of A{xv . . . , x n, 0),
A(x2, . . . , x nt 1) , . . . , A{xv . . . , x w, z — 1) are provable or some are refutable ;
3 y ( y < * & A(xlf. . x w, y)), according as some are provable or all are
refutable). S im ilarly w ith < in place of < .
P roof (for < ). U se *16 6 , * 1 6 7 , *(16 5) (also *5 8 b § 2 7 , *86 § 3 5 ).
C onsider th e d ivisio n o f tw o in tegers, a b y b. F o r exam p le , 13 = 5*2+ 3

w here 3 < 5. In w ords, w h en 13 is d iv id e d b y 5, th e q u o tie n t is 2 an d th e
rem ain der is 3. C u sto m a rily , th e d ivisio n process, a n d th e re w ith th e
q u o tie n t fu n ctio n [afb] a n d th e rem ainder fu n ctio n rm (a, b) are d efin ed
o n ly for b + 0. T o a v o id th e tro u b le o f discussing p a r tia lly d efin ed fu n c
tio n s n ow , we e x te n d th e d efin itio n s to b = 0 b y s e ttin g
th e case
[a/0] = 0, rm(tf, 0 ) = a. T h is preserves th e la w a = b[a/b ] + rm (a, b).
T h e n b \a {“ b d iv id e s a") if an d o n ly if r m {a, b) = 0.
The functions
*(178 ) [a fb l *(17 9 ) r m (a,b)
are num eralwise represented by respective form ulas Q{a, b, cj) and R {a, b t r)
such that , for an y num erals q and r:
*178a. Q ( a ,b ,q ) |- 3\qQ (a, b, cj). * 1 7 9 a . R ( a ,b ,r ) h 3 ! r R M , r).
§41 FORMAL CALCULATION 20 3
P ro ofs. *(179 ) an d * 1 7 9 a . L e t S ( a ,b ,r ) b e th e form ula
3q (q < a & a = b q -{-r & r< b ) V (b = 0 & r= a ).
T h e n let R (a, b , r) be
S(a, b , r) & V e ( e < r 3 -iS (tf, A, e)).
N o w * 1 7 9 a is im m e d ia te from * 1 7 4 a (w ith r as th e t).
T o estab lish (v) for *(17 9 ), consider a n y p air of n um bers a an d 6 , an d
le t r = rm (a, b) an d q[a, b] . C a s e 1 : b ^ 0. N o w a =
= 6g + r ; hence
b y (A), a= 6g-f-r. A ls o r < b \ hence b y *(16 5 ), b r < 6 . B y & -in tro d .
(or (C)), b a = 6 q + r & r < 6 . B u t q < a; so b y (E),
b 3^(gr<a & a=bq-\-r & r < 6 ). By V -introd.,
b 3 g(y< a & a = 6 g + r & r < 6 ) V ( 6 = 0 & r = a ) , i.e. b S (a , 6 , r). N o w
let e r. T h e n e < b (so b y *(16 5 ),
b e a n y n um ber < b e < 6 ). F o r a n y
n u m b er p, a ^ bp-\-e (since q , r is th e o n ly pair of num bers w ith r < b
such th a t a = b q + r ); so b y (A), b “ i a = 6 p + e . T h e n c e b y (C),
b ~ i ( a = 6 p + e & e < 6 ). In p a rticu la r th is h o ld s for p — 0, 1 , . . . , a \
so b y (E), b " i 3 ^ ( ^ < a & a — b q -\-e & e c b ) . B u t b ^ 0; h en ce b y
*(16 4 ), b “ i 6 = 0 . B y *(16 4 ), either b e = a or b i e = a . C o m b in in g
these results b y (C),
b -i(3q(q<a& a=bq+e& e<b)V (b=0&e=a))f i.e. b ~ iS (a, 6, c).
T h is w as for a n y e < r, i.e. it h olds for e = 0, 1 , . . . , r —1; so b y (E),
b V f(e< r D - i S ( « , 6 , e)). F ro m th is an d b S (a , 6, r) b y & -in tr o d .,
b S (a , 6, r) & V e(e<r 3 - i S ( a , 6, e)), i.e. b R(a, 6, r), as w as to b e
show n. Case 2 : 6 = 0 . Sim ilarly. (Su m m arizin g: F ro m th e form of
R(tf, b y r), b y (A), *(16 5 ), (C) an d (E), R (^, b f r) is num eralw ise d ecid a
b le. By fo rm al steps p arallelin g th e in terp reta tio n , we v e r ify th a t
R (a , 6 , r) is p ro v a b le ra th er th a n re fu ta b le w h en r= rm (a, b).)
N o w (vi) for *(179 ) follow s b y su b stitu tin g a n y n um erals a , 6 for a, b
in * 1 7 9 a , an d u sin g (v).
The function
*(180) rm(c, (i'-d)')
is num eralwise represented by a form ula B (c, d, i, w) such that, for any
num eral w :
*18 0 a. B (c , d, i, w ) b 3 !w B (c, d, i, w).
P roof. L e t “ c ” , “ d ” , “ i ” , fV ;, " w " d en ote c, d , i , q, r, r e s p e c tiv e ly ;

an d let B (c, d, i, w) b e R (c, (i'-d)', w). U se * 1 7 9 a , su b stitu tio n (* 66 ), *(179 )
an d (B).
T h e foregoing tre a tm e n t does n o t g iv e th e stron ger p r o p e r ty t h a t

b 3 !w P ( x 1, , . . , x n, w) for th e form ulas num eralw ise represen tin g

[a/b], rm (a, b) a n d rm (c, {i'-d)'). H o w e v e r u sin g * 1 4 6 a an d *14 6 b (w ith
* 1 2 3 , *14 0 , * 1 4 1 , * 1 4 2 a an d *14 3 a ), th is can be estab lish ed also, a n d
sim pler represen tin g form ulas can be g iv e n e q u iv a le n t to th e form er.
*178 b . h Q {a, b> q) ~ 3r ( a = b q + r & r < h ) V b = q = 0.
*179 b. b K (a, b, r) ~ lq { a = b q -\-r & r < b ) V (b = 0 & r= a ).
*18 0 b . b B (c, d, i, w) ~ 3 v ( c = ( i '* d ) '* v + w & w < ( i'-d ) ') .
*178c. b 3 lq Q (a ,b ,q ). *17 9 e . b 3!rR(tf, b, r).
*180c. b 3!wB( c, d, i, w).
L emma 18a. The results and (A) — (E) of this section,
*(164 ) — *18 0 c
excepting * 169 and * 174 a when t is not a num eral, * 174 b, * 178b, c, * 179 b , c
and *18 0 b , c, hold good for the form al system lacking A xiom Schema 13 but
having as additional particular number-theoretic axiom s the form ulas of
* 1 0 4 — * 1 0 7 and of * 1 3 7 or * 1 3 6 (“ R o b in so n 's sy ste m " ).
§ 42. GddePs theorem. F ro m a result o f P resb u rger 1 9 3 0 , m e ta -

m a th e m a tic a l proofs of c o n sisten cy an d com pleten ess, an d a decision
procedure, can be g iv e n for th e form al sy ste m w ith th e fo rm atio n rule
a n d a x io m s for • o m itted . (Cf. E x a m p le 2 § 79. Presb u rger d eals w ith
a classical sy ste m o f th e arith m e tic o f th e integers, b u t H ilb e rt an d
B e m a y s 1934 pp. 359 ff. a d a p t his m eth o d to essen tially th e present clas
sical sy ste m , an d J o a n R o ss has verified th a t th e a d a p ta tio n w orks for
th e in tu itio n is tic s y ste m as well.)
F o r th e fu ll sy ste m (or sy ste m s e ssen tially e q u iv a le n t to it), th ese
q u estion s p ro v e d to b e v e r y re fracto ry. C o n sisten c y proofs b y A c k e rm a n n
1924-5 an d v o n N e u m a n n 19 2 7 lead to th e result th a t th e sy ste m is
co n sisten t u nder th e restrictio n on th e use o f th e in d u ctio n p o stu la te
(A x io m S c h e m a 13) to th e case t h a t th e in d u ctio n v a ria b le x does n o t
o ccu r free w ith in th e scope o f a q u a n tifie r of th e in d u ctio n fo rm u la
A (x ). (Cf. T h eo rem 55 § 79. T h e restrictio n ex clu d es e.g. our proofs o f
* 1 0 5 , * 1 3 6 an d *14 8.)
T h is situ a tio n w as illu m in a te d in 19 3 1 b y th e ap p earan ce o f tw o
rem a rk a b le theorem s of G o d el “ on fo rm a lly u n d ecid a b le propositions
o f P rin cip ia M a th e m a tic a an d re lated s y ste m s" . W e d esign ate th e first
o f th ese theorem s, w h ich en tails th e o th er as corollary, as “ G o d e l's
th e o r e m ", a lth o u g h it is o n ly one of a series o f im p o rta n t co n tribu tio n s b y
its autho r. T h e se tw o theorem s, w h ich b eca m e th e m o st w id e ly n o ted in th e
su b je c t, b ear on th e w hole p rogram an d p h ilo so p h y o f m e ta m a th e m a tics.
T h e m e ta m a th e m a tic a l results presented th u s far in th is b o o k w ere
reach ed alo n g p a th s m ore or less su ggested b y th e in terp retatio n o f th e
§42 GODEI/S THEOREM 205
sy ste m . T h ese results of G o d el are o b ta in e d b y a k in d of m e ta m a th e m a ti-
ca l reasonin g w h ic h goes m ore d e e p ly in to th e stru ctu re of th e fo rm al
sy ste m as a sy ste m of o b jects.
A s is set fo rth in § 16, th e o b je c ts o f th e form al sy ste m w h ich w e s tu d y
are va rio u s form al sym b o ls, form al expressions (i.e. fin ite sequences of
fo rm al sym bols), a n d fin ite sequences of form al expressions. T h e re are an
en u m erable in fin ity of form al sy m b o ls g iv e n a t th e o u tset. H en ce, b y th e
m eth o d s o f § 1 , th e form al o b je c ts form an enu m erable class. B y sp ec
ify in g a p a rticu la r en u m eration o f th em , an d le ttin g our m e ta m a th e m a t-
ica l s ta te m e n ts refer to th e ind ices in th e en u m eratio n in ste a d o f to th e
o b je c ts en u m erated , m e ta m a th e m a tic s becom es a b ran ch o f n u m b er
th e o ry. T h e re w ith , th e p o ssib ility appears th a t th e fo rm al sy ste m
sh ould co n ta in form ulas w h ich , w h en considered in th e lig h t o f th e
enu m eration, express propositions o f its ow n m e ta m a th e m a tics.
I t w ill appear, on further s tu d y , t h a t th is p o ssib ility ca n b e e x p lo ite d ,
an d w ith th e use of C a n to r's d ia go n a l m eth o d ( § 2 ), a closed fo rm u la A
ca n b e fo u n d w h ich , in terp reted b y a person w h o kn ow s th is en u m eration ,
asserts its ow n u n p r o v a b ility .
T h is form p la A bears an a n a lo g y to th e proposition o f th e E p im e n id e s
p a ra d o x (§ 1 1 ). B u t n o w there is a w a y of escape from th e p a ra d o x. B y
th e co n stru ctio n of A ,
( 1) A m eans t h a t A is u n p ro vab le.
L e t us assum e, as w e hope is th e case, th a t form ulas w h ic h express false
propositions are u n p ro va b le in th e sy ste m , i.e.
(2 ) false form ulas are u n p ro vab le.
N o w th e form u la A ca n n o t b e false, becau se b y ( 1 ) t h a t w o u ld m ean
t h a t it is n o t u n p ro vab le, co n tra d ic tin g (2). B u t A can b e true, p ro v id e d
it is u n p ro vab le. In d eed th is m u st b e th e case. F o r assu m in g t h a t A is
p ro v ab le, b y ( 1 ) A is false, an d hence b y (2) u n p ro vab le. B y (in tu itive)
redu ctio ad absurdum , th is g iv e s th a t A is u n p ro v a b le , w h ereu p on b y
( 1 ) also A is true. T h u s th e sy ste m is in co m p lete in th e sense t h a t it fails
to afford a proof of e v e r y form u la w h ich is tru e und er th e in terp reta tio n
(if (2 ) is so, or if a t least th e p a rticu la r form ula A is u n p ro va b le if false).
T h e n e g a tio n - i A of th e fo rm u la is also u n p ro vab le. F o r A is tru e ;
hence ~ iA is fa ls e ; an d b y (2 ), - i A is u n p ro vab le. So th e sy ste m is in
co m p le te also in th e sim ple sense d efin ed m e ta m a th e m a tic a lly in th e la st
section (if (2 ) is so, or if a t le a st th e p a rticu la r form u las A an d - i A are
each u n p ro va b le if false).
T h e a b o v e is of course o n ly a p relim in ary h euristic ac co u n t of G o d e l's
reasonin g. B eca u se of th e n a tu re o f th is in tu itiv e argu m en t, w h ich sk irts
so close to an d y e t m isses a p ara d o x, it is im p o rta n t th a t th e s tr ic tly

fin ita r y m e ta m a th e m a tic a l proof of G o d e l’s theorem sh ould be a p p re c ia t
ed. W h e n th is m e ta m a th e m a tic a l proof is ex am in e d in fu ll d eta il, it is
seen to b e of th e n atu re of o rd in ary m ath em atics. In fa c t, if w e chose
to m a k e our m e ta m a th e m a tic s a p a rt of n um ber th e o r y (now in form al
ra th er th a n fo rm al n um ber th eory) b y ta lk in g a b o u t th e ind ices in th e
en u m eration , an d if w e ignore th e in terp retatio n s of th e o b je c t sy ste m
(now a sy ste m of num bers), th e theorem becom es a p rop osition of or
d in a ry e lem en ta ry n u m ber th eo ry. I ts proof, w hile e x c e e d in g ly lo n g an d
tediou s in th ese term s, is n o t open to a n y o b jectio n w h ich w o u ld n o t e q u a lly
in v o lv e p arts of tra d itio n a l m a th e m a tics w h ich h a v e been h eld m ost secure.
W e can g iv e th e rigorous m e ta m a th e m a tic a l proof now , b y b orrow in g
one lem m a from results of th e n e x t tw o ch apters. O u r n u m berin g of th e
lem m as an d theorem s corresponds to th e lo gical order.
In m a k in g use of th e id ea of en u m eratin g th e form al o b jects, p ra c tic a l
con sid eration s d ic ta te th a t th e indices of fo rm al o b je c ts sh ould be
correlated to th e o b je cts b y as sim ple a rule as possible. W e can m o d ify
th e a b o v e h euristic a rgu m en t (inessentially) b y usin g, rath er th a n an
en u m eration in th e usu al sense, an en u m eration w ith ga p s in th e n a tu ra l
num bers, i.e. a correlation of d istin ct n a tu ra l num bers to th e d istin c t
form al o b jects, n o t all of th e n a tu ra l num bers b ein g used in th e cor
relation. W e ca ll th is a G o d el num berin g, an d th e correlated n u m ber of
a form al o b je c t its G o d el num ber. (Som etim es sep arate G o d el n u m berin gs
are g iv e n of th e fo rm al sym b o ls, of th e form al expressions, an d of th e
fin ite sequences of form al expressions. I f t h a t is done, th e n w h en one
speaks of a n um ber as th e G o d el n u m ber of a sy m b o l, or of an expression,
or of a sequence of expressions, in each case a d ifferen t correlation is b ein g
referred to.)
R e la tiv e to a n y sp ecified G o d el num berin g, for a n y n w h ich is th e G o d el
n u m ber of a form ula, le t “ A n” d esign ate th e form ula. (For oth er n ’s,
w e need n o t define A w.) W e m a y w rite th is form ula A n also as “ A n{a)” f
sh o w in g th e free v a ria b le a for use w ith our su b stitu tio n n o ta tio n (§ 18).
L emma 21. There is a Godel num bering of the form al objects such that
the predicates A (a, b) and B (a, c) defined as follows are num eralwise ex
pressible ( §41 ) in the form al system .
A {a f b): a is th e G o d el n u m b er o f a form u la (n am ely A a(^)), an d b is th e
G o d el n um ber of a proof of th e form ula A a(a).
B (a, c): a is th e G o d el n u m b er o f a form ula (n am ely A a(a)), an d c is th e

G o d el n um ber o f a proof of th e form ula - i A a(a).
§42 GODEL’S THEOREM 207
N o w le t A (a, b) (a, c) b e p a rticu la r form ulas w h ich n um eralw ise
an d B
express th e p red icates A {a, b) an d B (a ,c), re sp e ctiv e ly , for th e G o d el
n u m b erin g g iv e n b y th e lem m a. T h e tw o form ulas A (a, b ) an d B (a, c)
co u ld a c tu a lly b e e x h ib ite d , a fte r w e h a v e th e proof of th e lem m a (to b e
co m p le te d in § 52).
Consider th e form u la mmyA(a, b) w hich a an d no oth er
co n tain s
v a ria b le free. T h is form u la has a G o d el num ber, ca ll it p, an d is th e n th e
sam e as th e fo rm u la w h ich w e h a v e d esign ate d “ A 9{a)” . N o w consider
th e form u la A P(p), i.e.
K ip)- V/> - i A ( p , b),
w h ic h co n ta in s no v a ria b le free. N o te th a t w e h a v e used C a n to r ’s d iago n a l
m e th o d in s u b stitu tin g th e nu m eral p for a in A p(a) to o b ta in th is form ula.
T o relate th is to th e p relim in ary h eu ristic ou tlin e, w e can in terpret
th e fo rm u la A v ip) from our p ersp e ctive of th e G o d el n u m b erin g as
exp ressin g th e proposition t h a t A P(p) is u n p ro vab le, i.e. it is a form u la
A w h ich asserts its ow n u n p r o v a b ility .
In th e m e ta m a th e m a tic a l argu m en t, th e assu m p tion s of th e h eu ristic
argu m e n t th a t th e sy ste m sh ould n o t allo w th e proof of either o f th e
form ulas A or - i A if false w ill b e replaced b y m e ta m a th e m a tic a l e q u iv a
lents. F o r th e u n p r o v a b ility o f A if false, th is e q u iv a le n t w ill b e th e (simple)
co n sisten cy of th e sy ste m (§28). F o r th e u n p r o v a b ility of - i A if false,
w e sh all need a stronger co n d itio n ca lled ‘ co-consistency’ w h ic h w e sh all
now define.
T h e fo rm al sy ste m (or a sy ste m w ith sim ilar fo rm atio n rules) is said
to b e co -consistent, if for no v a ria b le x an d form u la A (x ) are all o f th e
fo llo w in g tru e:
b A ( 0), b A( l ) , b A ( 2 ), ...; b -« V x A (x )
(or in oth er w o rd s if n o t b o th b A (n ) for e v e r y n a tu r a l n u m b er n an d
b " iV x A ( x ) ) . In th e c o n tr a ry case t h a t for som e x an d A (x ) all o f A(0),
A( l ) , A (2), ... an d also -n V x A (x ) are p ro v a b le , th e sy ste m is (^-in
consistent.
N o te t h a t co-consistency im p lies sim ple co n sisten cy. F o r if A b e a n y
p ro v a b le form ula co n ta in in g no free v a ria b les, w ritin g it as “ A ( x ) ”
w here x is a v a ria b le , all o f A ( 0 ), A ( l ) , A ( 2 ), . . . are p ro v a b le (under our
su b stitu tio n n o ta tio n § 18, eaeh of these is s im p ly A itself); an d h en ce
if th e s y ste m is co-consistent, - i V x A ( x ) is an e x a m p le of an u n p ro v a b le
§28).
form ula (cf.
T heorem 28. I f the number-theoretic form al system is {sim ply) consis

tent, then not b K i p ) ; and if the system is co-consistent , then not b ^ A P{ p ) .
Thus, if the system is to-consistent, then it is (sim ply) incom plete, with
A P(p) as an exam ple of an undecidable form ula. (G od el’s theorem , in th e
origin al form.)
P roof th a t, if th e sy ste m is con sistent, th en n o t b A p (p )- S u ppose

(for in tu itiv e red u ctio ad absurdum ) th a t b A P(p), i.e. suppose t h a t
A P(p) is p ro vab le. T h e n there is a proof of i t , le t th e G o d el n u m b er of
th is proof be k. Then A (p , k) is true. H en ce, since A (a, b) w as in tro d u ced
u n d er th e lem m a as a form u la w h ich num eralw ise expresses A (a, b),
bA ( p , k). B y 3-in trod ., b 3 b A (p ,b ). T h en c e b y *83a, b - i A ( p , b).
B u t th is is b ~ i A P(p). T h is, w ith our assu m p tion th a t b A P(p), con
tra d ic ts th e h yp o th esis th a t th e sy ste m is con sistent. T h erefo re b y re
d u ctio ad ab surdum , n o t bA P(p), as w as to be show n. (W e co u ld also
h a v e co n tra d icte d th e c o n siste n c y b y usin g V -elim . to infer b “iA(p, k)
from bA P(p).)
P roof th a t, if th e sy ste m is oo-consistent (and hence also con sistent),
th en n o t b A P(p). B y th e co n sisten cy an d th e first p a rt of th e theorem ,
A AP) is n o t p ro vab le. H en ce each of th e n a tu ra l num bers 0, 1 , 2, . . . is
not th e G o d el n u m b er of a proof of A P(p); i.e. A (p , 0 ), A (p , 1 ), A (p , 2),
... are all false. H en ce, since A (a, b) num eralw ise expresses A (a, b),
b- i A ( p , 0 ), b ~ iA (p, 1), b“"iA (p, 2 ) , ___B y th e (^-consistency, th e n
not b -i A (p , b). B u t th is is n o t b A P(p), w h ich w as to b e shown.
W e h a v e g iv e n th e origin al G o d el form of th e theorem first, as th e
proof is in tu itiv e ly sim pler an d follow s th e h eu ristic outlin e. R osser 1936
h as show n, h o w ever, th a t b y usin g a s lig h tly m ore co m p lic a te d e x a m p le
of an u n d ecidable form ula, the h yp o th e sis o f co-consistency ca n b e dis
pensed w ith , an d th e incom pleteness p ro v e d from th e (simple) co n
siste n c y alone. Consider th e form ula V b [-iA (a , b) V 3c (c < b & B (a, c))].
T h is has a G o d el n um ber, c a ll it q. N o w consider th e fo rm u la A Q(q), i.e.
A q(q ): Vb [-1A ( g , b) V 3c(c < b & B ( g , c ))].
W e ca n in terpret th e form u la A a(g) from our p ersp e ctiv e of th e G o d el

n u m b erin g as assertin g t h a t to a n y proof of A q(q) there e x ists a proof of
“ i A a(g) w ith an eq u a l or sm aller G o d el num ber, w h ic h under th e h y
po th esis of sim ple co n siste n cy im plies t h a t A a(g) is u n p ro vab le.
T heorem 29. I f the number-theoretic form al system is (sim ply) con

sistent, then neither bA g(g) nor b A a( g ) ; i.e. if the system is consistent,
then it is (sim ply) incom plete, with A q(q) as an undecidable form ula.
(R o sser's form of G o d e l's theorem .)
§42 godel ’ s theorem 209
P roof that, if the system is consistent, then not |- Aa(q). Suppose
that h Aa(q). As before (using q instead of p), A (q, k). Also, under our
hypothesis of consistency, the assumption that |- A„(qf) implies that
not [- “iA,(qf), i.e. -iA „(q) is unprovable. Hence, in particular, each of
B(q, 0), B(q, 1), .... B(q, k) is false. Since B {a, b) numeralwise expresses
B(a,b), therefore b -iB(qr, 0), (- —iB(qr, 1), . . \- -»B{q,k). Hence
by *166a, b Vc(c<fe D -iB(qr, c)). This with b A(q, k) gives by &- and
3-introd., b 3/>[A(qr, b) & 'ic{c<,b D -i B(q, c))]. Thence by *58b and
*86, b 3b[A(q, b) &-*3c(c<:b &B(q, c))]. Thence by *57b (and *70),
b 3/>-i[-iA(qr, b) V Bc(c<b & B(q, c))]. Thence by *85a,
b “iV^[-iA(qr, b) V 3c(c<,b & B(g, c))]. But this is b “lAj(qr). Hence as
before, not b AQ(q), as was to be shown.
P roof that, if the system is consistent, then not b As(qr). Suppose
that b ~iA„(<jf), i.e. that ~\A„(q) is provable. Then there is a proof of it;
let the Godel number of this proof be k. Then B(q, k) is true. Hence
1- B(qr, k). By *16 8 , h Vb[b>k 3 3c(c <^b & B(q, c))]. Also as before
(with q instead of p)t h "iA(q, 0), b ">A(q, 1), b -iA(q, k —1).
By *16 6 , b Vb[b<k 3 -iA(q, b)]. Now by * 1 6 9 , b V/>[-iA(q, b) V
3c(c^^ & B(q, c))]. But this is b AQ(q), Hence not b nA Q(q), as was
to be shown.
Observe that we have not shown outright that A P(p)t -iAp(p), Aq(q),
“i A0(q) are unprovable, but only that if the system is (simply) consistent,
A P(p)t Aq(q), -I AQ(q) are unprovable, and if the system is ^-consistent,
-iAp(p) is unprovable.
Consider our demonstration that A P(p) is unprovable, if the system
is consistent. If a demonstration of the consistency of the system were
now supplied, prefixing it to the former would complete a demonstration
that AP(p) is unprovable.
Supposing such a demonstration that A P(p) is unprovable to exist, we
could, using the representation of the formal objects by Godel numbers,
express it as a demonstration in informal number theory. We now ask
whether the latter demonstration could be formalized in the system.
In formalizing it, the formula Aj,(p) would itself be the formalized
statement of what is demonstrated, i.e. that A„(p) is unprovable. Thus
a formalized demonstration that AP(p) is unprovable would be a formal
proof of A P(p). By Theorem 28, such a proof cannot exist if the system is
consistent.
Thus, if we had an informal demonstration that A ^ ) is unprovable,
the demonstration would be incapable of being formalized within the
sy ste m , if th e sy ste m is con sisten t. T h e supposed inform al d em o n stratio n

w as to consist of tw o p arts, first a supposed proof of th e co n sisten cy of
th e sy ste m , an d second th e proof w e h a v e a lre a d y g iv e n (for th e first h a lf
of T h eo rem 28) th a t A v(p) is u n p ro va b le if th e sy ste m is con sistent.
B y sh ow in g t h a t th e second p a rt a c tu a lly can b e form alized in th e
sy ste m , w e h a v e a m e th o d of sh ow in g th a t th e first ca n n o t b e if th e
sy ste m is con sisten t. T h is g iv e s th e n e x t theorem . W e sh all re ca p itu la te
th is argu m e n t before s ta tin g th e theorem .
T h e assertion th a t th e sy ste m is (sim ply) co n sisten t ca n b e expressed
in th e form al sy ste m v ia th e G o d el n u m berin g v a rio u sly. L e t C (a , b) be
(D (a, c) be) th e p re d ic a te : a is th e G o d el num ber of a form ula, n a m e ly
A tt, an d b is (c is) th e G o d el n u m b er o f a proof of A a (of - i A a). T h ere
are form ulas C(a, b) an d D (a, c) exp ressin g C(a, b) an d D (a, c), respec
t iv e l y (§52). T h e origin al d efin itio n of co n sisten cy in § 2 8 is th en rendered
d ire c tly in to th e form alism b y th e form ula - \3a[3bC (a , b) & 3cD (a, c)].
B y th e second version of th e d efinition, an d th e fa c t t h a t -il=0 is
p ro v a b le (A x. 15), th e sy ste m is con sisten t, if an d o n ly if th e p articu lar
form u la 1 = 0 is u n p ro vab le. L e t r b e th e G o d el n u m ber of th is form ula.
Then A r(r) is th e sam e form ula, an d co n sisten cy is expressed by
- i 3 i>A(r, b) or V i> - iA ( r , b). L e t us ca ll one of these form ulas, a t our
preference, “ C o n sis” .
T h e assertion th a t A ^ ) is u n p ro vab le is expressed, v ia th e G o d el
n u m berin g, b y - i A (p , b )} w h ich is A v (P)-
T h e in tu itiv e d em o n stratio n of th e first h a lf of T h eo rem 28 is a d em
o n stra tio n th a t
(I) {the sy ste m is con sisten t} im plies { A P(p) is un p ro vab le}.
I t is now proposed t h a t th e entire m e ta m a th e m a tic a l d em o n stratio n of

(I) sh ould b e form alized in th e sy ste m , u sin g th e G o d el n um berin g, so
t h a t w e sh ould th en h a v e
(II) h Consis D A P(p).
N o w assum e m e ta m a th e m a tic a lly th a t b Consis. T h e n from (II) w e
sh ould h a v e by D -e lim in a tio n , b A j,(p). B y T h eo rem 28, th is is im
possible, if th e sy ste m is con sisten t. B y m e ta m a th e m a tic a l re d u ctio ad
ab su rd u m , th is w o u ld g iv e th e fo llo w in g theorem . (Th e proof cou ld
also b e b ased on T h eo rem 29, since A (q , b) D A Q(q).)
b -i
T heorem 30. I f the number-theoretic form al system is {sim ply) con

sistent , then not b C on sis; i.e. if the system is consistent , then there is no
consistency proof for it by methods form alizable in the system . (G odeFs
secon d theorem .)
§42 godel’s theorem 211
T h e proofs of T h eo rem s 28 an d 29 w ill b e co m p le te w h en w e h a v e
estab lish ed L e m m a 21 in C h a p te r X . F o r th e proof of T h eo rem 30, there
w ill rem ain th e g a p to b e filled in p assin g from (I) to (II). T h is is an
exercise in form alizin g an in fo r m a lly g iv e n proof of con siderable le n g th ,
w h ich w e sh all n o t ta k e th e space to ca rry th ro u gh in th is b ook.
H ilb e rt a n d B e rn a y s 1939 ca rry it o u t for a certain form al sy ste m Z u
(cf. pp. 283 ff., e sp e c ia lly pp. 306— 324), an d th en ce infer (pp. 324— 328)
th a t th e theorem h olds also for a sy ste m Z w h ich differs from our classical
n u m b er-th eo retic sy ste m in in essen tial respects (m ainly, in th e use of
predicate va riab les, cf. end § 37, an d in th e e q u a lity p o stu lates, cf. § 73).
Thence Theorem 30 holds for our classical s y s te m ; an d it can b e inferred
to h old for our in tu itio n istic sy ste m (at least w hen Consis is Vb - i A ( r , b)
or -i3 i> A (r, b), an d A (a, b) is s u ita b ly chosen) b y use of G o d el 1932-3
(cf. Theorem 60 (b2) § 81 an d N elso n 19 4 7 pp. 326— 327).
I t m a y b e rem arked th a t this exercise is required o n ly for th e sak e of
u sin g as th e form u la C onsis a d irect fo rm alizatio n of th e original d e f
in itio n of (simple) co n sisten cy or a close e q u iva len t. I n tu itiv e ly A p(p)
itself expresses an e q u iv a le n t, v ia th e lon g in tu itiv e proof of G o d e l's
theorem . F o | b y T h eo rem 28, if th e sy ste m is con sisten t, A P(p) is u n -
p ro v ab le, an d b y § 28, if A P(p) is u n p ro vab le, th e sy ste m is co n siste n t;
an d th e u n p r o v a b ility of A P(p) is expressed b y A P(p).
W h a t is th e sign ifican ce of these results for th e proposed p rogram
of m e ta m a th e m a tic s? We c e rta in ly hope th a t th e form al sy s te m is
con sistent. I f so, th en b y T h eo rem 29 it is n ecessarily incom plete. W e
h a v e n o t succeeded in fo rm alizin g inform al n u m ber th e o r y c o m p le te ly
e x p lic itly , so th a t each proposition or its n eg atio n is a con sequence b y
e x p lic itly sta te d rules of e x p lic itly sta te d ax io m s (§ 15).
T a k in g th e su p p osition of (simple) co n sisten cy in its fin ita r y m ean in g,
i.e. if th e c o n sisten cy is ca p a b le of b ein g p ro v ed m e ta m a th e m a ti-
c a lly , th e n th e sy ste m is in co m p lete also in th e sense th a t there are
expressible in it propositions true on fin ita r y grounds, b u t u n p ro vab le
fo rm ally , three such propositions b ein g expressed b y A P(p), A Q(q) an d
Consis.
F o r th e problem of p ro v in g co n sisten cy m e ta m a th e m a tic a lly , T h eorem
30 h as th e con sequence t h a t th e m eth o d s w h ich w e m u st tru st in th e
proof m u st include som e w h ich lie ou tsid e th e co llectio n of th e m eth o d s
form alized in th e sy ste m . T h is is not a priori in c o m p a tib le w ith th e oth er
requirem en t th a t th e m eth o d s n o t includ e all w h ich lie inside, our m is
tru st of som e of w h ich w as th e occasion for th e form alist pro ject o f a t
te m p tin g the proof. I t does pose th e ch allen ge to th e m e ta m a th e m a ticia n
to bring to bear methods of finitary proof more powerful than those

commonly used in elementary number theory.
There is a further implication for the completeness and consistency
problems. Suppose our system is (simply) consistent. Then, as in the
proof of the second half of Theorem 28, AP(p) is unprovable, but
h “"iA(p, 0), b ~iA(p, 1), b ""iA(p, 2), .... Thus we have a formula
A(x) (namely -tA(p, b)) such that b A(0), b A(l), b A(2), . . ., but not
b VxA(x) (which is AP(p)). Tarski 1933a gives the name ^-incompleteness
to this situation. If in this situation b ~tVxA(x) (which is -1 Ap(p)), the
system would be co-inconsistent. The discovery that a system may be
^-incomplete reveals the possibility that it may be co-inconsistent without
being simply inconsistent. The system obtained from our system by taking
“i Ap(p) as a new axiom definitely is, under the assumption that ours
is simply consistent. To see that this system is simply consistent, we
observe that if B and -iB were provable in it, then in the original system
B and -iB would be deducible from -iAp(p) (end § 20); hence by
-u-in trod., —1—1A P(p) would be provable; and by “i-elim. (or intuitionisti-
cally by IVa2 3 IVax from Corollary Theorem 17 § 35), AP(p) would be
provable, contradicting Theorem 28. There are evidently definable still
higher orders of completeness and consistency.
We don’t want our system to be co-inconsistent, even if it is consistent.
In particular, if the simple consistency were provable metamathematical-
ly, then the formula -iAp(p) would under the interpretation express a
proposition contradicting one that is true on finitary grounds; and in
case - 1AP(p) were provable, following Hilbert and Bernays (1939 p. 282)
we should call the system externally inconsistent, i.e. inconsistent with
respect to the finitary interpretation. Thus a proof of simple consistency
alone would not secure the formalized mathematics against the possi
bility of establishing something intuitively false. (This was first noted
by Finsler 1926 in connection with a system that is not formal in our
sense. Also cf. Godel ig 3i- 2a.)
Suppose our system is simply consistent. For brevity, call a formula
“true” (“false”), if it expresses a true (false) proposition under the inter
pretation. Let B(x) be a formula which numeralwise expresses a pred
icate B(x), which it shall express under the interpretation. Then for each
natural number x, the formula B(x) is provable if and only if it is true.
The formula VxB(x) is true if it is provable (in view of V-elim.); but not
in general conversely (e.g. AP(p)). The formula 3xB(x) is provable if it
is true (in view of 3-introd.); but we have not shown the converse meta-
mathematically (e.g. 3bA(p, b), from which -iAP(p) follows by *83a,
§42 godel’s theorem 213
and which we thus know to be false, but do not know without assuming
also ^-consistency to be unprovable).
Thus (cf. § 14) a proof of (simple) consistency would be sufficient to
justify the use of our classical system for proving “real” statements
intuitionistically by an excursion through the “ideal”, for “real” state
ments of the forms B(x) and VxB(x), but not so far as we have yet shown
for those of the form 3xB(x).
Suppose that we add to the system AP(p), AQ(q) or Consis as a new
axiom, and iterate the whole process, to obtain a succession of systems.
It can be shown that if these systems are consistent, then the class of
provable formulas of the form VxB(x) is successively enlarged. G o d el
1931-2 states that the same is true of the systems obtained by admitting
successively higher types of variables (at least, supposing co-consistency).
Except that we lack proofs of the appropriate consistency properties,
this shows (cf. § 14) that successively higher theoretical constructions do
add to the class of “real” statements of the original sort which are
comprised. Godel 1936 states also that in the higher systems infinitely
many of the previously provable formulas have very much shorter proofs.
Thus far we have GodeFs theorem only for our particular formal
system (except for the last remarks, which refer to a succession of systems).
The question arises now whether it may not depend on some peculiarities
of the present formalization of logic, and might be avoided in some other.
In the next chapters, besides completing the proof of the required lemma
for Godel's theorem, we shall reach a standpoint from which we can discuss
these questions for formal systems in general, with the formal system
studied here as an example (§§ 60, 61).
PA R T III
R E C U R S IV E F U N C T IO N S
Chapter IX
P R IM IT IV E R E C U R S IV E F U N C T IO N S
§ 43. P r i m i t i v e r e c u r s i v e f u n c t io n s . T o estab lish th e lem m a for

G o d el's theorem , w e shall d eve lo p an in tu itiv e th e o ry a b o u t a certain
class of n u m b er-th eo retic fu n ctio n s an d predicates, e v e n tu a lly sh ow in g
th a t e v e r y p red icate of th e class is num eral w ise expressible in th e form al
sy ste m (§ 49), an d th a t th e tw o pred icates A (a, b) an d B (a , c) of th e lem m a
b elo n g to th e class (§ 52). T h is w ill sav e us m u ch of th e lab o r of a step b y
step d eve lo p m en t w ith in th e form al system .
E x c e p t for th e a p p lica tio n ju s t described, th e th e o r y of these fu n ctio n s
an d pred icates w ill b e d eve lo p ed in d ep en d e n tly of th e form al sy ste m of
th e precedin g ch apters. In th is th eo ry, as in m e ta m a th e m a tics, w e sh all
use o n ly fin ita r y m ethods.
T h e series of th e n a tu ra l num bers
0, O', 0", O'", ....

or 0, 1, 2, 3, . . we described as th e class of th e o b je c ts gen erated from
one p rim itiv e o b je c t 0 b y m eans of one p rim itive op eratio n ' or + 1 .
T h is co n stitu te s an in d u c tiv e d efin ition of th e class of th e n a tu ral n um bers
(§ 6).
P ro o f b y in d u ctio n , as a m eth o d of p ro v in g a theorem T (y) for all n a t
ural num bers y, corresponds im m e d ia te ly to th is m ode of g en eratin g th e
num bers (§ 7). D efinition by induction (not to b e con fu sed w ith 'in d u c tiv e
d efin itio n ', §§ 6 , 53), also called recursive definition , is th e an alo go u s
m e th o d of defin in g a n u m b er-th eo retic fu n ctio n <p(y) or p red icate P (y ).
F irst <p(0) or P ( 0 ) (the v a lu e of th e fu n ctio n or p red icate for 0 as argum ent)
is giv en . T h e n , for a n y n a tu ra l n u m ber y, cp(y')
P (y') (the n e x t v a lu e
or
a fter th a t for y) is expressed in term s of y an d <p(y) or P (y) (the v a lu e for y ) .
A n a lo g o u s ly , w e can co n clu de t h a t under th ese circu m stan ces th e v a lu e
9 (y) or P (y) of th e fu n ctio n or p red icate is d efined for e v e r y n a tu r a l
n u m ber y. F o r th e tw o p arts of th e d efin ition enable us, as w e gen erate
a n y n a tu ra l n um ber y, a t th e sam e tim e to determ ine th e v a lu e <p(y) or
P (y)-
217
2 18 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
T o ex am in e th is in m ore d e ta il, let us w rite th e pair of eq u atio n s
(i)
to express th e d efin itio n of a fu n ctio n <p(y) b y in d u ctio n on y , w here q

is a g iv e n n a tu r a l num ber, an d yjy> z) is a g iv e n n u m b er-th eo retic fu n ctio n
o f tw o va riab les.
T h e n for exam p le , th e v a lu e <p(4) is d eterm in ed thus. T o gen erate 4,
w e gen erate su cce ssiv e ly 0 , 1 , 2, 3, 4. B y th e first eq u a tio n , th e v a lu e
<p(0) sh all b e th e g iv e n n u m b er q; th en b y th e second eq u a tio n , th e v a lu e
9 ( 1) shall be x( 0 , f(0)), i.e. (using th e v a lu e 9 (0 ) a lre a d y found) x(0> ?)#
w h ic h (since x(y* z) is a g iv e n function ) is a g iv e n n u m b er; a ga in th e v a lu e
9 (2 ) shall be x0» ?(!)), i.e. x0> x(°>9))> the value ? ( 3) sh all b e x(2> ?(2)),
i.e. x(2, x ( U x ( 0 } ?))); a n d fin a lly the value 9 (4) shall be x(3, 9 (3))> i-e.
x(3>x(2>x0>x(°>?))))*
T h u s w e h a v e a process b y w hich, to each n a tu ra l n u m b er y, on th e
basis of th e gen eratio n of y in th e n a tu ral n u m b er sequence, a corre
sp o n d in g n u m ber 9 (y) is d eterm ined. Since a n u m ber <p(y) is th u s a sso ciat
ed w ith y, for each y, a p a rticu la r n u m b er-th eo retic fu n ctio n 9 is d efin ed
w ith th ese num bers 9 (y) as its re sp ective valu es.
T h is fu n ctio n 9 satisfies th e eq u a tio n s ( 1 ), w h en ( 1 ) are con sidered as
fu n ctio n a l eq u a tio n s in an u n k n o w n fu n ctio n 9 , sin ce e v e r y p a rticu la r
e q u a tio n com prised in ( 1 ) (n am ely, 9 (0 ) = ^ , 9 (0 ' ) = x ( 0 , 9 (0)), 9 ( 1 ') =
x ( l , 9 ( 1 )), . . . ) is satisfied in th e course of selectin g th e su ccessive n u m
bers 9 (0 ), 9 ( 1 ), 9 (2 ) , ___A ls o th is 9 is th e o n ly fu n ctio n sa tisfy in g
( 1 ) as fu n ctio n a l eq u a tio n s, since th e process b y w h ic h w e d eterm in ed th e
su ccessive num bers 9 (0 ), 9 ( 1 ), 9 (2 ), . . . from th e eq u a tio n s ( 1 ) ca n b e
in terp reted as sh o w in g t h a t a n y fu n ctio n 9 sa tisfy in g th e eq u a tio n s m u st
h a v e th e v a lu e s selected.
In o th er defin ition s b y in d u ctio n , th e fu n ctio n 9 d efin ed d ep en d s on
a d d itio n a l v a ria b les %%t. . . , x nt called param eters , w h ich h a v e fix e d v a lu e s
th ro u g h o u t th e in d u ctio n on y.
E xample 1. C onsider in tu itiv e ly th e eq u a tio n s
r 0+0 = a,
\ a+b' = (a + b y ,
w h ic h w e en cou n tered in th e form al sym b o lism as A x io m s 18 a n d 19.
T h e se define th e fu n ctio n a+b b y in d u ctio n on b, w ith a as param eter,
a n d ' as a p re v io u sly k n o w n fu n ctio n . T h e n th e eq u atio n s
a-0 = 0 ,
{ a-b' = (a 'b )+ a
§43 PRIMITIVE RECURSIVE FUNCTIONS 219
d efin e a*b b y in d u ctio n on b , w ith a-\-b as a kn o w n fu n c tio n ; an d
r a° = lf
\ ab' = ab-a
defin e ab by in d u ctio n on b, w ith a-b as a kn o w n fun ction .
A n ex am p le of a d efin ition of a p red ica te b y in d u ctio n w ill b e g iv e n

la te r (E x a m p le 2 § 4 5 ).
W h a t n u m b er-th eo retic fu n ctio n s are d efin ab le b y in d u ctio n ? T o m a k e
th is qu estion precise, w e m u st sp e cify w h a t fu n ctio n s are to b e ta k e n
as kn ow n in itia lly , an d w h a t operations, in clu d in g w h a t form s of d efin itio n
b y in d u ctio n , are to be allo w ed in d efin in g fu rth er fu n ction s.
W e shall now select th e sp ecification s w ith a v ie w to o b ta in in g fu n ctio n s
d efin ab le b y in d u ctio n in an ele m e n ta ry m anner. T h ese fu n ctio n s w ill b e
ca lled 'p rim itiv e recursive'.
E a c h of th e fo llo w in g eq u a tio n s an d sy ste m s of eq u a tio n s (I)— (V)
defines a n u m b er-th eo retic fu n ctio n <p, w h en n an d m are p o sitiv e integers,
i is an in teger such th a t 1 • * •»
Imp X are g iv e n n u m b er-th eo retic fu n ctio n s of th e in d ic a te d n u m bers of
variab les.
(I) ?(*) =
(II) ?(*!>• *») = ?•
(III) 9 = *<•
(IV ) 9 = <MXl(*l. • • • >x n)> • • • > • • • >x n))-
f 9(0) = q>
(Va)
l <p( / ) = x(y> <p(y))-
(Vb)
r 9 (0 , = li>(x 2’ • • • * X„),
I 9 (y ',x 2, . . . , x n) = x(y- < ?{y ,x 2l . - . , x n ) , x 2 > - - - > x n )-
((Va) co n stitu te s th e case of (V) for n = 1 , an d (Vb) for n > 1 .)
A fu n ctio n is p rim itive recursive , if it is d efin able b y a series o f a p
p licatio n s o f th ese fiv e operatio ns of definition.
T h is d efin ition ca n be g iv e n in m ore d etail, a n a lo go u sly to th e d e f
in ition of p ro v a b le form u la for th e form al sy ste m (§ 19), s a y u sin g th e
second version, as follows.
W e refer to th e a b o v e eq u a tio n s an d eq u a tio n pairs (I)— (V) as sch e
m ata . T h e y are an alogo u s to th e p o stu lates, w ith (I)— (III) in th e role of
a x io m sch e m a ta (or m ore s tr ic tly , (I) to a p a rticu la r axio m ), an d (IV ) an d
(V) in th e role of rules of inference.
A fu n ctio n <p is ca lled an in itia l function, if 9 satisfies E q u a tio n (I),
220 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
or E q u a tio n (II) for a p a rticu la r n an d q, or E q u a tio n (III) for a p a rticu la r

n and i.
A fu n ctio n 9 is ca lled an immediate dependent of oth er fu n ctio n s, if 9
satisfies E q u a tio n (IV ) for a p a rticu la r n an d m w ith %v . . . , Xm as th e
o th er fu n ctio n s, or E q u a tio n s (Va) for a p articu lar q w ith x as th e o th er
fu n ctio n , or E q u a tio n s (Vb) for a p articu lar n w ith <J/, x as th e o th er fu n c
tions.
A fu n ctio n 9 is called primitive recursive, if there is a fin ite sequence
9i»---,9* 1 ) of (occurrences of) fu n ction s (called a prim itive re
cursive description of 9 ), such th a t each fu n ctio n of th e sequence is eith e r
a n in itia l fu n ctio n , or an im m e d ia te d ep en d en t of preced in g fu n ctio n s
o f th e sequence, an d th e la st fu n ctio n 9 *. is th e fu n ctio n 9 .
§ 44. E x p l i c i t d e f in it io n . T h e first problem of th is ch a p te r is to

recogn ize as p rim itiv e recursive vario u s fu n ctio n s w h ich m a y a lre a d y b e
kn o w n to us in oth er w a y s. (A n alo go u sly, in s tu d y in g th e form al sy ste m ,
w e d ed u ced from th e a x io m s fu rth er form al theorem s, an d d eriv ed rules
as gen eral m eth o d s for fin d in g still others.)
T h e sch e m a ta h a v e been g iv e n ste re o ty p e d form s to sim p lify th e d ef
in itio n o f th e class of p rim itiv e recursive function s. In th e rem ainder of
th is section , w e sh all learn to telescope several a p p lica tio n s of them .
S c h em a (I) g iv e s th e successor function as one of th e in itia l fun ction s.
In th is co n n ection , w e d esign ate it as 5 . T h e in itia l fu n ctio n s g iv e n b y
S ch e m a (II) w e c a ll th e constant functions, an d w e d esign ate th em as C£.
T h e in itia l fu n ctio n s g iv e n b y S ch em a (III) w e call th e identity functions,
an d w e d esign ate th em as £/?.
S c h em a (IV ) w e ca ll th e sch em a of definition by substitution. T h e exp res
sion for th e am b igu o u s v a lu e of 9 is o b ta in ed b y su b stitu tin g expressions
for th e a m b ig u o u s v a lu e s of Xi»* * •> Xm for th e v a ria b les of $ (cf. § 10).
T h e fu n ctio n 9 d efin ed b y an ap p lica tio n of th is sch em a w e som etim es
w rite as S ^ , x i >; • Xm)-
An explicit definition of a fu n ctio n consists in g iv in g an expression for
its am b igu o u s v a lu e c o n stru cted s y n ta c tic a lly from its in d ep en den t
v a ria b le s (w ith no o th er va ria b le s occurring free) an d sy m b o ls for g iv e n
fu n ctio n s, co n sta n ts, operators, etc. In particu lar, we say th a t a
fu n ctio n 9 definable explicitly from (or is explicit in) fu n ctio n s
is
a n d co n sta n ts ql9. . . 9q99 if an expression for its am b igu o u s
v a lu e <p(xv . . . , x n) can be g iv e n in term s of th e va riab les x Jf. . . , x n, th e
c o n sta n ts ql9. . - 9q» an d th e fu n ctio n s (cf. E x a m p le 2 § 10).
In th is case 9 can b e o b ta in e d from ^ , . . . , ^ 1 b y a series of a p p lica tio n s
§44 EXPLICIT DEFINITION 221
of S c h e m a ta (II) — (IV ). F o r S c h e m a ta (II) an d (III) in tro d u ce each co n
s ta n t an d each of th e v a riab les % , . . . , x n as a fu n ctio n of all th e v a ria b le s
x v . . . , x n ; an d th en the su b stitu tio n s used in b u ild in g u p th e expression
for th e am b igu o u s va lu e .., xn) all fit th e sta n d a rd form (IV ).
E xam ple 1. Consider th e e x p licit d efin ition
(a) ? ( * , 2, y) = K{x, 7)(y, 0(%)), 2 ).

C onsidering x, y an d 2 on th e righ t as each a fu n ctio n of x, z, y,
<?(x, Z, y) = X,(U\{x, z, y), ri{U\{x, z, y), 0(I/?(*, z, y))), C\{x, z, y)).
T h en c e w e see th a t th e follow in g series of ap p licatio n s of S c h e m a ta
(II) — (IV) can b e used to define 9 from £, yj, 0. T h e successive fu n ctio n s
used are nam ed or defined a t th e l e f t ; an d th e ap p licatio n s of th e sch e m a ta
are a n a ly ze d a t th e right. F o r ex am p le, a t S te p 5, S ch em a (IV ) is a p p lied
w ith n= 3 an d m = 1 an d w ith th e preced ing fu n ctio n s of S te p s 3 an d 4
as th e ^ an d x of (IV).
1 . £ — first g iv e n function .
2. y] — second g iv e n fun ction .
3. 0 — th ird giv e n fun ction .
4. U\(£, z, y) = x — (III), n = 3, i — 1 .
5. Q^x, z, y) = Q(Ul(x, z, y)) — (IV ), n = 3, m = 1 ; 3, 4.
6. U\{x, z , y ) = y — (III), n = 3, i = 3.
7. ty{x,z,y) = r\{Ul(x, z, y ) ,^ 1{x, z, y)) — ( I V ) , « = 3, m = 2 ; 2, 6 , 5.
8. C\(x, z , y ) = 2 — (II), n = 3, q = 2.
9. f(x ,z,y) = 'C ,{U \{x ,z,y), ty(x,z,y), C \ { x , z ,y ) ) — (IV ), n = 3,
m = 3; 1, 4, 7, 8 .
N o te th a t this defin ition of 9 from £, yj, 0 can be expressed sy m b o lic a lly

thus,
(b) 9 = S»K, U\, S|(y), Ul S>(0, (7?)), C 2

3).
If C 7], 0 are p rim itiv e recursive, th en so is 9 ; an d a p rim itiv e recursive

description 9 ,, . . . , 9 * of 9 is th en . . . , £ , . . . , 7), . . . , 6 , C7j, 0,, U\, C |, 9
where ..., y) ; . . . , 0 are d escription s of £, 7), 0, re sp ectively.
T h is use of th e id e n tity fu n ctio n s U f in th e an alysis of e x p lic it d efin itio n

is due to G o d el 19 3 4 .
S ch em a (V) is th e sch em a of primitive recursion , w ith o u t p aram eters
(Va) or w ith p aram eters (Vb). W e som etim es w rite th e fu n ctio n 9 so
defined as Rj(x) (for (Va)) or Rn(^> x) (ôr (Vb)). H o w e v e r in sp ea k in g of
a “ p rim itiv e recursion” , we shall n ow u n d erstan d th a t th e a p p lica tio n of
(V) m a y h a v e lu m p ed w ith it som e steps of e x p licit d efinition.
E x a m p l e 2. T o a n a ly z e th e p rim itiv e recursion for a+b ( E x a m p le
1 § 43), first le t us re sta te it w ritin g cp(b, a) for a+ bt
<p(0, a) = a [= U$a) ],
9(6'. «) = (?(*.«))' [= x(b><?{*>■ «).«)>
if X(b, c, a) = c' = S(t/23(&,c,«))].
T h is fits S c h e m a (Vb) w h e n th e rig h t m em bers are expressed as sh ow n
in b ra c k ets. S o w e a c co m p lish th e d efin itio n th u s:
1. S(a) = a ' ~ (I).

2. U\(a) = a — ( I I I) , n — 1 , i = 1 .
3. U\(b, c, a) = c — ( I II), n = 3, i = 2.
4. X(b, c, a) = S(U%(b, c, «)) — (IV ), n = 3, m= 1; 1,3.
f 9 (0 ,« ) = 171(a) _
»
H ( <p(i', a) = x(6. 9 (*. «). a)
(V b), = 2 ; 2 , 4.
T h is show s th a t a+b con sidered ascp(b, a) (i.e. 9 = },baa+bf cf.

E x a m p le 3 § 10 ) is p rim itiv e recursive, w ith S , U\, U\t x, 9 as a p rim itiv e
recursive d escription . W e ca n o b ta in a+b as 9 r(af b) (i.e. = lab a+b)
b y three m ore steps. S y m b o lic a lly ,
X baa+ b= R 2(U\, S ?(S ,£ /f)),
lab a+b = Sf(R 2(t/J, S?(S, 17®)), U\, V$.

T h is illu stra tes th e gen eral m eth o d . B y th e c o m m u ta tiv e p ro p e r ty o f
a+ bt 9 x(a, b) = 9 (a, b), so th e la st three step s co u ld b e o m itte d h ere;
b u t n o t e.g. in tr e a tin g th e recursion for ab.
W e n ow use th ese tech n iq u e s to estab lish th e p rim itiv e recursiveness
o f a series o f fu n ction s. E a c h o f th e fu n ction s listed b elo w a t th e le ft
is p r im itiv e recursive. T o v e r ify th is, th e reader m a y recognize, first,
t h a t th e e x p lic it defin itio n s an d p rim itiv e recursions e x h ib ite d a t th e
rig h t do gen erate p rim itiv e recursive fu n ction s, an d second th a t th e
fu n ctio n s gen erated are th e sam e as those d efined or n a m ed a t th e le ft.
a+ 0 — a ,
|l a+ft'
#1. a+b. = (a+i)'.
#2. «•&. |f[ «-0 = 0,
a-b' = a-b-\-a.
# 3 . ab *(also w r it te n : !r « ° = l.
a e x p b). |[ a6' = a6>a.
# 4 . a!. jfl 0!«'! == 1,a\-a'.
§45 PREDICATES, PRIME FACTOR REPRESENTATION 223
predecessor
# 5. pd(«) = < of a if a > 0, r pd(0) = 0,
0 if a = 0.
\ pd (a') = a.
if a > 5, r a — 0 = a ,
#6. a —b = j' 0a—b
if a < b. \ a — b' — p d (a—b).
# 7. min(a,6). m in (a,b) = b — {b—a).
#7a. m in (a l 9 . . . , a n) =
m i n ( . . . m i n ^ i n ^ , a 2), a 3) . . . , a n).
#8. m ax(a,6). m ax (a, b) = (a-{-b)—min(a,b).
#8a. m a x (a 1(. S im ila rly to #7a.
=j f[ 01 if a — 0, Sg(«) = 1-0, f sg(0) = 1,

#9. sg(«)
if a > 0. or s g(a) = 0°, sg(fl') = 0.
or <
{
f 0 if a = 0, sg(sg(«)), or |f sg(0) 0,
sg(«) = | [ 1
sg(a) = =
# 10 .
if a > 0. o rsg (a ) = m in(a, 1), l sg(a') = 1.
#11. 1 <*-b |. | a—b |= (a—b)-\-(b—a).

# 12. rm (a,b) (cf. § 41) .
| rm(0,6) = 0,
{ (a’,b) = (rm (a,6 ))'-sg | b—(rm(a,6))' |.
rm
r [o/b] = o,
#13. [*/*]• l [«'/*] = [«/*] + sg 1 b—{rm(a,b)Y |.
R emark 1. T h e p a rticu la r list (I) — (V) of sch em a ta for g e n era tin g
th e p rim itiv e recursive fu n ctio n s (B asis A) is a h a n d y one. I f co n sta n ts
be allow ed as p rim itive recursive fu n ctio n s of 0 va riab les, a b asis is
o b ta in ed b y ch a n gin g (II) to
( I I B) 9 = 0,
allow in g n= 0 or m = 0 in (IV ), o m ittin g (V a), an d allo w in g n = 1

in (Vb) (Basis B ). T h is basis em phasizes th e fu n d a m e n ta l role of 0 an d
T h e co n sta n t fu n ctio n s C°q for q > 0 are in tro d u ced b y su ccessive a p
p lication s o f (IV ) w ith n = 0, m = 1, 5 as th e # an d C q_t as th e x i an d
C q for n > 0 b y (IV ) w ith m == 0, an d Cq as th e # S u b sta n tia l red u ction s
in th e basis for g en era tin g p rim itiv e recursive fu n ctio n s h a v e been g iv e n
by P e te r 1934 (see also D a v id N elso n 1947 P a r t II) an d R a p h a e l R o b
inson 1947 . (It shall b e u nderstoo d th ro u gh o u t th e ch ap ter, o u tsid e o f
th e present rem ark an d R e m a rk 1 end § 47, th a t w e are u sin g B a sis A.)
§ 45 . Predicates, prime factor representation. The fo llo w in g
n otion of re la tiv e p rim itiv e recursiveness enters n a tu r a lly in to our th e o r y
for sh o w in g fu n ctio n s to b e p rim itiv e recursive, ju st as th e notion o f
d e d u c ib ility en tered in to our th e o r y for sh ow in g form ulas to b e p ro v ab le.
A fu n ctio n 9 is primitive recursive in ..., (briefly Y ) , if there is a
fin ite seq uence 9 * , . . 9 *. of (occurrences of) fu n ctio n s (called a primitive
recursive derivation of 9 from Y ) such th a t each fu n ctio n of th e seq u en ce
is eith er one o f th e fu n ctio n s Y (the assumed functions), or an in itia l
fu n ctio n , or an im m e d ia te d ep en d en t of precedin g fu n ctio n s, a n d th e
la st fu n ctio n 9 *. is 9 .
Sin ce th is d efin itio n h as th e sam e form as t h a t of d e d u c ib ility , to each
of th e gen eral properties of (- (§ 20 ) a corresponding prin ciple ca n b e
sta te d now . F o r exam p le , if 9 is p rim itiv e recursive in Y , an d som e o f
th e fu n ctio n s Y are p rim itiv e recursive, th e n 9 is p rim itiv e recu rsive in
th e rest of th e fu n ctio n s Y . A n e x a m p le (w ith l = 1 ) w ill b e g iv e n la te r in
w h ich “ if ^ 1, . . . , are p rim itiv e recursive, th en 9 is p rim itiv e re cu rsive”
is true, b u t “ 9 is p rim itiv e recu rsive in ..., is false (E x a m p le 2 § 55).
E xample 1. In E x a m p le 1 § 44, 9 is p rim itiv e recursive in £, y), 8 ,

w ith £, y], 0, XJ\t 0!, U \, C | , 9 as a p rim itiv e recursive d erivatio n .
O u r gen eral result on e x p lic it d efin itio n (§ 44) can b e s ta te d n o w th u s.
#A. A function 9 definable explicitly from functions Y and constants

q v . . . , q8 is primitive recursive in Y .
B y E ty(xv . . . , x n, y) w e m ean th e sum of th e n um bers i>(xv . . x n, y )
y< z
for all n a tu ra l num bers y such t h a t y < z, if z > 0 ; an d 0 , if z = 0 . I t
is a fu n ctio n of x v . . . , x nt z for a n y g iv e n fu n ctio n ty{xv . . x n, y). B y
II ( % , . . . , #n, y) w e m ean sim ilarly th e p ro d u ct of th e n um bers
y< z
. . . , x n, y) for y < z, if z > 0 ; a n d 1 , if z = 0 .
#B. The finite sum S y) and product II . . , x n, y)

V<z y< z
are primitive recursive in
P roof . T h e sum E ty(xl9 . . . 9x n, y ) is given from $(x l9. . . , x nt y) by
y< z
th e fo llo w in g recursion on z:
E y) = 0 ,
V<0
S ty{xl t . . . , x n, y ) = <\>(xl t . . . , x n,z ) + 2 , x n,y ) .
‘ y<z■' y<z
O th e r fin ite sum s an d p ro d u cts reduce to th ese b y e x p lic it d efin itio n ;
e.g. 2 <J;(y) = 2 ^(y) = 2 <J;(y)f 2 t|/(y) = 2 <\>{y+w'),

y< z y=0 y<z' w <y<z y < z —wf
2 ^(y) = 2 <}/(y) = 2 +(y+B»).

w Sv< z y —w y < z '—w
§45 PREDICATES, PRIME FACTOR REPRESENTATION 225
A lth o u g h in th is ch a p te r w e are d e ve lo p in g an in tu itiv e an d n o t a
form al th e o ry , w e sh all som etim es w ish th e conciseness o fe x p re ssio n w h ich
a lo gical sym b o lism affords, an d in p articu lar w e sh all h a v e a need for
such a sym b o lism in form in g n o tatio n s for p red icates an d fu n ction s.
F o r these tw o purposes, w e n ow in trod u ce a n ew lo gical sym b o lism . T h is
sym b o lism is to b e ta k e n as in form al an d m ean in gfu l, in co n trast to th a t of
th e form al sy ste m as su b je ct m a tte r for m eta m a th e m a tics. A n expression
in th e new in tu itiv e sy m b o lism is to be d istin gu ish ed from a form u la in
th e old form al sy m b o lism b y th e d ifferen ces in th e sy m b o ls e x c e p tin g
an d b y c o n te x t. T h is d istin ctio n b etw een tw o sym b o lism s is in tro d u ced
in th is b o o k (and in G o d el 1 9 3 1 ) for th e n am ed purposes, an d is n o t an
estab lish ed u sage in th e literatu re. (Our in tu itiv e lo gica l sym b o lism ,
e x c e p tin g “ = “ (.E ly ) ” , an d th e operators w ith " xcy” etc., is th e
form al sy m b o lism of H ilb e rt an d B e rn a y s 19 3 4 , 1 9 3 9 ; an d our form al
lo gica l sym b o lism , e x c e p tin g an d " 3 !y ” , is th a t of G e n tz e n
1 9 3 4 -5 .)
S y m b o ls in th e W o rd s in th e S y m b o ls in th e
in tu itiv e sym b o lism . E n g lis h lan gu age. form al sym b o lism .
Q = R- Q is e q u iv a le n t to R. Q~ R-
Q^R. Q im plies R (if Q, th en R). QZ>R.
Q&R. Q an d R. Q&R.
Q \/R . Q or R. QVR.
Q- n o t Q. ~*Q-
(y)R(y). for a l l y , i?(y). VyR(y).
(Ey)R(y). there e x ists a y 3yR (y).
such t h a t i?(y).
(E\y)R(y). th ere e x ists a u n iq u e y 3!yR(y).

such t h a t R(y).
iy)y<zR iy)- for all y < z, R(y). Vy(y<z d R(y)).
(Ey)v< M y ) - there e x ists a y < z 2y(y<z &R(y)).
such th a t R (y).
v-yy < M y ) - th e least y < z such

R(y), if (Ey)y<zR ( y ) ;
th a t
otherw ise, z..
Sim ilar n o ta tio n s are form ed b y u sin g “ (y)” , “ ( E y )” an d “ p y ” w ith th e

in e q u a litie s “y < , z ” , “w < y < z ” , “w < y < z ” , ’‘w < , y < z ’' , “w ^ y < z " . W h e n
th e in d ic a te d ran ge of y is e m p ty , th e “ {y)” expression is true, a n d th e

“ (Ey)” expression is false. W h e n th e in d ic a te d range co n ta in s no y such
t h a t i?(y), th e v a lu e of th e “ \iy” expression is th e card in al n um ber of th e
range.
In th e present th e o r y w e o ften t a lk a b o u t th e tru th v a lu es 'tru e'
(briefly t) an d 'false' (briefly f) of propositions, in stea d of th e propositions
th em selves. (C o n te x t w ill d istin gu ish th is use of “ t ” a n d " f ' to sta n d for
th e tru th v a lu e s of propositions from th e an alogo u s use of th e m in §§ 28,
36 in a v a lu a tio n procedure a p p ly in g to form ulas.) W h e n w e do th is, w e
h a v e im m e d ia te ly four ty p e s of function s, (a) F u n c tio n s from { 0 , 1 , 2 , . . . }
number-theoretic functions or b rie fly here functions.
t o { 0 , 1 , 2 , . . . } , ca lled
number-theoretic predicates
(b) F u n c tio n s from {0, 1 , 2 , . . . } to {t, f }, called
or b rie fly here predicates, (c) F u n c tio n s from {i, f} to {t, f}, called truth-
value functions or propositional connectives. W e use fiv e of th em ==,
& , V, d efined b y th e sam e ta b le s as were g iv e n in § 28 for th e re sp ective
form al operators 3 , & , V, -i. (d) F u n c tio n s from {t, f} to {0, 1 , 2, . . . } .
T h e fu n ctio n of th is ty p e w h ich correlates 0 to t an d 1 to f enters in to th e
d efin itio n o f 'rep resen tin g fu n ctio n ' g iv e n below .
O f course, w h en th e propositions are n o t b ein g id en tified w ith th eir
tru th va lu es, 'p re d ica te ' m eans propositional function of natural numbers
(§ 31). O u r p ractice of a t tim es ta lk in g a b o u t th e tru th va lu e s t, f in stea d
of th e propositions calls for com m en t. In fa c t it is im m a terial in m a n y
c o n te x ts w h e th er w e th in k of th e v a lu es of th e pred icates as propositions
or tru th v a lu e s t, f. T h is is b ecau se th e essential m a th e m a tic a l m ean in g of
th e propositions com es from th e d efin ition of th e pred icates w h ich ta k e
th em as valu es. F o r ex am p le , consider th e tw o propositions 3 < 5 an d
3 < 5 . T h e y are d istin ct propositions, d ifferin g in m eaning. A t first sigh t
so m eth in g appears to b e lost if w e id e n tify th em b o th w ith th e one
o b je c t t. H o w ev er, if w e id e n tify th e proposition 3 < 5 w ith t, a t th e sam e
tim e s ta tin g th a t th is is th e v a lu e of th e p red icate < for 3 an d 5 as ar
gu m en ts, w e express all th e m ean in g o f th e original proposition. In oth er
w ords, th e proposition 3 < 5 is sy n o n y m o u s w ith th e proposition th a t
th e p red icate < , in terp reted as h a v in g its va lu e s in th e d om ain {t, f},
ta k e s th e v a lu e t for 3 an d 5 as resp ective argum ents. (M oreover, here it is
im m a terial w h e th er " t " an d " f " m ean 'tru e' an d 'false' as w e p ro v id e
above^ or are s im p ly a n y tw o d istin ct o b je cts as in §§ 2 8 ,3 6 . T h e p red icates
under th e tw o in terp retatio n s are isom orphic, so th e a b stra c t m a th e
m a tic a l co n ten t o f th e proposition th a t th e v a lu e of 3 < 5 is t is th e sam e.)
In w o rk in g clo sely w ith fu n ction s, w e need to b e aw are of tw o m eanings
o f th e com m on fu n ctio n al n o tatio n , as n oted in § 10. F o r p red icates,
§45 P R E D IC A T E S , P R IM E FACTOR R E P R E S E N T A T IO N 227
th ere are three possible m ean in gs (or six if w e d istin gu ish b etw ee n propo
sitions an d tr u th values).
M ean ings of (<P {x lf . . x n)” . A lte r n a tiv e notation s.
1. T h e p red icate P {x l f . . . , x n). P, or . ,x n P (x 1, . . , , x n)

2. T h e v a lu e o f th e p red ic a te P N o a lte rn a tiv e n o tatio n .
for x l t . . x n as argu m en ts
(the am bigu ou s valu e).
3. T h e proposition th a t P (x l t . . . ,xn)
is true for all xlf . . . , x n.
T h ese three senses correspond, re sp e ctiv e ly , to th e n am e form ( §31 ) ,
co n d itio n al (§32), an d g e n e ra lity (§3 2), in terp retatio n of th e free v a ria
bles x x, . . x n in a form u la P (x v . . . , x n) of th e form al system .
E xample 2. T h e tw o sta tem e n ts
tm (or j £ ( 0) = t),
1 E (a ’) = E(a),
define th e p red icate E(a) (== {a is even}) b y recursion. — W e can define

a fu n ctio n e(a) b y th e p rim itive recursion
f e(0) = 0 ,
\ «(«') = sg(e(a))
(cf. # 9 § 4 4 ). T h e n E(a) = e (a )= 0 .
W e s a y th a t a fu n ctio n <?(xXi. . x n) is th e representing function of* a

p red icate P {x v . . . , #n), if <p ta k e s o n ly 0 an d 1 as va lu es an d satisfies th e
eq u ivalen ce
P (x v y{x lt . . . , x n) = 0 ;
S=
or in oth er w ords, w hen th e v a lu es of P are g iv e n as t an d f, if <p(xv . . . , x n)

is 0 w hen P (x l t . . . , x n) is t, and y(x v . . . , x n) is 1 w hen P (x l f . . . , x n) is f.
W e s a y th a t a p red icate P { x ly. . x n) is prim itive recursive , if its
represen tin g fu n ctio n <p(xv . . . , x n) is p rim itiv e recursive (e.g. E(a) in
E x a m p le 2). T h is d efin ition follow s G o d el 1 9 3 1 .
A s an oth er exam p le, w e list th e e q u a lity pred icate, th e representing
fu n ctio n b ein g show n a t th e righ t (cf. # # 10 , 1 1 ).
# 14 . a=b. sg | a — b |.
W e furtherm ore s a y t h a t a fu n ctio n 9 or p red icate P is prim itive re
cursive in pred icates an d fu n ctio n s Y , if th e corresponding sta te m e n t
holds replacin g th e pred icates am o n g P , Y b y th eir representing function s.
G o d e l 19 3 1 g a v e som e theorem s con cerning p rim itiv e recursive fu n c
tion s a n d predicates, w h ich w e sta te as follow s ( # # C — E ). T h e fa c ts

h a d also been o b ta in e d b y S k o lem 19 2 3 .
#C. A predicate P obtained by substituting functions Xi» • • • >X*» for

the respective variables of a predicate Q is primitive recursive in X\>. . . , Xm> Q-
P roof . I f th e g iv e n p red icate is Q(yv . . . , y m) w ith th e represen tin g
fu n ctio n ^( y x, . . ym), an d th e fu n ctio n s s u b stitu te d are Xi(xv • • • j x n)>
. . . , Xm(xi >. . . , x n), th e represen tin g fu n ctio n of th e n ew p red icate
Q(XliXV • * • » Xn)f • • • > Xm(x l* • * •» Xn)) is *MXl(^l* • • • >Xn)> • • • >Xm{X\> • • • t x n))'
T h is fu n ctio n is p rim itiv e recursive in ty, x 1#. . Xm b y S ch em a (IV ). W e
k n o w from § 44 th a t no g e n e ra lity is lost b y considering th e su b stitu tio n
as of th is p articu lar form ; an d sim ilarly in # D :
#D. The predicate Q(x v . . . , x n) is primitive recursive in the predicate Q.

The predicates Q(xv . . . , x n) V R (x v . . . , x n), Q(xv . . . , x n) & R (xv . . . , x n),
Q(x v . . . , x n) -> R (xv . . . , x n) and Q(xv . . . , x n)== R (x v . . . , x n) are p rim i
tive recursive in Q and R.
P roof . L e t th e represen tin g fu n ctio n s ofQ(x v . . . , x n) a n d R (x v . .
x n) x n) an d xix v • •> x n), re sp ectively. T h e n th e represen tin g
be ^(^1 , . . . ,
fu n ctio n of Q(xv . . . , x n) is . . . , x n)) ( # 9 ) , w h ich is p rim itiv e
recu rsive in ty. T h e represen tin g fu n ctio n of Q{x v . . . , x n) V R (x v . . . , x n) is
ty(xv . . . , x n)-xix v . x n), w h ic h is p rim itiv e recursive in ^ an d x- T h e
rest o f th e th eorem follow s b y k n o w n eq u ivalen ces for = in
term s of an d V (cf. C h a p te r V I , allo w in g for th e differences in th e s y m
bolism ).
#E. The predicates (E y)y<zR (x 1, . . . , x n, y) and (y)v<zR {xv . . . , x n, y )

and the function [jiyy<zR (x v . . . , x n, y) are primitive recursive in the predi
cate R .
P roof . Let x(x v • • • > x n> y) b e th e represen tin g fu n ctio n of
R (x v . . . , x n, y). T h e n II x(x v • •» x n> y) is th e represen tin g fu n ctio n of
y<z
(Ey)y<gR(xv . . . , x n,y ). T h is is p rim itive recursive in x b y # B . S im i
la rly , sg (£ x ( * i >--->x n>y)) is th e representing fu n ctio n of (y)y<8
y<z
R (x v ' . . . , x n, y) ( # 1 0 ) . W e illu stra te th e proof for \iyy<zR (x1, . . . , x n, y)
w ith an exam p le. L e t th e va lu es of x lf . . . , x n b e fixed , an d w rite
sim p ly “ x (y )” for x(xv • x n»y) w ith th e fix e d v a lu es of x v . . . , x n.
Su p p ose th a t z = 7 , an d th a t for y = 0 , 1 , . . . , 6 (first row below) x(y)
ta k e s th e va lu e s show n (second row).
§45 P R E D IC A T E S , P R IM E FACTOR R E P R E S E N T A T IO N 229
1 3
'O
y 0 2 4 5
II
x (y ) l 1 1 0 1 0 0
7i(y) = n x(s) l 1 1 0 0 0 0
8< V
<r(y) = E n{t) 0 1 2 3 3 3 3 3
t<v
T h e desired num ber \uyy<zR(y) is th e least y (first row) < z for w h ich
R(y) is true, i.e. for w h ich a 0 appears in th e second row , if there is such a
y. In our exam p le , there is, an d th e least is 3. T h is n um ber also appears
as th e la st n u m ber a(z) in th e fo u rth row. T h e d e v ic e illu stra ted w ill
e v id e n tly w o rk in a n y case. T o ch an ge th e ex am p le, if (Ey)y<zR(y), so
t h a t no 0 occurs in th e second row, th e n a(z) w ill be z, w h ich is w h a t
|xy v<zR{y) w as defined to b e in this case. T h e fu n ctio n a{z) w ritte n o u t in
fu ll is S II x(x i >• . . , x ni s). B y # B , th is is p rim itiv e recursive in x-
t<Z 8<t
In u sin g these theorem s, w e m a y com bin e several a p p lica tio n s in to
one step. B y # 14 w ith ^{x v . . . , x „ ) = x ( * n - . . , x n) is p rim itive
#C,
recursive in x> e -g- usin g § 44, c'-\~a—b is p rim itive recursive. B y
# # E , C an d § 44, (Ey)y<^[Xu Xn)R (xv .. . , x nf y) is p rim itive recursive
in R; e.g. usin g § 4 4 further, the follow in g is p rim itive recursive.
# 15. a<b. a,
“w < y < z Z \ “w < y < z " or “w < y < z ” ; for exam p le,
(y)w<v<zR ixv - > x n, y) = {y)v<z^ wR{x1, - . . , x y + w ) .
A set o f pred icates Qv . . Q m is mutually exclusive, if for each set of
a rgu m en ts n o t m ore th a n one of th em is tru e (cf. § 3).
#F. The function <p defined thus

•••,*») if Qi{xv . ■ - , x n),
? (* !,• • •. X „)
^n) i f Q m {,X 11• • *, X n ^, j
(Pm+i(Xi,• . . , x n) otherwise,
where Qv . . . , Q m are mutually exclusive predicates {or <p(xv . . . , x„) shall
have the value given by the first clause which applies) is primitive recursive in
<Pi.- • •, <Pm +i> Q m- (D efinition b y cases.)
P r o o f , for Qx, . . . , Q m m u tu a lly ex clu sive. F ir s t m e t h o d . Let
4<i,. . . , be th e represen tin g fun ction s of Qx, . . . , Q m. T h e n (o m ittin g
“ {x v . . . , x n) ” to s a v e space)
? = Sg ( W ‘ <Pl + • • • + Sg(+m) • <Pm + • ••• ‘ ’ ?.»+!•
S econd method .
? = m<9i+~+9m+i (01& y=<Pi) V ... V (Qm& y = 9 m) V_

((?!& ... &Qn &y=t pm+1).
P rime factor representation . Let the prime numbers in order of
magnitude be p0t plt p2, . . . f (i.e. pQ= 2, pt= 3, p2= 5, ...).
The fundamental theorem of arithmetic (Gauss 1801) states that a given
positive integer a can be factored into a product of prime factors which is
unique to within the order of the factors. Thus we have a unique rep
resentation of a of the form
(1) a = p?-p?-p?- ...-P? ... (* # 0),
where a{ is the number of times p{ occurs in a as factor (0 if p{ is not a
factor of a). We can regard the product ( 1) as extending indefinitely,
all but a finite number of the exponents being 0.
We now add to our list of particular primitive recursive functions
and predicates.
#16. a\b a divides b. a\b= (Ec)c<b[ac = b], or sgrm(6, a)
#17. Pr(a) = a is a prime number. Pr(a) = a > l & (Ic)1<0<o[c|a].
#18. pi = the t+ l-st prime f Po = 2,
number. l Pv = y^Pi<x<Pi'- +i Pr(%),
where the upper bound p{\ +1 for x is given by Euclid’s demonstration
that to any p there exists a prime > p and <; pl + l (§40). The com
bining of an application of # E with a primitive recursion is legitimate,
as it merely condenses what could be accomplished by first introducing
X(c) = \LXc<x<c[.n Vx{x)f and then writing the second recursion equation
as pf = tiPi)-
the exponent a{ of p{
#19. (a)i = in ( 1), if a ^ 0; (a)t = \LXx<a[pxi\a & pf\a\
0, if a = 0.
We may write ((a),), as (a),„ (((a),),)* as (a)ii t, etc.
the number of non lh(0, a) - 0,
# 20. lh(a) = ■ vanishing exponents
in (1), if a # 0; a) = j
\ lh(i,a)+l if pi\a,
[ lh(z',a) otherwise.
0, if a = 0. lh(a) = lh(a, a).
We can represent the finite sequences aQt..., a8 of positive integers by
the numbers a = $J°*... 'paB*\ then lh(a) is the length s-fl of the se
quence represented by a.
§46 COURSE-QF-VALUES RECURSION 231
Then if a = p$>-.. . ‘ p “ s (a0, . . . , a 8 > 0 ) and b = p b0° - ____-p ht‘
(■ K > . . bt > 0 ), a*b = p% ■ ... • p f ’p^+ l • • • . 1 Ps+t+v For any a and
any such b, a *\ = a, 1* 6 = b, 1*1 = 1 .
§ 46. Course-of-values recursion. In proving a theorem T(y)

b y induction, it m ay happen that the case T(y') of the theorem depends
not sim ply on the im m ed iately preceding case T(y), but on one or m ore
preceding cases. This kind of a proof b y induction w e have called a
‘course-of-values in d u ction 1. It can be reduced to a sim ple induction, b y
first proving the lem m a (s)8<y T (s) b y sim ple induction, after w hich the
theorem follow s b y settin g s = y (cf. *162a § 40).
The analogous situation arises in definition b y induction. T he function
value <p(0 ) is given o u trig h t; and the function value 9 (y') is expressed in
term s of y and one or m ore of the preceding values 9 (5 ) for s < y. The re
cursion is then called a course-of-values recursion. W e shall see that it can
be reduced to a prim itive recursion b y an analogous device (cf. P eter
1934)-
The tw o cases of the definition of 9 m ay be com bined (cf. *162b), b y
saying thajt 9 (y) is expressed in term s of y and 9 (s) for s < y. W hen
y = 0 , this m eans th at 9 (0) is given outright, since the set of values
9 (s) for s < y is then em pty.
More generally, let the function to be defined be 9 (3/, x 2, . . . , x n) w here
x n are param eters (rem aining fixed throughout the recursion). As
an auxiliary function, w e introduce
(i) ?(y; = i<ny p f u .....**),
called the course-of-values function (in y) for th e given function
G iven the sequence of th e values 9(5, x2, . . . , x n) of our original function

for s < y, b y ( 1 ) w e obtain the value 9 (y; x 2, . . x n) of the course-of-
values function. Conversely, given $(y; x2f. . . , x n), w e can extract all the
values 9 (s, x 2, . . . , x n) for s < y w ith th e help of # 1 9 thus,
(2 ) <p(s, x 2, . . . , x n) = ($ (y ; x 2, . . . , x „))s if s < y.
So in a sense th e know ledge of the value ?(y; x 2, . . x n) of th e course-
of-values function is equivalent to the know ledge of the sequence of
values 9 (0 , x 2y . . . , x n), . . . , 9 ( y— 1 , x 2> . . . , x n) of the original function.
# G . I f 9 satisfies the equation
(3) ■ *»..... x»\
\ 9 ( y ' ; x 2, . . . , x n) = y ( y ; x 2>. . . , x „ ) • p x {v’ ?,iv;x*
Then we obtain 9 from 9 by the explicit definition,
(5 ) <p(y, x 2,..., *n) = (9 (y' ; x2,..., *„))„.
E xample 1. Let
(a) <p(y) = II (y + <p(s)).
8<y
The sequence of the values of this function, and of its course-of-values
function, are as shown.
y 0 1 2 3
300
4
145920
...
?(y) 1 2 12
2U32
•.•
?(y) 1 21 21-32-512 21-32-512-7800 ...
N o te t h a t th e la st ex p o n e n t in 9 (y') is a lw a y s th e v a lu e o f 9 (y); e.g.
(9 (3))2 = 12 = 9 (2 ). T o a p p ly # G , n ote th a t b y (2 )
(b) <p(y) = 8<y
n (y + (?(y))s).
This is of the form (3 ), and by # # 1, 19, B and G, 9 is primitive re

cursive.
This version of #G accomplishes the reduction of course-of-values
recursion to primitive recursion, for cases when the course-of-values re
cursion is already given in the form (3) of a dependence of 9(5/, x 2, ..., xn)
on the number 9(y; #2, ..., xn) besides on y, x2, ..., xn.
We illustrate further how to reduce course-of-values recursions not
already so given to the form (3 ).
E xample 2 . recursion from a double basis.
9 (0 ) = ?o.
(a) 9 ( 1) =
. W ) = x(y. 9(y). 9 (y'))-
First we restate this in the more compact form for course-of-values
recursion (using # # 6 , F), thus
f0 if y = o,
9i if y = 1.
I
x(y—2, <p(y—2), ep(y—1)) otherwise.

Then we express 9(y—2 ), 9(y— 1) as (<?(y))y^2>(?(y))v^v respectively.
The method applies also to definitions of predicates by course-of-values
recursion.
§47 U N IF O R M IT Y 233
E xample 3. C onsider th e eq u ivalen ce
(a) T(y) = y = 23 V F (y ) V [ y = 2 17 - 3 (^ - 5 {^ & r ( ( y ) x) & T ((y )2)] V
[y = 2 19*3(l/)l • 5 (y)2 & r (( y ) j) & T ((y )2)] V [ y = 2 21 - 3 (l/)l & T{(y) x)],
w here F is a g iv e n pred icate. T h is defines T(y) b y co u rse-o f-va lu es in
d u ctio n on y . F o r w h e n y = 0 , all d isju n c tiv e m em bers on th e rig h t are
false e x c e p t p erh aps th e seco n d ; so T(0) == F ( 0). W h e n y > 0 , (y)i < y
an d (y )2 < y.
T h u s T (y ) is expressed in term s of y, F an d T(s) for s < y o n ly. L e t
r(y) b e th e represen tin g fu n ctio n o f T(y); an d le t T ((y )1), T((y)2) in th e
rig h t m em ber of (a) be expressed as (T(y))(y) = 0 , (?(y))^) 2= 0 , re
T has th e form
sp e c tiv e ly . N o w th e defin itio n o f
(b) T (y)= R < y,x(y))

w here, b y # # 2 , 3 , 14, 19, A , C an d D , R(y, z) is p rim itiv e recursive in F .
T h is m eans s im p ly th a t th e represen tin g fu n ctio n p o f R is p rim itiv e
recu rsive in th e represen tin g fu n ctio n o o f F ; so w e h a v e an eq u a tio n of
th e form
(c) T(y) = p(y, ?(y))

w here p is p rim itiv e recu rsive in o. B y #G, t is p rim itiv e recu rsive in p,
an d h ence in u; i.e. T is p rim itiv e recu rsive in F (and is p rim itiv e recur
siv e if V is).
I n th ese exam p le s w e do a lw a y s su cceed in red u cin g th e g iv e n cou rse-of-

v a lu e s recursion to th e form (3) b y use of (2). Closer e x a m in a tio n in § 47
w ill sh ow w h y , an d en able us to fo rm u late a version of # G w h ic h in
cludes t h a t reduction .
E xample 4. simultaneous recursion . T h e fu n ctio n v a lu e s 9 x(y)

an d 9 2{y) are expressed in term s of y an d v a lu e s 9 1 (s) a n d 9 2(s) for
s< y . R e d u c e to # G b y u sin g as a u x ilia r y fu n ctio n
9(y) = 2 ^ liy)3 ^ {y).
*§ 47. Uniformity. In ^ # A — G (considering for th e m o m en t o n ly

fu n ctio n s an d n o t p red icates), our concern w as n o t p rim a rily w ith a n y
p a rticu la r fu n ctio n s 9 an d b u t w ith m eth o d s b y w h ich a fu n ctio n 9 is
d efin ed from Y as u n sp ecified fun ction s. In sh ow in g for a p articu lar
such m e th o d t h a t a fu n ctio n 9 of n v a ria b les is p rim itiv e recursive in Y,
our a p p lica tio n s of th e sch e m a ta (I) — (V) d id n o t d ep en d on w h a t
fu n ctio n s are, so lo n g as th eir n u m ber l an d th e re sp ective num bers
tn v . . . , Wj o f argu m e n ts w h ic h t h e y ta k e is fixed . I n o th er w ords, w e
gave a primitive recursive derivation schema of 9 from Y, with a fixed
analysis. The definition of 'analysis' is analogous to that given in § 20.
For an application of Schema (II) it includes the specification of the n
and q ; for (III) of the n and i ; etc. (Analogously we often obtained derived
rules for the formal system by exhibiting a 'deduction schema' with
metamathematical letters standing for unspecified formulas, variables,
etc.) Under these circumstances, we say that 9 is primitive recursive
uniformly in Y.
We can also explain this uniformity notion as follows. For a particular
method of defining a number-theoretic function 9 from number-theoretic
functions Y, we can write 9 = F(Y) to express the fact that what
function 9 shall be is determined by what functions Y are. Then F is a
fixed mathematical function of higher type, namely one from l number-
theoretic functions Y of m lf..., mt variables respectively to a number-
theoretic function 9 of n variables. We call such a function F a scheme
function or schema or scheme or functional. We can also write y{x1$..., xn)
= F(Y ;xlf. . xn) (with the same F) to express the fact that (by the schema
F) what natural number <p(xv ..., xn) shall be is determined by what
functions Y are and what numbers x v ..., xn are.
For any fixed n and m, Schema (IV) constitutes a functional, which
we have already denoted by S^. For fixed n (and when n = 1, fixed q)
Schema (V) constitutes a functional Rj or Rn. The other three schemata
(I) — (III) define particular number-theoretic functions S, CJ, U% (or
constitute functionals with l == 0).
Now we say that a functional (or schema) 9 = F(Y) is primitive
recursive, or that 9 is primitive recursive uniformly in Y, if F is definable
explicitly from the functionals S*, R*, Rn and constants S, CJ, U£.
E xample 1. In Example 1 § 44 , 9 is primitive recursive uniformly
in £, y), 0 . Under the first version of the definition of uniformity, we see
this from the fact that the analysis of the primitive recursive derivation
schema £, tj, 0, U\t 01# ZJ\, C|, 9 (consisting of the explanations opposite
1 — 9 at the right) is fixed. Under the second version, we see it from the
fact that (b) expresses 9 = F(£, 7), 0) explicitly in terms of Sf, Sf, Sf and
U\, u\, Cl
Sometimes a method for determining a function 9 from functions Y
is specified only under some restriction on the Y's. To establish uniform
primitive recursiveness, we show then that there is a fixed succession of
applications of Schemata (I) — (V) which leads from Y to the same func
tion 9 as the given method, for any Y to which the given method applies
§47 UNIFORMITY 235
(or is intended to apply). The succession of applications of Schemata
(I) — (V) then leads in fact to some function 9 from any Y, since each
of (I) — (V) has this property. Thus: The given method constitutes a
functional 9 = F(Y) defined only for a restricted range of Y. The suc
cession of applications of (I) — (V) constitutes a completely defined
functional 9 = FX(Y), which is primitive recursive and such that F1(XF)
= F(Y) on the range of definition of F.
E xample 2. Let a function 9 be defined thus,
9s(x) if +2W=0 ,
? i ( * ) if + i W = o .
(a) ? (*) =
93(*) otherwise,
where <pv cp2, <p3, <p x, <p 2 are given functions such that, for each xy tyx(x) and
$ 2 {x) are each either 0 or 1 and not both 0 . Then we can write
<b) ?(*) = sg(îW)-?iW + sg(<M*))-<p,(*) + +i(*)-+i(*)-<Ps(*).
and conclude that 9 is primitive recursive uniformly in <px> 92, 93, $x, <J>2
(cf. the first proof of # F § 45 ). The 9 of (a) was only defined for tyx,
satisfying the restriction stated; but (b) defines a 9 without any restric
tion, which i<$the same 9 as the former when the restriction is satisfied.
For schemes involving predicates, we say that a function 9 or predicate
P is primitive recursive uniformly in predicates and functions Y, if the
corresponding statement holds replacing the predicates among P, Y by
their representing functions. The interpretation just explained applies
when the functions introduced as representing functions of the predicates
among Y are then treated as unrestricted function variables in applying
the schemata (I) — (V).
Using the explained interpretation, we can say that a function 9 (or
predicate P) is primitive recursive uniformly in Y, even when some of the
Y ’s are particular functions (or predicates). Then if any of those Y's
which are particular are primitive recursive, 9 (or P) is primitive re
cursive uniformly in the rest of the Y's.
If 9 is primitive recursive uniformly in 0, Y as function variables, and
we then take 0 to be 0*, the resulting function 9* is primitive recursive
uniformly in 0*, Y (and hence if 0* is primitive recursive, in Y). This
principle works whether 0* is a particular function, or a function variable,
including the case that it depends on additional number variables
cP as parameters. The principle is stated accurately as Lemma I.
To make it clear how many (and which) independent variables the
functions have, we write 0 = \sx. . .so0(s1, . . ., sq) (a function of q varia-
bles), 0* = . .sqcx. . ,cP 0Hc(s1, . . sq, cv ..., cP) (a function of q + p

variables), and Xsx. . .sQQ*(slt. .., sQ, cv . . cP) for the function of q
variables sv ....,s Q which we get from 0* whenever c1,...,c P are fixed
numbers.
Lemma I. Given a functional 9 = F(0, W) as follows
9 (^1,..., %n) = F(Xsx...s q Ofo.. .sq), Y ; xv . . xn),
let a functional 9 * = G(0*, Y) be defined as follows
* n , ^1>- • •> Cp) =
^(XSi. . . • • . Cp 0*($i>. . ., Sq, Cj,. . ., C p), T*, X^, • • •, X n, C-±, • . ., Cp) =

F(XSj . . »Sq 0^(-Sj, • . . , Sqy Cj, . . . , 6j))> T* t t • • • f %n)-
If F is primitive recursive, so is G.
P roof. The essence of the proof is that both explicit definitions and
primitive recursions remain such when parameters are introduced.
To give the proof in more detail, we use course-of-values induction on
the length k of a primitive recursive derivation schema 9 ^ . . . , 9 *. of
9 from 0, Y. Seven cases arise according as 9 (= cpk) is 0, or one of the Y's
(say or an initial function by Schema (I), (II) or (III), or an im
mediate dependent of preceding functions by Schema (IV) or (V).
Case 6 : (p(xlf..., xn) = <Kxi(*i> • • •, *«), • • •> X«(%> • • •, *«)), where <|/,
Xi>- • Xm precede 9 (= 9 *) in 9 ^ .. . , 9 *. Then
9*{X1» >Xn>^1*•••t Cp) :=
•••
V ( x t ( * V - > X n , Cl t . . . , C p) , . . . , x t ( * V - ’ X n , Cv . . . , C p), CV . . . , C P).
By the hypothesis of the induction, ^*, y*, • • •>Xm are primitive re
cursive uniformly in 0*, Y. By #A, 9 * is primitive recursive uniformly
in <J#*f Xi>* • Xm> and hence, in 0*, Y.
E xample 3. Not every number-theoretic function is primitive re
cursive. (Why? Cf. §§ 1 , 2 : Is every real number algebraic?) Let £(c) be a
particular function which is not primitive recursive. Let 9 be defined
from an unspecified function 0 thus,
9(*) = 5(6(0)).
Then, for each particular 0, the resulting 9 is a constant function, and so
is primitive recursive, by an application of Schema (II) with n = 1,
q = 5(0(0))'. A fortiori, for each 0, 9 is primitive recursive in 0, with C\
for q = 5(0(0)) as a primitive recursive derivation of 9 from 0. But,
because the analysis of this derivation depends on 0, we cannot conclude
that 9 is primitive recursive uniformly in 0. Indeed if it were, by Lemma I
§47 UNIFORMITY 237
ta k in g U \{s, c) (w hich is p rim itiv e recursive) as th e 0*(s, c), th e resu ltin g
fu n ctio n <p*(x, c) w o u ld b e p rim itiv e recursive, an d h ence so w o u ld b e
<p*(0 , c). B u t <p*(0 , c) = £([/|(0, c)) = £(c). — T h u s also L e m m a I does n o t
h old , if th e h yp o th e sis th a t F is p rim itiv e recursive, i.e. t h a t 9 is p rim itiv e
recu rsive u n ifo rm ly in 0, Y , is w eak en ed to : 9 is p rim itiv e recu rsive in
0, Y , for each 0, Y . — In th is ex am p le, of course 9 is p rim itiv e recu rsive
u n ifo rm ly in £, 0.
Since th e form er proofs a c tu a lly estab lish u n ifo r m ity :
# # A — G (second versio n s). Reread the original versions with “p rim

itive recursive u n iform ly” in place of “prim itive recursive” .
U s u a lly a co u rse-o f-va lu es recursion arises in th e fo llo w in g form . T h e
am b igu o u s v a lu e <p(y, x2, . . x n) is g iv e n in term s of y, x2, . . x n, o th er
fu n ctio n s an d pred icates Y , an d <p(s, x2, . . . , x n) as a fu n ctio n of s for th e
g iv e nx2, . . . , x n. T h e expression b y w h ich it is g iv e n is th e result of su b
stitu tin g <p(s, x 2, . . . , x n) for a fu n ctio n va ria b le 0(s) of a p rim itiv e re
cu rsive fu n ctio n al. T h is fu n ctio n a l h as th e p ro p e r ty t h a t its v a lu e is
not ch an ged, if va lu es of 0 (s) are ch an ge d for s > y o n ly. In o th er w ords,
there is a p rim itiv e recursive fu n ctio n a l F(Xs0(s), Y ; y , x 2, . . . , x n) such th a t
(6) cp(y, x 2, . . . , x n) = F(Xscp(s, x2, . . . , x n), Y ; y, x2, . . . , x n),
F(Xs01(s), Y ; y, x 2>. . x n) = F(Xs62(s), Y ; y, x2. . . . , x n)
w h e n eve r 0x(s) = 02(s) for all s < y .
x 2). . . , x n) is prim itive re

U n d er these circu m stan ces, w e s a y t h a t 9 (3/,
cursive uniform ly in 9 (s, x 2i. . . , x n) for s < y and Y.
L ik e term in o lo g y is used for pred icates (reading “P ” , “H ” , in
p lace of "9”, “= ” )•
In case w e are con siderin g th e d efin ition of 9 from Y o n ly for a restricted
range of Y , th e n (7) as w ell as (6 ) need o n ly h old on th is range.
E xample 4. L e t 9 (y, x) b e d efined b y
(a) ? {y, X) = y • P(?(o(y). *)) + iLZz<y[<?(z, x) I y]

w here p, a are g iv e n fu n ctio n s such t h a t a{y)
y for y > 0. T o see t h a t
<
9 (3/, x) is p rim itive recu rsive u n ifo rm ly in 9 (5 , x) for s < y an d p, a, le t
us insert an u n sp ecified fu n ctio n Xs0(s) in p la ce of Xs 9 (s, x) in th e rig h t
m em ber of (a), for co n ven ien ce ca llin g th e resu ltin g fu n ctio n Xi (y>x) :
(b) x i(y . *) = y ■ p(0(®(y))) + ^ < , [ 6 ^) | y \.

B y # # A , C, E , 16 (using th e second versions of A , C an d E ), Xi(y> %) is
p rim itiv e recursive u n ifo rm ly in 0, p, a ; an d c h a n g in g va lu e s of d(s) for
s > y o n ly w ill n o t ch an ge th e v a lu e of Xi {y, x) under th e restrictio n on <r.
23 8 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
E xample 5. W e can see d ire c tly from (a) of E x a m p le 1 § 46 ((b)

of E x a m p le 2 § 46) th a t 9 (3/) is p rim itiv e recursive u n ifo rm ly in 9 (s)
for s < y (in 9 (s) for s < y an d y). In ste a d of first w ritin g “ 0” in p la ce o f
“9 ”, w e need m erely to exam in e h o w th e righ t m em ber is co n stru cte d
o u t of 9 (5) regarded m o m e n ta rily as an u n specified fu n ction . S im ila rly
from (a) of E x a m p le 3 §46, T (y) is p rim itive recursive u n ifo rm ly in T(s)
for s < y an d V.
# G (third version). I f <p(y, x 2, . . . , x n) is prim itive recursive uniform ly
in 9 (s, x 2, . . . , x n) for s < y and Y , then 9 is prim itive recursive uniform ly
in Y. S im ilarly for a predicate {reading “P ” in place of " 9 ” ).
P roof , for a function 9. B y (6 ), (7) an d (2),
^ <p(y- x 2, . . •, X n) = F(Xs (9 (y; x2, . . . , x n))„ T; y, x2, . . x n)

= x(y> $ ( y ; x 2, . . . , x n) ,x 2, . . . , x n)
w here
(9) x (yf c ,x 2, . . . , x n) = F(Xs (c)„ Y ; y, x 2, . . . , x n).

B y L e m m a I, / is p rim itiv e recursive u n ifo rm ly in Xsc (c)8, Y ; an d hence b y
# 1 9 , in Y . N o w th e second version of # G applies.
T h e result for a p red icate P follow s b y go in g o ve r from P to its rep
resen tin g fun ction .
R emark 1 . C f . R e m a rk 1 end § 44. I f 9 , Y are functions of n f m l f . . . ,

m x > 0 variables , then 9 is prim itive recursive uniform ly in Y under B asis
B , if and only if under B asis A . F o r a n y p rim itive recursive d eriv atio n
sch em a of 9 from Y under B a sis A can b e tran sform ed in to one under
B asis B as a b o v e b y su p p ly in g a d escription under B asis B o f (of C®)
for each a p p lica tio n of (II) (of (Va)). C o n versely, g iv e n a p rim itive re
cu rsive d eriv atio n sch em a <plt . . . , 9 *. of 9 from Y under B asis B , one
u n d er B a s is A ca n be o b ta in e d b y th e follow in g process. S a y th a t n= l=
mx = 1 , i.e. th a t th e d eriv a tio n is of 9 ^ ) from ^(y). L e t a p aram eter c b e
in tro d u ced in to each of th e fu n ctio n s 9 ^ . . . , 9 *. S im ila rly to th e proof o f
L e m m a I, w e ca n th e n o b ta in a p rim itiv e recursive d e riv a tio n sch em a o f
y(x, c) from <p(y, c) under B a sis A . T o th is w e p refix ^(y, c) — § {U \{y, c))
an d su ffix 9 ^ ) = 9 {U \{x), C l(x)). — F o r ex am p le, if 9 (0) = ^(0), 9 (y') ==
X(y, ?(y))> th en 9 is p rim itiv e recursive u n ifo rm ly in 4>,X under B a sis B
(using su cce ssiv e ly ( I I B), (IV ) w ith n— 0, (Vb) w ith n= 1 ). H en ce it is
also under B a sis A .
§ 48. G o d e l ’ s p -fu n c tio n . The second problem of th is ch a p te r

is to sh ow th a t every p rim itiv e recursive p red icate is num eralw ise
§48 g o d e l ’s (3- f u n c t io n 239
expressible in th e form al sy ste m of C h a p ter IV , ev e n th o u g h th a t
sy ste m has fu n ctio n sy m b o ls o n ly for th e three fu n ctio n s ■.
W e sh all p ro v e th is in th e n e x t section, fo llo w in g a m eth o d of G o d el
(1 9 3 1 , 1934 ).
T h is proof is n o t essential to our program of form alizin g n um ber th eo ry.
I f it did not succeed, w e cou ld h a v e arran ged in stea d t h a t recursion
eq u atio n s for oth er fu n ctio n s besides + an d • sh ould b e axio m s of th e
system . In deed, b y an e n u m erab ly in fin ite sy ste m of p articu lar n u m b er-
th eo retic axiom s, w e co u ld in clu d e recursion eq u atio n s for all th e p rim i
tiv e recursive function s. H o w e v e r it is of som e interest th a t a fin ite sy ste m
suffices, th e m ore so th a t w e can g e t alon g w ith th e tw o ch ief fu n ctio n s +
an d • of tra d itio n a l arith m etic, w hen ta k e n w ith th e lo gical co n sta n ts
an d th e p red icate = .
G o d el has called a p red icate arithmetical, if it can b e expressed e x
p lic itly in term s of co n sta n t an d va ria b le n a tu ral num bers, th e fu n ctio n s
+ an d •, e q u a lity = , th e operations V, "“ of th e p rop osition al ca l
culus, an d th e qu an tifiers (x) an d (Ex), co m b in ed acco rd in g to th e u su al
s y n ta c tic a l rules. (This uses th e a d je c tiv e ‘a r ith m e tic a l’ in th e narrow er
sense, § 9.)
T h e reader m a y re a d ily g iv e th e d efin itio n m ore fu lly as an in d u c tiv e
definition, p arallelin g th e d efin itio n of form u la for th e form al sy ste m .
T h e arith m e tic a l pred icates are p recisely th ose w h ich ca n b e expressed b y
nam e form s in th e form al sy ste m und er th e usu al in terp retatio n of th e
sym b ols. ( B y com parison w ith th e form al tre a tm e n t in § § 3 9 an d 4 1,
a<b an d rm(c, d )= w are arith m etical.)
B u t, u sin g th e in tu itiv e sym b o lism , we sh all keep th e discussion
inform al for th e present. F o r th e a p p lica tio n to p rim itiv e recursive p red
icates, w e sh all require o n ly c o n stru c tiv e use of th e quantifiers.
In th e n e x t section, w e sh all need a m eth o d of d ealin g w ith fin ite se
quences a Q, . . . , a n of n a tu ral num bers a r ith m e tic a lly ; there w e ca n n o t use
th e fu n ction s ah, p { an d (a){ of §§ 44, 45 w ith w h ich w e h an d le d fin ite
sequences p rim itive re cu rsively in §§ 46, 47.
W e k n o w th a t th e p red icate rm(c, d )—w, w here rm(c, d) is th e rem ain
der w h en c is d iv id e d d, is arith m etical.
by
A set of p o sitiv e integers d 0, . . . , d n are said to b e relatively prim e , if
no tw o of th em h a v e a com m on p o sitiv e in tegral fa c to r e x c e p t 1. F o r
exam p le, 3, 4, 5 are r e la tiv e ly prim e.
Consider th e n -\-\ -tu p le s of th e v a lu es of th e fu n ctio n rm(c, d), for a
fix e d w +1-tuple of r e la tiv e ly prim e d ivisors d 0, . . . , d n, as c increases.
F o r ex a m p le (w ith n = 1 ), if dQ= 3, d1 = 4, t h e y are as follow s.
c 0 1 2 3 4 5 6 7 8 9 10 11 12 13 . . «
rm(c, 3) 0 1 2 0 1 2 0 1 2 0 1 2 0 1 . . .
rm(c, 4) 0 1 2 3 0 1 2 3 0 1 2 3 0 1 . . .
We see that, as c ranges from 0 to 11, the pair of remainders rm(c, 3),
rm(c, 4) assumes each of the 12 possible ordered pairs of numbers aQt
ax for a0< 3, ax< 4.
To establish this in general, let rm(c, dQ), rm(e, dx), ..., rm(c, dn) take
the respective values a0>av . .., an for c = j and also later for c = j+k.
Since j and j-\-k give the same remainder a{ on dividing by dt (i = 0,
..., n), their difference k must contain dt exactly; say that k = M*- Thus
k = bQdQ= bxdx = . • • = bndn.
Here k has each of d0, dx, ..., dn as factor. Since by hypothesis d0> ..., dn
are relatively prime, by the fundamental theorem of arithmetic (§ 45)
k must be a multiple of their product dQ- dx - ... •dn.
Therefore the ordered n + 1-tuple rm(c, i 0), Tm(c,dx), ..., rm (c,dn)
cannot return to a given sequence of numbers a0, ax,..., an after less
than dQ- dx * ... •dn consecutive values of c. But there are exactly d0- dx *
... •dn distinct sequences of numbers a0, ax, ..., an for aQ< dQ, ax < dv
..., an < dn. Each sequence is therefore taken once in any dQ- dx - ... •dn
consecutive values of c.
Following Godel 1934, we use this fact to construct a function (3(c, dt i)
with the two properties,
( 1) the predicate (3(c, d, i) = w is arithmetical, and
(2) for any finite sequence of natural numbers a0, av ..., an, there can be
found a pair of natural numbers c, d such that
P(c, d, i) = at (i = 0, 1, . . n).
As we know, a number c (c < dQ• dx•... •dn) can be chosen so that
rm(c, d{) = at for i = 0, 1, ..., w, provided d0, dv . . dn is a set of
numbers such that (a) dQ, dx, ..., dn are relatively prime, and (b) aQ< d0t
ai < dv ... ,a n < dn. Our problem will be solved, if we can obtain the
numbers d0, dx, ..., dn as the values of a function S(d, i) for i = 0, 1, ..., n
and a suitable number dy so that
(i) p(c, d, i) = rm(c, S(d, i))
alsa satisfies ( 1).
Now ( 1)' will be satisfied if we take
(ii) 8(d,i) — l + (^ + l)^-
For rm(c,d)=w is arithmetical, and 8(d,i), which we substitute for
d to get p(c, d} i)=w, is defined explicitly from 1, + and \
§49 THE NUMBER-THEORETIC FORMALISM 241
For the given sequence of numbers aQ, av . . ., an, let s be the greatest
of n, a0, av . . ., an, and take d = s\.
Then, (a) the numbers dt = §(d, i) for i = 0, 1 , .. n are relatively
prime. For if two of them ! - } - ( / + 1 )s! and 1+ (/+&+ l)s! had a
factor other than 1 in common, they would have a prime factor p in
common, and this factor p would divide the difference which is k*s\.
But p cannot divide $!, since then it would divide (j-\-\)s\, which is
impossible since it divides 1+ (/+ l)s!. Then also p cannot divide k,
since k < n < s and every number < s divides s\. Hence p cannot divide
k-s\; so by reductio ad absurdum (a) is proved.
Moreover, (b) for each i (i = 0 , 1 , . .., n), at < s < si < 1 + (^+ l)s!
= m i) = dt.
Julia Robinson 1 9 4 9 * shows that the predicate j (#16) and function ' can
be used in place of the two functions + and • in defining the arithmetical
predicates; and Church and Quine 1 9 5 2 show that a suitably chosen sym
metric 2 -place predicate can be used instead.
§ 49. Primitive recursive functions and the number-theoretic
fo rm a lism s T heorem I. If 9(34,. . xn) is a primitive recursive func
tion, then the predicate <p(xv . . . , x n)=w is arithmetical. (Godel 1 9 3 1 .)
P roof , by course-of-values induction on the length k of a given
primitive recursive description 9 ^ . . ., 9 *. of 9 (cf. § 43). The cases (I) — (V)
correspond to the five schemata by which cpk, i.e. 9 , may occur in the
description. (For a proof with a similar case structure, cf. that of Theorem
1 § 2 1 .)
Case (I): <p(x)t = x'. Then cp(x) = w == w—x-f-1, and w=x -f-1 is arith
metical.
Case (II): <p(xv . .., xn) = q. Then <p(xlt ..., xn) = w = w—q.
Case (IV). 9 (î> • • •>xn) ~ • • • >xn)>• • •>"£m(xi> • *•>xn)),
where by the hypothesis of the induction, ^(>3 , . . ., ym) —w,
Xi(xv . . . , x n)=yv . ,.,Xm(xlt . . •,x n)-=ym are arithmetical.
Then <p(xv . . . , x n)=w = (EyJ . . . (£yra)[xi(#i, .. .,x n)=y1 & ... &
Xm(xlt . . . , x n)=ym&^{yv . . . , y m)=w).
Case (Vb): 9 (0 , x2, ..., xn) = ty{x2, . . . , x n), 9 (/, x2t . .., xn) =
X(y> 9 (y> xn)>x2, .. ., xn), where p{x2, . . ., xn) = w and y(y, z,x2,...,
xn)—w are arithmetical. Suppose that y, *2, . . . ,xn, w are numbers such that
9 (y, x2, ..., xn) = w is true. Then there is a finite sequence of numbers
* 0> ^ 1* • • •>
(the va lu e s of <p(i, x %t. . x n) for i = 0, 1, . . . , y) such th a t
«0 — * • * * %n)>
"’’is
cT
«1
H
*
II
©
«2 X( ^ ^1> %2> * • • * %n) *
•
ay
"‘"’g
T
<3
K
ii
ca
X
w = a y.
B u t th e n there are num bers c, d for G o d el's (3-function such th a t (3(c, d, i)
= (i = 0, 1 , . . y), an d th e fa cts (A) can b e expressed u sin g th e
(3(c, d t i ) fs in stea d of th e a /s , th u s:
(Ec)(Ed){p(c, d , 0) = ty(x2, .. . , x n)
(B) & { i ) [ i < y - + P(c, d } i + 1 ) = x(i, p(c, d, i), x2, . . . , # * ) ]
& w = |3(c, d, y)}.
C o n ve rse ly, if (B) is true, th en for a n y c an d d g iv e n b y (B), th e num bers
(3(c, d t i) for i = 0 , 1 , . . , , y do c o n stitu te a sequence a0) % , . . . , a y
sa tis fy in g (A ); an d (A) im p lies th a t <p(y, x 2f . . . , x n) = w . Thus
9 {y, x2, . . . , x n) = w is e q u iv a le n t to (B). B u t (B) is an arith m e tic a l
p red icate of y, x2> . . . , x nt w t as w e see b y rew ritin g it in th e form
(Ec){Ed){(Eu)[${c, d, 0 ) = u & ty(x2t . . . , x n) = u ] & ( i ) [ i < y - >

(C) (Eu)(Ev)[$(cf d , i + l ) = u & $ ( c , d , i ) = v & x { i , v , x 2, . . . , x n) = u ] ]
& p(c, d, y ) = w } ,
an d ta k in g in to a c co u n t th e h yp o th e sis of th e in d u ctio n an d th e a r ith m e t
ica l ch aracter of (3(c, d, i)= w an d icy.
T h e a n aly sis of p rim itiv e recursion in term s of fin ite sequences of
n a tu ra l n um bers used here is a n u m b er-th eo retic a d a p ta tio n of D e d e
k in d 's an a ly sis of p rim itiv e recursion (1888 ).
Corollary . Every prim itive recursive predicate P { x lf . . x n) is

arithmetical.
F o r P ( x v . . . , x n) = 9 (^1, . . . , x n) = 0 , w here 9 is th e represen tin g
fu n ctio n of P (§4 5). C o n ve rsely, th e theorem follow s from its co ro llary
u sin g # # 1 4 , C. T h e theorem h o w e ver has th e form n ecessary to th e proof
b y in d u c tio n on th e le n g th of th e d escription 9 ^ . . . , 9 *.
By tra n sla tin g from th e in tu itiv e a rith m e tic a l sym b o lism in to th e

form al sym b o lism , w e o b ta in a form u la P ( x x, . . . , x n, w) w h ich expresses
<p{xv ..., x n) ~ w u n d er th e in terp retatio n of th e form al system . B y th is
m eans w e shall now p ro v e th e follow in g theorem .
T heorem 27. Every primitive recursive function <p(xv . . . , x n) is

numeralwise representable (§ 41) in the formal system of Chapter I V ; i.e.
there is a formula P ( x x, . . ., x n, w), containing no variables free other than
the distinct variables x v . . . , x n, w, such that, for each n-tuple of natural
numbers x v . . . , x n>
(v) if . . . , x n) = w , then b P(xv . . x n, w), and
(vi) b 3 !w P ( x 1, . . . , x n, w).
P roof . T h e co n stru ctio n of P ( x 1, . . . , x „, w), an d th e proof of th e

theorem , are b y a co u rse-o f-valu es in d u ctio n on k, w ith cases (I) — (V)
corresponding to those in th e proof of T h eorem I.
C ase (Vb). B y th e h yp o th esis of th e in d u ction , there are form ulas

Q ( x 2 , . . ., x n, w) an d R ( y , z , x 2 , . . . , x n, w) w h ich num eralw ise represent
th e resp ective fu n ction s ^{x2, . . . , x n) an d x(y> z >%2, • • •, x n), i.e. th e y h a v e
th e properties for these fu n ction s corresponding to (v) an d (vi) for
?(*i> . . . , x n).
A cc o rd in g to *(180) § 41 , G od eF s (3-function (3(c, d, i) [ = rm(c, 1 +
(i+ \)d = rm(c, (i'- d)')] is num eralw ise represented by a form u la
B (c, d, i, w) h a v in g a further p ro p erty *180a.
T h e form ula
3 c3 d { 3 u [B (c , d, 0, u) & Q ( x 2, . . . , x n, u)] & V i [ i < y D 3 u 3 v [ B ( c , d, i', u)
& B (c, d, i, v) & R (i, v , x 2, . . . , x n, u)]] & B (c, d, y , w)}
shall be th e form ula P (y , x 2,. . . , x n, w) to represent cp(y, x 2, . . . , x n). We

m ust show th a t it has th e properties (v) an d (vi).
T o establish (v), let y, x 2, . . . , x n, w be num bers such th a t 9 (y, x 2, . . . , x n)
— w. T h en there are num bers a0, al f . . . , a y as in th e proof of T h eo rem I,
and also num bers c an d d for these a0, av . . . , a y such th a t (3(c, d, i) = a t
(i — 0 , 1 , . . . , y). B y th e p ro p e rty (v) for B , Q an d R , th e follow in g s ta te
m ents h o ld :
1- B(c, d, 0, a0), h Q(x 2, . . . , x n> a0),

h B(c, d, 1, aj), h R(0, a0, x2> . . . , x n, a j,
. . .
h B(c, d,y, ay), 1- R(y—1, ay_v x2, . . . , x n, ay),
h B(c, d, y, w).
W e can th en ce show th a t b P {y, x 2, . . . , x n, w) b y & -in tro d ., 3-introd .
and *1 6 6 § 4 1 .
T o estab lish (vi), w e use an in tu itiv e in d u ctio n on y. I n d . s t e p . L e t
w= 9 (y, .... x n) a nd 11 = <p(y’, x 2, . . . . x„) == x(y> w , x 2, . . . . x n).
244 P R IM IT IV E R E C U R S IV E F U N C T IO N S CH . IX
B y (v) an d (vi) for R (and for b r e v ity using th e inform al presen tatio n ,
b e gin n in g § 38): (a) R (y, w}x2, ..., x n, u), an d (b) 3 !u R (y , u;, x2, ..., x n, u).
By (v) (as a lre a d y established for P ): (c) P ( y , x2, . . x n, w),
a n d (d) P ( y ', x 2, . . . , x n, 11). B y th e h yp o th esis o f th e in d u ctio n on y:
(e) 3 !w P (y , x 2, ..., x n, w). We m u st p ro ve 3 !w P ( y ', x2, ..., x n, w).
A ssu m e : (f) P ( y ' , x 2, . , x n, w). B y * 1 7 0 w ith (d), it w ill suffice to d ed u ce
a=w w ith w h eld co n sta n t. F o r & - an d 3-elim . from (f), assum e:
(g) 3 u [B (c, d, 0, u) & Q(x 2, . . . , x n, u)]f (h) V i [ i < y ' D 3 u 3 v [ B ( c , d, i', u)
& B ( c , d, i, v) & R (i, v , x2, ..., x n, u)]], an d (i) B (c, d, y ', w). F ro m (h)
u sin g * 1 3 8 a § 3 9 (or * 1 6 7 an d *16 6 ):
(j) Vi [ i c y D 3 u 3 v [ B ( c , d, i', u) & B ( c , d, i, v) & R (i, v , x 2, . . x n,u)]],
an d (k) 3 u 3 v [ B ( c , d, y ' , u) & B (c, d, y , v) & R ( y , v , x 2, . . . , x n, u)]. F o r
& - an d 3-elim . from (k), assu m e: (1) B (c, d, y ' , u), (m) B (c, d, y , v), an d
(n) R ( y , v , x2, . . . , xn, u). F ro m (g), (j) an d (m) b y & - an d 3 -in tro d .:
(o) P ( y , x2, . . . , x n, v). F ro m (o), (c) an d (e) b y * 1 7 2 , v=w , w h ich w ith
(n) g iv e s : (p) R ( y , w t x 2, . . . , x w, u). F ro m (p), (a) an d (b) b y * 1 7 2 u = u ,
w h ic h w ith (1) g iv e s : (q) B (c, d, y ' , u ) . F ro m (q), (i), *18 0 a an d * 1 7 2 , u = w ,
as w as to b e deduced.
C o r o l l a r y . E very p rim itive recursive predicate P (x v . . . , # n) is

num eralwise expressible in the form al system .
For if P ( x x, . . . , x n, w) num eralw ise represents th e represen tin g
fu n ctio n 9 of P , th en u sin g (vii) § 41 (obtained th ere b y * 1 7 3 , *(164 )),
P ( x j, . . . , x w, 0) num eralw ise expresses P .
L emma Theorem 27 and Corollary hold for R aphael R obinson’s

18b.
form al system (§ § 4 1, 76) consisting of the predicate calculus with thirteen
particular number-theoretic axiom s as follow s : A xiom s 14 — 2 1, and {the
form ulas of) *10 4 — * 1 0 7 and * 1 3 7 {or *13 6 ).
U sin g L e m m a 18a § 4 1 .
R em ark 1 . A m ore am b itio u s u n d erta k in g (relating to th e fu ll

sy ste m , n o t to R obin son's) w o u ld be to estab lish th e p r o v a b ility of
form ulas w h ich express th e recursion e q u a tio n s; e.g. for Case (V b), to
e s ta b lis h :
(1) 1- P(0, x 2) . . . , x n, w) ~ Q(x 2, . . . , x „, w).
(2) H P(y', x2, . . . , x n, w) ~

3 z [P (y , x 2, . . x„, z) & R ( y , z, x 2, . . . , x n, w )].
F ro m ( 1 ) an d (2) b y form al in d u ctio n on y (given b y th e h y p o th e
sis of an in tu itiv e in d u ctio n on k th a t |- 3 ! wQ( x 2, . . . , x n, w) and
b 3 !w R (y , z, x 2, . . x n, w)), th en :
(3 ) b 3 !w P (y , x 2, . . . , x n, w).
T o estab lish ( 1 ) an d (2), w e w o u ld b egin by form alizin g th e th e o r y

of th e ^ -function, w h ich w e g a v e in fo rm a lly in § 48. I t w o u ld suffice
to e s ta b lis h :
(a) b 3 c3 d B (c, d, 0, w).
(P) b 3 c 23d 2{ V i [ i < y D 3u [B (c1} d x, i, u) & B ( c 2, d 2, i, u)]] &

b (c 2, d 2> y '> w)}.
F o r th en th e four im p licatio n s of (1) an d (2) cou ld all be proved , u sin g

* 180c, (a) (for th e second), an d (p) (for th e fourth). — W e shall not ta k e
th e space to c a rry o u t th e form alization to establish (a) an d (P). H ilb ert
an d B e m a y s 1934 pp. 4 0 1— 4 19 do p r a c tic a lly th is in an oth er form al
system , w hence (in th e m ann er to be in d ic a te d before E x a m p le 9 § 74)
it can b e inferred th a t (a) an d (p) h old in our (classical or in tu ition istic)
system .
Chapter X
T H E A R IT H M E T IZ A T IO N O F M E T A M A T H E M A T IC S
§ 50. Metamathematics as a generalized arithmetic. As we

rem arked in § 42 (follow ing G o d el 1 9 3 1 ), b y selectin g a p articu lar enum er
atio n of th e form al o b jects, or a p articu lar correlation of d istin ct n a tu ral
num bers to th e d istin ct form al o b je cts (not usin g e v e r y num ber), an d th en
ta lk in g a b o u t th e correlated num bers in stead of th e form al o b je cts,
m e ta m a th e m a tic s b ecom es a b ran ch of th e a rith m etic of th e n a tu ral
num bers. In th is ch a p te r w e shall ca rry out such an arith m e tiz a tio n of
m e ta m a th e m a tics, usin g a G o d el num berin g sim ilar to th a t of H ilb e rt an d
B e rn a y s 19 3 9 .
H o w e v e r in stea d of ca rry in g ou t th e arith m e tiz a tio n d ire c tly, w e sh all
first represent th e form al sy ste m in an in term ed ia te w a y as a gen eralized
arith m etic, an d th en represent th e gen eralized a rith m e tic in th e o rd in ary
arith m etic. T h is w ill b rin g o u t som e an alogies w h ich are o f h eu ristic v a lu e ,
an d th e represen tation of th e sy ste m as a gen eralized arith m e tic w ill b e
o f in terest on its ow n acco u n t.
T h e a rith m etic of th e n a tu ral num bers deals w ith th e d om ain of o b je c ts
w h ich is gen erated b y s ta rtin g w ith one p rim itiv e o b je c t 0 an d a p p ly in g
one p rim itiv e operatio n ' or + 1 ( § 6 ).
A gen eralized a rith m e tic (for th e present purpose) is o b ta in e d b y
su p p o sin g one or m ore zeros, arid one or m ore successor operations. W e
sh all adhere to th e c o n v e n tio n (not th e o n ly useful one) th a t o b je c ts
gen erated from th e p rim itiv e s in d istin ct w a y s are d istin ct. T h ere are
several possibilities for represen tin g th e form al sy ste m as a gen eralized
arith m etic. H erm es 1938 considers an a rith m etic w ith th e e m p ty e x
pression as th e zero, an d th e operatio ns of su ffix in g one of th e form al
sym b o ls as th e successor operations.
T h e gen eralized a rith m etic w h ich w e select has a m ore co m p lic a ted
stru ctu re as an arith m etic, b u t is designed to represent d ire c tly th e
g ra m m a tic a l an d lo gical stru ctu re of th e form al o b jects. T h ere shall b e
r + 1 zeros 00, 0V . . . , 0 r, w here r is a n atu ral num ber to b e specified la te r;
an d there shall b e one successor operation a p p ly in g to an $ + 1 -tu p le of
246
§50 A GENERALIZED ARITHMETIC 247
argu m en ts, for each o f th e n a tu ra l n u m ber v a lu e s of 5 to b e sp ecified
later. T h e result of th e successor op eratio n ap p lied to x 0, xv . . . , x , as
a rgu m en ts is w ritte n “ (xQ, x v . . . , x*)” or also som etim es “ x 0(x 1 , . . x s)” .
W e ca ll th e o b je c ts b elo n gin g to th is gen eralized a rith m e tic entities.
W e ca n express th is b y an in d u c tiv e d efin itio n (analogous to t h a t g iv e n
for 'n a tu ra l n u m b er’ in § 6 ). 1 . 00, 0X, . . . , 0r are entities. 2. F o r each
a d m itte d s, if x 0, x v . . . , x 8 are entities , th e n (x 0, x 1#. . . , x,) is an entity.
3. T h e o n ly entities are th o se g iv e n b y 1 an d 2.
A s w e h a v e a lre a d y in d ica te d , tw o en tities sh all b e equal, if a n d o n ly
if t h e y are gen erated from th e zeros b y th e successor op eratio n s in th e
sam e w a y . T o s a y th a t x an d y are eq u al w e w rite " x X y ” (unequal,
“ x X y ” )- W e use " x ” rath er th a n “ = ” m erely to a v o id con fusion w ith
th e = of th e form al system .
A x io m s ch aracterizin g th e d o m ain of en tities ca n b e sta te d , an alo go u s
to P e a n o ’s for th e n a tu ra l num bers (§§ 6 , 7). In p articu lar, t h e y in clu d e
th e p rin cip le of proof b y m a th e m a tic a l in d u ctio n in th e form corre
spon d in g to th e m ode of gen eration of th e d om ain of en tities (or to th e in
d u c tiv e d efin ition ju s t g iv e n ) : I f th e en tities 00, 0V . . . , 0r each possess
a certain p ro p erty , an d if for each a d m itte d s, w h e n eve r en titie s x 0,
xv ..., x s possess th e p ro p e rty , th e e n t it y (x 0, x x, . . . , x 8) also possesses
it. th en all en tities possess th e p ro p erty . T h e s ta te m e n t of th e o th er
P ea n o ax io m s for th e gen eralized arith m e tic is le ft to th e reader.
T h e process of gen e ra tin g th e en tities p a r tia lly orders th e m (end § 8 );
w e w rite " x < y ” to s a y t h a t x is g en era ted before y in th e process of
gen eratin g y . E x p r e sse d in d u c tiv e ly : 1 . F o r each a d m itte d s a n d each
i < s, Xi -< (x0, x l t . . . , x s). 2. F o r each a d m itte d s an d each i <> s , if
x -< x*, th e n x -< (x0, x x, . . . , x s). 3. x -< y o n ly as required b y 1 a n d 2.
W e define a fu n ctio n of a n e n t it y x an d a n a tu ra l n u m b er i, w h ich
g iv e s th e predecessors o f a successor e n tity , th u s:
, , w / x <( if x X (x 0) x 1 ( . . x, ) a n d i < s,
' ” ^ | x, otherw ise.
W e n o w sp e c ify for th e rest o f th is c h a p te r th a t th e n u m b er r + 1

of th e zeros 00, 0 X, . . . , 0r sh all b e th irteen , an d w e n am e th e m as fo llo w s:
D, &, V, V, 3, =, + , •, ', 0, a, ,.
W e fu rth er sp e c ify th a t s a d m it th e v a lu e s 0 , 1 an d 2 . T h is co m p le tes th e

d efin ition of our gen eralized a rith m e tic as a d o m a in o f a b stra c t o b je c ts
w h ich ca n b e recogn ized an d d istin gu ish ed from one an o th er as in d iv id u a ls
b y th e m ode of th eir gen eration.
W e n ow h a v e to fix h o w our fo rm al sy ste m (as o rig in a lly in tro d u ced in
248 THE ARITHMETIZATION OF METAMATHEMATICS CH. X
C h a p ter IV ) is to b e represen ted in th e gen eralized arith m etic. T h is w ill

consist in g iv in g a correlation of en tities to th e o b je cts of th a t fo rm al
system . T h o se form al o b je c ts were ex p la in ed in § 16 as co n sistin g of form al
sym b o ls, fin ite sequences of form al sy m b o ls (called ‘form al exp ression s’),
and fin ite sequences of form al expressions. I t is n o t n ecessary to correlate
an e n t it y to e v e r y form al o b je c t, b u t o n ly to th ose form al o b je cts w h ich
are sign ifican t for th e m eta m a th e m a tics. F o r exam p le, an a s y n ta c tic a l
expression such as ((0 V 0 0 = w ill h a v e no e n tity correlated to it. A lso
there w ill be en tities w h ich are correlates of no form al o b ject.
T o th e first eleven of th e form al sy m b o ls listed in § 16 w e correlate
th e re sp ective en tities D , & , V, - i , V , 3, = , +> •, ', 0, w h ich w e are
d esig n a tin g n ow b y th e sam e sym b o ls, i.e. w e correlate r e sp e c tiv e ly th e
first eleven of th e zeros of th e gen eralized arith m etic.
T o th e v a ria b les a , b } c, d , . . . of th e form al sy ste m , w e correlate,
re sp e c tiv e ly , th e en tities
a> (i> a ), (,, a)), (,, (,, (,, a))), ...

(som etim es w ritte n a , a lf a n, a m, . . . ) , i.e. th e tw e lfth zero, an d th e
fu rth er en tities o b ta in e d th en ce b y repeated a p p licatio n s of th e su c
cessor operatio n of th e gen eralized a rith m etic w ith s = 1 an d w ith th e
th irte e n th zero as th e first predecessor.
T o term s an d form ulas such as r + s , r', r = s , A & B, n A , V x A (x )
we correlate th e en titie s ( + , r, s), (', r), ( = , r , s), (&, A , B ), (-i,A ),
(V ,x ,A (x )), re s p e c tiv e ly , w here r, s, A , B , x, A (x ) are n o w to be th e en tities
correlated to th e g iv e n r, s, A , B , x , A (x ), i.e. we repeat th e correlation
procedure on th e g iv e n r, s, A , B , x, A (x ).
E xample 1. The e n t it y co rrelated to th e form u la 3/>(—i =0) is

(3,
B u t h ereafter in d e sig n a tin g these en tities, e x c e p t w hen w e w ish to

em p h asize th eir stru ctu re (i.e. m ode of generation) as en tities, w e sh all
use th e form er expressions. F o r ex am p le, w h en V , x , A (x ) are en tities,
w e m a y w rite th eir successor (V, x , A (x)) as “ V x A ( x ) ” ; a n d w h en + ,
r, s are en tities, w e m a y w rite ( + , r, s) as “ r + s ” . T h is w a y of d esign a tin g
en tities correlated to o b je c ts of th e form al sy ste m w ill m ak e our s ta te
m en ts a b o u t th e en tities read as our form er sta te m e n ts a b o u t th e o b je cts
of th e form al syste m .
F u rth erm o re, in d ealin g w ith th e gen eralized arith m etic, it is co n ven ien t
to c a ll th e e n tity V x A ( x ) correlated to a form ula (i.e. th e e n t it y (V, x, A (x)))
sim p ly a “ fo rm u la ” , th e e n t it y r + s a “ te rm ” , e tc .; an d th e sy ste m of
th ese en tities “ th e form al sy ste m as a generalized a r ith m e tic ” in co n tra st
§50 A GENERALIZED ARITHMETIC 249
to th e form al s y ste m as o rigin a lly described, w h ich w e ca n d istin gu ish
w h en n ecessary as th e “ form al lin gu istic s y s te m ” .
P roofs an d d ed u ctio n s w ill b e represen ted b y en tities correspon din g to
th e m in th eir tree form (end § 24) rath er th a n to th em as fin ite sequences
of form ulas. T h u s, to g e t th e e n t it y correspon ding to a d e d u ctio n in
sequence form w ith a g iv e n an alysis, let th a t d ed u ctio n first b e p u t in to
tree form . In tree form it h as one of th e three form s
P p Q
D,
d ; D,
w here D is a form ula, an d P an d Q are d ed u ctio n s in tree form . In th e

gen eralized arith m etic, we con strue th ese to be th e e n tities (D),
(D, P ), (D, P , Q), re sp ectively, a t th e sam e tim e o f course co n stru in g D
P , Q to be th e en tities co rrelated to th e lin gu istic o b je cts D , P , Q . (In
p articu lar, a form u la D , an d a d ed u ctio n co n sistin g of th e single form u la
D , b ecom e differen t e n titie s ; th e e n t it y w h ic h th e la tte r becom es is th e
successor w ith s = 0 o f th e e n t it y w h ich th e form er becom es.)
E xample 2. T h e d ed u ctio n (6) of § 2 1 , after b e in g re w ritten in tree

form as in E x a m p le 1 § 24, b ecom es an e n t it y w h ic h w e ca n w rite as
(8 , (7, (1), (6 )), (5, (3, (1), (2)), (4))) w here th e num bers 1 — 8 a b b r e v ia te
th e form ulas o f (6 ) § 21 n o w considered as entities.
T h is co m p le tes th e correlation o f e n tities to th e sign ifica n t fo rm al

lin gu istic o b jects. D is tin c t en tities are co rrelated to d istin ct fo rm al
lin gu istic o b je c ts (exce p t in th e case o f tw o in e ssen tia lly d ifferin g proofs
or d ed u ctio n s in seq uence form w h ich b ecom e th e sam e in tree form ).
T h e proof o f th is, for th e case o f term s an d form ulas, d epen d s on th e
uniqueness o f th e scopes o f th e operators in th e term s an d form u las as
form al lin gu istic expressions (§ 17).
In th u s go in g o ve r from th e fo rm al lin gu istic sy ste m to th e gen eralized
arith m etic, w e h a v e effe cte d tw o ch an ges, eith er o f w h ich co u ld h a v e been
effected sep a ra tely . T h e first o f these is in th e stru ctu re a ttr ib u te d to th e
form al o b jects. In th e lin g u istic represen tation term s an d form ulas w ere
fin ite sequences of fo rm al sym b o ls, in w h ich th e sign ifica n t p a rts h a d
to b e recogn ized as subsequences, w h ile in th e gen eralized a r ith m e tic
th e y are co n stru cte d d ir e c tly o u t of th e sign ifica n t p a rts b y tn e gen eralized
successor operation. In th e la tte r th e a n a ly sis of expressions in to th eir
sign ifican t p a rts (includ ing th e use of parentheses) is tran sferred from
th e form al le v e l to th e e x p o sitio n of th e m e ta m a th e m a tics, w here w e d o
not let it w o rry us.
F o r ex am p le, consider (A) D (B) as a form u la of th e lin gu istic sy ste m ,
w here A an d B are form ulas (i.e. “ A ” an d “ B ” are m e ta m a th e m a tic a l

letters d esign a tin g form ulas). W e are th en ta lk in g e x p lic itly a b o u t a
fin ite sequence of sym b o ls, w ith th e four parentheses sh ow n b ein g
sy m b o ls of th e sequence, an d th e sequences d e sig n a te d b y " A ” an d “ B "
occu rrin g in th e in d ic a te d p osition in th e sequence (§ 16).
O n th e o th er h an d , consider (A) D (B) as a form u la in th e sy ste m
o f en tities, w here A an d B are likew ise form ulas. T h e n “ (A) D (B )”
m eans (D, A , B ), w h ich is th e successor of th e three en tities D, A, B.
T h e parentheses in “ (A) D (B )” an d th e parentheses an d com m as in
“ (D , A , B ) ” are n o t form al o b jects, b u t o n ly p art of our in tu itiv e n o
ta tio n for n a m in g th e e n t it y under consideration.
A s th e second ch an ge, w e are now ta k in g a differen t v ie w o f th e role
o f th e sy m b o lism in p resen tin g th e form al syste m . In C h a p te r I V , th e
o b je c ts of th e fo rm al s y s te m w ere considered to b e lin gu istic sy m b o ls or
m arks, an d o th er lin g u istic o b je c ts co n stru cted from such. In s tu d y in g
th e m w e h a d in p rin ciple to w a tc h th e d istin ctio n b e tw e e n an object an d
a name or designation for th e o b je ct, an d b etw een th e mention o f an ex
pression (as itself th e o b je c t und er consideration) an d th e use of it (in
d e sig n a tin g an o th er o b je c t or expressin g a proposition). T h is has been
em p h asized b y F rege (1893 p. 4), C arn ap (1934 pp. 153— 160) an d Q u in e
(1940 pp. 23— 37). T o m ak e a sta te m e n t a b o u t an o b je ct, o rd in arily a
n am e for th e o b je c t is used. (A n oth er m eth od, som etim es ap p licab le, is to
p o in t to th e o b je ct, or to use a lin gu istic co n stru ctio n w h ich am o u n ts to
p o in tin g to th e o b je ct, i.e. calls a tte n tio n to it in stea d of n a m in g it.) W e
d o n o t sp eak our friend, b u t w e sp eak th e nam e of our friend. I t is n o t
lik e ly t h a t w e sh all m ista k e our friend Jo h n for a sequence of four letters,
b u t in m e ta m a th e m a tic s as tre a te d in th e p reced in g ch ap ters w e d id
h a v e to be careful because w e were discussing o b je cts w h ich w ere th e m
selves lin gu istic.
O n e m eth o d of o b ta in in g nam es for lin gu istic o b je c ts is to p la ce th em
in q u o ta tio n m arks. T h e n am e of Jo h n is “ J o h n ” , an d th e nam e of th e
n a m e of J o h n is “ “ J o h n ” ” . T h e nam e of th e nam e of Jo h n consists of four
le tte rs enclosed in one set of q u o ta tio n m a r k s ; th e second set of q u o ta tio n
m a rk s used a b o v e is e m p lo y e d in n a m in g th a t.
A second m eth o d is th e use of separate m e ta m a th e m a tic a l letters
an d 'e x p re ssio n s as nam es for th e lin gu istic o b jects.
I t need h o t b e s tr ic tly fo rb id d en th a t a specim en o f a lin gu istic o b je c t
b e used as n am e o f th e o b je c t; th e n th e o b je c t h as tw o uses, its use as
th e o b je c t of s tu d y , an d its use as n am e of itself. In th e la tte r use, it is
called autonymous.
§51 RECURSIVE METAMATHEMATICAL DEFINITIONS 251
T h e m eth o d w e used in th e p reced in g ch ap ters w as a co m b in a tio n of
t h e second an d th ird m eth ods.
T h is problem of d esign atio n , w h ich is trou blesom e to tre a t e x p lic itly ,
is ex tran eo u s to th e m e ta m a th e m a tic s as m a th e m a tics. T h e issue ca n b e
a v o id e d b y u sin g o n ly nam es o f th e form al o b jects, an d not cla im in g to
e x h ib it th e o b je cts th em selves. W e fin d it co n ven ien t to do th is in th e
gen eralized arith m etic, considering n ow “ D ” , " V ” , " V x A ( x ) ” as nam es
o f certain o b je cts (the nam es are th e expressions inside th e q u o ta tio n
m arks), rath er th a n as th e o b je cts them selves. W e refrain from sp e cify in g
w h a t th e o b je cts are, o th er th a n th a t th e y b elo n g to a d om ain of a b stra c t
o b je c ts arran ged in a ce rta in w a y in relation to one an oth er, w h ich w e
are ca llin g en tities (cf. § 8 ). T h e o b je c ts nam ed co u ld b e form al sym b o ls,
fo rm al expressions, etc. in th e sense of C h a p te r I V , th o u g h w e n ow le a v e
th is open as irrelevan t for th e m e ta m a th e m a tics. (W hile w e ca n th u s
a v o id th e problem of d esign atio n in our m eta m a th e m a tics, it w o u ld h a v e
t o be faced in discussing th e ap p lica tio n of th e m e ta m a th e m a tic s to a
p a rticu la r lin gu istic system .)
B y go in g o ver from th e co n cep tio n of th e form al sy ste m in term s of
fo rm al sym b o ls, tre a te d as if th e y w ere m arks on paper, to an a b stra c t
sy s te m of o b jects, our m e ta m a th e m a tic s (i.e. th e s tu d y of th e form al
sy stem ) becom es a b ran ch of pure n um ber th e o r y en tire ly on a p ar
c o n c e p tu a lly w ith th e arith m e tic of th e n a tu ral num bers an d sim ilar
m a th e m a tic a l disciplines.
R emark 1. The usual co n ve n tio n or p ractice in inform al m a th e

m a tic a l w ritin g is to w rite all th e sym b o ls w ith o u t quotes, so th a t w hen
a sy m b o l is b ein g m en tio n ed rath er th a n used it is au to n y m o u s. In th is
b o o k our p ractice is to e m p lo y q u o tes s y s te m a tic a lly in m e ta m a th e -
m a tic a l passages to d istin gu ish th e m en tio n of m e ta m a th e m a tic a l
expressions from th eir use in d esig n a tin g form al lin gu istic expressions,
a n d elsew here o n ly for em phasis.
§ 51. Recursive metamathematical definitions. T h e a r ith m e tiz a -

tio n of m e ta m a th e m a tic s w ill b e co m p le ted in § 52 b y m a p p in g th e
gen eralized a rith m etic in to th e o rd in ary a rith m etic of th e n a tu ral
num bers. O u r m ain o b je c tiv e s are to co m p le te th e proof o f th e lem m a
for G o d e l’s theorem , an d to p ro v e T h eorem 3 1. B o th results w ill follow
from th e result th a t a certa in succession of n u m b er-th eo retic pred icates,
o b ta in e d b y th e m a p p in g from m e ta m a th e m a tic a l pred icates, are all
p rim itiv e recursive.
I t is in tu itiv e ly clear w h y these results h old, an d w o u ld h a v e to h old
for a n y form al sy ste m of like structure, w hen w e consider th e n a tu re of
th e definitions of th e m e ta m a th e m a tic a l p redicates in th e gen eralized
arith m etic.
B eca u se th e en tities are gen erated from zeros b y successor operations,
p red icates an d fu n ctio n s ca n be defined o ver th e en tities b y recursion.
W e use th is id ea now, w ith o u t sto p p in g to sta te an a c cu ra te d efin ition o f
*p rim itiv e re cu rsive’ for th e gen eralized arith m etic.
W e g iv e b elo w a series of th irteen d efinitions of m e ta m a th e m a tic a l
p redicates. E a c h defin ition is g iv e n b y listin g th e cases in w h ich th e p red
ica te is to be true. (A few clauses are starred for reference from § 52.)
E ach d efin ition is either e x p licit (the p red icate b ein g d efin ed n o t
ap p earin g in a n y of th e d efin in g clauses), or co n stitu te s a p rim itiv e
recursion (the v a lu e of th e p red icate for a g iv e n e n t it y d ep en d in g on
v a lu e s of itself for im m e d ia te ly preced in g en tities, or sim ilarly w ith
param eters), e x c e p t th a t in D n 5 a n d D n l 1 th e recursion is on tw o va ria b le s
sim u lta n e o u sly (one a n u m b er va ria b le in D n l l ) . F o r th e discussion of
a m e ta m a th e m a tic a l d efinition not o f th is n ature, cf. § 53.
T h e reader should v e r ify t h a t th e definitions do define th e p red icates
nam ed , as w e kn o w th em from earlier sections of th e b o o k (C h apter I V
an d § § 4 1 , 5 0 ) .
D efinitions of metamathematical predicates for the formal

NUMBER-THEORETIC SYSTEM AS A GENERALIZED ARITHMETIC
Dnl. y is a num eral. (A b b r e v ia tio n : 9?(y).)
1. y X 0.
2. y X n ' (i.e. y X (', n), cf. § 50), where n is a numeral.
Dn2. y is a variable. (A b b r e v ia tio n : 93 (y).)
1. y X a.
2. y X x, (i.e. y X („ x)), w here x is a variable.
Dn3. y is a term. (A b b re v ia tio n : %(y)-)
1. y X 0.
2. y is a variab le.
3— 5. y X r+ s or r-s, w here r an d s are terms, y X r', w here
r is a term.
D n4. D is a form ula. (A b b re v ia tio n : $ (D ).)
1. D X r = s , w here r an d s are term s.
2— 5. D X A D B , A & B or A V B , w here A an d B are form ulas .
D X "iA , w here A is a form ula.
§51 R E C U R S IV E M E T A M A T H E M A T IC A L D E F IN IT IO N S 253
6— 7. D X V x A ( x ) or 3 x A (x ), w here x is a va ria b le , an d A (x ) is a
form ula.
D n5. (t is a term , x is a v a ria b le, E is a term or form ula, and) D com es
from E b y th e substitution o f t for (the free occurrences of) x .
(A b b re v ia tio n : @ (D , E , t, x).)
1 . t is a term , x is a v a ria b le, E X x , an d D X t.
2— 3. t is a term , x i s a va riab le, E is 0 or a v a ria b le x x , an d D X E.
4— 5. E is a term or form ula, an d E is (e0, ex) a n d D is (e0, d x), w here
e 0 X , an d @ (d x, elf t, x) (so t is a term , an d x is a va riab le).
E is a term or form ula, an d E is (e0, elt e2) a n d D is (e0, d lt d 2),
w here e 0 X V or 3, ©(d-^ ev t, x) an d © (d 2, e2, t, x).
6 — 7. t is a term , an d E is (V, y , e2) an d D is (V, y , d 2), w h e re y is a
va ria b le, e 2 is a form ula, an d either y X x a n d @ (d2, e2, t, x),
or y X x an d D X E . S im ila rly for 3.
D n6. (E is a term or form ula, x is a v a ria b le, and) E contains x free .

(A b b re v ia tio n : ® g ( E , x).)
1 . E is a term or form ula, x is a v a riab le, an d @ (E , E , 0, x).
D n7. (t is i term , x is a v a ria b le, E is a form ula, and) t is free for x in E .

(A b b r e v ia tio n : $ ( t , x , E )). By recursion on E, With seven
clauses corresponding to 1— 7 D n 4 . F o r e x a m p le :
6 . t is a term , x is a v a ria b le, an d E X V y A ( y ) , w here y is a
va ria b le, A (y ) is a form ula, an d either E does n o t co n ta in
x free, or t is free for x in A (y ) an d t does n o t co n ta in y free.
D n 8. D is an axiom . (A b b r e v ia tio n : 21(D).)

1— 10. D X A D (B D A ), w here A an d B a r e form ulas (A x io m S ch em a
la). S im ila rly for A x io m S c h e m a ta lb , 3, 4a, 4b, 5a, 5b, 6 , 7, 8.
(N o t e : W e sep arate A x io m S ch em a 10 in to tw o cases (Clauses
11 an d 13), acco rd in g as th e A (x ) co n ta in s th e x free or not.
S im ila rly for A x io m S ch em a 1 1 .)
* 1 1 — 12. T h ere exists a t such th a t D X V x A (x ) D A (t), w here A (x )
co n tain s x free, t is free for x in A (x ), an d @ (A (t), A (x ); t, x).
(N o t e : T h a t x is a v a ria b le, A (x ) is a form ula, a n d t is a term ,
is in clu d ed here in th e stip u la tio n “t is free for x in A ( x ) ” .
In s te a d of presupposin g th e co n ven tio n of § 18 b y w h ich A (t)
stan d s for th e result of s u b s titu tin g t for x in A (x ), w e m a k e it
e x p lic it b y “ © (A ( t ) , A (x), t , x) ’ \) S im ila rly for A x io m S c h em a 1 1.
13— 14. D X VxA D A , w here x is a v a ria b le, A is a form ula, an d A
does n o t co n ta in x free. S im ila rly for A x io m S ch em a 1 1 .
15. D X A(0) & V x (A (x ) D A (x ')) D A (x ), w here A (x) is a form u la,
@ (A(0), A (x ), 0, x) an d @ (A (x '), A (x ), x ', x).
16— 23. D X a ' = b ' D a = b (A x io m 14). S im ila rly for A x io m s 15— 2 1.
D n 9. D is an immediate consequence of E . (A b b r e v ia tio n : (£(D, E ).)

1— 2. E X C D A (x ) an d D X C D V x A (x ), w here x is a va ria b le ,
A (x) an d C are form ulas, an d C does n o t co n ta in x free (R u le 9).
S im ila rly for R u le 12.
D n lO . D is an immediate consequence of E an d F .
(A b b re v ia tio n : g ( D , E , F).)
1. D an d E are form ulas, an d F X E D D (R u le 2).
D n 11. x is th e numeral for th e n a tu ral num ber %.

(A b b r e v ia tio n : 9?u(x, x).)
1. x X 0 an d x = 0.
2. x X n ' an d # = n', where 9?u(n, n).
D n l 2 . Y is a proof . (A b b re v ia tio n : ^ f( Y ) .)
1. Y X (D), where D is an axiom .
2. Y X (D, P), w here P is a proof, an d D is an im m e d ia te con
sequence of { P } 0.
3. Y X (D, P , Q), w here P an d Q are proofs , an d D is an im m e
d ia te con sequence of { P } 0 an d {Q }0.
D n l3 . A(a) is a form ula, ^ is a n a tu ral num ber, an d Y is a proof of th e

form u la A (x ) (as a p red icate of A ( a ) , Y ).
(A b b r e v ia tio n : $ f(A (tf), #, Y ).)
* 1. A(a) co n tain s a free, $ f ( Y ) , an d there is an x such th a t 9 iu (x, x)
an d @ ({ Y } 0, A (a), x, a).
2. A (a) does not co n ta in a free, ^ f ( Y ) an d { Y } 0 X A (a).
D n l 3 a . F o r each n d istin ct v a riab les x 1#. . . , x n an d form u la A ( x l f . . . , x w) :

Y is a proof of A ( x v . . . , x n). (A b b re v ia tio n :
$ fx i,...,x M,A(xlf...,xM) (Xl>• • •9xn>Y) or . . ., Xn,Y).)
Sim ilarly.
§ 52. Godel numbering. We n ow co m p lete th e a rith m etiz a tio n

o f m e ta m a th e m a tic s b y represen tin g th e generalized a rith m etic w ith in
th e arith m e tic of th e n a tu ra l num bers. F irst, w e correlate d istin ct odd
num bers to th e zero en tities, t h u s :
3 & V -i V 3 = + •
/ 0 a i
3 5 7 9 11 13 15 17 19 21 23 25 27
§52 GODEL NUMBERING 255
T h en , w h e n eve r x 0, . . . , x s are en tities to w h ich re sp ective num bers
xQf. . . , xs h a v e a lre a d y been correlated, w e correlate to th e successor
e n t it y (x0, . . . , x 8) th e n u m b er p %*-, . .-p 8s ( # 1 8 § 45).
B y a m a th e m a tic a l in d u c tio n correspon ding to th e d efin itio n of e n t it y
a n a tu ra l num ber > 0 is th e r e b y correlated to each e n tity . T h is n u m b er
w e ca ll th e Godel number of th e e n tity , or s a y t h a t it represents th e e n tity
(or fo rm al lin gu istic o b je c t to w h ich th e e n tity in tu rn is correlated).
Since o n ly ev e n num bers are correlated to successor e n tities (because
p 0 = 2 an d x0 ^ 0), an d b ecau se a g iv e n p o sitiv e in teger has th e form
Po°* • • * *Pss (xo> •••> % » > 0) f ° r a t m o st one s an d x 0, . . . , x s, d istin ct
num bers are correlated to d istin ct entities.
E xample 1. The G o d el n u m b er of 3i>(-i i>=^0), or as an e n t it y
(3. (,. a), (-., ( = . (,. a), 0))), is

Since x { •... 'px8s for 0 <, i < s, en tities x an d y in th e re latio n
ship x -< y are a lw a y s represen ted b y n a tu ra l num bers x an d y in th e
relationship x < y. (H o w ever x < y m a y h o ld for pairs of num bers x
and y correlated to en tities x an d y w ith o u t x -< y h o ld in g ; e.g. 3 < 5
bu t not 3 -< &.)
If x is an e n t it y o f th e successor form (x 0, . . x„) w ith s > i, an d x
is th e G o d el n um ber o f x , th e G o d el n u m b er of th e predecessor {x}< is
(*),. ( # 1 9 § 4 5 ).
W h e n w e pass from en tities to their G o d el num bers, a p red icate or
fu n ctio n o f en tities b ecom es a p red icate or fu n ctio n o f G o d el num bers.
A n u m b er-th eo retic p red icate or fu n ctio n o b ta in ed b y e x te n d in g th e
definition o f th e la tte r to a ll n a tu ra l num bers w e s a y corresponds to th e
original pred icate. In p articu lar, in th e case of a p red ic a te ^ ( x j , . . . , x „),
w e shall u n d erstan d b y the corresponding n u m b er-th eo retic p red icate
P (x 1, . . . , % „ ) th a t one w h ich is o b ta in ed b y ta k in g th e v a lu e to b e f
w henever n o t all o f x 1, . . . , x „ are G o d el n u m b e r s; i.e.
P (x v . . . , x n) = {x v . . . . x n are G o d el n um bers o f en tities x 1(. . . , x n, an d

$(Xi>. . x„)}.
S im ilarly w hen som e o f th e v a riab les o f th e original p red icate a lre a d y
range o ver n a tu ra l num bers (e.g. D n l 1 § 51).
Lemma 19. F or each of the predicates defined by D n l — D n l 3, Dnl 3a,

the corresponding number-theoretic predicate is p rim itive recursive.
P roof. T o illu stra te th e m eth o d , let us tre a t D n 3 , assum in g D n 2
alrea d y treated .
25 6 THE ARITHMETIZATION OF METAMATHEMATICS CH. X
W e can w rite D n 3 s y m b o lic a lly as follow s.

2(y) = yxo
v 3S(y)
(0 v [y x (+, {y}1( {y}2) & £({y}i) & st({y}2)]
v [y X (• ,{yli, ( y } 2) & £({y)i) & S((y}2)]
v [y x
(' .{yM&SKyW]-
T h e f i ve d is ju n c tiv e m em bers correspond to th e fiv e clauses of D n 3 as g iv e n
in § 5 1. F o r th e th ird clause w e o bserve th a t if y X r-j-s, i.e. y X ( + , r >s)>
for a n y en tities r a n d s , th en r X { y )i, s X { y } 2. T h e fa c t th a t y X ( + , r > s)
for som e en tities r an d s is th en expressed b y y X ( + , { y K, ( y } 2)-
In ( 1 ) let us replace th e zero en tities 0 b y th eir G o d el num bers
17, 19, 2 1 , 23, th e assum ed p red ic a te 9S(y) b y its correspon ding n u m b er-
th eo retic p red icate V (y), X by = , successor en tities ( x0, b y
Po0' ■ ■ ■ 'Pss’ predecessors ( y ) , by (y){, an d w rite T (y) in stea d of %{y)
for th e p red icate b ein g defined. T h is lead s fo rm a lly to th e fo llo w in g
n u m b er-th eo retic eq u ivalen ce.
T (y) = y=23
VF(y)
(2) V [ y = 2 17-3(!/)l-5 (!/)2 & r ( ( y ) x) & T ((y )2)]
V [ y = 2 19-3<v>i-5(v>* & r ( ( y ) x) & T ((y )2)]
V [ y = 2 21-3(!/)l & 7'((y)1)].
N o w (2) defines a p red icate T (y) b y co u rse-o f-va lu es recursion in th e
arith m e tic of th e n a tu r a l num bers, since (y)< < y for y ^ 0; an d b y
# G w ith # # 2 , 3, 14, 19, A , C, D an d our h y p o th e sis t h a t V is p rim itiv e
recursive, T (y) is p rim itive recursive (cf. E x a m p le 3 § 46).
I t rem ains to p ro v e t h a t th e p red icate T (y) d efined b y (2 ) is th e
n u m b er-th eo retic p red icate correspon ding to %{y)- F o r th is purpose, w e
p ro v e tw o propositions b y co u rse-o f-va lu es in d u ctio n on y :
(a) I f T (y) (b y (2)), then y is the Godel number of an entity y such that
£ ( y ) (b y ( 1 )).
(b) I f %{y) (b y (1)), and y is the Godel number of y , then T {y) (b y (2 )).
P roofs, (a) B y (2), T (y ) is tru e o n ly w hen one of th e d isju n c tiv e

m em bers (or “ cla u se s” ) on th e righ t of (2 ) is tru e ; so w e h a v e fiv e cases
to tre a t. Case 2 : V (y). T h e n , since V is th e p red icate correspon ding t o 93,
y is th e G o d el n u m b er of a v a ria b le y , i.e. 93( y ) ; an d b y th e correspon ding
clau se of (1), 2 ( y ) . C ase 3: y = 217-3<^-5(^ & r ( ( y ) x) & T ((y)2). Then
(y)x. (y )2 < y ; so from r ( ( y ) x) & r ( ( y ) a) b y the h y p o th e sis of th e in d u ctio n
on y, (y)t an d (y )2 are G o d el num bers of en tities r an d s such t h a t $ (r)
an d %(s). Then y (= 217 • 3(w)l • 5lvh) is th e G o d el n um ber of ( + , r, s).
§52 GODEL NUMBERING 257
C allin g th e la tte r y , th e n r X {y } i, s X { y } 2; so y X ( + , { y } i, { y } 2) &

2 ( { y } i) & % ( { y } 2)» an d b y th e re sp ective clause of ( 1 ), %(y)- T h e proof
of (b) is sim ilar, w ith cases from th e d isju n c tiv e m em bers (or “ clau se s” )
on th e rig h t of ( 1 ).
T h e o th er d efin ition s of p red icates in th e list D n l — D n l 3 , D n l 3 a are
h an d led sim ilarly, except as n o ted b elo w for th e recursions on tw o
va riab les an d for th e starred clauses. Af t e r tra n sla tin g (as from ( 1) to (2)),
th e n u m b er-th eo retic p re d ic a te d efin ed is tru e o n ly for G o d el n um bers
as argu m en ts, since in each clause of th e original d efin itio n each v a ria b le
e n tity is required to s a tis fy an earlier p red icate of th e list, or is a su c
cessor of en tities w h ich are fix e d or m u st s a tis fy such p red icates or th e
p red icate b e in g defined. F o r ex am p le , w e tra n sla te C lause lD n 9 as
e = 2 3,3 (e)l-5(e)a &d = 2 3.3 (e)'-5 ^ 3(d)2,1-5W2 & V ((d)2A)

F ((e)2) & *"((«),) & C F ((* ) 1((rf)2jl),
&
w here “(d)21” a b b re v ia te s “ ((d)2) 1,\ N o te th a t CF(e, x) is n o t q u ite th e

n u m b er-th eo retic p red icate correspon ding to g g j( E , x), since it is true w h en
e or x is n o t a G o d el num ber. B u t th is does n o t m a tte r here, since (e)t
occurs also in F((e)^) an d (d)2 l also in V ((d)2 l ), correspon ding to th e
stip u latio n s in lD n 9 th a t C is a form u la an d x is a va riab le.
D n 5 an d D n l 1 are recursions (of a sim ple kind) on tw o va riab les. F o r
D n 5 , e.g., let T (z, t, x) == {z = 2d-3e where S (d } e, t, x)}. T h e n T satisfies
a co u rse-o f-va lu es recursion on th e one v a ria b le z, an d S(d, e, t, x) ^
T (2d-3et t, x).
I t rem ains to consider th e starred clauses 11 an d 1 2D n 8, an d 1 D n 13
an d 1 D n l3 a . W e tra n sla te H D n 8 .as
(Et)t<d[ d = 2 3-32l l -3(<?),’1-5Wl’2-5<d>s & C F ((d)h2> (d)1A)

& F (t, (d)ifi, (d) 12 ) & S((^)2> (d)it2> Wi , i ) ] »
an d use # E . T h e b o u n d t < d is ju stified , since w hen A (x ) co n tain s
x free, t -< A (t) < V x A ( x ) D A (t) X D . We tran slate lDnl3 as
C F (a, 25) & P f(y) & (E n)n<y[N u(n, x) & S ((y )0, a, n, 25)].
T h is co m p letes th e proof of L e m m a 19. T h e gist of th is proof is th a t

th e p rim itiv e recursions in th e gen eralized arith m e tic b ecom e course-of-
va lu es recursions in th e o rd in ary a rith m etic , since th e G o d el n u m b erin g
preserves th e order relation sh ips alth o u g h it d estro ys th e relation sh ips of
im m ed ia te succession. I t is n ecessary to v e r ify th a t th e range of each
v a ria b le w h ich w e elected to in tro d u ce w ith a q u a n tifie r rath er th a n as a
fu n ctio n of th e in d ep en d en t v a ria b les of th e p red icate b ein g d efin ed
(e.g. t in H D n 8 an d x in 1 D i l i 3) can b e restricted .
Lemma 20. U nder the Godel num bering of this section, the predicates
A (a, b) and B (a , c) of Lem m a 21 § 42 are prim itive recursive .
P roof. B y Lem m a 19, since w e can express A {a, b) an d B (a , c) in
term s of th e n u m b er-th eo retic p red icate P f(a, x, y) corresponding to
th e p red ic a te ?Pf(A(^), x, Y ) of D n l 3 thus,
A (a, b) == P f(a, a , b)t B (a, c) = P f( 29-3«, a , c).
L e m m a 21 § 42 follow s n ow usin g C o ro llary T h eo rem 27 § 49,
T heorem 3 1. F or any given form ula A(a) (cf . D n l3 ) , the predicate

‘A ( x ) is provable* (as predicate of x, where x is the num eral for x) is ex
pressible in the form (E y )R (x , y) where R is prim itive recursive ; i.e. given a
form ula A (a), a p rim itive recursive predicate R (x, y) can be found such that
(E y)R (x} y) - b A ( x ) .
(S im ilarly for A ( x x, . . x n); cf. D n l 3 a.)
P roof. A form u la is p ro v a b le, if an d o n ly if there e x ists a proof of it.
Let a be th e G o d el n u m b er of th e p a rticu la r form ula A(a) of th e h yp o th esis
of th e theorem , an d set
R (x, y) = P f(a , x, y).
(For A ( x x, . . x n), set R (xv . . . , x n,y ) = P fA(xlf y).)
E xample 2. L e t @ (E , t, x) X {th e result of s u b s titu tin g t for x in
E , if t is a term , x is a v a ria b le, an d E is a term or fo rm u la; otherw ise, E } ;
an d le t yiu(x) X {th e n u m eral x for th e n a tu ra l n u m ber x}. T h ese m e ta -
m a th e m a tic a l fu n ctio n s @ (E , t, x) an d -Ku(#) ca n b e d efined b y recursion,
sim ilarly to th e p red icates @ (D , E , t, x) (Dn5) an d 9 iu (x , x) ( Dnl 1 ); an d
th e correspon ding n u m b er-th eo retic fu n ctio n s S(ef t, x) ( = e , w hen e , t , x
are n o t all G o d e l num bers) an d N u(x) are p rim itiv e recursive.
*§ 53. Inductive and recursive definitions. The defin ition s of

‘te rm ’ an d ‘fo rm u la ’ w ere g iv e n o rigin ally in § 17 as in d u c tiv e definitions.
O th e r ex am p les of in d u c tiv e defin ition s are th e d efin itio n of ‘n a tu ral
n u m b e r’ (§ 6 ), of ‘p ro v a b le fo rm u la ’ in th e first version (§ 19), of ‘p rim itiv e
recursive fu n c tio n ’ if phrased sim ilarly (§ 43), an d of ‘e n t it y ’ (§ 50). T h e
results of th is ch a p ter m a y b e vie w e d in term s of th e relationship b etw ee n
in d u c tiv e an d recursive d efinitions. W e b egin w ith a few rem arks a b o u t
in d u c tiv e defin ition s gen erally.
Inductive* d efin itio n s occur in tw o differen t roles, an d w e c a ll th e m
fundam ental an d non-fundam ental acco rd in gly. To w h ich c a te g o r y a
g iv e n in d u c tiv e d efin ition belo n gs m a y v a r y w ith th e c o n te x t or th e o r y in
w h ich it is b ein g used.
§53 INDUCTIVE AND RECURSIVE DEFINITIONS 25 9
For th e generalized arith m etic, th e d efin ition of 'e n tity ' is th e
fu n d am en tal in d u c tiv e defin ition . I t establish es th e dom ain of o b je c ts for
th e arith m etic. A n e n t it y is th ereafter und erstood to b e g iv e n , w h en and
o n ly w hen its m ode o f gen eratio n u nder th e in d u c tiv e d efin itio n of
'e n tity ' is g iven .
Then th e n o n -fu n d am en tal in d u c tiv e definitions, such as th ose of
'te rm ', 'fo rm u la' an d 'p ro v a b le form u la', a p p ly to o b je cts alrea d y kn o w n
in their s ta tu s as en tities. T h ese d efinitions each defin e a class of entities,
i.e. a subclass of th e entities. W e ca n a sk in th e a rith m etic w h eth er or not
a g iv e n e n tity belongs to th e su b class; an d w e can associate w ith th e
subclass a p red icate ta k in g th e v a lu e t for an e n t it y b elo n gin g to th e
subclass an d f for an e n tity n o t belo n gin g. W e ca n regard th e non
fu n d am en tal in d u c tiv e d efin itio n s as definitions of th ese predicates.
Thus th e fu n d a m en ta l in d u c tiv e d efin itio n establishes th e range
of a va ria b le , o ver w h ich one m a y su b se q u e n tly define pred icates b y non
fu n d am en tal in d u c tiv e d efinitions (including as a special case th e co n sta n t
p red icate t).
T h e m ann er in w h ich a n o n -fu n d am en tal in d u c tiv e d efin ition defines
a p red icate is th e follow ing. T h e direct clauses te ll us certain o b je cts for
w h ich th e p red icate ta k e s th e v a lu e t. T h e extremal clause sa y s th a t those
are th e o n ly o b je cts for w h ich th e v a lu e is t, so th a t w e can a ttr ib u te th e
v a lu e f w h en ever w e are able to see th a t th e d irect clauses do n ot require
th e v a lu e to be t.
T h e direct clauses g e n era lly includ e basic clauses , each of w h ich tells
us o u trig h t (or under h y p o th ese s in v o lv in g o n ly p re v io u sly defined
predicates) th a t th e v a lu e is t for a certain o b je ct, an d inductive clauses,
each of w h ich tells us th a t, if th e v a lu e is t for ce rta in o b je cts (and p o ssib ly
under h yp o th ese s in v o lv in g p rev io u sly d efined pred icates), th en th e v a lu e
is t for th e o b je c t related to those in a g iv e n w a y . (If b a sic clauses are
m issing, th en th e p red icate ta k e s th e v a lu e f for all argu m en ts. I f in
d u c tiv e clauses are m issing, th e d efin ition is sim p ly an e x p licit d efin ition
b y cases.)
N o n -fu n d a m e n ta l in d u c tiv e definitions ca n also b e used to define
predicates of m ore th a n one va riab le. S u ch a d efin ition som etim es has
the form of an in d u c tiv e defin ition of a class d ep en d in g on a param eter
(e.g. th a t of ' < ' § 6 ), an d in gen eral it can b e considered as th e in d u c tiv e
defin ition of a class of ordered ^ -tuples.
In d u c tiv e definitions, b o th fu n d a m en ta l an d n o n -fu n d am en tal, ju s tify
corresponding form s of 'proof b y m a th e m a tica l in d u ctio n '. T h o se cor
responding to th e in d u ctiv e definitions of 'n a tu ra l n u m b er' an d ' ent i t y '
h a v e a lre a d y been m en tio n ed (§§ 7, 50). A s an o th er exam p le, th e p rin ciple

o f in d u ctio n correspon ding to th e in d u c tiv e d efin itio n of ‘p ro v a b le for
m u la ' is t h i s : If e v e r y a x io m h as a certain p ro p erty , an d if w h en ever th e
prem ises for a form al inference h a v e th e p ro p e rty so does th e conclusion,
th en e v e r y p ro v a b le form u la has th e p ro p erty. (This in d u ctio n prin ciple
co u ld h a v e been in v o k e d for th e proof of T h eo rem 9 § 28, in stea d of
co u rse-o f-va lu es in d u ctio n on th e len g th of a proof. T h e n L e m m a s 12a
an d 12 b th em selves c o n stitu te th e basis an d in d u ctio n steps of th e proof.)
In th e sam e w a y , fu n d a m en ta l in d u c tiv e d efin itio n s (under th e con
v e n tio n th a t d iffe re n tly gen erated o b je cts are d istin ct) ju s tify ‘d efin ition s
b y in d u ctio n ' or ‘recursive d efinitions' of a fu n ctio n o ver th e dom ain
estab lish ed b y th e in d u c tiv e definition. (B u t a recursive procedure cor
respon din g to a n o n -fu n d am en tal in d u c tiv e d efin ition of a class w h ich
allow s an o b je c t to be recogn ized as in th e class b y d ifferen t successions
o f a p p licatio n s of th e d irect clauses m a y lead to m ore th a n one fu n ctio n
v a lu e for such an o b je ct, e.g. “ 9 (A) = 0 if A is an axio m , 9 (A) = 9 (B) + 1
if A is an im m e d ia te consequence of B , an d 9 (A) = 9 (B) + ? ( Q + 1
if A is an im m e d ia te con sequence of B an d C " does not define a sin gle
v a lu e d fu n ctio n 9 from p ro v ab le form ulas to n a tu ra l num bers.)
P red icates can be in tro d u ced from re cu rsively d efined fu n ctio n s
servin g as th eir represen tin g function s, or o ften as w e h a v e seen in § 51
b y recursive procedures d irectly.
In recursions (of such kinds as w e h a v e been con siderin g for th e sim ple
a rith m e tic in C h a p te r I X ) th e v a lu e of a fu n ctio n or pred icate, s a y of
one v a ria b le, for a n y g iv e n non-zero argu m en t is d eterm in ed from th e
va lu e s of th e sam e for o n ly argu m en ts preceding th e g iv e n a rgu m en t in
term s of th e order of gen eratio n of th e d om ain b y th e fu n d a m e n ta l
in d u c tiv e d efin ition . T h is h as th e consequence th a t w e can p ro ve b y a
correspon ding in d u ctio n th a t a re cu rsively d efined p red icate ta k e s th e
v a lu e t or f for e v e r y argu m en t. T h u s th e law of th e exclu d ed m id d le is
p ro v ed in tu itio n istic a lly to a p p ly to e v e ry proposition ta k e n as v a lu e of
a re cu rsively d efin ed predicate.
T h is is n o t in general so for an in d u c tiv e ly d efin ed pred icate, as th e use
o f th e e x trem a l clause to assign th e v a lu e f w h en ever th e d irect clauses
do n o t assign th e v a lu e t m a y le a v e us w ith o u t th e k n o w led ge of e ffe c tiv e
m eails to determ ine w h ich is th e v a lu e for a n y g iv e n argu m en t (s).
In a special case it is so, n a m e ly (e.g. for th e in d u c tiv e d efin ition of a
class) w h en th e order in w h ic h th e in d u c tiv e clauses in trod u ce m em bers of
th e class agrees w ith th e order of gen eration of th e o b je cts under th e fu n
d a m e n ta l in d u c tiv e defin ition . I t is in d u ctiv e d efinitions of th is sort
§53 IN D U C TIV E AN D R E C U R SIV E D EFIN ITIO N S 261
w h ich w e ca n recast as recursive d efin itio n s, as w e d id for those of ‘te r m ’

an d ‘fo rm u la’ in § 5 1.
A n in d u c tiv e d efin itio n n o t of th is sort is th a t of ‘p ro v a b le fo rm u la ’
in th e first version (§ 19). In th e second version , th e d efin itio n is set ap art
from th e m e ta m a th e m a tic a l d efin itio n s of § 51 b y th e fa c t th a t an ex iste n
tia l q u a n tifie r “ th ere e x ists a p roof Y ” is used w ith o u t a b o u n d b ein g
kn ow n for its va ria b le Y (in co n tra st to 1 1 — 12 D n 8 an d l D n !3 ) . (Cf. § 30.)
F ro m th is second version, w e o b ta in as th e corresponding n u m b er-th eo
retic p red icate (E y)[P f(y) & (y)Q= d ] (cf. D n l2 ) , w h ich is of th e form
E y)R (d, y)
(. w here R is p rim itiv e recursive.
W h e n th e form o f an in d u c tiv e d efin itio n (w ith ele m e n ta ry d irect
clauses) is specified in a n a tu ra l w ay, th e pred icates P (x v ..., x n)
d efin ab le b y use of in d u c tiv e d efinitions in th e n a tu ra l n um ber a rith m etic
are e x a c tly those expressible in th e form E y)R (xlt . . . , x n,y )
( w ith R
p rim itiv e recursive. T h e proof ca n b e g iv e n b y an exten sio n of th e a b o v e
m eth ods, as w as su ggested in K le e n e 1943 pp. 66— 6 7 ; or b y an o th er
m eth o d , in d ic a te d in K le en e 1944* p. 48 .
Ch a p t e r X I
GENERAL R E C U R S IV E F U N C T IO N S
§ 54. Formal calculation of primitive recursive functions.

E ach of th e sch e m a ta (I) — (V) of § 43 considered in tu itiv e ly is an
op eratio n d efin in g a fu n ctio n 9 from zero or m ore g iv e n function s.
A c t u a lly w e s ta te d th e sch em ata b y m eans o f equation s.
L e t us re view th e m an n er in w h ich th e eq u a tio n s d efine th e fu n ctio n
9 , to see w h e th er w e ca n n o t a n a ly z e our use of th e m in d eterm in in g
p a rticu la r va lu e s of 9 in to fo rm al operations.
E xample 1. L e t x b e a g iv e n fun ction , tw o o f w hose va lu e s are:
1. X(0,4) = 7. 2. x (1<7) = 7.
Let 9 be in trod u ced b y S ch em a (Va) w ith q= 4, th u s:
3. 9 (0) = 4. 4. 9 (y') = x (y, 9 (y))-

In § 43 w e co n vin ced ourselves th a t, for a n y n um ber y , th e recursion
eq u a tio n s for 9 determ ine th e corresponding v a lu e 9 (y) of 9 , if th e v a lu e s
of x are a lre a d y determ ined. In particu lar, w ith th e tw o va lu es of x ju s t
g iv e n , th e reasoning of § 43 tells us th a t 9 (2 ) = 7. W e now a sk : W h a t
sorts o f form al inferences w ill enable us to d educe th e eq u a tio n " 9 (2 ) = 7 ”
from E q u a tio n s 1 — 4 ?
S u b s titu tin g “ 0 ” for “ y ” in E q u a tio n 4 :
5. 9 ( 0 = X(0, 9(0)).
R e p la c in g " 9 (0 )” in th e righ t m em ber of E q u a tio n 5 b y “ 4 ” from
E q u a tio n 3:
6. 9(l) = x(0, 4).

F o u r m ore step s of these tw o sorts com plete th e d e d u c tio n :
7. 9 ( 1 ) = 7 — R ep la c em en t, 6 , 1 .
8 . 9 (2 ) = >X(1, 9 ( 1 )) — S u b stitu tio n , 4.
9. 9 (2 ) = x ( l, 7) — R ep la c em en t, 8 , 7.
10 . 9 (2 ) = 7 — R ep lacem en t, 9, 2 .
T h u s a su b stitu tio n an d a replacem en t operatio n suffice for th e d e-
d u ctio n from th e g iv e n eq u a tio n s 1 — 4 of th e e q u a tio n “ <p(2) = 7"
w h ich sta te s th a t th e v a lu e of <p for th e a rgu m en t 2 is 7. M oreover,
q u ite e v id e n tly , no succession of th ese tw o sorts of inferences ca n lead
from E q u a tio n s 1 — 4 to a n y oth er eq u a tio n w hose le ft m em ber is
“ 9 (2 )” an d w hose righ t m em ber is a num eral.
E xample 2. Su p p ose n o w m ore p a rtic u la r ly th a t x is th e co n sta n t

fu n c tio n (C|) defin ed b y th e eq u a tio n
—2- x(y»z) = 7,
w hile 9 is defined from x as in E x a m p le 1 . (Th en 9 is p rim itiv e recursive,
w ith x> 9 as a p rim itiv e recursive description.) N o w w e ca n d ed u ce
E q u a tio n s 1 an d 2 from E q u a tio n — 2 b y su b stitu tio n , as fo llo w s :
— 1. x(0> z) = 7 — S u b st., — 2 . 0 . x 0 » * ) = 7 — S u b st., — 2.

1 . x(0, 4) = 7 — S u b st., - 1 . 2 . x0> 7) = 7 — S u b st., 0.
C o m b in in g these tw o d ed u ctio n s w ith t h a t o f E x a m p le 1 , w e o b ta in
a d ed u ctio n of " 9 (2 ) = 7” from th e three eq u a tio n s — 2, 3, 4 d efin in g 9 a b
initio.
In th ese exam p les w e h a v e been con sid erin g q u estion s o f a form al

kind, w ith o u t h a v in g e x p lic itly set u p a form al sy ste m in a d va n ce . W e
sh all n o w estab lish a su ita b le form al sy ste m , an d m a k e our d iscussion
rigorous as a m e ta m a th e m a tic a l discussion referring to th is sy ste m .
T h e new form al sy ste m w e ca ll th e form alism (or form al system ) of
recursive functions. W e d escribe it n o w in th e lin gu istic m ann er, a n d
later (§ 56) as a gen eralized arith m etic.
The form al sym bols of th e sy ste m are as fo llo w s: = (equals), '(successor),
0 (zero), a, b ,c , . . . , a v a2> . . . (variables for n a tu ra l num bers), f, g , h t
• • •> fv f*> • • • (function letters, i.e. sy m b o ls for u n specified n u m b er-
th eo retic fu nction s), (,) (parentheses), an d , (com m a). A (p o ten tially)
in fin ite list o f v a riab les an d of fu n ctio n letters are supposed to b e g iv e n .
W e c a ll f, g , h, ..., f v f%, . . . " fu n c tio n le tte r s " ra th er th a n " fu n c tio n
sy m b o ls " here to distin gu ish th em from '. A ls o th e n am e is ap p ro p riate,
because t h e y w ill h a v e a role sim ilar to t h a t of th e p red icate le tters in th e
pure p red icate calculus, i.e. t h e y are to b e in terp reted as exp ressin g d if
ferent fu n ctio n s a t d ifferen t tim es, b u t there is no p o stu la te d rule of su b
s titu tio n for th em . In discussions in w h ic h ft g, h, etc. express fix e d
fu n ction s, an d are n o t b ein g tre a te d d iffe re n tly from e.g. we m ay
ca ll th e m " fu n c tio n s y m b o ls" .
T h e form al expressions 0 , O', 0 " , ... w e c a ll num erals. A s before
(§ 41), w e a b b re v ia te th e m r e sp e c tiv e ly b y " 0 " , " 1 " , " 2 " , . . . ; a n d w e
264 G E N E R A L RECU RSIV E FUNCTIONS CH. X I
co n tin u e to use th e co n v e n tio n w h e re b y “ x ” , “ y ” , etc. d esign ate th e

n um erals for th e n a tu ra l num bers d esign ate d r e s p e c tiv e ly b y ‘ V " , “ y ” ,
etc.
T h e term s are 0, th e va ria b les, an d expressions o f th e form r' w here r
is a term , or f(rx, . . . , r n) w here f is a fu n ctio n le tte r an d rv . . . , r n are
term s (n > 0 , o m ittin g th e parentheses w hen n = 0).
A form al expression r — s w here r an d s are term s is an eq u a tio n .
T h e eq u a tio n s are th e o n ly “ fo rm u las” for th is syste m . By a system
o f eq u a tion s w e m ean a fin ite sequence e0, . . . , es o f eq u a tio n s (not e m p ty ,
unless otherw ise stated ).
N o ax io m s w ill b e p ro v id e d ; an d w e sh all define o n ly ‘d ed u cib ility"
but not ‘p ro v ab ility".
T h e rules o f inference sh all be a one-prem ise su b stitu tio n rule R 1
an d a tw o -p rem ise replacem en t rule R 2 , as follow s.
R 1 : to pass from an e q u a tio n d co n ta in in g a v a ria b le y to th e eq u a tio n

w h ic h results from d b y s u b stitu tin g a num eral y for y .
R 2 : to p ass from an eq u a tio n r — s c o n ta in in g no v a riab les (the m a jo r

p r e m ise ) an d an e q u a tio n h (zl9 . . . , z P) = z w here h is a fu n ctio n
le tte r an d zv ..., z P, z are num erals (the m in o r p r e m ise ) to th e
eq u a tio n w h ich results from r = s b y re p lacin g an occurrence of
h (zv ..., z v) in s (or several such occurrences sim u ltan eou sly) b y z.
A d ed u ctio n o f an eq u a tio n e (the end equ a tion of th e deduction) fro m a
sy ste m (or set, p o ssib ly infin ite) E of eq u atio n s is to be one in tree form
(end § 24); i.e. it shall h a v e one of th e three form s
c w here c is one of th e eq u a tio n s of E ,
W w here W is a d ed u ctio n from E , an d c is an im m e d ia te con sequence

c b y R 1 o f th e e n d eq u a tio n of W , or
W X w here W an d X are d ed u ctio n s from E , an d c is an im m e d ia te

C con sequence b y R 2 of th e en d eq u atio n s of W an d X re sp e ctiv e ly .
I f th ere is a d e d u ctio n o f e from E , th e n e is d ed u cib le fro m E (in sym b o ls,

E h e).
E xample 2 (continued). T r a n sla tin g E q u a tio n s — 2 to 10 in to th e

n ew form alism (using th e form al fu n ctio n letters h for th e in tu itiv e
fu n ctio n letters “ <p” , “ x ” , an d b , c for th e in tu itiv e n u m b er va ria b le s
“y”, “ z ” ) , an d go in g o ve r from th e sequence to th e tree form , w e o b ta in
th e fo llo w in g figure.
§54 FORM AL C A L C U L A T IO N 265
4 2. f(b ')= h (b , f[b))
5. f ( l) = h ( 0 , f(0)) 3. f{0 )= 4 —21. h(b, c ) = 7
(a) 1 — 1 . h(0, c ) = 7
4 i- f(b ')= h (b ,f{b )) 6 . f ( l) = h { 0 ,4 ) l . A ( 0 , 4 )= 7 — 2 2. h { b ,c ) = 7

8. f ( 2 ) = h ( l ,f ( \ ) ) 7 .f(l)= 7 0. h ( l ,c ) = 7
9. f ( 2 ) = h ( l,7 ) 2. A ( l , 7 ) = 7
10. f { 2 ) = 7
T h is is a d ed u ctio n o f th e e q u a tio n / (2 ) = 7 from th e sy ste m of eq u a tio n s
h ( b ,c ) = 7, } (bi)
(b) />
(0 )= 4 , 1
f{ b ')= h { b ,f{ b )). /
(b2)
E xample 1 (continued). Consider th e p a rt of th e tree (a) w ith o u t the

(occurrences of) eq u a tio n s —2lt — 1 , — 22, 0, ca ll it (a2). T h is is a d e d u ctio n
of f( 2 ) = 7 from A ( 0 , 4 ) = 7 , h { \ ,7 ) = 7 an d (b2).
R emark 1. O n e m a y a sk : W h y ta k e th e tro u b le o f tr a n s la tin g ? The

reasons are of course th e sam e as th ose w h ic h d ic ta te d our use of a special
sym b o lism in th e n u m b er-th eo retic form al sy ste m o f C h a p te r I V , d istin c t
from th e sy m b o lism of in tu itiv e n um ber th eo ry. R e c a p it u la tin g : T h e
tran slatio n n o w from “ 9 ” , “ x ” * to f 9 h, b, c is ca lled for u nd er
th e lin gu istic co n cep tio n o f a form alism , becau se th e form al sy m b o ls m u st
be from a fix e d preassigned list of sy m b o ls considered in th e m e ta m a th e
m atics as m ere m arks. T o use th ese a u to n y m o u sly (§§ 50, 16) w ith o u t
risk of confusion, m o st of th e m sh ould b e from a special a lp h a b e t. U n d e r
th e co n cep tio n o f a form alism as a gen eralized a rith m etic , w e m a y if
w e w ish consider t h a t f is (i.e. t h a t “ f ” is a n am e for) “ 9 ” (but th e n it is
a nam e for “ 9 ” o n ly, a n d n o t som etim es for “ <J/\ etc.). T h e “ f ” used e.g.
in th e d efin itio n o f 'te rm ', on th e oth er h an d , is a n am e for an u n sp ecified
one of f t g , h, etc. (and if w e consider t h a t “ f ” is a nam e for “ <p” , “g ” f°r
“ h 99 for “x ’> e tc ., th e n “ f ” is a n am e for an u n specified one of
“ 9” , 'Y ', Y . etc.). T h e n e t result o f tra n sla tin g is to enable us to t a lk
clearly a b o u t th e sy m b o ls as d istin gu ish ed from th e fu n ction s, num bers,
etc. w ith o u t g o in g to th e tro u b le of c o n tin u a lly u sin g q u o ta tio n m ark s an d
related d evices (such as Q u in e's “ corners” 1940 ).
Suppose t h a t a fu n ctio n 9 of n v a ria b les has been d efin ed in tu itiv e ly

from l (;> 0 ) fu n ctio n s ^ 1 , of m v .. . , m x va ria b le s, re sp e ctiv e ly .
B efore w e ca n discuss w h e th er a g iv e n sy ste m E o f eq u a tio n s “ d efin es”
9 from in th e form alism , w e m u st s a y w h ich fu n ctio n le tters
f, g i, . . gi are to express 9 , . . . , <ph re sp ectively.
266 G E N E R AL RE C U R SIV E FUN CTIONS CH. X I
It is convenient to em ploy conventions w hich m ake these letters rec

ognizable from E itself. It suffices to consider system s E in w hich th e
first (leftm ost) sym bol of the last equation is a function letter f ; and w e
call this th e p rin cip a l function letter of E , and use it to express 9. The
d istinct function letters w hich occur in right m em bers of equations of
E but not in left m em bers w e call the given function letters of E . It suffices
to consider system s E in w hich there are l of these; and w e use them
gx, . . . , g t in the order of their occurrence in the preassigned list of function
letters to express t|/x, . respectively. It suffices to use system s
E in w hich f, gx, . . . , g l occur on ly in term s form ed w ith n, m v ..
argum ents, respectively (or w e could for the present purpose count as
d istinct a function letter w ith each num ber of argum ents, as w e did
predicate letters in § 31 ); we then call the other function letters
(if any) occurring in equations of E the a u x ilia ry function letters of E.
W hen / > 0 , w e shall need to have available as "assum ption equations'"
n ot on ly th e system E w hich w e associate w ith the schem e defining 9
from ..., but also equations giving the values of the functions
L et denote the set of the equations
S tfrv • • • >ym j)=*y where <h(yx, . . . , y mj) = y, for j = 1 and all m f
tuples y lt . . . , y mj of natural num bers {ylf . . . , y mj, y being th e num erals
for the num bers y x, >. . , y mj, y). This set of equations is infinite, w hen
l > 0 (unless m 1 + * .. + m x = 0); em pty, w hen l = 0.
W e bring these ideas together in the follow ing m etam athem atical
definition. A system E of equations defines 9 recursively in (or from)
^x, . . ^ 1 , if for each w-tuple x v . , x n of natural num bers:
EgJ”*^, E (- f(Xx, . . x n) = x , where f is the principal function letter
of E , g x, . . . , gi are the given function letters of E in order of their
occurrence in th e preassigned list of function letters, and x is a num eral,
if and only if cp(xv . . . , x n) — x.
In other words, E defines 9 recursively from §lf if (for f,
gi, as described): (- ffo , . . xn)= x where
x x, . . . , x n, x are num erals, if {completeness property) and only if {con
sistency property) f(x x, . . . , x w) = x £ E^.
E xamples 1 and 2 (continued). The principal function letter of (b)
is f ; h is ah auxiliary function letter; and quite evid en tly (b) defines
recursively the function 9 (where 9(y) = 4 if y = 0 , and 9(3/) = 7 if
y > 0). The system (bx) defines x ( = Cf) recursively; and (b2) defines
9 recursively from x> w ith h as th e given function letter.
T heorem II. I f 9 is prim itive recursive in ..., then there is a
system E of equations which defines 9 recursively from
I t is q u ite ev id e n t th a t w e o b ta in such a sy ste m E b y tra n sla tin g th e
sch em a a p p lica tio n s for a n y p rim itiv e recursive d eriv a tio n cpk
o f 9 from <J^, . . i nt o th e form alism , if w e choose th e fu n ctio n letters
s u ita b ly , an d if w e first arrange (if necessary) to h a v e each used in som e
sch em a ap p lica tio n (since our co n ven tio n s p ro v id e th a t <J/X, . . . , be
expressed b y fu n ctio n letters gv ..., g h re sp e ctiv e ly , all occurring in E ).
H o w e v e r w e shall g iv e th e m e ta m a th e m a tic a l a n a ly sis in d eta il, w ith fiv e
lem m as, to la y th e b asis for b rief tre a tm e n t of sim ilar m a tte rs later.
B y th e prin cipal branch of a d ed u ctio n , w e m ean th e b ran ch w h ich ,
traced u p w ard from th e en d eq u atio n , co n tain s th e m ajo r prem ise a t each
a p p lica tio n of R 2 . T h e eq u a tio n w h ich stan d s a t th e to p of th e p rin cip al
b ran ch w e c a ll th e prin cipal equation. T h e d ed u ctio n s (occurring as p arts
o f th e g iv e n ded uction) of th e m inor prem ises for th e ap p licatio n s of R 2
alo n g th e p rin cip al b ran ch w e c a ll th e contributory deductions.
E xample 2 (con tin u ed ). The p rin cip al b ran ch of (a), read dow n
w ards, con sists of th e eq u a tio n s n um bered 4lt 8 , 9, 10 . T h e p rin cip al
eq u a tio n is f( b f)= h (b , f(b)). T h e tw o co n trib u to ry d ed u ctio n s are th e
trees en d in g w ith 7 an d w ith 2 .
The p rin cip al b ran ch of a d ed u ctio n o f an eq u a tio n of th e form

f ( x x, . . . , x n) = x w here f is a fu n ctio n le tte r an d x v . . . , x n, x are
num erals, read dow nw ards, consists of zero or m ore a p p licatio n s of R l ,
follow ed b y zero or m ore a p p lica tio n s of R 2 . T h e a p p licatio n s of R l
su b stitu te resp ective num erals for th e va riab les of th e p rin cip al eq u a tio n ,
u n til th e le ft m em ber b ecom es f ( x x, . . x M). T h e a p p lica tio n s o f R 2
replace p arts h (zv ..., z P) in th e righ t m em ber (originally present or
resultin g in th e course of th ese replacem ents) b y re sp ective n um erals z ,
u n til th e rig h t m em ber becom es x .
B y th e identical schema , w e m ean :
<p (x v ...,* * ) = ...,* « ) .

L emma Ha. I f is an im m ediate dependent of
9 by one of
Schemata (I) — (V) or the identical schema, then the system E of equations
obtained by translating the inform al equations of the schema application into
the form alism {with any appropriate choice of the function letters) defines
9 recursively from tylf . . . ,
P roof of Lemma I la . T h e proofs parallel th e in form al reasoning
b y w h ich w e recognized t h a t th e sch em a ap p licatio n s define th e fu n ctio n s.
268 GENERAL RECURSIVE FUNCTIONS CH. XI
S chema (Vb). The E is o f th e form f( 0 , x 2, . . x n) = g ( x 2, . . x n)>

f ( y ,» x 2, . . . , x „ ) = h ( y , f (y, x 2, . . ., x„), x 2, . . . . x„). Choose any n—1
num bers x 2, . . . , x„. W e use in d u ctio n on xv B a s is : x x — 0 . T h e n
9 {xx, = ^{x2, . . . , x n). N o w E ^ ( E b f(Xi, . . . , x n) = x fo r
x = <p[xv . . . , x n), usin g th e first eq u a tio n of E as th e prin cip al equation *
a n d th e eq u a tio n g ( x 2, . . . , x n) — x from E | * as a m inor prem ise for R 2.
M oreover, to deduce i( x v ..., x n) = x for a n y num eral x , w e m u st use
th e sam e p rin cip al eq u a tio n an d th e sam e su b stitu tio n s of num erals b y
R 1 (apart from their order), since from no other eq u a tio n of E |£ , E
a n d b y no o th er su b stitu tio n s can w e o b ta in an e q u a tio n w ith f ( x x, . . . , x n)
(for % = 0 ) as left m em ber. T h e n th e replacem ent step b y R 2 is also

u n iq u e ly d eterm in ed. So E ^ , E bf ( x x, . . . , x n) — x w ith x a n u m eral
o n ly for x = <p(%, . . . , x n). I nduction s t e p : xx = y'. S im ilarly.
L emma lib . L e t D be a set o f e q u a tio n s [fin ite or in fin ite ), F be a
sy s te m o f e q u a tio n s w h o se le ft m em b ers c o n ta in n o fu n c tio n le tte r s w h ic h
o ccu r in [e q u a tio n s o f) D, a n d g be a fu n c tio n le tt e r o c c u r r in g in D. T h e n
D, F h g (y lf .. -> y m )= y w h ere yv y a re n u m e r a ls , o n ly i f
D i- g(yi» ■ ■ ■ ,ym)=y-
P roof of L emma lib . Consider a n y d ed u ctio n from D, F of an
eq u a tio n of th e described form g ^ , . . . , y w) = y . W e p ro v e b y co u rse-
o f-v a lu e s in d u ctio n on th e h e ig h t t of th is d ed u ctio n t h a t o n ly eq u a tio n s
o f D are used in it as assu m p tio n eq u atio n s (i.e. o ccur in it a t th e to p s o f
branches). T h e p rin cipal eq u a tio n is an eq u a tio n of D , since its first sy m b o l
is g, w h ich occurs in D an d hence n o t in th e le ft m em ber of an e q u a tio n
of F . E a c h co n trib u to ry d ed u ctio n is of h e igh t < t , an d term in ates in
an eq u a tio n h (zlt ..., z P) = z , where h is a fu n ctio n le tte r occu rrin g on
th e righ t side of th e p rin cip al eq u atio n , an d th u s in D . T h erefore, b y th e
h yp o th e sis of th e in d u ctio n , th e co n trib u to ry d ed u ctio n s use o n ly as
su m p tio n eq u a tio n s from D .
Lemma lie. L e t D F
a n d be a s in L e m m a lib. L e t G be th e set o f th e
e q u a tio n s o f th e fo r m g[y lt . . . , y m)z= zy } w here g is a fu n c tio n le tte r o c
c u r r in g in b o th D a n d F y lt . . . , y m, y
a n d a r e n u m e r a ls , w h ic h a r e d e d u c ib le
fr o m D. L e t f be a fu n c tio n le tte r n o t o c c u r r in g in D. T h e n
D, F b f(Xj, ..., xw
)=x w h ere x v ...,xn, x a re n u m e r a ls , o n ly i f
G, F b f(xu ..., xn)=x.

P roof of L emma H e, b y co u rse-o f-valu es in d u ctio n on th e h e ig h t
t of th e g iv e n d e d u ctio n o f f[ x v ..., xn)=x from D, F. T h e p rin cip al
e q u a tio n is an eq u a tio n of F , becau se its first sy m b o l is f, w h ich does n o t
§54 FORM AL CALCU LATION 269
occur in any equation of D. A ny m inor prem ise along the principal
branch is of the form h (z1, . . z p) = z where h occurs in the right
m em ber of the principal equation and hence in F. If h occurs in D ,
th en using Lem m a li b h (zv . . . , z P) = z is an equation of the form
g (y v . . . , y m) = y w hich is deducible from D , i.e. h (zv . . . , z P) = z 6 G.
If h does not occur in D , then b y the hyp oth esis of the induction
on t, h (zv . . z P) = z is deducible from G, F.
L emma lid . (1) Let f v .. . , f k be distinct function letters in order of
occurrence in the given lis t of function letters. (2) Let <px = ...,
(3 ) F o r i = Z+ l , . . . , k (k > l) t let E* define recursively from <pjiv
..., (qi >; 0; jn , . . . , j i Qi < i), w ith ft- as the p rin cip a l function letter,
and jiQ' as the given function letters. (4 ) Let the a u x ilia ry function
letters of E i {if any) be distinct from those of each E ?. for j ^ i and from
fi, .. ., ffc. Then f o r i = 1, . . . , k : , E l+ 1 . . . E k j- f ^ , . . . , x ni) = x
where x lf . . ., x ni, x are numerals, if and only if
fi(*i> . . X,H) = X 8 E * \
E xample £ (concluded). Let k == 2 ; l = 0; <px, cp2 be the prim itive re
cursive description 9; and E 1; E 2 be (bj), (b2) interchanging f and h.
E xample 3. Let <px, . . 9 *. (with k = 9 , 1 = 3) be the prim itive
recursive derivation £, y), 0, U \, 0X, U \, G |, 9 of E xam ples 1 §§ 44 and 45 ,
and E { (i = 4 , . . . , 9) be th e equation (ct) :
fM > c > b ) = a , (c4)
f5(a, c, b ) = f 3(fi {a, c, b)), (c«)
(c) fs(a, c, b ) = b , (c«)
f}(a, c, b) = f2(f6(a, c, b), f&(a, c, b)), (ct)
fs(a, c, b ) = 2, (C8)
f 3{a, c, b ) = f 1(fi (a, c, b), f,{a, c, b), fs{a, c, b)). (C9)
P roof of L emma lid . W e easily see, b y general properties of r , th at

E l+1. . .E * h X n*)=x, if f^X i, . . Xn. ) = x S E 9\ W e
prove the converse b y induction on k. B asis : k = l. Then E i+1 . . . E fc
is em pty. The conclusion follow s b y the con stitution of E^1’’’^ (noting
that it contains no prem ise for R1 and no m ajor prem ise for R 2 ).
I nduction step : k > l. B y hyp oth esis of the induction, only those
equations of the form f^ x ^ .. ., x ni) = x for i < k w hich 8 E 9* are
deducible from , E j+1 . . . E fe_1. B y H ypothesis (3 ) for i = k
270 G E N E R A L RECU RSIV E FUNCTIONS CH. XI
(and w ritin g "9” for ‘ V \ “ i
9 s” for ‘W ’, "f” for etc.), o n ly th o se
eq u a tio n s of th e form f ( x lf . . . , x n) = x w h ich £ are d ed u cib le from
an d by Lem m a lie no others b ecom e d ed u cib le
w h en is replaced in th e list of assu m p tio n eq u a tio n s by
Lemma lie . Let (plt . . . , <pk be a finite sequence of functions such that
(pk is 9 and for each i (i — 1 , . . . , & ) , either (A) <pt- is one of the functions
4 i, • • • > 4^ or (B) 9 t- is defined recursively by a system E t of equations from
9iiV •••> WiQi (ft ^ 0; jn> . . jiQi < i). Then there is a system E of
equations which defines cp recursively from 4 i> . . . , 4 ^
P roof of Lemma H e. I f it is n o t a lre a d y th e case t h a t each of th e
<J/s is in tro d u ced under (A) as one of th e <p’s a n d is th ereafter used u n d er
(B) as one o f th e <p/^, . . . , cp ^ for som e <pt-, w e ca n m a k e it so (increasing k)
b y in tro d u cin g som e a p p licatio n s of th e id en tica l sch em a (cf. L e m m a H a ).
T h e n , b y rearrangin g an d renum bering th e <p's an d E / s an d ch a n g in g th e
fu n ctio n letters in th e la tte r (if necessary), w e ca n b rin g a b o u t th e situ a tio n
described in L e m m a l i d , w ith k > /, cp* = 9, an d w ith i lf . . . , f t as th e
g iv e n fu n ctio n letters of E i+1 . . . E * . L e t E b e E i+1 . . . E fc.
P roof of T heorem II. B y L e m m a H a an d th e h yp o th e sis of th e

theorem , th e h yp o th ese s of L e m m a l i e are satisfied.
§ 55. General recursive functions. The sch e m a ta (I)— (V) are

n o t th e o n ly schem es o f d efin itio n of a n u m b er-th eo retic fu n ctio n , a b
in itio or from o th er n u m b er-th eo retic fun ction s, w h ich ca n b e expressed
b y sy ste m s of eq u ation s, u sin g in th e eq u atio n s o n ly fu n ctio n letters, ',
n u m b er va ria b le s an d num erals.
L e t us consider o th er ex am p le s, ca llin g th em all "recu rsio n s” . W e k e ep
th e eq u a tio n s in th e in form al la n g u a g e for th e tim e b e in g ; an d to k e ep
th e m o f th e sort described now , w e elim in ate certa in oth er m odes o f
expression w h ich w ere used in C h a p te r I X , e.g. II ( # B ) , W y< z ( # E )’
cases. ( # F ) .
T h u s (a) of E x a m p le 1 § 46 w e ca n w rite n ow
y) =
rc(0 , 1,
(a) iz{z',y) = (y + ?(*))• 7c(z,y),
<p(y) = 7r(y, y)
§55 GENERAL RECURSIVE FUNCTIONS 271
(defining the auxiliary function n as well as cp), while (a) of Example 2

§ 46 is already in the form under consideration. We showed in § 46 that
these course-of-values recursions are reducible to primitive recursion, i.e.
the same function can be defined by a series of applications of Schemata
(I) - (V).
As another very simple example, consider the recursion
(b) [ z)
* (0* =
l <p(y', *) = ? (y . a(y, *))•
This is not primitive, because the z, instead of being held fixed as a
parameter, has a(y, z) substituted for it in the induction step of the
definition. This recursion too can be reduced to primitive recursion.
Expanding (b) for y — 0, 1, 2, ... (as we expanded (1) in § 43), we find
that the value 9(3/, z) is
<t(0, <j (1, <t(2, ... a(y—3, a(y—2, c(y—l,z)))...))).
Consider the sequence of the numbers z, a(y—1, z), a(y—2, a(y—\,z))f
..., cr(0, a(l, a(2, ... a(y—3, c(y—2, c(y—1, z)))...))), which occur in
building up this value from the inside instead of as (b) gives it to us.
These are the values for u — 0, 1, 2, ..., y of the function |x(u, y, z) defined
by the primitive recursion
/b } f p(o. y >z ) = *.
1 l V-(u', y, z) = a(y—u', [i(M, y, z)).
Since the value for u = y is the same as the value 9(y, z)t
(b2) ? (y . z) = n (y . y, * ) ;
as can also be seen by using induction on u to prove that
(c) \x(u', y \ z) = y, <r(y, z)),
and thence that the 9 defined by (bx) and (b2) satisfies (b).
In a similar manner, Peter (19 3 4 , 19 3 5 a) showed that every recursion
(called “nested”) in which 9 (0 , z) is a given function of z, and 9(3/, z)
is expressed explicitly in terms of y , z, given functions (and constants),
and 9(3;, t) as a function of t, is reducible to primitive recursion.
Are there recursions which are not reducible to primitive recursion; and
in particular can recursion be used to define a function which is not
primitive recursive?
This question arose from a conjecture of Hilbert 1926 on the continuum
problem, and was answered by Ackermann 19 2 8 . Let (b, a) = a-\-bt
^ ( 6, a) == a-b, E,2(bt a) = ab; and let this series of functions be extended
by successive primitive recursions of the form £n/(0 , a) = a , £n,(&', a) =
5 n(5 n'(^» #)» #) (» >: 2 ), so that e.g. ^ (^ a) = aa'"a with 6 exponents.
27 2 G E N E R A L RECU RSIV E FU N CTIONS CH. X I
N o w consider 5 n(b , a) as a fu n ctio n 5(w, 6, a) o f all three variab les. L et a be

th e p rim itiv e recursive fu n ctio n d efin ed thus,
0 if n= 0,
(d) oc(n, a) = 1 if n= 1,
a otherw ise.
T h e n th e fo llo w in g recursion defines 5(w, 6 , a),
( 5 (0 , 6 , a) = a+b,
(e) I 5(n', 0 , a) = <x(n, a),
U « 6', a) = 5(», 5 K , 6, a), a).
T h is is an ex a m p le o f a “ d o u b le recursion” , i.e. one on tw o v a ria b le s
sim u ltan eo u sly. If th e fu n ctio n 5(n, 6 , a) defined b y (e) w ere p rim itiv e
recursive, th en th e fu n ctio n %(a) of one v a ria b le d efin ed e x p lic itly from it
th u s,
(f) Ua) = Z (a ,a ,a ),
w o u ld also be p rim itiv e recursive. A c k e r m a n n ’s in v e stig a tio n show s
t h a t 5 (a) grow s faster w ith increasing a th a n a n y p rim itiv e recursive
fu n ctio n o f a (just as 2 a gro w s faster th a n a n y p o ly n o m ia l in a), i.e. g iv e n
a n y p rim itiv e recursive fu n ctio n <p(a), a n a tu ra l n u m ber c can b e fo u n d
such th a t 5(a) > <p(a) for all a > c. T h u s 5(«), an d hence also 5(», 6 , tf)
(since 5 (#) com es from it b y th e e x p lic it d efin ition (f)), are n o t p rim itiv e
recursive. T h is ex a m p le w as sim p lified b y P e te r 1935 (cf. also H ilb e r t-
B e rn a y s 1934 pp. 330 ff.) an d R a p h a e l R o b in so n 19 4 8 .
A differen t m eth o d w as follow ed b y P e te r 1935 in c o n stru ctin g an o th er
exam p le. T h e class of th e in itia l fu n ction s d efin able b y S c h e m a ta (I) —
(III) is enum erable. T h e n th e class of th e p rim itiv e recursive fu n ctio n s
d efin ab le usin g S ch em a (IV ) or (V) ju s t once is enum erable, since th e
m+ 1-tu p le s <{/, xi, • • Xm for (IV ) or th e pairs <J/, x (or q, x) for (V)
form ed from an enu m erable class are enu m erable ( §1 ) . T h e n th e p rim itiv e
recursive fu n ctio n s d efin ab le usin g S ch em a (IV ) or (V) a second tim e are
en u m erable; an d so on. T h u s th e class of all th e p r im itiv e recursive
fu n ctio n s is enum erable, as w e cou ld also see b y en u m eratin g th e sy ste m s
E for T h eorem I I § 54. In particu lar, th e p rim itiv e recursive fu n ctio n s of
one v a ria b le are enum erable. H en ce b y C a n to r's d iago n a l m e th o d (§ 2)
t h e y ca n n o t com prise a ll th e n u m b er-th eo retic fu n ctio n s of one v a r ia b le ;
an d if
cp0 ( « ) , ? a (« ). • • •
is a n y en u m eratio n of th em allo w in g repetitions (i.e. a n y in fin ite list of
th e m in w h ich each occurs a t least once), th en <pa(#) + l is a n um ber-
theoretic function of one variable not in the enumeration, and so not
primitive recursive. The enumerating function <p(n, a) such that
<p(n, a) = <pn(a) is a function of two variables which is not primitive re
cursive, since <p0(^)+l = ?(a>#) + l. This of course only establishes
that number-theoretic functions <pa(a) + l and cp(n, a) can be found
which are not primitive recursive. What Peter did was to show that, for
a suitable enumeration (with repetitions) of the primitive recursive
functions of one variable, the enumerating function can be defined by a
double recursion (besides applications of Schemata (I) — (V)).
E xample 1. Do double recursions lead to any predicates which are
npt primitive recursive ? Yes, for 1—<p(a, a) takes only 0 and 1 as values,
and cannot occur in the above enumeration, so it is the representing
function of a predicate not primitive recursive. (Skolem 1944.)
P£ter 1936 studies &-fold recursions for every positive integer k. These
comprise primitive recursions for k = 1, double recursions for k = 2,
and so on. She shows that, for each successive k, new functions are ob
tained. Functions definable using (besides explicit definition) recursions
up to order k she calls “^-recursive”. She shows that every 2 -recursive
function is definable by a single double recursion of the form
r 9(0, b) = <p(«, 0) = 1,
(S> l ? ( » ', V) = «(«, b, 2 .
E xample 2 . To settle a point raised in § 45 , suppose 9 is 3 -recursive
but not 2-recursive, and ^ is 2-recursive but not 1-recursive, i.e. not
primitive recursive. Then “if is primitive recursive, then 9 is primitive
recursive’' is vacuously true, but “ 9 is primitive recursive in <]/’ is false,
since that would make 9 2-recursive.
These subjects are treated in Peter’s monograph 1951 (not available
during the writing of the present book).
It is not to be expected that the £-fold recursions with finite k exhaust
the possibilities for defining new functions by recursion. In 1950 Peter
uses “transfinite recursions” (first employed by Ackermann 1940) to
define new functions.
This brings us to the problem, whether we can characterize in any
exact way the notion of any “recursion”, or the class of all “recursive
functions”.
The examples (I) — (V), (a), (b), (e) (and others cited) of schemes of
d efin itio n of a fu n ctio n w h ic h w e h a v e th u s far agreed to ca ll ‘ ‘recursions’ '
possess tw o fe a tu r e s : (i) T h e y are expressed b y eq u atio n s in th e m ann er
w h ich w e a n a ly z e d fo rm a lly (for (I) — (V) p articu larly) in § 54. (ii) T h e y
are defin itio n s b y m a th e m a tic a l in d u ction , in one form or an oth er,
e x c e p t in th e tr iv ia l case w hen th e y are e x p lic it definitions.
T h e ch aracterizatio n of all “ recursive fu n ctio n s” w as accom plish ed in
th e d efin itio n of 'gen eral recursive fu n c tio n ’ b y G o d el 19 3 4 , w h o b u ilt
on a su ggestio n of H erb ran d . T h is d efin ition succeeds b y a b o ld gen eral
iza tio n , w h ich consists in choosing F e a tu re (i) b y itself as th e defin ition .
W e s a y th en th a t a fu n ctio n 9 is general recursive , if there is a sy ste m
E of eq u atio n s w h ich defines it re cu rsively (§ 54, w ith l= 0).
T h is choice m a y seem u n e xp ecte d , since th e w ord “ re cu rsive” has its
root in th e ve rb “ recu r” , an d m a th e m a tic a l in d u ctio n is our m eth o d for
h a n d lin g recurrent processes. T h e m ean in g of th e choice is n o t th a t
F e a tu r e (ii) w ill b e absen t from a n y p articu lar recursion, b u t th a t it
is transferred o u t of th e d efinition itself to th e a p p licatio n of th e d efinition.
T o show b y fin ita r y m eans th a t a g iv e n schem e has F e a tu r e (i), e x c e p t
in triv ia l situ atio n s, one w ill p resu m a b ly h a v e to m ake use of m a th e m a t
ica l in d u ctio n som ehow. B u t in definin g th e t o ta lit y of general recursive
fu n ction s, w e forego th e a tte m p t to ch aracterize in a d v a n c e in w h a t form
th e in tu itiv e principle of in d u ctio n m ust m an ifest itself. ( B y G o d e l’s
theorem § 42 w e k n o w th a t th e a tte m p t a t such a ch a ra cteriza tio n b y
th e form al n u m b er-th eo retic sy ste m is incom plete.)
In s ta tin g th e H erb ran d -G o d el defin ition of gen eral recursive fu n ctio n
e x a c tly , there is som e la titu d e as to th e d etails of th e fo rm alizatio n , so
th a t versions of th e d efin itio n can b e g iv e n w h ich are e q u iv a le n t to
G o d e l’s b u t a b it sim pler (cf. K le en e 19 3 6 , an d 1943 § 8 ). T h e present
version is t h a t of K le en e 19 4 3 , e x c e p t for in con seq u en tial ch an ges in R l
an d R 2 w h ich sim p lify § 56 slig h tly , an d th e inclusion of fu n ction s of 0
va riab les in th e trea tm en t. (To relate th e present tre a tm e n t to K le e n e
19 4 3 , w e n o te: ( 1 ) T h e inclusion of fu n ction s of 0 v a ria b les does n ot alter
th e n otion of general recursiveness for fu n ction s of n > 0 variables. F o r
one can sh ow th a t, if an a u x ilia r y fu n ctio n le tte r h occurs as a term
w ith 0 argu m en ts in th e assu m p tion equation s, all occurrences of th is
term m a y b e ch an ged to k(c), where k is a new fu n ctio n le tte r an d c a
new v a ria b le, w ith o u t a lterin g th e class of th e d ed u cible equ ation s
co n ta in in g o n ly th e p rin cipal fu n ctio n letter. A fte r th is: (2) O n e can
sh ow in a few lines th a t e x a c tly th e sam e eq u atio n s of th e form
f ( x x, . . . , x n) = x , w here f is a fu n ctio n le tte r an d x lf ..., x n, x are n u
m erals, are d ed u cib le from g iv e n assu m ption eq u atio n s b y th e present
R 1 an d R 2 as b y th e R 1 an d R 2 of 1 9 4 3 ; or w ith o n ly a little m ore
trou ble one can ca rry ou t th e tre a tm e n t of §§ 54 an d 56 w ith th e R 1
an d R 2 of 19 4 3 .)
A fu n ctio n 9 is general recursive in fu n ction s ^ 1 , . . . , tylt if there is a
syste m E of eq u atio n s w h ich defines 9 re cu rsively from <\>v . . . , <]>i (§ 54).
T h is in cludes th e defin ition of gen eral recursive fu n ctio n as th e case
1 = 0. For l > 0 (K leen e 194 3 ), w e are u su a lly considering a schem e or
fu n ctio n al 9 = F(^x, . . . , î) (§ 47) w h ich defines a n u m b er-th eo retic
fu n ctio n 9 of n va riab les from tj'z* for any ^ n u m b er-th eo retic
fun ction s ..., of m v .. . , m l va riab les re sp e ctiv e ly , or a n y such
function s su b je ct to som e sta te d restrictions. T h e n if th e E ca n be g iv e n
in d e p e n d e n tly of ^ 1 , . . (for th e fix ed n, l, mv . . . , mf), w e s a y th a t
th e schem e F is general recursive, or th a t 9 is general recursive uniform ly in
. . . , 4^- Sin ce our tre a tm e n t w ill a lw a y s g iv e u n ifo r m ity in th e fy’s
(subject to a n y restrictions stated ), w e u su a lly om it th e w ord “ u n ifo rm ly ”
e x ce p t for em phasis. (U n like th e p rim itiv e recursive case § 47, if th e
original schem e is for som e restrictio n on ^x, . . it is n ot im p lied th a t
th e schem e can n ecessarily be e x te n d e d to a gen eral recursive one d e
finin g a 9 w ith o u t restrictio n on th e <\>v . . . , ^ t.)
U sin g th e present te rm in o lo gy to re sta te th e results of L em m a s I l a
and H e , w e n ow h a v e :
T heorem I I (second version). If 9 is defined from tpi, . by

a succession of applications of general recursive schemes, then 9 is general
recursive in tpx, . . . ,
In particular, Schem ata (I) — (V) are general recursive. H e n c e : I f
9 is prim itive recursive in ^x, . . . , ^ 1, H is general recursive in 4>x, . . . , y*.
A n y prim itive recursive scheme is general recursive. I f 9 is prim itive re
cursive, it is general recursive.
T h e d efin ition of gen eral recursiveness has been sta te d for th e case
th a t th e fu n ctio n 9 is a lrea d y kn ow n, b y in tu itiv e use of th e sam e
equation s w h ich are form alized as th e E , or b y som e oth er m eans. T h is
a n ticip a tes our purpose of sh ow in g th a t va rio u s fu n ctio n s an d schem es,
kn ow n to us in d e p e n d e n tly of th e form alism of recursive fu n ction s, are g e n
eral recursive (as w e h a v e ju st done for th e p rim itive recursive fu n ctio n s
an d sch em es). F o r th e case t h a t th e 9 is not p rev io u sly kn ow n , w e th en
h a v e : A sy ste m E of eq u a tio n s defines re cu rsively a fu n ctio n of n v a r ia
bles from ..., if for each w -tuple xv ..., xn o f n a tu ra l num bers,
there is e x a c tly one num eral x such th a t E ^ ‘“^ , E b f ( x x, . . . , x n)=x,

where f is th e prin cipal fu n ction letter of E , an d g x, . . . , g t are th e g iv e n
276 G E N E R AL RE CU R SIV E FUNCTIONS CH. X I
fu n ctio n letters in order of th eir occurrence in th e g iv e n list of fu n ctio n

letters. I f so, th e fu n ctio n <p w h ich is d efined re c u rsively b y E is th e
fu n ctio n w hose v a lu e cp(%, . . x n) for th e n a tu ral num bers xlt . . xn
as argu m e n ts is th e n a tu ra l n um ber x for w h ich th a t x is th e num eral.
As w ith p rim itive recursiveness (§§ 45, 47), th e n o tio n of gen eral
recursiveness for fu n ctio n s ex te n d s to pred icates an d to m ix e d cases b y
use o f th e represen tin g fu n ctio n s for th e predicates.
§ 56. Arithmetization of the formalism of recursive functions.

T h e form alism of recursive fu n ctio n s w ill be trea ted n ow as a gen eral
ized a r ith m e tic of th e sort described in § 50. W e list b elo w recursive
d efin itio n s for th is gen eralized arith m etic, as in § 5 1 . S im u ltan eo u sly,
w e in d ic a te h o w to pass to th e sim ple a rith m etic b y G od ei num berin g,
as in § 52.
The gen eralized arith m e tic shall have s ix zeros, w ith re sp e ctiv e
G 5 d e l num bers, as follows.
f a
Z eros: = 0 i f
C o rrelated G o d el num bers: 15 21 23 25 27 29
W e p erm it successors to b e form ed of a n y p o sitiv e num ber of en tities,

i.e. all n a tu ra l n um bers are a d m itte d n ow as v a lu e s of s.
A s our list f, g , h, . . . of fu n ctio n letters w e use th e en tities f,
(,, (,, /')), . . . (som etim es w ritte n f, fu . . . ) , co n stru cted from th e zero
e n t it y f in th e sam e m ann er as th e (number) v a riab les from th e zero
e n t it y a.
A term of th e form f(rx, . . . , r n) w here f is a fu n ctio n le tte r an d
rv . . . , r n are term s (n > 0 ) w ill b e represented as th e e n t it y (f, rlf . . . , r w).
So in p a rticu la r th e fu n ctio n le tte r f an d th e term f (rx, . . . , rn) for n = 0,
a lth o u g h u s u a lly w ritte n alike lin g u istica lly, are n o t tre a te d as th e sam e
e n t it y ; in th e gen eralized arith m etic, if th e form er is f, th e la tte r is (f).
(W e s till h a v e a un iq ue e n t it y correlated to each sign ifican t lin gu istic
o b je c t, if w e consider t h a t th e om ission of parentheses, w h en i(rlf . . . , r w)
for n= 0 is w ritte n lin g u istic a lly as “ f ” , is m erely an a b b reviatio n .) F o r
exam p le, th e eq u a tio n f= 0 is th e e n tity ( = , (f), 0).
A sy ste m of eq u atio n s e0, . . . , es is represented as th e e n tity (e0, . . . , es).
In th e passage from th e gen eralized a r ith m e tic to th e n a tu ral num ber
arith m etic, b y G o d ei num berin g, th e clauses usin g a va ria b le n u m b er
s + 1 of predecessors are h an d led w ith th e h elp of an d 20 (§ 45 ).
F o r th ese clauses (and a few oth er definitions), w e ap p en d th e cor
respon din g n u m b er-th eo retic clause (or definition).
§56 ARITHM ETIZATIO N 277
D efinitions of metamathematical predicates and functions for the
FORMALISM OF RECURSIVE FUNCTIONS AS A GENERALIZED ARITHMETIC
Df l . y is a num eral. (A b b re v ia tio n : -Jl(y)-) S a m e as D n l §51.
D f2. y is a variable. (A b b re v ia tio n : 58 (y)-) S a m e as D n2.
D f3 . y is a function letter. (Abbreviation: % 2{y).) Like D f2 .

D f4. y is a term. (A b b r e v ia tio n : Jrrt(y)-)
1. y X 0.
2. y is a variab le.
3. y X r' (i.e. y X (', r)), where r is a term.
4. y x f(rlf . . . , r„) (i.e. y X (f, rv . . . , r„), cf. § 50), w here f
is a fu n ctio n le tter, an d rx, . . . , r n are term s (n > 0).
FL{{y)o) & (i)o<i<mv>T m ((y)i)-
D f5 . z is an equation. ( A b b r e v ia tio n : @q(z).)
1. z X r = s , w here r an d s are term s.
D f6. Z is a system of equations. (A b b r e v ia tio n : @@(Z).)

1. Z X (z0, . . . , z s), where z 0, . . z 3 are equation s.
l h ( z ) > 0 & ( i ) {<lh(s,£:?((z)t).
D f7 . (t is a term , x is a v a ria b le , e is a term or eq u atio n , and) d com es

from e b y th e substitution of t for x . (A b b r e v ia tio n : @ b(d, e, t, x).)
1. t is a term , x is a v a ria b le , e X x , an d d X t.
2— 3. t is a term , x is a va riab le, e is 0 or a v a ria b le X x , an d d X e.
4. t is a term , x is a v a ria b le , an d e is a term or eq u a tio n o f th e
form (e0, e^ . . . , e n) an d d is (d0, d 1( . . . , d m), w h e r e m = n,
e0 X i, d 0 X e0, @ b(d1( elr t, x ), . . . , ©&(d„, e„, t, x).
Tm (t) & V{x) & (Tm (e) V Eq(e)) & lh ( e ) > 0 & lh (tf)= lh (e) &
( ^ 0 ^ 2 7 & (d)0— (e)0 & (f)o<i<lh(e)^((^)*> (e)i> L x )-
D f8. (e is a term or eq u a tio n , x is a va riab le, and) e contains x.

(A b b r e v ia tio n : £ t(e, x).) L ik e D n 6.
D f 9. c is an im m ediate consequence o f d (b y R 1). ( A b b r e v ia tio n : £ n (c, d ).)

*1. d is an eq u a tio n , an d there e x is t a (variable) y an d a num eral
n, such t h a t d co n ta in s y , an d @ b(c, d, n, y).
Eq(d) & (E y )y<d(En)n<c[N (n ) & Ct(d, y) & Sb(c, d, n, y)].
I f d is an eq u a tio n an d Gt(d, y ), th en y is a v a ria b le -< d ;
an d if also @ b(c, d, n, y ), th en n «< c.
D flO . c is an im m ediate consequence of d an d e (by R 2).

(A b b re v ia tio n : ©tt(c, d, e).)
*1. e X h (z 1( . . . , zv)—z, w here h is a fu n ctio n letter, and
278 G E N E R AL RECU RSIVE FUNCTIONS CH. X I
zv . . . , are numerals (p > 0); d is an equation, con

zP, z
taining no variables, call it dx= d 2; and c is of the form
dj^Cg, where, for some term u containing a>
@b(d2, u, h {zlf . . . , z p), a) and ©b(c2, u, z, a).
Eq(e) &FL((e)h0) & {i)0<i<mehrN ^)i.il & ^ 2) & Eq(i) &
t y)y<d c m y) & c = 2 15 *3 (d)l*5 (c)2 & (Eu)u<d[Tm (u) &
Ct{u, 25) & Sb((d)2, u} (e)lt 25) & Sb((c)2, u, (e)2, 25)].
Since a has a smaller Godel number than any function letter,
a has a smaller Godel number than h(z1, ..., z P) ; and hence u
has a smaller G5del number than d.
Df 11. x is the numeral for the natural number x. (Abbreviation: 9fu(x, x).)
Same as Dnl 1.
Dfl2. (Z is a system of equations, and) Y is a deduction from Z (by
R1 and R2). (Abbreviation: 5)(Z, Y).)
1. Z is a system of equations (z0, . . . , z s), and Y X (z*) (i < s ).
SE (z) & (E i)i< m z)[ y = 2 ^ ] .
2. Y X (c, Yj), where Yx is a deduction from Z, and c is an
immediate consequence of {YJ q.
3. Y X (c, Y v Y2), where and Y2 are deductions from Z, and
c is an immediate consequence of {Y1}0 and {Y2}0.
D f 13. Y is a d ed u ctio n from Z of an eq u atio n of th e form f ( x 2, . . . , x n) = x ,
w here f is th e p rin cip al fu n ctio n le tte r of Z , xv . . . , xn are th e
num erals for th e n a tu ral num bers x v . . . , x n, re sp e c tiv e ly ,
an d x is a num eral (as a p red icate of Z , xv . . . , x n, Y , for each
fix e d n> 0). (A b b re v ia tio n : © n(Z, x v . . . , x n,Y ).)
1. Y is a d ed u ctio n from Z, { Y }0 X f ( x 2, . . . , x n) = x , w here
f is a fu n ctio n letter, f X {z s}1>0 if Z X (z 0, . . . , z s), etc.
D (z, y) & lh ((y )0>1) = » ' & F L {{y)Q10) & (y) 0, i , o = ( * W ) - M , o &
N u ((y ) & •••&N u ( M 0,1,«>*«)&^((y)o.*)-

( x, if y is a num eral x (i.e. if 9?u(y, x)),
Df l 4 . SRu-^y) = j
[ th e G o d el n um ber of y , otherw ise.
N u ~ 1(y) = y.xx<yN u (y ,x ).
D f 15. U( Y) = S R u -V W o .j).
T h e n U (Y ) = w h en ever Y is a d ed u ctio n of an eq u a tio n of
th e form r = x w here x is a num eral.
U (y) = N u ~ 1((y)02).
B y th e m eth o d s illu stra te d in § 52, an d th e in d icatio n s a c c o m p a n y in g
th ese defin ition s:
§57 THE (X-OPERATOR 279
Lemma F or each of the predicates and functions defined by

III.
D f l — D f 15, the corresponding number-theoretic predicate or a corresponding
number-theoretic function is prim itive recursive.
§ 57. T h e (x -o p e r a to r, e n u m e r a tio n , d ia g o n a l procedu re. We

shall now begin usin g th e least n u m ber o perator jxy (§ 45) w ith o u t a
bou n d on th e y. T h u s for a n u m b er-th eo retic p red icate R (y) such t h a t
(E y)R (y)f fxyJ?(y) = {the least (natural num ber) y such th a t R {y)}. For
the tim e being, w e use (xyi?(y) o n ly w h en th e e x isten c e c o n d itio n (Zsy)i?(y)
is fulfilled.
T h u s w e h a v e a new sch em a
(V ia) cp(xv ...,*») = y.y[x(xi. y )= 0 ]
for th e defin itio n of a fu n ctio n cp o f n v a ria b les (n > 0 ) from a n y fu n ctio n

/ of n+\ v a riab les such th a t
(la) {xn)(E y)[x(xlt

(*i) • • • y ) = 0].
By ta k in g R {xv . . . , x n,y ) = x ( * i, y )= 0 if x is g iv e n first,

or if R is g iv e n first b y ta k in g 7 to b e th e represen tin g fu n ctio n of R,
w e can also w rite th e sch em a thus,
(V Ib) <p(*i> . . . , x n) = \xyR(xlt y)
as a d efin ition of a fu n ctio n <p from a p red icate R such t h a t
(l b) (* 1) • • • {xn)(E y)R (xv y).
T heorem Schema (V ia ) w ith (l a) holding ((V Ib) with (lb ) holding)

III.
is general recursive. H en ce b y T h eo rem I I : A function <p defined from
functions and predicates Y by applications of Schem ata (I) — (V I) with
( 1 ) holding for the applications of (V I) is general recursive in Y . (Church
1936, K le en e 19 3 6 .)
P roof. W e recast (V I) in to a sch em a (V I') su ita b le for translation

into th e form alism of recursive fu n ctio n s, as fo llo w s :
tc(x 1( y) = n
s<y
x ( * i. • • • , * » . s),
(VI|)
(VI') x{z', 0, y) = y,
. . . , x n) = x{n(xu
?(*i> y), n (xv . . . , x n, y'), y),
(vi;)
where t ( w , v, y) is an auxiliary function w hich is left undefined for u = 0
or v > 0. (For another sim ple m ethod, see K leene 1943.)
First let us convince ourselves inform ally that (VI') is equivalent to
(VI). Consider any fixed values of x v . . . , x n, and w rite sim ply “ x(y)”
280 G E N E R AL R E C U R SIV E FUNCTIONS CH. X I
for “ x(xv .. x n) y ) ” , etc. T o illu stra te, suppose th a t for y — 0, 1,2,

(first row below ), x(y) ta k e s th e v a lu es sh ow n (second row).
y 0 1 2 3 4 5 6 7 ...
x(y) 3 1 2 0 9 0 1 5 ...
Ay) = n x(s) 1 3 3 6 0 0 0 0 ...
s<y
A cc o rd in g to (V I), 9 = ]xyR{y) = i.e. 9 is th e least v a lu e

of y (first row) for w h ich a 0 ap p ears in th e second row, n a m e ly 3 in th is
exam p le . T h is n um ber 3 is also id en tifia b le as th e u n iq u e y for w h ich a
successor (here 6 ) im m e d ia te ly precedes a 0 in th e th ird row . N o w ( V I 2)
g iv e s th is n u m b er an d no oth er as v a lu e to 9 , th u s. S u b s titu tin g 3 for y
in th e la s t eq u a tio n an d e v a lu a tin g ,
9 == t (7c(3), 7c(4), 3) = t (6, 0, 3) = 3.
I f w e su b s titu te a n y o th er n u m ber th a n 3 for y in th e la st eq u a tio n , w e

are u n a b le to e v a lu a te t. (If th e e x a m p le b e ch an ge d so th a t (Zsy)2?(y),
th e n w e g e t no v a lu e for 9 .)
N o w le t E 2 b e th e sy ste m of eq u a tio n s o b ta in ed b y tra n sla tin g ( V I 2)
u sin g s a y p, t, f, a v . . a n, b for ‘ V \ “ t ” , “ <p” , “ V ’ > • • - , “ * » ” > “y ” >
re sp e c tiv e ly .
T h e fo llo w in g proposition s (i) — (iv) tre a t th e situ a tio n for a n y fix e d
^ -tu p le x lf . . . , x n, w ith o u t u sin g ( 1 ). F o r (i) an d (ii), n m u st b e th e
fu n ctio n d efin ed from x or i? b y ( V I X).
(i) If (E y)R (x1, . . . , x n,y ) , then E ", E 2 b f(xv when x
= (J.yi?(î, . . . , x n, y). (ii) If E£, E2 b f(x v . . . . x n) = x where x is
a num eral , then (.E y)R (xv . . ., x nt y) and x = ^ yR {xli ..., x n, y).
T h e proofs clo sely p ara llel th e in form al ex p la n a tio n .
B u t 7u is p rim itiv e recu rsive in x ( # B § 45), so there is a sy ste m E x
of eq u a tio n s w h ich defines n re c u rsively from x (Theorem I I § 5 4 ). W e
can choose E x so th a t p is th e p rin cip al fu n ctio n letter, h is th e g iv e n
fu n ctio n letter, an d th e a u x ilia r y fu n ctio n letters do n o t occur in E 2.
L e t E b e E XE 2. T h e n c le a r ly E^, E b f(xv . .., x n) = x if Ejf, E 2 b
f ( xlt . . x n) =x; an d th e con verse follow s from L e m m a l i e §54 as
in th e proof of L e m m a l i d . U sin g th is in (i) an d (ii), an d co m b in in g th e
results in to one sta te m e n t:
(iii) b f(xr, ..., x n) = x for some num eral x, if and only if
E^, E
(E y)R (xlt\ x nt y), in which case x = (jlyR (x v . . . , x nt y).
..,
N o w usin g ( 1 ) w ith (iii), for each n -tu p le xv . . ., x n, E^, E b f(x x, . . .,
x n)= x w here x is a num eral, if an d o n ly if x = [iyR (xv . . . , x n, y);
i.e. E defines th e fu n ctio n \LyR{xv . . . , x n) y) re cu rsively from R .
§57 THE jx-OPERATOR 281
P roof of T heorem I V . N o w suppose th a t R (x lt ..., x n, y) is gen eral
recursive. L e t D b e a sy ste m of eq u a tio n s w h ich defines re cu rs ive ly th e
represen tin g fu n ctio n x of R t w ith h as p rin cip al fu n ctio n le tter, an d w ith
a u x ilia r y fu n ctio n letters n o t occurring in E . L e t F b e D E . N o w :
(iv) F h f ( x v . . . , x n) = x for some num eral x , if and only if
(E y)R (xv . . . , x n,y ) , in which case x = fxyi?(%, . . . , x n,y ) .
U sin g D f l 3 § 56, w e sta te th is result (o m ittin g th e fin al rem ark) in
sy m b o lic form , th u s:
(2) (E y)R (xv y) = ( £ Y ) @ n(F, * n, Y ) .
N o w le t / b e th e G o d el n um ber of F . B y th e d efin itio n o f 'th e cor

responding n u m b er-th eo retic predicate* § 52, S n(/, xv . . . , x ni y) = {y is
th e G o d el n u m ber o f an e n t it y Y such t h a t @ n(F, x v . . . , x ni Y )} . H e n ce
( £ Y ) @ W(F, x v . . x n, Y ) = (E y )S n(f , x l9 . . . , x nt y). So (2 ) g iv e s
(3 ) {E y)R {xlf . . . , x n,y ) = (E y)S n(f, * n, y).
B y L e m m a I I I , S n is p rim itiv e recursive.

In s ta tin g th e result in th e theorem , w e replace S n b y th e p re d ic a te
T n defined from it thus,
T n{z, xv v ., x n, y) = S n(z, xv . . . , x nt y) & (it)t<vS n{*> • • •. 0-
F o r g iv e n z t xv . . . , T n(z, x lf . . . , #n, y) is tru e for a t m o st one y

(cf. * 1 7 4 § 4 1 ) . T h e a d v a n ta g e of u sin g T n in stea d of S n as th e b a sic
p red icate o f th e th e o r y w ill ap p ear in § 58. B y # # D a n d E , T n is also
p rim itive recursive. B y th e in form al co u n terp arts of & -e lim ., *7 0 § 32
an d * 1 4 9 a § 40 (noting th a t, since 5 n is recursive,
S n(z, x v ..., x n, y) V S n{z, x v ..., x n, y)

in tu itio n istica lly , as w ill be discussed in §§ 60 an d 62),
i*i) T n{z> • j %n> y) ~^ E n(zf x^f . . . , x ni y),

(5 ) (E y )T n(zf % , . . . , x n9 y) = (.E y)S n(z , % , . . . , x nty).
T heorem I V. F or each n > 0: Given any general recursive predicate
R {xv . . . , x nt y), numbers f and g can be found such that
(6 ) (E y)R (xv . . . 9x n ty) = (E y)T x n9 y),
n (/,
(7) (y)R (x i, . . . 9x n9y) = (y)T n(g9 x l9 . . x n9 y),

where T n(zt x v . . . , x n, y) is the particular prim itive recursive predicate
defined above. S im ilarly w ith more qu an tifiers ; e.g. given a general recursive
predicate R {av . . . , a ni x, y), a num ber g such that
(8 ) (£ * ) ( ? ) £ (« !, . . . , a n, x ,y ) = {E x)(y)T n+1(g, av . . . . a n, x, y).
(E n u m eratio n theorem , K le e n e 19 4 3 .)
P roof (com pleted). F o rm u la (6 ) follow s from (3) b y (5). T o infer

(7), w e a p p ly (6 ) to th e p red icate R (xv . . x n, y), callin g th e / for th is
p red icate “g” , an d tran sform th e result b y th e inform al co u n terp arts
o f *30 § 26 an d *86 § 35 to
(y)R {xi, . . . , x n,y )
(y)T „ (g ,x v
= y).
Sin ce R is gen eral recursive, w e h a v e R \J R, an d hence (cf. *4 9 c § 27)
R = R . F o r (8 ), w e a p p ly (7) w ith n + 1 as th e n of (7), av . . . , a nt x
as t he x v . . . , x n, an d y as th e y, an d use th e inform al co u n terp art of
*72 §33.
D iscussion . By th is theorem , we o b ta in an enu m eration (w ith
repetitions) of th e ^ -v a ria b le pred icates of th e form (E y)R (xlf . . . , x n, y)
w here R is gen eral recursive b y ta k in g z = 0 , 1 , 2 , . . . in th e fix e d n + 1-
v a ria b le p red icate (E y )T n(z, x lt . . . , x n, y) of like form . B rie fly ,
(E y )T n(z, xv . . . , x n, y) 'en u m erates' th e p red icates of th e form
(E y)R (xlt . . . , x n, y) w ith a general recursive R . S im ilarly,
(y )T n(z, x v . . . , x n, y) enu m erates th e p red icates of th e form
(y)R (xv . . ., x nt y) w ith a gen eral recursive R \ e t c . — Sin ce T n, T n, T n+1,
etc. are p rim itiv e r e c u r s iv e :
Corollary. The class of the predicates expressible in a given form

consisting of a fixed succession of one or more quantifiers prefixed to a
predicate R is the same whether a general recursive R or a prim itive recursive
R be allowed. (K leen e 19 4 3 , gen eralizin g a lem m a of R osser 1936 p. 87.)
W e n ow use th e a b o v e enu m erations as th e b asis for an a p p lica tio n
o f C a n to r's d iago n a l m eth o d to p ro ve th e n e x t theorem . G iv e n a gen eral
recursive p red icate R (x, y), an d usin g (6 ) for n= 1 , there is a n um ber
/ such th a t
(9) _ _ ( E y W x .y ^ iE y W M .x .y ) .
S u b s titu tin g f for # in th is equ ivalen ce,
(10) ( E y ) R ( f ,y ) ^ { E y ) T 1( f ,f ,y ) .
T h en ce, u sin g th e in form al co u n terp art of *5 0 a § 27,
( 11 ) (Ey)R(f, y) * m T . i f , f, y).
F o rm u la ( 12 ) of th e theorem follow s, usin g th e inform al co u n terp art of
*86 § 35. T o p ro v e (13), u sin g (7) w e h a v e
(y)B(g> y) = ( y ) n t e » s> y) = iEy)Ti{g> g. y) ^ (Ey)Ti(g> g> y)-

T o infer (14) an d (15), g iv e n a n y general recursive R (x), le t R {x ,y ) =
R (x )& y = y or R {x ,y ) == R (U \{x ,y )) (§44), so th a t R {x ,y ) is gen eral
recursive an d R(x) = (E y)R (x , y ) = {y)R {xy y).
§57 THE fX-OPERATOR 283
T heorem V (P art I). Given any general recursive predicate R(x, y),
numbers f and g can be found such that
(12) (Ey)R{f,y)^{y)T1(f,f,y),
(13) (y)R {g .y) & (E y)T 1te ,g ,y ) .
A fortiori, given any general recursive predicate R{x)} numbers f and g
can be found such that
(H ) R{ f ) *( y) T1(f,f,y),
(15) {Ey)Tx{g, g ,y ).
R(g) &
D iscu ssio n . B y th is theorem , (y)T±(x, x, y) is an exam p le of a pred
ica te of th e form (y)R(x, y) w ith a recursive R w h ich is n ot expressible
in th e d u al form (Ey)R(x, y) w ith a recu rsive R. T h a t is,
{x){{y)T i{x, X, y) = (E y)R (x , y)]

ca n n o t h old for a n y gen eral recursive y); an d for a g iv e n
p red icate R (x,
R(x, y), t h e / o f ( 12 ) is a v a lu e of x w h ich refutes it. A fortiori, (y)T1(x,x,y)
is also n ot general recursive (cf. (14)).
T h e proof (above) am o u n ts sim p ly to th is: (E y )7 \(z , x, y) for z — 0,
1 , 2 , . . . isân enu m eration (w ith repetitions) of all pred icates of th e form
(E y)R (x, y) w ith R recursive. B y C a n to r's d iago n al m eth od , (E yjT^X ) x, y)
is a p red icate not in th e enum eration. The la tte r is e q u iv a le n t to
(y)Ti(*,*,y).
F ro m th e en u m e ra b ility of all syste m s E of equ ation s, w e can con clu de
w ith o u t th e present th e o ry th a t th e gen eral recursive p red icates are
enum erable, an d so also th e p redicates of th e form ( E y)R (x , y) w ith R
recursive. H en ce b y C a n to r's results ( § 2 ), th e y ca n n o t c o n stitu te all
n u m b er-th eo retic predicates. The ad d itio n al co n ten t of th e present
theorem is th a t an exam p le of a p red icate neith er gen eral recu rsive nor
expressible in th e form (E y)R (x, y) w ith R recu rsive is g iv e n w h ich is of
th e d u al form (;y)R (x , y) w ith R recu rsive; an d v ic e versa.
T o s a v e space, w e w rite th e n e x t p a rt of th e theorem for p red icate s of
one v a ria b le a ; b u t it holds likew ise for p red icates of n v a ria b les alf ..., an
for each n> 1 . In th e proof of (b) w e use a classical eq u ivalen ce, n o t
a v a ila b le to us in tu itio n istic a lly . W e ac co rd in g ly lab el (b) w ith a “ c ”
(cf. § 37). T h e results so lab eled in th is ch a p ter are all on th e n u m b er-
th eo retic level.
T heorem V (P art Consider the predicate forms

II).
R(a) (E x )R (a ’ x ) (x )(E y )R (a >x <y) {E x )(y )(E z)R (a ,x ,y ,z) ...

(x)R (a,x) (E x)(y)R (a, x, y) (x)(E y)(z)R (a, x, y, z) ...
where the R for each is general recursive. To each form with k + 1 quantifiers
(k > 0), there is a predicate expressible (a) in the negation of the form
((b)c in the other k -f- \-quantifier form ), but not in the form itself nor in
any of the form s with < k quantifiers. B y P a r t I, th e “ c ” is un n ecessary
io r k = 0. (K leen e 1 9 4 3 ; M o sto w ski 19 4 7 .)
P roof . F o r exam p le, ta k e k = 1 . F ro m (8) (for n = 1 ) w e infer
(16) (E x)(y)R (g, x, y) & (.E c)(y)T 2{g, g, x, y),

ju s t as from (6 ) w e inferred (11). T h u s (E x)(y)T 2(a, a, x, y) (eq u ivalen t
cla ssica lly to (x)(E y)T 2(a, a, x, y ) } cf. T h eorem 18 § 35) is n o t expressible
in th e form (E x)(y)R (a, x, y). S im ilarly, (x){E y)T 2{a, a, x, y) (eq u ivalen t
cla ssica lly to (E x)(y)T 2(a, a, x, y)) is n o t expressible in th e form
(x)(E y)R (at x ,y ) . A fortiori, these pred icates are n o t expressible in a n y
o f th e three form s w ith one or no quantifier.
T heorem F or each n > 0 : (a) E very general recursive predicate

V I.
P {x lt . . . , x n) is expressible in both of the form s (E y)R (xlf . . . , x n,y )
and (y)S (% , . . . , x n,y ) where R and S are p rim itive recursive. (b)c Con
versely , every predicate expressible in both of these form s where R and S are
general recursive is general recursive, (c) A predicate P {x lt . . . , x n) is
general recursive , if and only if P (x lt . . . , x n) and P (xlf . . . , x n) are each
expressible in the form (E y)R (xlt . . . , x n, y) with a general recursive R } and
(*i) • • • (x„)[P{x 1 , . . . , x n) y P (x v . (K leen e 19 4 3 , P o st 19 4 4 ,
M o sto w sk i 1 9 4 7 .)
P roofs, (a) T a k e P (x v . . . , x n, y) = P {x lt y = y , so th a t
P (x 1 , . . . y x n) = {E y)P {xlt y) == (y )P (% , y), an d a p p ly
C o ro llary T h eorem I V . (b) Suppose P {x v . . . , x n) = (E y)R (xv . . . , x n fy)
= (y)S(^x, .. . t x n,y ) . T h e n b y classical lo gic (cf. *8 5 § 35), P (x v . . . , x n)
= (E y)S (xv . . . , x n, y). B y th e classical la w of th e ex clu d e d m id d le,
P{Xi, . . . , x n)\J P (x lt . . . t x n). H en ce
P (x 1, • • • ,Xn) = ^ (^ 1, • . ., Xn, W [R (Xl> y) V S(xv . . . , x n, y)]),
w here th e second m em ber is gen eral recursive b y T h eo rem I I I . (c) T h e
la st h yp o th e sis m akes th e sta te m e n t v a lid in tu itio n istica lly .
In K le e n e 19 4 3 , 194 4 , a p red icate w as called “ e le m e n ta ry ” , if it ca n

b e expressed e x p lic itly in term s of co n sta n t an d v a ria b le n a tu ra l num bers,
gen eral recursive p redicates, th e operators & , V, of th e proposi
tio n a l calcu lu s, an d th e qu an tifiers, co m b in ed accord in g to th e u su al
s y n ta c tic a l rules.
§57 THE [/.-OPERATOR 285
T heorem V I I . (a) E very arithm etical predicate (§ 48) is “ elem entary ” ,
and (b) conversely, (c) E very predicate expressible by m eans of quantifiers
prefixed to a general recursive predicate is expressible in one of the form s of
Theorem V with a general recursive R . (d)c E very arithm etical predicate
is expressible in one of the form s of Theorem V with a general recursive R .
P roofs , (a) B y # # 1 , 2 , 14, C (§ § 4 4 ,4 5 ), e v e r y p red icate form ed
e x p lic itly usin g + , •, = is p rim itiv e recursive, an d hence gen eral re
cursive. (b) B y T h eo rem V I (a) an d C o ro llary T h eorem I § 49. (c) A s x
ranges over all n a tu ral num bers, th e w - f - 1 -tu p le of p rim itive recursive
fu n ctio n s (x)0, ...,(%)m (#1 9 § 45 ) ranges (w ith repetitions) o ve r all
w + 1 -tu p le s of n a tu ral num bers. Th erefore
(17) (E x0) ... ( E x m)A (x0, . . x m) = (E x)A ((x)0,

(18) (*0) . . . {xm)A (x 0, . . . , x m) = {x)A {{x)0, .... {x)m).
W e use these to co n tra ct co n secu tive qu an tifiers of th e sam e kind ,
(d) B y (a), th e inform al co u n terp art of T h eo rem 19 § 35, an d (c).
R emark 1. F o r a n y p red icateA (i, x),

(19) (i)i<a(Ex)A(i, x) = (E x)(i)i<aA (i, (*),).
B y s u b s titu tin g A (i, x) for A (i, x) an d usin g (the inform al co u n terp arts
of) *30, *85, * 86 , *58 an d *49 (cf. th e proof of C o ro llary T h eo rem s 8
an d 18), w e infer cla ssica lly th e d u al
c
(20) (Ei)i<a(x)A(i, x) = (x){E i)i<aA (i, (*),).
(But (20) does n o t h old in general in tu itio n istic a lly , b y E x a m p le 4 § 82.) —
S im p ly b y *9 5 an d * 7 7 ,{i)i< a{x)A (i, x) = {x)(i)i<aA (i, x ) ; an d sim ilarly
for (E i)i<a{Ex), {i)i<a(x)x<b an d {E i)i<a{Ex)x<b.
A lth o u g h there is a su b sta n tia l difference b etw een th e n otions
‘ a rith m etic a l' an d ‘e lem en ta ry ', w h ich requires tw o b asic theorem s
to b rid ge (Theorem s I an d IV ), h en ceforth in th e interest of u n ify in g
te rm in o lo gy w e shall u su a lly s a y “ a r ith m e tic a l" (even w hen p rim arily
we h a v e th e other n otion in m ind).
E xample 1. “ E le m e n ta r y " is used by K a lm a r 19 4 3 * in an oth er

sense, eq u iv a le n t to th e follow ing. A fu n ctio n is “ e le m e n ta ry " , if it can
be expressed e x p lic itly in term s of va ria b le n a tu ral num bers, th e co n sta n t
1 , th e fu n ctio n s + , • an d [afb], an d th e operatio ns S an d II. E x te n d in g
y<z y<z
th e notion to predicates, an d to th e case of assum ed fu n ctio n s an d
predicates, in th e fam iliar w a y , w e can recognize su ccessiv ely th a t th e
follow in g of our list # # 1— 21 of p rim itive recursive fu n ction s an d p red i
cates (§§ 44 ,4 5) are elem en tary, an d the follow in g of our results # # A — G
h old readin g “ e le m e n ta ry ” in place of “ p rim itive recu rsive” : # 1 , # 2 ,

# 13 , # # A — C, #3, #4, #9 (sg(«) = [l/ ( « + l) ] ) f # 10 , #D, #E
(II is II ), # 15 (the represen tin g fu n ctio n is s g [ ( « + 1 ) / ( 6 + 1)]), #6
*<< s<t +1 ________
( a - b = y.ccSa[ b + c < a V a < b ]), # 5 , # 7 , # 8 , # 1 1 , # 1 2 (rm(a, b)
— a —b[afb]), # 1 4 , # F , # 1 6 , # 1 7 , # 1 9 as a fu n ctio n of a for each
fix e d t ; also G o d e l’s ^ -fu n ction (§48). — N e x t w e show th a t (A) if 9
comes from # y {from y) by a prim itive recursion {Schema (V)), and
9 (y, x2, . . . , x n) < i](y, x2, . . . , x n), then 9 is elementary in # •/, {in 7 , 7]).
In th e proof of T h eorem I § 49 Case (V b), (B) is of th e form
{E c){E d)R {y, x2, . . . , x n, w, c, d), where R is elem en tary in • • • >* « ) = {\dR {y, x 2, . . . , x n, {t)0, {t)v {t)2))0. So b y # E it w ill
suffice to fin d a b o u n d elem en tary in 7) on th e t. B u t b y § 48, w e can
choose d = (m ax(y, a0, . . . . ay))\ where at = 9 (i, x2, . . . , x n), so th a t
d <, D w here D = ( y + E r\{i, x2, . . . , x n))\, an d c < II S {d, i) =
n (l + (* + l)rf) ^ C w here C = II (l + (f + l)Z>). So w e can ta k e
i<y iSy
t < 2E -3C -5D w here £ — 73(3;, x 2, . . x n). — N o w ^ ( # 1 8 ) is elem en tary,
for an e lem en ta ry b o u n d ca n b e found, e.g. p t < 22i ( = o n ly for i = 0),
as w e p ro ve b y course-o f-valu es in d u ctio n on i, thus. T ru e for i = 0 , 1 .
For i > 2, < P q 'P i * . . . * P i-\ — 1 (by reasoning as in th e proof
of E u c lid 's theorem §40) < 2 2° - 2 21 * .. .* 22i_1 (using th e h y p . ind.)
= 2 2°+ 2 l+ - + 2l_1 = 2 2l- i < 2 2t. — T h e n th e follow ing are e lem en ta ry :
# 19 , #20 ( i , a ) < a ) } # 2 1 . A lso (cf. # G ) (B) if cp comes from
(lh
X by the course-of-values recursion (3) § 4 6 , and <p(y, x 2, . . x n) <
?l(y> • • •> x r)> then 9 is elementary in x, y). F o r th en U p y il’x*’ ~"xn) is
^ **<2/
a b o u n d e lem en ta ry in tq for 9 (3/; x2> . . . , x n) in th e p rim itive recursion
(4) § 46. — I t follow s th a t th e nu m ber-th eoretic pred icates an d function s
correspon ding to th e m e ta m a th e m a tic a l ones defined b y D n l — D n l 3 a
§ 51 (cf. L e m m a 19 § 52) an d D f l — D f l 5 § 56 (cf. L e m m a II I) are elem en
ta ry . F o r recursions are used in those m eta m a th e m a tica l d efinitions in
th e gen eralized a rith m etic o n ly to introd uce (the represen tin g fu n ctio n s
of) pred icates, an d so (B) applies to each corresponding recursion in th e
sim ple a rith m etic w ith 73 = 1. In particu lar, S n ( D f l 3), an d hence (C)
T n {preceding Theorem IV ), and U ( D f l 5) are elementary. T h is corresponds
to th e result o f Ilo n a B e re czk i (1949 * unpublished) th a t (b y m eans of
d evices used in K a lm a r 194 3 ) each of th e p red icates an d fu n ction s
sh ow n in K le en e 1936 to b e p rim itive recursive is either elem en tary or
c a n b e replaced b y one w h ic h is elem en tary w ith o u t d eran gin g th e
§57 THE ^-OPERATOR 287
argu m en ts of th a t paper. — we can add 11or an elem entary R
So (D)
{in K alm ar s sense)” in Corollary Theorem IV . — K a lm a r (19 4 3 , 19 4 8 ,
1950 , 1950 a) uses his e lem en ta ry fu n ction s in p resen tin g G o d eF s theorem
an d other results w h ich are presented in th is b o o k u sin g p rim itiv e re
cursive fu n ction s. — M iss B e re c zk i (19 5 2 *) show s th a t th e p rim itiv e re
cursive fu n ctio n ha d efined by °a = 1, b'a = a^ (so th a t b+1a =
| 3(5, a) § 55) is n o t e lem en ta ry becau se aa grow s too fast as a increases.
She o b ta in s an oth er e x a m p le of a n o n -elem en tary p rim itiv e recursive
fu n ctio n b y co n stru ctin g a p rim itive recursive en u m eratin g fu n ctio n
(p{n, a) for th e elem en tary fu n ctio n s of one va ria b le (cf. § 55). F ro m th is
it follow s th a t there are n o n -elem en tary p rim itiv e recursive p red icates
(sim ilarly to E x a m p le 1 § 55).
T heorem V I I I 0. F or the prim itive recursive V and u defined below,

the predicate M (a, k) defined by induction on k thus,
M {a, 0) ^ V{a),
• M (a, 2 k -\-\) == (Ex)M(\>(a, x), 2k),
M {a, 2 k + 2 ) = (x)M {u{a, x), 2 k + \) ,
is not arithm etical. (K leene 19 4 3 .)
K a lm a r first o b ta in ed a result of th is sort, w h ich appears in S kolem
1936-7 pp. 86 ff. A lso cf. W a n g 1 9 5 3 , M o sto w ski 1 9 5 1 .
P roof L e t us in th e expression for S n(z, % , . . . , x n, y) (D f 13

(o p tio n al).
§ 5 6 ) replace “n '” b y " {%)'” , an d “N u {(y)QlA,x ^ & . . .& N u({y) 0>,
b y “ {i)i<i<(x)jyu{{y)<sti fi, {%)i— ! ) ” • T h e result is a p rim itiv e recu rsive
p red icate S(z, x, y) such th a t
(21) S(z, 2 n-pf'-.. = S n(z, xv . . . , x n, y).

L e t T {z, x, y) = S(z, x, y) & (t)t<yS(z, x, t). T h e n
(22) T (z, 2“ • ... • P I, y) - T n(z. xl t . . . , x n, y).

N o w let V{a) = T ((a)x— \,a , («)(a)o-— 1). T h e n for n > 1 ,
(23) V (2n ■ p f ' • . . . • p ? '- p ^ = T n(xv x v . . . . x n, y).

L e t u {a, x) — a*2x' (cf. # 2 1 § 45). N o w M (a, k) ta k e s for k = 0, 1 , 2 , . . .
th e va lu e s V [a), (E x)V (a*2X'), (x)(E y)V ({a*2x')*2v' ) , ----- S uppose M (a, k)
were arith m etical. B y T h eo rem V I I (d), th en it w o u ld b e expressible in
one of th e form s of T h eo rem V . M (a, k) =
F o r illu stratio n , suppose
(Ex)R(a, k, x) f o r a recursive R . T h e n (E x)R (22-3a', 2, x) == M (2 2 -3a', 2) =
(x)(Ey) V {{ (22 • *2X')*2*')= (x)(Ey) V (22 • 3°'- 5^'* 7 ^ > (x)(E y)T 2(a, a, x ,y ) .
B u t {E x)R {22 • 3a', 2 , x) is of th e form (E x)R (af x) w ith R recu rsive; an d
288 GENERAL RECURSIVE FUNCTIONS cm xi
in th e proof of Theorem V P a r t I I w e saw t h a t (x)(E y)T 2(a, a, x, y) is
n o t expressible in th is form ,
T h e reader m a y pass if he w ishes to §§ 60— 62, w h ich do n o t d epen d

e x c e p t in c id e n ta lly on §§ 58 an d 59. B u t § 58 co n ta in s a result w h ich
is fu n d a m e n ta l for §§ 63— 66 an d later.
§ 58. Normal form. P o s t ’s theorem. U s in g D f l 3 a nd D f l 5 (§ 56),

w e ca n re sta te th e d efin itio n of 'gen eral recu rsive fu n c tio n ’ (§§ 55, 54)
as follow s. A fu n ctio n <p(%, . . x n) is gen eral recursive, if an d o n ly
if there e x ists a sy ste m E of eq u a tio n s (w ith o u t g iv e n fu n ctio n letters)
such th a t
(24) (Xl) ... (xn) (E Y )B n(E, x v . . . , x n, Y ) ,

(25) (*,) . . . ( * .) ( Y ) [ @ « ( E , * x> Y ) - * U (Y ) = 9 (*1 ......... *„ )].
O n passin g o ver from th e gen eralized arith m e tic to th e sim ple a rith m e
tic b y th e G o d el n u m berin g, @>n becom es S n, U b ecom es U, E becom es its
G o d e l n u m b er e, an d (24) an d (25) g iv e
(26) (xn){E y)S n(e, xv . . . . x n> y),

(*i) . . •
(27) (*,) . . . (xn)(y )[S n(e, x v . . . , x n,y ) -> U (y)=<p(xv . . . , * „ ) ] .
B y L e m m a I I I , U as w ell as S n is p rim itiv e recursive.

I t follow s from (26) an d (27) t h a t th e fu n ctio n y (x v . . . , x„) ca n be
expressed in term s of th e n u m b er e thus,
(28) ? ( * !, . . . , * „ ) = U (tiyS n(e, xv . . . , y)).

B y (5 ) an d (4 ) § 5 7, now (26) an d (27) a n d therefore also (28) rem ain
v a lid w h en Sn is replaced b y T n.
T heorem IX . F or each n 0: Given any general recursive function
<p(xv x n), a num ber e can be found such that
...,
(29) (*x) . . . (xn){E y )T n{e, y),

(30) • • • j x n) = U ( w T n(e, Xi, . . . , x n, y)),
(31) (%) . . . (xn){y )[T n(e, xv . . . , x n, y ) -+ U (y)=<p(xv . . . , « » ) ] ,
where T n(z, xlt . . . , x n, y) and U (y) are the particular prim itive recursive
predicate and function defined above. (N orm al form theorem , K le e n e 19 3 6 ,
194 3)
T h e a d v a n ta g e o f u sin g Tn in stea d o f Sn is th a t (31) h old s for a n y
n u m b er e such t h a t (29) a n d (60) h o ld (whereas (26) an d (28) m ig h t h o ld
a n d (27) b e false, w h en e is th e G o d el num ber of a sy ste m E of eq u a tio n s
w h ic h la ck s th e c o n siste n cy p ro p e r ty for d efin in g 9 recu rsively).
§58 NORMAL FORM, POST'S THEOREM 289
W e n o w sa y , for a n y gen eral recursive fu n ctio n 9 , th a t a n y n u m b er
e (w hether or n o t it is th e G o d el n um ber of a sy ste m E o f eq u a tio n s
defin in g 9 recursively) su ch t h a t (29) an d (30) (and h ence (31)) h o ld
defines recursively or is a Godel number of 9 .
9
A n um ber e defines recursively (or is a Godel num ber of) a gen eral re
cursive p red icate P {x lt . . . , x n)t if it defines re cu rs ive ly th e represen tin g
fu n ctio n of P . In th is case,
(32) jP^ 1’ • • • ’ = (£ y) «(«. * !'•••• y ) & ^ (y)= °]

= (y)[r n(e, *lt y) -► tf(y)=0].
T o p ro v e c o n s tr u c tiv e ly th a t a fu n ctio n 9 is gen eral recursive, one
m u st e x h ib it (or im p ly a m eth o d for ob tain in g) eq u a tio n s E w h ich d efin e
9 recu rsively. T h u s to g iv e a gen eral recursive fu n ctio n e ffe c tiv e ly m ean s
to g iv e an E , or n o w a G o d el n u m ber e.
T h e th e o ry of th e G o d el num bers of recursive fu n ctio n s w ill b e tre a te d
in th e n e x t ch a p ter (§ 65).
E xample 1. Is e v e r y gen eral recursive fu n ctio n <p(xv . . x n) ex

pressible in th e form \iyR {xv ..., x nt y) w here R is p rim itiv e recu rsive
(and (l b) hplds) ? H int : U se E x a m p le 1 § 55 an d # E § 45. (A d ifferen t
m eth o d w as used b y P o st 1946 a.) — C all a fu n ctio n 0(y) “ u n iv e rs a l” ,
if for each gen eral recursive fu n ctio n <p(xv . . . , x n) there e x ists a p rim itiv e
recursive p red icate R (xv . . . , x nt y) w ith (l b) h o ld in g such t h a t
<p(xv . . . , x n) = Q([iyR(xv . . . , x nt y ) ) ; an d ca ll 0(y) “ o f large o sc illa tio n ” ,
if (x)(z)(E y)y>J d (y)= x. M a r k o v 19 4 7 c, 1949 show s th a t a su fficien t
an d cla ssica lly n ecessary co n d itio n th a t a p rim itive recu rsive fu n ctio n 0
be u n iversal is th a t it b e of large oscillation. K u z n e c o v 1950 an nounces
further results con cerning such fu n ctio n s 0.
Corollary. E very general recursive function 9 is definable by a p

plications of Schemata (I) — (V I) with ( 1 ) holding for the applications of
(VI). (Converse of T h eo rem I I I , for W e m p ty .)
F o r sim p lic ity , b egin n in g w ith T h eo rem I V , w e con fin ed our a tte n tio n
to fu n ctio n s an d p red icates gen eral recursive a b so lu te ly , i.e. recu rsive
in 4>i, . . . , for 1 = 0. N o w w e shall e x te n d th e th e o ry to re la tiv e gen eral
recursiveness, i.e. to l > 0.
W e sh all find th a t T h eo rem I X an d th e earlier results hold, if in stea d
of a p rim itiv e recursive p red icate T n, w e use one p rim itiv e recursive
in ijq, . . . , L e t us w o rk first w ith th e case th a t l= 1 an d ^ ( = £ x) is
a fu n ctio n of one v a ria b le (mt = 1 ).
B y d efin itio n (§ 55), 9 is gen eral recursive in if there is a sy ste m E
of eq u a tio n s w h ich defines <p re cu rsively from ^ (§54). I f there is a n y
such E , th e n w e can choose one in w h ich th e g iv e n fu n ctio n le tte r g
(w ith G o d el n um ber g) is th e first in th e fix e d enu m eration of fu n ctio n
letters.
N o w w e s ta te ap p ro p riate m o d ificatio n s of D f l 2 a nd D f l 3 § 56.
D f 12* (for l = m x = 1 ). (Z is a sy ste m of equ ation s, and) Y is a de

duction from E |, Z (b y R l an d R 2). ( A b b r e v ia tio n : ® ^(Z, Y ).)
*0. Z is a sy ste m of e q u a tio n s ; an d for som e n a tu ral n u m ber

u lf Y is (g(«i) = u) w here u = ^ (% ).
SE (z) & y == 2 e x p 2 15 • (3 exp 2° • 3(i/)o>m ) • 5i y ) &
{E u i ) U l < v [N u ( ( y ) o , i , i > u i) & N u ( ( y ) o, 2 > ^K))]-
1— 3. L ik e 1— 3 D f 12, e x c e p t read ing “ from E | , Z ” in place of
“ from Z ” .
D f l 3 * (for l = m1 = 1 ). L ik e D f 13, reading “ from E | , Z ” in p lace of

“ from Z ” , an d y)" in p lace of <tD { z ,y )i\
x v . . . , %w, Y ).)
(A b b r e v ia tio n : <5^(Z,
T h e reasoning used before to show th a t D (z, y) an d S n(z, xlf . . . , x n> y)

are p rim itiv e recursive (L em m a I I I § 56) show s n ow th a t D^(z, y) an d
S^( 2, x v . . . , #n, y) are p rim itiv e recursive in an d in p lace of (26) an d
(27) for a gen eral recursive fu n ctio n y(x lt . . ., x n), w e n ow h a v e for a
fu n ctio n <p(xv . . . , x n) gen eral recursive in ty,
(33) (*i) • • • (x„)(E y)Sj(e, x v . . . , x n, y),
(34) (*,) . . . {xn)(y)[S%(e, x lt . . . , x n,y ) ~ * U {y)=<p(xlt . . . , * , ) ] .

W e cou ld co n tin u e as b e fo r e ; b u t w e shall sh ow also th a t th e d ep en d
ence on can b e g iv e n a special form.
T h e co u rse-o f-va lu es fu n ctio n $(y) ( = II § 46 ( 1 )) for is p rim
ly
itiv e recursive in ^ ( # # A , B , 3, 18, §§ 44, 45).
U sin g § 46 (2 ), ^ (% ) in ODf 12* can be w ritte n as ($ (y))M, since u x < y,
or ev en as (4,(^))m1 for a n y v > y. Let D 1(w, z, y) be th e p red icate o b ta in ed
in p lace of D'^(z, y) w hen w e replace <];(%) in ODf 12* b y (w) u . T h e n D 1
is p rim itiv e re c u r s iv e ; an d b y course-o f-valu es in d u ctio n on y (w ith cases
correspon ding to th e four clauses in th e recursive d efinitions of D 1 an d
D *, cf. § 52),
(35) . (v)v> um $ ( v ) , z , y ) = D H z,y )].

U sin g D 1( w ,z ,y ) in p la ce of D ^ (z,y) in D f l 3 * , w e o b ta in a p rim itiv e
recursive p red icate S* (z£>, z, x v . . . , * n, y) such th a t
(36) (v)r>y[5i($(v)> z, x v . . y ) = S%(z, x v . . . , x n, y)].

§58 NORMAL FORM, POST’S THEOREM 291
W e define T \ and T ^ thus,
T ln {w, z ,x lt . . . , x n,y ) = S 1n (w ,z, x v . . . , x n,y ) & {t)l<^Sln{w ,z ,x 1, .
*1> • • • > y) — Tn{^{y), Z, A j , ..., x„, y).
T h en T ln is p rim itive recursive, an d is p rim itive recursive in tj/. U sin g
(36) first w ith y , y an d th en w ith y, t as the v, y,
(37) Tft(z, xv ..., x n, y) = S*(z, xlt . . . , x „ , y ) & (0t< ^ (z, xv . ..,x„, t).
N o w w e can con tin u e from (33) an d (34) w ith (37), as w e d id before
from (26) an d (27) w ith th e d efin ition of T n (preceding (4) § 57).
O ur sta tem e n ts here are to be und erstood in tu itio n is tic a lly as co n
sequences of the h yp o th esis th a t p articu lar va lu es of ^ are a v a ila b le on
dem and. T h is h yp o th esis ju stifies e.g. expressing a giv e n v a lu e ^(jq) as
($(v))w1 for a g iv e n v > uv w h ich requires th e other va lu e s am on g
+( 0), . . . , and b y it
*1. • • • - X„, t) V S%(Z, Xv .... X„, /)]

for a g iv e n y , w h ich requires th e va lu es ^(0), . . . , y (y — 1 ) (cf. th e proof
of (5)).
The results o b ta in ed in § 57 begin n in g w ith T h eo rem IV e x te n d
sim ilarly. "
T h e case of a n y l fu n ction s of mv . . mt variab les, re sp e ctive ly, can
be redu ced to th e case l = m1— 1 (by ta k in g ^ *(*) = ^ i( W i, . . . , (x ) ,„•),
^ = p ^ v . . . • p p ); b u t is also easily tre a te d d irec tly. F o r exam p le, if
l = tn1 = 2, m 2 = 0 , w e define th e co u rse-o f-valu es fu n ctio n ijq for ^1

(in b o th variables) thus, !jq(y,z) = II p t e x p (II p } e x p <jq(/, /)), so th a t
KU
^ x(s, t) = ( ^ ( y , z))8i t if s < y & t < z . W e define = <y2. D f 12 * an d
D f 13* are fo rm u lated for / = m x = 2 , m 2 = 0; an d w e in trod u ce succes
s iv e ly pred icates D ^1’ ^2, D 2,0, S ‘y ° , T%°, an d let
r t *■ * n . y) = Tn y ) > * 1 . • • • . * » . y)-
In general, w e com e o u t w ith a p rim itiv e recursive p red icate T "*1..... mi
of 2 variab les, an d a p red icate T^1 ^/ of n-\-2 v a riab les
p rim itive recursive in . . .,
W h en T* is a list of l fu n ction s an d predicates, th e foregoing applies b y
ta k in g as . . . , ^ th e list o b ta in ed from W b y replacin g the pred icates
(if any) b y th eir represen tin g fun ction s (cf. end § 55); an d w e th en w rite
T *' .... also as T * '.
T h e defin ition of 'a rith m e tica l predicate* (§ 48) is ad a p te d to / > 0 by
ad d in g the fu n ctio n s <Jq, . . . , to the in itial fu n ction s + and •; th a t of
" e le m e n ta ry p re d ic a te " (§ 57 preceding T h eorem V I I ) b y s u b stitu tin g
"g e n e r a l recu rsive in Y ” for " g e n e ra l re cu rsive” . (‘ U n ifo r m ity ' ca n b e

d efin ed in th e u su al m ann er, an d th e resu ltin g relation sh ip s in our
theorem s are u niform w h en th e g iv e n ones are uniform .)
T heorem X . L et lt m lt . . . , m t be fixed num bers ;> 0, and Y be l

functions and predicates of m lt . . . , m x variables, respectively. Then Theo
rem s I, I V — I X and corollaries hold good reading “general recursive in Y ” ,
“p rim itive recursive in Y ” , “ arithm etical in Y ” , “ elem entary in Y ” ,
" T j ” , “ V ^ " , “M ^ \ in place of “general recursive", “prim itive recur
sive" t “arithm etical"t “ elem entary", “ T n", “ V", “M " , respectively {with
U and o unchanged).
T h e th eorem s I , I V — I X th u s e x te n d e d w e cite u sin g stars. F o r e x
am p le : T heorem I X * . F or each n > 0: Given any function (p{xv .. . , x n)
general recursive in Y , a num ber e can be found such that
(38) (x,) . . . ( x n) ( E y ) T j( e ,x 1, . . . , x n,y ) ,
(39) <p(xv . . . , * „ ) = U {w T % (e, xlt . . . . x n, y)),
(40) (% ) . . . (xn)(y )[T j(e , x v y) - t/(y)= < p (*i. . . . . * „ )].
where U (y) is the prim itive recursive function, and T v[ { z , x 1, . . . , x n,y )
the predicate prim itive recursive in Y , defined above ; e.g. for l = m x
= l , T j ( z , x v . . . . x„, y) = r i ( $ ( y ) , z, xlt . . . . x n, y) where
T \{w , z, x v . . . , x n, y) is the particular prim itive recursive predicate defined
above.
A n u m b er e such t h a t (38) an d (39) (and hence (40)) h o ld is said to
define 9 recursively in (or from) Y or to b e a Godel num ber of 9 from Y
or of th e fu n ctio n a l 9 = F (Y ). T h e n o tio n e x te n d s to a p red icate P w ith
represen tin g fu n ctio n 9 as before.
C o r o l l a r y , (a) E very predicate P general recursive in arithm etical

predicates Y is arithm etical. {Sim ilarly, if P is general recursive in Y , 0 ,
and Y are arithm etical, then P is arithm etical in ©.) (b )° The predicate
M {a, k) of Theorem V I I I is not general recursive in any arithm etical
predicates.
P roof of (a). B y T h eo rem V I * (a), P is expressible b y a q u a n tifie r
p refix ed to a p re d ic a te R p rim itiv e recursive in Y . B u t b y C o ro lla ry
T h eo rem I * , R is th en a rith m e tic a l in Y ; so since Y are arith m etical,
R an d hence P is a rith m etical. See N o te 1 on p. 316 .
U n d er th e d efin itio n of " e le m e n ta ry p re d ic a te ” (preceding T h eo rem
V II § 57), th e gen eral recursions w ere ap p lied o n ly before th e lo gica l
§58 NORMAL FORM, POST’S THEOREM 293
operations. T h e m ean in g of C o ro llary (a) is t h a t p e rm ittin g gen eral re

cu rsive operatio ns a t all stages (i.e. interspersed w ith th e lo g ica l op eratio n s
in a n y m anner) does n o t lea d to a larger class of pred icates.
I t is n a tu ra l to inquire ju s t w h a t h ap pen s to p red icates fa llin g in a
g iv e n p la ce in th e scale of T h eo rem V P a r t I I (cf. T h eo rem V I I (d)) w h en
gen eral recursive schem es are ap p lied to them .
The d istin ctio n (for k> 0) b etw ee n th e tw o ^ -q u an tifier form s is
lost w h en p rim itiv e recu rsive operatio ns are ap p lied . For e x a m p le ,
a lth o u g h (Ex)Tx(af a t x) is n o t exp ressib le in th e oth er o n e -q u a n tifie r
form (x)R(a,x)f its n eg atio n (ExjTâ, a, x) is p rim itiv e recursive in
it (# D § 4 5 ), an d assum es th a t oth er form (cf. *86 § 3 5 ). For any
p red icate P (a), th e represen tin g p red icate P (a, w) (§ 41) of th e represen t
in g fu n ctio n (§ 45) of P (a) can b e expressed thus,
(41) P(af w) ss {P (a) & w = 0} V (P (a ) & w = 1}.
T h is is p rim itiv e recursive in P (a). W h e n P (a) == {E x)T x(a9 a, x), P (a , w)
is expressible in n eith er of th e o n e-q u a n tifie r fo r m s : e.g. if P (a , w) =
(x)R(af w, x), th en (x)R(a, 0, x) = P (a , 0) s= P(a) = {E x)T t {af a , x )f w h ich
co n tra d icts T h eo rem V if R is recursive.
M o sto w ski 1948 a g iv e s (classically) an e x a m p le of a p red icate , w h ic h
(b y (b) of th e n e x t theorem ) is gen eral recu rsive in p red icates o f th e
1 -q u a n tifier form s, b u t w h ich ca n n o t b e expressed in term s of p red icate s
of th e 1 -q u a n tifier form s b y th e operatio ns of th e p ro p o sitio n al calcu lu s.
T h e fo llo w in g theorem an d co ro llary are P o s t ’s on th e b asis o f an
a b stra c t (1948 ), as th e au th o r b eca m e aw are a fter w o rk in g o u t th e
present tre a tm e n t (in 1949)„
T heorem X I c , (a) I f a predicate P is general recursive in predicates

Qv . . . , Qi of the k-quantifier form s of Theorem V , then P is expressible in
both the k ~f 1 -quantifier form s, and (b) conversely.
T h e p roof of T h eo rem V I (b) establish es th e present (b), w h e n reread
ta k in g th e R an d S n ow to b e of ap p ro p riate ^ -q u an tifie r form s. T h e
proof of (a) w ill b e co m p le ted fo llo w in g th e lem m as.
L emma The representing predicate of the representing function of

IV c.
a predicate of either k-quantifier form is expressible in the k -quantifier
form with existence first.
P roof o f L emma IV . F o r ex am p le , suppose P (a) == (x)(E y)R (af x , y)

w ith recursive R. Then
P (a ,w ) = {(x )(E y )R (a ,x t y) & w = 0 } V {(x )(E y)R (a ,x ,y) & w = \} (b y (41))
^ ■ {(x J iE yJ R ia , xlf y 1) & w = 0} V {(E x2){y2)R {a, x 2, y 2) & w = 1}

(cf. *8 5, * 86 )
= ( E x ^ x ^ y ^ E y ^ iiR ia , x lf yx) & 0} V {R (a, x 2> y 2) & w = 1}]

(cf. *89— *92)
^ (E x)(y)(E z)[{R (a, (y)0, z ) & w = 0} V {*(« , x, (y),) & w = l}] (b y (18)).
T h e last expression is of th e form (E x)(y)(E z)R (a, w, x, y , z) w ith re

cu rsive R. (This an d th e n e x t lem m a could also be p ro ved for g e n e ra lity
first.)
L emma V. Let %(aly . . . , a m) be the course-of-values junction for

ty{alt a m). I f the representing predicate of ^{aly . . . , a m) is expressible
...,
in the k + 1 -quantifier form with existence first, so is the representing
predicate of $(a, . . . , a).
P roof of L emma V . F o r ex am p le, ta k e m = k = 1 . B y h yp o th esis,
ty(a )= w = (Ex)(y)Q (a, w, x, y) w ith recursive N ow
ty(a )= w ^ w = U & {i)i< M i ) = { w ) i

i<a
= w = R & (i)i< a{E x)(y)Q (i, (w)0 x, y)

i<a
= W = n pW i & {E x)(y){i)i<aQ(i, (w)t, (x)0 y) (b y R e m a rk 1 § 57)

i<a *
= {E x)(y){w = II p'f'i & (i)i<aQ{i, (w)it (x)it y)} (cf. * 9 1 , *89).

i<a
T h e last expression is of th e form (E x)(y)R (at w, x, y) w ith recursive 2?.
P roof of T heorem X I (a). F o r exam p le, ta k e l == m = k = 1. T h e

h yp o th e sis th a t P is gen eral recursive in Q m eans (§ 55) th a t P is gen eral
recursive in th e represen tin g fu n ctio n of Q. B y L e m m a s I V an d V ,
ty(a )= w =s (E x)(y)R (af w, x, y) w ith recursive R . W e sh ow first th a t
P (a) is expressible in th e 2 -q u a n tifier form w ith g e n e ra lity first. U sin g
T h eo rem V I * (a) w ith its proof from T h eorem I V * (7),
P (a) = (f)T}($(<), g, a, t) = (f)(s)[$ (/)= s -> T\(s, g, a, <)]

= V )(s)[(E x){y)R (t, s, x, y) - * T\(s, g, a, *)]
= W(s) (*)(£>)[/?(<, s, x, y) -+ T\(s, g, a, /)] (cf. *96, *98)
= W ( £ y ) [ R ( W o , W x, ( * ) „ y) -+ r } ( ( * ) lf g, a, (*)„)] (b y (18)),
v h ich is of th e desired form . T o express P{a) in th e oth er 2 -q u a n tifier

form , w e h a v e sim ila rly u sin g T h eo rem I V * (6 ),
t) = (E t)(E s) W ) = s & T \(s, f, a, t)}, etc.

P («) ^ ( £ * ) r i( $ ( 0 . /, «,
Corollary 0. F or each of the k-\-\-qu an tifier form s, the predicate of

Theorem V (b) which is expressible in the other k-\-\-qu an tifier form but
not in the form itself is not recursive in predicates expressible in the form s
with k {or fewer) quantifiers.
We get a T h eo rem X I * (w ith corollary) by re p lacin g “ gen eral re
cu r s iv e ” for th e form s of T h eo rem V (b) b y “ gen eral recu rsive in Y ” .
*§ 59. General recursive functions and the number-theoretic

f o r m a l i s m . In th is section , “ (i)” — “ (vii)” w ill refer to § 4 1 .
W e s a y th a t a n u m b er-th eo retic p re d ic a te P {x x, ..., x n) is resolvable
in a form al sy ste m (or decidable w ithin th e sy ste m ), if there is a n u m eral-
w ise d ecid ab le form u la P ( x x, . . . , x n) (cf. (iv)) w ith no free va ria b le s
other th a n th e d istin c t va ria b le s x x, . . . , x n such th a t, for each n -tu p le
xv ..., x n of n a tu ra l num bers,
(viii) P {x v = I- P ( x 1, . . . , x n).
In th is c a ie , P ( x x, . . . , x n) resolves P {x ly ..., x n) (w ith th e o b v io u s
correspondence of form al to in tu itiv e variab les).
A n u m b er-th eo retic fu n ctio n <p(xlt ..., x n) is reckonable in a fo rm al sy s
te m (or calculable w ithin th e sy ste m ), if there is a form u la P ( x x, . . . , x n, w)
w ith no free v a ria b les o th er th a n th e d istin c t v a ria b le s xv . . . , x n, w
such th a t, for each x v . . . , x n, w f
(ix) <?(xv . . . t x n) = w s= b P ( x lf
In th is case, P ( x x, . . . , x n, w) reckons <p(xv . . . , x n).
T heorem 32. Let S be the number-theoretic form al system of Chapter I V

{or Robinson's system described in Lem m a 18b § 49). I f S is sim p ly con
sistent, then: {<p is general recursive} = {9 is num eralw ise representable in S }
s= {<p is reckonable in S}.
P roof . W e estab lish three im p licatio n s.
(a) I f (p is general recursive, then is num eralwise representable in S.

 • • • > y )= w =

Sn (^, ) • • •, Xnt w) & 5n {c, X^f • • *, Xnt Z).
By C o ro llary T h eo rem 27 § 49, since S „ (z , xv . . x„, w) is p rim itiv e
recursive, it is num eralw ise expressed b y a formula S(z, xv . . . , x n, w).
W e sh all p ro ve th a t th e form ula
x v . . . , x n, w) & V z ( z < w D i S ( e , x lt . . . , x n, z)),

S(e,
call it “ M (xv . . . , x n, w)", num eralw ise represents

\uyTn(e, x v . . . , x n, y). L e t x v be a n y fixed w -tuple. T o estab lish
(v) for M, suppose that (iy T n(e ,x lt . . , , x nf y) — w. T h e n usin g (42),
an d (E) an d (C) § 41 with (i) an d (ii) for S, (- M ( x lf . . . , x n t w ). Thus
w e h a v e (v) for M. B u t th en w e also h a v e (vi) b y * 1 7 4 a (w ith w as
th e t). So i i y T n (e, x v . .., x n, y) is numeralwise representable.
B y T h eo rem 27, since U(y) is p rim itiv e recursive, it is num eralw ise
representable.
B y th e reasoning for C ase (IV) in th e proof of T h eo rem 27, th en th e
co m p o site fu n ctio n U (\iy T n (e , x l9 . . . , x n, y)), i.e. (by (30)) <p(xlt . . x n),
is num eralw ise representable.
In th is proof, besides th e p red icate calculus, A x io m s 14 — 2 1 , th e
replacem ent p ro p e rty o f e q u a lity , an d results a lre a d y em p lo y e d in
o b ta in in g T h eo rem 27, we h a v e used o n ly (E) an d * 1 7 4 a (w ith th e
num eral w as th e t). H en ce, usin g L em m as 18a § 41 an d 18b § 49, th e
im p lica tio n (a) holds also for R o b in so n 's form al system .
(b) I f s is sim ply consistent, and <p is num eralwise representable in S,

then 9 is reckonable in S (and any form ula P which num eralwise represents
<p reckons cp (and b y § 41 is num eralwise decidable)).
Consider a n y fix e d xv . . . , x n. W e h a v e (v) b y h yp o th esis. T o estab lish

th e converse, assum e th a t (- P(xx, . . . , x n, w). N ow
x n) = w V <p(%, . . . , x n) ^ w (cf. * 1 5 8 §4 0 ). B u t if
<p(#i, . . . ,
<p(xv ...,x n) ^ w , th en usin g (vii) w e co n tra d ict th e sim ple co n sisten cy.
(c) I f cp is reckonable in S, then <p is general recursive.
S u ppose cp(%, . . . , x n) is reckon ed b y th e form u la P ( x 1? . , . , x n, w).

I f S is th e fu ll n u m b er-th eo retic syste m , let P / P be th e p rim itiv e re
cu rsive p red icate correspon ding b y th e G o d el n u m b erin g to ^$fp (cf.
D n l 3 a § 51 an d L e m m a 19 § 52, or T h eorem 31 § 52). I f S is R o b in so n 's
sy ste m (L em m a 18b), w e m u st first m ake th e ap p ro p riate ch an ges in
D n 8 . N o w (%) . . . (xn)(E y)P fp (x 1, . . . , x „ , (y)0, (y),) an d =
([xyP/p {xv . . . , x n, (y)0, (y),)),,. H en ce usin g T h eo rem III §57, 9 is
gen eral recursive.
Corollary . U nder the hypotheses of the theorem, and if

(xn)[P (x x, . . . , x n) V P (x v . . x ^ ] : {P is general recursive }
(%) . . . =
{P is num eralwise expressible in 5 } = {P is resolvable in 5 }.
§59 THE NUMBER-THEORETIC FORMALISM 29 7
P roof . L e t <p b e th e represen tin g fu n ctio n of P. W e estab lish fou r
im plicatio n s,
{P is general recursive} ^ {«p is num eralwise representable in S }

^ {P is num eralwise expressible in S }
^ {P is resolvable in S (and any form ula P which num eralwise expresses

P resolves P )}
{P is general recursive },
where the sim ple consistency of S is a further hypothesis (f) and (g), fo r
and also ( % ) . . . (xn) [P (xlf . . . , x n) V P ( % , x n)] for (f) (to m a k e

th e proof of (f) v a lid in tu itio n is tic a lly ).
(d) F ro m (a) b y th e d efin itio n o f gen eral recursive pred icate.
(e) A s in th e proof of C o ro llary T h eo rem 27 § 49.
(f) W e o b ta in (iv) as in § 4 1. W e h a v e (i) b y h yp o th esis. T o estab lish

th e con verse, assum e th a t b P(*i> . . . , * * ) . N ow
P (x v ..., x n) V P (x lt B u t if P (x v . . . , x n), th e n u sin g (ii) w e
co n tra d ict th e sim ple co n sisten cy.
(g) Su p p o se t h a t th e form u la P ( x 1, ..,, x n) resolves P (x lf ..., x n).

S im ila rly to (c),
(*i) • • • (*n)(E y)[P fp(xv

• • • >* «. y) V P f-,A x v • • •>*»> y ) ] (since P is
num eralw ise decidable) an d
P (x v . , . 9x n) s= P fp (x v . . . , x n,v.y[Pfj>(xv . . . , x n,y ) V P f^ P(x l t . . . , x n,y )])
(using (viii) an d th e sim p le co n siste n cy o f S). (T h e sim ple co n siste n c y
of S is im p lied b y (viii), if (E x^ ... (E xn)P (x v . . x n).)
The eq u ivalen ce of re c k o n a b ility (resolvability) to gen eral recur
siveness w as p ro v e d b y M o sto w sk i 19 4 7 for a n y s im p ly co n sisten t S
co n ta in in g th e u su a l n u m ber th e o r y an d su ch t h a t (R x) p rim itiv e re
cursive p red icates are re so lv ab le in S (cf. C o ro llary T h eo rem 27) and
(R 2) th e p red icates P fA are p rim itiv e recursive (cf. D n l 3 a a n d L e m m a 19).
O th er references w ill b e g iv e n in § 62. C f. R . M, R o b in so n 1950 a b stra c t.
Let T b e a list . . . , fa of fu n ctio n s o f m lt v a ria b les, re
sp e c tiv e ly . C hoose d is tin c t p red icate le tters Qi» . . . » Qi , an d annex
th em to th e sto c k o f form al sy m b o ls for S. E x t e n d th e d efin itio n o f
form u la for S b y p ro v id in g t h a t Q i (t1, . . . , t mj, t) b e a form u la for each j
(| = 1, . . . , £ ) t v , . . , t m., t. L e t
a n d term s b e th e set o f th e
form ulas Q ,(y v . . . , y mjt y ) & V z(Q ,(yv . . . , y mji z) D y = z ) w here

btyv . = y, for / = 1, . . . , / an d all m r tu p les y v . . . , y mj of
n a tu ral num bers. L e t notion s sim p ly consistent (num eralwise decidable,
num eralwise expressible, num eralwise representable, resolvable, reckonable)

in (or from) Y be o b ta in ed b y ch a n gin g to “ F q ^ " qJ h” (or in
w ords, " p r o v a b le ” to “ d ed u cib le from F ^ g ”) in th e form er d ef
inition s. T h eo rem s 27, 3 1, 32 an d corollaries (including (a) — (g)) h old
w ith th ese n otions an d " F q *” [q * b ” > " p r im itiv e recursive in Y ” and
"g e n e ra l recursive in Y ” in p lace of th e resp ective form er notions. (Also

P is num eralwise expressible uniform ly in (or from) Y , if a form u la P can
b e g iv e n in d e p e n d e n tly of Y w h ich num eralw ise expresses P from Y
for each ch oice of Y under co n sid e ra tio n ; an d sim ilarly for num eralw ise
re p resen ta b ility, re so lv a b ility an d re ck o n a b ility . The theorem s h o ld
w h e n th e notions for P or cp re la tiv e to Y are u n d erstoo d th ro u g h o u t as
uniform .)
E q u iv a le n tly , w e m a y in stea d choose d istin ct fu n ctio n letters gv ..., g lt
e x te n d th e d efin itio n of term to in clude g ;(t1} . . . , t m.) w h en t lf . . . , t m/
are term s, an d su b stitu te h” (§ 54) for “ h” in th e form er
definitions.
I f w e le t Y in clude predicates, w e m a y proceed in either of these
w a y s , ta k in g as ^ th e represen tin g fu n ctio n of w h en th e /-th o f th e
Y ’s is a p red icate Q j ; or in stea d for each such j we m a y in tro d u ce a pred
ic a te le tte r Q , to b e used w ith m j argum ents, an d le t th e assu m p tion
fo rm u la correspon ding to th is j an d a g iv e n m r tu p le y v . . . , y m- b e
Q j(y v or -I Q i(y v • • •, ymf) acco rd in g as Q ,(ylt . . . , y m,) or
Q i(y ........... ym,)-
§ 60, Church’s theorem, the generalized Godel theorem. We

n o w u n d erta k e to answ er fin a lly th e qu estion w h eth er inform al m a th e
m a tic s ca n b e c o m p le te ly fo rm alized (§ 15). W e k n o w b y G o d e l's theorem
(§ 42) t h a t th e p a rticu la r form al sy ste m of C h a p te r I V does n o t co m
p le te ly form alize in tu itiv e num ber th eo ry.
In in form al n u m ber th e o r y we consider propositions d ep en d in g
on param eters. In fin ite ly m a n y p articu lar propositions arise, acco rd in g
to th e n a tu ra l n um bers ta k e n as v a lu es b y th ese param eters. W e describe
th is situ a tio n b y sa y in g t h a t th e propositions are th e v a lu es of predicates.
G e n e ra lly w e h a v e und er con sideration a n um ber of p red icates sim ul
ta n e o u s ly . H o w ev er, le t us discuss th e fo rm alizatio n in relation to one of
th em , s a y a p red icate P (x) of one v a riab le
The form al n u m b er-th eo retic sy ste m of earlier ch ap ters p rovid es
one illu stra tio n of h ow form alization can b e carried out. L e t us p u t
§60 THE GENERALIZED GODEL THEOREM 299
aside all th e finer details, an d consider o n ly th e b are sk eleto n of w h a t
a form al sy ste m m u st p ro v id e if it is to serve its purpose.
T h e form al sy ste m m u st h a v e som e d om ain of ‘form al objects*. A m o n g
these, if th e sy ste m is to c o n stitu te a fo rm alizatio n of th e th e o r y of th e
p red icate
P (x) for x
P (x),
=
P(
there m u st b e p articu lar d istin ct form al o b je c ts w h ich
w e id e n tify as expressing th e propositions 0), P ( l ) , P ( 2 ), . . . ,
0 , 1 , 2 , . . . . W e m a y c o n v e n ie n tly d esign ate these form al
i.e.
o b je c ts b y “ A (0 )*’ , “ A (l)**, “ A(2)**, . . . , re sp e ctiv e ly , i.e. “ A(x)** for

x = 0, 1,2,...; an d ca ll A (x ) ‘th e form ula exp re ssin g P {x )\ W e do
this w ith o u t m a k in g a n y assu m p tio n as to w h e th er there b e such th in g s
as num erals x , a v a ria b le x , an d a form ula A (x ) from w h ich A ( x ) com es
b y s u b stitu tin g x for x.
N e x t, there m u st b e a c a te g o r y of form al o b je c ts called ‘proofs*.
E a c h proof m u st be a proof ‘of* a p articu lar form al o b je c t, w h ich m a y
or m a y n o t be th e o b je ct A ( x ) for a g iv e n n a tu ra l n um ber x. L e t “ SR(x, Y)**
stand for th e m e ta m a th e m a tic a l proposition th a t Y is a proof of A (x ).
T h e o b je c t A ( x ) is said to b e ‘provable*, if there is a proof o f it. L e t
“ b A(x)** sta n d for th e proposition th a t A ( x ) is p r o v a b le ; i.e.
{43) , ( E Y ) 9 i< * ,Y ) s h A (x ).
W h a t is th e n atu re of th e p red icate 9t(x, Y ) ? O u r purpose in form al

izin g a th e o ry is to m ake e x p lic it th e con dition s w h ich d eterm ine w h a t
propositions h old in th e sense of b ein g p ro v a b le in th e th e o r y (§ 15), or
in brief to g iv e an e x p lic it d efin ition of w h a t co n stitu te s a proof. T h is
purpose w ill b e accom plish ed , for th e th e o ry of th e p red icate P (x) as
expressed b y th e form ula A ( x ) , o n ly if there is a preassigned procedure,
n o t requiring a n y m a th e m a tic a l in v en tio n on our p a rt to a p p ly , b y w h ich ,
w h en ever w e are g iv e n a p a rticu la r n a tu ral n um ber x an d form al o b je c t Y ,
w e can te ll w h eth er Y is a proof of A (x ) for t h a t x. T h a t is, th ere m u st
b e a decision procedure or algo rith m for th e q u estio n w h e th er 9t(x, Y )
holds (§ 30). W e sh all also express th is b y sa y in g th a t 9l(*, Y ) m u st b e
an “ e ffe c tiv e ly decidable** m e ta m a th e m a tic a l pred icate.
W e h a v e n o t s tip u la te d w h a t k in d o f stru ctu re th e d om ain o f th e
form al o b je c ts sh ould h a v e . H o w e v e r each form al o b je c t Y m u st b e
ca p a b le of b ein g g iv e n as a fin ite o b je c t, as otherw ise it w o u ld m a k e no
sense to sp eak of a decision procedure for $t(x, Y ) . T h is m a y m ean th a t
each Y can b e gen erated from a fin ite n u m ber of in itia l o b je cts b y a
fin ite n um ber of ap p licatio n s of recognized operations, as in our co n
cep tio n of a gen eralized a rith m etic (§ 50); or th a t each Y ca n b e g iv e n
as a figure c o n stitu te d o u t of a fin ite num ber of occurrences of sy m b o ls
from a preassigned enu m eration of sym b o ls. B y fam iliar m eth o d s (§§ 1,
50, 52) w e ca n th e n g iv e an e ffe c tiv e enu m eration of th e form al o b je c ts ,
or perh aps m ore c o n v e n ie n tly an e ffe c tiv e G o d el n u m berin g, i.e. a
1-1 correspondence b etw ee n th e form al o b je cts an d a su bset of th e
n a tu ral num bers. T h e effective n ess m eans th a t, g iv e n a n y form al o b je c t
Y w e can a lw a y s fin d th e corresponding num ber y , an d in v e rse ly g iv e n
a n y n a tu ra l n u m b er y w e can a lw a y s d eterm ine w h e th er it correspon ds
to a form al o b je c t, an d if so find th a t o b je ct Y . T h e r e b y w e correlate t o
th e e ffe c tiv e ly d ecid ab le m e ta m a th e m a tic a l p red icate 5ft(x, Y ) an ef
fe c tiv e ly d ecid ab le n u m b er-th eo retic p red icate R {x, y), w here R (x , y) = {y
is th e n a tu ra l n u m ber correlated to a form al o b je c t Y such th a t 9t(#, Y ) } .
Then (£ Y )R (x, Y ) = (E y)R (x, y ); an d b y (43),
(44) (E y)R (x, y) - b A (x ).

W h a t k in d of a n u m b er-th eo retic p red icate ca n m eet th e co n d itio n
on R (x ,y ) of b e in g e ffe c tiv e ly d ecid a b le? F o r th e form al sy ste m of
C h a p te r I V , an d a n y fo rm u la A (a) in th a t sy ste m , th is p red icate R (x, y)
(corresponding b y th e G o d el n u m berin g to ‘ Y is a proof of A ( x ) ') is
p rim itiv e recursive, as w e sh ow ed in p ro v in g T h eo rem 31 § 52.
Any gen eral recursive p red icate is e ffe c tiv e ly d ecid able. For any
gen eral recursive fu n ctio n <p is e ffe c tiv e ly calcu lable. G iv e n a sy ste m E
of eq u a tio n s d efin in g 9 re cu rsively, an e ffe c tiv e process for fin d in g th e
v a lu e of 9 for g iv e n argu m e n ts x v . . . , x n is afford ed b y d ed u cin g e q u a
tio n s from E u n til one f ( x 1? . . . , x n) = x expressin g th a t th is v a lu e is x
is found. S u c h a s y ste m E a lw a y s does e x is t an d such an eq u a tio n can
a lw a y s b e ded u ced , acco rd in g to th e d efin itio n s of 'gen eral recu rsive
fu n ctio n ' (§ 55) an d ‘E defines <p re cu rsively' (§ 54). (W e can also see t h a t
9 is e ffe c tiv e ly ca lcu lab le, b y considering T h eo rem I X (29) an d (30)
§ 58, or C o ro llary T h eo rem I X .) T h e n if R is a gen eral recursive p red ic a te ,
w e ca n decide w h e th er it is tru e or false for g iv e n a rgu m en ts xv ..., x n,
b y c a lc u la tin g th e v a lu e of th e represen tin g fu n ctio n for th ose argu m e n ts,
a n d read in g w h e th er t h a t v a lu e is 0 or 1 .
T h e con verse of th is has seem ed also to b e true. E v e r y ex a m p le o f a
fu n ctio n (predicate) a c k n o w led g ed to be e ffe c tiv e ly ca lcu la b le (de
cid ab le), for w h ich th e q u estio n has been in v e stig a te d , has tu rn ed o u t to
b e gen eral recursive. T h is h eu ristic evid en ce an d o th er con sid eration s
led C h u rch 1936 to propose th e fo llo w in g thesis.
T hesis I. E very effectively calculable function {effectively decidable

predicate) is general recursive.
T h is thesis is also im p licit in th e co n cep tio n of a co m p u tin g m a ch in e
§60 THE GENERALIZED GODEL THEOREM 301
fo rm u la te d b y T u r in g 1936 -7 a n d P o s t 19 3 6 . W e p o stp o n e th e d iscussion

o f th e e v id e n ce for th e th esis to th e n e x t tw o ch ap ters, an d proceed a t
on ce to consider its im p licatio n s.
T h e th esis an d its con verse p ro v id e th e e x a c t d efin itio n o f th e n o tio n o f
a ca lcu la tio n (decision) proced ure or algo rith m , for th e case o f a fu n ctio n
(predicate) of n a tu ra l num bers, w h ich w e w ere n o t re a d y to p ro v id e in
§ 30. T o g iv e a decision procedure for a p red icate P {x) th u s m eans to
g iv e a gen eral recursive p red ic a te i?(*) such t h a t P (x) = R {x). B y
T h eo rem V (14) an d (15) § 5 7, w e h a v e th e fo llo w in g theorem , o rig in a lly
sta te d b y C h u rch 1936 w ith a d ifferen t e x a m p le o f an “ u n s o lv a b le ”
decision problem .
T heorem There is no decision procedure {or algorithm) for

X II.
either of the predicates {y)T x{x, x, y) or (E y)T x (x, x, y).
F ro m th e con sideration s in th e first p a rt of th is section , b y a p p ly in g
T h esis I to th e R of (44), w e o b ta in a second thesis.
T hesis I I . F or a given form al system S in which the values of a pred
icate P (x) are expressed by designated distinct form ulas A (x ) (x = 0 , 1 ,
2 , . . . ) , the predicate 'A ( x ) is provable in S ’ is expressible in the form
(E y)R {x} y) where R is general recursive, i.e. there exists a general recursive
predicate R such that (44) holds. {S im ilarly for a predicate P {x lt . . . , x n)
of n variables.)
R emark 1. H ere w e h a v e considered o n ly th e form ulas A ( x ) in
w h ich w e are interested. I f w e ta k e o th er form ulas in to a c co u n t, an d
th e s y ste m satisfies our u su al co n cep tio n s of w h a t it sh ou ld accom plish ,
w e can a n a ly z e further. L e t “ *g(A, Y ) ” d en o te ‘Y is a proof of A ' as a
p red icate of a n y tw o form al o b je c ts A , Y . T h is p red icate *JS(A, Y ) w ill
b e e ffe c tiv e ly d ecid ab le, A ( x ) w ill b e an e ffe c tiv e ly ca lcu la b le fu n ctio n of
#, an d th e e ffe c tiv e d e c id a b ility of 9i(#, Y ) w ill result by ta k in g
9t(*, Y ) = sj$(A(x), Y ) . F o r an e ffe c tiv e G o d el n u m b erin g of th e fo rm al
o b jects, le t“A a” s ta n d for th e o b je c t (if an y) h a v in g a as its n u m b e r ;
an d le t “ b A a” express t h a t th is o b je c t is a p ro v a b le fo rm u la (“ b A 0”
b ein g false w h en a is n o t a G o d el num ber). T o ^$(A, Y ) w e correlate P {a, y).
U sin g T h esis I : (a) There is a general recursive predicate P such that
{E y)P {a, y) = b A a. (b) The Godel num ber ol{x ) of A ( x ) is a general
recursive function of x. T h e sis I I follow s b y ta k in g R {x, y) == P { ol{x ), y).
T h esis II fo rm u lates th e (m inim um ) stru ctu ra l requirem en t on a
form al system 5 in order for it to serve as a fo rm alizatio n of th e th e o r y
of a p red icate P , w ith th e propositions P{x) for x = 0 , 1 , 2, . . . expressed
b y th e re sp ective form ulas A ( x ) . A s in th e case of T h esis I, th e con verse

h olds (as sta te d below ).
I t rem ains to sta te th e con d ition s on S re latin g p r o v a b ility of th e for
m u las A ( x ) w ith th e propositions P(x) w h ich th e y are in ten d ed to express.
F o r th e fo rm alizatio n to b e correct (or consistent) for P (x), it is required
th a t
(45) b A(x) - P (x),

i.e. A (x ) is p ro v a b le in S o n ly w h en P(x) is true. I f th e sy ste m S is also
to c o n stitu te a complete fo rm alizatio n of th e th e o ry of th e pred icate, w e
m u st also h a v e , co n versely, th a t P (x) -> p A ( x ) , i.e. A (x ) is p ro v a b le
w h en ever P (x) is true. C o m b in in g th is w ith (45),
(46) h A (x ) = P(x)
in th e case th a t S is co m p le te as w ell as correct for P {x).
C o m b in in g th is w ith th e stru ctu ra l requirem ent g iv e n m T h esis I I :
T o g iv e a correct form al sy ste m for th e p red icate P (x) en tails fin d in g a
gen eral recursive p red icate R such th a t
(47) (E y)R (x ,y) -* P (x);

to set u p one th a t is also co m p lete, such th a t
(48) (E y)R (x, y) = P (x).

N o w suppose P (x) is th e p red icate {y)T 1(x, x, y). T h e n there is no
gen eral recursive R such th a t (48) holds for all x \ for th en b y T h eorem
V ( 12 ), g iv e n a n y general recursive R } there is a num ber / such th a t (48)
fails for x = /. T h u s :
T heorem X I I I (P art I). There is no correct and complete form al

system for the predicate (y)T 1(xt x, y). (A gen eralized form of G o d e l’s
th eorem , K le en e 19 4 3 .)
T o ex am in e th e situ a tio n in m ore d etail, consider a n y form al sy ste m

S an d ch oice o f form ulas A (x ) in S to express (y)T 1(xt x, y) for x = 0,
1 , 2 , -- L e t R b e th e gen eral recursive p red icate such th a t (44 ) holds
(given b y T h esis I I ) ; an d consider th e / of (9) for th is R. Suppose S is
correct for (y)T 1 (x>x f y); th e n b y (45),
(49) b A (x ) -> (y)T x(x ,x ,y ).

B y th e 'in fo rm a l co u n terp art of *86 § 35, (9 ) an d (44 ),
(50) (y)T , (/, /, y) = (E tiT yU , f ,y ) - {T y)R {fi y)

A ssu m e b A ( f) . T h e n b y (49), {y)T 1 (/, /, y); an d b y (50), b A (/). By
red u ctio ad ab surdum , b A ( f ) ; an d b y (50), {y)T 1(ft f f y). T h u s:

§60 THE GENERALIZED GODEL THEOREM 30 3
T heorem X III (P art In particu lar , suppose S is a form al system
II).
with^ distin ct form ulas A (x ) designated as expressing the propositions
(y)7 \ {%> x, y) for x = 0 , 1 , 2 , . . . . Then a number f can be found such that:
I f S is correct for {y)T x{xt x , y ), then (y ff^ f, /, y) & b A (f ) ; i.e. the propo
sition (y)T 1(f>/, y) is true , but the form ula A (f) expressing it is unprovable.
T h u s no form al sy ste m can b e co m p le te for th e purpose of p ro v in g th e
tru e (and o n ly th e true) propositions ta k e n as th e v a lu es of a certain p re
assigned in tu itiv e p red icate ( y ) r x(A;, x , y). N o th in g is assum ed a b o u t th e
form al sy ste m , e x c e p t th a t it fu lfils th e stru ctu ral requirem en t expressed
in T h esis II, an d y ie ld s o n ly results correct und er th e in terp retatio n o f
th e form u las A ( x ) as exp ressin g th e va lu es of th e p red icate ( y ) r x(%, x, y).
T h e se assu m p tion s c o n stitu te a v e r y considerable a b stra c tio n from
th e p a rticu la r form al sy ste m s w e h a v e stud ied. In those, a proof con sisted
in a p p lica tio n s of liste d p o stu lates. W e h ad ground s for belief in th e cor
rectness o f each of th ese p o stu la tes sep arately , an d hence in th e sy ste m
as a w hole. T h e G o d el incom pleteness, w e n ow see, does n o t d ep en d on th e
n a tu re o f th is in tu itiv e eviden ce.
T o em p h asize th is, w e can im agin e an om niscient n u m ber-th eorist.
W e sh ould e x p e c t th a t his a b ility to see in fin ite ly m a n y fa c ts a t once
w o u ld en able h im to recognize as correct som e principles of d ed u ctio n
w h ich w e cou ld n o t d isco ver ourselves. B u t a n y correct form al sy ste m
for (y )T 1(xf x t y), w h ich h e co u ld re veal to us, tellin g us h o w it w orks
w ith o u t te llin g us w h y , w o u ld still b e incom plete.
T o u n d ersta n d th e m ean in g of th e propositions (y )T x(x, x, y), o n ly
th e n o tio n of a p a rticu la r e ffe c tiv e ly ca lcu lab le p red icate (indeed, of
one w h ich is p rim itiv e recursive), an d of th e u n iversal q u a n tifie r u sed
c o n s tr u c tiv e ly , are required. L esser co n cep tu a l presupposition s, if a n y
m a th e m a tic a l in fin ite is to b e allow ed, are h a rd ly co n ceivab le.
In u sin g th is p red icate ( y ) r x(x, x, y) on th e m e ta th e o re tic level, as
a lre a d y m ean in gfu l to us, w e h a v e n o t assum ed th a t each v a lu e of it
is eith er tru e or false. W h a t w e can co n clu de b y o n ly fin ita r y reasoning
w ith th is p red icate is enough, ta k e n w ith T h esis I I , to rule o u t th e
p o ssib ility of our ever h a v in g a correct an d co m p le te form al sy ste m for it.
H ere w e h a v e been d e a lin g w ith th e incom pleteness of a n y (correct)
form al sy ste m as a fo rm alizatio n of th e ex istin g in tu itiv e th e o r y of th e
p red icate ( y ) r x(x, x, y). B e ca u se th is in terp retatio n has been h a n d le d
in a fin ita r y w a y , th e theorem ca n b e considered as m e ta m a th e m a tic a l
in th e b roader sense o f th e term .
F o r form al sy ste m s h a v in g som e o rd in ary fo rm a tiv e an d d e d u c tiv e
properties, th e theorem can also b e fo rm u lated m e ta m a th e m a tic a lly in
th e narrow er sense, replacin g th e reference to th e in terp retatio n of th e
form ulas A(x) as exp ressin g th e va lu es of th e p red icate (y)T1(x, x, y)
b y co n siste n cy an d com pleten ess d efined as intrin sic properties of th e
s y s t e m ; e.g. b rie fly th u s:
T heorem X III (P a r t Let S be a formal system having distinct

I I I).
formulas T(x,y), -iT(x, y), A(x) (also written “Vy -iT(x, y)”) and
—iA(x) {also written “-i Vy nT(x, y)”) (x, y = 0, 1,2, . ..). Suppose that
(A) T^x, x, y) -> b T ( x , y ) and Tx{x, x, y) -> b - i T ( x , y ),
(B) pV y-iT(x, y) -> (y){ b “iT(x, y)}, and (C) for some general re
cursive R , (44) holds. Then a number f can be found such that: If S is simply
consistent in the sense that for no x, y both b T(x, y) and b “iT(x, y),
then b A (f). If S is also {^-consistent in the sense that for no x both
(y){ b (x, y)} and b V y iT(x, y), then "»A(f). Thus if S is simply
“i T — b
and {^-consistent, it is simply incomplete in the sense that for some x neither
b A(x) nor b nA(x).
P r o o f . A ssu m e th a t S is sim p ly con sistent, an d (for red u ctio ad
absurdum ) th a t b A (f). T h e n b y (44), (.E y)R (f , y); b y (9), (E yjT ^ f, /, y );
an d b y (A), b T(f, y ) for som e y. B u t also from b A (f) b y (B), b “iT (f, y )
for th isy , co n tra d ictin g th e sim ple con sisten cy. B y reductio ad absurdum ,
b A (f). N o w assum e th a t S is also co-consistent. F ro m b A (f) b y (50),
(y)^ i(/> /> y); bY (A), ^T(f,y)};
(y){\- an d by th e co-consistency,
b -i Vy - i T {f, y ), i.e. b - i A (f).
M ostow ski 19 5 2 , w h ich w as n o t a v a ila b le d urin g th e w ritin g of the
present b o o k , com pares va rio u s proofs of G o d ek s theorem .
T h e incom pleteness theorem s X I I an d X I I I ap p ear in our presen tatio n
as ap p licatio n s of th e cases of T h eorem V for th e re sp ective p red icate
form s R(x) an d (Ey)R{x, y). T h is th em e is d eve lo p ed a t le n g th in K le en e
19 4 3 *. W e con clude this section w ith som e further rem arks a b o u t th e
p red icate form [Ey)R[x, y).
C o n v e r s e o f T h e s is I I (P art I). There is a correct and complete
formal system for any predicate of the form (Ey)R{x, y) with R general
recursive. {Similarly for {Eyj) ... {Eym)R{xlf . . xn,y lf .. ., ym) with
R general recursive {n, m > 0).)
In m o re d eta il (for n = m = 1 ): Given any general recursive predicate
R{x, y), a formal system S with distinct formulas A (x ) for x — 0, 1, 2,
can be found such that (44) holds.
In E x a m p le 1 th e correctness of th e syste m rem ains a h yp o th esis.
§60 THE GENERALIZED GODEL THEOREM 30 5
H o w e v e r E x a m p le 3 pro vides a q u ic k an d u n im p each a ble d em o n stratio n .
(Here (44) is th e (viii) of § 59, b u t in gen eral (iv) is la c k in g for re so lva
b ility .)
E xample 1. By C o ro llary T h eo rem IV §57, there is no loss of

g e n e ra lity in ta k in g R to b e p rim itiv e recursive. L e t S b e th e n u m b er-
th eo retic form al sy ste m o f C h a p te r I V , R ( x , y) b e a form u la w h ich n u m er-
alw ise expresses R (x, y) (C orollary T h eo rem 27 § 49), a n d A (x ) b e th e
form ula l y R ( x , y ). T h e n cle a rly (E y)R (x, y) -> b A ( x ). So (44) holds,
if th e sy ste m h as th e p ro p e rty th a t b 3 y R ( x , y) o n ly w hen ( E y)R (x f y).
T h is is a co n sisten cy p ro p e rty w h ich (if {E x){E y)R {xt y)) im plies sim ple
co n sisten cy, so b y G o d e l’s second theorem (Theorem 30 § 42) w e ca n n o t
e x p e c t to fin d an e lem en ta ry proof of it ; an d in fa c t for th e classical
form al sy ste m it is n o t clear h o w a n y proof of it can b e g iv e n w ith o u t
usin g classical lo gic in th e m eta lan gu age.
E xample 2. S im ila rly ta k in g as S R o b in s o n ’s sy ste m (L e m m a

18b § 49) w ith o n ly th irteen n u m b er-th eo retic axiom s. A le n g th y b u t
elem en tary proof of th e required co n siste n cy p ro p e r ty (for an R ( x , y)
co n stru cted q.s in th e proof of C o ro llary T h eo rem 27) w ill b e g iv e n in
§§ 7 7 — 79 (cf. T h eo rem 53 (b) § 79 ). — A fortiori (or d ire c tly , cf. T h eo rem
53 (a)), R o b in so n ’s sy ste m is sim p ly con sistent.
E xample 3. Let E b e a sy ste m of eq u a tio n s in th e form alism of

recursive fu n ctio n s w h ich defines re cu rsively th e represen tin g fu n ctio n
of R, w ith f as p rin cip al fu n ctio n letter. L e t th e p red icate le tte r be
ad d ed to th e sto c k of form al sym b ols. L e t 5 b e th e sy ste m h a v in g th e
eq u a tio n s of E as its axio m s, an d h a v in g as its rules of inference R l ,
R 2 an d th e follow in g, w here x an d y are num erals:
f ( * .y )= o
X{x).
T h e c o n sisten cy p ro p e rty (w ith <^?(x) as th e A (x )) is im m ed iate.
E xample 4. L e t th e form al sy m b o ls of S com prise o n ly 0 , ', an d

th e tw o p red icate letters 2^ an d (w ith co m m a an d parentheses). T h e
form ulas sh all b e th e expressions K .( x , y) an d c2?(x) for y = 0 , 1 , 2 , -----
T h e p o stu la te s sh all b e an a x io m sch em a 1 an d a rule of inference 2 ,
as follow s. F o r A x io m S c h em a 1 , x an d y are s tip u la te d to b e n um erals
such t h a t R (x , y).
l. K(x,y). 2. K(x, y)
=3(x).
T h is form al sy ste m m a y seem un o rth o d o x. B u t th e stip u la tio n on th e

x an d y for A x io m S ch em a 1 is an e ffe c tiv e one (since R is gen eral re
cursive), ju s t as e.g. th a t on th e t for A x io m S c h e m a ta 10 an d 11 § 19.
A fifth ex a m p le is g iv e n as E x a m p le 2 end § 73.
Converse of T hesis II (P art

There is a form al system S such
II).
that , given any general recursive predicate R (x ,y ), distinct form ulas A (x)
for x = 0, 1 , 2 , . . . can be found such that (44) holds. (Sim ilarly for any n ,
m ; or for all n, m sim ultaneously.)
P roof . Let A (z, x) (z, x = 0, 1, 2, . . . ) b e th e form ulas o b ta in e d by
a p p ly in g P a r t I (for n = 2, m = 1 ) to T ^ z, x, y )\ an d let A ( x ) (x = 0,
3, 2, . . . ) b e A (f, x) for th e / of (9) § 57.
R emark 2. B y a d d in g new p o stu lates to S (in P a r t I or II), w e

o b ta in sy ste m s 5 ' such t h a t ( E y)R (x, y) ->■ b A ( x ) , b u t n o t n ecessarily
h A (x ) - (E y )R {x ,y ).
B y T h esis I I an d its con verse, th e sam e pred icates are expressible in
th e form (E y)R (x} y) for som e gen eral recursive R as are expressible b y
b A (x ) for som e form al sy ste m S an d effe ctiv e d esign atio n of form ulas
A ( x ) for x= 0, 1 , 2 , . . . . B rie fly , th e p red icate form (E y)R (xf y) co in
cides w ith th e n otion of p r o v a b ility in som e form al system .
B y th e result m en tio n ed a t th e end of § 53 (w ith C o ro llary T h eo rem IV )
in d u c tiv e defin ition s (w ith c o n stru ctiv e d irect clauses) lead to precisely
th e sam e class of pred icates. T h is fa c t is clo sely related to th e foregoing,
in v ie w of th e role o fte n g iv e n to in d u c tiv e d efin itio n s in d efinin g form al
system s.
R ecursive enum erability . A set or class C of n a tu ra l num bers is

recursively enumerable , if there is a gen eral recursive fu n ctio n 9 w h ich
en u m erates it (allow ing repetitions), i.e. such th a t 9 (0 ), 9 ( 1 ), 9 (2 ), . . . is
an en u m eration (allow ing repetitions) of th e m em bers of C. (P ost 1944
includes also th e e m p ty class as recu rsively enum erable.)
T heorem A class C having a member is recursively enumerable ,

X IV .
if and only if the predicate x £ C is expressible in the form (E y)R (x , y) with
a general recursive R.
In m ore d e ta il: (a) I f 9 enumerates C , then x £ C == (E y)R (x ,y) with
an R p rim itive recursive in 9 . (b) If x 8 C = (E y)R (x, y) and C has a
member m , then C is enum erated by a function 0 prim itive recursive in R
(K leen e 1936 ).
§60 THE G E N E R A L IZ E D GODEL THEOREM 307
P roofs, (a) x B C = (£y)[<p(y)==%] (cf. # 1 4 § 4 5 ). (b) L e t
(y)o if ?((y) o- (y)i).

w if /?((y)o, (y)i)
(cf. # # D , F , 19).
T h u s T h esis I I is e q u iv a le n t to sa y in g th a t th e class C o f th e n um bers

x for w h ich A (x) is p ro v a b le is re cu rsively enu m erable (if it has a m em ber).
W e call a set or class C general recursive , if th e p red icate x£C is gen eral
recursive. P arap h rasin g results of § 5 7 : A gen eral recursive class C is a
fortiori re cu rsively enum erable, if it has a m em ber (Theorem V I (a));
likew ise its co m p lem en t C (# D § 45). C lassically , a class C is gen eral
recursive, if b o th C an d C are re cu rsively enum erable (Theorem V I
(b) or (c)). T h e class of th e #'s such th a t {E y)T x{x, x f y) (in sym b o ls,
x {E y )T x{x, x ,y )) is re cu rsively enum erable (b y (29) §5 8 , (E yjT ^ e, e, y ),
so e is a m em ber), b u t n o t gen eral recursive (Theorem V (15)); its co m
p lem en t x (y)T 1(x, x, y) is neith er re cu rsively enu m erable nor gen eral
recursive (Theorem V (12) an d (14)).
Corollary. I f a class can be enumerated {allowing repetitions) by a

general recursive function, it can be enumerated {allowing repetitions) by
a prim itive recursive function. (R osser 19 3 6 .)
U sin g (a), th en C o ro llary T h eo rem I V , th en (b).
E xample 5. An in fin ite class C is gen eral recursive, if an d o n ly

if it is recu rsively enu m erable w ith o u t repetitio n s in order o f m a g n itu d e
(K leen e 1936 ). (H i n t : U se T h eo rem II I.) E v e r y in fin ite re c u rsively
enum erable class co n tain s an in fin ite gen eral recursive subclass (Post
1944 ). I f an in fin ite class is re cu rsively enu m erable allo w in g repetition s,
it is re cu rsively enum erable w ith o u t rep etitio n s (K leen e 1936 ).
E xample 6. P roblem ; to define c o n str u c tiv e ly (i.e. in tu itio n istica lly)

‘general recursive fu n ctio n ' from ‘re cu rsively enu m erable class'. W e m u st
a v o id th e a p p licatio n of th e la w of th e ex clu d ed m iddle w h ich occurs in
sa y in g th a t a class is either e m p ty or h as a m em ber, an d th e n on -in-
tu itio n istic step s in th e proof of T h eo rem V I (b). Solution b y co m b in in g
the first of th e fo llo w in g tw o propositions, T h eo rem V I (c), an d th e second
(or T h eorem X I V ) : A fu n ctio n y{x v x n) is gen eral recursive, if an d
...,
o n ly if th e class of th e num bers Xn)-p * i-. . .•p% 1 is gen eral re
•••’
cursive. A p red icate x £ C is expressible in th e form (.E y)R {x , y) w ith
gen eral recursive R , if an d o n ly if th e class (call it { 0 } + C ') con sistin g o f
0 an d th e successors of th e m em bers of C is recu rsively enum erable.
W e o b ta in a n o tio n 'enum erable recu rsively in Y * b y read in g " fu n c tio n

9 gen eral recursive in Y ” in place of "gen eral recursive fu n ctio n 9 " in
th e defin ition of recursive en u m erab ility. T h e results e x te n d to this.
§ 6 1. A s y m m e t r i c f o r m o f G o d e l ’ s t h e o r e m . T h eo rem X I I I § 60
generalizes T h eo rem 28 § 42 (Godehs theorem in th e original form ),
w ith A (f) corresponding to th e A P(p). P a r t I I I of Th eorem X I I I w as
fo rm u lated m e ta m a th e m a tic a lly in th e narrow er sense. W e sh all now
ta k e u p th u s a gen eralizatio n of T h eorem 29 (the resp ective R osser form
o f G o d ei's theorem ).
In p lace of (y)T i(^, %, y), w e n ow use th e s lig h tly m ore co m p lic a ted
p red icate {y)[T x{{x)l t x ,y)_ V (££)0<y7 \ ( ( * ) o, #, z)], or its eq u iv a le n t
(£ y )!T i(M i. y) & (* )* < y L ( M o . * )].

Let “ W 0( x ,y ) ” a b b re v ia te T 1((x)1, x , y ) & ( z ) ls£yT 1((x)0,x ,z ) an d
" W ^ x, y ) ” a b b re v ia te 7 \((x )0, x, y) & {z)s<yT 1{{x)l , x, z).
L e t x be fixed . Suppose there is a num ber y 1 such th a t (i) T 1((x)0, x, y x)
an d (ii) {z)z<y T 1{{x)1, x, z). T h e n there can be no n um ber y 0 such t h a t
(iii) T 1((x)1, x , y 0) an d (iv) {z)z<yJ ^ { x ) a, x, z). F o r (i) an d (iv) im p ly
Vi > Vo* ai*d (ii) an<3 (iii) im p ly y 0 > y^ T h u s
(51) (E y)W 1{x, y) -> (E y)W 0(x, y).
Sin ce th e pred icates W 0(x, y) an d W x(x} y) are p rim itive recursive
(using # 4 f:A, C, D , E , 19 § 45), th e th e o ry of th e p red icate (E y)W 0(x, y)
ca n b e co m p le te ly form alized, an d likew ise a t least as m u ch of th e th e o r y
of ( E y )W 0(x , y) as is g iv e n b y th e su fficien t co n d itio n (E y)W 1(x, y) o f (51)
(b y th e C onverse of T h esis I I an d R e m a rk 2 § 60). W e sh all n o w sh ow
t h a t a form al sy ste m S w h ich form alizes a t least th is m uch , if con sisten t,
ca n n o t b e com plete.
A c c o rd in g ly le t S be a n y form al sy ste m in w h ich there are form ulas
B(x) an d iB (x ) for # = 0, 1 , 2 , . . ., all d istin ct. W e shall n o t m ak e
a n y re strictive assum ption s as to w h a t kin d of sym b o lism 5 has, or in
B(x) com es from a form ula B (x ) b y su b stitu tin g a num eral
p a rticu la r th a t
x for a v a ria b le x, or th a t nB(x) com es from B(x) b y p refixin g a certa in
sy m b o l —1 . T h e d e d u c tiv e rules of 5 shall be such th a t
(52) (E y)W 0(x ,y ) -* h B (x ), (53) (E y)W 1( x , y ) ^ b iB ( x ) .
T h e sy ste m S is to serve th e purpose of g iv in g an e x p licit criterion of

w h a t co n stitu te s proof for th e form ulas B(x) an d nB(x), an d hence
(alth ou gh w e n ow a v o id sp ecify in g th a t B(x) an d -iB(x) should express
certain predicates) it is d em an d ed as before b y T h esis I I th a t there e x ist
§61 A SYMMETRIC FORM OF GODEL’S THEOREM 309
gen eral recursive pred icates R0(x, y) an d R-^x, y) such th a t
(54) (Ey)R0(x,y)= b B ( x ) , (55) (Ey)R1(x,y)

= b n B (x ).
B y th e (simple) consistency of S w e shall m ean th a t for no n a tu ral num ber
x, b o th b B (x ) an d |- “ i B ( x ) ; an d b y th e (simple) completeness th a t
for e v e r y x , eith er j- B (x ) or h -n B (x ).
T heorem X V . There is no simply consistent and complete formal

system satisfying (52)— (55).
In m ore d e ta il: Given any formal system S with distinct formulas B (x )
and - i B ( x ) (x = 0, 1 , 2, . . .) and general recursive predicates R0(x, y)
and Ri(xf y) such that (52) — (55) hold, a number f can be found such that,
if Sis simply consistent, then neither h B (f)n o r b ~ iB (f). (R o sser'sfo rm of
G o d e l's theorem , in a gen eralized version.)
P roof . A ssu m e t h a t S is sim p ly con sisten t, or in sym b o ls
(56) h B (x ) & |- - i B ( x ) .
B y T h eo rem I V (6 ) § 5 7, there are num bers f0 an d fx such th a t, if

we put / = 2/o*3/l, th en
(57) (Ey)R0(x,y) s (Ey)Tx(U, x,y) = (Ey)TM »*>y).

(58) (Ey)R1(x, y) ^ ( E y ) ^ , x, y) s (Ey)Tt((f)v x,y).
In th e rest o f th e proof, each tim e w e use (52) — (58), w e su b s titu te th e
n um ber / for th e v a ria b le x. T o show b y red u ctio ad absu rd u m th a t
b B (f), suppose th a t
(a) b B (f).
T h e n b y (54), (Ey)R0(f, y); an d b y (57),
(b) ( E y ) T M » f 9y).
A lso b y (a) an d th e sim ple co n sisten cy ((56)),
(c) b -« B ( f) .
H en ce b y (55), {Ey)R1{f, y); b y (58), (Ey)T1{(f)1, f, y ); w hence
(d) ( y ) T M v f, y).
B y (b) an d (d), (Ey)[TMo> A y) & f>«)]. i.e. (Ey)W1(f, y );

and b y (53), b “ «B (f), co n tra d ictin g (c). H ence, re je ctin g th e assu m p tio n
(a) b y red u ctio ad absurdum , b B (/).
By sim ilar steps, or sim p ly by o b servin g th e s y m m e tr y b etw ee n
(52), (54), (56), (57) an d (53), (55), (56), (58): b i B (/).

D iscussion . W e m o tiv a te d th e con dition s for S b y su ggestin g th a t

B (x ) should express (E y)W 0 (x, y) an d - i B ( x ) its n egatio n . L e t us ca ll
this, for con ven ience, th e preferred interpretation. U n d er th e preferred
in terpretatio n , - i B (f) corresponds to th e A Q(q) of § 42 an d expresses a
true proposition. H o w e v e r th e preferred in terp retatio n is n o t m en tio n ed
in th e theorem itself. T h e con dition s for S are en tirely sy m m etrical.
T h ere is n o th in g to keep us e q u a lly w ell from in terp retin g - i B ( x ) as
expressing (E y)W 1(xf y) an d B (x ) th e n eg atio n of th a t. T h e n B(f) cor
responds to A q(q) an d expresses a true proposition, w hile - i B ( / ) is false.
B e tw e e n these extrem es, there are m a n y in term ed ia te possibilities for th e
in terpretatio n . W e shall illu stra te th is further in con n ection w ith th e
fo llo w in g exam p les of sy ste m s S for T h eorem X V .
E xam ple 1. In th e n u m b er-th eo retic form alism of C h a p ter IV ,

since T ^ a ) ^ a, b) an d T 1((a)0f a, c) are p rim itive recursive, b y C orollary
T h eorem 27 § 49 th e y are num eralw ise expressed b y form ulas A (a, b)
an d B (a, c), re sp e ctive ly. L e t B (x ) b e 3i>[A(x, b) & >fc(c<,b D - i B ( x , c))],
an d - i B ( x ) b e - i 3 i [ A ( x , b) & V c(e D - i B ( x , c))]. T h e n (52) ca n be
show n to h old b y m eth o d s used in th e first p a rt of th e proof of T h eo rem
29 § 42. A lso b y steps show n there an d D -in tro d .,
b B (x ) D -iV i> [ -iA ( x , b) V 3 c (c < b & B ( x , c))]. C o n trap o sin g (b y *13 ),
b A (x , b) V 3 c (c < b & B (x , c))] D n B ( x ) . T h is w ith th e m eth o d
of th e second p art of th e proof of T h eorem 29 g iv e s (53). N o te th a t th is
proof requires, besides th e p red icate calculus w ith e q u a lity , A x io m s
14 — 21 an d C orollary T h eorem 27, o n ly *16 6 a , *16 8 , * 1 6 6 an d * 1 6 9
(w ith t a num eral). H en ce b y L e m m a 18a (end § 4 1 ) an d L e m m a 18b
(end § 49), it also holds good for th e sy ste m of R obin son. F o r th e
n u m b er-th eo retic sy ste m of C h a p ter I V , w e h a v e (54) an d (55) for som e
recursive R0 an d Rv b y T h eo rem 31 § 5 2 (and for R o b in so n ’s sy ste m ,
b y th e m eth o d of th e proof of T h eorem 31). In th is exam p le, th e - i of
- i B ( x ) for T h eorem X V is a c tu a lly th e - i of th e n u m b er-th eo retic for
m alism ; an d under th e usual in terp retatio n of th e n u m b er-th eo retic s y m
bolism , B (x ) an d - i B ( x ) h a v e th e preferred in terpretatio n . N o w le t S
b e th e n u m b er-th eo retic sy ste m of C h a p te r I V (or R o b in so n ’s), w ith
this ch oice of B (x ) an d - i B ( x ) . B y th e theorem , S (if it is s im p ly co n
sistent) is sim p ly incom plete, an d so is e v e ry sim p ly con sisten t en
largem en t oi S o b ta in ed b y ad d in g m ore p o stu lates (in such a w a y th a t
(54) an d (55) still h old for som e recursive R 0 an d R t). S u ch an enlargem ent
o f S m a y even be a t v a rian ce w ith th e preferred in terp retatio n of B (x )
an d - i B ( x ) in S , p ro v id e d o n ly th a t th e new p o stu lates do n o t co n flict
§61 A S Y M M E T R IC FORM OF G O D E l/S THEOREM 311
w ith th e preferred in terp retatio n in a su fficie n tly ele m e n ta ry w a y as to
g iv e rise to a sim ple in con sisten cy.
E xample 2. Let the symbolism for a formal system S include the

numerals and four predicate symbols W0, Wlt B an d (or in place of
th e last, an operator - 1 ). Let Shave as its p o stu la tes tw o a x io m sc h e m a ta
an d tw o rules of inference, as follow s. F o r A x io m S c h e m a 1, x a n d y
are num erals such th a t W Q(xt y ), a n d for A x io m S c h em a 2 su ch t h a t
W i(x ,y ).
T h e sim ple co n sisten cy o f th is sy ste m is im m e d ia te from (51). T h is

sy ste m S , or a n y sim p ly co n sisten t ex ten sio n of it o b ta in e d b y a d d in g
m ore p o stu la tes (w ith (54) an d (55) rem ain in g tru e for som e recu rsive
R 0 a n d R x) is sim p ly in co m p lete. In th is S th e form ulas B (x ) a n d - i B ( x )
are p ro v a b le o n ly as required b y (52) a n d (53). W e are u n restricted b y
th e in terp retatio n in en largin g S .
T h eo rem X I I I § 60 is th e case of T h eo rem V § 5 7 for th e p red ic a te form

E y)R (x , y),
(. in a m e ta m a th e m a tic a l ap p licatio n . L ik e w ise T h eo rem X V
ad m its a version en tire ly in term s of p red icate form s. L e t u s co m p are
T h eorem s X I I I an d X V u sin g th e la n g u a g e o f re c u rsively en u m erab le
classes (cf. T h eo rem X I V § 60).
R emark 1. T h eo rem s X I I I an d X V m a k e it ab su rd t h a t a n y o f
th e th ree classes ^(E y^T^x, x f y ), x {E y)W Q{x) y) and x (E y)W 1(xt y) be
e m p ty . — M em bers ca n b e fo u n d th u s. U sin g (29) T h eo rem I X § 58,
(E y)T t (et e ,y ). C hoose a n y recu rsiveR such th a t (x)(E y)R (xf y ), an d
choose / for th is R b y (6) T h eo rem I V § 5 7 ; also choose a n y recu rsive R
such t h a t (x)(y)R (xf y ) t an d choose g for th is R b y (7). L e t e0 = 2 <7*3/
an d ex = 2/-3flf. T h e n (E y)W 0(e0t y) an d (E y)W 1(ev y).
In T h eo rem X I I I w e h a v e a fix e d re cu rsive ly enu m erable class C 0

of n a tu ra l num bers (n am ely x {E y )T x{xt x, y)) w hose co m p le m en t C3
(= x (y )T x(xt x, y)) is n o t re c u rsively enu m erable (F igu re 1). In T h eo rem
X V w e h a v e tw o fix e d re c u rsively enu m erable classes C 0 a n d C x (n a m ely,
x {E y)W ^ x , y) an d x (E y )W 1(x>y ) f re sp e ctiv e ly ), w h ich are d isjo in t (b y
(51)) an d such th a t, for e v e r y sep aratio n of a ll n a tu ra l n u m bers in to tw o
d isjo in t classes C 2 an d C 3 w ith C 0 C C 2 an d Cx C C 3, th e classes C 2 a n d C 3
are n o t b o th re c u rsively en u m erable (Figure 2). (In stead o f (x)(x £ C2V
x £ C 3) it suffices to h a v e (x)(x £ C2V x £ C 3), w h ic h is w eak er in tu itio n -

312 G E N E R AL RECU RSIVE FUNCTIONS CH. X I
istica lly.) O n ly for th e preferred in terpretatio n is C 3 th e co m p lem en t of

C Q. In p ro v in g T h eo rem X I I I (P art II), w e supposed g iv e n a n y recu rsively
Figure 1. Figure 2.
enu m erable class D z co n ta in ed in C 3 (nam ely, x(E y)R (x, y)), an d fo u n d a
n u m b er / co n ta in ed in neith er C0 nor D z (Figure la). In p ro v in g T h eorem
X V , w e supposed g iv e n tw o d isjo in t recu rsively enum erable classes D%
an d D 3 co n ta in in g C 0 an d Cv re sp e ctiv e ly (nam ely, x (E y)R 0(x, y) an d
x (E y)R 1(x, y)), an d fou n d a n um ber / co n tain ed in neith er D 2 nor Dz
(F igu re 2a). T h e m e ta m a th e m a tic a l p h raseo lo gy in th e a b o v e proof can
Figure la. Figure 2a.

o f course b e b y p a s se d , b y first u sin g (54) an d (55) in (52), (53) an d (56),
so t h a t th e h yp o th e se s becom e
(52a) {E y)W 0{ x ,y ) - + (E y)R 0(x ,y), (53a) (E y)W 1{x,y) -> {E y)R 1(x ,y),
(56a) (E y)R 0(x, y) & (E y)R 1(x, y)
(K leen e 1950 ). T h ese results e x te n d to classes enu m erable re cu rsively in
Y ; cf. end § 60 an d T h eo rem X § 58.
T h e num bers x for w h ich B (x ) is p ro v a b le are re cu rs ive ly enu m erable
((54), T h eo rem X I V ; b y (52) an d R e m a rk 1 , b B ( e 0)).
T heorem X V I.I f S as described in Theorem X V (<om itting (55)) is

sim p ly consistent , the num bers x for which B (x ) is unprovable in S are
not recursively enum erable , or equivalently there is no general recursive
predicate Q (x , y) such that (Ey)Q (x, y) == b B (x ). (A fter R osser 19 3 6 .)
P roof. F o r if x[bB (x)] w ere re c u rsively enum erable, th en ta k in g
C 2 = x [ p B (x )] an d C 3 ~ x [\- B ( x ) ] , w e w o u ld h a v e th e situ a tio n show n

in F ig u re 2 w ith C 2 an d C 3b o th re cu rsively enum erable. (In tu itio n istic a lly ,
(x)(x 8 C 2 V a: £ C 3) ; c f . *5 1 a § 27.) T h e proof m a y also b e g iv e n b y o b serv

in g th a t (52) — (56) w o u ld h o ld replacin g “R ^ by " Q yi an d b
b y <f b B ( x ) ,,>w hereupon th e form er proof th a t b B ( f) & becom es

§61 A SYMMETRIC FORM OF GODEL'S THEOREM 313
a d ed u ctio n of th e lo gical co n tra d ictio n b B ( f ) & b B ( f) . (W e use th e

h yp o th eses regardin g i B ( x ) an d th e sim ple c o n sisten cy o n ly to infer
th a t (E y)W 1(x,y) -> b B ( i).)
T he decision problem for a formal system (cf. § 3 0 ). B y T h e sis I

§ 60, 'e ffe c tiv e ly d ecid ab le (calculable)' m eans 'gen eral recu rsive' in th e
case of a n u m b er-th eo retic p red icate (function). T o g iv e an e x a c t sense to
'e ffe c tiv e ly d ecid ab le (calculable)' for a m e ta m a th e m a tic a l p red ic a te
(function), we can require th a t th e corresponding n u m b er-th eo retic
p redicate (a corresponding n u m b er-th eo retic function) b e gen eral re
cursive, in th e case of a n y p a rticu la r form al sy ste m S , th e o b je c ts o f
w h ich a d m it an e ffe c tiv e G o d el n u m b erin g (as t h e y m u st, if S is to serve
our purpose in form alizin g, b egin n in g § 60). F o r exam p le , to g iv e a
'decision procedure' for p r o v a b ility in S , i.e. for th e p red icate b A (where
A is a m e ta m a th e m a tic a l v a ria b le ran gin g o ve r all form ulas, or o v e r all
form al o b jects, of S), th en m eans to g iv e a gen eral recursive p red ic a te
R(a) such th a t R(a) == b A a, in th e n o ta tio n of R e m a rk 1 § 60. (A second
m eth od of m a k in g 'e ffe c tiv e ly d ecid ab le (calculable)' e x a c t for a m e ta
m a th e m a tic a l p red ic a te (function) is in d ic a te d a t th e en d o f § 70.) T h e
follow ing can b e read either in term s o f our in tu itiv e co n cep tio n o f a
decision procedure a p p ly in g to form ulas of S , or in term s o f th is e x a c t
m a th e m a tica l definition. T h e co n d itio n th a t B (x ) b e an e ffe c tiv e m e ta
m a th e m a tica l fu n ctio n of x (or t h a t its G o d el num ber (3(#) b e a gen eral
recursive n u m b er-th eo retic fu n ctio n of x) m u st b e m et, if S is to serve
th e purpose of fo rm alizatio n for th e form ulas B (x ).
Corollary . L et S be as described in Theorem X V (om itting (54) and

(55)), and such that B (x ) can be effectively found from x (or that in some
specified effective Godel num bering, its Godel num ber (3(%) is a general re
cursive function of x). I f S is sim ply consistent, then its decision problem is
unsolvable, i.e. there is no decision procedure for- determ ining whether a
form ula is provable in S.
P roof. For if there w ere a m eth o d for d eterm in in g e ffe c tiv e ly
w heth er a n y g iv e n form ula of th e sy ste m is p ro v a b le, one could , g iv e n
a n y n um ber x, fin d th e correspon ding form ula B (x ) a n d th e n a p p ly th e
m eth o d to th a t form ula. B y T h esis I § 60, th is w o u ld im p ly t h a t th e class
* [ b B (x )] is gen eral recursive. T h e n a fortiori th e classes # [ b B (x )] an d
x [\- B (x )] w o u ld b o th be re cu rsively enum erable, an d hence there w o u ld

be gen eral recursive p red icates R 0 an d Q such th a t (E y)R 0(x t y) == b B (x )
(for (54)) an d (E y)Q (x,y) = b B (x ) (cf. end § 6 0 , an d R e m a rk 1 ), co n

tra d ie tin g th e theorem . — O th erw ise s ta te d : Were there a gen eral re
cu rsive R such th a t R(a) == h A a, th en ta k in g R 0(xt y) = 2?(p(#)) an d
Q(x, y) — R (p(*)), w e w o u ld co n tra d ict th e theorem as a b o ve.
T heorem I f the number-theoretic form al system of Chapter I V

33.
{or Robinson s system described in Lem m a 18b §49) is sim p ly consistent ,
then its decision problem is unsolvable, and rem ains unsolvable when the
system is extended by adding postulates in any w ay such that the system
rem ains sim ply consistent .
B y th e co ro llary w ith E x a m p le 1 . Sin ce in E x a m p le 1 th e —i o f - i B ( x )
is th e “ i of th e n u m b er-th eo retic sy ste m , ‘sim ple co n sisten cy ' as used in
T h eo rem X V coin cides w ith ‘sim ple co n siste n cy ’ as d efin ed (in § 28) for
th e n u m b er-th eo retic sy ste m . (F or th e d efin ition o f ‘ d e c id a b le ’ b y use
o f a G o d el n u m b erin g, it m a y b e un d ersto o d t h a t th e n u m b erin g is t h a t
o f §§ 50, 52. U s in g E x a m p le 2 § 52, th e n (3(x) is p rim itiv e recursive.)
R educibility , degrees of unsolvability . M u ch o f th e w o rk on d e

cision problem s is d e v o te d n o t to o u trig h t solutions b u t to red u ction s
of one decision pro b lem to another.
T o ‘re d u c e’ th e decision p roblem for a p red icate P (or th e ca lcu la tio n
p ro b lem for a fu n ctio n 9 ) of n v a ria b les to th e re sp e ctive problem s for
l fu n ctio n s an d p red icate s ..., , Qlt ..., Q x%(a b b rev ia ted Y ) m eans
in t u it iv e ly to fin d a uniform m e th o d of procedure b y w h ich , g iv e n a n y
w -tuple of argu m e n ts xv ..., x nt one co u ld decid e w h e th er or not
P {x lt . . x n) is true (calcu late th e v a lu e <p(xv ..., x n)), if, a t each sta g e
o f th e procedure, he h a d a v a ila b le th e v a lu e s of th e fu n ctio n s tylf ..., ^
a n d th e tr u th or fa ls ity of th e v a lu e s of th e p red icate s Qv ..., Qh, for
su ch argu m e n ts as he m ig h t th e n n a m e ; or b rie fly , to e stab lish t h a t
P is e ffe c tiv e ly d ecid ab le (9 is e ffe c tiv e ly calculable) from Y .
T o o b ta in a precise m a th e m a tic a l n o tio n to correspond to th is in tu itiv e
n o tio n , w e n a tu r a lly e x te n d C h u rc h ’s thesis (Thesis I § 60) to in clu d e th e
case of / > 0 assum ed fu n ctio n s a n d p red icates Y (callin g it th e n T h esis
I * ) . T h e e v id e n ce for T h esis I w ill also a p p ly to T h esis I * . T h e co n verse
o f T h esis I * holds.
F o r ex am p le , to reduce th e decision p roblem for a p red icate P {a)
to th a t for an oth er p re d ic a te Q{a) n o w m eans to fin d a p red icate R(a)
gen eral recu rsive in Q(a) an d such t h a t P (a) == R (a), or b rie fly to estab lish
th a t P (a) is gen eral recu rsive in Q{a). W e infer th is from our in tu itiv e
n o tio n of red u ctio n b y T h esis I * ; or if w e ta k e it as a d efin ition , w e ap p e a l
to T h esis I * in asserting t h a t th e defined n otion agrees w ith our in tu itiv e
co n c e p tio n o f reduction .
§61 A SYMMETRIC FORM OF GODEL'S THEOREM 315
P ost 1944 fo rm u lated several m a th e m a tic a l re d u c ib ility co n cep ts.
T h e m o st gen eral of th em , w h ich P o st ta k e s from T u r in g 19 3 9 , is e q u iv a
le n t to th e n o tio n w e o b ta in from T h esis I * . I f th e decision problem for
P (a) is reducible to th a t for Q {a\t an d is u n so lva b le, P o st sa y s fu rth er
t h a t it is of equal lower degree of unsolvability th a n th a t for Q(a) a c
or
co rd in g as th e decision problem for Q(a) is or is n o t reducible to t h a t for
P (a) (cf. § 3). T h e degrees of u n s o lv a b ility are a t least p a r tia lly ordered
(cf. §8). Each of the p red icates (E x)T x(a9 a, x )9 (x)(E y)T 2(a, a, x, y),
(Ex)(y)(Ez)TB(af a, x, y, z), ... (cf. T h eorem V P a r t I I (b) §57) has a
decision problem of h igh est degree of u n s o lv a b ility for p red icates of th e
respective forms (E x)R (a9x )f (x)(Ey)R(at x}y), (Ex)(y)(Ez)R(a, x t y , z)t...
w ith R recursive, as w e sh all p ro v e la te r (E x a m p le 2 § 6 5 ; th is h o ld s
in tu itio n istic a lly , if ta k e n to m ean s im p ly th a t each p red icate o f th e
re sp e c tiv e one o f th e form s w ith a recursive R is recursive in th a t
n on -recu rsive pred icate, w h ich is w h a t is d ir e c tly proved ). C la s s ic a lly
th e n each of these p red icates after th e first h as a decision problem of
h igh er degree of u n s o lv a b ility th a n th e preced in g (using also C o ro lla ry
Theorem X I § 58), an d th e p red icate M (a, k) o f T h eo rem V I I I § 5 7 h as
one of still h igher degree of u n s o lv a b ility (b y T h eo rem X C o ro lla ry (b)
§ 58 an d th e proof of T h eo rem V I I I ) . U sin g T% in stea d of T n, w e o b ta in
degrees of u n s o lv a b ility ascen d in g from t h a t (b y T h eo rem X I * end § 58
a n d E x a m p le 2 § 6 5 for l > 0 ). D a v is (1950 ab stract) explores those
degrees of u n so lv a b ility . P ost 1944 raises, w ith o u t answ ering, th e
q u estio n w h e th er there ex ists a low er degree of u n s o lv a b ility th a n
t h a t of th e decision problem for (E x)T x(a9 a, x).
E xample 3. L e t T h eo rem 33 b e re sta ted u sin g th e stronger con

siste n c y p ro p e r ty o f E x a m p le 1 (or 2 ) § 60 in p la ce o f sim p le co n sisten cy.
In th is w eak er form T h eo rem 33 ca n be p ro v ed from E x a m p le 1 (or 2 ) § 60
b y ta k in g th e R (x t y) there == T x(x9 x t y) an d u sin g T h eo rem X I I § 60.
F r o m th is proof it follow s, b y th e results of E x a m p le 2 § 65 ju s t cited ,
th a t, u nd er th e stronger c o n sisten cy h yp o th esis, th e decision p rob lem for
th e form al sy ste m of C h a p te r I V (or R o bin so n 's, L e m m a 18b), or for a n y
e xten sio n h a v in g t h a t c o n sisten cy p ro p e rty for th e sam e or som e oth er
R (x , y) w h ic h n um eralw ise expresses R {x t y) (= T x(x9 x 9 y ))9 is of th e
h igh est degree o f u n s o lv a b ility for 1 -q u a n tifier predicates. — F o r th e
sy ste m of C h a p te r I V (or R o bin so n 's), w e can also show th is (w ith
a n o th er choice of R ( x , y) an d R (x, y) in th e stronger co n sisten cy h y
pothesis) from th e p roof of T h eo rem 33 as giv e n , thus. L e t / b e chosen
b y (6 ) w ith T t ((x)2f (x)2f y) as th e R , an d g b e chosen b y (7) for a n y recur-
316 GENERAL R E C U R S IV E F U N C T IO N S CH . XI
sive R su ch th a t (x) iy)R (x, y ). T hen (Ey) T x(x, x, y) = (Ey) W 0{2 ®*3A5 *, y ).
T his reduces th e decision problem for (E y )T x(x, x, y) to th at for
(E y)W 0(x, y). So th e decision problem for th e system is of the highest
degree for 1-quantifier predicates, if the system has th e property th a t
th e converse of (52 ) holds for th e B (x) of E xam ple 1 (as R obinson’s
system does, b y Theorem 53 (c) § 79 ). (This B (x) is of the form 3 y R (x , y)
w here R (x, y) num eral w ise expresses W 0(x, y), b y § 41 (C) and (E).)
In th e case Y are unspecified functions and predicates, one m ay discuss
the reduction of th e decision problem for P (the calculation problem for <p)
to th e respective problem s for Y in the sense of obtaining a procedure
uniform in Y as w ell as in x v . . . , x n; or briefly, of establishing th a t P
is effectively decidable (<p is effectively calculable) uniform ly from Y .
Then Thesis I* and Converse are to be stated reading "uniform ly” in
h yp oth esis and conclusion.
Note i. Proof of (a) bottom p. 292 should be amplified as follows. The application
of Corollary Theorem I* leads to (I) an expression for the predicate R in terms of
the logical operations of the predicate calculus with number variables, 0, ', -j-, •, =
and the representing functions 4i> • • •> 4z °f the predicates T \ By the method of
proof of Theorem I* and Corollary, each of 4*1, • ••, 4z enters into th at expression
only in parts of the form “ 4(î> • • •, am) = w” . Each such part can be replaced by
“ {Q{ai> • • •» am) & w = 0} V {Q(av ..., am) Sc w = \}” where Q is the predicate re
presented by 4- Thus we obtain (II) an expression for the predicate R in terms of the
logical operations of the predicate calculus with number variables, 0, ', + , •, = and
the predicates T themselves. Now we are ready for the final step (p. 292). —
Actually, an expression (I) exists (which is what ‘ R is arithm etical in T” means for
predicates T , under our definition p. 239 with bottom p. 291) if and only if an ex
pression (II) exists. W hy?
Ch a p t e r XII
P A R T IA L R E C U R SIV E FU N C T IO N S
§ 62 . C h u rch ’s th e s is . One of the m ain objectives of th is and the
n ext chapter is to present the evidence for Church's thesis (Thesis I
§ 60).
Since our original notion of effective calculability of a function (or
of effective decidability of a predicate) is a som ew hat vague in tu itive
one, the thesis cannot be proved.
The in tu itive notion how ever is real, in that it vouchsafes as effectively
calculable m any particular functions (§ 30 ), and on the other hand
enables us to recognize that our know ledge about m any other functions
is insufficient to place them in the category of effectively calculable
functions.
For an exam ple of the latter, let R (xt y) be an en ectiv ely decidable
predicate, and consider the function eyR(x, y) (Godel 1931) defined clas
sically thus,
\ the least y such th at R(x, y), if (E y)R (x, y),
zyR(x, y) = j
[ 0, otherw ise.
This definition does not (of itself) provide a calculation procedure. G iven
x , w e can search through the propositions R (x, 0 ), R(x, 1), R (x, 2 ), . . .
in succession, looking for one th at is true, as far as w e p lea se; i.e. w e can
in principle com plete the exam ination of the first n of them , for an y finite
n. If the given x is such th at (E y)R (xt y), b y persisting long enough w e
shall even tu ally encounter a first y for w hich R (xf y) is true, w hich y is
the value of th e function syR(x, y). B u t if x is such that (Ey)R (x, y),
we shall never learn this b y persisting in the search, w hich w ill rem ain
forever uncom pleted. The com pletion of the exam ination of all Kq
propositions, w hich the classical definition envisages, is im possible for
a hum an com puter.
For som e choices of R (x ,y ), the function eyR (x,y) m ay nevertheless
be effectively calculable, not “im m ediately" on the basis of its definition,
but because of the existence of som e other procedure for determ ining the
value, w hich unlike the one suggested b y the definition itself is effective.
318 PARTIAL RECURSIVE FUNCTIONS CH. XII
For example, w hen R(x,y)~ (x)0{y)0+(x)1(y)1=(x)2t there is k n o w n

to b e such a procedure (cf. E x a m p le 2 § 30).
The fu n ctio n eyR(x, y) is e ffe c tiv e ly calcu lab le, if an d o n ly if th e
p red icate (Ey)R(x, y) is e ffe c tiv e ly d ecidable. F o r if (Ey)R{x, y) is effec
t iv e l y d ecid ab le, th en g iv e n #, to ca lcu late eyR(x, y), w e can first d ecid e
w h e th er or n o t (Ey)R(x, y) is true, an d acco rd in g to th e answ er eith e r
search for th e least y su ch th a t R(x, y) or ta k e 0 as th e va lu e. C o n ve rsely,
i f zyR{x, y) is e ffe c tiv e ly ca lcu lab le, th en g iv e n x> w e ca n decid e w h e th e r
or n o t (.Ey)R(x, y) is true, b y first ca lcu la tin g eyR(x, y), an d th en ascer
ta in in g w h eth er or not R(x, zyR(x, y)) is true.
T h e in tu itio n ist finds no ju stifica tio n for th e b elief th a t w e can a lw a y s
tell, for a g iv e n p red icate P(y), (Ey)P(y). T h is is h is
w h eth er or n o t
A or not A w ith
gro u n d for n o t a c c e p tin g th e la w of th e ex clu d ed m id d le
his m ean in g o f “ o r” (§ 13). H is argu m en t, ap p lied to R(x, y) as th e P(y),
is an argu m en t th a t w e h a v e no basis to suppose th a t, for a n y R , th e pred
ica te (Ey)R(x, y) is e ffe c tiv e ly d ecidable.
C h u rch 's thesis, b y su p p ly in g a precise d elim itatio n of ‘all e ffe c tiv e ly
ca lcu la b le fu n ctio n s', m ake s it possible to prove, for certa in p red icate s
R(x, y), e.g. Tx(x, x, y) (Th eorem X I I § 60), th a t there is no uniform m e th o d
o f so lvin g th e problem w h e th er or n o t (Ey)R(x, y). T h e r e b y B ro u w er's
a rgu m en t, t h a t H ilb e rt's b elief in th e s o lv a b ility of e v e ry m a th e m a tic a l
p roblem is u n p ro ven , is n ow stren gth en ed to an a c tu a l disproof, w h e n
s o lv a b ility is ta k e n to m ean uniform s o lv a b ility an d C h u rch 's thesis is
accep te d . T h e relation sh ip of C h u rch 's thesis to in tu itio n ism w ill b e
discussed fu rth er b elo w (§ 82).
T h e in tu itio n ist does n o t regard th e defin ition g iv e n a b o v e for syR(x, y),
la c k in g a proof of e ffe c tiv e c a lc u la b ility , as p ro p erly d efin in g a fu n ctio n .
B u t our discussion o f syR(x, y) ca n refer in tu itio n istic a lly to th e p red ic a te
{ R(x, w) & (z)z<wR(x> z)} V {(.Ey)R{x, y) & w= 0}. C la ssica lly , th is pred
ica te, ca ll it < tP{xi w)i,i is th e represen tin g p red icate of zyR{x,y). B u t
in tu itio n is tic a lly , w e m a y n o t be able to p ro ve th a t (x)(E\w)P(x, w),
i.e. t h a t P(x, w) is a represen tin g p red icate of a fu n ctio n (cf. § 4 1 ;
(x)[(Ew)P(x, w) == (Elw)P(x, w)]f cf. * 1 7 4 b , * 1 7 1 ) .
W h ile w e ca n n o t p ro ve C h u rch 's thesis, since its role is to d elim it
p recisely an h ith erto v a g u e ly co n ceiv ed to ta lity , w e require ev id e n ce
t h a t it ca n n o t co n flict w ith th e in tu itiv e n otion w h ich it is su pposed to
co m p le te ; i.e. w e require evid e n ce th a t e v e r y p articu lar fu n ctio n w h ich
our in tu itiv e n otion w o u ld a u th e n tic a te as e ffe c tiv e ly ca lcu lab le is gen eral
recursive. T h e thesis m a y b e considered a h yp o th esis a b o u t th e in tu itiv e
n o tio n of e ffe c tiv e c a lc u la b ility , or a m a th e m a tica l d efinition of e ffe c tiv e
§62 church's thesis 319
c a lc u la b ility ; in th e la tte r case, th e ev id e n ce is required to g iv e th e
th e o r y b a se d on th e d efin itio n th e in ten d ed sign ifican ce.
The con verse o f C h u rch 's thesis, i.e. t h a t e v e r y gen eral recu rsive
fu n ctio n 9 is e ffe c tiv e ly ca lcu lab le, w e ta k e to b e a lre a d y con firm ed b y
th e in tu itiv e n o tio n (cf. § 60). W e use here th e d efin itio n o f ‘E d efines
9 re cu rs iv e ly ' w h ich sa y s th a t, g iv e n x v . . . , x n, a d ed u ctio n from E of
an eq u a tio n f(xx, . .. ,x n)= x exp ressin g th a t th e v a lu e y(x v . . . , x n)
is x a lw a y s ex ists (or if w e b ase th e co m p u ta tio n procedure on T h eo rem
I X (30), w e use (29); or if w e b a se it on C o ro llary T h eo rem I X , w e use ( 1 )).
In co n clu d in g t h a t w e h a v e an e ffe c tiv e c o m p u ta tio n procedure, th e
e x iste n tia l q u a n tifie r w h ich ap pears in th e d efin itio n of ‘E defines 9 recur
s iv e ly ' (or in (29), or in (1)) m u st b e u n d erstoo d c o n s tr u c tiv e ly (§ 13);
an d likew ise th e e x iste n tia l q u a n tifie r in th e d efin itio n of '9 is gen eral
recu rsive', w h ich sta te s t h a t th ere e x ists an E d efin in g 9 re c u rsively (or
in T h eo rem I X , t h a t a G o d e l n u m b er e ca n b e fo u n d ; or in C o ro lla ry
T h eo rem I X t h a t a fin ite sequence o f a p p licatio n s of (I) — (V I) ca n b e
found).
In other words, we should not claim that a function is effectively
calculable oq the ground that it has been shown to be general recursive,
unless the demonstration that it is general recursive is effective (cf.
Church 1936 Footnote 10).
W e n o w sum m arize th e e v id e n ce for C h u rch 's thesis (and T h esis I * ,
end § 61) u nder three m ain h ead in gs (A) — (C), an d one o th er (D) w h ich
m ig h t b e in clu d ed und er (A). Som e of th is evid e n ce w ill b e g iv e n in m ore
d eta il in la te r sections.
(A) H eu ristic evidence.

( A l) E very p a rticu la r e ffe c tiv e ly c a lcu la b le fu n ctio n , an d every
operatio n for d efin in g a fu n ctio n e ffe c tiv e ly from o th er fu n ctio n s, for
w h ich th e q u estio n h as been in v e stig a te d , h as p ro v ed to b e gen eral re
cursive. A g re at v a r ie t y of e ffe c tiv e ly ca lcu lab le fu n ction s, of classes of
e ffe c tiv e ly ca lcu lab le fu n ctio n s, a n d o f operatio ns for d efin in g fu n ctio n s
e ffe c tiv e ly from other fu n ction s, selected w ith th e in ten tio n of e x h a u stin g
kn ow n ty p e s , h a v e been in v e stig a te d .
(A 2 ) The m eth o d s for sh o w in g e ffe c tiv e ly ca lcu la b le fu n ctio n s
to be gen eral recursive h a v e been d eve lo p ed to a degree w h ich v ir tu a lly
exclu d es d o u b t th a t one co u ld describe an e ffe c tiv e process for d e
term in in g th e va lu e s o f a fu n ctio n w h ich co u ld n o t b e tran sform ed b y
these m eth o d s in to a gen eral recursive d efin ition of th e fun ction .
(A3) T h e e x p lo ra tio n o f v a rio u s m eth o d s w h ich m ig h t b e e x p e c te d
to le a d to a fu n ctio n ou tsid e th e class of th e gen eral recursive fu n ctio n s
has in every case shown either that the method does not actually lead
outside or that the new function obtained cannot be considered as ef
fectively defined, i.e. its definition provides no effective process of cal
culation. In particular, the latter is the case for the Cantor diagonal
method. (An illustration of the former will be given in Example 1 § 65.)
(B) Equivalence of diverse formulations.
(Bl) Several other characterizations of a class of effectively cal
culable functions with the same heuristic property ((A)) exist. These
have turned out to be equivalent to general recursiveness, i.e. the classes
of functions which they describe are coextensive.
In fact three notions arose independently and almost simultaneously,
namely general recursiveness, k-definability (successive steps toward
which were taken by Church 1 9 3 3 and Kleene 1 9 3 5 ; cf. Church 1 9 4 1 )
and computability (Turing 1 9 3 6 -7 , Post 1 9 3 6 ). The equivalence (i.e. co
extensiveness) of the X-definable functions with the general recursive
functions was proved by Church 1 9 3 6 and Kleene 1 9 3 6 a (also cf. the ref
erence to work of Rosser in Church 1 9 3 6 Footnote 16). The equivalence
of the computable to the X-definable functions (and hence to the general
recursive functions) was proved by Turing 1 9 3 7 .
The notion of a function reckonable (§ 59) in a certain formal system
S1 described (very briefly) in Godel 1 9 3 6 is a fourth equivalent of general
recursiveness, under the hypothesis that Sx is simply consistent (as
Rosser remarked in a review 1 9 3 6 a).
Still another approach is given by Post (1 9 4 3 , 1 9 4 6 ) in terms of what
he calls canonical and normal systems. What this gives directly, as
it is presented, is an equivalent of recursive enumerability, but then as
in Example 6 § 60 we obtain an equivalent of recursiveness.
The fact that several notions which differ widely lead to the same class
of functions is a strong indication that this class is fundamental.
(B2 ) Of less weight, but deserving mention, is the circumstance
that several formulations of the main notions are equivalent; i.e. the
notions possess a sort of “stability”.
Thus, for general recursiveness, the formalism may be chosen in
several ways (§55). Also one may give a formulation (f.i-recursiveness)
not based on any formalism but using instead Schemata (I) — (VI)
(Theorem III § 57 and Corollary Theorem IX § 58), or one using Schemata
(III), (IV) and (VI) with x+y, x-y and 8* (= 1 if x = y, = 0 if x ^y ) as
initial functions (Kleene 1 9 3 6 b). (Also cf. Julia Robinson 1 9 5 0 .)
The notion of X-definability has the variants k-K-definability (studied
by Rosser, cf. Kleene 1 9 3 6 a Footnote 12) and k-8-definability (Church
§62 c h u r c h ' s t h e s is 321
1935). A lso there is a parallel developm ent, started b y Schonfinkel
1924 and Curry (1 9 2 9 ,1 9 3 0 , 1 9 3 2 ) and continued by R osser (1 9 3 5 , 1942a*)
(also cf. Curry 1948-9), w hich leads to a notion that w e m ay call com
binatory definability, proved equivalent to X-definability b y Rosser.
The details of the definition of com p u tab ility can also be varied,
as we shall see later (Chapter X III).
The system S x of Godel 1936 is the first system in a hierarchy of system s
Si {i = 1, 2 , 3 , . . . ) using successively higher typ es of variables (cf. § 12).
Godel rem arks, “ It can be show n m oreover, th a t a function w hich is
reckonable in one of the system s Sit or even in a system of transfinite
order, is reckonable already in Sv so th a t th e concept ‘reckonable* is in
a certain sense ‘absolute*, w hile alm ost all hitherto know n m etam athe-
m atical concepts (e.g. provable, definable, etc.) depend very essentially
on the system w hich is taken as the b asis.” E x a ctly the sam e functions
are reckonable in our num ber-theoretic system of Chapter IV or R ob
inson's described in Lem m a 18b § 49 (by Theorem 32 § 59 and the
equivalence of reckonability in G odel’s S 1 to general recursiveness). The
equivalence of reckonability in these system s is under the hypothesis of
sim ple consistency for G odel’s system s and our system of Chapter IV.
(The sim ple consistency of R obinson's system w ill be proved as Theorem
53 (a) § 79 .) Tw o other system s (Z°) and (Z00) having the sam e class of
reckonable form ulas are given in H ilbert-B ernays 1939 Supplem ent II;
these are form alizations of ^-recursiveness ((Z00) utilizing also the norm al
form ). (M ostowski 1947 bases his version of Theorem V § 57 on the notion
of resolvability of a predicate P in a system S (§ 59 ), w ith S subject
only to som e quite general conditions. As w e shall see below in connection
w ith (D l) and T hesis II, only a general recursive function can be reckon
able in an S w hich is a form al system w ith effective ru les; but M ostowski
considers also non-constructive generalizations of form al system s in
our sense.)
(C) Turing's concept of a com puting m achine.
Turing's com putable functions (1936-7) are those w hich can be com puted
b y a m achine of a kind w hich is designed, according to his analysis,
to reproduce all the sorts of operations w hich a hum an com puter could
perform, working according to preassigned instructions. T uring’s notion
is thus the result of a direct attem p t to form ulate m athem atically the
notion of effective calculability, w hile the other notions arose differently
and were afterwards identified w ith effective calculability. Turing's
form ulation hence con stitutes an independent statem ent of Church’s
thesis (in equivalent term s). P ost 1936 gave a sim ilar form ulation.
The work referred to under (A) (especially (Al)) was not all carried
out originally for general recursiveness or the special notions of
recursiveness subsumed under general recursiveness (§ 55), but much of
it was done for X-definability (in Kleene 1 9 3 5 ) or computability (in Turing
i 9 3 6 ~7 ). But by (B) the heuristic and other evidence accumulated in study
ing the various notions all applies to any one. The accumulation of
methods shown to be general recursive under (Al) contributes to (A2).
The case under (A2 ) will be presented in this chapter in connection
with the theory of partial recursive functions (cf. § 6 6 ). We shall take
up computability in the next chapter, proving the equivalence of com
putability to general recursiveness in §§ 68 , 69 (cf. (Bl)) and incidentally
the equivalence of some differing formulations of computability (cf. (B2)),
and giving the evidence under (C) in § 70.
(D) Symbolic logics and symbolic algorithms.
Church 1 9 3 6 gave the following arguments (in substance), as showing
“that no more general definition of effective calculability than that
proposed above can be obtained by either of two methods which naturally
suggest themselves” (p. 358).
(Dl) Suppose that we are dealing with a function <p(.r) and a formal
system such that the following is true. The set of the axioms is finite
or (if infinite) effectively enumerable, and likewise the set of the rules of
inference; and each rule of inference is an effectively performable oper
ation. We can effectively recognize a formula P(x, w ) which attributes
a number w as value to 9 for a given argument x, and effectively read
from it this number. The formulas P(x, w ) attributing the correct
and only the correct values to 9 are provable in the system; i.e. 9 is
‘reckonable’ § 59 (except that here we are not insisting that P(x, w )
come from some P(x, w) by substituting x, w for x, w). If the interpre
tation is allowed that the effectiveness of the metamathematical functions
and predicate just mentioned implies that the number-theoretic functions
and predicate corresponding to them under a suitable Godel numbering
are general recursive, then 9 is general recursive. For by reasoning
as in the proof of Theorem IX § 58 or (c) § 59, for some general recursive
^ and R, (x)(Ey)R(x, y) and cp(x) = §{\xyR(x, y)); whereupon Theorem I I I
§ 57 applies.
(D2) Consider a symbolic algorithm for the calculation of the values
of a function <p(x), which shall consist in a method by which, given any
x, a finite sequence Exo, Exu . . EXVx of expressions (in some notation)
can be obtained, in the following fashion. Given x, the first expression
Ex0 can be effectively found. Given # and the expressions Exi for i <; 7,
§63 PARTIAL RECURSIVE FUNCTIONS 323
it can be e ffe c tiv e ly recogn ized w h e th er th e algo rith m h as term in ated
(i.e. w h e th er j = r x), an d if so, th e v a lu e < p(x) can be e ffe c tiv e ly fo u n d ;
w hile in th e co n tr a ry case, th e n e x t expression E a;>;.+1 can be e ffe c tiv e ly
found. A g a in , if th e e ffe c tiv e fu n ctio n s an d p red icate d escribed becom e
gen eral recursive und er som e G o d el num berin g, th en <p is gen eral
recursive. F o r w e can reason as in ( D l) , regardin g (x, E xo) n ow as an alogo u s
to an ax io m , an d th e operatio n of passin g from (x, E x0, . . . , E x/) t o
(x, E xo, . . . , E^., E X' j +l) as an alogo u s to a rule of inference.
In brief, ( Dl ) an d (D2) sh ow th a t if th e in d iv id u a l operations or rules
of a form al sy ste m or sy m b o lic algo rith m used to define a fu n ctio n are
gen eral recursive, th en th e w hole is general recursive. So w e co u ld in
clud e ( Dl ) an d (D2) as p articu lar exam p les of operations or m eth od s of
definition under ( Al ) .
N o te t h a t ( Dl ) an d (D 2 ) refer to form al sy ste m s an d sy m b o lic a l
go rith m s h a v in g a special kin d of structure, exem p lified b y p a rticu la r
form al sy ste m s an d algo rith m s w e know . W e h a v e elsew here (§§ 30, 60, 61)
used “ a lg o rith m ’ ’ m ore b ro a d ly to m ean a n y ca lcu latio n (or decision)
p ro c e d u re ; an d w e gen eralized th e n otion of a form al sy ste m likew ise in
con n ection w ith T h esis I I an d T h eorem X I I I (§ 60). T h ere is o f course
no circu la rity in ad d in g th e evid en ce p ro v id e d b y algo rith m s an d form al
syste m s of th e special sorts to th e case for C h u rch ’s thesis, an d afterw ard s
a p p ly in g th e thesis (as in §§ 60, 61) to th e discussion of algo rith m s an d
form al sy ste m s in th e broader sense.
If w e consider o n ly sy ste m s sa tisfy in g T h esis I I (for n+l variables),
th e fu n ction s (of n variables) w h ich are reckon able in vario u s form al
syste m s (i.e. each one in som e system ) are all gen eral recursive, an d
hence all are reckon able in one sy ste m (e.g. a n y one of th e sy ste m s
m en tion ed under (B 2 )).
§ 63. Partial recursive functions. A s a t th e b egin n in g of § 62, let

R (x, y) be an e ffe c tiv e ly d ecid able pred icate. Consider the procedure w h ich
consists, for a g iv e n x, in d ecid in g as to the tru th or fa ls ity of each of th e
propositions R (x, 0), R ( x ,l) , R (x, 2), ..., su ccessively, u n til one is
found to b e true, an d ta k in g th e second argu m en t y of th a t one R (x, y).
T h is procedure leads to a n a tu ral num ber y in a fin ite num ber of steps,
if (E y)R (x, y) an d o n ly then. Therefore it can be considered as an a l
gorith m for ca lcu la tin g a m a th e m a tica l fu n ctio n of x d efined o ver th e
subset x(E y)R (x, y) of th e n a tu ral num bers. T h e fu n ctio n c a lcu la te d is
T h e least y such th a t R (x, y ) ’ or in sym b o ls iyR (x , y)'.
I t m a y be im possible to e x te n d th e d efinition of th is fu n ction [iyR(x, y)
t o all n a tu ra l num bers, in such a w a y th a t there w ill b e an algo rith m for

c a lc u la tin g th e resu ltin g co m p le te ly d efined n u m b er-th eo retic fu n ction .
W e n o ted in § 62 th a t th e p a rticu la r exten sio n e yR (x, y) is e ffe c tiv e ly
ca lcu lab le, if an d o n ly if th e p red icate (. E y)R (x , y) is e ffe c tiv e ly d ecid ab le;
an d th e m e th o d show s th a t no exten sio n of \iyR (x, y) to all n a tu ra l n u m
bers is e ffe c tiv e ly ca lcu lab le, unless (E y)R (x, y) is e ffe c tiv e ly d ecidable.
T h is ca n b e s ta te d in term s of th e th e o ry of gen eral recursive function s.
C h u rch 1936 called a fu n ctio n <p(%, . . . , x n) d efined o ve r a subset of th e
^ -tu p le s of n a tu ra l num bers potentially recursive, if there exists a gen eral
recu rsive fu n ctio n <p'(xv . . . , x n) such th a t <p'(%, . . . , x n) = <p(%, . . . , x n)
for each w -tuple xv . . . , #n for w h ich <p(%, . t x n) is defined. N o w
if R (x, y) is gen eral recursive, th en \iyR {x, y) is p o te n tia lly recursive if
an d o n ly if (E y)R (x, y) is gen eral recursive. F o r
(59) eyR (x, y) = \lw [R{ x , w ) V {(E y)R {x, y) & ^ = 0 } ] ;

an d therefore b y # D § 45 an d T h eo rem I I I § 57, if (E y)R (x, y) is gen er
al recursive, th en zyR (x, y) is a gen eral recursive exten sio n <p'(x) of
{jiyR (x, y). C o n v e rse ly , if fxyR (x, y) is p o te n tia lly recursive w ith <p'(x) as a
gen eral recursive exten sio n , th en (E y)R (x , y) == R (x, an d therefore
(E y)R (x, y) is gen eral recursive.
E xam ple 1 . H en ce b y T h eo rem V (15) § 5 7 , \xyTx{ x ,x ,y ) is n o t
p o te n tia lly recursive (K leen e 19 3 8 , 19 4 3 *), an d eyT ^ x , x, y) is not gen eral
recursive (K leen e 1936 ).
A n a lgo rith m for c a lc u la tin g a fu n ctio n 9 m a y , for a g iv e n x, fa il to

lead to a n u m b er as v a lu e of y(x) either b y not te rm in a tin g (so t h a t no
m a tte r h o w m a n y step s h a v e a lre a d y been perform ed, th e rules of th e
algo rith m ca ll for a n e x t step), or b y term in a tin g b u t w ith o u t g iv in g a
n u m ber as v a lu e . W e can m o d ify a n y g iv e n algo rith m so th a t w hen ever,
for a g iv e n x, th e g iv e n algo rith m term in ates w ith o u t p ro d u cin g a n u m ber
as v a lu e , th e n ew algo rith m g iv e s 0 as va lu e. T h e n ew algo rith m c a lcu la te s
an exten sio n 9' of 9 d efin ed e x a c tly w hen th e original (and th e new)
a lgo rith m term in ate.
E x am ple 1 (continued). H en ce any algo rith m w h ich w ill lead to

th e n u m b er \iy T x{x, x, y) for e v e r y # such th a t (E y ^ ^ x , x, y) ca n n o t
term in ate for e v e r y x (using T h esis I § 60).
I f there is an algo rith m for d ecid ing, g iv e n #, w h e th er th e fu n ctio n
<p(x) ca lc u la te d b y a g iv e n algo rith m is d efined or not, th en a new a l
go rith m can b e set u p w h ich ca lcu lates an exten sio n <pf(x) of 9 ^ ) to all
n a tu ral num bers. (Also cf. E x a m p le 5 § 64.)
§63 PA R T IA L R ECU RSIVE FUNCTIONS 325
E xample 1 (continued). Hence there is no algorithm for deciding,
given x, whether p y T ^ x , x , y) is defined or not, as we can also see directly
from its condition of definition (£ y )7 \(# , x , y) (cf. Theorem X I I § 60).
W e ca n also m o d ify a g iv e n algo rith m w ith o u t e x te n d in g th e fu n ctio n
<p{x) so th a t, for a g iv e n x , th e algo rith m w ill a lw a y s fail to te rm in a te
w h e n eve r <p(#J is u n d efin ed (as is a lre a d y th e case for th e a lgo rith m for
\iy R {xt y) d escribed ab o ve). T o do so, w e arran ge th a t, w h en th e g iv e n
algo rith m term in ates w ith o u t g iv in g a n u m ber as v a lu e of <p(x), th e new
algo rith m calls for ad d itio n a l steps w h ich w ill co n tin u e ad in fin itu m . In
discussing algo rith m s in th e rest of th is ch ap ter, w e sh all o fte n t a c it ly
assum e th a t th e algo rith m s are of th is kind.
S u p p o se th a t x(x) is d efined o ve r all n a tu ra l num bers, an d <]>(#) o v e r a
proper su bset of th em c o n ta in in g all th e n um bers ta k e n as v a lu e s b y x(#), a n d
th a t b o th fu n ctio n s are e ffe c tiv e ly ca lcu lab le. T h e n *s c o m p le te ly d e
fin ed an d e ffe c tiv e ly ca lcu lab le. T h e fu n ctio n <{/(#) m ig h t b e ^ y T ^ x ^ .y ) .
T h u s an e ffe c tiv e ly ca lcu lab le fu n ctio n restricted to a proper su b set of th e
n a tu ra l num bers m a y b e u sefu l in c o n stru ctin g an o th er e ffe c tiv e ly c a lc u la
ble fu n ctio n defin ed o ve r all n a tu ra l num bers.
Also there are problems in foundations which call for a function, needed
only on a proper subset of the natural numbers, to be effectively cal
culable. This occurs in the theory of constructive ordinals (Church-
Kleene 1 9 3 6 , Kleene 1 9 3 8 , Church 1 9 3 8 ), and in studies of the intu-
itionistic logic (cf. § 82).
T h ese con sideration s in d ic a te th e d e sira b ility of in c lu d in g p a r tia lly
defined fu n ctio n s und er our tre a tm e n t of e ffe c tiv e c a lc u la b ility . We
sh all a c co rd in g ly e x te n d th e class of th e gen eral recu rsive fu n ctio n s to
ta k e in ce rta in in c o m p le te ly d efin ed fu n ctio n s, ca llin g th e re su ltin g class
of fu n ctio n s th e p a rtia l recu rsive fu n ctio n s'. T h e te ch n ica l a d v a n ta g e s
of th u s e x te n d in g th e class of th e gen eral recu rsive fu n ctio n s, e v e n if our
purpose w ere o n ly to su p p o rt C h u rch 's thesis for th e case o f c o m p le te ly
defin ed fu n ctio n s (Theses I an d I* ) , w ill b eco m e fu lly ap p aren t in §§ 65
an d 6 6 . A t th e en d of th e present section, w e use th e p a rtia l recu rsive
fu n ctio n s in s ta tin g C h u r c h ’s thesis for th e case o f p a r tia lly d efin ed
fu n ctio n s (Theses I t an d I * t) .
T o fix our te rm in o lo gy, le t us n ow c a ll a fu n ctio n from a n y su b set
(proper or im proper) of th e n -tu p le s of th e n a tu ra l num bers to th e n a tu ra l
num bers a partial function. In oth er w ords, a p a rtia l fu n ctio n 9 is a fu n c
tio n w h ich for each n-tu p le x lt . . . , x n of n a tu ra l num bers as argu m e n ts
ta k e s a t m ost one n a tu ra l n u m b er <p(xlt . . . , x n) as v a lu e . F o r an n -tu p le
x v . . ., x n for w h ich 9 has a n a tu ra l n u m b er as v a lu e , w e s a y 9 (or
< ...,
p ( x lt is
x n)) ; for an w-tuple ..., for which has no
d e fin e d x v x n 9
natural number as value, we say (or <p(x1} ..., is (some

9 x n )) u n d e fin e d
times written “a”). The of apartial functionis the set

ra n g e o f d e fin itio n
of the n-tuples .. for which

x v ..., x n is defined. When this < ? {x lr x n)
consists of all w-tuples, we have anordinary number- ( c o m p le te ly d e fin e d )
theoretic function; otherwise an function. When it

in c o m p le te ly d e fin e d
is empty, we have the function.

c o m p le te ly u n d e fin e d
To obtain the definition of 'partial recursive function’ (Kleene ), 1938
we adapt the Herbrand-Godel definition of 'general recursive function’

topartial functions, asfollows.
For the case that 4i> ..., 4* are partial functions (of .. m v , fm l
variables, respectively) we now understand naturally that (cf.

§54) is the set of the equations g,(jq, .. ., where 4i(y1, ..., y u j)= = y y m j)
= y, for all = , ..., and in the case of each for those mrtuples
j 1 l j
y lf . .., for which 4* is defined. Then for a partial function the

y m . 9
definition of E d e fin e s (or 9 4i, ..., 4*given in §54

r e c u r s iv e ly in fr o m )
reads correctly, if (for emphasis) we now replace "if and only if

< p (x v. f., x n) =by "if and only if
x ” is defined and < p (xv . . . , x n)
9 (%, ..., x n) = The second phrasing of the definition there (with

x ” .
acompleteness andaconsistencyproperty) canbeused, if wenowunder

stand E JPfor apartial function inlike sense to that just explained for
9
Finally we say (corresponding to the definition in §55) that

apartial function is 9 i, ..., *>if there is asystem
p a r t ia l r e c u r s iv e in 4 4
Eof equations which defines recursively from4p ..., 4*- 9
In the case of ascheme = F(<Jq, ..., ^) where 9 range over

partial functions (subject to any stated restrictions), we say that F is
p a r tia l , or that is
r e c u r s iv e 9 ,.., ^ if
p a r t ia l r e c u r s iv e u n ifo r m ly in 4
(for fixed .. .,
n , l, m vthereis suchanEindependent of 4i- ..., 4i*
m t)
As before, we omit the word "uniformly" except for emphasis.

W e nowhave, for the case the is not previously known (as at the 9
end of §55): AsystemE of equations defines recursively a partial re

cursive function of variables frompartial functions i>•••» if f°r
n 4
eachw-tuple . .., of natural numbers there is at most one numeral

x v x n
x such that E b f(x1, ..., x„)=x (where f, gv are

as before). Here no completeness property is required. The function
whichis definedrecursivelybyEis thefunction suchthat ..., 9 < p (xv x n)
isdefinedforagivenw-tuple . .., #n,if and@ nlyif thereisanx forthis

x v
x v ..., in which case

x nt .. ., where # is the number for
< p (xv x n) = x
which the x is the numeral.

§63 PARTIAL RECURSIVE FUNCTIONS 32 7
T h e p a rtia l recursive fu n ctio n s in clu d e th e gen eral recursive fu n ctio n s
as those for w h ich th e ran ge of d efin itio n con sists of all th e w -tuples
xv ..., x n of n a tu ra l num bers.
W hen R {xXt • . x n, y ) is a gen eral recu rsive p red icate , th e p a rtia l
fu n ctio n piyR (xx, . . . , x n9y) (defined if an d o n ly if (E y)R (xlt . . x nt y),
in w h ich case its v a lu e is th e least y such t h a t R (xlt . . . , x nt y)) is p a rtia l
recursive. W e h a v e a lre a d y sh ow n th is (w ith o u t th e present term in o lo gy)
as (iv) in th e first p a rt of th e proof of T h eo rem IV § 57. (T h en
\iyR {xX9 . . . , x n, y) is gen eral recursive e x a c tly w h en (l b) § 5 7 holds.)
E xam ple 1 (concluded). \LyTx{x, x, y) is partial recursive.
For p a rtia l
Xm(xv • • •» *«)) to
Xm(xi> • • •} arexn)
fu n ction s Xi> • • •» X«,
b e d efined w h en an d o n ly w h en
we ta k e
all d efin ed an d th eir va lu e s c o n s titu te an w -tu p le

• • •> xn)>
+(Xi(*i* • • • * * » ) , - • - *
• • •>
for w h ich
th is th e of • • •>xn)>
is defined, e x c e p t w h en w e h a v e otherw ise specified. W e ca ll
weak sense Xiixv • • •> Xm{xv • • •> x n)).
ve n tio n sh all a p p ly likew ise to d efin ition s b y su b stitu tio n n o t in th is
T h is co n
stan d ard form (cf. § 44). T h e a m b ig u ity w h ich th e co n v e n tio n rem o ves
arises w h en ^ is a co n sta n t fu n ctio n , or b ecom es such in one v a ria b le
for som e su b stitu tio n for th e o th er va riab les. F o r exam p le , sh all 0 • x ( x )
h a v e th e v a lu e 0 or b e und efined, for an x w h ich m akes x ( x ) u n d efin ed ?
A cc o rd in g to our co n ven tio n , it sh all b e und efined.
We also e m p lo y partial predicates w ith th e sam e co n ven tio n . For
exam p le, b y s u b stitu tin g p a rtia l fu n ctio n s if(xl9 . . . , x n) an d x (% , . . . , x n)
in to th e c o m p le te ly defined p red ic a te y x—y 2t w e o b ta in a p a rtia l p red
ica te ip(xv . . x n )= x (x i> • • • * * » ) • T h is p red icate , for g iv e n x v . . . 9x n9
is d efin ed if an d o n ly if an d x are b o th d efined, in w h ich case it ta k e s
a true proposition as v a lu e if an d x h a v e th e sam e v a lu e , a n d a false
proposition as v a lu e if ^ an d x h a v e d ifferen t values.
Similarly, by substituting partial predicates Q(xv . . . , x n) and
R (xlt . . . , x n) into the truth-value function Y x = Y 2 ('equivalence', §45),
we obtain a partial predicate Q(xx, . . . , x n) = R (x v . . . , x n), defined if
and only if Q(xv . . . , x n) and R (x Xt . . . , x n) are both defined, in which
case it asserts the equivalence of those two propositions, being true or
false according as the two are equivalent ornot.
W e now introduce
particular x v
..., ( t ^ { x Xf
x n, that if either of ty{xx,

. .
•••>**)” to express, for
x n) ~
and x ( x i>
x (x i,
. . . , x n) • • •> xn)
is defined, so is the other and the values are the same (and hence if
either of ..., x n) and
^ ( x x,
The difference in the meaning of (i)

x (x v xn) xn)
•••> is undefined, so is the other).
= x ( x v •••> and• • •> xnY’
328 PARTIAL RECURSIVE FUNCTIONS Ctt. XII
(ii) “ ty{xv %(xlf ..., x n)” com es w h en one of ty(xl9 an d

x (% , . . . , x n) is undefined. T h e n (i) is undefined, w hile (ii) is tru e or false
acco rd in g as th e oth er is or is n o t undefined. W e d istin gu ish = an d as
weak an d complete equality , re sp e ctiv e ly . O ur use of ~ co n stitu te s an
e x ce p tio n to th e co n v e n tio n sta te d ab o ve.
S im ila rly “ Q(xlt . . x n) ^ R (xv . . . , * tt)” sh all express, for p a rticu la r
xv ..., x ni th a t if either of Q(xv . . x n) an d R (xlt . . x n) is defined*
so is th e o th er an d th e tw o v a lu e s are eq u iv a le n t propositions (and hence
if eith er is un defin ed, so is th e other). W e d istin gu ish == an d ^ as weak
an d complete equivalence.
N o w (*,) . . . (xn)[i>(x i, ~ x(*i. • • • • *»)]> or ty{xv . . . , x n) ~
x(xv . . . , x n) w hen x v . . . , x n have th e g e n e ra lity in terpretatio n ,
expresses th a t an d x are eq u al as function s, i.e. t h e y h a v e th e sam e
range of definition, an d o ve r th is com m on range t h e y agree in v a lu e .
S im ila rly (*x) . . . (xn)[Q (xv = R (xv x n)}, or Q(xv . . . ,x „ )
^ . . x n) u nder th e g e n e ra lity in terp retatio n o f x v . . x n, e x
i? (% ,
presses th a t Q an d R are eq u a l as predicates.
W e s a y t h a t a p a rtia l fu n ctio n <p(%, . . . , x n) is th e representing function of
a p red icate P (x v . . . , x n), if <p(xv . . . ,x n) ta k e s o n ly O a n d 1 as va lu es, an d
P (x v ...,#») ^ <p(xv . . tfw) = 0 ;
or in oth er w ords, if acco rd in g as th e v a lu e of P {x v ..., x n) is t, f or u,

t h a t of y{x lt . . x n) is 0 , 1 or u.
W e s a y th a t a p a rtia l fu n ctio n 9 or p a rtia l p red icate P is partial
recursive in p a rtia l p red icates an d fu n ctio n s T , if th e corresponding
sta te m e n t h olds replacin g th e p red icates am o n g P , T b y th eir represen tin g
function s.
T h e role o f th e tw o e q u a lity pred icates = an d w ill b e d ifferen t.
T h e w e a k e q u a lity = w ill serve as an op eratio n in b u ild in g p a rtia l
recu rsive predicates. W e sh all see in a m o m en t (Theorem X V I I # # 14 ,
O) th a t ty(xv ..., %n) = x ( x 1> . . . , # w) is p a rtia l recu rsive in and x-
T h e co m p le te e q u a lity w ill b e used in expressin g our th e o r y a b o u t
p a rtia l recursive function s. T h e p red icate ty{xlt ..., x n) ~ x (% , . . . , x n)
is n o t a lw a y s p a rtia l recursive w h en ^ an d x are p a rtia l recursive (cf.
E x a m p le 7 § 64).
Sim ilar rem arks a p p ly to th e tw o eq u ivalen ces == an d ^ (cf. T h eo rem
X V II # D ^ a n d E x a m p le 8 § 64).
The p articu lar fu n ctio n s an d pred icates of # # 1 — 21 (§§44, 45),
b ein g p rim itiv e recursive, are gen eral recursive (Theorem I I § 55) an d
therefore p a rtia l recursive. R e w ritin g S c h e m a ta (IV ) an d (V) (§ 43) w ith
in p lace of “ = ” , an d u sin g our co n ven tio n to read th e expressions
on th e righ t in th e w e a k sense, th e n o tio n '9 is p rim itiv e recursive in Y'
ta k e s on a m ean in g for th e case t h a t th e assum ed fu n ctio n s an d p red icate s
are p a r tia lly defined. F o r ex am p le , in a p rim itiv e recursion (V a), for a
g iv e n y , 9 (y') is defined, if an d o n ly if 9 (y) is d efin ed an d %(y, z) is d efined
w hen z is th e v a lu e of 9 (y) (hence b y in d u ctio n , o n ly if all of 9 (0), 9 ( 1 ) , . . . ,
9 (y) are defined). The notion e x ten d s from fu n ction s to p red icate s
v ia represen tin g function s. N o w th e form er proofs of # # A — G (§§ 44—
47) a p p ly , p ro v id e d w e u n d ersta n d th e resu ltin g fu n ctio n s an d p red icates
in th e su itab le w e a k senses, e.g. S ^(y) is d efined w h en an d o n ly w h en all
y<z
o f ^( 0 ), . . . , ^(2 — 1) are defined, QV R w hen an d o n ly w h en b o th
Q an d R are defined, (E y)v< zR (y) w hen an d o n ly w hen R ( 0), ..., R (z — 1)
are, etc. In each case th e ap p ro p riate w e a k sense is e a s ily inferred from
th e proof. B u t also, w ith these senses for (IV ) an d (V), th e pro o f o f
T h eorem I I (§§ 54, 55) carries over. H en ce:
T heorem X V II. (a) A n y function 9 definable from partial functions

Y by a succession of applications of partial recursive schemes is partial
recursive in Y. Schem ata (I) — (V) are partial recursive, (b) The functions
and predicates of — 21 are partial recursive ; and — G hold
reading “partial recursive” for “prim itive recursive” and using the weak
senses of the resulting functions and predicates (call th em th en # # A * — G * ).
W e used th e fi-operator a b o v e to form [xyl?(%, . . x n, y) as a p a rtia l
fu n ctio n from a c o m p le te ly d efin ed p red icate R {xv . . . , x n, y). N o w
w hen R (x v . . . , x n, y) is a n y p a rtia l pred icate, w e ta k e \iyR (xlf . . . , x nt y)
to b e d efin ed w h en an d o n ly w h en there is a y such th a t R (xlf . . . , x n, y)
is true an d R (x v . . . , x n, 0), . . . , R (x v . . . , x ni y — 1 ) are all d efined , in
w h ich case its v a lu e is th e least such y. F o r ex am p le , if R(0) ~ R ( l ) = f,
R {2) ^ t, th en jxyR{y) ~ 2 , b u t if R { 0) ^ f, R { 1 ) ^ u, R { 2 ) ^ t, th e n
fiyR {y) ~ u.
N o w w e ca n con sider (V I) § 5 7 read in g \iy“ for “ = \iy” as a sch em a
for a n y g iv e n p a rtia l fu n ctio n x(xv . . x n, y ) or p a rtia l p red ic a te
R (xlf . . . , x n) y). L e t th e eq u a tio n s E b e set u p as in th e p roof of T h eo rem
I I I . I f yiyR(xlf . . . i %ri,y) is d efined, th en E^, E f- f(*i> • • •> x n) —x
w hen x = \jjyR{x1} . . . , x n, y) a n d for no o th er n u m e fa l x , ju s t as before
((iii) § 57). C o n ve rsely, if E^, E b f(x v . . . , x n)=*x w here x is a num eral,
then th e present co n d itio n s for [iyR (xlt . . x n, y ) to be d efined are m e t
(noting th a t for # B t II x (% , . . . , x n,s ) is d efin ed o n ly if yXx i> • • • >x n> s)
s<y
for s = 0, 1 , . . y — 1 are all defined). T h u s :
T heorem Schema (V I) is partial recursive .

X V I I I ( = T h eo rem I I I f ).
H en ce b y T h eo rem X V I I : E very function 9 definable from partial functions
and predicates Y by applications of (I) — (V I) is partial recursive in Y .
N o w le t us see w h a t h ap pen s to th e proof of th e norm al form theorem
(Theorem I X § 58) w h en w e ta k e <p(%, . . . , x n) to b e a p a rtia l recursive
fu n ctio n . L e t E define 9 recu rsively. N o w (24) an d (26) do n o t necessarily
h old , b u t in stead, o m ittin g th e u n iversal quantifiers, w e o b ta in (from
(26)) (E y)S n(e, x nf y) as th e co n d itio n on xv . . . , x n th a t
<p(%, . . . ,x n) b e defined. A ls o (25) an d hence (27) h o ld w ith in p lace of
an d (28) holds, n o t w ith th e g e n e ra lity in terp retatio n , b u t for each
n -tu p le %, . . xn for w h ich <p{xv . . . , x n) is defined. B u t under our
m ean in g o f th e ^ -operator ap p lied to a co m p le te ly d efin ed p red icate,
U ([iyS n(e, x v .,., x n, y)) is d efin ed e x a c tly w hen (E y)S n(e, x v ..., x nt y).
H e n ce b o th m em bers of (28) h a v e th e sam e ran ge of d efin ition , so (28)
w ith “ = ” replaced b y holds for all xt , ..., x n. T h e proof g iv e n for
T h eo rem I X * (under T h eo rem X ) goes th ro u gh w ith th e sam e m o d ifi
ca tio n s, p ro v id e d th e assum ed fu n ction s <|^, . . . , ^ are c o m p le te ly d efined
(otherw ise e.g. ( £ % ) Ul<v[iV«((y)0iU) « ,) & N u ((y)02, ^ K ) ) ] in 0D f l 2 *,
likew ise $(y), m ig h t b e u n d efin ed in som e case in w h ich for th e proof
w e w o u ld need it to b e d efin ed ). T h u s :
T heorem X IX . (a) ( = T h eo rem Given any partial recursive

IX * ) .
function <p(xv * . . , x n)f a num ber e can be found such that
(60) <p(*i, . . . , x n) ~ U (jiy T n(e, x v . . . . x n, y))
(so that (E y)T „(e, xv . . x n, y) is the condition of definition of the function

<p(%, . and
(61a) (#,) . . . (xn)(y )[T n(e, x v . . . , x n,y ) ~ * U (y) ~ y (x lt . . . , * , ) ] .
(b) ( = T h eo rem I X * f ) . S im ila rly reading tfpartial recursive in Y ,#,

in place of “partial recu rsive *, “ T n**, respectively , where “ Y "
stands for any l ( > 0) com pletely defined functions and predicates of
m v . . . , m t variables , respectively.
E x te n d in g th e d efin itio n s g iv e n in § 58, w e s a y t h a t a n y n u m b er e
su ch t h a t (60) (and h ence (61a)) h olds defines 9 recursively or is a Godel
num ber of 9 ; an d sim ilarly in th e case o f a fu n ctio n 9 p a rtia l recu rsive
in c o m p le te ly defin ed fu n ctio n s Y . T h e n otions e x te n d to a p red icate P
w ith represen tin g fu n ctio n 9 as before.
B y th e proof of T h eo rem X I X , if E is a sy ste m of eq u atio n s d efin in g
9 re cu rs ive ly (recursively from co m p le te ly d efined Y , w ith su ita b le
g iv e n fu n ctio n letters), an d e is th e G o d el n u m ber of E , th en e d efin es 9
re cu rsively (recursively from Y ) .
E xample 2. Let / an d g be G o d el num bers of p a rtia l recursive

fu n ctio n s ^ an d re sp e ctiv e ly . T h e n th e p red icate ${x) ~ x(x) ex‘
pressed b y
(y)[{T A f. x, y) -> {E z) { T ^ , x, z) & U (y) = U (z))}

& { ^ ( g , x, y) - (E zX T M , x, z) & U{y) = U {z))}].
Corollary. E very partial recursive {unction <p {{unction 9 p a rtia l

recursive in completely defined functions and predicates Y ) is definable
{definable from Y ) by applications of Schemata (I) — (V I).
T h e algo rith m g iv e n b y (60) for c a lc u la tin g a p a rtia l recursive fu n ctio n
9 is of th e k in d w h ich does n o t term in ate w hen <p{xv . . . , x n) is und efined.
B y th e theorem , a p a rtia l recursive fu n ctio n <p(xv . . . , x n) h as a range
o f d efinition of th e form x {E y)R {xv . . . , x n, y) w here R is p rim itiv e re
cu rsive, or in other w ords for n = 1 , th e range of d efin ition (if it h as a
m em ber) is recu rsively enu m erable (Theorem X I V § 60).
E xample 3. H en ce b y T h eo rem V (12) § 5 7 , no p a rtia l fu n ctio n w ith

x{y)T i{x, x, y) as its range of d efin itio n is p a rtia l recursive. I t is im
possible to d evise a n y algo rith m w h ich w ill lead to som e n a tu ra l n u m ber
for e x a c tly th ose x ’s such t h a t x, y) an d no others (b y T h esis
V (a) below ). In p articu lar, th e fu n ctio n 9 d efin ed th u s,
9( * ) ~ f ° “ (y)Jl{X X> y)>

1 u otherw ise,
is n o t p a rtia l recursive. T h is fu n ctio n is e ffe c tiv e ly ca lcu lab le for # ’s in

its range of defin ition , an d is p o te n tia lly recursive.
N o w w e s ta te C h u rch 's thesis for th e case of p a rtia l fu n ctio n s, k e ep in g

in m ind E x a m p le s 1 an d 3. W e ca ll a p a rtia l fu n ctio n <p{xv ..., x n)
potentially partial recursive , if there is a p a rtia l recu rsive fu n c tio n
9 ,(#1, . . . , x n) such th a t y'{x lt ..., x n) = y{x lt ..., x n) on th e ran ge
of d efin itio n of 9 . T h e thesis w ill b e sta te d in tw o p arts, acco rd in g as w e
require th a t th e e ffe c tiv e ca lcu la tio n procedure lead to no fu n ctio n v a lu e
o ff th e range of d efin itio n of th e fu n ctio n in question , or m erely disregard
w h a t h ap pen s o ff th is range.
E xample 4. Show th a t: If cp{x) is p o te n tia lly p a rtia l recursive an d

has a range of d efin itio n of th e form x{E y)R {x, y) w here R is gen eral
recursive, th en 9 ^ ) is p a rtia l recursive.
T hesis I+. (a) The function which an y algorithm calculates {the function
being undefined for each n-tuple of argum ents for which the algorithm leads
to no natural num ber as value) is partial recursive . (b) E very partial function
which is effectively calculable {in the sense that there is an algorithm by
which its value can be calculated for every n-tuple belonging to its range
of definition) is potentially partial recursive .
T h e thesis ca n also b e p hrased to a p p ly to pred icates.
For l (;> 0) assum ed fu n ctio n s an d pred icates Y , w e h a v e a corre
sp o n d in g T h esis I * t , in clu d in g th e case o f u n ifo rm ity.
M u ch of our tr e a tm e n t o f fu n ctio n s p a rtia l recursive in l assum ed
fu n ctio n s an d p red icate s Y w ill b e lim ite d (for l > 0 ) to th e case each
o f th e fu n ctio n s an d p red icate s Y is co m p le te ly defined, or is in c o m p le te ly
d efin ed b u t p a rtia l recu rsive or p a rtia l recursive in c o m p le te ly d efin ed
fu n ctio n s. T h e se kin d s of in c o m p le te ly d efin ed fu n ctio n s have been
in tro d u ced to m eet th e requirem en ts of th e th e o r y o f algo rith m s (in
clu d in g redu ctio n s of decision problem s), b eca u se it m a y b e im p o ssib le
to co m p le te th e d efin itio n of such a fu n ctio n an d still h a v e an a lgo rith m
for it. O u tsid e th e th e o r y of algorith m s, a like reason for n o t c o m p le tin g
th e d efin itio n s of in c o m p le te ly d efined n u m b er-th eo retic fu n ctio n s is
n o t re a d ily a p p a ren t (cf. E x a m p le 4 § 64).
E xample 5. I f ^(^) is a p a rtia l fu n ctio n w ith th e range o f d efin itio n

x {E y)R {x , y) (or xR{x)) w here R is a c o m p le te ly d efin ed p red icate, th e n
th ere is a c o m p le te ly d efin ed fu n ctio n tyc{x) prim itive, recu rsive in <[/, R
such th a t (p is p a rtia l recu rsive in <J/\ F o r le t C = x { E y )R (x , y ); le t 0
en u m erate re cu rsiv e ly { 0 } + C ' (cf. E x a m p le 6 § 60); an d le t
4>c(y) = 2®iv) • 3 ^ {y)) w here 7)(0) = 0, r\{x') c^ ^ (# ).
N ow <|/(*) ~ (<i'c( ^ [ ( ^ °( y ) ) o = ^ '] )) i-
E xample 6. W 0{x, \iy[W 0{x, y) V W x{x, y)]) is a p artial, but not

p o te n tia lly , recursive p red icate (cf. §61 ) .
§ 64. The 3-valued logic. In th is section w e sh all in trod u ce n ew

senses o f th e p ro p o sitio n al co n n ectiv es, in w h ich , e.g. Q{x) V R{x) w ill b e
d efin ed in som e cases w h en Q{x) or R{x) is und efined.
I t w ill b e co n ve n ie n t to use tr u th tab les, w ith three “ tr u th v a lu e s "
t (‘tru e ’), f ('false') an d u ('und efined '), in d escribin g th e senses w h ic h
th e co n n e ctiv e s sh all n o w h a ve.
Som e rem arks are a p p ro p riate to ju s t if y our use of tru th ta b le s here
from th e fin ita r y sta n d p o in t, an d to e x p la in h o w w e are led to choose th e
p a rtic u la r ta b le s g iv e n below .
§64 THE 3-VALUED LOGIC 333
W e w ere ju s tifie d in tu itio n is tic a lly in u sin g th e classical 2 -v a lu e d
logic, w h en w e w ere u sin g th e co n n ectiv es in b u ild in g p rim itiv e an d
gen eral recursive pred icates, since th ere is a decision procedure for each
gen eral recursive p r e d ic a te ; i.e. th e la w of th e ex clu d e d m id d le is p ro v e d
in tu itio n is tic a lly to a p p ly to gen eral recursive predicates.
N o w if Q(x) is a p a rtia l recursive pred icate, there is a decision p ro
cedure for Q(x) on its range o f d efin ition , so th e la w o f th e e xclu d ed m id d le
or ex clu d e d “ th ir d ” (sayin g th a t, for each x, Q(x) is either t or f) ap p lies
in tu itio n istic a lly on th e range o f definition. B u t th ere m a y b e no algo rith m
for decid in g, g iv e n x, w h e th er Q(x) is d efined or n o t (e.g. there is none
w h en Q(x) is p y T ^ x , x, y ) = 0 ) . H en ce it is o n ly c la ssica lly an d n o t in
tu itio n is tic a lly th a t w e h a v e a la w of th e ex clu d ed fo u rth (sayin g th a t,
for each x, Q(x) is eith er t, f or u).
T h e th ird “ tru th v a lu e ” u is th u s n o t on a p ar w ith th e oth er tw o t an d f
in our th eo ry. C on sideratio n of its sta tu s w ill sh ow th a t w e are lim ite d
to a sp ecial k in d of tru th tab le.
In assertin g e.g. th a t Q(x) V R(x) is p rim itiv e or p a rtia l recursive (uni
form ly) in Q an dR , w e assert th e existen ce of an algo rith m for o b ta in in g
th e tru th vali\e of Q(x) V R(x) from those of Q(x) an d R (x). A g a in , in th e
p a rtia l case, it w ill h a v e a d ifferen t sta tu s from t an d f.
S u ppose w e are to p ick a tru th ta b le for Q V R , so t h a t Q(x) V i?(#)
w ill b e p a rtia l recursive (uniform ly) in Q an d R . L e t us discuss th is heuris-
tic a lly , for th e m o m en t, id e n tify in g p a rtia l recursiveness w ith e ffe c tiv e
x t w h e th er
d e c id a b ility . W e a sk to b e able to d ecid e b y an algo rith m , g iv e n
Q(x) V R(x) is t or f (if it is defined) from in form ation th a t Q(x) is t or is f
(if it is defined) an d like in fo rm atio n a b o u t I?(a:). In fo rm atio n t h a t Q(x)
is u is n o t u tiliz a b le b y th e a lg o r ith m ; u m eans o n ly th e absen ce o f in
form ation th a t Q(x) is t or is f. I f in case Q(x) is it, th e algo rith m g iv e s e.g.
t as v a lu e toQ(x) V R {x), th e decision to do so (for th e g iv e n x an d 2?(#))
m ust n o t h a v e depen d ed on in fo rm atio n a b o u t Q(x) (since none w as
availab le). In p articu lar, if w ith o u t ch a n gin g th e v a lu e o f R (x), th a t of
Q(x) w ere ch an ge d to t or f, th e sam e decision w o u ld still b e m ade.
We reach th e sam e conclusion, if we ask in stea d m erely th a t
Q(x) V R(x) b e p a rtia l recursive, w h e n eve r Q an d R are p a rtia l recursive.
In general, an algo rith m for Q(x) V R(x) (ab initio) w ill h a v e access to
inform ation a b o u t Q(x) an d R(x) o n ly b y u tiliz in g algo rith m s for Q(x)
an d R(x) w h ich are in corp orated in to it. A decision reach ed in pursuing
th e algo rith m for Q{x) V R(x) th a t e.g. t is th e v a lu e m u st h a v e been based
on in fo rm atio n a b o u t Q(x) an d a b o u t R(x) w h ich h ad been prod u ced a t
som e fin ite sta ges in pu rsu in g th e algorith m s for Q(x) an d R (x). A t a n y
334 P A R T IA L RECU RSIV E FUNCTIONS CH. X II
sta g e in th e algo rith m for Q(x ), w e shall either h a v e fo u n d o u t t h a t

Q(x) is t, or t h a t Q(x) is f, or w e sh all n o t h a v e learned th e tru th v a lu e o f
Q(x). I f Q{x) is a c tu a lly u, w e ca n n o t learn th is b y pursuing th e algo rith m ,
b u t if a t all o n ly in som e oth er w a y , as b y som e m eta th e o re tic reasonin g
a b o u t th e algo rith m . F o r special Q’s w e m ig h t s u b stitu te an oth er al
go rith m w h ich w o u ld also te ll us w h e n Q(x) is u, b u t w e ca n n o t do th is
in general. T h u s, if w h en Q(x) is u, Q(x) V R(x) receives th e v a lu e t, th e
decision m u st (in th e gen eral case) h a v e been m ad e in ignorance a b o u t
Q{x), an d in th e face of th e p o ssib ility th a t, a t som e sta g e in th e p ursuit of
th e algo rith m for Q(x) la te r th a n th e last one exam in ed , Q(x) m ig h t b e
fo u n d to b e t or to b e f.
P roofs in term s of th e th e o r y of p a rtia l recursive fu n ctio n s con firm in g
these h eu ristic argu m en ts (and e x te n d in g them to oth er operators besides
th e propositional con nectives) w ill b e g iv e n a t th e end of th e sectio n
(Theorem X X I an d E x a m p le s 6 — 8).
W e con clude th a t, in order for th e propositional co n n ectiv es to b e
p a rtia l recursive operatio ns (or a t least to produce p a rtia l recursive pred
ica tes w h en ap p lied to p a rtia l recursive predicates), w e m u st choose
ta b le s for th em w h ich are regular , in th e follow in g se n se : A g iv e n colu m n
(row) co n ta in s t in th e u row (colum n), o n ly if th e colu m n (row) con sists
e n tir e ly of f s ; an d likew ise for f.
W h e n w e e x te n d e d #D from p rim itive to p a rtia l recursiveness b y
ta k in g o ve r s u b s ta n tia lly th e form er proofs ( o f T h eorem X V I I § 63),
w e w ere u sin g th e p rop osition al co n n ectiv es in th e weak senses , w h ich are
described b y th e 3 -v a lu e d ta b les (the weak tables ) o b ta in e d from th e
classical 2-v a lu e d ta b les b y su p p ly in g u th ro u gh o u t th e row an d colu m n
h ead ed b y u. T h ese are regular ta b les (trivially).
N ow we in tro d u ce strong senses of th e p ro p o sitio n al co n n ectives,
described b y th e fo llo w in g strong tables.
Q QVR Q&R Q -+ R Q= R
R t f u R t f u R t f u R t f u
Q t f Q t t t t Q t t f u Q t t f u Q t t f U
f t f t f u f f f f f t t t f f i u
u u u t u u u ufu u t ni l u u u u
O f these tab les, o n ly those for V, & an d differ from th e resp ective
w e a k tables. H en cefo rth V, & , an d = ap p lied to p a rtia l pred icates
shall be u n d ersto o d in these stron g senses, except w h en otherw ise
stated .
§64 THE 3-VALUED LOGIC 33 5
E xample 1. N ow (61a) can be restated thus:
(61b) (xj) ... (xn)(y )[T n{e, x v y) U (y)=< ?{x1, .. . , * , ) ] .
S ta te m e n ts of th e form “ I f (?{xlt . . . , x n) is defined, th en . whe r e
th e con clusion is m eaningless w h en <p(xlf . . . , x n) is und efined, ca n b e
u n d erstoo d th en as uses of th e stro n g -> . (Such sta te m e n ts occur a lre a d y
in § 63.)
T h ese stro n g ta b le s are u n iq u e ly d eterm in ed as th e stron gest possible

regular exten sio n s o f th e classical 2 -v a lu e d tab les, i.e. th e y are regular,
an d h a v e a t or an f in each p osition w here a n y regular exten sio n of th e
2 -v a lu e d ta b les can h a v e a t or an f (w hether t or f b e in g u n iq u e ly d e
term ined).
W e g iv e th e fo llo w in g three tab les as ex am p les of irregular tab les.
T h e present stro n g 3 -v a lu e d lo gic (K leene 1938 ) is n o t th e sam e as th e
original 3 -v a lu e d lo gic of L u k a sie w icz (1 9 2 0 ; cf. L ew is an d L a n g fo rd
1932 pp. 2 1 3 ff.), w h ich differs from it b y h a v in g t in stea d of u in R o w
3 C olu m n 3 of th e ta b les for -> an d == (labeled here as - > L an d = L).
Q ^ R Q ^ R Q^R
R t fu R t fu R t f
u
Q t t fu Q t t fu Q t tff
f ttt f ftu f ftf
u tut U uu t u fft
W e fu rth er con clu d e from th e in tro d u c to ry discussion th a t, for th e
definitions of p a rtia l recursive operations, t, f, tt m u st b e su scep tib le of
an oth er m ean in g besides (i) 'tru e ', 'false', 'u n d efin ed ', n a m e ly (ii) 'tru e ',
'false', 'u n k n o w n (or v a lu e im m a terial)'. H ere 'u n k n o w n ' is a c a te g o r y in to
w h ich w e can regard a n y proposition as fallin g, w hose v a lu e w e either do
not k n o w or choose for th e m o m en t to d isre g a rd ; an d it does n o t th en
exclu d e th e oth er tw o possibilities 'tru e' an d 'false'.
E xample 2. Suppose th a t, for a g iv e n x , w e k n o w Q(x) to b e u n d efin ed

an d R(x) to b e false. T h e n u sin g t, f, u as 'tru e', 'false', 'u n d efin ed ' ((i)), w e
can co n clu de b y th e e n tr y o f R o w 3 C olum n 2 in th e ta b le for V th a t
Q(x) V R(x) is undefined.
E xample 3. S u ppose th a t, for a g iv e n x, w e kn o w Q(x) to b e true.

T h en , usin g t, f, u as 'tru e ', 'false', 'u n k n o w n ' ((ii)), w e ca n con clude b y
the e n tr y of R o w 1 C o lu m n 3 th a t Q(x) V R(x) is true. T o d raw th is
conclusion b y usin g th e ta b les w ith M ean in g (i), w e w o u ld need to use th e
classical la w of th e ex clu d ed fourth, th u s: E ith e r R(x) is t, f or u ; an d ill

each of R o w 1 C olu m n 1 , R o w 1 C olum n 2 an d R o w 1 C olu m n 3 a t
appears.
F ro m th is stan d p o in t, th e m ean in g of Q V R is b ro u gh t o u t clearly b y

th e sta te m e n t in w ords: Q V R Q is true (here n o th in g is said
is true, if
about R) or if R is tru e (sim ila rly); false, if Q an d R are b o th fa ls e ; defined,
o n ly in these cases (and hence undefined, otherw ise).
T h e stro n g 3 -v a lu e d lo gic can b e ap p lied to c o m p le te ly d efined pred
ica tes Q(x) an d R (x), from w h ich com posite pred icates are form ed
usin g V, == in th e usual 2-v a lu e d m eanings, thus, (iii) Suppose
th a t there are fix e d algorith m s w h ich decide th e tru th or fa ls ity of Q(x)
an d of R (x), each on a su bset of th e n a tu ral num bers (as occurs e.g. a fter
co m p le tin g th e definitions o f a n y tw o p artial recursive p red icates clas
sically). L e t t, f, u m ean 'd e cid ab le b y th e algorith m s (i.e. b y use of o n ly
such in fo rm atio n a b o u t Q(x) an d i?(#) as can be o b ta in ed b y th e a l
gorithm s) to b e tru e', 'd e cid ab le b y th e algorith m s to b e false', 'u n d e-
cid ab le b y th e algo rith m s w h eth er true or false', (iv) A ssu m e a fix e d sta te
o f kn o w led ge a b o u t Q(x) an d R(x) (as occurs e.g. a fter pursuing algorith m s
for each of th em u p to a g iv e n stage). L e t t, f, u m ean 'kn o w n to b e tru e',
'k n o w n to b e false', 'u n k n o w n w h eth er true or false'.
T h e fo llo w in g three classical eq u ivalen ces
(62) Q & R ^ qVW , (63) Q ^ R ^ Q V R ,
(64) Q ^ R ^ (Q -+ R )& (R -> Q )

hold, as th e reader m a y v e r ify b y co n stru ctin g th e tab les for th e righ t
m em bers, an d co m p arin g th em w ith th e g iv e n tab les for th e left m em bers.
B u t e.g. Q & (R V R) ^ Q (cf. * 5 2 § 27) does n o t (when Q is t an d R is u,
th e le ft m em ber is u an d th e righ t is t).
T h e proofs of (62) — (64) b y use of th e 3 -v a lu e d ta b le s can be con strued
as sh ow in g th a t if either m em ber is d efined th e other is an d has th e sam e
v a lu e (as = asserts), u sin g th e la w o f th e e x clu d e d th ird on th e ranges
o f d efinition. T h is la w w e h a v e g iv e n w hen Q an d R are p a rtia l recursive;
an d th e eq u ivalen ces in th e gen eral case are su b je c t to it in tu itio n istica lly
as an h yp o th esis. Sim ilar rem arks a p p ly to (65).
S tro n g senses are g iv e n to th e bo u n d ed quantifiers, thus. F o r each
2 > -0 , (E y )y<zR {y) ^ R ( 0) V . . . V R (z — 1 ); an d (E y)y< 0R {y ) ^ f. E x -
pressed im words, (E y)y<zR (y) is true, if i?(y) is tru e for som e y < z;
false, if R (y) is false for all y < z; defined, o n ly in these cases.
S im ilarly, for each z > 0 , (y)v<zR (y ) = R ( 0) & . . . & R (z — 1 ); and
(y)v<0 R(y) ~ t.
W e th en h a v e
(65) ( y ) v <zR (y) ~ {Ey)v< M y ) -
S im ilarly, w e g iv e stro n g senses to th e u n b o u n d ed qu an tifiers, b y

(E y)R (y) as th e stron g d isju n ctio n of R (y) for y = 0, 1 , 2 ,
th in k in g of
an d of (y)R (y) as th e stro n g con ju n ction .
In w ords, (£y)i?(y) is true, if R (y) is true for som e n a tu ral n um ber y;
false, if i?(y) is false for e v e r y y ; defined, o n ly in these cases. (Th en
(y)i?(y) ^ (£ y )^ (y ).) U n lik e th e b ou n d ed quantifiers (Theorem X X (b)),
th e u n b o u n d ed qu an tifiers are of course n o t p a rtia l recursive operations
(Theorem V § 57).
E xample 4. Consider th e class of the pred icates expressible in th e

form {E y)R {x, y) w ith R p a rtia l recursive. T h e definitions of all in co m
p le te ly defined p redicates of th is class can be co m p le ted w ith o u t go in g
ou tside th e class (in fa c t, w ith a p rim itive recursive R ). F o r if r is a G o d el
n um ber o f R (x, y), th en (E y)[T 2(r, x, (y)0, (y)x) & t/((y) 1) = 0 ] is co m
p le te ly defined an d = (E y)R (xf y) on th e range of d efin ition of th e la tter.
S im ila rly for th e form (y)R (x, y). (Cf. K le en e 1943 p. 5 7 T h eorem V I.)
T heorem X X . (a) The {strong) predicate Q(xv . . . , x n) is partial
recursive in the predicate Q. The {strong) predicates
Q(xv . . . , %w) V R {xv . . . , x n), Q(xlt . . . , * « ) & R (xv . . . , x n),
Q(xv . . . , x n)- > R {xv . . . , x n) and Q{xlt . . . , * „ ) = R {xv . . . , x n) are
partial recursive in the predicates Q and R.
(b) The {strong) predicates {E y)y< zR {xv . . . , x n,y ) and
{y)v<zR {xli . . . , x n, y) are partial recursive in the predicate R.
(c) (Strong) defin ition b y cases. The function 9 defined by
<P1(^1. • • • >* » ) if Q i{xlt

9(*i> ~
(Pm(-^li • • •> *^n) if Qm(xlt • • •» ^n)»
where Qv ..., Q m are m utually exclusive {under the interpretation that

9 (#1, . . . , # „ ) shall <Pi{xv . . . , x n) if Qi{x1} . . . , x n) is true, disregarding
9 j{x1, . . . , x n) and Qj{xv . . . , x n) for all j ^ i ), is partial recursive in
9l> **•> Q l> • • •> Q m -
P roofs, (a) T o tre a t QVR for ex am p le, let ^>{xl f . . . , x n),%{xv . . . , x n)

an d (p{xlf ..., x n) be th e represen tin g fu n ction s of Q(xlt .. ., x„),
R{xv ..., x n) an d Q{xlf ..., x n) V R{xv ..., x^, re sp ectively. Consider
th e equation s
338 P A R T IA L R E C U R S IV E F U N C T IO N S CH . X II
CT(0) = 0, t (1,1) = 1,
(A)
I 9(*lt ...,*„) = o(x(*i. •••»*«)).
. • • • ’ x ») = •••>*«))>
9(*lf ...,* „ ) = ----- , *„). x(*i. • ..,* .)),

w here a and t are p a r tia lly d efin ed a u x ilia r y fu n ction s. B y tra n sla tin g
th ese eq u a tio n s (A) in to th e form al sym b o lism of recursive fu n ctio n s
(§ 54), w e o b ta in a sy ste m E d efin in g 9 re cu rsively from x- — The
m e th o d illu stra te d ap p lies to any regular tab le. (A lte r n a tiv e ly , one
c a n tr e a t th e o th er co n n e ctiv e s th en ce, n o tin g th a t Q h as a lre a d y been
tr e a te d in D t of T h eo rem X V I I , since th e stron g ta b le for Q agrees w ith
th e w e a k one, an d u sin g (62) — (64).)
(b) B y (65) it w ill su ffice to tr e a t (E y)y < z. Let x(x v y)

and <f(xv z) b e th e represen tin g fu n ctio n s of R (x x, ..., x„, y)
and ( E y )v< zR (x lt y ), re sp e ctive ly. T h e n (using # # 6 , B f) th e
eq u a tio n s
a(0) = 0, t (1) = 1,
(B) ?(#!, •••»#«> z') = <*(x(xl>■ ■ •>xn, Z—y)),
<p(î. • • •.*«>z) = t( n x(xv • ••>*„,y)) y< z
tra n s la te in to a s y ste m E d efin in g 9 re cu rsively from y.
(c) F irst method . L ik e (a); e.g. for m — 2 an d n— 1,
ffi(0, x) = ?].(*), <t2(0, x) = <p2(*),
<p(*) = *). ?(*) = ° M x).x)-

Second method , for Qv sim u lta n e o u sly d efin ed .
9 ~ w & Q x & y = 9 i) V • • • V (Qm & y = 9 m)].

R emark 1 . F o r c o n sisten cy w ith our u sage in (V I') § 5 7 , an d a t th e
b eg in n in g o f § 54 w here w e first h an d le q u a si-fo rm a lly before tran s
la tin g , w e w rite (A) — (C) here w ith “ = B u t considered in tu itiv e ly ,
as law s o b e y e d b y th e p a rtia l fu n ctio n s ap p earin g in th em , (V I') an d
(A) — (C) sh o u ld b e w ritte n w ith
E xample 5. If <p(x)
is p a rtia l recursive, an d *<p(x) is d efin ed ' is
gen eral recursive, th e n <p(x) is p o te n tia lly recursive. F o r let
y'(x) ~ f x y [ y = <p(x) V (<p(x) is d efin ed & y — o)]-
L e t T* b e a seq uen ce o f p a rtia l fu n ction s . . . , <J»j. B y an extension
Y ' of T w e m ean a seq uen ce ijq, ■ . <Pi o f p a rtia l fu n ctio n s w h ich are
exten sio n s r e sp e c tiv e ly of ipi, i.e. such t h a t for i = 1 , ..., l,
<K'(y1 . • • ym<) = <k(yi, • • •. Vm,) on th e range of d efin itio n o f <J/j. S im i
la r ly if Y include predicates.
T h e reader w ish in g to a d v a n c e r a p id ly to th e m ain results of th e c h a p te r
in § 66 m a y o m it P a r t (b) o f th e follow in g theorem .
T heorem X X I. (a) If
or y(x) ~ F (Y ; x) is a partial re
<p ~ F(Y)
cursive functional, and F ^ ; x-f) k, where Y x are particular functions
and x v k are particular natural numbers, then for every extension Y / of Y j ,
F (^ i/ ; xi) — k.
(b) Let a function <p be defined from a function ^ by an operation of the
form <p(x) ~ F(^(#)), where F(a) is a function from {u, 0, 1, 2, . . . } to
{u, 0, 1 , 2 , . . . } . I f F(u) ^ k, where k is a natural number, but for some
natural number m, not F(m) ~ k , then there is a partial recursive function
^ {taking only m as values) for which the resulting function 0 for (a), > 1 for (b)); an d th e theorem ca n be
sta te d for p redicates, w ith t an d f ta k in g th e p lace of 0 , 1 , 2 , . . . as d efined
values.
P roofs , (a) B y h yp o th esis, there is a sy ste m E of eq u atio n s w ith

g iv e n fu n ctio n letters G an d prin cip al fu n ctio n le tte r f, such th a t, for
any n a tu ra l num ber x an d p a rtia l fu n ctio n s Y: E^T, E b f(x )= y
w here y is a num eral, if an d o n ly if F ( Y ; # ) = y . But F ^ j;^ ) = kf
w here k is a n a tu ral n u m b er; so E ^ 1, E b f( jcx) = fe. N o w if Yx is a n y
exten sio n of Y x, th en E j l C E j 1'; hence also E ^ ,E b f ( x 1) = f e ;
an d therefore F( Y / ; x x) — k. (If th e fu n ctio n a l F(Y) is d efined o n ly under
som e restriction on th e ran ge of Y , th e theorem applies o n ly to exten sio n s
T} s a tis fy in g th e restriction.)
(b) Let —m + 0* pjyT^x, x , y) an d p{x)
~ W i9(:x )—^ & y = 0 ] .
B y T h eorem X V I I I , fy{x) is p a rtia l recursive, an d p(x) is p a rtia l recursive
if cp(x) is. W e show n ow th a t p{x) is n o t p a rtia l recursive. I f {E y)T x{Xy x , y),
th en\LyTx{x, x , y) is undefined, hence ^(jt) is undefined, hence b y h y
pothesis <p(x) == k t hencep(x) is defined. T h u s {E y)T x{Xy x , y) -> (p(x) is
defined}. S im ila rly , [E y)T x{Xy x t y) -> {p(*) is u n d efin ed }; or b y co n tr a
position (cf. * 1 3 § 2 6 ), {p(#) is defined} -> {E y)T x{Xy x f y). T h u s {p (x) is
defined} == [E y ^ ^ x , x, y) == {y)T x{Xy x, y). B y E x a m p le 3 § 6 3 , p(x) is
therefore not p a rtia l recursive.
In E x a m p le s 6 — 8 , w e g iv e (a) an d (b) parts sep arately , so as to

illu stra te b o th p a rts of th e theorem , a lth o u g h th e conclusion of th e (b)
p a rt im plies th a t of th e (a) p art.
E xam ple 6. C an w e im p ro ve upon T h eorem X V I I I b y stre n gth en in g

iyR (x , y)
I to b e th e lea st y such th a t R (x, y) is true irresp ective of w h eth er
R (x, 0), . . R (x } y -~ 1 ) are all d efined (write it th en \iy R {x , y)) ?
(a) N o, since w h en R (x, 1 ) is t, th en \i'yR (x, y) ch an ges from 1 to 0 w hen
R (x, y
0) is ch an ged from u to t. In m ore d e t a il: L e t x i x > ) b e th e rep
resenting fu n ctio n of R (x, y) an d let <p(x) ~ \i,yR {x i y) ~ F(x; x). L e t
Xx, % be choices of x> x sucb th a t Xi(xv 0) ^ u, Xiixi> 0 — 0. T h e n
F (x i; xt) ~ 1 . N o w b y (a) o f th e theorem , if cp is p a rtia l recursive in x,
th en F(x { ; x x) ~ 1 for e v e r y ex ten sio n x[ of Xi- B u t F ( x i; x x) ~ 0 for an
exten sio n xi such th a t Xi\x i* 0) — 0- Th erefore cp is n o t p artial recursive
in x- (b) W e ca n ev en fin d a p a rticu la r p a rtia l recursive R for w h ich
li'yR (x, y) is n o t p a rtia l recursive. T o see this, le t R (x, y) b e of th e form
y = ^ ( # ) V y = l . T h e n [ify R {x ,y ) ch anges from 1 to 0 w h en <]>(#) is
ch an ge d from u to 0. In d e ta il: L e t <p(x) [i'yR(x, y) ~ F(^(%)). T h e n
F(u) ~ 1, b u t F(0) ^ 0. H en ce b y (b) of th e theorem a p a rtia l recursive
can be chosen so th a t , y . A lso (b) for som e p a rtia l recursive ^ an d x> — x(x )
is n o t p a rtia l recursive. (First ta k e x to b e C j § 44; th en choose a ^ to go
w ith th is x b y (b) of th e theorem .)
E xam ple 8 . T h e ta b le for Q ^ R is irregular; e.g. w hen R ^ t,

th en Q = R ch an ges from f to t w h en Q is ch an ged from u to t. H en ce
(a) Q(x) ^ R(x) is n ot p a rtia l recursive in Q an d R , an d (b) for som e
p a rtia l recursive Q an d R , Q(x) ^ R(x) is n o t p a rtia l recursive. — A n y
irregular ta b le ca n b e d e a lt w ith sim ilarly.
§ 65. Godel numbers. U (p yT n(z,

W e now a b b re v ia te xn, y))
as “Q n(z ,x v . . . , * n)” or ev en as ”{z}(xlt . . . , x n)” or “z(xv . . . ) x n)>\
B y T h eorem X V I I I , is a p a rtia l recursive fu n ctio n of n + 1 va riab les
(and hence, for each fix ed z, <!>n(z, x v . . . ,x n) is a p artial recursive fu n ctio n
of th e n va riab les x v . . . , x n). R e w ritin g T h eorem X I X (60), a n y p a rtia l
recursive fu n ctio n cp(xv . . . , x n) of n va riab les can b e o b ta in ed from <S>n>
thus,
(66 ) <p(xv . . . , x n) ~ <bn{e, x v . . ., x n) ~ e(xv . . . , x n),
w here e is a n y G o d el n um ber of <p. S im ilarly, U ([LyT^(zf xv . . ., x n, y))
§65 GODEL NUMBERS 341
w ill be w ritte n "(t>J (z , xv . . . . x n)” , “ {z}'F ( * 1, . . . , * „ ) ” or “z ¥ (x1, . . . , x „ ) ”
for c o m p le te ly defin ed fu n ctio n s Y , w ith correspon ding rem arks. S u m
m arizing :
T heorem X X II (= Th eo rem s The function

X V III + X IX ).
® n(z, x n) is p a rtia l recursive , and 4>n(z, x n) for z = 0, 1 , 2 ,
is an enum eration (with repetitions) of the partial recursive functions
of n variables. S im ilarly, for com pletely defined Y , is partial recursive in
Y and enumerates (with repetitions) the functions of n variables which are
partial recursive in Y . (E n u m eratio n theorem for p a rtia l recu rsive
functions.)
T h is th eorem is o n ly possible b ecau se a p a rtia l recu rsive fu n ctio n

m a y b e u n d efin ed for som e sets o f argum ents.
L e t us re ca p itu la te th e u su a l C a n to r d iago n a l argu m e n t for c o m p le te ly
defined function s. Suppose C is a class o f such fu n ctio n s o f v a rio u s
num bers o f v a r ia b le s ; an d t h a t (z, xv ..., x n) enu m erates (w ith re p
etitions) th e n -v a ria b le fu n ctio n s of C (for som e fix e d n> 1 ). T h e n
(#i, x v x2, . ..,#*) + 1 ca n n o t 8 C , b ecau se otherw ise w e w o u ld h a v e
<%> q>* 2> • • • . * » ) + ! = $ (?, q,

for som e n u m b er q, w h ich is im possible. I f C is closed u nd er th e op eratio n
of passin g from a fu n ctio n cp(z, x v . . x n) to <p(xl9 x lf x 2, . . . , x n)-\-\,
th en $>(z, xv ..., x n) ca n n o t 8 C . In p articu lar, th is show s th a t th ere is
no corresponding en u m eration theorem for th e gen eral recu rsive fu n ctio n s.
B u t w ith p a rtia l fu n ction s, w e w o u ld h a v e in stea d
0% , qf x 2, .. . , x n) + l ^ <t>(q, q, x 2, .
w h ich is n o t im possible, b u t sim p ly m eans t h a t <&(#, q, x 2, ..., x n) m u st

be u n d efin ed. In p articu lar, w e th u s p ro v e th a t th e p a rtia l recu rsive
fu n ctio n O n(z, xv . . . , x n) is u n defin ed, w hen z = % = q w here q is a n y
G o d el n um ber of <&n(xv x v x 2, . . . , x n)-\- 1 .
T h e R ic h a rd p a ra d o x ( § 1 1 ) arose b y a tte m p tin g to m a in ta in sim u l
ta n eo u sly th a t 0 (£, xv ..., x n) is in th e class an d is c o m p le te ly defined.
(In § 1 1 , n = 1.)
D ia g o n a l reasoning w ith p a rtia l fu n ctio n s w ill b e e x p lo ite d fu rth er
in estab lish in g T h eorem X X V I I § 66 .
E xam ple 1. Let ri(zv . . . , z r) b e a g iv e n p a rtia l recursive fu n ctio n .

For each zv ..., z r> l et (pZl,...,Zr(xlt x n) b e th e p a rtia l recursive
...,
fu n ctio n w h ich is defined for an n -tu p le x v . . x n o n ly if rj(zv . . . , z r)
is defined, an d of w h ich in th is case r\(zv . . . , zr) is a G o d el num ber.
Then <pzlt„.,zr(%lt . . . » x n) considered as a function <p(zv ..., zr, xv ..., x n)

of all r-\-n variab les is p artial recursive. F o r
(67) (?(zv . . . , Z r9Xv . . ., * w) — V • • •» Zr)> %1> * * * >*»)•
(This is an e x a m p le for th e first a lte rn a tiv e under (A3) § 62.)
T heorem X X III. F or each m , n > 0 , there is a prim itive recursive

function S™{z, y lt . . y w) {defined below) such that, if e defines recursively
9 (yx, . . . , y m> xlf . . . , # n) as a function of the m -\-n variables y l9 . . . , y m,
% , . . . , # n, then for each fixed m -tuple y v . . . , y m of natural numbers,
S™(e, y v • • •» Vm) defines recursively 9 ^ , . . . , y m, x v . . . , #„) as a function
of the rem aining n variables x v . . . , xn. S ta te d in th e X -notation (§ 10):
I f e defines l y t . . . y w% . . . x n f { y v . . . , y m, x v . . . , x n) recursively, then
S™(e, y v . . . , y m) defines X% . . . * wcp(yv . . . , y m, x v . . . , x n) recursively .
S im ilarly reading (tprim itive recursive function S™'mi"'”mi{z,y1, . . . , y m) ,f,
<(defines recursively from T ’ m ^>Zac£ 0/ <(prim itive recursive function
S™(z, y lt . . . , y m)” , C(defines recursively” , respectively, where Y are l
com pletely defined functions and predicates of m lt variables,
respectively.
P roof (for l = 0). L e t 5®(2) = z. F o r m > 0, choose num bers y x, . . . , y m.
W ith th ese fix e d , let 9 (yv ..., y m, xv . . . , #w) be w ritte n “<p{xv . . . , # n)”
N ow
(68 ) <p(xv . . . , X n) ~ ®m+n{e> Vi > . . • , ym> Xv . . ., Xn).

L e t D b e a sy ste m of eq u a tio n s d efin in g O m+n re cu rsively, w ith g as
p rin cip al fu n ctio n letter, an d n o t co n ta in in g f. L e t C con sist of th e e q u a
tion s of D follow ed b y th e eq u a tio n
f{& 1, • • • >a n) • • •1 ••• >&n)

w h ich w e o b ta in b y tra n sla tin g (68 ). T h is sy ste m C d efines <p(xlf . . . , x n)
recu rsively. Let d, f, g, av ..., a n be th e G 5 d e l num bers of D , f, g ,
av ..., a n, re sp ectively. L e t S%(z, y v . . y m) =
d * [2 e x p 2 15 •3 2/*pl 1*,•, ^ n*5 2a#3iVrw(2!)’p2rw(,/l)***/p^+(J/m),pwl+2*,,,,pm+w+i]
(cf. § 56, # 2 1 § 45, E x a m p le 2 § 52). T h e n S%(z9 y l9 . . . , y m) is p rim itiv e

recursive as a fu n ctio n of z, y lf y m; an d for th e fix e d y lt . . . , y m,
...,
S™(e, y lf . . . ; y w) is th e G 5 d e l num ber of C, an d hence defines f( x v . . . , x n)
recu rsively.
E xam ple 1 (concluded). W e do n o t g e t a larger class of fu n ctio n s

9 (zv . . zr, x v ..., x n) b y s ta rtin g w ith a p artial recursive r\ (as above)
§65 GODEL NUMBERS 343
th a n w ith a p rim itiv e recursive r). F o r let th e (pfo, . . . , z r, x, of
(67) h a v e
of 9 Zl.... zr{xv
e as G o d el num ber. T h e n
zr)”
. . x n),
for th e sam e
i.e.
S rn (e, zv . . . , zr) is a G o d e l n u m b er
S rn(e, zv . . . , zr) is a p rim itiv e recursive
<p2l.... ,r{x i> • • •> *«)•
E xample 2 . F o r a n y gen eral recursive p red ic a te R (xv . . . , x „ ,y )

(w ith n > 1 ), le t e b e a G o d el n u m b er of th e p a rtia l recursive fu n ctio n
~hxxzx2 . . . x n piyR{xv . . . , x n, y). F o r a n y fix e d x x, th e n S \( e ,x x) is a
G o d el n um ber o f lz x 2 . . . x n \xyR{xx, . . . , x n,y ) . T h is fu n ctio n is d efin ed
for a g iv e n n -tu p le z, x 2, . . . , x „ , if an d o n ly if (E y)R (x x, . . . , x n, y) .
H en ce b y T h eo rem X I X ,
(69) (E y)R {x1, . . . , x n, y ) = (E y )T x x), z, x2l . . . , x n, y).

S u b s titu tin g S * (0, x j for z,
(70) (E y)R (x x, . . . , x n, y ) = ( £ y ) T n(S ‘ (e, x x), S ln (e, x x), x 2, . . . , x n, y).
T a k in gn — 1 an d w r itin g " a ” for “ x an d “x ” for " y ” , th is show s t h a t

(E x)R (at x) is p rim itiv e recursive in (E x)T x(a, a, x). T h u s th e decision
problem for {E x)T x{at a, x) is of h igh est d egree o f u n s o lv a b ility for pred
ica tes o f th e fdrm (E x )R (a t x) w ith gen eral recu rsive R (cf. en d § 61).
F u rth erm ore, b y th e d efin itio n of S \t ax^ a 2 -+ SJ(^, a x) ^ S \(e , a2).
T h u s there is a gen eral recu rsive fu n ctio n ^(a) such th a t
[a x^ a 2 -> 7*4 W } & {(• E x)R (a, x) = (E x ^ T ^ a ), <J;(a), x)}; in P o s t's
term in o lo gy {1944 ), th e decision p roblem for th e set a(E x)R (af x) is
1-1 reducible to t h a t for a f t x ^ ^ a , a, x). T h a t th is is so for e v e r y set
a(E x)R (a , x) w ith gen eral recu rsive R is a p ro p e r ty of th e presen t set
# (£ # )7 \(a , a t x) w h ic h P o st sh o w ed is possessed b y an o th er re cu rsiv e ly
enu m erable set “K " ca lled b y h im “ th e co m p le te s e t" (1944 p. 295 a n d
T h eo rem p. 297). W e o b ta in lik e resu lts for th e form s w ith m ore q u a n ti
fiers a n d for l> 0. F o r ex a m p le , from (70) w ith n = 3 (l = 0), w e infer
th a t
(E x)(y)(E z)R (a, x, y, z) = (E x)(y)(E z)T 3(Sl(e, a), S \(e, a), x, y , z)

(cf. *71, *72 §33). Thus (E x)(y)(E z)T z(at a, x, y t z) has a decision p ro b
lem of h igh est degree of u n s o lv a b ility for p red icates of th e form
(E x)(y)(E z)R (a, x , y , z) w ith R recursive, an d th e d ecision p rob lem for
a n y p re d ic a te o f th is form is 1-1 red u cib le to its decision problem .
E xample 3. O m ittin g th e 2 in E x a m p le 2, w e o b ta in t h a t th e d ecision

problem for a(E x)R (a>x) is 1-1 re d u cib le to t h a t for a (E x )T 0(a , x) (and
hence b y E x a m p le 2, th e p re d ic a te s ( E x ^ ^ a , a t x) a n d (E x)T 0(at x)
344 PARTIAL RECURSIVE FUNCTIONS CH x n
h a v e decision problem s of th e sam e degree of u n s o lv a b ility ) ; an d sim

ila rly w ith m ore q u an tifiers, e.g.
E x)(y){E z)R (a, x, y, z ) = (E x ){y)(E z)T 2{Sl2 (e, a), x, y, z),

(■
w here e is a G o d el n um ber of \LzR(a9 %, y, z).

T h e m ajo r p a rt of § 66 can be und erstoo d w ith o u t th e rest of th is
section.
T he A- notatign (to be used in § 82). When

• • • ym * 1 • • . ? (y i, • • • , *«)
is p a rtia l recursive an d e defines it recu rsively, w e shall use
9(yV • * • > ym> %1> • • • > %n)
as an a b b re v ia tio n for S™(e, y l9 . . . , y m). U s u a lly w e o m it th e su bscrip t

“ (e)” . T h u s (for m = 0)“ A x 1 . . . x n <p(%, . . . , x n)” w ill sta n d for som e
G o d e l n um ber of th e fu n ctio n \x 1 . . . x n y(x lt . . . , x n) ; an d (for m > 0)
<iA x 1 . . . x n <p(yv . . . , y m, x v . . . , x n)” w ill stan d for som e p rim itiv e
recursive fun ction , w hose v a lu e for each w -tu p le y v . . . , y m of n a tu ra l
n um bers as argu m en ts is som e G o d el n um ber o f th e fu n ctio n
\X i . . . x n <p(yv . . . , y m, xv . . . , x n).
U sin g th is a b b re v ia tio n w ith th e ab b re v ia tio n
" { *} (*!. . x n)” for <Dn(z, xv . . . , * „ ) ,

.
w e h a v e , for a n y w -tuple tv . . . , tn of n a tu ral num bers,
. . . x n <p(yl9 . . . , y m, xv . . . , x n)}(tlf
— 9 (y v • • • > ymJ v • • • > ^n) •
L ik e n o ta tio n w ith “ A m a y b e used, w h en T are l co m p le te ly
defin ed fu n ction s an d p red icates of m lt . . . , va riab les, re sp e ctiv e ly .
L emma Lem m a I § 4 7 holds reading “partial recursive'y in place

V I.
of “p rim itive recursive \ T h u s (for p = 1 ): I f <p{xlf . . . , x n) is partial
recursive uniform ly in functions 0, Y , and <p*(%, . . x n, c) is the function
obtained when 0 in the definition of 9 is taken to be a function 0* depending
on a param eter c, then 9 * is partial recursive uniform ly in 0*, Y .
P roof . B y h y p o th e sis th ere is a sy ste m E of eq u atio n s such th a t,
for each choice of 0, Yf th e sy ste m E defines th e resultin g fu n ctio n 9
re cu rsiv e ly from 0, T . L e t Y b e <|^, . . . , ; an d sa y th a t th e g iv e n fu n ctio n
le tte rs of E (expressing 0, Yt resp ectively) are t, gv ..., g lf an d th e
p rin cip al fu n ctio n le tte r is f. L e t c b e a va ria b le n o t o ccu rrin g in E . L e t
§65 G O D EL NUM BERS 345
E J result from E b y ch a n g in g sim u ltan eo u sly each p a rt h(rx, . . . , r8)
where h is a fu n ctio n le tte r an d rx, . . . , rs are term s to h ^ , . . rs, c).
L e t E * b e th e sy ste m of eq u atio n s
g 1(a1, . . . » a mi, c) = g i ( a 1, . . . , a TOi), . . . , • • • > am/, c) = g i(a x, . . . , a ^ ) ,
w here g x, . . . , g z are d istin c t fu n ctio n letters n o t o ccurrin g in E . L e t
E* b e E * E | . W e show th a t, for each choice of 0*, Y , th e sy ste m E *
defines <p* re cu rsively from 0*, Y .
Consider a n y fix e d choice of 0*, Y an d c. L e t (E^*)c b e th e set of th e
equ ation s t ( s x, . .., sfi, c)= u w h ich £ E X\ L e t (E^)c b e th e set of th e
resp ective eq u atio n s t ( s 1# . . s q) = u , i.e. th e set of th e eq u a tio n s
expressing th e va lu es of th e fu n ctio n 0 w hen w e ta k e 0(sx, . . . , sq) =
0*($x, . . sq> c) for th e fix e d c.
We easily see th a t if (E?)c, E ^ ; ”'£j, E f(x1, . . x„)=x, then

(E?*)c, E f e * k f ( x 1( . . . , x n ic ) = x , a fortiori e J’ J j ; ; ; ^ ,E * h
f(x1( ...,x„,c)=x.
We now d em o n strate th e converse. Let ( E ^ ' ^ ) c be th e set o f
th e eq u a tio n s g ^ , . . . . y mj, c ) = y w here g ^ y ^ . . . . y m,) = y 6 E ^ " ’^ .
L et (E *)c be%th e sy ste m of eq u atio n s w h ich results b y s u b stitu tin g c

for c th ro u gh o u t E * . W e s a y th a t an eq u a tio n e is a c-equation , if there
is a d ed u ctio n of e from E x E * an d e ith e r e 8 (E^*)c, or e S (E ^ * '^ )c,
or th e p rin cip al eq u a tio n (§ 54) of th a t d ed u ctio n o f e 8 E * an d th e prin

cip al b ran ch co n ta in s an a p p lica tio n of R 1 w h ich su b stitu te s c for c.
W e ca n p ro v e (b y in d u ctio n on th e h e igh t of a g iv e n such d ed u ctio n of e)
th a t, if e is a c-equ ation , th e n (E^*)c, ( E | j “ ;|p c, ( E * ) c b e. In th is
resulting d ed u ction , e v e r y occurrence of a fu n ctio n sy m b o l h is in a p a rt

of th e form h(rx, . . . , rs, c). A n y eq u a tio n of th e form f ( x x, . . . , x n, c ) = x
d ed u cib le from Ex E* is a c-equation . If in th e re su ltin g
d ed u ctio n ju s t described w e ch an ge sim u ltan eo u sly each p a rt h(rx, . . . , rs, c)

to h (rx, . . . , r8), we o b ta in a d ed u ctio n of f ( x x, . . . , x n) = x from
(e ?),. E t r f j . e .
T heorem X X I V . (a) I f y (x x, . . . , x n) is partial recursive uniform ly

in partial recursive functions 01( . . . , 0r, then there is a partial recursive
function <${zx, . . . , z r, x v . . . , x n) such that, when tv . . . , tr are any GSdel
numbers of Qx, . . . . 0r, respectively, y (x x, ~ <p(<x, . . . , t r, x x, . . . . x n).
(b) I f cp is partial recursive uniform ly in partial recursive functions
0X, . . . , 0r, then there is a p rim itive recursive function rj(zv . . . , z r) such
that , when tlf . . . , t r are any Godel numbers of 0X, . . . , 0r, respectively ,
Godel num ber of 9 .
73(Zlf . . . , Zr) is a
yl/so both fa rts hold reading t(partial recursive in 01# . . . , 0r, Y ” , “partial
recursive in Y ” , “ Godel num ber from Y ” in place of “partial recursive in
. , . , 0r” , “partial recursive ” , “ Godel num ber ” , respectively , w A iri Y
l com pletely defined functions .
P roofs (for Z = 0). (a) B y use o f L e m m a V I , th e result o f rep lacin g
0 <(slf . . . , sa.) b y Qqfai, sv . . sq%) (i = 1 , . . r) in th e d efin itio n of
9^, ..., x n) from 01, . . . , 0r is a fu n ctio n y (zlf . . zr, x v . . . , x n) p a rtia l
recu rsive in , O ar. But <&3i, . . . , ® Qr are p a rtia l recursive, so
<p(zlf ..., zrt x lt ..., x n) is p a rtia l recursive. A lso
• • •, S«i) ^ $«<(*,•, Si, • • •. %•), SO <p(*1( — 9^!, . . . , tr, x1, . . . , x n).
(b) L e t £ b e a G o d e l n u m b er of 9 (2^, . . zr> xlt . . x n), a n d ta k e
*}(*i> . . . , z r) = S£(e, zv . . . , z r).
E xample 4. B y our proof of T h eorem X I I I § 60 (going b a c k u pon
t h a t of T h eo rem I V § 57), th e / of F ig u re l a § 61 can b e a n y G o d el n um ber
o f th e p a rtia l recursive fu n ctio n [iyR (xf y). A s \iyR {x , y) is p a rtia l recursive
u n ifo rm ly in R , b y T h eo rem X X I V (b) (applied to th e represen tin g
fu n ctio n of R) there is a gen eral recursive (a c tu a lly , p rim itiv e recursive)
fu n ctio n r\{z) such th a t, for a n y G o d el n um ber r of R (x, y), r\(r) is an / for
F ig u r e la . T h is show s t h a t th e re cu rsively enu m erable set C 0 (n am ely,
£ ( E y ) 7 ,1(#, x, y)) is w h a t P o s t 1944 calls a creative set. (Th e G o d el n u m b er
r of R (x, y) ta k e s th e p la ce here of P o s t's “ basis B” for th e re cu rs ive ly
en u m erable set x {E y)R {x f y).)
E xample 5. O u r ex am p le s of pred icates n o t gen eral recu rsive g iv e n
in T h eo rem V § 57 w ere of n > 1 va riab les. — T h e p red icate {Ex)R{x)
o f 0 v a ria b les is n o t gen eral recursive u n ifo rm ly in th e p red icate R .
F o r suppose it were. T h e n b y T h eo rem X X I V (a) for n= 0 (applied to
th e represen tin g function s), there w o u ld b e a p a rtia l recursive p red ica te
Q(z) such t h a t (Ex)R(x) = Q(t) for a n y G o d el num ber t of R (x). N o w
le t e b e a G o d el n u m ber o f th e p red icate T x{at a , x). T h e n ^(S}(^, a)) w o u ld
b e a gen eral recursive p red icate of a, an d {E x)T x{at a , x) = Q {S\{et a)).
B u t [E x ^ ^ a , a , x) is n o t gen eral recursive (Theorem V (15)).
I ndefinite description. T heorem X X V . F or each n ^ O , there is

a partial recursive function vn(z, xlf . . ., x n) with the following property.
Suppose r is a Godel num ber of a partial recursive predicate R (xv . . . , x n, y).
Then vn(r, x lt . . . , x n) is defined if and only if (E y)R (xv . . x n, y),
§65 GODEL N UM BERS 347
in which case its value is a number y such that R {xv . . . , xm y). W e u su a lly
a b b r e v ia te vn(r, xv . . . , x n) as l<vyir)R (xlf . . . , x n, y ) ” or ev en
4tvyR (xv . . . , x nt y )” t an d read it “ a y (depending on r) such t h a t
R(xlt . . . t xnfy)”.
S im ilarly for l completely defined functions and predicates Y , reading
44partial recursive in Y ” , (<Godel number from Y ' \ in
place of 14partial recursive” t (<Godel num ber” , “vn” , “vy(r)” , respectively.
P roof (for 1 = 0). W e define
v n(zf xl9 . . . , x n) ~ (|iy [ T n+1(z9 xl9 . . . 9x nt (y)0, (y)x) & C 7((y)i)= 0 ])0.
D iscussion . T h e least n u m ber o perator \iy an d th e in d efin ite de

scrip tio n vy correspond to tw o d ifferen t e ffe c tiv e procedures for c a l
c u la tin g a y such th a t R {xv ..., x n$ y), each w ith re sp ective lim itatio n s.
L e t i? be e ffe c tiv e ly decid able. T o ca lcu la te \iyR {xv . . . , x n, y), w e
first t r y to se ttle b y th e algo rith m for R w h e th er R (x lf . . . , x n90) is tru e
or false. I f it is true, w e ta k e 0 as th e n um ber s o u g h t; if false, w e th e n t r y
n e x t to settle w h eth er R (xv ..., x nt 1) is tru e or fa ls e ; an d so on. I f
before fin d in g a y for w h ich R (xv ..., x n, y) is true, w e com e to one for
w h ic h R (xv \ .., x n, y) is undefined, w e are u n a b le to g e t p a s t th is y, b u t
m u st (in obedience to th e rules of our c a lcu la tio n procedure) co n tin u e
a d in fin itu m in th e fu tile effort to settle w h e th er R (xv ..., x n, y) is tru e
or false for it. F o r a p articu lar R, w e m ig h t fin d an oth er procedure w h ich
w o u ld g e t around th is o b stacle, b u t w e ca n n o t in gen eral (cf. E x a m p le 6
§6 4 ).
To ca lcu la te vyR (xv . . . , x n, y ) f w e d istrib u te our efforts to se ttle
w h e th er R (xv . . . , x n, y) is tru e or false for va rio u s y 's , so th a t, w h ile
our efforts to se ttle w h e th er i? (% , . . . , x n9 0) is tru e or false are in progress,
if after a certain n um ber of step s t h e y h a v e n o t led to a decision, w e th en
set to w o rk also on R (x lt ..., x n, 1), a n d so on. T h e search, if n o t te r
m in ated , w ill for each y e v e n tu a lly c a rry th e a lgo rith m for d eterm in in g
th e tru th or fa ls ity o f R (x v ..., x n, y) for th a t y a rb itra rily far. A s
soon as R (x lt ..., x n9 y) is fo u n d to b e tru e for a n y y , w e a c c e p t t h a t y
w ith o u t a tte m p tin g to se ttle w h e th er it is th e least. T h e y o b ta in e d b y
th is procedure m a y v a r y w ith th e algo rith m for R (or th e G o d el n u m b er r)
w h ich is used. A c c o rd in g to th e fo llo w in g e x a m p le (stated for n = 0),
th is is u n a v o id a b le in general.
E xample 6 . Su p p ose v'(z) is a p a rtia l fu n ctio n w ith th e tw o p ro p

erties: (a) I f r is a G o d e l n u m b er o f a p a rtia l recursive p red icate i?(y),
th e n (E y)R (y) -> i?(v'(r)). (b) I f r an d s are G o d e l num bers o f th e sam e
348 PA R T IA L R E CU RSIV E FUNCTIONS CH. X II
predicate, then v'(r)^v'(s). We shall show that then v' is not partial
recursive. For each k} let partial recursive predicates Rk and Sk be defined
thus:
Rk(y) ~ y = 0 V y = l+ 0-lLzT1(k>k, z), Sk{y) ^ y ^ O ^ z T ^ k , kf z) V y = l .
Let rk and sk be any Godel numbers of Rk and Sk, respectively. Given
k , Rk(0) and Sk( 1) are true; so by (a), v'(rk) and v'(sfe) are both defined.
If (EzjT^k, k, z)} then (y)[i?fc(y) ^ Sfc(y)], i.e. Rk and Sk are the same
predicate; and hence using (b), v'(rk)=v(sk). Thus
(EzjT^k, kt z) -> v'(r*)=v'(sfc), or contraposing (cf. *12 § 26 ),
v '(r k) ^ v (sk) -> (EzjT^k, k, z). Conversely, if (2lz) 7 \(&, k, z), then R k(y)
is true only for y = 0 and Sk(y) only for y = 1, so by (a), v'(rfc) =£ v'(sfc).
Thus v'(r*) ^ v'(s*) = {E z^^k, k , z) = (.z)Tt{k, k, z). The expressions
defining Rk(y) and Sk(y) are partial recursive predicates of the two
variables k, y ; say those predicates have Godel numbers r and s, respec
tively. We can take rk = Sj(r, k) and sk = S\(s, k) in the above, obtaining
the equivalence v'(Sj(f, k)) ^ v'(S}(s, k)) = fyT^k, k , z). If v' were
partial recursive, the left member would be a general recursive predicate
of k\ but the right is not (Theorem V ( 14)).
§ 66. The recursion theorem. T heorem XXVI. For any n > 0 ,
let F(£; xv ..., xn) be a partial recursive functional, m which the function
variable £ ranges over partial functions of n variables. Then the equation
...,*„) ~ F(£; xlf . . . , x n)
has a solution 9 for £ such that any solution 9' for £ is an extension of 9,
and this solution 9 is partial recursive.
Similarly, Y are l partial functions and predicates,
£(%, .. •, xn) ~ F(£, Y ; #*)
has a solution 9 for £ swcA 2Aa£ any solution 9' for £ is aw extension of 9,
awi solution 9 is partial recursive in Y. (The first recursion theorem.)
P roof (for l = 0 , w = 1). Let 90 be the completely undefined
function. Then introduce 9!, 92, 93, ... successively by
9 i W ^ f(?o; *)> ?*(*) — F(<Pi;*), ^ f(9 2; x ) ,...
Since 90 is completely undefined, cpx is an extension of <p0; then by Theorem
XXI (a), 92 is an extension of <pv cp3 of 92, etc. Let 9 be the “limit function”
of 90, <pv 92, . . . ; i.e. for each x, let cp(x) be defined if and only if cps(x)
is defined for some s, in which case its value is the common value of
9*(#) for all s > the least such s. Now:
§66 THE RECURSION THEOREM 349
(i) F or each x, cp(%) ~ F(ip; x). For consider any x. Suppose <p(*) is
defined. Then for some s, cp(x) ~ +i(%) [by definition of 9] ~ F(9„; x)
[by definition of <p,+1] ~ F(9; x) [by Theorem XXI (a), since 9 is an
extension of 9,]. Conversely, suppose F(9; x) is defined; call its value k.
Since F is partial recursive, there is a system F of equations defining F(£; x)
recursively from £, say with f as principal and g as given function letter;
so now there is a deduction of f(x)=fc from Ef, F. Let g(yi)=z1, . ..,
g(yp)=z p (where zt = 9(y,)) be the equations of E | occurring in this de
duction. But 9(yx) = 9Sl(yi), • •. , 9(y„) = 9sp(yP) for some sx, . . . , s P.
Let s = m a x ( S j , . .., s P). Then 9(yx) = 9,(yi), .... 9(y„) = 9,(y,). So
g(yi)=*x» • • "%&*)=** 8 E| s- Thus E! S«F b i(x)=k. Hence k ~
F(9s;x) ~ 9m (x) ~ 9(x).
(ii) I f for each x, <p'{x) ^ F(cp'; %), then 9' is an extension of 9. It will
suffice to show by induction on s that, for each x } if <ps(x) is defined, then
cp'(x) <ps{x). B asis : s = 0. True vacuously. I nd . step . Suppose for a
given x that <ps+1(#) is defined. Then 9S+1(^) ^ F(9S; x) ~ F(9'; x) [by
Theorem XXI (a), since by hyp. ind. 9' is an extension of 9J <p'(x).
(iii) I f F defines F(^; x) recursively from and E comes from F by
substituting the prin cipal function sym bol f for the given function sym bol g,
then E defines 9 recursively. It will suffice to show that E b f(x)=ft,
if and only if <ps(x) k for some s. We easily see that if <ps(x) = k, then
E b f(x)=fe. For the converse, we show by induction on h that if there
is a deduction of f(x)=ft from E of height h, then y s{x) = k for some s.
The deduction can be altered if necessary, so that in each inference by
R2 with a minor premise of the form i(y)—z only one occurrence of f (y)
in the major premise is replaced by z (Act 1). The occurrences of f in
equations of the deduction can be classified in an evident manner into
those which come from an occurrence of f in F, and those which come via
the substitution of f for g from an occurrence of g in F. Now consider the
inferences by R2 with minor premise of the form f (y)=z in which the
f of the part f(y) replaced comes from a g in F. Say there are p such infer
ences, the minor premises i(yi)=zv ..., f(yp)= z p of which do not stand
above other such premises. Each of these p premises occurred above
the endequation of the given deduction before Act 1; so using the hy
pothesis of the induction, z1c^ f s ^ ) ~ <p8(yi), • • • > ^ <?8 P(yp) ^ 9 s{yp)
where s = max(s1, ..., sp). Now consider the tree remaining from the
deduction after Act 1, when all the equations above i(y1)=zv ..., f(yp) = z P
are removed (Act 2). In this tree, let each occurrence of f which (before
Act 2) came from a g of F be changed back to g (Act 3). The f s in question
all occurred in the right members of equations, since g being the given
fu n ctio n s y m b o l of F occurs in F o n ly on th e rig h t; so no f is ch an ge d b y
A c t 3 in w h a t w as a m inor prem ise for R 2 before A c t 3 or in th e en d -
eq u a tio n f(x )= ft. F in a lly , le t th e fs of i(yt)=zlf . ..,f (yP)=zP be
ch a n g e d to g (Act 4), w h ich restores th e inferences b y R 2 w h ich A c t 3
spoiled. T h e resu ltin g tree is a d ed u ctio n of f ( x ) = f t from E |« , F . H en ce
k ~ F(<p,;*) ~ <pJ+1(*).
E xample 1. Consider th e p r o b le m : to find a p a rtia l recursive fu n ctio n

9 such th a t
(a) <p(*) = * <p(*);

i.e. to solve th e eq u a tio n Z,(x) cm £(*) for O b v io u s ly a n y p a rtia l fu n ctio n
satisfies th is eq u ation . The p a rtia l fu n ctio n w ith th e least range of
d efin itio n w h ich satisfies is th e co m p le te ly u ndefined fu n ction . T h is is
th e solu tio n 9 g iv e n b y th e theorem (w ith F(£; x) cm £(#)).
E xample 2. T o fin d a p a rtia l recursive fu n ctio n 9 such th a t
(a) <p(*) = * ? W + i ;
x
i.e. to so lve £(, ) £M £(#) + l for C O n ly th e c o m p le te ly u n d efin ed p a rtia l
fu n ctio n satisfies. T h is of course is th e solutio n 9 g iv e n b y th e theorem
(w ith F K ; * ) ~ « * ) + l).
E xample 3. T o fin d a fu n ctio n 9 p artial recursive in x such th a t
/a\ f ? (°) — i >

() l 9 (/) =* xiy. 9 (y))
(Schem a (Va) § 43). O n ly one fu n ctio n 9 satisfies for a g iv e n x> an d w e
a lre a d y k n o w b y T h eo rem X V I I (a) th a t it is p a rtia l recursive in x-
H o w e v e r to see h o w th e th eo rem ap p lies, w e rew rite (a) as
(b) » ( , ) ~ F ( » , x ; * ) w h O T F « , x ; . ) ~ { 9 (^ * " ^ _ 1)) a x>0

(e q u iv a le n tly ,
F(C X ‘>x ) — ^ [{^ = 0 & w=q} V {x>0 & w = x ( x - 1- 1, ^(^•J~ 1))}]).

S in ce F(£, x ^ ) *s p a rtia l recursive (using T h eorem s X V I I , X X (c); or
X V I I , X V I I I , X X (a)), b y th e theorem 9 is p a rtia l recursive in x-
E xample , 4. W e g iv e a new proof of T h eo rem X V I I I (w hich proof in

va rio u s guises ap p eared in K le en e 1935, 1936, 1943). Let
<p(x) ~ \xy\x(x >y ) = 0 ] . T h e n 9 (3;) cm <p(x, 0) where
(a) 9(*. y)— V^t>vlx(x>0=°]-

§66 TH E RECURSION THEOREM 351
B u t <p (x ,y) is th e p a rtia l fu n ctio n <p w ith th e least ran ge of d efin itio n su ch
th a t
(b) ?(*. y) — y x{x, y) = 0,

if
<p(x, y') if x(*. y) # 0.
B y T h eo rem X X (c) (w ith th e first proof), th e rig h t side o f (b) is o f th e

form F(<p, x ; x ,y ) w here F(£, x ; %, y) is p a rtia l re cu rsive; so b y th e presen t
theorem cp(x, y )} an d hence <p(x), is p a rtia l recursive in x-
D iscussion . T h e theorem for Z = 0 asserts th a t w e ca n im pose a n y
relation sh ip of th e form
(72) <p(%, . . . , x n) ~ F ( < p , x n)

exp ressin g th e am b igu o u s v a lu e y{x Xi . . . , x n) of a fu n ctio n 9 in term s
o f 9 itself an d xv ..., xn b y m eth o d s a lre a d y tre a te d in th e th e o r y o f
p a rtia l recursive fu n ctio n s; an d con clu d e t h a t th e p a rtia l fu n c tio n w ith
th e least range o f defin itio n w h ich satisfies th e relatio n sh ip is p a r tia l
recursive.
M oreover th e case of th e theorem for l > 0, in w h ich
(73) <p(xv ..., x n) ~ F(<p, Y ; xv x n)

is th e relation sh ip im posed, can b e used to e x te n d th e b o d y of th e m eth o d s
a v a ila b le for use in further application s.
In our exam p le s of special kin d s of “ recursion” (§§ 43, 46 an d b e g in n in g
§ 55) th e am b igu o u s fu n ctio n v a lu e <p(xlt ..., x n) w as expressed in term s
of va lu e s of tn e sam e fu n ctio n for sets of a rgu m en ts p reced in g th e g iv e n
w -tuple xv ..., x n in term s of som e special ordering of th e n -tu p les. W e
n ow h a v e a gen eral k in d of “ recursion” , in w h ich th e v a lu e <p(xv ..., x n)
can b e expressed as d ep en d in g on other va lu e s of th e sam e fu n ctio n in
a q u ite a r b itra ry m anner, p ro v id e d o n ly th a t th e rule of d ep en d en ce is
d escribable b y p re v io u sly tre a te d e ffe c tiv e m eth od s.
T h e g iv e n “ recursion” m a y n o w b e am b igu o u s as a d efin itio n of an
o rd in ary (i.e. c o m p le te ly defined) n u m b er-th eo retic fu n ctio n 9 , in th e
sense th a t it is satisfied b y m ore th a n one such fu n ctio n (E x a m p le 1 , or (b)
in E x a m p le 4 w hen x(#, y) does n o t va n ish for in fin ite ly m a n y v a lu e s of y).
B u t n ow w e choose as th e solu tio n w h ich in terests us t h a t p a rtia l fu n ctio n
w h ich is defin ed o n ly w h en th e recursion requires it to be. T h e g iv e n
“ recursion” m a y b e in co n sisten t as a d efin itio n o f an o rd in ary fu n ctio n
(E xa m p le 2 ); a ga in th e d iffic u lty is escaped n o w th ro u g h th e fa c t t h a t it
is o n ly a p a rtia l fu n ctio n w h ich w e are seekin g as th e solution. B o th
these situ atio n s can arise w hen th e F is gen eral recursive (E x a m p le s 1
an d 2). W h e n F is in c o m p le te ly defined, th e recursion m a y also d ir e c tly
352 P AR TIAL RECU RSIVE FUNCTIONS CH. X II
d em a n d under our u su al co n ven tio n (§ 63) th a t 9 be un d efin ed for som e

a rgu m en ts (e.g. E x a m p le 3 w hen % is g iv e n to be th e co m p le te ly u n d e
fin ed fu n ctio n ), as w ell as in d ire ctly th ro u gh an in co n sisten cy (not
n ecessarily as o b v io u s as in E x a m p le 2).
G iv e n a p a rticu la r relation sh ip of th e form (72) or (73), it m a y be a
d ifficu lt problem to recognize for w h a t argu m en ts xv ..., x n th e fu n ctio n
v a lu e <p(xlt ..., x n) m u st b e defined. T h is problem is separate from th e
problem , w h ich th e first recursion theorem solves, of recognizing th e
p a rtia l recursiveness of th e solutio n 9 h a v in g th e least range of d efinition.
W e can use th e theorem in p resenting th e case under (A2) § 62. O u r
m eth o d s for sh ow in g g iv e n e ffe c tiv e ly ca lcu lab le fu n ction s to be p a rtia l
recursive (and hence w hen co m p le te ly defined, general recursive) are
now d eve lo p ed to th e p o in t w here th e y seem a d eq u a te for h an d lin g a n y
e ffe c tiv e d efin itio n of a fu n ctio n w hich m igh t be proposed. T o describe
in o rd in ary la n gu a ge a process for ca lcu la tin g a new fu n ctio n 9 , we w o u ld
h a v e to e x p la in e ffe c tiv e ly h ow a n y fu n ctio n v a lu e <p(xlf . . ., x n) is to
be o b ta in ed from va lu es of 9 a lre a d y ca lcu lated . In th is exp la n atio n ,
we w o u ld n o rm a lly e m p lo y e ffe c tiv e ly ca lcu lab le fun ction s p rev io u sly
stu d ied , th e co n n ectiv es of th e propositional calculus, also p o ssib ly
b o u n d ed qu an tifiers (unbounded qu an tifiers w o u ld not be effective), an d
description s of th e form T h e least num ber such t h a t ’ . T h is v o c a b u la r y
tran slates in to operations a lre a d y trea ted in our th e o ry (cf. p a rtic u la r ly
T h eorem s X V I I I an d X X ) . T h e e x p la n a tio n as a w hole w o u ld th en
com e under th e first recursion th eorem ; n a m e ly w e could express it
as a sta te m e n t th a t 9 is to s a tis fy (72) for a certain F, an d is to be d efined
o n ly w hen th e e x p la n a tio n leads to a v a lu e , i.e. 9 is to be th e fu n ctio n of
least range of d efin ition s a tisfy in g (72). T h e n b y th e theorem w e can
con clu de th a t 9 is p a rtia l recursive.
Som e other operations besides those ju st n o ted can b e h an d led b y
m eth o d s a lre a d y considered, e.g. th e d efinition of several fu n ction s
sim u ltan eo u sly (cf. E x a m p le 4 § 46). S till other notions w e m igh t e m p lo y
w e w o u ld e x p e c t to b e able to ex p la in first in term s of th e ab o v e v o c a b u
lary. O u r results are fo rm u late d so as to co n stitu te, n o t o n ly m eth o d s for
sh ow in g p articu lar fu n ctio n s to be p artial (or general) recursive, b u t
also tools for enlarging our sto ck of m eth od s as th e need m a y appear.
T heorem For each n > 0: Given any partial recursive func

X X V II.
tion fy(z, x n), a number e can be found which defines ty(e, xv . . . , x n)
recursively , i.e . such that
(74a) {«}(*,, . . . , x n) ~ x n).
S im ila rly , for l completely defined functions and predicates Y , reading
t(partial recursive in Y ” , t(defines recursively from Y ” , “{ in place of
(tpartial recursive” , ((defines recursively” , “{ }” , respectively. (Th e re
cursion theorem , K le en e 19 38 .)
P roof (for 1 = 0). The fu n ctio n 4>(S*(y, y ), x n) is p a rtia l

recu rsive (cf. T h eo rem X X I I I ) . L e t f d efine it recu rsively, a n d ta k e
e = SU f, /). T h e n e defines re c u rsively th e fu n ctio n of n v a ria b le s o b ta in e d
b y su b s titu tin g th e n u m ber / for th e v a ria b le y in <J>(S* (y, y), x v . . . , x n);
i.e. e defines re cu rsively <j^(S*(/, /), x lt . . x n); i.e. e defines re cu rs ive ly
ty(e, xv . . x n), as w as to b e show n.
D iscussion . T h e theorem ca n b e read as s a y in g t h a t for a n y p a rtia l
recu rsive fu n ctio n ^ th e eq u a tio n
z{xlt ..., x n) ~ ty(z, xv ..., x n)

can b e so lve d for z. F o r th e n o ta tio n , cf. § 65 e sp ec ia lly (66 ). I f w e w rite
(p(xv . . . , x n) for th e fu n ctio n e(xv . . . , x n) d efin ed re cu rsiv e ly b y th e
solutio n e of th is eq u a tio n , (74a) ca n b e w ritte n
(74b) <p(xlt . . . , x n) ~ <\>{et x n).

B u t <J>(z, xv ..., x n) ca n be a n y p a rtia l recursive fu n ctio n of 1 va riab les.
T h u s th e theorem sa y s th a t a p a rtia l recursive fu n ctio n 9 ca n be fo u n d
w hose am b igu o u s v a lu e <${xv . . . , x n) is g iv e n from a G o d e l n u m b er
e of itself an d th e num bers x v . . . , x n b y a n y preassigned p a rtia l recu rsive
fu n ctio n ty. T h e n 9 can b e used in c o n stru ctin g its ow n am b igu o u s v a lu e ,
since in b u ild in g ty(e, x v , . , , x n) w e can use e in p arts of th e fo rm
® * (e,u v . . . , u n) (briefly e(uv . . . , u n))t a n d <1>n(e, u v . . . , u n) is
(p(uv . . . , u n). T h is can be em p h asized b y s ta tin g a corollary. (Th e versio n
of th e co ro llary for l > 0 is left to th e reader.)
Corollary . Given a partial recursive functional F(£; z, %n),

where £ ranges over partial recursive functions of n variables , there can be
found a partial recursive function 9 and a Godel num ber e of 9 such that
(75) cp(xv . . . , x n)~ F ( < ? ;e ,x 1, . . . , x n).
Given a partial recursive functional F(£, 0X, . . . , 0r ; z, w v . . . , w r, x v . . . , x n),
where £ ranges over partial recursive functions of n variables , and 0lf . . . , 0r
over partial recursive functions of specified num bers of variables , there can
be found a partial recursive function 9 (wlf . . . , wri xv . . . , x n) and a Godel
num ber e of 9 such that , when tlf . . . , tr are an y Godel num bers of%Xi . . . , 0r,
respectively ,
(76) 9 (^1 , . . . , tTi x^ . . . , # n) F(9, 0i , • • •, 0r, ^1 , • • •, ^ 1 , • • •, ^«)*
354 P A R T IA L RE C U R SIV E FUNCTIONS CH. X II
P roof (for r = 0). By Theorem XXIV (a), we can find a partial

recursive function $ of n-\~2 variables such that
F(£; z,xv ..., xn) ~ <J/(s, 2 , xlf ..., #w)
when s is any Godel number of Now let ..., #n) ^(z, z,
xv ..., #n). Then by the theorem we find a number e for (74). Writing
<p(xl9 . . . , x n) for the function defined recursively by e, we have
. . . , x n) ~ <[>(*, %, . . xn) ~ ty(e, e, xv .. ., x n) ~ F(<p; e} xv .. .,xn).
D iscussion (concluded). When 9 is partial recursive uniformly in
we may consider that 9 depends partial recursively on £ as a function;
when <p(xlt . .., xn) is a partial recursive function of xlf ..., and a
Godel number of that 9 depends partial recursively on £ as an object.
The gain in Theorem XXVII over XXVI is that now in imposing a
relationship expressing the ambiguous value (p(xv . .., xn) partial
recursively in terms of 9 itself and xlf ..., xn> we can let 9 enter not
only as a function but also as an object. Our conclusion then is that some
partial recursive function 9 satisfies the relationship; but in general
different solutions 9 will be obtained from different selections of the
Godel number / of ^(5^(y, y), xlf . .., xn) in the proof of Theorem XXVII.
Theorem XXVII does not quite include XXVI as the case 9 enters
into the “recursion” only as a function (so that the F(£; z, xv ..., xn) of
the corollary reduces to F(£; xv . . xn)), because the method of proof of
Theorem XXVII does not (at least without some further argument)
show that the partial recursive solution 9 obtained is necessarily the one
having the least range of definition. So Theorem XXVII via its corollary
can take the place of Theorem XXVI in supporting Theses I, I*, It (b)
and I*t (b), but not It (a) and I*t (a).
Of course it is only after we have Church's thesis that we have the
means of regarding every effectively calculable function as a number-
theoretic object. That then an effectively calculable function can be
found to satisfy any “recursion” in which it enters as an object besides
as a function is a consequence of the thesis which proves useful in further
developments.
E xample 2 (concluded). To treat this by Theorem XXVII instead
of XXVI, take the x) for Theorem XXVII - d>1{z9x) + l:
E 'am ple #3 (concluded). Similarly taking ty(zt x) ~
F(X# ^(z, x),x'*x) — \zw[{x=0 & w=q} V {x> 0 & w=x{x —1 , x—1))}].
E xample 5. Given fixed partial recursive functions Qv ..., 05,
to find a partial recursive function 9 with the following properties. If
0iW = ?(*) = 1- If OiW = 1. ?(*) ^ 02(?(O3W))- If 01(x) = 2,
y(x) ~ 04(«9>x) where a^ r is some number which defines Xy 9 (05(.r, y))
recursively. Note first that if e is any Godel number of 9 , then
9 (Gsfc, y)) ^ ®i(«, 05(%, y)). Now teacy®^, 06(*, y)) is a fixed partial
recursive function. If g be a Godel number of it, then S\(g, ef x) is
a Godel number of XyOê, 05(x, y)), i.e. of Xy 9 (05(#, y)). Now the above
three properties are imposed on 9 by the equation
<p(x) p^[{0iM=O & w— 1 }
V {0!(*)=1 & zc= 02(9(03W))} V {04(*) = 2 & tt' = 04(5'2(g, c, *))}].
This is of the form (75) for the corollary, since (using Theorems XVII,
XVIII, XX, XXIII) the right side is of the form F(9 ; e, .r) where F(£; z , x)
is a fixed partial recursive scheme function, and e may be any Godel
number of 9 . For the theorem
^{z,x) OL \IW[{0XW = 0 & w= 1}
V {0j W = 1 & ic=0 2(O1 (r, 03(*)))} V {01 (-t) = 2 & Zt' = 04(S?(g, 2, .r))}].
The problem arose as above with particular functions 6 a, ..., 05 in Kleene
19 3 8 . We can generalize it by considering 0 a> ..., 05 as unspecified
partial recursive functions. Let tv ..., t5 be any Godel numbers of them.
If h is a Godel number of Xztcxy^^z, <b2(te*, y)), then S'l(h, e, /5, x) is a
Godel number of Xy Oa(£, 0 2(/5, xt y)), i.e. of Xy 9 (65(a*, y)). Now the
properties are imposed on 9 by
<p(*) ~ 9(/1( ..., ts, x) ~ (iK'[{01(*)=0 & w= 1}
V {0X(*)= 1 & w = 0,(?(e»(*)))} v { 01(-v) = 2 & w = 04(S®(/«, C, tb, *))}].
By the second part of the corollary, we can find a partial recursive
9 (ie?a, ..., w5, x) and a Godel number e of 9 to satisfy (b).
Chapter XIII
COM PUTABLE F U N C T IO N S
§ 67. Turing machines. Suppose th a t a person is to co m p u te

th e v a lu e of a fu n ctio n for a g iv e n set of argu m en ts b y fo llo w in g pre
assigned e ffe c tiv e in structions. In perform ing th e c o m p u ta tio n he w ill
use a fin ite n um ber of d istin c t sy m b o ls or token s o f som e sort. H e can
h a v e o n ly a fin ite n um ber of occurrences of sy m b o ls under o b servatio n at
one tim e. H e ca n also rem em ber others p rev io u sly observed, b u t again
o n ly a fin ite num ber. T h e preassigned in stru ctio n s m u st also b e finite.
A p p ly in g th e in stru ctio n s to th e fin ite num ber o f o b served an d rem em
bered sy m b o ls or token s, he ca n perform an a c t th a t ch an ges th e situ a tio n
in a fin ite w a y , e.g. he ad ds or erases som e occurrences of sym b ols, sh ifts
his o b servatio n to others, registers in his m em o ry those ju st observed .
A succession of such a c ts m u st lead h im from a sy m b o lic expression
represen tin g th e argu m en ts to an oth er sy m b o lic expression represen tin g
th e fu n ctio n valu e.
W e n ow inquire w h e th er it is n o t possible to a n a ly z e th e a c ts th e
co m p u te r ca n perform in to certa in “ a to m ic ” acts, such th a t a n y per-
form able a c t w ill be e q u iv a le n t to som e succession of ato m ic acts. T h e
a to m ic a c ts w o u ld be co m b in atio n s of th e fo llo w in g: recogn izin g a single
ob served occurrence of a g iv e n sy m b o l, erasing th is occurrence, m ark in g
dow n a single occurrence o f a sym b o l, d isp la cin g th e p o in t of o b serva tio n
of a g iv e n a rra y of sy m b o ls to an a d ja ce n t p o in t, an d alte rin g th e re
m em bered inform ation .
S u ch an an aly sis w as u n d erta k en b y T u rin g (1936 - 7 , received for p u b li
ca tio n 28 M a y 1936) in th e form of a d efinition of a kin d of co m p u tin g
m achine. A sim ilar an alysis w as proposed b rie fly b y P o st (19 3 6 , received
for p u b lic a tio n 7 O c to b e r 1936).
T h e' evid en ce th a t th e an aly sis is com p lete, i.e. th a t to a n y fu n ctio n
w h ich is e ffe c tiv e ly ca lcu la b le a T u rin g m ach in e can b e fou n d w h ich
co m p u tes it, w ill be re view ed after w e h a v e b ecom e fam iliar w ith th e
T u r in g m ach ines (§ 70).
W e b egin b y fo rm u la tin g th e id ea of a T u r in g m achine, as follow s.
356
§67 TURING MACHINES 357
T h e m ach in e is supplied w ith a linear tape, (poten tially) in fin ite in b o th
direction s (say to th e left an dright). T h e ta p e is d iv id e d in to squares.
E a c h square is ca p a b le of b e in g blank, or of h a v in g printed u pon it a n y
one of a fin ite list sv . . ., s* (j > 1) of sym bols, fix e d for a p a rticu la r
m achine. I f w e w rite “ s0” to sta n d for “ b la n k ” , a g iv e n square ca n th u s
h a v e a n y one of / + 1 conditions s0, . . . , s,-. T h e ta p e w ill b e so e m p lo y e d
th a t in a n y “ situ a tio n ” o n ly a fin ite n u m ber ( > 0) of squares w ill b e
prin ted.
T h e ta p e w ill pass th ro u gh th e m ach ine so th a t in a g iv e n “ situ a tio n ”
th e m ach in e scans ju s t one square (the scanned square). T h e sy m b o l on
th is square, or s0 if it is b la n k , w e ca ll th e scanned sym bol (even th o u g h
s0 is n o t p ro p erly a sym bol).
T h e m ach in e is ca p a b le o f b ein g in a n y one o f a fin ite list q 0, . . . , q *
(k > 1) of {machine) states (called b y T u r in g “ m ach in e co n figu ra tio n s” or
“ w -c o n fig u ra tio n s” ). W e ca ll q 0 th e passive (or term inal) sta te; and
qv ..., q w e ca ll active states. T h e list q 0, . . . , q * is fix ed for a p ar
ticu la r m achine.
A {tape vs. machine) situation (called b y T u r in g “ co m p lete c o n fig u
ra tio n ” ) consists in a p a rticu la r p rin tin g on th e ta p e (i.e. w h ich squares
are prin ted, an d each w ith w h ich of th e j sym bols), a p articu lar po sitio n
o f th e ta p e in th e m ach in e (i.e. w h ich square is scanned), an d a p a rticu la r
s ta te (i.e. w h ich of th e & + 1 sta te s th e m ach in e is in). I f th e sta te is a c tiv e ,
w e ca ll th e situ atio n active; otherw ise, passive.
G iv e n an a c tiv e situ atio n , th e m ach in e perform s an {atomic) a d (called
a “ m o v e ” b y T u rin g). T h e a c t perform ed is d eterm in ed b y th e scan n ed
sy m b o l sa an d th e m ach in e s ta te q c in th e g iv e n situ atio n . T h is p air
(stt, q c) w e ca ll th e configuration. (It is active in th e present case th a t q c
is a c tiv e ; otherw ise passive.) T h e a c t alters th e three p a rts of th e situ a tio n
to p rodu ce a resu ltin g situ atio n , thus. F irst, th e scan n ed sy m b o l sa is
ch an ge d to sb. (B u t a= b is p erm itted , in w h ich case th e “ c h a n g e ” is
identical.) Second, th e ta p e is sh ifted in th e m ach in e (or th e m ach in e
sh ifts alo n g th e tape) so th a t th e square scanned in th e resu ltin g situ a tio n
is eith er one square to th e left of, or th e sam e square as, or one square
to th e righ t of, th e square scan n ed in th e g iv e n situ atio n . T h ird , th e
m ach in e s ta te q c is ch an ge d to q d. (B u t c = d is p erm itted.)
N o a c t is perform ed, if th e g iv e n situ atio n is passive.
T h e m ach in e is used in th e follow in g w a y . W e choose som e a c tiv e
situ a tio n in w h ich to sta rt th e m achine. W e call th is th e in itia l situation
or in put. O u r n o ta tio n w ill b e chosen so th a t th e s ta te in th is situ a tio n
(the in itia l state) is q v T h e m ach ine th e n perform s an a to m ic act. If th e
358 COMPUTABLE FUNCTIONS CH. XIII
situation resulting from this act is active, the machine acts again. The
machine continues in this manner, clicking off successive acts, as long
and only as long as active situations result. If eventually a passive
situation is reached, the machine is said then to stop. The situation in
which it stops we call the terminal situation or output.
The change from the initial situation to the terminal situation (when
there is one) may be called the operation performed by the machine.
To describe an atomic act, we use an expression of one of the three
following forms:
%L<\d, s6^qd,
The “L”, “C”, “R ” indicate that the resulting scanned square is to the
left of, the same as (“center”), or to the right of, respectively, the given
scanned square.
The first part of the act (i.e. the change of s0 to sb) falls into four cases:
when a = 0 and b > 0 , it is “prints s6” ; when a > 0 and b — 0 , “erases sa” ;
when a, b> 0 and a ^ b t “erases s0 and prints s5” or briefly “overprints
sb” ; when a = b, “no change”. We often describe this part of the act as
“prints s6” without regard to the case.
To define a particular machine, we must list the symbols sv ..., s;*
and the active states qv ..., qk, and for each active configuration (s0, qc)
we must specify the atomic act to be performed. These specifications
may be given by displaying the descriptions of the required acts in the
form of a {machine) table with k rows for the active states and / + 1
columns for the square conditions.
E xample 1. The following table defines a machine (“Machine 21”)
having only one symbol and only one active state qv
Name of Machine Scanned symbol
machine state So Sl
21 qi sxCq0 Si^qi
Suppose the symbol s2 is actually a tally mark “I”. Let us see what
the machine does, if a tape of the following appearance is placed initially
in the machine so that the square which we identify by writing the
machine state qt over it is the scanned square. The conditions of all
squares not shown will be immaterial, and will not be changed during the
action.
qi
1 1 1
The machine is in state qv and is scanning a square on which the symbol
sx is printed. In this configuration, the atomic act ordered by the table
is s1Rq1; i.e. no change is made in the condition of the scanned square,
the machine shifts right, and again assumes state qv The resulting
situation appears as follows.
qi
1 1 1
The next three acts lead successively to the following situations, in the
last of which the machine stops.
<h
1 1 1
<Ji
1 1 1
q0
1 1 1 1
Machine 91 performs the following operation: It seeks the first blank

square at or to the right of the scanned square, prints a I there, and
stops scanning that square.
Now we define how a machine shall ‘compute' a partial number-
theoretic function 9 of n variables (cf. § 63). The definition for an or
dinary (i.e. completely defined) number-theoretic function is obtained
by omitting the reference to the possibility that (p(xv ..., xn) may be
undefined.
We begin by agreeing to represent the natural numbers 0, 1 , 2 , ... by
the sequences of tallies I, II, III, ..., respectively, the tally “I” being
the symbol sx. There are y + 1 tallies in the representation of the natural
number y.
Then to represent an m-tuple yv ..., ym (m ;> 1 ) of natural numbers
on the tape, we print the corresponding numbers of tallies, leaving a
single blank between each two groups of tallies and before the first and
after the last.
E xam ple 2. The triple 3, 0, 2 is represented thus:
We say that (the representation of) a number y

(or of an w-tuple
ylf .. ., ym)
on the tape is (scanned) in standard position when the ,
scanned square is the one bearing the last tally in the representation of
y (or of ym).
Now we say that a given machine $01 computes a given partial function
9 of n variables (n > 1), if the following holds for each n-tuple %,
of natural numbers. (For the case n = 0 , cf. Remark 1 below.) Let
xt, . . xn be represented on the tape, with the tape blank elsewhere, i.e.
outside of the %-f- ... + # n + 2 n + 1 squares required for the repre
sentation. Let $0i be started scanning the representation of xv ..., xn
in standard position. Then $01 will eventually stop with the n-\-1-tuple
xv ..., xn, x represented on the tape and scanned in standard position,
if and only if <p(xlf .. ., xn) is defined and <p(%, ..., xn) = x. (If <p(x1} ..., xn)
is undefined, $0t may fail to stop. It may stop but without an n-\-\ -tuple
%, ..., xn) x scanned in standard position.)
Example 2 (concluded). If 9 (3 , 0, 2) = 1 and $0£ computes 9 , then
when $0t is started in the situation
qi
with all squares other than those shown blank, it must eventually stop
in the situation
q0
where the condition of the squares other than those shown is immaterial.
Although only one symbol sx or “I” is used in stating the arguments
and in receiving the function value, others may be used in the progress
of the computation. For each n > 1 , each machine (with its first symbol
s-lserving as the tally) computes a certain partial function of n variables.
A partial function 9 is computable, if there is a machine $01 which
computes it.
W e h a v e n o t a tte m p te d here to reproduce th e d eta ile d fo rm u latio n of
T u r in g 19 3 6 - 7 , b u t o n ly his gen eral co n cep tio n o f th e b eh a v io r of th e
m achines. A lth o u g h he n o ted a v a r ie t y o f a p p licatio n s of his m ach ines,
he con fined his d eta ile d d eve lo p m en t to m ach in es for co m p u tin g d u a l
exp an sion s of real num bers x (0 < x < 1 ). T h e successive d igits w ere to b e
p rin ted ad in fin itu m on a lte rn a te squares of a 1 -w a y in fin ite tap e, w h ile
th e in te rven in g squares w ere reserved for te m p o ra ry notes servin g as
scratch w o rk in th e co n tin u in g co m p u ta tio n . It should be an e a sy
exercise, a fte r §§ 68 an d 69, to sh ow th a t a m ach in e e x ists w h ich does
this, if an d o n ly if th e n -th d ig it in th e d u al exp an sio n is a co m p u ta b le
fu n ctio n of n. A critiq u e, of h elp to one w h o w o u ld s tu d y T u r in g ’s
p ap er in d etail, is g iv e n in th e a p p e n d ix to P o s t 1 9 4 7 . O u r tre a tm e n t
here is closer in som e respects to P o st 19 3 6 . P o st 1936 considered
co m p u ta tio n w ith a 2 -w a y in fin ite ta p e an d o n ly 1 sy m b o l.
O u r m ain o b je c tiv e n e x t is to p ro ve th e eq u ivalen ce of c o m p u ta b ility
w ith p a rtia l recursiveness, or, w hen o n ly c o m p le te ly d efin ed fu n c
tio n s are considered, w ith gen eral recursiveness (§§ 68 , 69).
I t is of som e in terest to see a t th e sam e tim e w h eth er a 1 -w a y in fin ite
ta p e an d 1 sy jn b o l w ill n o t suffice. W e s a y t h a t a m ach in e 50i 1/1 com putes
9 , if it co m p u te s 9 su b je c t to th e fo llo w in g restrictions, an d w e s a y t h a t
9 is 1/1 com putable , if it is 1/1 co m p u te d b y som e m ach in e 501: (i) T h e
m ach in e 501 h as o n ly one sy m b o l s*. M oreover, w hen 50£ has been sta rte d
as described for a n y w -tuple xv ..., x n, it b e h a v e s so th a t th e fo llo w in g
((ii) — (iv)) are th e case, (ii) A square w h ich in th e in itial situ a tio n is to
th e left of th e represen tatio n of x lt . . . , x n (i.e. to th e left of th e b la n k
preced in g th e first t a lly of x x) w ill n o t b e scan n ed in a n y su b seq u en t situ
ation. H en ce th e co m p u ta tio n can e q u a lly w ell b e perform ed on a 1 -w a y
in fin ite ta p e (infinite to th e right), (iii) I f (p(xv . . . , x n) is d efined , th en
in th e term in al situ a tio n th e represen tation of x v . . . , x n} <p(xv . . . , x n)
starts on th e sam e square as d id th a t of xv ..., x n in th e in itia l situ a tio n ,
an d all squares to th e righ t of th e represen tatio n of xv ..., x n, <p(xv ..., x n)
are b la n k (also b y (ii), th ose to th e le ft) . (iv) T h e m ach in e 501 e v e n tu a lly
stops, o n ly if <p(xv ..., x n) is defined.
I t w ill follow from th e results of §§ 68 an d 69 th a t 1/1 c o m p u ta b ility
is e q u iv a le n t to c o m p u ta b ility ; an d h ence th a t all w /s c o m p u ta b ility
n otions are e q u iv a le n t, w here w is 1 or 2 acco rd in g as (ii) — (iv) are re
quired or n o t, an d 5 is an u p p er b o u n d for th e num bers of sy m b o ls d ifferen t
m ach ines m a y h a v e (00 , if there is no fin ite u p p er bound ). (In th is scale,
our origin al n otion of c o m p u ta b ility is 2/oo co m p u ta b ility .)
O th e r v a ria n ts of 'c o m p u ta b ility ’ can b e fo rm u lated , e.g. w e m ig h t
o m it (iii) an d (iv) w hile re tain in g (ii), or in stea d of (ii) w e m ig h t assum e
t h a t th e m ach in e is supplied w ith o n ly a 1 -w a y in fin ite tap e, an d require
it to sto p if a situ a tio n is reached from w h ich th e m ach ine ta b le orders
a m o tio n leftw a rd from th e leftm o st square. In ste a d of represen tin g th e
num bers on th e ta p e b y tallies, w e m ig h t use du al n o tatio n or d ecim al
n o tatio n . I t should not b e h ard to show, a fter §§ 68 an d 69, th a t each of
th ese other c o m p u ta b ility n otion s is eq u iv a le n t to ours (cf. § 70).
N o w w e e x te n d our notion s to th e case of co m p u ta tio n from l co m
p le te ly defined fu n ctio n s $ lt ..., (briefly Y ) of m lf . . . , m t va riab les,
re sp e c tiv e ly . H ere th e id ea is m odified b y assum in g th a t a n y v a lu e of
one of th e fu n ctio n s Y , if d em an d ed in th e course of th e co m p u ta tio n ,
w ill th ereu pon be su p p lied (cf. end § 61).
A m ach in e for th is purpose m a y h a v e am on g its a c tiv e sta te s qv . . q*
ones from w h ich a new k in d of a c t (not ato m ic in character) is perform ed.
L e t q c be one of these states. W h en th e sta te is q c, an d (for a certa in i
d e p e n d in g on c) an m r tu p le yv . . . , y m. is scanned in sta n d a rd p osition,
th e a c t perform ed b y th e m ach in e shall con sist in su p p ly in g n e x t to th e
rig h t of th is m r tu p le ^ 1(y1, . . . , ymt) + 1 tallies an d a b la n k , a t th e sam e
tim e d isp la cin g b y . . . , y m*-)+2 squares to th e righ t all p rin tin g
w h ich p re v io u sly e x iste d to th e rig h t of th e scan n ed square, an d th en
assu m in g sta te qd (depending on c) w ith th e resu ltin g w ^ -j-l-tu p le
yv ..., y m., ^ i(yv ..., y mi) scan n ed in stan d ard position. W h e n th e s ta te
is q c b u t an w r tu p le is n o t scanned in stan d ard position, w e shall consider
th a t th e m ach in e perform s an id e n tica l a c t (w ith resu ltin g situ a tio n th e
sam e as th e original situ atio n ). In th e m ach in e ta b le , corresponding
to each such sta te q c, w e sp e c ify o n ly th e p air iq d, as in o th er respects
th e a c t is determ in ed from t h e situ a tio n b y th e d efin ition s ju st g iv e n
an d th e fu n ctio n s Y .
M o d ify in g our m ach in e n o tio n in th is w a y , w e o b ta in th e three n o tio n s :
m achine from computes cp from Y , 9 is computable from Y . I f (for
Y , 50?
fix e d n, l , m v. . . , m t) 9 is co m p u ta b le from Y b y a m ach in e 9D?, th e ta b le
for w h ich is in d ep en d en t of Y , w e s a y th a t 9 is computable uniform ly from
Y , or t h a t th e fu n ctio n a l 9 ~ F(Y) is computable. (T h e th eo rem s of
§§ 68 an d 69 estab lish u n ifo rm ity , w h en ever u n ifo r m ity is assum ed.)
In d efin in g 50? 1 /1 com putes 9 from Y , an d 9 is 1/1 computable from Y,
w e a d d th e fo llo w in g to our form er list (i) — (iv) of restriction s: (v) F o r
each s ta te q c for w h ich th e ta b le e n tr y is of th e form iq d, th e s ta te q c
is reach ed o n ly w h en som e m r tu p le yv ..., y m{ is scan n ed in stan d ard
p o sitio n an d all squares to th e righ t of th e scanned represen tation of
y lf ..., y mt are blan k.
§68 COMPUTABILITY OF RECURSIVE FUNCTIONS 363
W e can easily form u late th e d efin itio n of ‘c o m p u ta b ility ’ from assum ed
fu n ction s, s a y from ^ (w ith l = mx = 1 ), in an eq u iv a le n t w a y , in w h ich
th e m ach ine perform s o n ly a cts a to m ic in ch aracter, b u t is supplied w ith
a ta p e h a v in g a (p oten tially) in fin ite p rin tin g represen tin g th e sequence
of th e va lu es of This p rin tin g m a y b e on a second tap e, or on alte rn a te
squares of th e one tap e.
R em ark 1. In th is ch ap ter, outside th e present rem ark an d passages

referring to it, w e shall u n d erstan d th a t we are d ealin g w ith fu n ction s of
n > 1 variables. Sin ce w e h a v e n o t p ro v id e d for represen tin g n -tu p les of
n a tu ral numbers on th e ta p e for n= 0 , w e s a y a m ach in e com putes a
fu n ctio n 9 of 0 va riab les, if it co m p u tes the fu n ctio n cp(x) of 1 v a ria b le such
th a t <p(x) ~ 9 . F o r an assum ed fu n ctio n of 0 va riab les, th e n o n -id en tica l
a c t from a sta te q c w ith ta b le e n try iq d is to be perform ed w h en a n y
1-tu p le x is represented on th e tap e. W ith these definitions, it w ill follow ,
as soon as w e h a v e p ro ved T h eorem s X X V I I I — X X X for fu n ctio n s of
> 0 va riab les, th a t th e y also h old for fu n ctio n s of 0 variables.
§ 68. Computability of recursive functions. T heorem X X V IIl.

Every partiak recursive function 9 is 1/1 computable. E very function cp
partial recursive in l com pletely defined functions Y is 1/1 computable from W.
T h e proof, g iv e n a t th e end of th is section, w ill be b a sed on C o ro llary
T h eorem X I X § 63 (for 1= 0 an d a gen eral recursive 9 , on C o ro llary
T h eo rem I X §58). A n in tu itiv e ca lcu latio n b y S c h e m a ta (I) — (V I) is
accom plish ed b y repetitio n s of a few sim ple operations, such as c o p y in g
a num ber p re v io u sly w ritte n (at a d eterm in ate earlier position), a d d in g or
su b tra ctin g one, d ecid in g w h e th er a g iv e n n um ber is 0 or n o t 0 . W e
shall first co n stru ct som e m ach ines to perform such operatio ns as th e s 6 .
W e b egin b y in tro d u cin g som e n o tatio n s w h ich w ill b e co n ven ien t
here, in clu d in g m o d ificatio n s of n o tatio n s used in § 67.
A g iv e n square of th e ta p e m u st n ow h a v e one of tw o con d ition s s 0
an d sv or b la n k an d p rin ted w ith a t a lly “ I” , or w h ite an d b la ck . In th e
illu stration s in th is section, it w ill b e co n ven ien t to show th ese tw o co n
ditions as “ 0 ” an d “ 1 ” , re sp ectively.
T o id e n tify th e scanned sy m b o l in our illustration s, w e n ow u su a lly
w rite sim p ly a b ar over it, in stead of th e m ach in e sta te as in § 67.
Som etim es a sm all num eral “ 1” or “ 2” w ill b e w ritte n w ith th e b ar to
in d icate a certain m ach in e sta te (the sam e in all occurrences of th a t
num eral), to w h ich w e w ish to call a tte n tio n w ith o u t in d ic a tin g a t th e
m o m en t w h ich of q 0, . . . , q fc it is.
364 COM PUTABLE FUNCTIONS CH. X III
In d escribin g an a to m ic a c t, w e now w rite “ P ” (“ p rin ts” ) for ch a n g in g

0 to 1 (i.e. s0 to Sj), “ £ ” (“ erases” ) for ch an gin g 1 to 0 (i.e. sx to s0), a n d
n o th in g for th e ch an ge s a to s5 w hen a = b. W e o m it th e “ C ” w h ich w e
used in § 67 to sig n ify no ch an ge in th e scan n ed square. W e n o w w rite
sim p ly “ 0” , “ 1” , . . . , “ ft” for th e m ach in e sta te s q 0, . . . , q fc. For
exam p le , th e a c t w ritte n “ s + q o ” in § 67 is n ow w ritte n as “ P 0 ” w h en s0
w as th e scanned sy m b o l in itia lly , an d as “ 0” w h en sx w as th e scan n ed
sy m b o l in itia lly .
A s a fu rth er exam p le, w e repeat th e ta b le for M ach in e 21 (E x a m p le I
§ 67), an d th e form er illu stra tio n of th e operatio n it perform s (show ing
now o n ly th e in itia l an d fin al situations).
N a m e of M achin e S can n ed sym b o l

m ach in e s ta te 0 1
2t 1 P0
S ta r tin g from th e in itial situ atio n
T 1 1 0,
M ach in e 51 reaches th e term in al situ atio n

1 1 1 T.
M ach in e 21 prin ts on th e first b la n k square a t or to th e righ t of th e
scan n ed square, an d stop s there.
In our illu stra tio n s prior to “ P roof of T h eo rem X X V I I I ” , none o f th e
squares to th e le ft or righ t of th ose show n is scanned in a n y in term ed ia te
situ a tio n d u rin g th e m ach in e action . T h u s th eir co n d itio n is im m aterial.
F o r each m ach in e co n stru cted prior to “ P ro o f of T h eo rem X X V I I I ” ,
w e sh all s ta te w h a t is th e le ftm o st square scan n ed in a n y in term ed ia te
situ atio n , in each case w h e n it falls ou tsid e th e in te r v a l b etw ee n th e
in itia lly and te rm in a lly scanned squares in clu sive. T h is in fo rm atio n
w ill b e used in co n clu d in g th a t th e m ach ines w e b u ild in th e proof of th e
theorem s a tis fy th e restriction (ii) § 67.
In th is sectio n w e sh all ge n e ra lly th in k o f a sequence of y + 1 co n
se c u tiv e p rin ted squares preced ed an d follow ed b y a b la n k square as
represen tin g th e n a tu ra l n u m ber y. T h e sam e b la n k square m a y b e
regarded sim u ltan eo u sly as th e la st square in th e represen tation o f one
n um ber an d th e first in th e represen tation of another. T h is enables us
to th in k of* a n y p rin tin g on th e (2-w a y infinite) ta p e as con sistin g of
a fin ite succession of represen tation s of n a tu ral num bers. M ore th a n one
co n secu tive b la n k squares (say 2 + 1 of them ) b etw ee n tw o gro ups o f
co n secu tive p rin ted squares w e ca ll a gap (of z squares) b e tw e e n th e
§68 C O M PU TAB ILITY OF RECU RSIV E FUNCTIONS 365
num bers represented. T h e b la n k portion of th e ta p e before th e first, or
a fte r th e last, p rin ted square w e also call a gap (of oo squares). A s in
§ 67, w e do not th in k of m successive represen tation s of n a tu ra l num bers on
th e ta p e as represen tin g an m -tu ple, unless there are no ga p s b etw ee n them .
Sim ilar rem arks a p p ly to a 1 -w a y in fin ite tap e, p ro v id e d th e first
(leftm ost) square is b la n k. T h ere m a y or m a y n o t b e a (finite) g a p before
th e first num ber.
N o w consider th e problem of co n stru ctin g a m ach in e SB to perform th e

fo llo w in g o p e ra tio n : G iv e n a num ber on th e tap e, n o t th e leftm o st, to
m o ve th is num ber to th e left so as to close th e g a p (if an y) b etw ee n it
an d th e preceding num ber. T h e g iv e n n um ber is to be scanned in sta n d a rd
position before an d after th e operation. T h e leftm o st square scan n ed in
a n y situ atio n durin g th e a ctio n sh all be th e righ tm o st p rin ted square
of th e p reced in g num ber.
W e b egin w ith an illu stratio n , sh ow ing a ty p ic a l in itia l situ a tio n
(S itu atio n 1), som e proposed in term ed ia te situ atio n s n ot n ecessarily
co n secu tive (Situ ation s 2 — 16), an d th e desired term in al situ a tio n
(S itu atio n 17), for one plan for m ech an izin g th e operation. E x p la n a tio n s ,
an d th e tabl£ of a m ach ine w h ich w ill perform th e operatio n acco rd in g
to th is plan, w ill follow . T h e “ SB,” , “ 932” , " » 8” , a t th e righ t refer to
an an alysis of th is m ach in e as a co m b in atio n of sim pler m achines, g iv e n
afterw ards.
1. 1 0 0 0 1 1 T 0
2. 1 0 0 0 1 T 0 0
3. 1 0 0 0 1 1 0 0 ®1
4. 1 0 0 1 1 1 0 0
5. 1 0 0 T1 1 1 0 0 1r *
6. 1 0 0 1 1 I 0 0
7. 1 0 0 1 1 0 0 0
8. 1 0 o 1 1 0 0 0 ®1
9. 1 0 1 1 1 0 0 0
10. 1 0 l1 1 1 0 0 0 1r ^2
«
11. 1 0 1 1 T 0 0 0 '
12. 1 0 11 0 0 0 0
13. 1 0 1 1 0 0 0 0 * ®1
14. 1 1 1 1 0 0 0 0
15. 1 T2 1 1 0 0 0 0
16. 1 0 1 1 0 0 0 0 :1 31 } ®»
17. 1 0 1 1 1 0 0 0
366 COM PU TABLE FUNCTIONS CH. X III
T h e step from S itu a tio n 1 to S itu a tio n 2 is an erasure of th e scanned 1,

an d a m o tio n once to th e left. T h e n (2 — 3) th e m ach in e seeks th e first
b la n k square at or to th e left of th e scanned s q u a r e ; (3 — 4) it prin ts
there, an d goes one m ore square to th e l e f t ; (4 — 5) it recognizes (in
th is illustration ) an d registers th a t it is now on a b la n k square, and
goes one square to th e r i g h t ; (5 — 6) it goes to stan d ard position o ver th e
n u m ber n o w b ein g scanned. A lto g e th e r 1 — 6 m o v e th e n um ber as a
w h ole one square to th e left. T h e n 6 — 11 repeat th is operation. T h e
sam e c y c le of steps is sta rte d aga in (11 — 14), b u t is con clu ded d if
fe re n tly (14— 17) a fte r 1 in stea d of 0 is d iscovered on th e leftm o st
square ex am in e d (at S itu a tio n 14).
T h e reader m a y v e r ify th a t th e m ach ine d efined b y th e follow in g
ta b le w ill ca rry o u t th e operatio n in this illu stration , passin g th ro u gh
S itu a tio n s 1 — 17 an d oth er in term ed ia te situ a tio n s; an d he m a y co n
vin ce h im self n o w (or a fter th e n e x t discussion) th a t it w ill a lw a y s perform
th e op eratio n described, sta rtin g from a n y in itial situ atio n of th e kin d
described. W e le a v e a dash “ — ” in p lace of the ato m ic a c t order a t th ose
places in th e ta b le w here th e a c t ordered is im m a terial to us, because th e
co n figu ratio n is one w h ich w ill n o t arise in our use of th e m ach ine, i.e.
w hen it is sta rte d from an in itia l situ a tio n of th e kin d described. H o w e v e r
th e defin ition of th e m ach in e m a y b e un d ersto o d to b e co m p le ted b y
su p p ly in g “ 0” a t those places.
N a m e of M achin e S can n ed sy m b o l
m ach in e sta te 0 1
93 1 — EL2
2 PL3 L2
3 R4 R5
4 LI R4
5 — ER6
6 P0 R6
In d isco verin g th is m ach in e, w e first a n a ly ze d the w hole operation w e
w ish it to perform in to sim pler (not necessarily atom ic) acts, an d then
fu rth er a n a ly z e d these in to th e a to m ic a cts to be perform ed b y th e
m achine. W e can th in k of the m ach ine as o b ta in ed b y co n n ectin g to geth er
several sim pler m ach in es w h ich perform one or a succession of th e steps in
th e p relim in ary an alysis. W e now introduce n o tatio n s w h ich w ill be
useful in describing a m ach in e as a co m b in atio n of sim pler m achines.
If tw o operations are to be perform ed su ccessively, and a m ach ine
3£ perform s the first of them , and a m achine ?) the other, then the com bin ed
§68 CO M P U T A B IL IT Y OF R ECU RSIVE FUNCTIONS 367
op eratio n w ill b e perform ed b y th e m ach ine w e g e t b y id e n tify in g th e
p assive (or term inal) sta te of X w ith th e first a c tiv e (or initial) s ta te of
|). T h e o u tp u t of X th u s b ecom es th e in p u t of 9). T h e resu ltin g m ach in e
w e den ote b y “ X f T (Th en ( X $ ) 3 = X(® 8).)
W e use “ Xw” for n > 0 to m ean X . . . X (n facto rs); an d “ X0” for
a m ach in e w hose ta b le h as o n ly 0 as entries, so th a t X°D an d ?)3£° b o th
perform th e sam e operatio n as ?).
W e m a y also w ish th e o u tp u t of one m ach in e X to becom e th e in p u t
of either a m ach in e | or a m ach in e Q, d ep en d in g on som e circu m stan ce
arising durin g th e a c tio n of X. W e can p ro vid e for th is b y e x te n d in g our
n otion of a T u r in g m ach in e to allow tw o term in al sta te s 0^ an d 02. H ere
such 2 -term inal m achines w ill b e used o n ly as com pon en ts in th e co n
stru ctio n u ltim a te ly o f m ach in es as defined in § 67 w ith one term in al
state. U p o n id e n tify in g th e sta te 0± of X w ith th e in itia l sta te of an d
th e s ta te 02 of X w ith th e in itia l sta te of 8> w e o b ta in a m ach ine w h ich
w e d en ote b y “ X j
W e sh all n ow express M achin e 33 as a co m b in atio n of several m achines.

F irst, w e define a m ach in e 93x w h ich perform s th e operatio n illu stra te d
by 1 — 5 or b y 6 — 10 or b y 11 — 15 ; i.e. such th a t, w h en sta rte d
scan n in g a n u m ber in sta n d a rd position, it erases th e scanned square,
p rin ts on th e first b la n k square to th e left, a n d goes to sta te 0X or 02
scan n in g t h a t square, acco rd in g as th e square n e x t left of t h a t is b la n k or
prin ted . T h e le ftm o st square scanned d u rin g th e op eratio n is th e square
n e x t left of th e te rm in a lly scan n ed square. T h is m ach in e h as th e fo llo w in g
ta b le (cf. L in es 1 — 3 of th e ta b le for M achin e 93).
1 — EL2
2 PL3 L2
3 R 01 ro2
T h e op eratio n illu stra te d b y 5 — 6 or 10— 11 is perform ed b y th e

m ach in e w ith th e ta b le (cf. L in e 4 of th e ta b le for M achin e S3):
532 1 L0 Rl
T o perform 15— 16, sim ilarly w e h a v e (cf. L in e 5 for M ach in e S3):
533 1 — ERO
T h e op eratio n illu stra te d b y 16— 17 is perform ed b y M ach in e SI.
368 CO M PU TABLE FUNCTIONS CH. X III
The action of Machine 93 can be described in terms of 31, 93x, 932 and
933, thus. We first use 93x, then according as the terminal state is 0Xor 02,
we use 932 or 933. In the case that 932 is used, its output is fed back into 93x.
In the case that 933 is used, its output is fed into 31. We express this by
the formula
(a)
where the dots express that the output of 932 is fed back as input for 93x.
The notation is suggested by that for repeating decimals. If the operation
performed by 93x each time produced an output with state 0V the oper
ation performed by 93x932 would be repeated ad infinitum. This would
actually happen if Machine 93 were started scanning the leftmost number
on the tape in standard position; Machine 93 would then keep moving
this number square after square to the left.
Formula (a) indicates how to construct the table for Machine 93 from
those for 31, 93x, 932 and 933. In the table for 93x, we replace the terminal
states 0X and 02 by the initial states for 932 and 933, respectively, re
numbering these suitably as 4 and 5; etc. (cf. the table for 93 as first given) ^
Now let us build a machine (for each fixed m > 1) such that:
Machine Qm, when started scanning in standard position the represen
tation of an m-tuple y lf . . . , y m of numbers, with all squares (or at least
the first y 1-j~2 of them) to the right of the m-tuple blank, copies y x
following the m-tuple without a gap, and stops scanning the copy in
standard position. The leftmost square scanned during the action is the
first (blank) square of the given m-tuple.
For example, from the initial situation shown next (Situation 1), Q4
shall reach the situation shown last (Situation 16). A plan for the action
is indicated by the intervening situations shown.
§68 CO M P U T A B IL IT Y OF RECU RSIVE FUNCTIONS 369
1. 0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 0 0 0 0 1
2 . 0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 7 00 0 Jf«J I $ 4
3. 0 1 1 0 I 1 0 1 1 1 1 0 1 1 0 1 00 0 '
4. 0 1 1 0 I2 1 0 1 1 1 1 0 1 1 0 1 00 0 h i |s
5. 0 1T 0 0 1 0 1 1 1 1 0 1 1 0 1 00 0 :
6. 0 1 1 0 0 1 0 1 1 1 1 0 1 1 0 7 00 0 K
7. 0 1 1 0 0 1 0 1 1 1 1 0 1 1 0 1 01 0 JK I 3>4
8. 0 1 I
0 0 1 0 1 1 1 1 0 1 1 0 1 01 0 ' J
9. 0 1 72
0 0 1 0 1 1 1 1 0 1 1 0 1 01 0 Is 1
10. 0 T 0 0 0 1 0 1 1 1 1 0 1 1 0 1 1 0 0 ' (5
11. 0 1 0 0 0 1 0 1 1 1 1 0 1 1 0 1 07 0 il ®‘ j
12 . 0 1 0 0 0 0 1 1 1 10 1
1 1 0 1 1 T 0 hi[s4
13. 0 T0 0 0 1 0 1 1 1 1 0
0 T1 0 0 0 1 0 1 1 1 1 0 1
1 1 0
0
1
1
1 1
1
0 ' J
14. 1 1 0 Ji ® ! ( s
15. 0 1 1 10 1 0 1 1 1 10 1 1 0 1 1 1 0 11 ®4 J
16. 0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 1 1 I 0
M achin e S , acco rd in g to the plan, w hen sta rted scan n in g in stan d ard
position a num ber w h ich is follow ed b y a gap , goes righ t tw o squares,
prin ts an d stops there.
M achine ® , w hen sta rte d scan n in g in stan d ard position a n um ber
w h ich is n o t th e le ftm o st-o n th e tap e, goes left to stan d ard position on
th e n e x t num ber to th e left. H en ce 5)m, w hen started scan n in g in stan d ard
position a num ber to th e left of w h ich at least m num bers occur, goes le ft
passin g o ver m — 1 in terven in g num bers to stan d ard position on th e m -th
n um ber to th e left.
M achin e @, w hen sta rted in stan d ard position on a num ber, d ecides
w h eth er th a t n um ber is 0 or greater th a n 0, an d assum es sta te 0X or 02
accordin g to w h ich is th e case, w ith th e n um ber still scanned in sta n d a rd
position. T h e leftm o st square scanned during th e operatio n is th e square
n e x t left of th e scanned square.
M achin e fj, w hen sta rted scan n in g a p rin ted square, erases an d goes
left one square.
M achin e © goes one n um ber (& m goes m num bers) to th e righ t, ju s t
as % goes one num ber (<
J)m goes m num bers) to th e left.
M achine w hen sta rted scan n in g in stan d ard position a n u m ber
w hich is not th e righ tm o st on th e tap e, fills u p th e g a p (if any) b etw ee n
th a t n um ber an d th e n e x t one to th e right, increasing th e first n um ber b y
the n u m ber of squares th a t co n stitu te d th e gap , and stops scannin g th e
resu ltin g n um ber in stan d ard position.
370 COM PUTABLE FUNCTIONS CH. X III
If w e can find m ach ines ©, <

2), (S, @, § as d escribed , th en M achin e
w ill be g iv e n b y th e form u la
(b) 3 . -
T h e o p eratio n perform ed b y is repeated , so long as th e ap

p lic a tio n of @ g iv e s s ta te 02; b u t w hen it g iv e s 0X, th e a ctio n is term in ated
b y ^)©w.
W e g iv e ta b le s for 3), @ an d ip, le a v in g those for (£, g an d © to th e reader.
1 L2 LI
2 L2 0
© 1 — L2
2 R 01 R 02
$ 1 — R2
2 PR2 L3
3 — EL0
B y several a p p lica tio n s of w ith su itab le v a lu es of m, w e ca n c o p y
w ith o u t ga p s a n y p e rm u ta tio n (allow ing repetitions) of th e num bers
represen ted righ tm o st on th e ta p e w ith o u t ga p s betw een . F o r exam p le,
from th e in itia l situ atio n
0 1 1 1 0 1 0 1 1 1 1 0 1 I 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,
th e m ach in e $•[ ( = (Q4)4) reaches th e term in al situ atio n
0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 1 1 1 0 1 0 1 1 1 1 0 1 T0 0 ,
th e w h ole q u adru p le b ein g co p ied ; an d $ 4^2 reaches
01 1 1 0 1 0 1 1 1 101 101 1 1 0 1 T 0 0 0 0 0 0 0 0 0,
th e first an d fo u rth n um bers b ein g copied.
T h e m ach in e defined th u s
(c)
w hen sta rte d in stan d ard p osition on an m -tu p le y lf ..., y m, w ith all
squares (or a t least th e first y 1+ . . . Ar y mAr 2 m -\- 1) to th e righ t b la n k ,
copies th e w -tu p le to th e righ t after le a v in g a one-square gap , an d stops
scan n in g th e c o p y in sta n d a rd position. T h e leftm o st square scanned
du rin g the' a ctio n is th e first (blank) square of th e g iv e n m -tuple. F o r
e x a m p le , from th e a b o v e in itial situ atio n , reaches th e term in al sit
u atio n
0 1 1 1 0 1 0 1 1 1 1 0 1 1 0 0 1 1 1 0 1 0 1 1 1 1 0 1 7 0.
§68 COMPUTABILITY OF RECURSIVE FUNCTIONS 371
W e n o w w ish a m ach ine 2 w hich, started scannin g a n um ber in sta n d a rd
p o sitio n , w ill erase all num bers (if any) to th e left of this u p to th e first
g a p encountered, an d return to stan d ard position on th e num ber o rigin ally
scanned. T h e leftm o st square scanned in th e operation shall b e th e square
n e x t righ tm o st in th e gro up of co n secu tive b la n k squares c o n stitu tin g th e
gap . F o r exam p le, from th e situ atio n
0 0 1 1 1 0 1 1 1 1 0 1 1 0 1 0 1 1 7 0,
M achin e 2 reaches th e term in al situ atio n
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 T 0.
T h e co n stru ctio n of 2 is left to th e reader.
P roof of T heorem X X V I I I for 1 = 0. T h e proof is b y an in d u ctio n

based on C o ro llary Th eorem X I X § 63. T h ere are six cases, acco rd in g
to w h ich one of th e sch em ata (I) — (V I) is ap p lied last in th e g iv e n
defin ition of 9 b y these sch em ata. F o r each case, w e m u st show h ow to
co n stru ct a m ach ine w h ich 1/1 com p u tes 9 .
T o describe th e in itial situ atio n in the c o m p u ta tio n of <p(xlf . .., x n)
for a g iv e n n -tu p le xv . . ., xn w e now w r it e :
1. x^f
, Xn,
H ere each “x ” stan d s for * * + 1 co n secu tive p rin ted squares, th e com m as
stan d for b la n k s form ing p art of th e represen tation of th e w -tuple, an d
th e b ar in d icates th a t th e represen tation of xn is scanned in stan d ard
position. U n d er th e d efinition of c o m p u ta b ility (§ 67), all o th er squares
to th e left an d righ t are also b la n k. T h e term in al situ atio n , w h ich 9JL
m ust reach if <p(xlf . . .,x„) is defined, w e m a y w rite s im ila r ly :
, Xj , . . . , Xnf (p(xv . . . , #n),
T h e first co m m a here an d in the a b b re v ia tio n for th e in itial situ atio n

(S itu atio n 1 ) refer to th e sam e square of th e tap e, an d here aga in all
squares not in d ic a te d are b la n k (for (iii) § 67). W hen sta rted from
S itu a tio n 1 , 9D?9 shall sto p e v e n tu a lly , o n ly if <?(xv . . . , * „ ) is d efined (for
(iv)). F ro m S itu a tio n 1 , m a y go a rb itra rily far to th e right, b u t m ust
n ever scan a square to th e left of th e one represented here b y th e first
co m m a (for (ii)).
C ase (I) (Schem a ( I ) ) : <p(x) = x'. Let
C ase (II): <?(xv ..., x n) = q. Let
C ase (I I I) : <p(xlf . ..,*„) = Let = S«-/+i*
C ase (I\ ). ty{x\> • • •> x tl) — • • •> • • •» %m{%i> • • *» x.n)').

B y the h yp o th esis of th e in duction , there are m achines 9)?^, . . . , 9Jixm,
w h ich 1/1 co m p u te Xv • • •, Xm> i>, re sp e ctiv e ly . T h e p la n of th e c o m p u ta

tio n o f y {x lf . . x n) is as follow s. S ta r tin g from S itu a tio n 1, w e c o p y th e
w -tuple % , . . . , x n w ith a o ne-square g a p (shown b y th e d o u b le c o m m a ):
2. , x^f . • ., x nt , x , x nt
T h e g a p m ark s th e b eg in n in g of a te m p o ra ry record, to b e erased later.

M ach in e perform s th is c o p y in g operation. M ach in e if sta rte d
from th e situ a tio n , xv ..., x n, , w ill go to th e situ a tio n
, x lt ..., x n, Xi{xv , p ro v id e d Xi(xi< • • • >*») is defined. In

p erform in g th is op eratio n , 9KXl does n o t in a n y in term ed ia te situ a tio n
scan a square to th e le ft o f th e first square in , xv ..., x„,
H e n ce if 9JiXi is s ta rte d from S itu a tio n 2, th e presence o f th e a d d itio n a l
p rin tin g , xv ..., x„, to th e le ft w ill n o t a ffe ct its actio n . So
from S itu a tio n 2, 2RXl w ill go to th e follow in g, p ro v id e d Xi(xi> •••,*„)
is d e fin e d :
3. , Xi, . . ., Xn, , Xx, . . ., Xn, Xl{xl> • • •>x n)>

N e x t we copy xv ..., xn, w ith o u t a g a p , b y M ach in e 3™+ 1, o b ta in in g :
4. ,x 1, . . . , x n, , x 1, . . . , xn, Xiixi> . • •, xn), % ,..., x„,
A p p ly in g W y 2, if Xiix i> . . . , x n) is d efined, w e o b ta in :
5. , Xx, . . ., Xni , Xx, . . ., Xnt Xl(Xl> • ’ * > x n)> X11 • • • f x nt Xz(xl> • • •» x n)i
C o n tin u in g in th is m ann er, w e e v e n tu a lly o b ta in th e fo llo w in g, p ro v id e d

a ll th e fu n c tio n v a lu e s ap p earin g are d efin ed :
6. Xx, . . ., Xn, , XXf . . ., Xnt Xlixl ’ • * •» x n)> ••• i x l> • * *i x n> %m{xl> • • • >x n)>
>
Xl(Xl> • • • >^n)» • • • >Xm(x l> • • • >x n)t <MXl('*‘'l> • • • • x n)> • • • > • • •» x n))t •
B y M ach in e 2 w e c a n th e n erase e v e r y n u m b er b e tw e e n th e scan n ed

n u m b er <Kxi(*i, • • - . *« ), • • - , X«(*i> • • • . *«)) (non-inclusive) a n d th e g a p
(represented b y th e d o u b le co m m a), a fte r w h ich w e ca n close u p th e g a p
b y M ach in e 58 to o b ta in :
7. , x x, . . ., Xn, <J*(Xl(^1'l> • • • ’ •*'«)> • • •> Xm(x l> • • •> Xn))> >i*C.

, xx, ..., xn, <p{xx, . • ., Xn),
T h e entire operatio n , w h en <p(*1( . . . , x n) is defined, is perform ed b y th e
m ach in e defined t h u s :
3K9 = t „ 9 K Xl X + i ^ x 2 • • • 3 n + i ^ x m\J(m-l)(n+1) + 1 S(m-2)(n+1)+2 ’ • •
C o n ve rsely, u sin g th e p ro p e r ty (iv) of th e m ach ines • • •» 50?Xm an<^

90^, M ach in e a p p lied to S itu a tio n 1 w ill e v e n tu a lly stop, o n ly if all
§69 RECURSIVENESS OF COMPUTABLE FUNCTIONS 373
of X l Xl
( > • • • > x n) > * • • > Xm x l
{ > • • •t% n)f 4 K X l( ^ l» • • • t n
% )> • • • > X w (% j • • • >% n ))
are d efined, i.e. o n ly if y {x v ..., x n) is defined.
Ca se (V ): For n > 1 (Schem a (V b)), cp(0, x 2, . . . , x n) ~ ty(xl9 ..., x n),

?(/, , x n) ~ x(y> 9(y> • • • * *«)> • • • >*»)• By th e h y p o th e sis
o f th e in d u ctio n , there are m ach in es 90?^ an d 90?x w h ic h 1/1 c o m p u te
^ an d x> re sp e ctiv e ly . W e ca rry o u t a series of op eratio n s to o b ta in th e
fo llo w in g situ a tio n (corresponding to S itu a tio n 6 for Case (IV )), if th e
required fu n ctio n va lu e s are defined. A t each p la ce w here a ch oice is
in d ic a te d , th e u p p er a lte r n a tiv e ap p lies under th e co n d itio n sta te d , th e
low er o th e r w is e ; i.e. for a g iv e n y , th e low er a lte rn a tiv e is ta k e n on th e
first y choices, an d th e up p er on th e y + l-s t choice.
# y %%2> • • • t * y >X 2> • • •> x nt ty(X 2> • • • > ^ n )>
r ------ , x n), if y = 0.
y ’ X 0, <K*2> • • • » * » ) . • • • . * » » X(°> <K*2> • • •. *»)> *2> • • •. Xn).

_1 f X(0. <M * 2> • • • • *» ). * 2. • • • - Xn), if y — 1 = 0.
,
V i X(^» ^(*^2’ • • •, Xjijt X2, •• -, Xy^)t X21 • • • 1
x 0 » x(0, <M*2> • • •>*»). % • • • . * » ) . * 2> • • •>*»)>
0 ( x ( U x { 0 ,i>{x2 , . . . , x „ ) , X 2 , . . . , x n) , X 2 , . . . , X „ ) , if y —2 = 0.
y “ 2' i 2 .........
T h e n e v e r y th in g b e tw e e n th e la st n u m b er an d th e g a p is erased, an d
th e g a p is closed up. T h e entire op eratio n is perform ed b y th e m ach in e
901^ defined t h u s :
S lig h t ch an ges a d a p t th e tre a tm e n t to th e case n= 1 (Schem a (Va)).
Case (V I): <p(xl9 . . . , x n) ~ w [% {xl9 .,., x n9 y ) = 0] (for n ^ 1; cf.

R e m a r k 1 end § 67). H a n d le d like C ase (V) (but m ore sim ply).
P r o o f o f T h e o r e m X X V III f o r l> 0. W e n o w h a v e as an ad d itio n a l

case th a t <p is one, s a y tyit of th e assu m ed fu n ctio n s Y . In th is case <p is
1/1 co m p u te d from Y b y th e m ach in e 90^ h a v in g a sin gle a c tiv e sta te
q x w ith th e ta b le e n tr y iq 0.
§ 69. Recursiveness of computable functions. In th is section th e
n o ta tio n s w ill b e th o se o f § 67 ra th er th a n § 68.
T h eo r em X X IX . E very computable partial function <p is partial

recursive. E very partial function 9 computable from l com pletely defined

functions Y is partial recursive in Y .
P roof for 1 = 0. I t is possible to represent th e ta p e vs. m ach in e
situ atio n s of a g iv e n m ach in e SH b y form al expressions (as ind eed w e
sh all do for an o th er purpose in § 7 1 , w here th e expressions w ill b e ca lled
“ P o s t w o rd s” ). T h e theorem ca n th en b e considered as an a p p lica tio n
o f (D 2 ) § 6 2 (exten d ed to p a rtia l fu n ctio n s; here E x?.+1 depend s o n ly
on E xj).
I f d eta ils are to b e su pplied, th e fo llo w in g proof is a little m ore direct.
G 5 d e l num bers are assigned d ire c tly to th e ta p e vs. m ach in e situ a tio n s;
an d in stea d of g iv in g G d d el num bers to d ed u ctio n s or proofs (as for
T h eo rem I X § 58 an d ( D l) § 62), w e sim p ly n um ber th e successive situ
ation s reached from a g iv e n situ a tio n b y a g iv e n m ach in e as th e 0-th ,
1 -st, . . . , z - t h , ___T h e n if one w ill read “ fu n c tio n ” for “ p a rtia l fu n c tio n ”
an d “ gen eral re cu rsive” for “ p a rtia l recu rsive” , he need n o t use C h a p te r
X I I . T h e p roof can b e sh ortened slig h tly b y usin g T h eo rem X X V I § 66 .
In esta b lish in g th e G o d e l num berin g of situ ation s, w e describe th e
co n d itio n o f th e ta p e o n ly re la tiv e to th e scanned square, in stea d of
re la tiv e to som e fix e d square. T h is suffices for our purpose here, since
ab so lu te position on th e ta p e does n o t enter in to th e d efin itio n of co m
p u ta b ility . (For 1 js c o m p u ta b ility , ab so lu te position or a t least p o sitio n
re la tiv e to th e first square of th e represen tation of xv ..., xn in th e
in itia l situ a tio n does m a tte r, b y (ii) an d (iii).)
W e b egin b y assign in g a G o d el num ber to th e (condition of the) ta p e
to th e left of a n y p articu lar square. L e t th e con d ition s of th e su ccessive
squares to th e left of th a t square b e sWo, su , s M2, ___Sin ce o n ly a fin ite
n um ber of squares are p rin ted , all b u t a fin ite num ber of th e su b scrip ts
Ui are 0 . T h e G o d el num ber u of th e ta p e to th e le ft of th e square sh all b e
II pf*f i.e. II pv* where t is a n y n um ber w h ich exceed s e v e r y i for w h ich
i i< t
Ui J=. 0. Then u itself is such a n u m b er; so 11 = II pf* ---■ H p {^ K
i<u i<u
A G o d el n um ber is assigned sim ilarly to represent th e co n d itio n of th e
ta p e to the righ t of a g iv e n square.
If in a g iv e n situ atio n , th e Godel n um ber of th e ta p e to th e left of th e
scanned square is n, th e scanned sym b o l is s a, th e m ach ine sta te is q Cf
and th e Godel n um ber of th e ta p e to th e righ t of th e scanned square is
v, then the Gddel n um ber of the situ atio n sh all b e 2 M*3r**5c * 7 r.
§69 RECU R SIV EN ESS OF CO M PU TABLE FUNCTIONS 375
E xample 1. The Godel number of the situation

q4
Si Si S3 Si So Si
w here all squares n o t show n are b la n k is 2 5l-3l -23 *3 1 - S 4 * ? 20*31.
I f th e G o d el n u m ber of th e situ a tio n is 2U• 3 ° • 5 C• 7V, th e co n fig u ra tio n is

(sa, q c). I f c> 0 , an d th e a c t w h ic h th e m ach in e perform s from th is
co n figu ratio n is of th e form sbL q d, th e G o d el n u m b er p («, v) of th e re
su ltin g situ atio n is
[2 e x p n p ^ i ' ] . 3 (w)o • 5d • [7 exp {2b • U p f i } ] ;

i< u i< v
an d sim ilarly if th e a c t is sbR q d. I f th e a c t is sbC q d, th e n po c (w, v) is

2 u . 36 . 5 d . 7 « E a c h of th ese fu n ctio n s pac is p rim itiv e recursive.
N o w p defin ed as follow s is p rim itiv e recu rsive ( # F § 45), a n d h as th e
p ro p e rty th a t, if w is th e G o d el n u m b er of an a c tiv e (passive) situ a tio n ,
then p(w) is th e G o d el n um ber of th e n e x t (same) situ a tio n :
P a .c (M . 0 Ms) if Ml = « & Ms = C
pM (a = 0, c= 1,
w otherw ise.
T h e n c e w e define a p rim itiv e recursive fu n ctio n 0 th u s:
f 0 (^, 0 ) = w,
\ Q(w, z') = p(0(^, z)).
If w is th e G o d el n um ber of a n y situ atio n , th en §(w, z) is th e G o d e l n u m b er
of th e situ a tio n after th e n e x t z acts, if th e m ach in e perform s a t le a st z
a cts from th e g iv e n situ a tio n ; an d Q(w} z) is th e G d d el n u m b er o f t h e te r
m in al situ a tio n reach ed from th e g iv e n situ atio n , if th e m ach in e perform s
< £ a c ts from th e g iv e n situ atio n .
N e x t, for each n > 1 (cf. R e m a r k 1 en d § 67), w e define a p rim itiv e
recursive fu n ctio n x n w ith th e fo llo w in g p ro p erty . I f th e w -tuple xv ..., xn
is scanned in stan d ard position, th e s ta te is q c, th e G o d el n u m b er of th e
ta p e to +he le ft o f th e re p resen ta tio n o f xv
x n is u t an d th e G o d e l
...,
n um ber of th e ta p e to th e rig h t o f t h a t re p resen ta tio n is v, th e n r n(xv . . . ,
x n, c} u , v) is th e G o d el n u m ber of th e situ atio n . F ir s t w e d efine th u s :
î{xv c, u, v) =
[2 e x p { ( n Pl)-p°x i- n p (^ + i} y 3 l -5 H 7 e x P { 2 °- n p™*}].
3 76 COMPUTABLE FUNCTIONS CH. XIII
T h e n , for n = 1, 2, 3, . . . :
Tn+i(*i, • • • >%n> ^n+l» G v)
W hen xv . . xn
niXn+V G (T n fe, •• *n-l, *n> G
is scan n ed in stan d ard position w ith sta te
V))o> »)•
q lt an d
th e ta p e is b la n k elsew here, th e G o d el num ber of th e situ a tio n is
x n(xv . . x n, 1 , 1 , 1 ). W h e n % , . . x n, x is scanned in stan d ard p osition
w ith s ta te q 0, th e G o d e l n u m b er of th e s itu a tio n is xn+1(xv ..., x n, x, 0, u, v)
for som e u an d v ; an d co n versely.
N o w , if 9 is th e p a rtia l fu n ctio n of n va riab les co m p u te d b y th e g iv e n
m a ch in e 501, an d xv ..., x n is a g iv e n n -tu p le, th en cp(xlt . . . , x n) is defined,
if a n d o n ly if th ere ex ists a q u a d ru p le (z, x } u, v) of num bers such th a t
0(Tn(*!, = x n+i(x i> . . . , x n,x , 0, u, v), in w h ich case x
is th e v a lu e of <p{xlf . . . , x n). A c c o rd in g ly ,
?(*i, • • • >*») ~
((ji[0(Tn(% , . . x ni 1, 1, 1), (t)0) = xn+1(xlt . . . , x nt (t)l9 0, (t)2, (O3)])l*
Th erefo re, b y T h eo rem X V III § 63, 9 is p a rtia l recursive (or if 9 is

c o m p le te ly defined, b y T h eo rem I I I § 57, 9 is gen eral recursive).
P roof for l > 0. S a y e.g. there is one assum ed fu n ctio n <[/ of one
v a ria b le (i.e. I = m1 = 1). N o w , for each c for w h ich th e ta b le e n tr y
correspon din g to q c is of th e form \q d, w e replace “ pfl,c(Wo> W s ) if
(w)1= a & (w)2= c (a = 0 , . . . , / ) ' ' in th e d efin ition of p(w) b y “ p c(w)
if Q c(w )” , w here Q c is th e p rim itiv e recursive p red icate an d pc th e fu n ctio n
p rim itiv e recursive in ^ d efin ed t h u s :
Qc(w) = ( £ y ) 1/< „,(£ :w ) u < J ^ ) r < « ,C iK' = Tl(y - C> U’ V)],

Pc(w) = Ta( Y , + (Y ), d, U, V) where
y = v-yv<w{Eu)u<w{Ev)v<w\-w=î{y> c>u >»)].
U = y-uu<J E y ) y<w (Ev)v<w[w = T 1{ y ,c ,u ,v ) ] , etc.
T heorem XXX (= The following

T h eo rem s X X V I I I + X X IX ).
classes of partial functions are coextensive , i.e. have the same m em bers :
(a) the partial recursive functions , (b) the computable functions , (c) the 1/1
com putable functions. S im ilarly with l completely defined assum ed functions
Y.
§ 70. T u r i n g ’ s t h e s i s . T u r in g 's thesis th a t e v e r y fu n ctio n w h ich
w o u ld n a tu r a lly be regard ed as co m p u ta b le is co m p u ta b le under his
defin itio n , i.e. b y one of his m ach ines, is eq u iv a le n t to C h u rch 's thesis b y
T h eo rem X X X . W e sh all n o w e x a m in e th e p a rt of th e evid en ce for it
§70 Turing ’s thesis 377
w h ich pertain s to th e m ach in e co n cep t, i.e. w h a t w e liste d as (C) in § 62.

W h a t w e m u st do is to co n vin c e ourselves t h a t a n y a c ts a h u m a n co m
p u te r co u ld ca rry o u t are a n a ly z a b le in to successions of a to m ic a c ts o f
som e T u r in g m achine.
" T h e b e h a v io r of th e co m p u ter a t a n y m o m en t is d eterm in ed b y th e
sym b o ls w h ich he is ob servin g, an d his 's ta te of m in d ’ a t th a t m o m e n t.”
T h e n u m ber of sy m b o ls w h ich he ca n recognize is fin ite. " I f w e w ere to
allow an in fin ity of sym b o ls, th e n there w o u ld b e sy m b o ls d ifferin g to
an a rb itra rily sm all e x t e n t .” (T u rin g 1936-7 pp . 249— 250.) T h e w o rk
lead in g from th e problem sta te m e n t to th e an sw er m u st b e carried o u t
in som e " s y m b o l sp a c e” (P ost 1936 ), i.e. som e s y s te m a tic arran gem en t of
cells or bo xes, each o f w h ich m a y b ear (an occurrence of) a sy m b o l.
T h ere is a fin ite b o u n d to th e n u m ber of occurrences o f sy m b o ls (or o f
b o xes w here a sy m b o l m a y occur) w h ich he ca n ob serve a t one m o m en t.
H e can also rem em ber sy m b o ls p re v io u sly ob served , b y a lte rin g his s ta te
of m ind. H o w e v e r " th e n u m ber of sta te s of m in d w h ich need to b e
ta k e n in to acco u n t is fin ite. . . . I f w e a d m itte d an in fin ity of sta te s of
m ind, som e of th e m w ill b e 'a rb itra rily clo se’ an d w ill b e co n fu se d .”
(T u rin g 193^-7 p. 250.) B u t th e c o m p u te r’s a c tio n m u st lead from a q u ite
d iscrete o b je c t, n a m e ly th e sy m b o l a rra y represen tin g som e n a tu r a l
n um ber (or n -tu p le of n a tu ra l num bers) as argu m en t (s), to a n o th er su ch
o b je c t, n a m e ly th e s y m b o l a r ra y represen tin g th e correspon ding fu n c tio n
va lu e. T h e possible sta te s of m in d are fix e d in a d v a n c e of n a m in g th e
p articu lar a rgu m en t (s), as w e are con sid erin g c o m p u ta tio n b y a preas
sign ed m eth o d , an d do n o t allo w m a th e m a tic a l in v e n tio n in th e m id st
of th e co m p u te r’s perform ance. E a c h a c t he perform s m u st c o n s titu te a
d iscrete ch an ge in th e fin ite sy ste m co n sistin g o f th e occurrences of
sy m b o ls in th e s y m b o l space, th e d istrib u tio n o f o b served squares in th is
space, a n d his s ta te of m ind.
T h ese lim ita tio n s on th e b e h a v io r of th e h u m a n co m p u te r in c o m p u tin g
th e v a lu e of a n u m b e r-th e o re tic fu n ctio n for g iv e n argu m en ts, b y fo llo w
in g o n ly preassigned rules, are of th e sam e k in d as en ter in th e co n
stru ctio n of a T u r in g m ach ine. T h e ta p e is th e sy m b o l space for th e m a
chine, an d th e m ach in e s ta te corresponds to th e co m p u te r’s sta te of m ind.
T h e h u m a n co m p u te r is less restricted in b e h a v io r th a n th e m ach in e,
as follow s: (a) H e ca n ob serve m ore th a n one s y m b o l occurrence a t a
tim e, (b) H e can perform m ore co m p lic a te d ato m ic a c ts th a n th e m ach in e,
(c) H is s y m b o l space need n o t be a one-d im ension al tap e, (d) H e ca n
choose som e oth er sy m b o lic represen tation of th e a rgu m en ts an d fu n ctio n
v a lu e s th a n th a t used in our d efinition of c o m p u ta b ility .
W e sh all exam in e va rio u s possibilities under (a) — • (d), an d see b rie fly
h o w each ca n b e reduced to an e q u iv a le n t in term s of T u rin g m achines.
W e sh all u s u a lly sp eak as th o u g h o n ly one were b ein g reduced, b u t our
m eth o d s w o u ld serve to reduce a n y co m b in atio n of th em su ccessively.
U n d e r (a), w e rem ark th a t e.g. 17 an d 21 an d 100 can each b e o b
served in a single ac t. B u t a lon g sequence of sy m b o ls can o n ly b e
o b served b y a succession of acts. F o r exam p le, w e ca n n o t tell a t a glan ce
w h e th e r 157767733443477 and 157767733443477 are th e sam e; “ we
sh o u ld h a v e to com pare th e tw o num bers figure b y figure, p o ssib ly
tic k in g th e figures o ff in pen cil to m ake sure of their n o t b ein g co u n ted
tw ic e .” (T u rin g 1936-7 p. 251.)
I f 17 an d 21 an d 100 are n o t o n ly ob served as u n its b u t m a n ip u la te d
as th o u g h each occupies a single cell of th e sy m b o l space, w e need o n ly
red efin e th e sy m b o ls so t h a t each of these co n stitu te s a single sy m b o l,
in order to reduce th e co m p o u n d o b se rv a tio n to a sim ple o b serva tio n of
th e k in d u sed b y a T u r in g m achine.
In a c tu a l co m p u tin g w e som etim es use certa in m ark s (accent, ch eck ,
m o v a b le p h y sic a l poin ter, etc.), w h ic h m a y b e p la ced on a g iv e n square
in a d d itio n to an o rd in ary sym b o l. I f there are j of th e o rd in ary sy m b o ls,
and n of th ese special m arks, a n y subset of w h ich m a y b e p la c e d on a
g iv e n square, th e n u m b er of th e square con dition s is m erely increased
from 7 + 1 to ( ; + 1 ) • 2 *.
As a n o th er e x a m p le of b e h a v io r in v o lv in g co m p o u n d o b servatio n ,
su ppose th a t th e fo llo w in g sequence of sy m b o ls is p rin ted ,
. . .4 4 0 1 3 8 5 7 8 9 2 6 4 ...,
th a t th e ob server's a tte n tio n is cen tered a t th e figure 7 near th e m idd le,

a n d t h a t h e observes c le a r ly a t m o st fiv e figures cen tered a t th is 7 ; th u s
th e sequence o f th e fiv e d ig its 85789 to geth er w ith his sta te o f m in d
d eterm in e his n e x t a c t. Som e d ig its further off m a y b e v a g u e ly ob served ,
b u t w ith o u t a ffe c tin g his a c t. T h e a c t sh all b e of one of th e k in d s p er
form ed b y T u r in g m ach ines, w ith each sep arate sy m b o l occurrence (not
gro u p s of five) o c c u p y in g a square. F o r ex a m p le , if th e n e x t a c t is 0 L q d,
th e p rin tin g becom es
...4 4 0 1 3 8 5 0 8 9 2 6 4 ...,
w ith 38508 observed . S u ch b e h a v io r ca n b e reduced to T u r in g m ach in e

b e h a v io r as follow s. S a y t h a t th e sy m b o ls are th e te n A r a b ic d igits.
T h e b e h a v io r ca n b e considered as th a t of a gen eralized T u r in g m ach ine,
in w h ic h th e co n figu ratio n (determ ining th e act) is (e, /, a, g, h , q c)
w here e, /, a, g} h are th e d igits o c c u p y in g th e fiv e squares cen tered a t
th e scanned square. C orresponding to each sta te q c o f th is g e n e ra liz e d
m ach in e, w e in tro d u ce a set of 104 sta te s q cefgh (e, /, g, h = 0 , . . . , 9 ); an d
w e m o d ify th e ta b le so th a t upon reach ing sta te q c, a series of T u r in g
m ach in e a c ts is perform ed, co n sistin g of in spection s of th e tw o a d ja c e n t
squares on each side, lead in g to sta te q cefgh w hen th e four squ ares in
q u e s tio n are occu p ied b y th e re sp ective d igits e, /, g, h. N o t o n ly th e
sta te s q cefgh m u st b e ad d ed , b u t also som e states to be assum ed d u rin g
th e ac tio n lead in g from q c to q cefgh. D e ta ils are left to th e reader. N o w
th e act th e gen eralized m ach ine perform ed from th e co n fig u ra tio n
(e, /, a, g, h, q c) shall b e perform ed from th e co n figu ratio n (a, q cefgh).
T h is red u ction is an illu stratio n of th e rem ark th a t one ca n re m em b er
a fin ite num ber of p re v io u sly o b served sy m b o ls b y h a v in g c h a n g e d o n e ’s
sta te o f m ind w hen th e y were observed.
I t m ig h t be th o u g h t th a t th e p rin tin g on still o th er squ ares m a y
co n stitu te p art of th e o b servatio n , e.g. th a t on certa in s p e c ia lly m a rk ed
squares (finite in num ber). I f these squares are so lo c a te d in th e sy m b o l
space th a t th e co m p u ter can fin d th em an d return b y a c ts o f th e kin d s
perform ed b y T u rin g m ach ines (cf. th e discussion o f (c) to follow ), th is
kin d of co m p o u n d o b servatio n ca n b e reduced in a sim ilar fash ion to th e
preceding.
U n d er (b), th e co m p u ter ca n alter other squares besides th e sca n n ed
square. T h e new ob served square need n o t b e a d ja c e n t to th e original.
H o w e v e r there is a fin ite b o u n d to th e c o m p le x ity of th e a c t, if it is to
c o n s titu te a sin gle a c t of th e com pu ter. M ore c o m p lic a te d a c ts w ill require
renew ed m o tiv a tio n b y reference to th e o b served d a ta an d th e s ta te
o f m ind a t in term ed iate situ atio n s b etw een th e g iv e n an d re su ltin g ones.
(Ind eed it ca n be argued th a t th e T u rin g m ach in e a c t is a lre a d y co m p o u n d ;
a n d consists p s y c h o lo g ic a lly in a p rin tin g an d ch an ge in s ta te o f m in d ,
fo llo w ed b y a m o tio n an d an oth er ch an ge of m ind. P o s t 19 4 7 does th u s
sep arate th e T u rin g a c t in to tw o ; w e h a v e n o t here, p rim a r ily b eca u se it
sa v e s space in th e m ach in e ta b le s n o t to d o so.)
A ll sim ple alte ratio n s of th e situ atio n , n o t in th e T u r in g m ach in e form ,
w hich are re a d ily proposed, e.g. p rin tin g a fter m o tio n in stea d o f before,
are ea s ily expressed as successions of th e ato m ic a c ts of a T u r in g m ach in e.
(M uch m ore co m p lic a te d operations, w h ich co u ld h a rd ly b e regard ed as
sin gle acts, h a v e a lre a d y been so tre a te d in § 68 .)
T u r n in g to (c), co m p u tin g is co m m o n ly perform ed on 2 -d im en sio n al
paper, an d th e 2 -dim ensional ch aracter of th e p aper is so m etim es u sed
in ele m e n ta ry arith m etic. T h e o r e tic a lly , w e m u st also consider th e p o s
s ib ility of still oth er kinds of sy m b o l space. T h e sy m b o l space m u st b e
s u ffic ie n tly regular in stru ctu re so th a t th e co m p u te r w ill n o t beco m e lo st
in it d u rin g th e co m p u ta tio n .
F ro m a g iv e n square or cell of th e space, there w ill b e a fin ite n u m b er
m -\ -1 of w a y s of m o v in g to th e sam e or an a d ja c e n t cell, ca ll th e m
M 0, . . . , M mw here M 0 is th e id en tica l m otion. F o r exam p le , in th e p la n e
ru led in to squares, m = 4 (no m otion , le ft, up, righ t, dow n), or if d ia g o n a l
m o tio n s are also allow ed, m = 8. T h e com p u ter, w hose a c t from a g iv e n
situ a tio n m u st b e determ in ed b y w h ich one of a fin ite n u m ber of co n
fig u ra tio n s is e x istin g, co u ld n o t use m ore. W e lose no g e n e ra lity in
su p p o sin g t h a t there are th e sam e n u m ber of direction s of m o tio n from
e v e r y ce ll; in case there are few er from som e cells, th e term inus of th e rest
o f th e m -\ -1 m o tio n s m a y b e d efin ed to b e th e g iv e n cell, i.e. th ese as w e ll
as M q m ay b e ta k e n to be id en tical.
T h e n u m b er of cells w h ich ca n u ltim a te ly b e reach ed is th erefore
co u n ta b le . T h e sam e cell m a y b e reach ed b y d ifferen t successions o f
m o tio n s, e.g. in th e plan e, d ow n an d th en righ t leads to th e sam e sq u are
as righ t an d th en dow n.
W e sh all suppose t h a t an en u m eration w ith o u t repetitio n s ca n b e
g iv e n of all th e cells, such th a t th e fo llo w in g is th e case. T o each o f th e
w a y s of m o v in g Mt (i j= 0, . . . , m), there is a co m p u ta b le fu n c tio n [xt*
su ch th a t, if x is th e in d e x in th e enu m eration o f th e g iv e n cell, th e n
[Li(x) is th e in d e x of th e cell reach ed b y th e m o tio n M{. T h is su p p o sitio n
is realized b y a n y re a d ily im agin e d sy m b o l space.
U s in g th is en u m eration , let th e cell num bered x in th e en u m eratio n
(x = 0, 1 , 2 , . . . ) correspond to th e x -th square c o u n tin g rig h tw a rd from
a ce rta in square (called th e 0-th) on a linear tap e.
U s in g m eth o d s from § 68, w e ca n set u p a T u r in g m ach in e w h ich w ill
fin d th e |xt(^)-th square, w h en sta rte d on th e x -th square, if a d is
tin g u ish in g m ark is k e p t on th e 0 -th (or — 1-st) square. T h e c o m p u ta tio n
for th is purpose can b e done b y m ark in g squares w ith accen ts, a fterw ard s
erased, w ith o u t in terferin g w ith th e p rin tin g a lre a d y on th em . T h is
en ables us to reduce c o m p u ta tio n in th e g iv e n sy m b o l space to co m
p u ta tio n on th e linear ta p e of a T u r in g m achine.
F o r th is red u ctio n , w e d id n o t assum e th a t from a n y ce ll a d ja c e n t to
a g iv e n cell one of th e m o tio n s returns us to th e g iv e n cell, i.e. th a t e v e r y
m o tio n in th e space has an inverse. T h is w o u ld b e th e case in a n y or
d in a r y s y m b o l space. A n e x c e p tio n is represen ted b y th e co m p u te r w h o
receives a sign al a t in te rv a ls b y ear.
The s y m b o l space m ay con sist of several d isco n n ected su bsp aces,
each h a v in g its ow n scan n ed cell, as e.g. in th e case of a co m p u te r w h o
sim u lta n e o u sly reads a sy m b o l on a p ap er b y ey e, read s a n o th er in b raille
on a ta p e b y h an d, an d receives a sign al b y ear. I f th ere are r su ch
subspaces, w e can reconstrue th e cells to b e th e r-tu p le s co n sistin g
o f a cell from each of th o se re sp e ctiv e su bspaces.
In regard to (d), w e m a y argu e th a t a n a tu ra l n u m b er y is g iv e n in th e
original sense ( § 6 ), o n ly if som e seq uen ce o f y + 1 o b je c ts, s a y y+1
tallies, is g iv e n ; an d h en ce t h a t a procedure for c o m p u tin g a fu n c tio n 9
from its a rgu m en t (s), w h en b o th are expressed in som e o th er n o ta tio n ,
w o u ld n o t so lve th e c o m p u ta tio n p roblem for 9 , unless th e c o m p u te r
ca n also proceed from th e o th er n o ta tio n for a n u m b er y to th e seq u en ce
of y+1 tallies, an d v ic e versa.
A c c o rd in g to our o th er argu m en ts, T u r in g m ach in es co u ld th e n b e
b u ilt w h ich , g iv e n th e o th er n o ta tio n for y w o u ld s u p p ly th e y-f 1 tallies,
an d v ic e versa. D e ta ils ca n b e arran ged as in th e d efin itio n o f c o m p u
ta tio n w ith in one sy ste m of n o tatio n . T h u s for d ecim al n o ta tio n , th e
first m ach in e sta rte d in th e first of th e fo llo w in g situ a tio n s w o u ld g o
to th e second.
qi
1 2
q<>
1 2 1 1 1 1 1 1 1 1 1 1 1 1 1
F o r th e fam iliar sy ste m s o f n o ta tio n , such as th e d u a l or d e cim a l, th e

e x isten c e o f such a p air of m ach in es ca n b e established .
W e h a v e been d efen d in g T u r in g ’s thesis for n u m b er-th eo retic fu n c tio n s ;

b u t T u r in g m ach in es a p p ly e q u a lly w ell to expressions in a n y la n g u a g e
h a v in g a fin ite list of sym b o ls. B y u sin g th em as ju s t illu stra te d for th e
case of co n v e rtin g one n o ta tio n for a n a tu ra l n u m b er in to a n o th er, w e
g e t a d irect w a y o f ch ara cte riz in g ‘e ffe c tiv e ’ op eratio n s on exp ression s
in such lan gu age s, as an a lte r n a tiv e to requ irin g a co rrespon din g n u m b e r-
th eo retic fu n ctio n u nder a p a rticu la r e ffe c tiv e G o d e l n u m b erin g to b e
gen eral recu rsive or c o m p u ta b le ( § 6 1 ). T h e m e th o d e x te n d s to la n g u a g e s
h a v in g an en u m erable in fin ity o f sy m b o ls, w h e n eve r th e sy m b o ls ca n
b e con sid ered e ffe c tiv e ly as com p osed in tu rn from th e sy m b o ls o f som e
fin ite lis t; e.g. to th e form al n u m b er-th eo retic sy m b o lism , b y re ga rd in g
th e va ria b le s a, b, c, ... as a, a [t a Hf ... (§§ 16, 50).
*§ 71. The word problem for semi-groups. Church's original
example of a decision problem which, on the basis of his thesis, is un-
solvable was constructed in terms of X-definability (1 9 3 6 ). The corre
sponding example given above (Theorem XII § 60) is constructed in terms
of general recursiveness (following Kleene 1 9 3 6 , 1 9 4 3 ). Turing 1 9 3 6 - 7 gave
examples constructed in terms of computability. This can be done e.g.
as follows. A machine -JJi is determined by its table, containing (/+ 1 )&
entries. We can set up a Godel numbering of the tables, and thus of the
machines. Then the function £ defined as follows is not computable:
%(x) = 0, ii x is the Godel number of a machine Wtx, and the partial
function <px of one variable which computes has 1 as value for %as
argument, and %(x) = 1 otherwise. So by Turing's thesis (or via the
equivalence of general recursiveness and computability, by Church's
thesis) there is no algorithm for deciding whether any given number #
is the Godel number of a machine which, when started scanning x in
standard position with the tape elsewhere blank, eventually stops
scanning *, 1 in standard position. As another example, there is no al
gorithm for deciding whether any given machine, when started from any
given initial situation, eventually stops. For if there were, then, given any
number #, we could first decide whether ^ is the Godel number of a
machine and if so whether started scanning ^ in standard position
with the tape elsewhere blank eventually stops, and if so finally whether
x, 1 is scanned in standard position in the terminal situation.
These first examples of decision problems proved unsolvable are
problems arising directly in connection with one of the mathematical
notions (X-definability, general recursiveness, or computability) originally
identified with effectiveness by the Church-Turing thesis.
A second class of examples, a step removed from these, are the
decision problems for certain formal systems, e.g. Theorem 33 § 61
(also cf. § 76, Church 1 9 3 6 p. 363).
A decision problem proved unsolvable by Post 1 9 4 7 and Markov 1 9 4 7
is of interest as constituting the first example in which an existing problem
from outside the field of logic and foundations has been treated.
The problem which Post and Markov prove unsolvable was proposed
by Thue 1 9 1 4 . Suppose that a finite list alt . .., a-m (m > 1 ) of distinct
symbols is given; let us call them letters, and the list of them the alphabet.
A finite sequence of zero or more (occurrences of) the letters, we call a
word (in, or formed from, that alphabet); in the language of Chapter IV
(§ 16), a word is simply a formal expression, when a1? ..., am are the
formal symbols, except that now we always include the empty expression.
§71 THE W ORD PROBLEM FOR SEM I-GROUPS 383
A w ord C is part of an oth er w ord D , if D is of th e form U C V w here U

a n d V are w ords (possibly e m p ty ).
N o w suppose further th a t a fin ite list (A x, B x), . . . , ( A n, B n) (n > 1)
of pairs of w ords is g iv e n ; w e call th is list the dictionary . W e s a y th a t
tw o w ords R an d S are im m ediately equivalent (by th e g iv e n d ictio n ary),
if R an d S are of th e resp ective form s U A fV an d U B tV , or of th e resp ective
form s U B * V an d U A tV , for som e w ords U an d V an d p air (A t-, B ,)
(1 < i < n ) ; in oth er w ords, if R is tran sform able in to S b y replacin g a
p art A i b y its correspondent B* in th e d ic tio n a r y , or in versely. W e
c a ll tw o w ords P an d Q equivalent (by th e g iv e n d ictio n ary), if there is a
fin ite sequence R 1? . . . , R* (/ ;> 1) of w ords such th a t R x is P , an d R z is Q ,
an d R*_x is im m e d ia te ly e q u iv a le n t to R f (t = 2, . . . , / ) .
T h u e ’s (general) problem is to find an algorith m for d ecid in g, for a n y
g iv e n a lp h a b e t an d d ictio n a ry, w h eth er a n y tw o g iv e n w ords are e q u iv
alent. T h e problem is also kn ow n as th e word problem for sem i-groups.
T heorem X X X I . The word problem for sem i-groups is unsolvable;
in fact, there is a p a rtim la r alphabet and dictionary, such that there is no
algorithm for deciding whether any two words (formed from that alphabet)
are equivalent (by that diction ary) . (Post , M arkov
1947 .) 1947
P roof . O u r m eth o d of proof w ill con sist in p ick in g an a lp h a b e t an d

a d ictio n a ry, such th a t, if w e h ad a decision procedure for th e eq u ivalen ce
of a n y tw o w ords, w e could th en ce o b ta in a d ecision procedure for th e
p red icate (E'y )T x(x, x, y), co n tra d ictin g T h eorem X II §60 (based on
C h u rch ’s thesis). T h e u n s o lv a b ility of T h u e ’s problem for th e case of th is
p articu lar a lp h a b et an d d ictio n a ry of course im plies th e u n s o lv a b ility of
th e gen eral problem .
I t is co n ven ien t now to consider th e eq u ivalen ce relation sh ip b etw ee n
tw o w ords in term s of th e p o s sib ility of fo rm ally d ed u cin g th e second
w ord from th e first b y use of 2n rules of inference, as follow s (i = 1, . . . , n ):
U A tV U B ,V
(a) (b)
U B .V , U A .V ,
w here U an d V are a n y w ords (possibly em p ty ) in th e a lp h a b e t a x, . . . , a m.

W e shall first s tu d y sem i-Thue system s in w h ich o n ly th e n rules of
inference (a) are ad m itte d , b u t n o t their inverses (b).
W e shall describe a m eth o d b y w hich, g iv e n a n y p a rticu la r T u rin g
m ach in e, w e can set u p a sem i-T h u e sy ste m such th a t th e ta p e vs.
m ach in e situ atio n s are represented b y w ords of th e sem i-T h u e sy ste m ,
an d th e ato m ic a c ts of th e m ach in e correspond to ap p licatio n s of th e
rules (a).
The a lp h a b e t of th e se m i-T h u e sy ste m sh all con sist of s0, . . . , s,
(representing th e square co n d itio n s for th e m ach ine), q 0, . . q k (rep
resen tin g th e m ach in e s ta te s ), an d an a d d itio n a l sy m b o l h (j+ k + 3
letters altogeth er).
In a g iv e n ta p e vs. m ach in e situ atio n , s a y th a t th e sm allest u n b ro ken
piece of ta p e co n ta in in g a ll p rin ted squares an d th e scanned square consists
of r squares (r > 1 ). L e t th e con dition s of these squares from left to
righ t b e s^, . . . , s^. L e t th e p-th. of these squares b e th e scanned square
(1 < p <L r). L e t th e m ach in e sta te b e q c. T h e n th e situ a tio n sh all be
represented b y th e w ord
• • • s ^ q c S ^ +1 . . . s ^ ,
w h ich w e ca ll th e P ost word for it. (W hen is a w ord a P o st w ord ?)
E xam ple 1. The P o s t w ord for th e situ a tio n of E x a m p le 1 § 69

is h s 1s 1s 3s 1q 4s 0s 1h.
The n rules of inference (a) of th e sem i-T h u e sy ste m sh all com prise
one or m ore rules correspon ding to each of th e ( / + \)k a c tiv e co n figu
ra tio n s of th e T u r in g m achine. I f th e ta b le e n tr y for an a c tiv e co n
figu ra tio n (sa, q c) is of th e form sbZ,qd w ith b+ 0 , there sh all b e j+ 2
corresponding rules as follow s (e = 0, . . . , / ) ,
U s gs 0q cV U h s 0q cV
U s eq ds 6V , U h soq ^ V ;
if th e ta b le e n try is s 0L q d, there shall be ( j+ 2 ) 2 rules as follow s

(e>/ = o , ...,/),
Us^SqqeS/V U h s aq cs/V U s es aq ch V U h s aq ch V
U s eq ds0s ,V , U h s 0q ds0s ,V , U s eq dh V , U h s 0q dh V .
S im ila rly there shall b e j + 2 rules, if th e ta b le e n tr y is of th e form sbR q d

w ith b+ 0 ; an d ( j+ 2 ) 2, if of th e form s02?qd. I f th e e n tr y is of th e form
sbC q d, there sh all be one rule, as follow s,
U s aq cV
U s bq dV .
In these n rules th e A / s are all d istin ct. G iv e n th e P o st w ord for an
a c tiv e situ atio n , e x a c tly one of these n rules (a) is ap p licab le to it as
prem ise, n a m e ly th e rule or one of th e rules corresponding to th e con
figu ra tio n (s'a, q c) in th a t situ atio n , an d in o n ly one w a y , i.e. w ith o n ly
one ch oice o f th e U an d V . (For a rule in w h ich th e first sy m b o l of th e
A t- is h, th e U w ill a lw a y s b e em p ty .) T h e ap p lica tio n of th e rule g iv e s as
co n clu sio n th e P o st w o rd for th e situ atio n resu ltin g b y th e ato m ic a c t
§71 THE WORD PROBLEM FOR SEMI-GROUPS 385
of th e m ach in e from th e g iv e n situ atio n . G iv e n th e P o st w ord for a p a ssiv e
situ atio n , none of th e n rules is applicable.
H en ce a g iv e n w ord Q is d ed u cib le from a g iv e n P o st w ord P in th e
sem i-T h u e sy ste m , if an d o n ly if Q is th e P o st w ord for a situ a tio n w h ich
th e T u r in g m ach in e w ill reach from th e situ a tio n represen ted b y th e
P o st w ord P.
T h e p a rtia l recursive fu n ctio n 0 • p yT ^ x , x } y) is d efin ed an d has th e
v a lu e 0, if an d o n ly if {E y)T x(x9x 9y) (cf. § 6 3 ). B y T h eo rem X X V I I I ,
there is a T u rin g m ach in e w h ich 1/1 co m p u tes O -p yT ^ x , x, y). W e set u p
th e sem i-T h u e s y ste m corresponding to th is m ach ine. In th is se m i-T h u e
syste m , from th e P o st w ord for th e situ a tio n in w h ich a n u m b er x is
scanned in sta n d a rd position w ith sta te q x an d th e ta p e is b la n k elsew here
(i.e. th e P o st w ord h sx . . . s ^ h w ith x+ 1 occurrences o f sx), w e ca n
ded uce th e P o st w ord for th e situ a tio n in w h ich th e n u m b er pair x, 0
is scanned in stan d ard position w ith sta te q 0 an d th e ta p e b la n k else
w here (i.e. th e P o st w ord h s t ... s ^ s ^ h w ith x+ 1 occurrences o f sx
p reced in g th e s0), if an d o n ly if (EyYT^x, x, y). Sin ce b y use o f C h u rch 's
thesis there is no algo rith m for d ecid in g w h eth er (£‘y ) r i (^, x , y) (Th eorem
X I I ) , there can b e no algo rith m for decid ing, for a n y tw o g iv e n w ord s
P an d Q in th e sem i-T h u e sy ste m , w h eth er Q is d ed u cib le from P , i.e.
w h eth er Q follow s from P b y th e rules (a).
T o estab lish th e theorem , it rem ains to e x te n d th is result to th e fu ll
T h u e sy ste m in w h ich th e inverse rules (b) are a d m itte d . T h is is a c
com plish ed b y th e follow in g lem m a.
L emma For the rules (a) corresponding to a given Turing machine

V II.
as described above: If P is a Post word, Q is deducible from P by the rules
(a) and (b), and Q contains q 0, then Q is deducible from P by the rules (a)
only.
P roof of L emma V I I , b y co u rse-o f-va lu es in d u ctio n on th e le n g th
l o f a g iv e n d ed u ctio n of Q from P b y th e rules (a) an d (b). L e t th e d e
d u ctio n b e Rx, ..., R i, w here Rx is P an d R* is Q. T h e case for / = 1
is triv ia l, an d w e n ow suppose l > 1. Since P is a P o st w ord, an d th e
rules (a) an d (b) each preserve th is p ro p erty, each of Rlf . . . , Rt is a
P o st w ord, an d hence co n ta in s e x a c tly one occurrence o f a q, i.e. o f one
o f q 0, . .., qk. N o w Rz, i.e. Q, co n ta in s q 0; an d b y th e choice o f th e rules
(a) to correspond to th e a c tiv e co n figu ratio n s of th e T u r in g m ach in e,
each o f th e A / s con tain s q c for som e c^ 0. H en ce Rt m u st com e from
R*_i b y one of th e rules (a). So if there are a n y ap p licatio n s of th e rules
(b) in th e g iv e n d ed u ction , th e last w ill b e in th e step from R f_x t o R*
for some t < l. Then R *+1 comes from R* by one of the rules (a); but also
since the rules (b) are the inverses of the rules (a), R i_1 comes from R*
by one of the rules (a). But R t is a Post word, and to such a word at
most one of the rules (a) is applicable and in at most one way. Hence
R*_x and R ,+1 are the same word. Hence we can shorten the given de
duction of Q from P by omitting R*Rm . Applying the hypothesis of the
induction to the shortened deduction of Q from P by the rules (a) and (b),
we conclude that there is a deduction of Q from P by the rules (a) only.
It is easily seen that the word problem for semi-groups is equivalent
via Godel numbering to a problem of the form whether (Ey)R(x, y) is true
for given where R is general recursive. Since our proof of Theorem
XXXI is by reducing the problem whether {EyjT^x, x,y) to the word
problem, the word problem is of highest degree of unsolvability for
decision problems of predicates of the form (Ey)R(x} y) (cf. end § 61 and
Example 2 § 65).
Some further results along the same line as Theorem XXXI are
contained in Markov’s papers 1 9 4 7 , 1 9 4 7 a, 1 9 4 7 b, 1 9 5 1 , 1 9 5 1 a, 1 9 5 1 b.
Also cf. Hall 1 9 4 9 and Boone 1 9 5 1 abstract.
The definition of equivalence of two words P and Q (in symbols,
P ~ Q) which we used above in the word problem for semi-groups can
be expressed inductively thus, where (extremal clause) P ~ Q only
as required by the following (direct clauses): 1 . A* ~ B* (i = 1 , ..., n).
2 . U ~ U. 3. If U — V, then V — U. 4. If U ~ V and V — W, then
U — W. 5—6. If U — V, then Ua* — Va, and a,U — a,V (i = 1, .
Turing 1 9 5 0 * shows that the word problem for semi-groups with cancellation,
which we obtain by adding the two following clauses, is likewise un-
solvable: 7—8 . If Ua, ~ Vat- or a*U ~ a*V, then U ~ V (i = 1, . . m).
PA R T IV
M A T H E M A T IC A L L O G IC (A D D IT IO N A L T O P IC S )
Chapter X IV
THE PREDICATE CALCULUS AND AXIOM SYSTEMS
§ 72. GOdel’s completeness theorem. We resume the study of
the predicate calculus, continuing from the point reached in § 37. Say
that F is a predicate letter formula containing free only the distinct
variables zlt ..., zq (q ;> 0) and containing only the distinct predicate
letters P^ ...,P , (s ^ 1). An assignment of objects zx, from
some non-empty domain D as values to zx, ..., z, and of logical functions
P v ..., P„ over D as values to Px, ..., P„ we say satisfies F (or is a sat
isfying assignment to zlt ..., z„ Plf ..., Ps for F), if under the valuation
rules given in §§ 28, 36 and 37 F then takes the value t. As defined in
§ 37, F is satisfiable {valid) in a non-empty domain D, if some (every)
assignment to zlt ..., z„ P1( ..., P, in D satisfies F. As to the
notation, we are now writing the logical functions “Pi{al, ..., an.)”,
"A {a, h)’’, etc., instead of .... an;.)”, ”l(a, b)” , etc. as we did in
§§ 36, 37. In the case that the domain is the natural numbers, logical
functions are simply number-theoretic predicates, when we do not
make a distinction between propositions and truth values t or f (cf.
(b) § 45 and remarks there); and indeed we shall sometimes call them now
predicates.
T heorem 34oC. If a predicate letter formula F is irrefutable (i.e. if
~iF is unprovable, § 41) in the predicate calculus, then F is satisfiable in
the domain of the natural numbers. (Godel’s completeness theorem for the
predicate calculus, 1930.)
Modifications of Godel’s proof appeared in the 2nd (1938) edition of
Hilbert-Ackermann 1928 and in Hilbert-Bernays 1939. Henkin 1949 gave
a proof employing a minimum of knowledge of the deductive properties
of the predicate calculus. We give a proof which is intermediate in this
respect between Hilbert-Bernays’ and Henkin’s. There is also a proof by
Rasiowa and Sikorski 1950 using algebra and topology.
P roof of T heorem 34 (preliminaries). By Theorem 19 §35, any
predicate letter formula F is equivalent to a prenex predicate letter
389
390 AXIOM SYSTEMS CH. XIV
form ula, w h ich (b y th e m eth o d of proof) has th e sam e d istin ct free
v a riab les zv . . . , z g an d p red icate letters Pv . . . , P s as F . B y T h eorem
21 § 37 an d th e v a lu a tio n ta b le for ~ § 28, or b y paralleling th e proof of
T h eorem 19 se t-th e o re tica lly , th e pren ex form of F is satisfied b y a g iv e n
assignm ent zlt . . . , z q, P v . . . , P s, if an d o n ly if F is satisfied b y it.
F o r th e rem ainder of th e proof (including L em m as 22 an d 23), w e
shall assum e th a t F is prenex. For illustration , suppose th a t F is
¥ x 1 3 y 1 V x 2V x 33 y 2V x 4B ( z 1, z 2, xv y v x 2, x 3, y 2, x 4) w here B (zv z 2, xv yv
x 2, x 3, y 2, x 4) con tain s no q uan tifiers an d o n ly th e d istin ct va riab les
show n (so q— 2).
In ste a d o f sp eak in g of th e v a lu e w h ich a p red icate letter fo rm u la
A (u v . . . , u P) ta k es w hen its free va riab les an d pred icate letters u v . . . , u p,
P 4, . . . , P s ta k e re sp e ctiv e ly th e va lu es uv ..., u p, P lf ..., P s, it w ill
u su a lly be m ore co n ven ien t n o ta tio n a lly to perm it th e su b stitu tio n of
num erals for th e free va riab les of p red icate le tte r form ulas, an d th en to
sp eak of th e v a lu e ta k e n b y A ( u 1, . . . , u v) (where u v . . u P are th e
num erals for th e n a tu ral num bers u lt . . . , u P) w hen P 4, . . . , P s ta k e th e
v a lu es Pv ..., Ps. (Here A(ux, . . . , u P) is a predicate letter form ula w ith
num erals in th e sense o b ta in ed b y ex te n d in g th e n otion of ‘^ -pred icate
le tte r form ula' § 37 to allo w all num erals an d n o t sim p ly 1, . . . , k (and
variables) as term s. A p red icate letter form ula w ith num erals co n ta in in g
no v a riab les free or b o u n d is a proposition letter form ula with num erals.)
We can th en tre a t th e problem of choosing th e lo gical fu n ctio n s
Pv ..., P s as a problem of choosing a va lu e (t or f) for each of the form ulas
P 5(ax, . . . , anj) where j = 1, . . . , s and av . . . , anj range over all n r
tu ples of n a tu ral num bers. L e t these form ulas be enu m erated in som e
m anner w ith o u t repetitions as Q0, Q v Q 2, . . . .
B y th e v a lu a tio n rules for V an d 3, F is satisfied b y zv z2, Plf ,.., P s,
if for each n a tu ral n um ber x lf there is som e n a tu ral n um ber y x d ep en d in g
on xx (w rite it such th a t for each x 2 an d x3, there is som e y2
d ep en d in g on x lt x 2, x 3 (write it“y 2{xv x 2, #3)” ), such th a t for each #4,
(I) B ( z 1; z 2, x 1( y ^ x j , x 2, x 3, y 2{xv x2, x3), x 4)
(where<
y 1(x1) is th e num eral for th e n a tu ral num ber y i(% ), etc.) has th e va lu e
t. W e n ow t a k e z 4, z2>y ^ x ^ , y 2(xv x 2, x3) to be 2° • 3 1, 2° * 32, 2 1 * 3Xl, 22 • 3Xl *
5X2 • 7^3, re sp ectively. T h u s w e d eterm ine an infin ite class F0 of proposition
le tte r form ulas w ith num erals (nam ely th e form ulas (I) w hen xv x 2, x3, %4
range o ve r all qu ad ru p les of n a tu ral num bers, an d zlf z2, y^ x ^ , y 2{xlf
x 2, x3) are as ju st specified) such th a t if P v . . . , P s join tly satisfy these
form ulas, i.e. g iv e th em all th e va lu e t, th en zv z2, Pv . .., P s sa tis fy F .
§72 GODEl/S COMPLETENESS THEOREM 391
W e s a y th a t a class of form ulas is consistent in a form al sy ste m S , if
th e form al sy ste m o b ta in ed b y ad jo in in g th e form ulas o f th e class as
axio m s to th e p o stu la tes of 5 is sim p ly co n sisten t (§ 28), i.e. if for no
form ula A are b o th A an d - i A d ed u cib le in S from form ulas o f th e class.
T h e proof of T h eorem 34 w ill b e co m p le ted b y L e m m a s 22 an d 23,
w h ich relate to a n y p ren ex p red icate le tte r form u la F an d th e class
F0 o b ta in e d from it as illu strated .
L em m a 2 2 oC. I f F0 is consistent in the propositional calculus, then F is

satisfiable in the dom ain of the natural numbers.
P roof. By the preliminaries, it will suffice to show that the formulas
of F0 can be jointly satisfied under the valuation procedure of the prop
ositional calculus by an assignment of truth values (t or f) to each of
Go, Qv Q 2> ___ (Our m eth o d n o w w ill sh ow th is for a n y class F0 of
form ulas con sisten t in th e pro p o sitio n al calcu lu s an d list Q 0, Qv Q 2,
... of d istin ct form ulas prim e for th e pro p o sitio n al calcu lu s in c lu d in g
all co m p o n en ts of F0 prim e for th e p rop osition al calculus, § 25.)
L e t R 0 b e Q 0 or - i Q 0 acco rd in g as F0 b Q 0 or n o t F0 b Q 0 in th e
p rop osition al calcu lu s (where “ F0 b” m eans d ed u cib le from form u las
of F0), an d ad jo in R 0 to F0 to o b ta in a new class Fx. T h e n Fx is also con sisten t
in th e p roposition al calculus. F o r in th e first case (i.e. w h en F0 b Qo
an d R 0 is Q 0), th e a d d itio n of R 0 does not increase th e class of th e form ulas
w h ich can b e d ed u ced ; an d in th e second case (i.e. w h en F0 b Qo an d

R 0 is “ iQo), if Fx b A an d F* b ~»A for som e A , th e n F0, - i Q 0 b A
an d F0, - i Q 0 b A , an d th en ce b y i - i n t r o d . an d elim ., F0 b t h Qo
b Q 0, co n tra d ic tin g th e case h yp o th esis.
S im ila rly for each n a tu ra l n u m b er i, le t R i b e Q i or - i Q i acco rd in g as
Fi b Q i or JFi b Q i, an d a d jo in R i i
to F to o b ta in Fi+1; th e n th e co n
siste n c y o f Fi+1 follow s from th a t of F*.
W e n ow assign to Q i th e v a lu e t or f acco rd in g as R i is Q i or R i is - i Q i .
F o r th is assign m en t, n o t o n ly R 0, Rv R 2, . . . b u t also th e form ulas
of F0 ta k e th e v a lu e t. F o r le t H b e a form ula of F0. T h e d istin ct p arts of
H prim e for th e p rop osition al calcu lu s b elo n g to th e list Q 0, Qv Q 2, . . . ;
s a y t h e y are Q i , . . . , Q i;. C onsider th e form ula H & R ^ & . . . & R ^ ; ca ll
it "A". L e t i = l+ m a x ^ , . . . , i t)\ th en Fi b A , since H , R ^ , . . . , R ^
a ll b elo n g to Fi. T h e assigned va lu e s are th e o n ly ones for w h ich R ^ ,
. . . , R {, are all t. H en ce if H w ere n o t t for th is assign m en t, th en A w o u ld
b e id e n tic a lly false, an d - i A w o u ld b e id e n tic a lly true (§ 28); so b y T h e o
rem 10 § 2 9 -I A w o u ld b e p ro v a b le in th e p rop osition al calcu lu s, a fortiori
Fi b ~iA, w h ich w ith Fi b A w o u ld co n tra d ict th e c o n sisten cy o f Fi.
L emma If F is irrefutable in the predicate calculus, then F0 is
23.
consistent in the propositional calculus.
P roof . N o te th a t by our choice of th e num bers zv z2 an d fun ction s
yi(xi) an d y 2(*n x 2> xb) : (A) T h e num bers zx and z2, y i(% ) for x x = 0 ,
1, 2 , ..., an d y 2(%, x 2> x s) f° r x i > x 2> x z = 1 2 ,... are all d istin ct.
(B) xx < y x(%) an d % , y ^ ) , x2, x 3 < y 2(xv x2} x3).
T o sh ow b y redu ctio ad absurdum th a t F0 is con sisten t, suppose th a t
for som e A, b o th A an d - i A are d ed u cible in th e propositional ca lcu lu s
(using p red icate le tte r form ulas w ith num erals) from form ulas o f F0.
T h e n usin g w e a k -i-e lim ., there is a d ed u ctio n of th e form ula cC7 &
from th e sam e form ulas, i.e. from a fin ite set of form ulas of th e form (I).
L et " S ” , “ T” , “ 2", ... d en ote v a riab les d istin ct from each oth er an d
from x 1} y lt x 2, x 3, y 2, x 4, b u t such th a t zx is z ± an d z 2 is z 2. B y ch a n gin g
each num eral u w here it occurs in this d ed u ctio n n ot as a p a rt of an oth er
num eral to th e correspon ding va ria b le uf w e o b ta in a d ed u ctio n of
e2f & 1 <£? in th e (predicate letter) propositional calculus, a fortiori in th e
(pure) p red icate calcu lu s w ith all va riab les held co n sta n t, from a fin ite
set of form ulas of th e form B ( z 1, z 2, x lt x 2> xz)> xi) > an(^
th en ce b y use of V -elim . (noting th a t x 4 is d istin ct from th e oth er v a riab les
show n), from form ulas of th e form
Vx4B(z1; z2, xv y^X j), xg, x 3, y2(x1( x 2, x 3), x4).

T h u s w ritin g o u t th e d istin ct assu m ption form ulas (say there are l of
them ),
Vx4B(z4, z2, x\, y4(x}), x \, x \, y 2{x\, x \, x § , x4),

Vx4B(z4, z2, x{, yx(*f), x \, x \, y 2{x\, x\, x\), x4),
Vx4B(z4, z 2 , x\, y x(x 0 . x \, x \, y 2{x\, x \, x \), x4) h <3 & - i <3.
T h e form ula co n ta in s no variab les. T h e v a riab les w h ich
a p p ear free righ tm o st in th e assu m p tio n form ulas of (1) are
y 2(x \> x l, x \), . . ., y 2(x\, x l2, x l3). Sin ce th e l assu m p tio n form ulas are
d istin ct, i.e. x {} x{, x \ for j = 1, . . . , / are d istin c t triples of n a tu ra l
num bers, b y (A) these l va ria b le s are d istin ct from each oth er as w ell as
from z v z 2. B y (B), w e can choose one of th em , sa y y 2(î> x h * 3)* w h ich is
of greater in d ex (i.e. com es la te r in th e list 0, 1 , 2, . . . ) th a n a n y of
th e v a ria b les w h ich ap p ear free elsew here in (1) (i.e. n o t rightm ost)
e x c e p t perhaps z x or z 2, an d hence is d istin ct from th ose va ria b le s also.
B y 3-elim ., follow ed b y a ch an ge of b o u n d v a riab les (*74 § 33, n o tin g
th a t y 2 is d istin ct from th e other va riab les shown) an d tw o V -elim in atio n s
§72 g o d e l ’s c o m pl e t e n e s s t h e o r e m 393
(noting th a t x 2, x 3 are d istin ct from each oth er an d th e o th er v a ria b les

shown),
V x 2V x 33 y 2V x 4B ( z 1( z 2, x \, y ^ x l), x 2, x 3, y 2, x 4),
(2) V x 4B ( z1, z 2, x \, yx{x\), x\, x\, y2(x\, x\, x\), x 4),
V x 4B( z 1, z 2, x \, y1(x[), x\, x\, y2(x\, x\, x\), x 4) h & -i<^.
N o va riab les are va ried , since th e v a ria b le y2{x\, x\, x\) of th e 3-elim .
does n o t occur in a n y of th e other assu m p tion form ulas (ef. L e m m a 7 b
§24 ).
A g a in b y (A) th e v a riab les y^ x f), y 2(xf, xt> xt)> • • •> y%{%i> x \> x l) w h ich
ap p ear free righ tm o st are d istin ct from each oth er as w ell as from
zt an d z 2. B y (B) w e can choose one of th em w h ich is o f greater in d e x
th a n any of th e v a riab les ap p earin g free elsew here in (2 ) e x c e p t
z x an d z 2, an d so m u st be d istin ct from those also. T h is v a ria b le m a y b e
y x{x X) or s a y y 2(xf, x\, xf). I f it is y^ x f), b y 3-elim ., * 7 4 an d V -elim . w e
replace th e first assu m p tion form ula b y V x 13 y 1V x 2V x 33 y 2V x 4B ( z 1, z 2, x v
y v x 2, x 3, y 2, x 4), i.e. b y F . I f it is y 2(xf, x \, x l), w e in stea d replace th e
second assu m p tion form ula b y V x 2V x 33 y 2V x 4B ( z 1, z 2, xf, y t (xf), x 2, x 3, y 2,
x 4), an d if th is is a d u p lic a te of th e first (i.e. if x \ = xf) w e o m it it.
W e co n tin u e in th is m anner. A fte r each use of 3-elim . (applied to a
y ap p earin g righ tm o st an d d istin ct from all th e oth er v a ria b les shown)
an d *74 , w e a p p ly V -elim . to th e x 's w h ich are “ u n c o v e re d ” b y th e 3-elim .,
an d th en om it th e resu ltin g a ssu m p tio n form ula, if it is a d u p lic a te of
another, so th a t th e y 's w h ich ap p ear righ tm o st a t th e n e x t sta g e w ill
again be d istin ct from each other. E v e n t u a lly w e o b ta in sim p ly
(3) F h
F ro m (3) b y & -elim . an d -i-in tr o d .,
(4) h -iF ,
co n tra d ictin g th e h yp o th e sis of th e lem m a t h a t F is irrefutable.
C o r o lla r y 1 oC. E very predicate letter form ula G which is valid in the
dom ain of the natural numbers is provable in the predicate calculus {and
hence, b y T h eo rem 21 § 3 7 , is valid in every non-em pty dom ain) . (A n o th e r
version o f G o d el's com pleten ess theorem , 19 3 0 .)
P roof . {G is valid in the domain of the natural numbers}

-> {~ iG is not satisfiable in that domain} {~ iG is refutable, i.e.
“ i - i G is provable, in the predicate calculus} [by the theorem, applied
with i G as th e F , and contraposed (cf. * 1 4 § 26)] -> {G is p ro v a b le in
the sam e} [b y -i-elim].
C o r o llar y 2c. If a F is satisfiable in some
p r e d ic a te le tte r f o r m u la
(:
n o n -e m p ty ) , then F
d o m a in in the domain of the natural
is s a tis fia b le
numbers. (Low enheinTs theorem , 1 9 1 5 , also called th e L o w e n h e im -

Skolem theorem.)
P roof . B y contraposing C o ro llary 1 ; or th u s: (F is sa tisfiab le
in som e dom ain} -> { - i F is n ot v a lid in th a t dom ain} -> {—»F is n o t
p ro v a b le, i.e. F is irrefu tab le, in th e p red icate calcu lu s} [b y T h eo rem
21 contraposed] -> { F is satisfiab le in th e d om ain of th e n a tu ra l num bers}
[b y th e th eo rem ].
Ldwenheim’s proof of his theorem , an d th e sim pler proof g iv e n b y

Sk o lem 1920 , e m p lo y th e set-th e o retic ax io m of choice (Zerm elo 190 4 ,
cf. § 13). T h e proof v ia G o d e l 1s theorem is n o n -co n stru ctive to a lesser
degree. T h e n o n -in tu itio n istic step (in th e present treatm en t) occurs in
th e proof of L e m m a 22, w here w e assum e th a t Ff b Q , or Ft b Q f.

B y fo rm alizin g a p a rt of th eir proof of G o d e l’s com pleten ess theorem
w h ich co n ta in s th e n o n -co n stru ctiv e step, H ilb e rt an d B e rn a y s o b ta in
a m e ta m a th e m a tic a l com pleten ess theorem for th e p red icate calcu lu s
(1939 pp. 252— 253), w h ich w e shall fo rm u late as T h eo rem 36.
T heorem 3 5 oC. The satisfying

predicates P lt . . , , P S for F in Theorem
34 can be chosen so that Pj{av . . . , aUj) = (.E x)(y)R j(a1, . . . , an.f x, y) ==
(x){E y)S j{a1> . * . , anj} x, y) where R j and S j are prim itive recursive
a = 1.
P roof. In th e fo llo w in g, ‘ 'recu rsiv e” can m ean gen eral recursive
(th ou gh a c tu a lly th e sta te m e n ts h old in th e m ean in g p rim itiv e re cu rsiv e );
th en b y C o ro llary T h eo rem I V § 57, w e ca n ta k e Rj an d S 5- in th e con
clusion to b e p rim itiv e recursive.
Su p p ose w e h a v e set u p a G o d el n u m berin g of th e p ro p o sitio n le tte r
form ulas w ith num erals b y th e m eth o d s of §§ 52 an d 56. T h e n if a an d b
are th e G o d el num bers of form ulas A an d B , re sp e c tiv e ly , t h a t of A D B
is p(a, b) an d of - 1 A is v(a), for certain recursive fu n ctio n s p an d v. L e t
H(a)' = {a is th e G o d el n u m b er of a form ula b elo n gin g to F0}; th en H is
recursive. I f th e en u m eration Q 0, Qv Q 2, ... w as chosen s u ita b ly , th e
fo llo w in g fu n ctio n s w ill be recu rsive: x(i) = {the G o d el n u m ber of Q J ,
a i(av . . . , a n/) = {th e i such th a t P ^ , . . ., a nf) is Q J .
B y th e defin ition of th e sa tisfy in g p redicates as g iv e n in th e proof
§72 godel ’s completeness theorem 395
of Lemma22, if P,(ax, . . a»y) is Qit then Pj(av . . . , anj) == {Rt is QJ =
{Ft- b Qt}. Let F(i, a) =s {a is the Godel number of a formula A such that
Ft- b A}. Then
(a) P »&rij) = F( 0Cj(î, . . x(0C5(fli, . . . , #7jy))).
Now we shall investigate the predicate F(i, a).
To say that F0 h A in the propositional calculus means the same
as that A is provable in the formal system obtained by adjoining F0 as
axioms to the propositional calculus. Let F0(a) == {a is the Godel number
of a provable formula of this system}. By the methods of §§51, 52 (cf.
Dnl2), since H is recursive, F 0(a) = (Ey)R(a,y) with a (primitive) re
cursive R] i.e. F0(a) is expressible in the existential 1 -quantifier form of
Theorem V Part II § 57.
By the definition of ¥i+1 from Ft and the 3-rules, {Fi+1 h A }= {F,-,
Rt b A } — {F,- b R*3 A }. T a k in g into acco u n t the definition of R*
by cases, we thus have
J F (0, a) = F 0(a),
\ F (i', a) = [F(i, x(*)) & F (i, ^(x(i). «))] V [F (i, x(*)) & F (i, |x(v(x(*)), «))]•
This shows that F is recursive in F 0; for applying #D § 45, and going

over from the predicates F and F 0 to their representing functions b and v are (primitive) recursive. (In fact by the result of
Peter 1 9 3 4 cited in § 55, F is primitive recursive in F 0.)
By (a), P j is recursive in F , a,-, x; hence since oq and x are recursive, in
F ; and hence in F 0, which is expressible in the existential 1 -quantifier
form. Therefore, by a theorem of Post (Theorem XI § 58), Pj is ex
pressible in both the 2-quantifier forms of Theorem V, as was to be shown.
T heorem 36°.The addition to the postulate list for the predicate
calculus of an unprcvable predicate letter form ula G for use as an axiom
schema would cause the number-theoretic system as based on the predicate
calculus and Postulate Group B (§19) to become <*>-inconsistent (§ 42).
{In fact , a certain form ula would become refutable which expresses a true
proposition of the form (y)D (y) where D (y) is an effectively decidable pred
icate.) (Hilbert-Bernays completeness theorem, 1 9 3 9 .)
Note the partial analogy to Corollary 2 Theorem 10 §29 for the prop
ositional calculus.
396 AXIOM SYSTEM S CH. X IV
M e t h o d o f pr o o f . W ith o u t loss of g en erality, w e can ta k e G to b e

closed, so th a t q= 0 (cf. end § 32). L e t F b e a p ren ex form of - i G , so t h a t
b y T h eo rem 19 w ith *30 an d *49,
(i) b G ~ -i F in th e p red icate calculus.
B y (i), since G is u n p ro vab le, so is i F , i.e. F is irrefutable.

H en ce b y L e m m a 23 (the proof of w h ich w as fin itary),
(ii) F0 is con sisten t in th e p rop osition al calculus.
T h e c o n sisten cy of F0 is e q u iv a le n t to th e p roposition th a t F0 b <3 & -i< 3.
Ifr b e th e G o d e l n u m ber of Z l & - i <3, th en F0 b S en == (E y)R (r, y)

= (y)R (r , y) for th e p rim itiv e recursive R (a, y) used in th e proof of T h e o
rem 35. L e t D (y) == R (r, y). T h e n D (y) is p rim itiv e recursive, an d (ii) is
e q u iv a le n t to
(“ .) (y)D (y).
By C o ro llary T h eorem 2 7 § 49, D (y) is num eralw ise expressed in th e
n u m b er-th eo retic form alism b y a form ula D (y). T h e n from (iid),
(iii) (y)[t- D(y)]

in th e n u m b er-th eo retic form alism .
A c c o rd in g to L e m m a 22, classically,
^ {F0 is co n sisten t in th e proposition al c a lc u lu s } ->
{F is satisfied by certa in p red icates P lf . . . , P s}.
T h e prem ise of th e im p lica tio n (iv) can b e expressed in th e sy m b o lism

of th e n u m b er-th eo retic sy ste m b y th e form ula V y D ( y ) . T h e co n clu sio n
ca n also be expressed in th e n u m b er-th eo retic sym b o lism , w h en w e use
th e expressions for Pv
P s g iv e n in T h eorem 35. L e t R ^ a *, . . . , a nj, x , y )
...,
num eralw ise express R j(a lf . . . , anj, x , y). T h e n 3 x V y R i (a1, . . . , a^-, x , y)
expresses P j(a lt . . . , an]) in th e sym b o lism . T h e p rop osition t h a t F is
satisfied b y P v . . . , P s is th en expressed b y th e form u la F * w h ic h w e
o b ta in from F b y s u b s titu tin g th e form ulas 3 x V y R i (a1, . . . , aW;-, x , y )
(/ = 1, . . . , s) for th e re sp e ctive p red icate letters P i (a1, ..., a nj) (assu m in g
t h a t th e b o u n d va ria b le s h a v e been s u ita b ly chosen). Then th e im
p lic a tio n (iv) is expressed b y th e form u la V y D ( y ) D F * .
W e n o w propose th a t th e in form al classical d em o n stratio n of (iv)
sh ould b e form alized in th e classical n u m b er-th eo retic sy ste m as a p ro o f
of th is form ula, so t h a t w e sh ould th e n h a v e
(v) b V yD (y) D F *
§72 g o d e l ’s c o m pl e t e n e s s t h e o r e m 397
in th a t sy ste m . B y co n trap o sitio n (* 1 2 § 2 6 ),
(vi) b “"*F* D —iV y D (y ).
F ro m (i) b y su b stitu tio n (Theorem 15 § 34),
(vii) f- G * ~ —i F * .
U sin g (vii) in (vi),
(viii) b G * D —iV y D (y ).
In (iii) an d (viii), th e " b ” refers to th e u n a u gm e n te d n u m b er-th eo retic

form alism . N o w if G b e ad d ed to this sy ste m as an axio m sch em a, th en
G * becom es an a x io m (b y th e new schem a), an d th en (since (iii) an d (viii)
still hold) D (y ) for y = 0, 1 , 2 , . . . an d - i V y D ( y ) are p ro v a b le sim u l
ta n e o u s ly , i.e. th e au g m e n te d sy ste m is ^ in c o n s is te n t. (Th us V y D ( y )
becom es refu tab le, alth o u g h it expresses th e true proposition (y)D (y).)
W e sh all not ta k e th e space to ca rry o u t th e fo rm alizatio n of a g iv e n
inform al d em o n stratio n called for here in th e step from (iv) to (v), ju st
as w e did n o t for th e step from (I) to (II) in th e proof of T h eorem 30 § 42.
H ilb e rt an d B e rn a y s 1939 (pp. 205 f f ., esp ecially pp. 243— 252) ca rry
o u t th e fo rm alizatio n of th e corresponding p art of their proof of G o d e l’s
com pleten ess theorem in an oth er form al sy ste m , w h en ce it can be in
ferred th a t 'th eo rem 36 holds for ours (cf. th e rem arks on th a t sy ste m
p reced in g E x a m p le 9 § 74).
T heorem Given an enum erably in fin ite (or finite) class of pred
3 7 oC.
icate letter form ulas F 0, F v F 2, . . . , if every conjunction of a fin ite num ber
of them is irrefutable in the predicate calculus, then they are join tly satisfiable
in the dom ain of the natural num bers , b y a sa tisfy in g assignm ent of n a tu ral
num bers z0, z v z2, . . . to th e d istin ct v a ria b les z 0, z v z 2, . . . w h ich occur
free, an d of p red icates P 0, Plf P 2, ... to th e d istin ct p red icate letters
P0, Pj, P2, . . . w h ich occur, in form ulas of th e class. T h e lists z 0, z v z 2, . . .
an d P0, Px, P2, . . . m a y b e fin ite or infin ite. (G o d el’s com p leten ess
theorem for in fin ite ly m a n y form ulas, 1930 .)
P roof . F o r exam p le, suppose n ow th a t for som e k, Fk is
V x *l 3 yfclV x *2V x fc33 yfc2V x fc4B (ZeSl. Ze*,. X k l ’ Y k l ’ X k2 ’ X lc3’ Y k 2 ’ X k i) - T h e n W e

ta k e zehl, Zek i, y k l(xkl), y k2{xk l, xk2, xk3) to be 2 ° - 3 e*>, 2°• 3e*2, 2l-3 k -5 Xk\
2 2• 3k • 5Xkl • 7 Xk2• 1 \ Xkz. T h e set F0 is to be th e sum of th e sets F00, F10, F20, . . .
form ed as w as illu stra ted .
Corollary 1oC. I f for each assignm ent in the dom ain of the natural
numbers to the free variables and predicate letters of the form ulas G 0, G v
G 2, . . ., one of those form ulas takes the value t, then some disjunction of a
fin ite number of them is provable in the predicate calculus.
C o r o lla r y 2c . If
F lt F 2, . . . are jointly satisfiable in some non
F 0,
em pty dom ain [or even if each conjunction of a finite number of F 0, F lf
F 2, . . . is satis fiable in a respective non-em pty dom ain (G odel 1930 )),
then F 0, F lf F 2, . . . are join tly satis fiable in the dom ain of the natural
numbers. (A gen eraliza tio n of L o w e n h e im ’s theorem , Skolem 1920 .)
P roof of Corollary 2. { E a c h co n ju n ctio n of a fin ite n um ber of

F 0, F x, F 2, . . . is satisfiab le in a resp ective n o n -e m p ty dom ain} ->
{th e n eg atio n of each such co n ju n ctio n is not v a lid in every d o
m ain} -> {each such co n ju n ctio n is irrefu table} (b y T h eo rem 21 § 37)
{ F 0, Fv F 2, . . . are jo in t ly satisfiab le in th e d om ain of th e n a tu ra l
num bers} (b y th e theorem ).
T heorem 3 8 oC. I f the enum eration F 0, F lt F 2, . . . of form ulas in

Theorem 3 7 is effective (if infinite), the join tly satisfying predicates Po,
P lf P 2, . . . can be chosen so that P j(a lf . . . , an?) == (E x)(y)R j(av . . . , anj)
x , y) = (x)(E y)S j(a1, . . . , a nj, x, y) where R j and S j are prim itive recursive
(j = 0, 1 , 2 , . . . ) . T h e h yp o th e sis th a t th e enu m eration F 0, F v F 2, . . . is
e ffe c tiv e ca n b e m ad e e x a c t b y requiring, for a su itab le G d d el num berin g,
t h a t th e G o d el n u m ber of Fk b e a gen eral recursive fu n ctio n of k (cf.
§ 6 1 ) , or b y use of T u r in g m ach ines in th e m anner su ggested a t th e end
of § 7 0 .
P roof . N ow th e class aH (a) (cf. th e proof of T h eo rem 35) is re

c u r siv e ly enu m erable (but n o t n ecessarily recursive). H o w e v e r b y ta k in g
F 0(a) = ( E n )F 0(a , n) w here F 0(af n) = {a is th e G o d el n um ber of a
p ro v a b le form u la of th e sy ste m o b ta in ed b y ad jo in in g th e first n of th e
form ulas F0 as axio m s to th e proposition al calculus}, an d usin g (17) § 57
(or end § 53), w e still h a v e F 0(a) = ( E y)R (a , y) w ith a recursive R.
T h e sign ifican ce o f G o d e l's com pleten ess theorem an d L o w e n h e im 's

theorem (including th e version s g iv e n in § 73) w ill b e discussed in § 75 ,
w h ich m a y b e read w ith o u t th e starred § 74, p ro v id e d th e reader w ill
a c c e p t a few p lau sib le sta te m e n ts referring to § 74. In § 76 so m ew h at
m ore use is m ad e of § 74.
I f F is d ed u cib le from G in th e p red icate ca lcu lu s w ith a p o s tu la te d
substitution* rule (end § 37), th e n F is v a lid in e v e r y d om ain in w h ich G i s ;
an d hence in terd ed u cible form ulas are v a lid in th e sam e dom ains. T h e
con verse is tru e w hen o n ly 0- an d 1 -p lace p red icate v a riab les occur, b y
use o f th e o r y c ited in § 7 6 ; b u t n o t in general, b y H asen jaeg er 19 5 0 .
§73 THE PREDICATE CALCULUS WITH EQUALITY 399
§ 73. T h e predicate calculus with equality. A tre a tm e n t of

e q u a lity may be co m b in ed w ith th e p red icate calculus b y a d d in g to th e
p o stu la tes th e follow in g ax io m an d ax io m schem a, w here x is a va ria b le,
A (x) is a form ula, an d a an d b are d istin ct va riab les free for x in A ( x ) :
22. a=a. 23. a = b D (A(a) D A(b)).
F o r th e pure predicate calculus with equality, ‘te rm ’ shall m ean va riab le,
an d ‘fo rm u la ’ shall h a v e th e sense of equality and predicate letter form ula,
w h ich w e o b ta in from ‘p red icate letter form ula' (§ 31) b y a d d in g to th e
d efin ition a clause w h ich sta te s th a t, if s an d t are term s, s = t is a form ula.
A n e q u a lity an d p red icate le tte r form ula co n ta in in g no p red icate letters
oth er th a n th e d istin ct letters P lf . . . , P s w e call a letter form ula in — ,
P • • •>Ps*
l>
(A) A x io m 22 (w hich is *10 0 § 38) an d a —b D (a = c D b=c) (A xio m

16 § 19, w h ich is now an a x io m b y A x io m S ch em a 23) w e ca ll th e open
equality axiom s for =. The n fo llo w in g axio m s b y S ch em a 23,
a=b ID (P(aj, . . . , a j __j , a, a 2-^_^, . . . , a n) ZD P(a^, . . . , a 4-_^, b, a 4-_^j, . . ., a n))
(i — 1, . . . , n ) , w here P is an n -p lace p red icate le tte r arid a x,
dii_x, a, b, a ^ j , . . . , a w are som e n + 1 d istin ct variab les, w e ca ll th e
open equality axiom s for P . T h e closed equality axiom s are th e closures
of th e re sp ective open e q u a lity axio m s (w ith w h ich th e y are in ter-
d ed u cib le, end § 32). B y “ E q ( = , P lf ..., P 8)” w e den ote the co n ju n ctio n
of th e closed e q u a lity ax io m s for — , P 2, . . . , P s.
E x a m pl e 1. If <3 ta k e s tw o argu m en ts, E q ( = , <3) is th e form u la

y4 a [a = a ] & V a V b V c [a = b ZD ( a = c D b = c )] &
V a V b V c[a = b D {£l{a,c) D c 3 (A ,c ))]'& V a V A V c [tf= A D (<Vt(c,a) D c 3 ( c ,i) ) ] #
(B) F ro m th e e q u a lity ax io m s for = (A xio m s 22 an d 16), w e can

d ed u ce in th e p red icate calcu lu s th e re fle x iv e (*100), sy m m etric (*10 1)
an d tr a n s itiv e (*102) properties of e q u a lity , an d b o th th e special re
p la cem en t properties (*108, *109) as in § 3 8 . F ro m a n y axio m a = b ID
(A(a) ZD A(b)) b y S ch em a 23 w ith A (x ) n o t co n ta in in g a or b free (in
p articu lar, *10 8, *10 9 or an e q u a lity ax io m for a p red icate le tte r P) an d
*10 1 w e can deduce a = b ID ( A ( a ) ~ A ( b ) ) (as we d id in a s lig h tly d ifferen t
fo rm at w h en w e inferred * 1 1 5 an d * 1 1 6 from *10 8, *10 9 an d *1 0 1 ).
N o w an d in § 7 5 , w h en w e are d ealin g w ith a class of e q u a lity an d

p red icate le tte r form ulas c o n ta in in g p red icate letters o n ly from a g iv e n
list, w e sh all use “ Q ” to sta n d for som e p a rtic u la r 2-place p re d ic a te
le tte r n o t in th e list. G iv e n a form ula of th e c la ss, call it “ E ” or “ E ( = ) ” ,
b y “ E Q” or “ E ( Q ) ” w e sh all m ean th e p red icate le tte r form ula o b ta in e d

from it b y replacin g sim u lta n eo u sly each p a rt of th e form s = t w here
s an d t are term s b y Q(s, t).
T h e notion s of ^ -id e n tity an d ^ -e q u a lity (§3 6 ), an d of sa tis fia b ility
an d v a lid it y in a g iv e n n o n -e m p ty dom ain, are e x te n d e d to eq u ali
ty an d p red icate le tte r form ulas b y p ro v id in g th a t a=b sh all h a v e
t as v a lu e (f as valu e) w h en a an d b assum e th e sam e (different)
o b je c ts from th e dom ain as th eir v a lu e s ; or if w e h a v e first su b stitu te d
num erals, t h a t a=b h a v e th e v a lu e t or f accord in g as a= b or a b.
T h u s a le tte r form u la F in = , P x, . . . , P s w ill be satisfied b y zv . . . , zg,
P lf . . . , P s, if an d o n ly if F Q is satisfied b y zv ..., zq, Q, P v ..., P s w ith
Q(a, b) == a = b .
In G o d eF s exten sio n of his com pleten ess theorem to th e p red icate
ca lcu lu s w ith e q u a lity (1930 ), it is n ecessary to allo w th e a lte rn a tiv e th a t
th e d o m ain b e fin ite, since e.g. a=£b & V c (c = a V c = b ) is satisfiab le in
a n d o n ly in a d o m a in o f tw o o b jects.
T heorem 39(oC). Theorems 20 (§36 ), 2 1 c (§ 3 7 ), 34 oC and 3 7 oC (§ 72 )

and their corollaries hold reading “equality and predicate letter form ula” ,
“predicate calculus with equality” , “satisfiable {valid, each assign
ment) in the dom ain of the natural numbers or a {and every, or a) non
em pty finite dom ain” in place of “predicate letter form ula” , “predicate
calculus” , “satisfiable {valid, each assignm ent) in the dom ain of the
natural num bers” , respectively. (The theorem s th u s e x te n d e d w e c ite
u sin g a sta r “ * ” ; th e m ark s <<0” an d “c ” a p p ly to th e starred th eorem
w h en th e y a p p ly to th e unstarred.)
P roofs. T h eo rem 3 4 *. L e t F b e a le tte r form u la in = , P x, . . . , P ,

w h ich is irrefu tab le in th e p red icate ca lcu lu s w ith e q u a lity . By (A),
E q (= , Pv . . . , P s) is p ro v a b le in th e p red icate ca lcu lu s w ith e q u a lity .
H en ce b y *4 5 § 27, F & E q ( = , P x, . . . , P s) is irrefu tab le in th e p red icate
ca lcu lu s w ith e q u a lity , a fortiori in th e p red icate ca lcu lu s (using e q u a lity
a n d p red ic a te le tte r form ulas). T h e n b y T h eo rem 15 § 3 4 , F Q & E q ( Q ,
P lf . . . , P s) is irrefu tab le in th e pure p red icate calculus. H en ce by
T h eo rem 34, F Q & E q ( Q , Pv . . . , P 3) is satisfiab le in th e d om ain o f th e
n a tu ra l num bers. T h e p roof is co m p le ted b y (a) of th e fo llo w in g lem m a.
Lemma 24 °. (a) I f F Q & E q ( Q , P v . . . , P s) is satisfiable in a given

non-em pty dom ain D , then F is satisfiable in a non-em pty dom ain D *
w ith the sam e or a lesser cardinal number.
(b) I f F ^ & E q ( Q , Pfci, . . . , P kSk) where P * i, . . . , P*** are the predicate
letters of F k (k = 0, 1 , 2, . . .) are join tly satis ft able in a non-em pty dom ain D y
then F 0, F x, F 2, . . . are jointly satisfiable in a dom ain D * with 0 < D* < D .
P ro o fs , (a) Su p p ose g iv e n a sa tisfy in g assign m en t z lf . . . , zQ,
Q ,P lt Ps for T h e n Q ,P V . . . . P s m u st
F Q & E q (Q , P ,, . . . , P s).
sa tis fy th e fo llo w in g fo r m u la s : (i) VaQ (a, a), (ii) ' id ib [Q(a, b) D Q(b, a )],
(iii) V a V b V c M a , b) & Q (b, c) => Q(a, c)], (iv) V [Q (a , b) D
( P i( a x» • • • > 3-i- 1> a > 3-i+l> • • • > a ny) ' ' P j ( a i> • • •> i> b , a i+ l> ,■ ))]
(/ = 1, . . ., s; i = 1, . . ., n 3). W e see th is b y p ara llelin g s e t-th e o re tic a lly
th e reasonin g g iv e n p ro o f-th e o re tic a lly under ( B ) ; or b y n o tin g th a t th e
co n ju n ctio n of these form ulas is im p lied b y (in fa ct, is e q u iv a le n t to)
E q ( Q , P x, . . ., P s) in th e pure p red icate calculus, an d usin g T h eo rem 21
§ 37 an d th e v a lu a tio n ta b le s for & an d 3 (or ~ ). (For p ro v in g T h eo rem
3 4 *, th is step can b e a v o id e d b y usin g th is co n ju n ctio n in p lace of
E q ( Q . P 1, . . . , P , ) . )
F ro m th e form of (i) — (iii) an d th e v a lu a tio n procedures for V , D
an d & , w e see th a t th e lo gica l fu n ctio n Q(a, b) w h ich fulfils th em m u st
b e re flex iv e, sy m m e tric an d tra n s itiv e . (A relation Q(a, b) w ith these
three properties w e call an equivalence relation .) H en ce th e d om ain D
falls in to equivalence classes w ith respect to th e relatio n Q, i.e. m u tu a lly
e x clu siv e n o n -e m p ty classes such t h a t a n y tw o m em bers a an d b of D
b elo n g to th e sam e class, if an d o n ly if Q (a , b). T h e n since th e form ulas
(iv) are satisfied, th e v a lu e of P j(a v . . . , an) for a n y j (j — 1, . . ., s) is
u n ch an ged b y ch a n g in g a n y one of its a rgu m en ts to an oth er in th e sam e
eq u iva len c e class.
N o w let us ta k e th e eq u iva len c e classes as a n ew d om ain D * (w ith
0 < D* < D ) ; an d define o b je cts z*, from D * an d lo gical fu n ctio n s
Q*, P * , . P * over 19* th u s: z* is th e e q u ivalen ce class to w h ich Zj
.
belo ngs, i.e. z f = bQ(zjf b); an d Q*(a*, 6*) (Pf{a*> . . . , t f * y)) sh all h a v e
th e v a lu e w h ich Q(a, b) (P j(a1, . . ., anj)) ta k e s for a n y a w h ich 8 a* an d
6 8 6 * (for a n y a , 8 a*, . . . , 8 4 y). T h e n z*, . . z*, Q*, P * , . . .,P *
s a tis fy in 19* a n y form u la w h ich zv . . . , zQ, Q, P lf . . ., P s s a tis fy in D ;
in p a rticu la r th e y sa tis fy F Q in D *. B u t ()*(a *, 6*) == a * = 6*. H en ce
* i , • • •, P f , •. ., P * s a tis fy F in 19*.
T heorem 40oC (— T h eo rem 38*). L et one of the predicate letters
P 0, P x, P 2, . . . occurring in F 0, F x, F 2, . . . , say P 0, be a 2-place predicate
letter. I f the enum eration F 0, F x, F 2, . . . is effective {if in fin ite ), then there
are a dom ain D * and join tly satisfyin g predicates P * f P * , P * , . . . for
Theorem 3 7 * such th a t : I f D * is in fin ite , and for a suitable enum eration
S0, of it P * {sa, s„) = a ’ = b, then P * ( s 0i, . . s0 )
(E x)(y)R *(av . . . , anj, x, y) s= (x)(E y)S *(av . x, y) where R f a n d S *
are p rim itive recursive (j = 1, 2, 3, . . . ) .
P roof. B y T h eo rem 38, jo in tly sa tisfy in g pred icates Q , P 0, P t for

th e form ulas Fj? & E q ( Q , P m , . . P***) (k = 0 , 1,2, . . . ) ca n b e chosen
w h ich are expressible in b o th th e 2-q u an tifie r form s of T h eo rem V § 5 7,
an d hence b y T h eo rem XI § 58 are gen eral recu rsive in 1-q u a n tifier
pred icates.
By th e d efin itio n of th e pred icates Pf in th e proof of L e m m a 24,
(1) P f { S a v .... Sa„.) ss (£Ci) ... (Ecn,) [Cj € S0l & . . . & Cnj £ Sa„. &
P j( c l> • • • > £»/)] = (^l) • • • (c»y) [î 0 Soj & . . . & Cnj £ San —> P j (c1( . . ., Cnj) ] .
L e t z b e som e m em ber of s0 (w hich is n o t e m p t y ) ; th en s0 — cQ(z, c).
B y th e d efin itio n o f P j , P 0(d, c) holds if an d o n ly if P *(d*, c*) w here d*
an d c* are th e eq u iv a le n c e classes to w h ic h d an d c, re sp e ctiv e ly , belong.
B u t b y h yp o th esis, P * (sa, sb) ~ a ' — b. H en ce if d Z s a, th en c £ s a, if
an d o n ly if P Q(dt c). Sin ce no sa is e m p ty , (a) (Ed) [d 8 s j . T h u s
| c £ s 0 = Q ( z ,c ) ,
W \ c £ s „ s (Ed) [d Z s a & P 0(d, c)] = (d) [ d Z s a -> P 0(d, c)].
A p p ly in g T h eorem V P § 5 7 or X I * § 58 to th e second line, th is show s
t h a t th e p red icate c 8 sa is gen eral recursive in Q , P 0. In m ore d e ta il: C o n
sider th e second line o f (2) as of th e form
H (c) eee (E d)[K (d) & P 0(d, c)] = (d)[K (d) -> P 0(d, c)].
L e t q, k, p, h, f be fu n ctio n letters expressing th e represen tin g fu n ction s
of Q , K , P 0, H ,c s a, re sp e ctiv e ly . B y T h eorem V I * or X I * , H(c) is gen eral
8
recursive (uniform ly) in K (d ), P Q(d, c), i.e. there is a sy ste m of eq u atio n s
(con tain in g k, p, h) d efin in g re cu rsively (the represen tin g fu n ctio n of)
H(c) from (those of) K (d ), P Q(d, c). B y L e m m a V I § 65, w e can in trod u ce
a p ara m e ter to o b ta in eq u a tio n s E d efin in g re cu rs iv e ly H (c , a) from
K (d ,a ), P Q(d ,c). Then E w ith th e three eq u a tio n s f(c, 0 ) = q ( z , c),
k(d , a ) = f ( d , a), f(c, a ') = h ( c , a) defines cZ sa re cu rsively from Q, p »
as w e see w ith th e h elp of in d u ctio n on a.
Since Q , P 0 are in tu rn gen eral recursive in 1-q u a n tifier pred icates,
so is c 8 s a; an d hence b y T h eo rem X I , c C s a is expressible in b o th
th e 2 -q u a n tifie r forms.
U sin g th e (E x)(y) -expressions for cZ sa an d P j(cv . . . , c n;) in th e
m idd le expression of (1), a d v a n c in g th e q u an tifiers (b y th e inform al
an alo gs of *91 an d *8 7 § 35), an d co n tra ctin g (b y (17) § 5 7 ) , w e o b ta in
an (E x) (y)-expression for P * (sai, . . satij). An (x)(E y) -expression is
obtained similarly, using instead the (x)(Ey\-expression for P (c1, . . cUj)

and the right member of (1) (and applying *91, *96, *95 *87, *98, *97,
(17), (18)).
We now consider also applied predicate calculi with equality, in which

the terms and formulas are constructed using the logical symbolism of
the predicate calculus with certain individual symbols ev . .., eg, function
symbols ilf ..., fr and predicate symbols = , Px, ..., Ps (but no predicate
letters). The number-theoretic definitions of Term' and 'formula’ (§ 17)
provide an example (with q = 1 , r = 3, s = 0). Although commonly each
function or predicate symbol takes a number n > 1 of arguments, we
may allow n > 0 , in which case individual symbols may be included
among the function symbols, and proposition symbols among the pred
icate symbols. We shall write e.g. "f(s, t)” for the term constructed by
placing terms s and t in the respective argument positions of a 2 -place
function symbol f, even though in a given system some other manner of
combining the symbols may be used, e.g. "f” may stand for + and
"f(s, t)” for s+t.
The predicate calculus with equality is dealt with at length in Hilbert-
Bernays 1 9 3 4 pp. 164 ff.
We shall see (Theorem 41 (b)) that in an applied predicate calculus
with equality, Axiom Schema 23 is replaceable by a finite list of particular
axioms without changing the provability and deducibility notions.
The idea has already been used in setting up the number-theoretic system
without Axiom Schema 23; in that case the particular axioms which
replace Schema 23 did not appear as postulates except Axioms 16 and
17, since the rest were deducible from the other number-theoretic axioms.
(C) For the applied case, we read "predicate symbol P other than = ”
in place of "predicate letter” in (A). The open equality axioms for an n-place
function symbol f shall be the n formulas
a= b D f(a2, . . a ^ , a, am , . .., a*) = f(alf . . ., a ^ , b, at-+1, .. ., a„)
(i = 1 , ...,« ). These are deducible in the predicate calculus from
Axiom 22 and axioms by Axiom Schema 23; e.g. (with n = 2 ),
a—-b D f(a, c)=f(b, c) is deducible by *3 and D-eliin. from
a= b D (f(a, c) = f(a, c) D f(a, c)=f(b, c)), which is an axiom by Axiom
Schema 23, and f(a, c)=f(a, c), which is deducible by substitution from
Axiom 22. Except in the case of = , the open equality axioms for a symbol
are what we previously called the special replacement properties (cf.
Theorem 23 § 38). We use "Eq( = , P1} . . ., Pg, iv .. ., f,.)” for the con
junction of the closed equality axioms for = , Plf . . ., Ps, fp . . ., fr.
(D) G iv e n th e e q u a lity ax io m s for = , P x, . . P g, iv . . . , fr, w e can
use th e results of (B) an d th e form er m e th o d of proof to estab lish T h eo rem
24 an d its corollaries for th e case th e p a rt r w h ich is replaced does n o t
sta n d w ith in th e scope of a n y p red icate sy m b o l or letter, or fu n ctio n
sy m b o l, oth er th a n = , P 1? . . . , P s, flt . . . , fr.
(E) In a n y sy ste m in w h ich T h eo rem 24 holds under th e restrictio n

ju s t sta te d , e v e r y a x io m by A x io m S ch em a 23 c o n ta in in g o n ly th e
p red ica te sy m b o ls or le tters = , P x, . . . , P S an d th e fu n ctio n sy m b o ls
iv . . . , fr is p ro v ab le. F o r under th e stip u latio n s for th e schem a, b y
a p p lica tio n s of T h eorem 24 w h ich replace b y b each occurrence of a in
A (a) w h ich entered b y su b stitu tio n for x in A (x ), we o b ta in a=b
b A (a) ~ A ( b ) w ith no v a ria b le s varied . T h e n c e b y * 17 a an d D -in tr o d .,
b a = b D (A(a) D A (b)).
T heorem 4 1. (a) In the predicate calculus with equality , the reflexive

(*100), sym m etric (*10 1), transitive (*102) and replacement (Theorem 24
and corollaries) properties of equality hold.
(b) T b E in the applied predicate calculus with equality having
= , P lf . . . , P 8, iv . . . , fr as its predicate and function sym bols , if and only
if T, E q ( = , P j, . . . , P s, fj, . . . , fr) b E in the predicate calculus.
(c) T b E in the pure predicate calculus with equality , when T, E are
letter form ulas in — , F lt . . . , P s, if and only if T , E q ( = , P 1# . . . , P s) b E
in the predicate calculus.
P roofs, (a) B y (A) — (D).
(b) By ( A ) - (E).
(c) S im ila rly, if w e ca n first ex clu d e th e p o ssib ility th a t th e d ed u ctio n
T b E requires axio m s b y A x io m S ch em a 23 c o n ta in in g oth er p red ica te
le tters P s+1, . . . , Fs+t besides P x, . . . , P s. T h e d ed u ctio n V b E is a
d ed u ctio n in th e p red icate calcu lu s of E from T , a —a an d a xio m s b y
A x io m S ch em a 23. B y th e m eth o d of R e m a rk 1 § 34, w e can th en ce
o b ta in a d ed u ctio n of E from F, a = a an d axio m s b y A x io m S ch em a
23 w h ich co n ta in o n ly th e p red icate letters 'Pv . . . , P s.
E xample 2. A s a fifth ex a m p le for th e C on verse of T h esis I I § 60,

for a p rim itiv e recu rsive R (x ,y ) (cf. E x a m p le 1 §6 0 ), le t <pv
b e a p r im itiv e recursive d escrip tion of th e represen tin g fu n ctio n cp ( = cpfc)
of R. Let i{, . . . , f fc be d istin ct fu n ctio n sym b o ls (to express <px, . . . , <pk,
re sp ectively). L e t th e term s an d form ulas of S be co n stru cted usin g the
lo gica l sym b o lism of th e p red icate ca lcu lu s w ith th e in d iv id u a l sy m b o l 0,
th e fu n ctio n sy m b o ls ', f lf . . . , f*. an d th e p red icate sy m b o l = . L e t S
§74 ELIMINABILITY OF DESCRIPTIONS 405
h a v e th e postu lates of th e p red icate calculus an d as p a rticu la r a xio m s
th e equ ation s o b ta in ed b y tra n sla tin g th e sch em a a p p lica tio n s for th e
description y lt . . . , cpfc (as E x a m p le 1 § 44 w as tra n sla te d to g et E x a m p le 3
§ 54, b u t w ith 1 = 0), to geth er w ith th e open e q u a lity axio m s for = ,
iv . . . , f fc. L e t A (x ) be th e formula 3 y f fc(x, y ) = 0 . In S th e su b stitu tio n
rule R 1 of § 54 holds as a d erived rule (b y § 23), an d likew ise th e re
p la cem en t rule R 2 (by T h eo rem 41 (a) an d (b)). H en ce if, for som e y,
R (x, y) is true, i.e. <$k(x, y) = 0, th en ik(x, y)= 0 is p ro v ab le inS, an d b y
3-in trod ., so is A (x ). C o n ve rsely, A (x ) is p ro v a b le in 5 o n ly if (E y)R (x, y),
as w ill b e show n in T h eo rem 5 2 § 79.
*§ 74. Eliminability of descriptive definitions. A t va rio u s stages

in th e inform al d eve lo p m en t of a m a th e m a tica l th e o ry ad d itio n s m a y
be m ad e to th e sto ck of co n cep ts an d n o tatio n s. If th e d eve lo p m en t is
form alized, a t th e correspon ding stages new form ation rules an d p o stu la tes
are a d d ed to a g iv e n form al sy ste m S 1 to o b ta in an oth er S 2. T h u s th e
form ulas (provable form ulas) of becom e a subset of those of S 2. T h e
new fo rm atio n rules in trod u ce n ew form al sym b o ls or n o tatio n s, an d th e
new p o stu la tes p ro vid e for th eir use d e d u c tiv e ly . W e shall w rite ‘ ‘ hi ”
(“ h2” ) for th e d e d u c ib ility relation in St (in S 2).
U n d er such circum stan ces, w e s a y th a t th e new n o ta tio n s or sym b o ls
(w ith th eir postulates) are elim inable (from S 2 in S 2), if there is an ef
fe c tiv e process b y w h ich , g iv e n a n y form ula E of S 2, a form u la E ' of 5 X
can b e found , such t h a t :
(I) If E is a form ula of S v then E' is E.
(II) h2 E ~ E ' .
(III) If T b2 E , then T hi E '.
H ere P is D [, . . if T is D x, . . . , D ,. W e call (I) — (III) th e elim i

nation relations. (In E x a m p le 13, sligh t m o d ificatio n s are called for in
our form u latio n of ‘ ‘e lim in a b ility ” .)
W h en th e elim in ation relations hold, th en furtherm ore:
(IV) T h2 E , if and only if T f hi E '.
(V) If T, E are form ulas of S v then T h2 E if and only if T hi E .
P roofs . (IV) F o r c o n versely to (I I I) : If V' hi E ', then a fo rtio ri

T' h2 E ' ; th en ce b y (II), F h2 E .
(V) B y (I) w ith (III) or (IV ).

Thus an eliminable extension of St to obtain S 2 is inessential, in that
by (V) it gives no enlargement of the class of the original formulas which
are provable, while by (II) any new formula is equivalent in the enlarged
system to one of the original formulas.
E x a m p l e 1. E l im in a b il it y o f e x p l ic it d e f in it io n s . In studying
our formal system, we regarded 1, <, 3!x, x2, etc. as merely ab
breviations in the presentation of the metamathematics (cf. end § 17).
Alternatively, we could have treated them as successive additions to the
formal symbolism. In this case, each time we should have added a
new formation rule (if A and B are form ulas , so is A ~ B; 1 is a term ;
if s and t are terms, s e t is a form u la ; if x is a variable and A(x) is a
form ula, 3!xA(x) is a form u la ; if s is a term, s2 is a term ; respectively),
and a defining axiom or axiom schema
({(A ~ B) D (A D B) & (B D A)} & {(A 3 B) & (B D A) D (A ~ B)},
1=0', a < b ~ 3c(c'-\-a=b), 3!xA(x) ~ 3x[A(x) & Vy(A(y) D x=y)],
a2= a-a, respectively, with appropriate stipulations about A, B, x, A(x), y).
These additions are eliminable, with ' the operation previously regarded
as unabbreviation. Generally this is the case under the following con
ditions (briefly stated). The defining axiom or schema is an equivalence
(an equation). Then St shall be the predicate calculus (the predicate
calculus with equality) and possibly additional particular axioms and
axiom schemata. Theorem 14 § 33 (Theorem 24 (b) § 38) is used in proving
(II). The two conventions for “permanent abbreviations’' end § 33
shall apply; the first of these is used in treating the case for Rules 9
and 1 2 in proving (III), and the second for Axiom Schemata 1 0 and 1 1 ,
and an additional convention can be adopted to fix which bound varables
are used in the E' for each E. For each additional axiom schema,
whenever A is an axiom of S 2 by it, A' must be provable in S t .
E x a m p l e 2 . Let S 2 be our system of the classical predicate calculus
with the addition of Axiom Schema 9a of Lemma 11 § 24, which by *9 5
§ 35 is redundant in S2. (An axiom or axiom schema or several such are
redundant in a formal system S, if the axioms involved are all provable
in the system left after omitting those axiom(s) or axiom schema(ta)
as postulates.) Let S t be what remains from S 2 when &, V, 3 and their
postulates are omitted. Then the elimination relations hold, when ' is
the replacement of &, V and 3 by the equivalents given by *60 and *61
§ 27 and *83 § 35. (In treating the case for Axiom Schema 6 in the proof
of (III), assume A DC, B d C, n A D B and -iC, and deduce both B
and -i B with the help of * 1 2. Then use -i -introd. and elim., and D-introd.)
E xample 3. In our n u m b er-th eo retic sy ste m , f- - i A ~ A 3 1 = 0.
T h is enables - i to b e elim in ated , if (redundant) p o stu la te s 8 ' (or 8 1') a n d
15" are first ad d ed , w here 8 ' is ((A D 1 = 0 ) 3 1 = 0 ) 3 A , etc. (B u t l b
includ es 7'.)
A n elim in ation theorem ca n b e lo o ked u p o n as a n ew sort o f d e riv e d

rule (§ 20 ) for th e s tu d y of th e origin al sy ste m Sv I t enables us, w h en our
purpose is to estab lish th e p r o v a b ility o f a form u la E in Sv to u n d e rta k e
in stea d to p ro v e E in a su ita b le sy ste m S 2 h a v in g m ore m ach in e ry th a n
Sv A lso b y it a n y form u la E n o t o f S x w h ich w e p ro v e in S 2 ca n b e con
strued as an a b b re v ia tio n for a p ro v a b le fo rm u la E ' o f S v
A n elim in ation theorem h as as corollaries th a t th e sim ple co n siste n cy
of S 2 is im plied b y th a t of St (using (V), cf. § 28) an d th e decision p rob lem
for S 2 is reduced to th a t for S x (b y (IV ), cf. § 30, end § 61).
E xample 4. N on-eliminability of the recursive definition of •.

T h e fu n ctio n sy m b o l • w ith its recursion eq u a tio n s (A xio m s 20 a n d 21)
is not elim in able from th e form alism o f C h a p te r I V , c a ll th a t S 2. F o r
by P resb u rger’s result (beginn ing § 42) th e rem ain in g sy s te m Sx is
sim p ly con sisten t an d its decision problem is solvab le. If • w ere
elim in able, th e sam e w o u ld be true o f S 2, co n tr a d ic tin g T h eo rem 33 § 6 1.
E liminability of descriptive definitions . In a d escrip tiv e d efin ition ,

an o b je c t / is defined as th e w such t h a t F (w ) (in sy m b o ls, iw F (w ))f
where F is a p red icate for w h ich it is kn o w n th a t there is a u n iq u e w
such th a t F(w) (in sym b o ls, (E \w )F (w )). If w is th e o n ly in d ep en d en t
v a ria b le for F , th en iwF(w) is an in d iv id u a l /. I f F d ep en d s on n o th er
in d iv id u a l va riab les, w rite it "" F (x v ..., x nf w )” 9 th e n vwF(xv ..., x nt w)
is a fu n ctio n f(xv ..., x n). W e sh all in clu d e th e first case in th e second b y
con siderin g an in d iv id u a l as a fu n ctio n o f 0 v a riab les (acco rd in gly, w e
ta k e "function s y m b o l’ to in clu d e "individual s y m b o l’ in th is section).
T h e lo gic of d e scrip tiv e d efin itio n s w as tre a te d b y W h ite h e a d a n d
R u ssell 19 10 (pp. 30— 32, 66— 7 1 , 173 — 186 in th e 2 n d ed. 19 2 5 ).
T h e ir e lim in a b ility w as estab lish ed b y H ilb e rt an d B e r n a y s 1934 pp.
422— 457. A n o th e r p roof w as in d ic a te d b y R osser 19 3 9 . T h e se proofs
estab lish th e e lim in a b ility o f iw as a form al operator, th e adm ission of
w h ich in to a g iv e n form alism in tro d u ces all th e po ssib ilities for de
scrip tiv e defin ition s a t once. N o ta tio n a lly it is sim pler to use fu n ctio n
sym b ols. A m a th e m a tic ia n d e v e lo p in g a th e o r y w ill o rd in arily in tro d u ce
n ew fu n ctio n sy m b o ls su cce ssiv e ly as th e need for th em arises. W e sh all
estab lish th e e lim in a b ility of fu n ctio n sy m b o ls in tro d u ced in th is m anner.
408 AXIOM SYSTEMS CH. X IV
I t suffices to consider the in tro d u ctio n of one new fu n ctio n sy m b o l a t a

tim e.
T heorem 42. Let S x have the formation rules of an applied predicate

calculus with equality (§ 73), and let the postulates of S x be those of the
predicate calculus with additional particular axiom s and axiom schemata
such that the equality axiom s for the function and predicate sym bols of S x
are provable [or let S x be sim p ly an applied predicate calculus with equality
and possibly additional particu lar axiom s and axiom schemata ).
L et x 1, . . . , x n, w (n > 0) be distinct variables, and F ( x 1, . . x n, w) be
a form ula which contains free only x v . . . , x n, w , and in which x v . . . , x n
are free for w. Suppose that
(i) 3 !w F ( x I, . . . . x n, w)
is provable in Sv
L et S 2 be obtained from S x by adjoining a new n-place function sym bol f
and the new axiom
(ii) F ( x 1, . . x n, f(x j, . . x n)).
Then the new function sym bol f and its axiom are elim inable, i.e. (I) —
(III) (and h ence (IV) an d (V)) hold for a certain effective correlation '
(w hich w ill b e sp ecified in th e proof), provided each of the additional
axiom schemata has the property that , if E is an axiom of S 2 by it , then E '
is provable in S v
P roof is p ro v id e d b y L e m m a s 25 — 3 1. B y T h eo rem 41 (b), it is im
m a te ria l w h e th er w e con sid er th e lo gic to b e th e p red icate ca lcu lu s or
th e p red icate ca lcu lu s w ith e q u a lity , w hen w e k n o w th e e q u a lity a xio m s
for th e fu n ctio n an d p re d ic a te sy m b o ls to b e p ro v a b le. H en ce from th e
o u ts e t th is is im m a te ria l in th e case of Sv an d it w ill b e in th e case of S2
as soon as w e learn (in L e m m a 27) th a t th e e q u a lity a xio m s for th e n ew
S 2.
fu n ctio n s y m b o l f are p r o v a b le in
L e m m a 25. In the predicate calculus with equality , if u, v and x are

distinct variables , F ( v ), C (v), C(u, v ), A , B , A ( v ), B (v ) and A ( x , v) are
form ulas , u is free for v in F (v ) and C(u, v ), A and B do not contain v free ,
and F (v ) does not contain u or x free :
*18 1. 3 !v F (v ) b 3v[F(v) & C(v)] ~ V v [ F ( v ) D C (v)].
*18 2 . 3 !v F (v ), C (v) bv 3v[F(v) & C (v)].

*18 3 . 3 !v F (v ) b 3u[F(u) & C (u ,u )] ~ 3u[F(u) & 3 v [F ( v ) & C(u, v )]].
*18 4 . 3 !v F (v ) b 3v[F(v) & (A D B (v))] ~ A D 3 v [F ( v ) & B (v )].
*18 5 . 3 !v F (v ) b 3 v [F ( v ) & (A (v) 3 B)] ~ 3 v [F ( v ) & A(v)] 3 B .

*186. (- 3v[F(v) & A & B(v)] ~ A & 3v[F(v) & B(v)].
*187. 3!vF(v) h 3v[F(v) & (A V B(v))] ~ A V 3v[F(v) & B(v)].
*188. 3!vF(v) h 3v[F(v) & -iA(v)] ~ -i3v[F(v) & A(v)].
*189. 3!vF(v) h 3v[F(v) & VxA(x, v)] ~ Vx3v[F(v) & A(x, v)].
*190. h 3v[F(v) & 3xA(x, v)] ~ 3x3v[F(v) & A(x, v)].
(Cf. David Nelson 1947 Lemma 23 pp. 347—348.)
P roofs. *181. Assume 3!vF(v), F(v)&C(v) (for 3-elim.) and F(t)
where t is a new variable (for 3-introd.). Then using *172 §41, v=t;
and hence by replacement, C(t). By 3- and V-introd. and change of
bound variables, Vv[F(v) 3 C(v)]. The variable of the V-introd. was t,
so no variable has been varied. By 3-elim. and 3-introd.,
3 v[F(v) &C(v)] 3 Vv[F(v) 3 C(v)]. For the converse, assume 3!vF(v)
and Vv[F(v) 3 C(v)]. From 3!vF(v), we get 3vF(v). Preparatory to
3-elim., assume F(v).
*183. 3u[F(u) & C(u, u)] h 3u[F(u) & F(u) & C(u, u)] (*37 §27)
1- 3u3v[F(u) & F(v) & C(u, v)] (*80 § 35) \- 3u[F(u) & 3y[F(v) & C(u, v)]]
(*91). The converse can be based similarly on *79, with help from *181,
*4 § 26 (with *69 § 32), etc.
*184. 3!vF(v) h 3v[F(v) & (A 3 B(v))] ~ Vv[F(v) 3 (A3 B(v))]
(*181) ~ Vv[A 3 (F(v) 3 B(v))] (since using*3, h B 3 (A 3 C) ~ A 3
(B 3 O) ~ A 3 Vv[F(v) 3 B(v)] (*95) ~A3 3v[F(v) & B(v)] (*181).
*187. 3!vF(v) h 3vF(v) b 3v[F(v) & (A V B(v))] ~ 3v[(F(v) & A)
V (F(v) & B(v))] (*35) ~ (A & 3vF(v)) V 3v[F(v) & B(v)] (*88, *91 with
*33) ~ A V 3v[F(v) & B(v)] (*45).
*188. Use successively *181, *58b, *86.
L emma 26. In the predicate calculus with the equality axioms for the
function and predicate symbols of Sx only but with f admitted to the symbolism
(a fortiori, in the predicate calculus with equality), the conjunction of (i) and
(ii) is interdeducible with
(iii) f(xx, ..., x„)=w ~ F(x1( ..., x„, w).
Hence (iii) is provable in S2.
P roof . From (ii) and f(xx, . ..,x„)=w , we deduce F(x,, .. .,x n,w)
by the replacement property of equality, which by (D) § 73 requires
only the predicate calculus with the equality axioms for = and the
function and predicate symbols of F(xx, w). From (i), (ii) and
F(xx, .. ., x n, w), we deduce f(xx, ...,x „ )= w by *172. From (iii), we
deduce (i) with the help of *171, and (ii) by substituting f(xx, ..., xB)
for w in (iii) and using Axiom 22 § 73.
410 AXIOM SYSTEMS CH. XI v
F o r th e rest of th e proof of T h eorem 42 , w e shall use “ F ( x x, . . . , x n, w ) ”

u nder th e co n ven tio n w h ich applies to perm an en t ab b re v ia tio n s (cf.
end § 33 ) th a t, for any terms t v . . . , t n, s, “ F ( t x, . . . , tn, s),J shall d en ote
th e result of su b stitu tin g t 1# . . . , t n, s for x 1# . , . , x n, w in F ( x 1, . . . , x n, w)
a fte r any le g itim a te ch an ge of bound va riab les in F ( x 1? . . ., x w, w)
w h ich m akes th e su b stitu tio n free.
Lemma In the system of Lemma 26, the equality axioms for f are
27.
deducible from (iii). Hence they are provable in S 2.
W ere th is n o t so, a d d in g o n ly (ii) to th e axio m s of Sx w o u ld n o t g iv e
e ffe c tiv e use of f in S 2 (cf. T h eo rem 41 (a) an d (b)).
P ro o f . F o r exam p le, if n = 2 (and w ritin g a, c for x x, x 2), w e can

p ro v e a = b 3 f(a, c ) = f ( b , c) thus. A ssu m e a=b. By replacem en t,
F (a , c, f(b, c)) ~ F (b , c, f(b, c)). T h en c e usin g (iii), f(a, c ) = f ( b , c) ~
f(b, c ) = f ( b , c); w hence (A x io m 22 an d *18 b ), f(a, c ) = f ( b , c).
Lemma 28 . Let t lf . . . , t n be terms, v be a variable not occurring in

t x, . . t n, and
be a form ula in which f(tx, . . . , t n) is free for v . Then
C(v)
C (f(tx, . . t w)) ~ 3 v [ F ( t x, . . . , t w, v) & C(v)] is deducible from (iii) in
the predicate calculus with equality, and hence is provable in S2.
P ro o f . A ssu m e C (f(tx, . . . , t n)). S u b s titu tin g in (ii) (cf. Lem m a 26 ),
F ( t x, . . . , t w, f( tx, . . . , t n)). N o w use 3- an d 3 -in tr o d . C o n ve rsely,
assum e (for 3-elim .) F ( t x, . . . , t n, v) & C (v). U sin g (iii), f( tx, . . . , t tt) = v .
B y replacem en t in C (v), C (f(tx, . . . , t n)).
A prim e form u la is one co n ta in in g no lo gical sy m b o l, i.e. here it is a

p red icate s y m b o l w ith term s as argu m en ts. A term or form ula of Sv
i-less. A term of th e form f( tx, . . . , t n)
tv
i.e. one n o t c o n ta in in g f, w e ca ll
w here ..., t n are term s we ca ll an i-term ; an d if t x, . . . , t n are f-less,
w e s a y it is a p la in f-term . An occurrence of a term in a form ula is bound
{free), if it is (is not) w ith in th e scope of som e (any) q u a n tifie r V y or 3 y
w here y is a v a ria b le of th e term .
E xample 5. L e t P an d Q b e p red icate sym b o ls, g b e a fu n ctio n sy m b o l

d istin c t from f , an d x an d y b e d istin ct variab les. In th e fo llo w in g form ula,
th e second occurrence of an f-te rm is b o u n d ; th e c cher six are free.
1. V x { P (f(y ), f(g(x))) & Q (x , x)}
D P (f(y ), f(g(f(y)))) & Q(f(y). %))•

T h e f-term s f(y) an d f(g(x)) are plain, b u t f(g(f(y))) is n ot, as th e f-term
f(y) is n ested w ith in it.
L emma 29. There can be correlated effectively to each form ula E of S 2

a form ula E ' of (called th e prin cipal {-less transform of E ) in such a
S x
w ay that (I) and (II) hold , no free variables are introduced or rem oved ,
an d the operators of the predicate calculus are preserved , i.e. (A D B )'
is A ' D B ' ( A & B ) ' is A ' & B ' , ( A V B) ' is A ' V B ' , ( - i A ) ' is - i A ' ,
(V x A (x ))' is V x A '( x ) (where A '(x ) is (A (x))') a w i (3 x A (x ))' is 3 x A '( x ) .
P roof . T h e co n ditio n th a t th e operators are preserved d efines E'

b y recursion on th e n um ber g of occurrences of lo gical sy m b o ls in E , pro
v id e d w e s u p p ly as basis a d efin itio n of E ' for th e case E is prim e. W e do
th is b y in d u ctio n on th e n u m ber q of occurrences of f-term s in E , thus.
If g = q = 0, E ' shall b e E .
If g = 0 an d q > 0, select th e first occurrence of a p la in f-term
in E , s a y it is f( tx, . . . , t n). L e t v b e a v a ria b le w h ich does n o t occu r
in E . L e t C (v) result from E b y c h a n gin g th e occurrence of f ( t x, . . t n)
u n d er con sideration to v. T h e n E ' shall b e 3 v [ F ( t 1} . . . , t n, v) & C '(v )].
( B y * 1 81 , we could e q u a lly w ell use V v [ F ( t 1, . . . , t n, v) D C '(v )].) T h ere
is an a m b ig u ity here regard in g th e choice of th e b o u n d v a ria b le v an d
th e m ann er in w h ich th e b o u n d v a riab les of F ( x x, . . . , x rt, w) are c h a n g e d
w h e n n ecessary to m a k e th e su b stitu tio n of tv . . . , t n, v for x lf . . . , x n, w
free. B u t d ifferen t le g itim a te choices lead to co n gru en t form u las (§ 33).
W e m a y suppose som e co n v e n tio n su pplied to fix w h ich one is E ' itself.
I n our illu stra tio n s, it w ill suffice to use a n y con gruen t o f E '. N o te th a t
C (v) is prim e an d co n ta in s ju s t q— 1 occurrences of f-term s.
The properties of ' m en tio n ed in th e lem m a follow by a corre
sp o n d in g in d u ctio n on g (using T h eo rem 14 §33 for (II)), w ith in
d u c tio n on q w ith in th e basis (using L e m m a 28).
A s u g g e stiv e a b b re v ia tio n is to w rite “ F^1’••,’t* C ( v ) ,, for
3 v [ F ( t 1, v) & C ( v ) ] .
E xample 5 (continued). F o rm u la 1 is reduced to a c o n g ru e n t o f

its p rin cip al f-less tia n sfo r m thus.
2. V x { F ^ P ( v 2, f ( g ( x ) ) ) & Q ( x , x ) }
=> n ; p ( v 2, % ( % » ) ) & F v 12Q ( v 12, f(y)).
3. V x { F $ t F«W P ( v 2, v 3) & Q (x , x)}
3 Fv2F?„ p(v2. f(g(vn))) & Fv12F^13Q(v12, v13).

4. Vx{ F ; F « P ( v2) v3)&Q( x, x)}
3 Ft u)p(v2, v3) & F ^ F ^ 3Q(v12, v13).
Let us ca ll th e p refixes F^1*-*1* “ F -q u a n tifie r s” . By *18 4 — * 1 9 0
(since (i) g iv e s us th e assu m p tio n form u las 3 !v F (v )), F -q u a n tifie r s ca n b e
p erm u ted in St w ith th e operators of th e p red icate calculus an d w ith
each oth er (ch anging bound v a ria b les as necessary), su b je ct to th e
restrictio n th a t F *1” •,fcw m a y not b e advanced (leftw ard) o ver an or
d in a r y or F -q u a n tifie r w h ich b in d s a n y v a ria b le o f tv ..., tn (since for
* 1 8 9 an d * 19 0 th e F (v ) m u st n o t co n ta in th e x free). T h is restriction as
b e tw e e n tw o F -q u a n tifie r s m eans sim p ly t h a t tw o F -q u a n tifie r s w h ic h
result from e lim in a tin g f-term s one n ested w ith in th e other m u st be k e p t
a lw a y s in th e sam e re la tiv e order (the one corresponding to th e inn er
f-te rm b e in g to th e left). A s b etw ee n an F - an d an o rd in ary quantifier,,
it m eans th a t an F -q u a n tifie r resulting from th e elim in atio n o f a b o u n d
occurrence of an f-term m a y be a d v a n c e d leftw a rd o n ly u p to th e rig h t
m o st o f th e o rd in ary q u an tifiers w h ich b o u n d th e occurrence.
M oreover b y *18 3 , a d ja c e n t like F -q u a n tifie rs F|j*"*,tn an d F ^ - ,tw c a n
b e co n tracted .
A ls o * 1 8 2 co n stitu te s an in tro d u ctio n rule for F -q u an tifiers.
E xample 5 (concluded). F o rm u la 1 is an a x io m V x A (x ) D A (t)

o f S 2 b y A x io m S c h em a 10, w ith f(y) as th e t ; an d w e chose th e sam e
b o u n d va ria b le s in elim in atin g corresponding f-term s of A (x ) an d A ( t)
in th e red u ctio n to F o rm u la 4. N ow b y *18 6 , *19 0 , *18 6 (w ith *3 3 ),
* 1 8 4 an d * 1 8 3 (w ith a ch an ge of b o u n d variables), w e can a d v a n c e th e
th ree F -q u a n tifie r s F * u , F * m, Fîg resulting from th e elim in ation of th e
th ree occurrences of f(y) as th e t of th e sch em a a p p lica tio n to th e fro n t,
a n d c o n tr a c t th em to one F -q u a n tifie r F ^ . T h is g iv e s us th e fo llo w in g
form ula, as an e q u iv a le n t in S x of F o rm u la 4 an d hence of th e p rin cip al
f-less tran sform of F o rm u la 1.
5. F ^ [V x { F ^ F |WP ( v * v 3) & Q (x , x)}

3 F^F*<; i>P(v 2) v 3) & Q(vj, Vl)].
N o w th e scope o f F ^ in 5 is an ax io m of Sx by A x io m S ch em a 10, w ith
vt as th e t. So b y F -in tro d . (*182), F o rm u la 5 is p ro v a b le in Sv an d h en ce
also th e p rin cip al f-less tran sform of F o rm u la 1.
L emma 30. If E is an axiom of S 2, then \~x E '.
P roof . If E is an a x io m o f S 2 b y a n y a x io m sch em a of th e p ro p
o sitio n a l ca lcu lu s, th e n E ' is an a x io m of Sx by th e sam e sch em a, since
' preserves th e operators of th e calcu lu s (L em m a 29).
A xiom S chema 10: E is V x A ( x ) D A (t), w here t is free for x in A (x ).

Case 1: t is f-less. T h e n b y ch oosin g th e sam e b o u n d v a ria b le s for
co rrespo n din g step s in th e red u ctio n s o f A (x ) a n d A (t), w e ca n o b ta in
a co n gru en t of E ' w h ich is of th e form V x B ( x ) D B (t) w here t is free for
x in B (x ). T h is is an ax io m of St b y A x io m S c h e m a 10, w ith th e sam e t.
H en ce hi E '.
E xample 6. Let E be
Vx{P(f(y), f(g(x))) & Q(x, x)} D P(f(y), f(g(t))) & Q(t, t),
w here t is f-less. Then
V x {F ^ F ^ ' P K , v 2) & Q (x , x)} Z> F ^ F ^ P K , v 2) & Q (t, t)
St
k
is co n gru en t to E ', an d is an ax io m of b y A x io m S ch em a 10.
Case 2: th e gen eral case. S a y th a t A (x ) co n ta in s occurrences of

f-term s an d l free occurrences of x , an d th a t t co n ta in s m occurrences of
f-term s. (Illu strated w ith k= 2, l= 3, m= 1 b y E x a m p le 5.) T h e n A (t)
co n ta in s k occurrences of f-term s w h ich o rigin ate from A (x ) an d th u s
correspond to th e f-term s of A (x ), an d Im others co n sistin g o f m in each
o f th e l occurrences of t w h ich ta k e th e p la ce o f th e free occurrences o f
k
x in A (x ). In th e red u ctio n of E to a co n gru en t of E ', w e ca n use F -
q u an tifiers "with th e sam e b o u n d v a ria b le in elim in atin g each o f th e
pairs of correspon ding f-te rm occurrences from A (x ) an d A (t), a n d F -
q u a n tifiers w ith oth er d istin ct b o u n d v a ria b les in e lim in a tin g from A (t)
th e Im f-term occurrences in th e t's. S a y th a t in th e elim in atio n , th ese l
occurrences of t beco m e t * , . . . , t f (v n , v 12, v 13 in th e exam p le). S in ce
none of th e k f-term occurrences in A (t) w h ich o rigin ate from A (x ) ca n b e
n ested w ith in a n y of th e Im w h ich en ter b y th e s u b s titu tio n of t for x ,
an d since th e s u b stitu tio n is free, th e Im F -q u a n tifie r s u sed in e lim in a tin g
th e la tte r ca n b e a d v a n c e d to th e front, an d in such an order t h a t each
gro u p of l of th em b elo n gin g to corresponding f-term occurrences in th e
l occurrences of t are a d ja c e n t, after w h ich each gro u p ca n b e co n tra cte d .
W e th u s o b ta in as an e q u iv a le n t in S x o f E ' a fo rm u la o f th e form
(A) . . . F®^1’--’Swn[VxB(x) D B ( t* ) ] ,
w here t * results from each of t* , . . . , t f b y th e id e n tifica tio n of v a ria b le s

in th e co n tractio n , an d is free for x in B (x ). N o w V x B ( x ) D B ( t* ) is an
a x io m of Sx b y A x io m S c h em a 10, an d (A) is p ro v a b le th en ce b y m
a p p lica tio n s of F -in tro d . (*182).
Axiom (ii): E is F(x1, ..., xn, i(x v ..., xw)). Su p p ose w occurs free I
tim es in F ( x 1, . . . , x n, w). T h e n E ' co n tain s l F -q u a n tifie rs. T h ese ca n b e
advanced and contracted to give 1F^ ,"*,x»F(x1, . . . , x n, v ), i.e.
lv{F(x1, . . x n) v) & F(x1? . . x nt v)}, w h ich is p ro v a b le from (i).
Lemma 3 1. I f E is an im m ediate consequence of F {of F and G) in S 2,
then E' is an im m ediate consequence of F ' {of F ' and G ') in S 2.
F o r each of R u le s 2 , 9 an d 12 , b ecau se ' preserves th e operators, an d
does n o t in tro d u ce free v a ria b les (so th a t th e C of R u le 9 or 12 is n o t
tran sfo rm ed in to a fo rm u la C ' c o n ta in in g x free).
R emark 1. I f f is n o t 0 or ', th e in d u ctio n sch em a satisfies th e

p ro viso of th e theorem for ad d itio n a l ax io m sch em ata, as is seen b y th e
reasonin g used for A x io m S c h em a 10 Case 1 .
E xample 7. Let b e th e n u m b er-th eo retic sy ste m of C h a p te r I V .

Let S 2 result from S x b y ad jo in in g th e fu n ctio n sy m b o l rm (to express th e
rem ainder function ) w ith th e ax io m
3 q {a = b q Jr im { a t b ) & rm {a, b) < b ) V {b = 0 & rm {a, b )= a )

(cf. * 179 b, c § 41). T h e n b y th e theorem (togeth er w ith R e m a rk 1 ), rm
an d its ax io m are elim inable. — L e t th e n o ta tio n [afb] for th e q u o tien t
an d th e a x io m lr {a = b [a /b ]-\-r & r < b ) V b = [ a / b ] = 0 (cf. * 1 7 8 b , c) also
b e ad join ed. B y su ccessive a p p licatio n s of th e theorem , b o th rm an d
lajb] can b e elim inated.
E xample 8. L e t S x b e th e n u m b er-th eo retic form al sy ste m , R ( x , y)

b e a form u la c o n ta in in g free o n ly x an d y , an d R{x, y) be th e p red icate
w h ich R (x , y) expresses under th e in terpretatio n , (a) Consider th e
classical sy ste m Sv an d suppose R is such th a t: (A) hi 3 y R ( x , y ).
T h e n u sin g * 1 4 9 a n d * 1 7 4 b , a fu n ctio n sy m b o l f in tro d u ced w ith th e
a x io m R ( x , f(x)) & V z ( z < f ( x ) D - i R ( x , z)), so th a t under th e (classical)
in terp reta tio n (at least) f(x) expresses th e fu n ctio n [iyR{x, y) (beginning
§ 5 7, assu m in g a b o u t S ± th a t (A) im plies (l b) § 57 (w ith n= 1)), w ill b e
elim in able, i.e. th e elim in atio n relations h old from th e sy ste m S2 th u s
o b ta in ed . U sin g * 1 4 9 a in stea d of *14 9 , th e sam e holds in the in tu itio n istic
Sv w h en besides (A) also: (a) hi R (x , y) V i R ( x , y ). (b) N o w consider
th e classical Sv w ith o u t supposin g (A). Let R t( x , w) be R (x , w) V
{ - i 3 y R ( x , y) & w = 0 } , an d R t ( x , w) b e R t( x , w) & V z ( z < w D - i R +(x, z)).
By *5 1, hi 3 y R ( x , y) V i 3 y R ( x , y). Thence usin g cases (V-elim.),
hx 3 w R t( x , w ); an d b y *14 9 an d *174 b : ( 1 ) hi 3 !w R t(x , w). So a
fu n ctio n s y m b o l f in tro d u ced w ith th e ax io m R t ( x , f(x)) w ill a lw a y s
b e elim in able classically. U n d er th e (classical) in terpretatio n , now f(x)
expresses s)>R{x, y) (beginn ing § 62; cf. (59) §63). T h e elim in a b ility holds
in th e in tu itio n istic Sv w h en R is such th a t (a) holds an d also:
((3) hi 3 y R ( x , y) V - i 3 y R ( x , y). F o r ((3) enables us to dispense w ith * 5 1 ,
an d (a) an d (($) w ith * 1 5 8 an d R e m a rk 1 (b) § 29 g iv e hi R *(x , w) V
§ 74 ELIMINABILITY OF DESCRIPTIONS 415
n R ^ ( x , w), w h ich enables us to use * 1 4 9 a in stead of *14 9 . (c) L e t <p(x) b e
<p(x)=w is a rith m e tic a l
a n y fu n ctio n of w h ich th e represen tin g p red icate
(§ 48). L e t R ( x , w) be a form u la expressing<p(x)=w. T h e n th e f(x) o f (b)
expresses zw [(p[x)= w ], i.e. <p(x). T h u s in the classical number-theoretic
system , for any function (p[x) such that cp(x)=w is arithm etical, a form ula
F (x , w) containing free only x and w can be found such that F ( x , a;) is true
under the interpretation exactly when w = <p(x), and a new function sym
bol f expressing 9 with the axiom F ( x , f(x)) is elim inable. In p a rticu la r,
such a form ula F (x , w) can be found for any general recursive function y(x)
(b y T h eorem V I I (b) § 5 7 ); and for m any functions [under classical inter
pretation) which are not general recursive, e.g. z y T t [x, x, y) (cf. (b) or b e
gin n in g § 62, an d E x a m p le 1 § 63). (d) C o n versely, if an a x io m of th e form
F (x , f(x)) ch aracterizes f(x) as expressin g a fu n ctio n <p(x) (i.e. if for each
x, F (x , w ) is true under th e in terp retatio n e x a c tly w hen w = (p[x)),
th en (p[x) = w is arith m etical. — T h e situ a tio n for th e in tu itio n istic
sy ste m w ill be considered fu rth er in § 82 (E x a m p les 1 an d 2 ).
F o r th e n e x t tw o exam p les, w e sh all suppose th a t ( 1 ) — (3) of R e m a r k

1 § 49 h a v e been establish ed , either b y form alizin g § 48 d ire c tly , or b y
b orrow in g fro m H ilb e rt-B e r n a y s 1934 pp. 401— 419. The H ilb e rt-
B e rn a y s tre a tm e n t is in a form al sy ste m w h ich (disregarding o b v io u s ly
inessential differences, in clu d in g their use of p red icate va riab les, cf. end
§ 37) results from our classical n u m b er-th eo retic sy ste m b y a d d in g an
o perator (w ith ap p ro p riate postu lates) w h ich , ap p lied to a fo rm u la
R (x , y ), g iv e s a term exp ressin g th e fu n ctio n zyR (x, y) w here R (x, y) is
th e p red icate expressed b y R (x , y ). B y E x a m p le 8 (b), each use of th is
o perator can be elim inated. A p p ly in g th e process to th e form ulas
d isp la y e d fifth on p. 4 16 an d se v en th on p. 419, ta k in g th e a and
h(n, p(m, n f• / + 1)) to be sim p ly a va ria b le w , w e are led to (a) an d (p) o f
R e m a rk 1 § 49. T h is tre a tm e n t can be a d a p te d to th e in tu itio n istic sy ste m .
(H ilbert an d B e rn a y s w rite th e o perator [ix; their ex has an o th er m ean in g,
cf. 1939 pp. 9 f f ., H ilb ert 19 2 8 .)
E xample 9. E liminability of further primitive recursive

definitions , having those for + and • (cf. E x a m p le 4). Su p p o se
9 X, . . . , 9 k is th e p rim itiv e recursive description of a fu n ctio n 9 ( = 9 *.).
Let S x be th e n u m b er-th eo retic sy ste m w ith ad d itio n al fu n ctio n sy m b o ls
fx, . . . , f ^ (expressing 9 X, . . . , <pk__v re sp ectively), an d w ith eq u a tio n s
o b ta in ed b y tra n sla tin g th e sch em a ap p licatio n s for <pv . . . , 9 *.^ (as
in § 54) ad jo in ed as axiom s. S a y e.g. th a t 9 com es from <{/, x (where <J/, x
are from th e list <pv . . . , 9 *_1) b y S ch em a (Vb) § 4 3 w ith n= 2 . A p p ly in g
R em ark 1 §49: ( 1) hi P (0, x, w) ~ Q(x, w), (2 ) hi P ( y ', x, w) ~

3 z [ P ( y , x , z ) & R ( y , z , x , w ) ] , (3) hi 3 !w P (y , x , w). W e also assum e (as h y p o
th esis of an in d u ctio n on k) : (4) hi g(x)= w ~ Q (x , w), (5) hi h (y , z, x ) = w
~ R ( y , z, x , w) (where g, h express <{/, re sp ectively). T h e theorem
(w ith (3 )) tells us t h a t th e ad d itio n o f f to S x w ith P (y , x, f(y , x)) as th e
a d d itio n a l a x io m is elim in able. B u t in w ith f ad d ed to th e sym b o lism ,
P (y, x, f(y , x)) is in terd ed u cib le w ith f(0 , x ) = g ( x ) & f(y ', x ) = h ( y , f(y , x ), x),
as w e e a s ily see (using ( 1 ) — (5), an d in one d irection , in d u ctio n on y ).
H e n ce the p air of eq u a tio n s f( 0 , x ) = g ( x ) , f(y ', x ) = h ( y , f(y, x), x) m a y b e
u sed in stea d of the fo rm u la P ( y , x , f(y , x)) as th e a d d itio n a l axiom s, w ith
th e sam e results. (This p air of eq u a tio n s is o b ta in ed s lig h tly d iffe re n tly
in H ilb e r t-B e r n a y s 1934 on p. 421.)
E xample 10. Let Sx b e th e n u m b er-th eo retic form alism . By th e

pro o f o f T h eo rem 32 (a) § 59, there is a form u la P (z, x , w) such th a t,
if e is a G o d e l n u m b er of a gen eral recu rsive fu n ctio n <p(x), th en P (e , x , w)
nu m eralw ise represents <p(x). T h e n 3 !w P (e, x , w) is p ro v a b le for each
n a tu r a l n u m b er M u st 3 !w P (e , x , w) be p ro v a b le ? (It c e rta in ly is for
som e ch oices of 9 an d e, e.g. b y fo rm alizin g v ia § 56 reasoning g iv e n in
§ 54.) B y T h eo rem 31 § 52, { hi 3 !w P (z, x , w )} = ( E y)R (z , y) for som e
p r im itiv e recu rsive R. L e t 0 b e o b ta in e d from th is R b y T h eorem X IV
(b) § 60. T h e n 0 is p rim itiv e recursive, an d 0(y) for y = 0, 1 , 2 , . . . is an
en u m eratio n of th e n um bers z for w h ic h 3 !w P (z , x , w) is p ro vab le. U n d e r
th e in terp reta tio n , 3 !w P (z , x , w) is tru e o n ly w h en z is th e G o d el n u m b er
o f a gen eral recursive fu n ctio n o f one va riab le. Su p p ose th a t S ± h as th e co n
sis te n c y p ro p e r ty t h a t 3 !w P (z , x , w) is p ro v a b le o n ly then . N o w for each y ,
0(y) is th e G o d e l n u m b er of a gen eral recursive fu n ctio n , w h ich ca n b e
w r itte n ®i(0(y), x) (begin n in g § 6 5 ) ; an d so ®i(0(x), z) + l is a gen eral
recu rsive fu n ctio n <p(x), for a n y G 5 d e l n u m ber e of w h ic h 3 !w P (e, x , w)
is u n p r o v a b le in Sv — F o r such an e, le t S 2 b e o b ta in ed from S* b y
a d jo in in g a n ew fu n ctio n sy m b o l f w ith th e a x io m P (e , x , f(x)). T h e n f
(and its axiom ) are n o t elim inable. F o r, usin g * 1 7 4 a an d th e result (3) of
R e m a r k 1 § 49 for th e p rim itiv e recursive fu n ctio n U (cf. th e p roof of
T h eo rem 32), w e e a sily sh ow t h a t P (e , x , t) hi 3 !w P (e, x , w). H en ce,
u sin g th e n ew ax io m , h 2 3 !w P (e , x , w). If f w ere elim inable, th e n
S v w h ic h is n o t th e case. T h u s
3!w P (« , x , w) w o u ld b e p ro v a b le in
in the number-theoretic system , there is a form ula P (x , w) containing free
only x and w such thaty under a consistency assu m ption : P (x , w) num eral-
w ise represents a general recursive function 9 , but a new function sym bol
f {expressing 9 ) w ith the axiom P (x , f(x)) is not elim inable.
F o r a g iv e n F , f an d form ation rules (here th o se o f S 2), w e sh all m ean
by a transform o f a form ula E a n y fo rm u la D such t h a t E ~ D is
d ed u cible in th e p red icate calcu lu s w ith e q u a lity from th e form u la (iii).
R emark 2. (a) The tran sform s of E includ e its p rin cip al f-less
tran sform E ' (b y L e m m a 28 an d th e proof o f L e m m a 29) an d all form ulas
o b ta in ab le th en ce b y m an ip u latio n s of F -q u a n tifie r s b ased on L e m m a 25
(b y L e m m a 26). (b) A n y tw o f-less tran sform s E 1 an d E 2 o f a fo rm u la
E o f S 2, since eq u iv a le n t in S 2 (b y L e m m a s 26 an d 27), are e q u iv a le n t in S 1
(b y (V)). T h is holds for a n y S* for T h eo rem 42 w ith th e g iv e n form ation
rules (w ith ou t th e proviso for ad d itio n al a x io m sch em ata, since th e p roof
of E 1 ~ E 2 in S 2 uses ax io m s b y these o n ly o f S t), an d in p a rtic u la r w hen
S j is s im p ly th e p red icate calculus w ith e q u a lity an d (i) as an axio m .
R eplaceability of undefined functions by predicates. T heorem

43. (a) Let S 2 be an applied predicate calculus with equality which has a
sym bol f for a function of n variables (n > 0); and let S t come from S 2
by om itting f and su pplyin g instead a sym bol F for a predicate of n -\-1
variables with the axiom 3 !w F ( x x, . . . , x n, w) [or the postulates of S 2
m ay be those of the predicate calculus with the equality axiom s for the function
and predicate sym bols of S 2> in which case in form ing we also om it the n
equality axiom s for f and su pply instead the n -\-1 equality axiom s for F).
For any form ula E of S 2, let E ' be any particular f -less transform of E
{with the present f and F, under form ation rules allowing both in the sym
bolism). F or any form ula E of S v let E ° be the form ula of S 2 obtained from E
by replacing sim ultaneously each part of the form F ^ , . . . , t w, s) where
t lt . . . , t n, s are terms by f ( t x, . . . , t n) = s . T h en :
(V ia ) hi E ~ E 0/. (V Ib ) h2 E ~ E '° .
( V ila ) {T h E } -> { F h {V h. E } -> { T ° h 2 E °} .

E '} . ( V llb )
(b) L ikew ise when S 2 has additional particular axiom s B x,
and axiom schemata 93x, . . . , 9 3 * , and S 1 has as additional particular
axiom s B{ , . . . , B^. and additional axiom schemata 2li, . . . , such that ,
if E is an axiom of S 2 by 93*- (of S t by 21*), then hi E ' ( h2 E °). (Cf. H il-
b e rt-B e r n a y s 1934 pp. 460 ff.)
F ro m (V ia ) — ( V llb ) it follow s th a t:
( V illa ) {r h2 E } se {F hi E '} . (V U Ib ) {T hi E } » {T ° h2 E °} .
P roofs. ( V illa ) . F o r co n versely to (V ila ) : If F hi E ', th en b y

( V ll b ) , F ° h2 E '° ; an d th en ce b y (V Ib ), V h2 E .
P roof of T heorem 43. (a) We begin w ith the version in which

the logic is the predicate calculus with equality. Let S3a come from S2
by adding F to the symbolism with i(xv . . ., xw) = w ~ F(x1, ..., xn, w) as
axiom. Regarfling this axiom as an explicit definition of F, the additions
are eliminable with ° as the correlation, by Example 1. But by Lemma 26,
the axiom f(x1, . . xn)= w ~ F (x 1, . .., xn, w) can be replaced in S3a
by the pair of axioms 3!wF(x1, .. ., xn, w) and F(xx, . .., xn, i ( x lt . . ., xn))
without changing the deducibility relationship. So the elimination
relations hold also from the resulting system S3, i.e.:
(lb) If E is a formula of S2, E° is E.
(lib) b E ~ E ° . (IHb) {r b E} {r° h E°}.
But Theorem 42 applies to S 3 (as its S2), and the result of the elimi
nation of f from S 3 is Sv Thus, using Remark 2 (b) if ' indicates some
other than the principal f-less transform:
(la) If E is a formula of Sv E' ~ E.
(Ha) h3 E ~ E '. (Ilia) {F b E} -> {Ff b E'}.
Now (Via) — (Vllb) follow, e.g.:
(Via) By (lib) and (Ha), b E ~ E ° ~ E ° '. But E ~ E°' is a
formula of Sv Hence by (Va), hi E ~ E°'.
(Vila) {F b {r b E} (a fortiori) -> {Ff b E'} (by (Ilia)).
The version with equality axioms follows by Theorem 41 (b), or can
be treated directly thus. The equality axioms for F are provable in S3a.
To pass to S3, we first add them as axioms, then replace (iii) by (i)
and (ii), and finally omit the equality axioms for f (using Lemma 27).
R em a rk 3. Any two f-less transforms E1 and E2 of E are equivalent
in Sv by the proof. Since this equivalence is already established in
the system Sx of (a) of the theorem, it suffices to satisfy the conditions
of (b) for any one convenient way of choosing the transforms.
R em a rk 4. For the version with equality axioms, and for F a pred
icate symbol: The entire discussion beginning with Theorem 42 and
including the definition of Transform" holds good, when, for certain
values of i, we exclude the formation of terms f(t1? . . ., tn) and formulas
F(tx, . . ., t n, s) with t{ for any of these values of i containing f, and
omit the equality axioms for f and F for these values of i.
R em a rk 3. With equality axioms, in Theorem 43 (b) the additional
postulates may make some of the n equality axioms for f redundant in
S2. Are the corresponding equality axioms for F redundant in 5X? Sup
pose more particularly that for certain values of i there exist proofs of
th e i- t h e q u a lity axio m s for f from th e rem ain in g p o stu la tes of S 2 in
w h ich proofs no f-term s f( t1, . . . , t n) occu r w ith t i for a n y o f these v a lu e s
of i co n ta in in g f. T h e n th e e q u a lity a x io m s for F for these v a lu e s o f i
are red u n d a n t in Sv p ro v id e d th a t (b) is satisfied w hen ' m eans p rin
cip a l f-less tran sform , or a n y o th er f-less tran sform in th e altered
sense o f R e m a rk 4. F o r th e n b y R e m a rk 4, u sin g (V II) th e p rin cip al
f-less tran sform s o f th e e q u a lity a x io m s in qu estion for f are p ro v a b le
in th e a ltered S lt an d th en ce th e correspon ding e q u a lity axio m s for F .
E x a m pl e 11. (a) L e t S 2 be th e full n u m b er-th eo retic sy ste m or

R o b in so n 's (L em m a 18b § 49). B y (b) of th e theorem , w e ca n replace th e
fu n ctio n sy m b o l • b y a p red icate sy m b o l, in th e fo llo w in g m anner.
Say the new p red icate s y m b o l is •, w ritte n p reced in g its three argu m en ts.
We adjoin as new ax io m s
a=b D ( - { c ,d ,a ) Z )- (c, d, b )), 3!c • (a, b, c ) ;
an d if w e use p rin cip al -le s s transform s, w e ch an ge A x io m s 20 an d 21 to
1 b [-(a , 0, b) & b = 0], 3c [• (a, b \ c) & 3 d [-(a , b, d) & c = d + a ] ] ,

re sp e ctiv e ly , b u t these m a y b e sim plified to
• (a , 0, 0), 3d[-(a, b ', d + a ) & • (a, b , d)].

T h e oth er tw o e q u a lity axio m s
a=b D (-(a, c ,d )0 - (b, c, d)), a=b D (• (c, a, d)Z> • (c, b, d))

for • as a p red icate sy m b o l need n o t b e ad jo in ed , if S 2 is th e fu ll n u m b er-
th eo retic sy ste m , since th e y are in fa c t p ro v ab le in th e sy s te m d escribed
(b y R e m a rk 5 w ith th e proofs of th e e q u a lity ax io m s for • as a fu n ctio n
sy m b o l § 38); b u t if S 2 is R o b in so n 's syste m , w e ad jo in th em in p la ce of
th e form ulas of *10 6 an d * 1 0 7 as axiom s. B y a second a p p lic a tio n of th e
theorem w e can further replace + . In d oin g so, if A x io m 18 is ch a n g e d to
its p rin cip al + - le s s tran sform or to + (<z, 0, a), a—a w ill still b e p ro v a b le
(and w ill rem ain so if 0 is replaced sim ilarly under (b) below ). B y a th ird
a p p lica tio n w e ca n fu rth er replace ' . I n d o in g so, w e ch a n ge th e in d u ctio n
sch em a s a y to A(0) & V x (A (x ) D 3 y [ '( x , y) & A (y)]) D A (x ). Thus we
o b ta in a sy ste m 5 X w ith o u t fu n ctio n sym b o ls in th e o rd in a ry sense, i.e.
for n > 0, related to S 2 b y (V ia ) — ( V U I b ) , w here n o w ' an d ° d en ote
successive elim inations o f several sym b ols, (b) I f w e w ish a sy ste m la c k in g
also in d iv id u a l sym b ols, w e can still further replace 0 b y an ap p licatio n
w ith n = 0, or w e can elim in ate 0 before re p la c in g ' (next exam ple).
E x a m pl e 12. U sin g * 1 3 7 an d A x io m 15, 0 —b ~ V a (a '^ b ) (which is

of th e form (iii)) is p ro vab le. T h is fa c t ca n b e u tilized to elim inate 0 b y
T h eorem 42 after som e p relim in ary tran sform ations.
E xample 13. E liminability of a defined sort of variables.

Let St be the predicate calculus (say under formation rules using indi
vidual, function and predicate symbols) with additional particular axioms
and axiom schemata. Suppose that for a certain formula M(w) containing
free only w, the formula 3wM(w) is provable in Sv Let S 2 come from St
by adjoining a new sort of variables a, b, c, . .., admitting these in
constructing terms and formulas and for Rules 9 and 1 2 , restricting the
x for Axiom Schemata 1 0 and 11 to be a variable of the original sort, and
adding the following three axiom schemata, where x is a variable of
the new sort, A(x) is a formula, and t is a term free for x in A(x) and
for w in M(w).
0. M(x). 10. M(t) D (VxA(x) 3 A(t)). IT. M(t) D (A(t) D 3xA(x)>.
Given any formula E of S2, let El result from E by replacing each part
of the form VxA(x) by one of the form Vx[M(x) D A(x)], and of 3xA(x)
by 3x[M(x) & A(x)], where x is a variable of the original sort not oc
curring in A(x) or M(w); then let Et come from El by prefixing
M(yx) & ... & M(ym) D where yv . . ., ym are exactly the distinct
variables of the new sort which occur free in E (and hence in E l); and let
E' come from Et by substituting for ylf .. ., ym distinct variables
yv .. ., y m of the original sort not occurring in Et (and hence not in E).
Then (I) — (III) hold with the following modifications, provided that
for each additional axiom schema of Slt to each axiom A of S2 by it,
A' is provable in Sv For m > 0 , (II) becomes: E yi- y”H hl1-*™ E'.
In (III), if corresponding variables are to be held constant for cor
responding assumption formulas, for each y the same y should be
substituted in E and each of T for which y is held constant (so in this
case the operation ' is not specified for each single formula by itself).
Before treating (II), Theorems 1, 2 and 14 can be extended appropriately
to S2. We can use Lemma 8 a § 24 to reduce (III) to the case of it with
T empty. To treat this, write El as <<El(y1, .. ., ym)”. We can show by
induction that, if h2 E, then M(yx), ..., M(ym) hi Et(y1, ..., y m) holding
yv ..., y m constant. (For Rule 2 , say e.g. Al contains only one new
variable y, write it “Al(y)”, and Bl none. By hyp. ind., M(y) hi Al(y)
and M(y) hi Al(y) D Bl. By Rule 2, M(y) hi BL By 3-elim. and *74,
3wM(w) hi BL But hi 3wM(w). Hence hi BL) — Similarly intro
ducing sevefal sorts of variables successively.
§ 75. Axiom systems, Skolem’s paradox, the natural number
sequence. Suppose we are dealing with an axiom system (§ 8 ) having
§75 skolem’s paradox , the number sequence 421
as its primitive or undefined notions a set or domain D of individuals,
certain individuals zlf ..., zq from Df and certain predicates Plt ..., Ps
over D. An axiom of the system which can be expressed by a formula in
the symbolism of the predicate calculus (i.e. the restricted or first order
predicate calculus, cf. § 37) with individual symbols ev ..., eq to express
zl9 ..., zq and predicate symbols Pr1# ..., Prs to express Plt ..., Ps,
respectively, we call elementary. If each axiom is elementary, and the
axioms are finite in number, we call the axiom system elementary or
of first order.
Now we can always choose closed formulas to express elementary
axioms, since under the generality interpretation any formula is syn
onymous with its closure (cf. end § 32). Moreover, since the axioms of
an elementary system are finite in number, we can form the conjunction
F(ex, ..., eq>Prx, ..., Prs) of the closed formulas expressing the axioms.
Finally, by changing the individual symbols ev ..., ea to respective
distinct variables zlf ..., zq not occurring in F(ex, ..., etf, Prx, ..., Pr„)
and the predicate symbols Prx, ..., Prs to distinct predicate letters
Pj, ..., Ps, we obtain a predicate letter formula F(z1, ..., zQ, Px, ..., Ps)
or briefly F. A simple illustration (with q = 0 , 5 = 1 ) has already been
given in § 37.
Expressing the axiom system thus by a predicate letter formula
F helps to emphasize the standpoint of formal axiomatics (§ 8 ), from
which the set D, the individuals zlt ..., zq and the predicates Pv ..., Ps
of the axiomatic theory are undetermined except as the axioms charac
terize them. Every predicate letter formula can be considered as express
ing an axiom system with the free variables and predicate letters oc
curring in it representing the undefined individuals and predicates.
When we interpret the logical symbols classically, and treat the
predicates “extensionally” as simply logical functions, the notions of
set-theoretic predicate logic (§ 37) become applicable to the discussion of
axiomatic systems. To say that the axioms are satisfied (in the intuitive
sense, § 8 ) by some non-empty system of objects (which we also ex
pressed in § 8 by saying that the axioms are ‘non-vacuous') now means
exactly that the formula F is satisfiable (in the set-theoretic sense) in some
non-empty domain. Predicate letter formulas which are satisfiable but
not valid are the ones which are of interest as axiom systems. A valid
predicate letter formula does not restrict or characterize the individuals
and predicates expressed by its free variables and predicate letters; but
rather expresses a law of logic applicable to all choices of those individuals
and predicates in any non-empty domain.
In fo rm al a x io m a tics, w ith o u t th e further step of form alizin g th e proc
esses of logical d ed u ctio n so as to o b ta in a form al sy ste m (§ 15), theorem s
are d ed u ced from th e ax io m s on th e basis of th e m eanings of th e lo gical
sym b o ls. W h a t it m eans on th is basis for a proposition to b e a theorem
can b e expressed in se t-th e o re tic p red icate logic as follow s. Consider a n y
p rop osition of th e a x io m a tic th e o r y w h ich is expressible b y a p red icate
le tte r form u la in P x, . . . , P s. W e can a lw a y s choose th a t form u la to
co n ta in no va ria b le s free oth er th a n z lf . . . , z G, an d to co n ta in these o n ly
free. N o w a n y such form u la B ( z 1, . . . , z G, P x, . . . , P s) or b rie fly B e x
presses a tru e p roposition or theorem of th e a x io m a tic th eo ry, p recisely
if every assign m en t of in d iv id u a ls zlt ..., zq from som e n o n -e m p ty
d om ain D to zv ..., z q an d of pred icates Pv ..., P s o ver D to P 1? . . . , P s
w h ich satisfies F also satisfies B . In v ie w of th e v a lu a tio n ta b le for D
§ 28, th is is e q u iv a le n t to sa y in g th a t th e form ula F D B should b e v a lid
in e v e r y n o n -e m p ty dom ain.
N o w suppose t h a t th e a x io m a tic th e o ry is form alized (§ 15) b y a d o p tin g
th e d e d u c tiv e rules of th e p red icate calculus as th e m eans of d ed u cin g
theorem s, under th e p roviso t h a t th e va riab les zv ..., z q be held c o n s ta n t;
i.e. w e n o w s a y p ro o f-th e o re tic a lly th a t B expresses a theorem , if in th e
p red icate calcu lu s F \- B w ith zv ..., z Q held co n sta n t. ( B y R e m a rk s
1 an d 2 (a) § 34, w e ca n th e n a lw a y s fin d a d ed u ctio n of B from F in
w h ich no p red icate le tters oth er th a n P 1? . . . , P s occur an d z lt ..., zq
occur o n ly free.)
B y th e D -ru les, n o tin g th a t, since F con tain s no va riab les free e x c e p t
z ly . . . , z G, no v a riab les are varied , this is e q u iv a le n t to sa y in g th a t
h F D B in th e p red icate calculus.
W e are n o w in a p o sitio n to estab lish th a t th e fo rm alizatio n of d ed u ctio n
for e lem en ta ry a x io m a tic theories b y th e p red icate calcu lu s is b o th
correct (or con sistent) an d a d e q u a te (or com p lete), i.e. th e p red icate ca l
culus enables o n ly an d all those form ulas to b e d ed u ced from F w h ich
express propositions t h a t are true of a n y sy ste m s a tis fy in g th e axiom s.
For {[- F d B} = { F d B is v a lid in every n o n -e m p ty dom ain},
b y T h eo rem 21 § 3 7 an d C o ro llary 1 T h eo rem 34 § 7 2 .
T h e q u estio n w h e th er th e theorem s are con sisten t w ith th e axio m s
(just answ ered affirm a tiv e ly ) is of course q u ite separate from th e q u estio n
w h e th er th e a xio m s th em selves are con sistent. Prior to H ilb e rt's proof
th e o r y or m eta m a th e m a tics, proofs of co n sisten cy of an a x io m a tic
sy ste m or th e o ry w ere b y e x h ib itin g a m o del for th e th e o r y (§ 14). T h e
co n sisten cy p ro p e rty p ro v ed im m e d ia te ly in th is case is th e sa tisfia b ility
of F in som e n o n -e m p ty dom ain.
§75 skolem’s paradox , the number sequence 423
We gave a heuristic argument in § 14 that this property implies con
sistency in the sense of non-existence of a contradiction (one theorem
denying another) in the theory deducible from the axioms. The converse
that from any unsatisfiable system of axioms a contradiction must
necessarily follow by a finite number of logical steps was then by no
means clear.
It is only with the step taken by the modern formalists of formalizing
deduction that consistency in the sense of non-deducibility of a contra
diction becomes amenable to exact discussion. We now Have as the con
sistency property that for no formula A, both F h A and F b nA with
zv . . zq held constant.
By the -i-rules § 23 (since F contains only zlt ..., zQfree), this property
is equivalent to 'not (- -iF ’, i.e. to ‘F is irrefutable’. The formalist’s
transformation of the consistency problem may thus be described (for the
case of elementary axiom systems) as the replacement of satisfiability
by irrefutability.
By Theorem 21 (which takes the place now of the reasoning given in
§ 14) and Godel’s completeness theorem (Theorem 34), satisfiability
and irrefutability are equivalent.
The purpose of the formalistic transformation in the notions of de
ducibility and consistency is to obtain notions which are finitary.
A reduction from the non-enumerably to the enumerably infinite is
achieved, as validity and satisfiability refer to the totality of logical
functions, which is non-enumerable, while the proof-theoretic equivalents
provability and irrefutability refer only to the enumerable infinity of
formal proofs. In metamathematics, the reasoning with the notions is
also finitary. Although the equivalence proof, as given by Godel’s com
pleteness theorem, cannot belong to metamathematics, it is significant
for metamathematics that the set-theoretic notions are actually equiva
lent to the proof-theoretic ones when one reasons on the non-finitary
plane to which the set-theoretic notions belong.
We see now that the decision problem for provability in the pure
predicate calculus includes the decision problem for provability in every
axiomatic theory having an elementary axiom system (by asking whether
a certain predicate letter formula F D B is provable), and also the de
cision problem whether any given elementary axiom system is consistent
(by asking whether n F is unprovable).
Axiom systems used in mathematics often employ = in the role of
an ordinary or logical term which must be understood in advance
rather than as one of the undefined predicates which the axioms charac-
terize. T h e foregoing rem arks w ill a p p ly , if w e first s u p p ly som e ad d itio n al
axio m s for e q u a lity , form alized as E q ( = , P r^ . . . , P r s) or E q (Q , P x, . . . , P s) ;
or w e can in stead form alize th e ax io m s as th e y sta n d b y an e q u a lity an d
p red icate le tte r form ula, an d d ed u ctio n from th e axio m s b y th e p red icate
calcu lu s w ith e q u a lity , an d th en use th e exten sio n of G o d el's com pleten ess
theorem to th a t. T h e tw o m eth od s g iv e results w h ich are e q u iva le n t v ia
L e m m a 24 (a) an d T h eorem 41 (c) § 73, th o u g h se t-th e o r e tic a lly th e form er
does not narrow th e in terp retatio n s w h ich sa tisfy th e axio m s to those in
w h ich Q is e q u a lity b u t allow s Q also to be an eq u ivalen ce relation.
E xample 1. T h e ax io m sy ste m L I — L 3 for linear order (end § 8)

is expressed b y th e fo llo w in g e q u a lity an d p red icate le tte r form u la
(call it “ F ( = , <£?)"), w ith <3 expressin g < :
V a V £ V c [< C ? M ) & a (b ,c ) D & (a,c)] & V * V J [ - i (<3(<i ,£) & a=b) &
-i ( £ l (a ,b ) & 3 l {b ,a )) & -i (a = b & ba 9 ))] & [^{ a ,b ) V a = b V d {b ,a )].
T h e sam e axio m sy ste m is expressed b y th e p red icate le tte r form ula
E q (B , <£?) & F (B , <C3), w ith S expressing = (cf. E x a m p le 1 § 73).
O u r rem arks also a p p ly in d ir e c tly to ax io m sy ste m s h a v in g fu n ction s

flt ..., fr am on g their p rim itiv e notions, as w ith th e h elp of = these
can b e replaced b y th e represen tin g predicates of th e function s, as w as
discussed from th e p ro o f-th eo retic stan d p o in t in T h eo rem 43 § 74.
E le m e n ta r y axio m sy ste m s occur fre q u e n tly in m a th em a tics, if w e use
th e term 'e le m e n ta ry ' m ore w id e ly to includ e sy ste m s w h ich can b e
tran sform ed b y w ell-k n o w n d evices so as to b ecom e e le m en ta ry in th e
sense fo rm u lated a t th e b e gin n in g of th e section. F o r exam p le th e axio m s
for groups, an d H ilb e rt's axio m s for g e o m e try w ith th e c o n tin u ity axio m
o m itted , are e lem en ta ry in th e w ider sense. (In th e first, th e gro up
operatio n can b e replaced b y its representing pred icate, usin g = ; an d
in b oth , axio m s for = can be supplied. A n exam p le is w o rk ed o u t in
H ilb e r t-B e r n a y s 1934 p p . 3— 8 , 380— 381.)
T h e foregoing discussion for th e case of an elem en tary ax io m sy ste m
ca n b e paralleled for th e case of an enum erable in fin ity of elem en tary
axio m s as follows. L e t F 0, F x, F 2, . . . b e p red icate le tte r form ulas express
in g th e resp ective axio m s an d co n ta in in g free o n ly th e v a ria b les z 0, zv
z 2, . . . w h ich sta n d for th e u n d efin ed in d iv id u a ls of th e a x io m a tic
th e o ry. N o w {B is a " th e o r e m " se t-th e o re tica lly} = {e v e ry assign m en t
w hich satisfies all of F 0, F lf F 2, . . . satisfies B } = {for e v e ry assign m en t,
one of B , n F 0, - i F 1, - i F 2, . . . is t} = {som e d isju n ctio n of a fin ite
n um ber of B , - i F 0, - i F j , n F 2, . . . is pro vab le} (b y T h eorem 21 an d
C o ro llary 1 T h eorem 37) = {(- F D B for som e co n ju n ctio n F of a fin ite
§75 sk o l e m ' s pa r a d o x , t h e n u m b e r se q u e n c e 425
n u m ber o f F 0, Fv F 2, . . . } (using *62, *59) = { F 0, Fv F 2, . . . h B
w ith z 0, zv z 2, . . . h eld co n sta n t} = { B is a “ th eo rem " p ro o f-th eo re ti-
ca lly}. S im ila rly {th e axio m s are “ co n siste n t" se t-th e o re tica lly} =
{ F 0, F 1# F 2, . . . are jo in tly satisfiable} = { e v e ry co n ju n ctio n F of a
fin ite n um ber o f F 0, Fv F 2, . . . is irrefutable} (b y T h eorem s 21 an d 37) ==
{for e v e r y A , n ot b o th F 0, Fv F 2, . . . b A an d F 0, Fv F 2, . . . b " iA ,
w ith z 0, zv z 2, . . . h eld co n sta n t} = {th e axiom s are “ co n siste n t" p ro o f-
th eo retically }. Thus as before th e set-th e o retic an d p ro o f-th eo re tic
notions are e q u iva len t. B u t th e form er m eth o d of reducin g th e decision
problem s for d e d u c ib ility from th e axio m s an d for co n sisten cy o f th e
axio m s to th a t for p r o v a b ility in th e p red icate calcu lu s fails, since n o w
there are q u a n tific a tio n s w ith respect to th e fin ite co n ju n ctio n s F of
F 0, F lt F 2, . . . (how ever cf. R e m a rk 3 § 76).
U s in g th e starred form s of T h eorem s 21 an d 3 7 ( c f . T h eo rem 39 § 73),
th ese results e x te n d to th e case = is used as a lo g ica l notion, an d th e
a x io m s are expressed b y e q u a lity an d p red icate le tte r form ulas.
A xiomatic set theory. T h e a x io m sy ste m s for set th e o r y of v o n

N e u m a n n * 9 2 5 , of B e rn a y s 1937-48 an d of G o d el 1940 * are ele m e n ta ry
(in th e w id er sense).
A s G o d el's axio m s are sta te d , there are three p rim itiv e notions, (£13
(to b e a class), (to b e a set) a n d 8 (to b elo n g to), besides w h ich = is
used as a lo gical notion. A ll sets are classes, an d no oth er o b je c ts are
con sidered; so th a t th e classes c o n stitu te th e d om ain. T h e a x io m sy ste m
can th en b e expressed b y an e q u a lity an d p red icate le tte r fo rm u la
F ( = , <C¥, B) where <3? (a) an d B(a, b) express Wl(a) an d a 8 bf re sp e c tiv e ly ,
or b y a p red icate le tte r form u la E q ( C , e2f, B) & F (C , <C5f, B) w here C (a , b)
expresses a=b.
T h is ax io m sy ste m is e x tre m e ly pow erful. F ro m it w ith a p p ro p ria te
d efin itio n s th e usu al classical an alysis an d m u ch of gen eral set th e o r y ca n
b e d educed. In p articu lar, th e ex isten ce of an in fin ite set is p o s tu la te d (b y
th e a x io m of in fin ity ), an d also th e ex isten ce to a n y set of a set w h ic h in
cludes th e subsets of th a t se t; so it is d ed u cib le v ia C a n to r's th eorem
(Theorem C § 5) t h a t there e x ists a n o n -en u m erab ly in fin ite set of sets.
B u t b y L o w e n h e im ’s theorem (C orollary 2 T h eorem 3 4 *, cf. T h eo rem
39), if th e form ula F ( = , <^7 ,B ) expressin g th e a x io m s is satisfiab le a t all,
as it appears to b e from its presum ed in terp reta tio n b y set th e o ry , it is
satisfiab le in a fin ite or en u m erab ly in fin ite dom ain. (E x a m in a tio n o f
th e ax io m s rules o u t th e case of a fin ite dom ain.) T h u s w e can in te r p r e t
th e p rim itiv e notions so th a t there are o n ly en u m erab ly m a n y sets a n d th e
a xio m s are all true (i.e. an enum erable m odel exists for a x io m a tic set
th e o ry, w ith = in its u su al m eaning), even th o u gh a theorem in th e
th e o r y asserts th a t there are n o n -en u m erab ly m a n y sets. T h is is th e
S k o lem " p a r a d o x ” (1922 - 3 ).
By th e Skolem exten sio n of L o w en h ein T s theorem to th e case of
jo in t s a tis fia b ility of an enum erable in fin ity of form ulas F 0, F lf F 2, . . .
(C orollary 2 T h eo rem 3 7 *), th e " p a r a d o x ” applies e q u a lly to a x io -
m a tiz a tio n s of set th e o r y usin g in fin ite ly m a n y axiom s, such as th ose of
F r a e n k e l 1922 an d of S k o lem 1922 - 3 .
L ig h t is shed on th e " p a r a d o x ” b y tw o observation s. O n ly those p ar
ticu la r subsets of a g iv e n set are d efin able w ith in th e a x io m a tic th e o r y
w h ich can b e c o n stru cted b y operations, or sep arated o u t from th e set b y
properties (i.e. predicates), a v a ila b le in th e th eo ry. T h e b asic operatio ns
for b u ild in g sets (or processes for co n stru ctin g predicates) p ro v id e d b y th e
a xio m s are fin ite or a t m ost en u m erab ly in fin ite in num ber. T h e itera tio n
o f th em th en g iv e th e m eans for d efin in g o n ly en u m e ra b ly m a n y subsets
o f a g iv e n set. T h is exp la in s th e p o ssib ility o f in terp retin g th e a x io m
sy ste m , i.e. of sa tis fy in g th e form ula(s) expressing th e axiom s, in an
en u m erable dom ain.
O n th e oth er h an d, to enu m erate a set is to g iv e a 1-1 correspondence
o f th e set w ith a p a rticu la r enum erable set, s a y th e set of th e n a tu ra l
num bers (§ 1 ). A 1-1 correspondence can be considered as th e set of th e
correspon ding pairs.
T h u s it m a y b e possible for th e subsets of a g iv e n in fin ite set d efin ab le
w ith in th e th e o r y to b e enu m erable from w ith o u t th e th eo ry, a n d y e t
b e n on -en u m erable w ith in th e th eo ry, b ecau se no en u m eratin g set of
correspon ding pairs is am o n g th e sets d efin ab le w ith in th e th e o ry . T h e
co n stru ctio n of th e en u m eratin g set of pairs is acco m plish ed b y ta k in g
in to ac co u n t th e stru ctu re of th e ax io m sy ste m as a w hole, a n d th is
co n stru ctio n is n o t possible w ith in th e th eo ry, i.e. usin g o n ly th e operations
p ro v id e d b y th e axiom s.
The situ a tio n is sim ilar to th a t in G o d eF s in com pleten ess or u n
d e c id a b ility th eo rem (Theorem 28 § 42), w here, if w e suppose th e n u m b er-
th eo retic fo rm al sy ste m to b e con sisten t, w e ca n recognize th a t A P(p)
is tru e b y ta k in g in to v ie w th e stru ctu re of th a t sy ste m as a w hole, th o u g h
w e ca n n o t recognize th e tr u th of A P(p) b y use o n ly of th e p rin cip les of
inference fdrm alized w ith in th a t sy ste m , i.e. n o t b A P(p).
A lth o u g h th ere is th is " e x p la n a tio n ” , the " p a r a d o x ” still con fronts
u s w ith th e fo llo w in g alte rn a tiv e . E ith e r w e m u st m a in ta in t h a t th e
co n ce p ts of an a r b itra ry su b set of a g iv e n set, an d o f a n on -en u m erable
§75 sk o l e m ' s p a r a d o x , t h e n u m b e r s e q u e n c e 427
set, are a priori co n cep ts w h ich elude ch aracterizatio n b y a n y fin ite or
en u m e ra b ly in fin ite sy ste m of elem en tary a x io m s ; or else (if w e stic k to
w h a t can be e x p lic itly ch aracterized b y ele m e n ta ry axiom s, as w e m a y
w ell w ish to in con sequence of th e set-th e o retic p ara d o xes § 1 1 ) w e m u st
a cce p t th e set-th e o retic con cepts, in p a rticu la r th a t of n o n -en u m erab ility ,
as b e in g re lative , so th a t a set w h ich is n on -en u m erable in a g iv e n
a x io m a tiz a tio n m a y becom e enum erable in an oth er, an d no ab so lu te
n o n -en u m era b ility exists. T h is re la tiv iz a tio n of set th e o r y w as proposed
by Skolem (19 2 2 -3 , 19 2 9 , - ).
19 2 9 30
The L o w en h eim theorem , since it leads to SkolenTs “ parado x” ,
ca n be regarded as th e first of th e m odern incom pleteness theorem s.
F o r further discussion, see Skolem 19 3 8 .
A xiomatic arithmetic. P o stu la te G rou p B of our form al n u m b er-

th e o re tic sy ste m p rovid es an ex am p le of an ax io m sy ste m for th e th e o r y
of th e n a tu ral num bers con sistin g of an e ffe c tiv e ly en u m erable in fin ity
of elem en tary axiom s, i.e. th e form ulas exp ressin g th e axio m s are ef
f e c t iv e ly enum erable (cf. T h eorem 38 § 72). T h e fu n ctio n s are of course
replaceable by their represen tin g predicates. R y ll-N a r d z e w s k i 1 9 5 2 *
show s th a t no fin ite subset of these axio m s w o u ld suffice for th e d ed u ctio n
of th e sam e class of theorem s.
Another question is whether these axioms do completely characterize
the natural number sequence. Godel’s completeness theory for the
predicate calculus provides us with a proof of the following theorem,
which was originally obtained in another way by Skolem (1 9 3 3 , 1 9 3 4 ;
cf. 1938 ).
W e sh all consider axio m s for th e sequence of th e n a tu ra l n um bers
(call th e set of th em N), u sin g as p rim itiv e notions th e in d iv id u a l 0 an d
th e p red icate a '= b , i.e. in th e X -notation (§ 10) \a b a '= b , an d perh ap s
o th er p rim itiv e notions. T h e ax io m s shall be expressible b y e q u a lity a n d
p red icate le tte r form ulas, w ith z expressin g 0 an d P 0(a, b) exp ressin g
a '= b .
In discussing assign m en ts to th e free v a riab les an d p red icate le tters
of a n y such form ula, w e let “D ” sta n d for th e dom ain, “ 2” for th e in
d iv id u a l assigned to z, an d “ P 0(a, b)” for th e p red icate assigned to P 0(a, b).
T h e n (D , z, PQ{a, b)) is a m a th e m a tic a l sy ste m in th e sense of § 8 , co n
sistin g of a set or dom ain, a m em ber of th e set, an d a b in a r y p red icate
o ve r th e set.
T heorem 44°. Any finite or effectively enumerable infinite class of

equality and predicate letter form ulas which can be join tly satisfied so that
(D , z, P 0(a, b)) is (N , 0 , a '— b) can also be join tly satisfied {with 0 < D < H0)
so that {D, z, P 0(a, b)) is not isom orphic to (N , 0, a '= b ).
P roof. B y h yp o th esis, th ere is a sa tisfy in g assign m en t for th e given
formulas jointly in which the domain D is N , z has the value 0, and
P 0(a, b) has the value a ' = b.
L e t P x, P 2 an d P 3 b e o th er d istin ct p red icate letters, w h ich either d o
n o t o ccur in th e g iv e n form ulas, or h a v e th e resp ective v a lu e s a+b=c,
a*b= c an d {% ){Ey)T%{at a, x, y) in th e g iv e n assignm ent. W e sh all e x te n d
th e g iv e n class of form u las b y a d d in g seven form ulas (if n o t a lr e a d y
included) w h ich are satisfied w h en th e g iv e n assign m en t is e x te n d e d
(if necessary) b y assign in g P x, P 2 an d P 3 th e va lu e s ju s t m entioned .
W e a d d four closed form ulas, s a y
(cf. E x a m p le 11 (a) § 74), w h ich under th e d escribed assign m en t exp ress

th e recursion eq u atio n s for a-\-b an d a-b as p araph rased in term s of th e
represen tin g pred icates an d a*6 = c , to g e th e r w ith th e tw o
form ulas
(b) V a V b 3 !c P 1 ( a , b , c ) , V a V b 3 !c P 2(a, b, c),
w h ic h express th a t a+b=c an d a -b = c are represen tin g p red icates.

S in ce T 2(a, b ,x t y) is p rim itiv e recursive, b y C o ro llary T h eo rem I § 49
it is a rith m e tic a l (§ 48), an d so (replacing ', + an d * b y th eir represen tin g
predicates) w e ca n fin d a le tte r form ula T 2(a, b , x , y) in = , P 0, P x, P 2
w h ic h expresses it u nd er th e d escribed assignm ent. W e th e n a d d th e
form ula
(C) V a ( P 3(a) ~ V x 3 y T 2(a, a, x , y)).
L e t th e form u las o f th e re su ltin g (enlarged) class b e F 0, Fv F 2, . . . .

A s in th e proof of L o w e n h e im 's th eorem (C orollary 2 T h eo rem 37*),
th e h y p o th e sis of T h eo rem 3 7 * is s a tis fie d ; an d w e use T h eo rem s 3 7 * a n d
40 to o b ta in a n o th er s a tis fy in g assign m en t for F 0, F x, F 2, . . . jo in tly .
Let D * t z* f P * , P * f P * , P * b e re sp e c tiv e ly th e d o m ain an d th e v a lu e s
of z, P 0, P x, P 2, P 3 in this.
S u p p o se (for redu ctio ad absurdum ) th a t (D *, z*, P * (a * t b*)) is
isom orph ic (§ 8 ) to (N , 0, a ' = 6), i.e. D * is in fin ite an d ca n b e en u m erated
as s0, sl9 s2, / . . so th a t z* = s0 an d P j (sa, sb) = a '= b .
T h e n b y T h eo rem 40,
P | ( s a) ^ (E x)(y)R *(a, x, y)
for som e p rim itiv e recursive R *.
§75 SKOLEM^S PARADOX, THE NUMBER SEQUENCE 429
In § 43 w e reasoned t h a t w h en th e v a riab les range o ve r th e n a tu ra l
num bers, an d 0 an d ' h a v e th eir u su al m eanings, th e recursion eq u a tio n s
fo r + an d • h a v e th e u su al fu n ctio n s + an d * as th eir u n iq u e solution.
S in ce th e form ulas (a) an d (b) are satisfied , th a t reasoning (w ith m inor
rearran gem en ts to fit th e use of th e represen tin g p red icates in s te a d o f
th e functions) applies n o w to sh ow th a t P * (sa, s6, $c) == a+b=c an d
P * (s a, sb, s c) = a-b—c. Then sim ilarly our proof of T h eo rem I § 49
sh ow s th a t th e form ula T 2(a, b , x , y) now expresses a p red icate
T $ (a * ,b * ,x * ,y * ) such th a t T%{$ai sb, sv) = J2(a, 6, x, y).
s mt H en ce
V x 3 y T * ( a , a, x , y) expresses a p red icate T *(a *) su ch t h a t
(ii) T * (sa) 3 (x)(E y)T t ( a ,a ,x ,y ) .
B y th e v a lu a tio n rules for ~ (E x a m p le 1 § 28) an d ¥ , since (c) is
satisfied ,
(iii) P*{a*) ss T*(a*).

C o m b in in g (i) — (iii), (Ex)(y)R%(a, x, y) = (x)(E y)T 2(a, a, x f y). B u t b y
T h eo rem V (16) § 5 7 , th e p red icate (x)(E y)T 2(a, a , x, y) is n o t expressible
in th e o th er 2-q u an tifie r fo rm ; for a certa in n u m ber g,
(E x )(y )R * (g ,x ,y ) ^ (x)(E y)T 2(g, g, x, y).

B y redu ctio a d absurdum , (D *, z*, P *(a*, b*)) is n o t isom orph ic to
(iV, 0, a'=b).
D iscussion . By th e theorem , no fin ite or e ffe c tiv e ly enu m erable
in fin ite set of e lem en ta ry axio m s can ch aracterize th e n a tu ra l n u m ber
seq uence 0, 1 , 2 , . . . , a, a ' , ___A n y such set w h ich are tru e of th e n a tu ra l
n u m b er sequence m u st also b e tru e under an oth er in terpretatio n . W e
s ta te d th e theorem for th e n a tu ra l num ber sequence as a sy ste m o f th e
form (N , 0, a '= b ), b u t b y th e re p la c e a b ility of a ' b y a != b it applies
also to th e n a tu ra l n u m ber sequence as a sy ste m o f th e form (N , 0, ').
In p articu lar, th e axio m s of P o stu la te G ro u p B of our form al nu m b er-
th eo retic sy ste m (§ 19) a d m it an in terp retatio n (using th e lo gical sy m b o ls
and = in th eir u su al m eanings) oth er th a n th e in ten d ed one.
T h is in com pleten ess o f P o stu la te G ro u p B as a ch aracterizatio n of
th e n a tu ra l n u m ber sequence is u n d ersta n d ab le w h en we com pare
P e a n o 's fifth a x io m (the prin ciple of m a th e m a tic a l in d u ctio n , § 7) w ith
A x io m S c h em a 13. P ea n o 's fifth ax io m asserts th a t
(I) 4(0) & (x)(A(x)

A (x')) - (x)A(x)
-
h olds for all n u m b er-th eo retic pred icates A (x). T h ese pred icates c o n stitu te
a n on -en um erable to ta lity . B u t th e b u n d le of axio m s g iv e n b y A x io m
S c h e m a 13 o n ly express th a t (I) holds for those p red icates A (x) w h ich
are expressible b y form ulas A (x ) of th e sy ste m , i.e. o n ly for e n u m era b ly

m a n y predicates. P ea n o 's fifth ax io m is not elem en tary. W e can express
it in th e sym b o lism of th e second order p red icate calcu lu s (§37), b y usin g
a g e n e ra lity q u a n tifie r w ith a p red icate va ria b le <C?, th u s:
Vc3[<^(0) & Vx(^{x) 3 a(x')) 3 Vx&{x)].

T h ese id eas h a v e a co n n ection w ith G od eFs theorem on fo rm a lly
u n d ecid a ble propositions (Theorem 28 or 29 § 42).
O n e m a y th in k of th e form ula A v(p) or A g(q) (w hich is true b u t u n -
pro v a b le , if th e n u m b er-th eo retic syste m is sim p ly consistent) as e x
pressing a proposition w h ich can be " p r o v e d ” from P ea n o 's axiom s,
b u t o n ly b y m a k in g use of in d u ctio n w ith som e in d u ctio n p red icate
A (x) w h ich is n o t expressible in th e sy ste m under th e in ten d ed in ter
p retatio n . (This su ggestio n w ill receive con firm ation la te r; cf. end § 7 9 ,
n o tin g (II) § 42.)
T h e u n p r o v a b ility of A P(p) becom es u n d erstan d ab le also from S k o le m ’s
result (Theorem 44), on th e gro u n d th a t A P(p), a lth o u g h true of th e n a t
ural num bers, is false under one of th e other in terpretatio n s w h ich
sa tis fy th e axiom s. T h e n th e u n d e c id a b ility of A P(p) in th e num ber-
th eo retic form alism appears as a phenom enon of th e sam e kin d as th e
im p o ssib ility of p ro v in g either E u c lid 's parallel p o stu late or its n eg atio n
from th e o th er axio m s of g e o m e try ( § 8 ). T h e g iv e n axiom s are n o t
ca tego rica l.
In d eed co n versely, as rem arked a b o v e (using C o ro llary 1 T h eorem 3 7 *),
a form ula is p ro v ab le, if it is true under all in terpretatio n s w h ich m ake
th e axio m s true. So th e kn o w n u n p r o v a b ility of A P(p) m akes it absurd
th a t A P(p) sh ould b e tru e under all th e in terp retatio n s w h ich sa tis fy th e
axiom s. T h u s G o d eF s theorem s 28 an d 3 7 * afford an oth er proof of
T h eo rem 44 for th e case th e class of form ulas for T h eorem 44 is P o s tu la te
G ro u p B of our form al sy ste m (restated as e q u a lity an d p red icate le tte r
form ulas). B u t (as rem arked a b o ve , usin g T h eorem 2 1 *), if a n y enum er
ab le class of e q u a lity an d p red icate letter form ulas are jo in tly satisfiable,
th e form al sy ste m o b ta in ed b y ad jo in in g th em as axiom s to th e p red icate
ca lcu lu s w ith e q u a lity is (sim ply) consistent. So g iv e n any class of
form ulas for T h eorem 44, b y ad jo in in g P o stu la te G rou p B , an d ca rry in g
o u t th e proof of T h eorem 28 in th e resulting sy ste m (or using T h eorem
X I I I P a r t I I I § 60), w e g e t T h eo rem 44 in general.
M o sto w ski 1949 g iv e s an in terestin g ex am p le (suggested b y S k o lem 's
" p a r a d o x ” ) of a proposition in a x io m a tic set th eo ry, w h ich he d em on
strates to be u n d ecidable b y sh ow in g it to be true under one in terp retatio n
§75 skolem's paradox , the number sequence 431
a n d false u nder another. K reisel 1950 deals w ith sim ilar problem s.
T h e proof of T h eo rem 44 w e g a v e first, an d th a t b ased on T h eo rem 28
(or X I I I ) , are sta te d for th e case of an ele m e n ta ry a x io m sy ste m or one
h a v in g an e ffe c tiv e ly enu m erable in fin ity of e lem en ta ry axio m s. S k o le m 's
proof does n o t restrict th e en u m eration of th e axio m s to b e e ffe ctiv e .
T h e a d d itio n a l g e n e ra lity is n o t essential, w h en w e are con siderin g th e
theorem as an incom pleteness theorem for form al a x io m a tic s from th e
sta n d p o in t th a t th e aim of a x io m a tiz a tio n is to m ak e th e assu m p tion s of
th e th e o r y e x p licit.
R emark 1. O u r (first) proof of T h eorem 44 can b e m o d ified to secure

a d d itio n a l gen erality. Suppose n ow th e class of th e form ulas for T h eo rem
44 is m erely arithm etical, in th e sense th a t, under a G o d el n u m berin g
estab lish ed b y th e m eth o d s of §§ 52 an d 56, th e p red icate ‘x is th e G o d el
n u m ber of a form ula of th e class', c a ll it “ C (x)” f is arith m etical. T h e n
b y T h eorem V I I (d) § 5 7 , C (x ) is expressible in one o f th e form s of
T h eo rem V , s a y a ^ -quan tifier form . L e t B(x) refer sim ilarly to th e g iv e n
class o f form ulas exte n d e d b y ad d in g som e fin ite list (as ab o v e w e ad d ed
th e seven (a) — (c)); then , h o w ever th a t list is chosen, B(x) is expres
sible in thev sam e ^ -quan tifier form . B y T h eo rem X I V (b) § 6 0 (tak in g
R (x , y) == B (x))t th e class xB (x) is enu m erated b y a fu n ctio n 0 (&) re
cu rsive in ^ -q u an tifier p red ic a te s (in th e case of th e ^ -quan tifier form w ith
e x iste n c e first, even in q — 1-q u a n tifier predicates). B u t th e proofs of
T h e o r e m s 38 an d 40 h o ld good , w h en th e h yp o th e sis th a t th e enu m eration
F 0, F x, F 2, . . . is e ffe ctiv e is o m itte d , an d th e conclusion is altered b y
ch a n g in g " p r im itiv e re cu rsive" to " p r im itiv e recursive in 0" w here
0(A) is th e G o d el num ber o f F * ; an d hence (using T h eorem X I § 58, an d
(17) an d (18) § 5 7 ) if 0 is recursive in ^ -q u an tifier p red icates, w h en th e
alte ratio n in th e con clusion con sists in s u b s titu tin g th e q+2- for th e
2 -q u a n tifie r form s. So th e proof of T h eo rem 44 goes th ro u gh now , b y
u sin g a case of T h eo rem V for a q + 2 - in stead of a 2 -q u a n tifie r form . —
W e ca n s till further generalize th e C(x) for T h eorem 44 to b e a rith m e tic a l
in th e p red icate M o f T h eo rem V I I I § 57, in clu d in g am o n g th e ad d ed
form u las three to express th e d efin ition of M, an d usin g T h eorem s I * , V * ,
V I P , X I * (w ith M as th e *F) in stea d of I, V , V I I , X I . (The second proof
of T h eo rem 44 ca n also b e carried o u t under m ore gen eral h yp o th ese s,
b y usin g gen eraliza tio n s o f T h eo rem 28 or X I I I to " n o n -c o n s tr u c tiv e
lo g ics".)
G o d e l’s u n d e c id a b ility theorem h o w ever is n o t restricted to th e case

th e a x io m s are ele m e n ta ry (Theorem X I I I § 60). B u t w e ca n ch aracterize
the natural number sequence completely by Peano’s axioms, the fifth

of which is non-element ary, if we grant the notion of all predicates over
the domain. Suppose we have a consistent formal system containing
these categorical axioms for the natural numbers. Under the conception
of a formal system which we are entertaining, a formal system may
have only an enumerable infinity of formal objects. So only an enumerable
infinity of formulas can be substituted in the system for the predicate
variable c2f of the fifth Peano axiom. Thus for deductive purposes within
the system, just as before, (I) is available only for an enumerable infinity
of predicates. In fact, by Godel’s theorem (Theorem XIII), there is a
consequence of the axioms under the interpretation which is not provable,
i.e. not deducible from the axioms by the logic formalized in the system.
So when we have non-elementary axioms, not all formulas need be prov
able which are true under all interpretations which satisfy the axioms.
(In our example, there is essentially just one such interpretation.) The
incompleteness which appeared in the axiom system in the case of
elementary axioms is transferred to the deductive apparatus, if we
undertake to avoid it by using non-elementary axioms.
As Skolem expresses it ( 1 9 3 4 p. 160), "... the [natural number]
series is completely characterized, for example, by the Peano axioms,
if one regards the notion 'set' or 'propositional function’ as something
given in advance with an absolute meaning independent of all principles
of generation or axioms, But if one would make the axiomatics conse
quent, so that also the reasoning with the sets or propositional functions
is axiomatized, then, as we have seen, the unique or complete character
ization of the number series is impossible.”
This situation is discussed in Henkin 1 9 5 0 , which came to the author’s
attention after this section was written (the first draft in 1947). Other
papers are e.g. Mostowski 1 9 4 7 a (cf. Kemeny’s review 1 9 4 8 ) and Rosser
and Wang 1 9 5 0 (cf. Skolem’s review 1 9 5 1 ).
§ 76. T h e d e c i s i o n p r o b l e m . T heorem 54. The decision problem
for the pure predicate calculus {pure predicate calculus with equality) is
unsolvable, i.e. there is no decision procedure for determ ining whether a
predicate letter form ula {an equality and predicate letter formula) is provable
in the calculus. (Church 1 9 3 6 a, Turing 1 9 3 6 -7 .)
P roof, for the predicate calculus. We saw in § 75 that the decision
problem for provability in any axiomatic theory having an elementary
axiom system reduces to that for provability in the pure predicate cal-
§76 THE DECISION PROBLEM 433
cuius. T h u s to p rove this theorem of C hurch, it w ill suffice to fin d an
e lem en ta ry a x io m a tic th e o ry for w hich, on th e basis of C h u rch 's thesis
(§ 60), there can be no decision procedure.
A c c o rd in g to T h eorem 33 § 6 1 , an ex am p le of such an a x io m a tic
th e o ry is p ro v id e d b y th e form al sy ste m of R o bin so n d escribed in L e m m a
18b § 49, if th a t sy ste m is sim p ly con sistent. A proof of its sim ple con
sisten cy, w h ich th e present theorem is num bered to follow , w ill be g iv e n
in § 79 (Theorem 53 (a)).
W e repeat th e reasoning in d eta il (already o u tlin ed in § 7 5 ) . L e t S 2
be th e sy ste m of L e m m a 18b w ith th irteen p articu lar axiom s.
(A) B y ap p licatio n s of T h eo rem 43 (see E x a m p le 11 (a) § 7 4) , we

find an oth er sy ste m 5 X in w h ich th e fu n ctio n sym b o ls ', + , * are replaced
b y re sp ective p red icate sym b o ls, th e n um ber of th e axio m s b ein g in
creased to nineteen. B y ( V i l l a ) , a form ula E of S 2 is p ro v a b le in S 2, if
an d o n ly if th e form u la E ' of S ± is p ro v a b le in Sv
(B) We can th en replace th e ax io m s by th eir closures, w ith o u t
ch a n gin g th e p r o v a b ility n o tio n (end § 32).
(C) M oredver, b y th e &~rules, since th e ax io m s are fin ite in num ber,

we can likew ise replace th em b y their co n ju n ctio n as a single axiom .
(D) W e can furtherm ore ch an ge th e n o ta tio n to em p lo y a v a ria b le z

not occurring in th e axiom (s) in p la ce of th e in d iv id u a l sy m b o l 0, w ith th e
u n d erstan d in g th a t th en p r o v a b ility shall m ean d e d u c ib ility from th e
axiom (s) in th e p red icate calcu lu s w ith z held co n sta n t. I f C be th e result
of m a k in g th is ch an ge in a form ula D of S 1 not co n ta in in g z, th en b y
R e m a rk 2 (b) § 34, C is p ro v a b le now , if an d o n ly if D w as p ro v a b le
before. (This tre a tm e n t of 0 se p a ra te ly from the fu n ctio n sy m b o ls can
b e a v o id e d b y usin g in stea d E x a m p le 11 (b) § 74 in (A).)
(E) W e m a y further ch an ge th e n o ta tio n to e m p lo y p red icate letters

in p lace of th e p red icate sym b ols. I f B results from a form ula C b y th is
ch an ge of n o tatio n , th en b y tr iv ia l ap p licatio n s of T h eorem s 15 an d 16
§ 34, B is n ow p ro v ab le, if an d o n ly if C w as before. (In § 7 5 w e to o k
B to co n ta in no v a riab les free e x c e p t z, b u t th a t w as done to m ak e th e
se t-th e o retic n otion s a p p ly p roperly, an d is un n ecessary now.)
(F) F in a lly , by th e D -ru les, B is p ro v ab le in th e last described

sy ste m , i.e. F h B in th e p red icate calculus w ith z held co n sta n t, w here F
is th e form u la (con taining o n ly z free) expressing th e axio m s now , if
an d o n ly if F D B is p ro v ab le in th e p red icate calculus.
T h e w h ole process b y w h ic h from th e form ula E o f S 2 w e fin d a p red

ica te le tte r form u la F D B such th a t { h E in S 2} == { b F D B in th e
p red icate calcu lu s} is e ffe c tiv e . (It cou ld be represented v ia G o d el n u m b er
in g b y a gen eral recu rsive fu n ctio n as discussed in § 6 1, an d it co u ld be
effe cte d b y a T u r in g m ach in e as a t end § 70.) Th erefore, if th ere w ere a
decision procedure for p r o v a b ility in th e pure p red icate calculus, there
w o u ld b e one for 5 2, w h ich w o u ld consist, g iv e n a form u la E of S 2, in
fin d in g th e correspon ding p red icate le tte r form ula F D B , an d a p p ly in g
th e procedure for th e p red ic a te ca lcu lu s to th e latter. B u t b y T h eo rem 33,
if S 2 is s im p ly con sisten t, there is no decision procedure for p r o v a b ility
in S 2.
T h e a rgu m en t applies also to th e p red icate ca lcu lu s w ith e q u a lity ,
since b y T h eo rem 41 (b) p r o v a b ility in th e sy ste m of L e m m a 18b is
e q u iv a le n t to p r o v a b ility in a sy ste m con sistin g of th e p red icate c a l
culus w ith e q u a lity an d seven p a rticu la r axiom s.
R emark 1. T h e order in w h ich th e redu ction steps (A) — (F) are

perform ed is im m aterial, so lo n g as (B) an d (C) precede (F), (A) precedes
(B), (C) an d (E), an d if (D) precedes (B) th e v a ria b le z is e x e m p te d from
th e closure op eratio n of (B). (Or (C) m a y b e o m itte d , an d (F) per
form ed once for each a x io m ; cf. *4 an d *5 § 26.)
R emark 2. T h e p roof of T h eorem 54 m a y be b ased on T h eorem X I I

§ 60 in stea d of T h eo rem 33, thus. B y th e foregoing reduction s (A) — (F)
on th e sy ste m of L e m m a 18b w ith E x a m p le 2 § 60, or b y (the m e th o d of)
th ose redu ction s on th e sy ste m of E x a m p le 2 § 73, an d usin g th e con
siste n c y p ro p e rty to b e estab lish ed in § 79 (Theorem 53 (b) or T h eo rem
52): F or any fixed prim itive {or general) recursive predicate R (x ,y ), there
is an effective procedure by which , given any number x, a predicate letter
form ula K x can be found such that
(1) {E y)R {xf y) s= { |- in the predicate calculus }.
T h eo rem 54 th en follow s from T h eo rem X I I b y ta k in g R (x, y) = T ±{x, x, y).
T h is p roof from T h eo rem X I I w ith E x a m p le 2 § 73 is e ssen tially C h u rch 's
origin al p ro o f; th a t from T h eo rem 33 essen tially M o sto w ski an d T a rsk i's
(1949 a b stract).
R emark 3. O u r co n cep tio n of a form al sy ste m S im plies t h a t th e

form ulas o f S sh ould b e e ffe c tiv e ly enum erable, or a d m it a G o d el n u m ber
in g, so th a t g iv e n a n y fo rm u la A w e can e ffe c tiv e ly fin d its n u m ber x,
an d in v e rs e ly g iv e n a n y num ber x w e ca n e ffe c tiv e ly decid e w h e th er it
is th e G o d el n um ber of a form ula an d if so find th a t fo rm u la Ax. T h e n
(cf. R e m a rk 1 (a) § 6 0 ) : There is a general recursive predicate R such that
(2) { b A x in 5 } as (E y)R {x , y).
So th e decision problem for p r o v a b ility in a n y sy ste m S is e q u iv a le n t

to th a t for a p red icate of th e form (E y)R (x, y). B y R e m a rk 2, or b y th e
(first) proof of T h eo rem 54 w ith th e la tte r p a rt of E x a m p le 3 § 6 1, th e
decision problem for th e p red ica te calcu lu s is of th e h igh est degree of
u n so lv a b ility for pred icates of th is form (cf. p reced in g E x a m p le 3 § 61).
T h u s (com binin g ( 1 ) an d (2 )), th e decision problem for p r o v a b ility in
a n y form al sy ste m is reduced to th a t for th e p red icate ca lcu lu s (either
classical or in tu itio n istic). T h is generalizes our rem ark (§ 75) th a t th e
problem for a n y a x io m a tic th e o r y w ith an e le m en ta ry a x io m sy ste m re
duces to i t ; b u t of course th e redu ction s b y g o in g o u t of th e sy ste m to a
G o d el n u m b erin g v ia (2), an d th en ce in to th e p red icate calcu lu s v ia ( 1 )
are v e r y indirect.
B y §§ 37, 72, 73, 75 , p r o v a b ility in th e p red icate calcu lu s w ith e q u a lity

is e q u iv a le n t to v a lid it y in e v e r y n o n -e m p ty d o m a in ; so n o w there is
no decision procedure for th e la tte r p ro p e rty of an e q u a lity an d p red icate
le tte r form ula. T ra h te n b ro t 1950 proves th e an alogo u s theorem for v a lid it y
in every n o n -e m p ty fin ite dom ain.
R eductions and special cases . B eca u se so m any p a rtic u la r

question s (e.g. F e r m a t’s <fla st th eo rem ” § 13) an d decision problem s
reduce to th e decision problem for th e p red icate calcu lu s, m u ch w o rk
has been done on it, lead in g to p o sitiv e results of tw o s o r ts : (a) reduction s
of th e gen eral problem , an d ((3) solutions of special cases. T h e results are
o fte n presented in a d u al set-th e o retic form , in w h ich th e problem is to
decide as to th e s a tisfia b ility of a p red icate le tte r form ula in som e n on
e m p ty dom ain (§§ 72, 75), rath er th a n as to its p ro v a b ility .
A n e a rly exam p le of (a) is S k o le m ’s norm al form (Skolem 1920 , H il-
b e rt-B e r n a y s 1934 pp. 158 ff). T h e Skolem p ro o f-th eo retic (satisfactio n -
theoretic) norm al form is a p ren ex form ula (Theorem 19 § 35) in w h ich
all th e e x iste n tia l (generality) qu an tifiers com e first. G iv e n a p red icate
le tte r form ula G , there can b e fou n d e ffe c tiv e ly a p red icate le tte r form u la
M (N) of th is form , such th a t M is p ro v ab le in th e p red icate ca lcu lu s
(N is satisfiab le in a g iv e n dom ain), if an d o n ly if G is. T h u s th e d ecision
problem for p r o v a b ility (satisfiab ility) for p red icate le tte r form ulas
ge n e ra lly is reduced to th e sam e problem for Skolem norm al form s
(cf. th e la tte r p art of § 61). T h e norm al form M (N) is not in gen eral
e q u iv a le n t to G , b u t M (n N ) is interdeducible w ith G ( - 1G) in the
p red icate ca lcu lu s w ith a p o stu la ted su b stitu tio n rule (§ 37). Skolem
used his satisfaction-theoretic normal form in simplifying the proof of
Lowenheim’s theorem and generalizing it, and Hilbert and Bernays 1 9 3 9
employ it in proving Godel’s completeness theorem in a way which makes
the formalization referred to in the proof of Theorem 36 § 72 reasonably
simple.
An example of (p) is the solution of the decision problem by Lowenheim
1 9 1 5 (simplified by Skolem 1 9 1 9 ), and independently by Behmann 1 9 2 2 ,
for the case of predicate letter formulas containing only predicate letters
with 0 or 1 argument. Equivalently, by Remark 1 §34, the decision prob
lem is solved for the \-place predicate calculus, i.e. the calculus with
only 0 - and 1 -place predicate letters. (Cf. end § 72, Hilbert-Bernays 1 9 3 4
pp/ 179—209.)
Reductions and special cases of the decision problem have remained
an active field of research, since Church showed that there can be no
general solution (Theorem 54). The literature is too extensive to be cited
here, and the reader is referred to Church’s bibliography and the review
sections of the Journal of Symbolic Logic (cf. the preface to the bib
liography of this book). Quite a number of the results are described
in Hilbert-Bernays 1 9 3 4 and 1 9 3 9 . Church 1 9 5 1 discusses special cases.
A xio m atic We take up now a method of Tarski ( 1 9 4 9
t h e o r ie s .
abstract) for investigating the decision problems for axiomatic theories.
We shall consider theories formalized on the basis of a logical calculus,
which may be either the predicate calculus or the predicate calculus with
equality. (Tarski uses the latter.) We shall usually say “formal system S”
for uniformity with our previous terminology, where Tarski says “theory
which emphasizes the mathematical application.
By the logical constants we shall mean the six logical symbols D, &,
V, - 1 , V, 3, if the logical calculus is the predicate calculus; these and
also = , if it is the predicate calculus with equality. The terms and formulas
of a system are to be constructed using besides these logical constants
a finite number of individual, function and predicate symbols, called the
non-logical constants (but no predicate letters). The postulates besides
those of the logical calculus shall be a finite or infinite set of non-logical
axioms.
Following Tarski, we call such a system finitely axiomatizable, if the
non-logical axioms are finite in number or all but some finite set of
them are redundant (Example 2 § 74). We say of such systems that S2
is an extension of S1 (or is a subsystem of S2), if each formula
provable in Sx is provable in S2; S 2 must then have all the non-logical
constants of S v but it mayhave others inaddition. AnextensionS 2 of S 1
is a fin ite extension, if all but a finite number of the axioms of S are
2
provable in S v W e say briefly that S 1 is undecidable to mean that the

decision problemfor provability in S x is unsolvable.
FollowingTarski, wesaythat Sisessentially undecidable, if Sis (simply)
consistent, and every (simply) consistent extension of S is undecidable.
Rosser 1936 showedthat systems like our number-theoretic systemof
Chapter IV (if they are consistent) have this property (cf. Theorem33
§61). Then the formalized systems of axiomatic set theory of von Neu
mann 1925 , of Bernays 1937-48 and of Godel 1940 (if consistent) are
examples of systems Swhicharebothessentiallyundecidable (sincethey
include the usual number-theory) and finitely axiomatizable. Mostowski
andTarski (1949 abstract) werethefirst tonotethe existence of asystem
5 which is both essentially undecidable and finitely axiomatizable,
andalsosimpleenoughtobeeasilyinterpretableinvariousothertheories,
inthe sensetobedefinednext. This provides thebasis forthe application
of the methodof Tarski whichis giveninTheorem45 (b) and (c). Astill
simpler example of anessentially undecidable andfinitely axiomatizable
systemisthatofRaphaelRobinson (1950 abstract),whichhasthirteennon-
logicalaxiomsasdescribedinLemma18b§49onthebasis of the predicate
calculus, or sevenonly (Axs. 14, 15, 18—21, andthe formula of *137,
or equivalently of *136) as it was described by Robinson on the basis
of the predicate calculus with equality. Robinson states that none of
these sevencanbeomittedwithoutsacrificingtheessentialundecidability.
Tarski says that two systems S x and S are com patible, if they have
2
the same non-logical constants and a (simply) consistent common

extension. Now consider any two systems S 1 and S 2 which in general
do not have the same non-logical constants. First we take the case
the logic is the predicate calculus with equality. Then S is consistently
2
interpretable in S v if S t and S 2 have a consistent common extension S3,

inwhichthere is provable, for each w-placepredicate symbol P (function
symbol f) of S whichS x lacks, aformulahavingtheformP(xx, ..., xn) ~
2
F (xlt ..., xn) of an explicit definition of P (the formf(x v ..., x n) w

=
~F(xx, ...,xn,w), i.e. (iii) of Lemma 26 §74) where the variables

shown are distinct and F(xx, ...,xw ) (F(xx, ...,xn,w)) contains only
these variables free and as non-logical constants only ones of and
possibly additional individual symbols. For the case the logical calculus
is the predicate calculus and S 2 has function symbols which lacks,
furthermore S t shall have = among its constants, and in S there shall
3
be provable the equality axioms for the predicate andfunction symbols

of The situation is illustrated (rather trivially) by the S and S
S v S v 2 3
in the proof of Theorem43 §74 (but either for (a), or for (b) assuming
consistency).
T heorem 45. (a) I f S is u n d e c id a b le , th e n every s y s te m S ± w h ic h la c k s
n o n e o f th e c o n sta n ts o f S e x cep t p o s s ib ly in d iv id u a l s y m b o ls a n d o f w h ic h
S is a fin ite e x te n s io n is u n d e c id a b le .
(b) I f S is e s s e n tia lly u n d e c id a b le a n d fin ite ly a x io m a tiz a b le , th e n every
s y s te m S x w h ic h la c k s n o n e o f th e c o n sta n ts o f S excep t p o s s ib ly in d iv id u a l
s y m b o ls a n d w h ic h h a s a c o n s is te n t c o m m o n e x te n s io n S z w ith S [in p a r tic u la r
every sy ste m S 1 c o m p a tib le w ith S ) is u n d e c id a b le .

(c) I f S is e s s e n tia lly u n d e c id a b le a n d fin ite ly a x io m a tiz a b le , th e n every
sy s te m S t in w h ic h S is c o n s is te n tly in te r p r e ta b le is u n d e c id a b le . (Tarski
1949 abstract.)
P roofs , (a) By the reductions (B), (C), (D) applied only to the
individual symbols of SwhichSxlacks, and(F). (Tarski takes the axioms
tobeclosedabinitio, anddealswiththecasethat andShavethesame S x
constants; then (B) and (D) are not required.)

(b) Let be the systemhaving the axioms (and constants) of both
S 2
and Then S is a subsystemof S3; and hence, since S is con

S . 2 3
sistent, so is S2. Also S is an extension of S; and hence, since S is es

2
sentially undecidable and S is consistent, S is undecidable. But S 2 2 2
is a finite extension of Now (a) applies with S as its S. S v 2
(c) W e shall treat in detail the case S has a function symbol f as its
only constant (except perhaps individual symbols) which lacks, for S x
the predicate calculus as the logic. If Shas more suchfunctionsymbols,

we merely iterate the application of Theorem42 andits lemmas; andif
Shas suchpredicatesymbols, weuseExample §74likewise. Theresults 1
will then hold also for the predicate calculus with equality, as (by
Theorem41 (b)) extending the logic to that is equivalent to assuming
that the equality axioms for all the function and predicate symbols
are present inall the systems considered(whichonlymakestheargument
easier).
Let S bethecommonextensionof andSdescribedinthe definition
3 S x
of consistent interpretability (with S as the S2).

Let- S4a be the subsystemof having as its constants those of S 3 S v
f, and the additional individual symbols (if any) belonging to S or

occurringinF(x1, ..., xn,w), andhaving as its non-logical axioms those
of and of S, the equality axioms for the predicate and function
symbols of and (iii). S v
Since S4ais a subsystemof S and S is consistent, S4ais consistent.
3 3
Hence, since S4ais an extension of S and S is essentially undecidable,

S4a is undecidable.
Using Lemmas 26 and 27 and Remark 2 (a) §74, in the list of
non-logical axioms for S4awe can replace each one Aof the non-logical
axioms of S by its principal f-less transformA', and (iii) by (i) and
(ii), without changing the class of the provable formulas; call the re
sulting undecidable systemS4.
By Theorem 42 §74, the function symbol f and its axiom (ii) can
be elim inated fromS 4, leaving a systemS which (by (IV) §74) is also
5
undecidable.
But S 5 h as as its constants only those of and possibly individual
S x
symbols, and S 5 is a finite extension of So by (a) (with S as its S),

S v 5
S t is undecidable.
R emark 4. For the conditions on S in the definition of consistent
3
interpretability, instead of (iii) we may have provable in S a formula

3
having the formf(xx, ..., xn)=t(xx, ..., xn) of an explicit definition of

f, where t(xx, ..., xn) is a termcontaining only the variables shownand
as non-logical constants only function symbols of Sx except possibly
individual symbols. For then t(xx, ...,xn)=w is an F(xx, ...,xn, w).
E xample . By (b), since Robinson's system, call it S, is essentially
1
undecidable ari finitely axiomatizable, everyformal systemSxwiththe

constants (andperhaps others, e.g. ), the provable sentences
0
of whichexpresstruepropositions about natural numbers, isundecidable,

if we accept the simple consistency of a commonextension S (say that
3
onewhichhas the axioms andconstants of boil Sxand5) as guaranteed

by the truth. To make the undecidability of such a systemSxa meta-
mathematical result, it remainstosupplyametamathematical consistency
proof for the S3. —By (c) and Remark4, since ' is definable explicitly
from as anindividual symbol and +, the same holds for suchsystems
1
with the constants =, +, • (at least).

Starting fromMostowski and Tarski's example of a finitely axioma
tizable andessentially undecidable system, Mostowski and Tarski (1949
abstract), Tarski ( a,
1949 babstracts), JuliaRobinson
1949 (1949 abstract,
1949 ) and Raphael Robinson abstract) obtain in rapid succession
(1949
theundecidabilityof avarietyof mathematical theoriesinthe arithmetic

of integers and rationals, rings, groups, fields, lattices and projective
geometries.
Ch a p t e r XV
CONSISTENCY, CLASSICALAND INTUITIONISTIC SYSTEMS
§ 77. Gentzen’s formal system. InExample §73wefoundwhat 2
may be described as a direct way of deducing A(x) fromthe axioms

of in the predicate calculus when
S This -direct way can
(E y )R (x , y ).
lead fromthe axioms to A(x) only when To establish the

(E y )R (x t y ).
consistency property, i.e. that A(x) is deducible only when ), (E y )R (x , y
what we must do is to showthat a roundabout way of proceeding in

the predicate calculus can lead fromthe axioms to A(x) only when the
directwaydoes. Intheformalismofrecursivefunctions, thecorresponding
consistency problemwas trivial (§54, Example 3§60), preciselybecause
nootherthanthe direct wayof proceedingfromthe assumptionformulas
was allowedby the rules of the system. This leads us to inquire whether
theremaynot be atheoremabout the predicatecalculusassertingthat, if
a formula is provable (or deducible fromother formulas), it is provable
(ordeducible) inacertaindirect fashion; inotherwards, atheoremgiving
a normal formfor proofs and deductions, the proofs and deductions in
normal formbeing in some sense direct.
Atheoremof this sort was obtained by Gentzen - *. W e shall
1934 5
present it in §78, and apply it in §79 to obtain the consistency results

referred to in Example §60 and Example §73 (and used in §76),
2 2
as well as the consistency of number theory with the restricted rule of

induction (mentionedat the beginning of §42). These consistency proofs
canbegivenbyothermethods, as by Ackermann - , von Neumann
1924 5
1927 and Herbrand 1930 , - . All are somewhat long. Gentzen's is

1931 2
one of the easiest to follow, as the proof of his “Hauptsatz” or normal

formtheorem (of which it mainly consists) breaks down into a list of
cases, each of which is simple to handle. Another application of this
theoremisgivenin§80. Except incidentally, §§81and82areindependent
of §§77—80.
Gentzen's normal formfor proofs in the predicate calculus requires
a different classification of the deductive steps than is given by the
postulates of the formal system of predicate calculus of Chapter IV
440
§77 gentzen's formal system 441
(§19). The implication symbol D has to be separated in its role of
mediatinginferences fromits role as acomponent symbol of the formula
being proved. In the former role it will be replaced by a new formal
symbol (read "gives” or "entails”), to which properties will be
assigned similar to those of the informal symbol “b” in our former
derived rules.
Gentzen's classification of the deductive operations is made explicit
by setting up anewformal systemof the predicate calculus. The formal
system of propositional and predicate calculus studied previously
(Chapters IV ff.) we call now a H ilbert-type system , ?ud denote by H .
Precisely, H denotes any one or a particular one of several systems,
according to whether we are considering propositional calculus or pred
icate calculus, in the classical or the intuitionistic version (§23), and
accordingto the sense inwhichwe are using ‘term' and ‘formula' (§§17,
25, 31, 37, 72—76). Thesamerespectivechoices will applytothe Gentzen-
type system G \ which we introduce nowand the G2, G3 and G3a later.
The transformation or deductive rules of G 1 will apply to objects
whichare not formulas of the system but are built fromthemby an
H ,
additional formation rule, so we use a new term ‘sequent' for these

objects. (Gentzen says "Sequenz”, which we translate as "sequent”,
because we have alreadyused "sequence” for any succession of objects,
where the German is "Folge”.) Asequent is a formal expression of the
formAlf ..., A| -► Bj, ..., Bmwherel , m ^ 0 andAv ..., Au Bx, ..., Bm
are formulas. The part Alf ..., A, is the antecedent, and Bv ..., Bm
the succedent of the sequent Ax, ..., A x-^B^ ..., Bm .
When l,m
interpretation for G1 astheformula A±&
1, the sequent Alf ..., Aj->BX , ..., Bmhas the same
. . . & A t D Bt V . . . V Bm
The interpretation extends to the cases with / = 0 or m —
forH.
0by
regarding Ax&... &At for 0 (the "empty conjunction”) as true and
1 =
Bx V . . . V Bmfor m — 0 (the "empty disjunction”) as false.

A formula occurs in(orbelongs to) asequent Ax, ..., A t ->Blf ..., Bm,
if it is one of the l+ m occurrences of formulas A1#..., A lf Blf ..., Bm;
Forexample, G ?andG B S
and similarly for occurrence of a formula in antecedent or succedent.
7& but not occurinthesequent G G 4, ?&S G
Avariable (symbol, quantifier, etc.) occurs in asequent, etc., if it occurs
?.
in some formula of the same.
As in Chapter V, we use Greek capitals ‘T”, "A”, "0”, "A”, etc.
to stand for finite sequences of zero or more formulas, but now also as
antecedent (succedent), or parts of antecedent (succedent), with sep
arating formal commas included.
442 C O N S IS T E N C Y CH . X V
P ostulates for the formal system G 1

S tipulations : A , B , C , D are fo rm u las; r , A , 0 , A are fin ite sequences
of zero or m ore fo rm u las; x is a v a ria b le ; A (x ) is a fo rm u la; t is a term
free for x in A ( x ) ; an d b is a v a r ia b le free for x in A (x ) an d (unless b is x)
n o t occu rrin g free in A (x ).
R estriction on variables (for tw o of th e p o stu la tes as in d ic a te d ):

T h e v a ria b le b of th e p o s tu la te sh all n o t occu r free in its conclusion.
(W h en th e A (x ) does n o t co n ta in th e x free, th e n A (b ) is A (x ) no m a tte r
w h a t v a ria b le b is; w e agree in such a case to choose for th e a n a ly sis
a b n o t occu rrin g free in th e conclusion, so t h a t th e restrictio n is m et.)
The differen ce b etw ee n th e classical an d in tu itio n istic sy ste m s G\

is secured b y th e in tu itio n istic restriction s ta te d for tw o o f th e p o stu lates.
A x io m schem a.
C -*C .
L o g ic a l rules of inference for th e pro p o sitio n al calcu lu s.
Intr<ed uction o f in succed ent. in a n teced en t.
A, r->0,B A-> A, A B, r-»0

•mJ
r-j.©, a d b . a d b , a, r-»A, 0 .
r -* 0, a r-*©, b A,r-»0 B,r-*©
A
r-»0, a & b . A&B,T->.0. A&B,r->-0.
r-*-©, a r~>0 , b a , r-»© b , r-*©
vy
r-*©,AVB. r^ e.A V B . A v b , r -»0.
a , r ->■© r-*©, a
1
r - * 0 , “ 1A , a , r -> © .
w ith 0 e m p ty for
th e in tu itio n is tic sy stem .
A d d itio n a l lo gical rules of inference

for th e p red ic a te calculus.
In tro d u ctio n o f in succedent. in a n tec ed en t.

r - * 0 , A (b) A (t), r - * ©
V
r - * 0 , V x A (x ), V x A (x ), r - * 0 .
su b je c t to th e
restriction on variables.
§77 gentzen ’s formal system 443
1 r - * © , A (t)
r - * 0 , 3 x A (x ).
A (b ), r - * 0
3 x A (x ), r - * © ,
su b je c t to th e
restriction on variab les.
S tru c tu ra l rules of inference.
in su cced en t. in an teced en t.
r-»© r-*©
T h in n in g
r-»0, c, c, r-*©.
w ith 0 e m p ty for th e
in tu itio n istic sy ste m .
r -> ©, c ,c c, c, r - * 0
C o n tra c tio n
r - * 0 , c. c, r 0.
r -*• A , C, D , 0 A, D , c, r 0
In te rc h a n g e
r - * A,D, c, 0. A , C, D , F —> 0 .
A -*A ,C c ,r -» 0
C u t.
a , r -» A ,0.
F o r th e classical sy ste m G l , th e p o stu la te s e x c e p t th e tw o D -ru les

fa ll in to a d u a l-sy m m e tric arran gem en t, th u s. T h e rules for & a n d V
are d u al to each other, th e one set b e in g tran sform ed in to th e o th er b y
th e in terch a n ge of & w ith V an d -*► w ith S im ilarly, th e V - a n d 3-rules
are d ual. T h e a x io m schem a, th e -i-r u le s , an d th e stru ctu ra l rules o f th e
fo u r kinds, are each self-d ual.
T h e rules in th e le ft co lu m n w e ca ll succedent ru les ; an d w e d e n o te
th e m b rie fly b y &”, " -► -i” , % V M, " - * 3 ” , % r ,
C”, / ” , re sp e ctiv e ly . T h e rules in th e rig h t co lu m n w e c a ll
antecedent rules , an d d en ote b y “ D -V ', “& etc.
T h e lo gica l rules c o n s titu te in tro d u ctio n s o f a lo g ic a l sy m b o l, b u t
som etim es in th e su cced en t (left colum n), a n d som etim es in th e a n te c e d
en t (righ t colum n). T h e form u la in w h ich th e lo gica l sy m b o l is in tro d u c ed
is ca lled th e prin cipal form u la ; an d th e one or tw o form ulas sh ow n e x
p lic itly in th e prem ise(s) th e side form ula(s).
E xample 1. T h e u p p erm o st rule in th e rig h t co lu m n is “ D -in tr o -

d u c tio n in th e a n te c e d e n t” , or “ th e D -a n te c e d e n t ru le” , or b r ie fly
“ D T h e p rin cip al form u la is A D B , th e first prem ise h as A as side
fo rm u la, an d th e second B .
444 CONSISTENCY CH. XV
T h e lo g ica l rules of G 1 are m ore or less sim ilar in form to th e re sp e c tiv e
d eriv ed rules of T h eo rem 2 § 23, w ith th e u n d efin ed form al sy m b o l
a p p ea rin g n o w in p la ce o f th e d efin ed m e ta m a th e m a tic a l sy m b o l
An in tro d u c tio n in th e su cced en t corresponds to an in tro d u c tio n in
T h eo rem 2, an d in th e a n tec ed en t to an elim in atio n in T h eo rem 2.
T h e p resen t a x io m sch em a a n d stru ctu ral rules correspond to gen eral
p rop erties o f th e form er b as liste d in L e m m a 5 § 20.
T h e tree form (end § 24) is used in th e co n stru ctio n of proofs in G 1;
an d “ b S ” , w here S is a seq u en t, is used to express th a t th e seq u en t S
is p ro v a b le .
In e x h ib itin g proofs (or p a rts thereof), it is ted iou s to show se p a r a te ly
all th e a p p lica tio n s o f th e one-prem ise stru ctu ral rules. W e shall a d o p t
th e c o n v e n tio n t h a t a d o u b le line (w ith or w ith o u t c ita tio n of a n o th er
rule) s ta n d s for a seq uence of zero or m ore th in n in gs (“ T ” ), c o n tra ctio n s
(“C”) an d in terch a n ges (“ / ” ) (follow ing th e a p p lica tio n of th e o th er
rule w h e n a n o th er is cited).
E xample 2. F o r a n y form u las A , B an d C , (a) an d (b) are proofs

in G1 in tu itio n is tic a lly , (c) o n ly classically.
(a) B B C -*C
A A B 3 C, B C ^
A A B,AD(B d C),A->C j 3-*

A, A 3 (B 3 C), A 3 B -*C
A 3 (B 3 C), A D B ^ A D C “* D ^
A 3 B (A D (B 3 C)) 3 (A D C) ->D_>
(A 3 B) 3 ((A 3 (B 3 C)) 3 (A 3 C)). D
(b) A-» A (c) A -» A

A, “iA—>B — A, “iA
---------------------------- ► 3
-iA-» A 3 B ^ -i-iA A
-► -iA 3 (A 3 B). ~*D -»■ - i - i A 3 A.
Gl, w e v e r ify b y in d u ctio n :
F r o m th e p o stu la te list for
L emma 32a. If j- r —* B x, . . B m in the intuitionistic system G\,

then m — 0 or m — 1.
L emma 32b. If b T -^B ,, . . . , B m in the intuitionistic system Gl
without using the rule —i—>•, then m — 1.
§77 gentzen ' s formal system 445
L e m m a 3 2 a expresses th e w h o le difference b etw ee n th e in tu itio n istic
a n d th e classical sy ste m G 1; b u t it is u seful to ob serve t h a t w e n eed ed to
restrict o n ly tw o of th e p o stu la te s for th e in tu itio n istic sy s te m G l to
secure th is difference.
O u r first o b je c tiv e is to sh ow (as d id G en tzen ) t h a t sy s te m G 1 is e q u iv a le n t
to our form er sy ste m H, in th e sense th a t, for a n y fo rm u la E , |— > E
in G l , if an d o n ly if h E in H. T h is result w ill b e co n ta in ed in the
n e x t tw o th eorem s for T e m p ty .
T heorem 46. I f T b E in H with all variables held constant, then

in G l . W hen the given deduction in H uses the postulates only for
j- T - » * E
certain of the sym bols D , & , V, - i , V , 3, then the resulting proof in G l uses
the logical rules only for D and for the sam e sym bols.
P roof, b y co u rse-o f-valu es in d u ctio n on th e le n g th o f th e g iv e n de
d u ctio n in H of E from T. S ix te e n cases arise, as follow s, a c co rd in g t o the
a n a ly sis for E in th e g iv e n d ed u ctio n .
Case 0: E is one of th e form ulas V. T h e n th e fo llo w in g is a p roof

in G l of r - » E .
E -* E
r->E .
T h e oth er cases w ill b e num bered as th e resp ective p o stu la te s la , lb,

2, 3, 4a, 4 b , 5a, 5b, 6, 7, 8 or 81, 9 — 12 of H.
Case lb : E is an ax io m b y A x io m S c h em a lb , i.e. E is
(A D B) D ((A D (B D C)) D (A D C)) for som e form ulas A , B and C.
W r itin g T before th e in th e en dsequen t of E x a m p le 2 (a), an d d o u b lin g
th e line o ve r it (to represent thin nings), m akes t h a t tree in to a proof in
G l of T ~ * E .
Case 2: E is an immediate consequence of two preceding formulas

by Rule 2, i.e. for some pair of formulas Aand B, Eis Band the two
preceding formulas are A and ADB. By the hypothesis of the in
duction, thereareproofs inGl of Aand F-ÂDB, Graftingthese
twoproofs ontothe followingtree, weobtainaproof of T-* B.
A-*A B-*B ^
T-* A D B A D B, A -*B
_ _ —_ _ ——-------L--—-- Cut
r-*A a, r~*B ^
— ------- - .........................—Cut
r-*B.
Case 8 or 81. See Example 2 (c) or (b), respectively.
446 C O N S IS T E N C Y CH. X V
C a s e 9: E is an im m e d ia te consequence of a p reced in g fo rm u la
b y R u le 9. T h e n E is C D V x A ( x ) , an d th e preceding form u la is C D A (x ),
w here C does n o t co n ta in x free. L e t I \ b e th e subsequence of T com prisin g
th o se of th e form ulas T on w h ich th a t precedin g form ula C D A (x )
d ep en d s in th e g iv e n d e d u ctio n of E from T. B y o m ittin g from th e g iv e n
d e d u c tio n all form ulas b elo w th e form ula C D A (x) an d those a b o v e
w h ic h d ep en d on o th er assu m p tio n form ulas th a n r^, w e o b ta in a d e
d u c tio n o f C D A (x ) from Fv A p p ly in g th e h yp o th e sis of th e in d u ctio n
to th is d ed u ctio n , there is a proof in G l of 1 ^ — D A (x ). Since th e v a ria
b les are h eld co n sta n t in th e g iv e n d ed u ctio n of C D V x A (x ) from T,
n on e o f th e form u las co n ta in s free th e va ria b le x of th e ap p licatio n
o f R u le 9. T h is an d th e fa c t th a t C does n o t co n ta in x free are used in
v e r ify in g th a t th e restriction on va riab les is satisfied for th e in
th e follow in g.
C-»C A (x )-» A (x )
T1- > C d A(x) C d A(x),C -^ A ( x)
c, rt-» A (x ) “ ut
C, T j —► V xA (x)
T -> C D VxA(x).
Case 12: E is an im m e d ia te consequence of a p reced in g form ula
b y R u le 12. D u a l to C ase 9.
W e in tro d u ce th e n o ta tio n to be used in T h eorem 47. L e t F b e som e

p a rtic u la r closed form ula. S a y th a t 0 is B lf ..., B m (m > 0). T h e n le t
0 ' be B lt ..., B m_1 (e m p ty if m <> 1), 0 " b e B m if m > 1 an d - i (F D F )
if m = 0, “ i 0 b e - i B ^ . . . , ~ i B m, an d - i 0 ' be “ i B 1, . . . , (em p ty
if m ^ 1).
I f T, -i© ' |- 0 " in H with all variables held constant ,

C o r o l l a r y 1.
and provided for the intuition istic system s that m < 1, then f- 0
in G L
U se cases acco rd in g as m = 1, w = 0 or w > 1.
C o r o l l a r y 2. F or the classical (intuitionistic) system s when l,m ;> 1

(/ ;> 1, m = 1): I f h A x & . . . & A t D B x V . . . V B m in H , then
h A 1# . . A { B lt . . B m in G L
T heorem 47. I f b F - > 0 in G l, then T, - i 0 ' f- 0 " in H with all
variables held constant. (In p a rticu lar: If |- F —> E in G l , then T j- E
in H w ith all variables held constant.)
§77 gentzen ' s formal system 447
M oreover, for the in tuition istic {classical) system s, when the given proof
in G \ uses the logical rules only for certain of the sym bols D, & , V, - i , V , 3,
then the resulting deduction in H uses only the D- postulates {the D- and
—i -postulates) and the postulates for the sam e sym bols, provided in case the
sym bols include V but not & the V-postulates include A xiom Schema 9a
of Lem m a 11 § 24.
P roof . W e p ro ve th e first sta te m e n t of th e theorem b y cou rse-o f-
v a lu e s in d u ctio n on th e h e ig h t (i.e. n um ber of levels) of th e g iv e n proof
in G\ of T —> 0 , u sin g cases acco rd in g to th e p o stu la te of G 1 ap p lied last
in th is proof.
T h e a d d itio n a l d eta ils g iv e n in th e second p a rt of th e theorem w ill
be ve rified a fter w e h a v e gone th ro u gh th e cases. I n all b u t th e cases
an d subcases in d ic a te d b y a single or d o u b le star, th e d em o n stratio n o f
th e existen ce o f th e resu ltin g d ed u ctio n is alm o st im m e d ia te b y use o f
gen eral properties of b an d, in th e case of a lo gical rule of G 1, an a p
p lic a tio n of th e corresponding d erived rule (other th a n a i-r u le ) of
T h eo rem 2 § 23, or in th e case o f —> V an d 3 of th e correspon ding stro n g
rule of L e m m a 10 § 24. In th e cases m ark ed b y a single star, w e use also
th e in tu itio ftistic - i -rules ( i - i n t r o d . an d w e a k -i-e lim . § 2 3 ) an d som e
tim es * 1 § 26. In th e cases m ark ed b y a d o u b le star, w e fu rth er use th e
classical (strong) - i -elim in atio n rule (Theorem 2). N o n e of th e nine cases
n o t m en tio n ed b elo w requires either subcases or starring.
Ca se 1: th e a x io m schem a. B y gen eral properties of h C b C in H.

Case 2: -> D . W e h a v e b y th e h yp o th e sis of th e in d u ctio n th a t
A , T , "i 0 b B in H w ith all v a riab les held co n sta n t, an d w e m u st infer
th a t T, 0 b A D B likew ise. T h is w e can do b y D -in tro d .
Case 3: D S ubcase 1: A e m p ty or 0 n o t e m p ty . T h e n (A , 0 ) "

is 0 " . B y h y p . ind., A , - i A b A an d B , T, - i 0 ' b 0 " . T h en ce, u sin g
D -e lim ., A D B , A , T , A , - i 0 ' b 0 " . S ubcase 2 * * : A n o t e m p ty an d
0 e m p ty . B y h y p . ind., A , - i A b A an d B , T b ~i (F D F ). B y D -e lim .,
A D B , A , T, - . A b - i ( F D F ). U sin g * 1 , A D B , A , T, - A b F D F . i
H en ce, b y -i-in tr o d . an d i - e l i m . , A D B , A , T , i A ' b n A " b A ".

Case 10*: - > i . S ubcase 1: 0 e m p ty . F ro m A , T f- - ( F F), i d
w e infer T b ~ iA b y *1 an d -i-in tr o d . S ubcase 2 : © n o t e m p ty . F ro m

A , T , - i 0 ' b 0 " , w e infer V, i 0 ' , - i 0 " b i A b y -i-in tr o d .
Case 1 1 : —i S ubcase 1 * : 0 e m p ty . U se w e a k - i -elim . S ubcase 2 * * :

0 n o t e m p ty . U se -i-in tr o d . an d —i-elim .
448 C O N S IS T E N C Y CH. X V
Case 12 : - * V . B y h y p . ind., T ,- 1 © b A (b ) w ith all va riab les h eld

co n sta n t. B y stron g V -in tro d ., A (b) bb V x A (x ). H en ce F, - i 0 b V xA (x),
w ith all v a ria b les in clu d in g b h eld co n sta n t, since b y th e restriction on
v a ria b les for - > V , F and 0 do n o t co n ta in b free.
Case 16: a n y one-prem ise stru ctu ra l rule. L e t th e prem ise be 0,

an d th e con clusion b e F* 0L Then T* ( 0 +) com es from F (0 ) b y
p erm u tin g form ulas, su p p ressin g repetitio n s o f form ulas, or in tro d u cin g
new form ulas. S ubcase 1: 0 t is 0 . F ro m F, - i 0 ' b 0 " , w e infer T*,
“i 0 ' b 0 " b y gen eral properties of h S ubcase 2 * : 0 t is not 0 , an d 0 is
e m p ty . B y h y p . ind., F b —«(F D F ). T h en c e b y gen eral properties of b,
r t , -i© * ' b “ i ( F D F ). Th ence by *1 an d w eak -i-e lim ., F^, -i0t'
b 0 *". S ubcase 3 * * : 0 * is n o t 0 , an d © is n o t e m p ty . T h e n 0 t is n o t
e m p ty , an d a t least one o f 0 an d ©t consists of m ore th a n one form ula.
B y h y p . ind., T , - i 0 ' b 0 " . B y w e a k -i-e lim ., T , - i © ', - i 0 " b “ i ( F d F),
i.e. T, - i 0 b i (F d F). T h e n c e b y gen eral properties of h F^, -i©t
b - i ( F D F ), i.e. r t , - i 0 t ' , - i 0 t " b - i ( F D F ). T h e n c e b y * 1 , -i-in tr o d .
an d -i-e lim ., r t , - i 0 t ' b © t".
Case 17 : C u t. T r e a te d as C ase 3, e x c e p t w ith o u t usin g D -elim .

S ubcase 1: A e m p ty or 0 n o t e m p ty . S ubcase 2 * * : A n o t e m p ty an d
0 e m p ty .
T o v e r ify th e second sta te m e n t of th e theorem , first suppose th a t

th e g iv e n proof in Gl of 0 w here 0 is B lt . . . , B m is in tu itio n istic
a n d does n o t use th e - i -rules. T h e n usin g L e m m a 32b, w e see t h a t none
of th e s in g ly or d o u b ly starred cases or subcases ca n occur. T h e sta te m e n t
th e n follow s from th e m an n er o f tre a tm e n t o f th e u nstarred cases, b y
Lem m a 11 § 24.
I f th e p roof does use th e - i -rules b u t is in tu itio n istic, th e sta te m e n t
p erm its th e in tu itio n istic - i -p o stu la te s o f H to be used in th e resultin g
d ed u ctio n , an d w e need o n ly v e rify , u sin g L e m m a 32a, th a t none of th e
d o u b ly starred cases or subcases can occur.
Corollary . W hen l, m > \ (and hence for the intuition istic system s ,
m = 1): I f b A x, . . . , A j B lt . . . , B m in G 1, then
b A x & . . . & Aj D B x V . . . V B m in H .
§ 78. Gentzen’s normal form theorem. E xample 1. T h e p roof (a)

co n ta in s a c u t, w h ile (b) is a proof w ith o u t cut of th e sam e
sequent.
§78 gentzen ' s normal form theorem 449
T h e first proof uses an unn ecessarily co m p licated form u la ^x^l{x) in

th e a n teced en ts o f th e righ t bran ch. T h is co m p licatio n is u n rave lled
b y m eans of th e cu t. T h e second proof proceeds d ir e c tly , w ith o u t in
tro d u cin g co m p licatio n s th a t are su b seq u en tly un ravelled .
T h e sign ifican ce of proofs w ith o u t c u t, as in a sense proofs in norm al

form , is further em phasized b y th e su bform u la p ro p e r ty (L e m m a 3 3a
below ).
W e define ‘su b fo rm u la’ of a g iv e n form ula thus.
1. I f A is a form ula, A is a subformula of A . 2 — 4. I f A an d B are
form ulas, th e subformulas of A an d th e subform ulas o f B are subform ulas
of A D B , A & B and A V B . 5. I f A is a form ula, th e subform ulas o f
A are subform ulas of n A . 6 — 7. I f x is a va riab le, A (x ) is a form ula,
an d t is a term free for x in A (x ), th e subform ulas subform ulas
of A (t) are
o f V x A ( x ) an d 3 x A (x ). 8. A form ula h as o n ly th e subform ulas required
by 1 — 7.
E xample 2. T h e subform ulas of c3f D (-i D 2) are th e fiv e

form ulas <£? D (-i<£7 D 2), <C?, -i<£¥ D 2, ~i<C7, 2.
E xample 3. (a) The subform ulas of Vi>Vc(2(c) & <Z!(t?)) are th e

form ulas VbVc(B(c) &a(b)), V c(2(c) & <3(t)), 2(u) & c3(t), 2(u ), <3T(t)f
for e v e r y term t n o t c o n ta in in g c free an d e v e r y term u. (b) T h e o n ly
su bform u la o f <^(c) is J7(c).
F ro m th e p o stu la te list for G 1, w e v e r ify b y in d u ctio n :

L em m a 33a. Each formula occurring in any sequent of a proof in G1
without a cut is a subformula of some formula occurring in the endsequent.
(S u b form u la prop erty.)
L em m a
33b. Each formula occurring in the antecedent (succedent)
of any sequent of a proof in Gl without a cut or application of the D - or
—i-rules is a subformula of some formula occurring in the antecedent (suc
cedent) of the endsequent.
G e n tz e n ’s “ H a u p t s a t z ” or norm al form theorem (Theorem 48 below)
asserts t h a t th e cu ts can a lw a y s be elim in ated from th e proof of a n y
seq u en t in w h ich no v a ria b le b o th occurs free an d occurs bound.
T h e restrictio n th a t no v a ria b le occur b o th free an d b o u n d in th e end-
seq u en t does n o t d e tra ct from th e usefulness of th e theorem . F o r w hen
va ria b le s do occur b o th free an d b o u n d in a g iv e n seq uent, b y replacin g
th e form ulas b y others con gru en t to th e m w e can o b ta in a sequent
sa tis fy in g th e restriction an d p ro v ab le if an d o n ly if the g iv e n one is
p ro v a b le (b y T h eorem 47, L e m m a 15b § 33, * 18a an d * 18b § 26, an d C orol
la r y 1 T h eo rem 46). T h e restriction is necessary, as w e illu stra te now .
E x a m pl e 4. Consider th e proof
T h e c u t ca n n o t b e elim in ated . F o r b y L e m m a 33a, no seq u en t in a proof

w ith o u t c u t of Vi>Vc(#(c) & < 3 f(A ))-» c3(c) can co n ta in th e sym b o ls D
an d ; so th e Z>- an d -i-r u le s can n o t be used. H en ce L e m m a 33b applies.
B u t th e tw o lists of subform ulas in E x a m p le 3 h a v e no form ula in co m
m on, so th a t no axio m satisfies th e requirem ent of L e m m a 33b. (The
m eth o d of t h i s ex a m p le w ill be d evelo p ed further in § 80.)
In p ro v in g th e norm al form theorem , w e use a form al sy ste m G2

o b ta in e d from G l b y ch a n gin g tw o of th e p o stu lates, as follows.
T h e c u t is replaced b y th e fo llo w in g rule, called “ m ix ” . H ere M is
a form u la (the mix formula)', II, ®, 2 , Q are sequences of zero or m ore
form ulas such th a t b o th O an d 2 con tain M ; and and 2 M are th e
results of suppressing all occurrences of M in <1> an d 2 , resp ectively.
M ix
E x am ple 5. The following is a mix.
M ix
The 3 -► of G l (d istin gu ish ed as 3 - ^ ) is replaced b y a n ew one

( 3 - * 2), due for th e classical sy ste m to K e to n e n 1944. H ere A , B , T, 0
are as a b o v e ; an d 0 ° is 0 for th e classical sy ste m an d e m p ty for th e
in tu itio n istic.
r -» 0 °,A B ,r-*© ^
A D B, r - * 0. D->>2
A n y cu t can b e acco m plish ed b y a m ix (below left), an d v ic e ve rsa

(right), w ith th e h elp of TCI steps.
a a, c cfr-*©w. n-*a> s -* q
_____________________ M ix «====*============== -
A , T c —> A c , 0 n —* 0 M, M M , S M —* £2
_ _ _ _ n, <d m, q . u
S im ilarly, a n y 3 — c a n be accom plish ed b y a 3 —* 2 (below ), n o tin g

th a t for th e in tu itio n istic case A is e m p ty b y L e m m a 3 2 a ; an d v ic e
ve rsa (left to th e reader).
A -*A ,A B, T - * 0
A, r - * A , 0 ° , A B, A, T - * A , 0
A D B A A 0 = 5 -> 2
T h is show s th a t th e tw o ch an ges do n o t alter th e class o f th e p ro v a b le

seq uents, as w e sta te w ith ad d itio n al d e ta ils in L e m m a 34.
A proof in G l or in G 2 is said to h a v e th e pure variable property , or
to be a pure variable proof , if no v a ria b le b o th occurs free a n d occurs
b o u n d in th e proof, an d for each a p p licatio n of V or 3 -*> th e v a ria b le
b of th e ap p lica tio n occurs o n ly in seq uen ts a b o v e th e conclusion o f th e
a p p lica tio n (where if th e A (x ) does n o t c o n ta in th e x free, w e choose
for th e a n a ly sis a b o ccurring o n ly thus).
L emma 34. I f |- T - > 0 in G l , then b T - * © in G 2 ; and conversely.

The proof in G2 contains a m ix , only if the proof in G l contains a c u t ;
and conversely. Either proof contains applications of exactly the sam e-nam ed
logical rules as the other. I f either proof has the pure variable property , so
does the other. Lem m as 32a— 33b {stated above for G l and the cut) hold
good also for G2 and the m ix.
L emma 35. A given postulate application of G 1 or G2 rem ains an
application of the same postulate when, in the sequent(s) of the application
{nam ely, the axiom in the case of the axiom schema, and the prem ise{s) and
conclusion in the case of a rule of inference), a variable is changed in exactly
its free occurrences {or in exactly its bound occurrences) to another variable not
occurring either free or bound in the sequent{s) in question.
Lem m a 35, in th e case of an a p p lica tio n of V or 3 depend s on
th e fa c t th a t th e b an d th e x do n ot need to b e th e sam e v a ria b le (as
th e y did for th e correspon ding rules 9 an d 12 of th e sy ste m H ).
L emma 36. A given postulate application of G 1 or G2 rem ains an
application of the sam e postulate when, in the sequent{s) of the application,
a given term is substituted for {the free occurrences of) a given variable,
provided in the case of a n —► V or 3 that (i) the term does not contain the
b of the application and (ii) the variable is not the b of the application,
and also provided in every case that (iii) the term is free for the variable in
each form ula of the sequent{s).
C o n d itio n (i) insures th a t th e restriction on v a riab les rem ains satisfied
for an V or 3
L emma 37. Given a proof in G 1 or G2 of a sequent in which no variable

both occurs free and occurs bound, by changes in free and bound occurrences
of variables in sequents of the proof {each postulate application rem aining an
application of the sam e postulate), we can obtain a proof of the sam e sequent
in which proof no variable both occurs free and occurs bound.
P roof. Let x lf . . . , x n b e th e d istin ct v a riab les w h ich occur free
in th e en d seq u en t an d occu r bound (elsewhere) in th e proof, and
a1, ..., a p be th e oth er d istin c t v a ria b les w h ich occur b o th free an d b o u n d
in th e proof. L e t y lf . . . , y n, b x, . . . , b P be d istin ct va ria b le s n o t occurring
in th e proof. C h an ge xv . . . , x n in their b o u n d occurrences o n ly to yv . . . , y w,
re sp e ctiv e ly , an d a x, . . a P in th eir free occurences o n ly to b j, . . . , b^.
B y Lem m a 35, th e figure rem ains a p r o o f; an d b y th e h y p o th e sis th a t th e
en d sequ en t co n ta in s no v a ria b le b o th free an d b ou n d , th e en d seq u en t is
u n ch an ged .
L emma 38'. Given a proof in G \ or G2 in which no variable occurs

both free and bound, by changes only in free occurrences of variables in it
{each postulate application rem aining an application of the sam e postulate),
we can obtain a pure variable proof of the sam e sequent.
§78 gentzen ’s normal form theorem 453
P roof. Su p p ose that in th e g iv e n proof in Gl or G2 there are e x a c t l y
q a p p lica tio n s of th e rules V an d 3 —► w ith re sp e ctive v a ria b le s c x, . . . , cg
(not n ecessarily d istin ct) as th e b 's o f th e ap p licatio n s. Select any q
d is tin c t v a ria b le s d lf . . . , d* n o t occurring in th e g iv e n proof. C hoose
one o f th e a p p lica tio n s w h ic h is u p p erm o st (i.e. h as no o th er o v e r it),
s a y its b is cv S u b s titu te d x for c x th ro u g h o u t all seq u en ts a b o v e th e c o n
clusion of th e a p p lica tio n , b u t now here else. U n d er th e restrictio n on
v a ria b les for V an d 3 ct ca n n o t occu r free in th e con clusion o f th e
a p p lica tio n , so u sin g L e m m a 35 a ll p o stu la te a p p lica tio n s re m a in v a lid .
R e p e a t th e procedure, each tim e w o rk in g on one of th e a p p lic a tio n s of
V or 3 w h ich is u p p erm o st am o n gst th o se n o t y e t tre a te d , u n til
th e b 's o f all q of th e a p p lica tio n s h a v e b een ch a n g e d from c ’s to d ’s.
B e ca u se each s u b stitu tio n alters o n ly seq u en ts a b o v e th e co n clu sio n of
an a p p lica tio n of V or 3 — t he en d seq u en t is un altered.
T heorem 48. Given a proof in G \ {in G2) of a sequent in which no

variable occurs both free and bound , another proof in G l {in G2) of the sam e
sequent can be found which contains no cut {no m ix). T h is proof is a pure
variable proof. The only logical rules applied in it are ones which were
applied in the given proof. (G en tzen 's H a u p ts a tz or norm al form th eo rem ,
I934 -5-)
P roof , red u cin g th e theorem to a lem m a. B y L e m m a s 34, 3 7 an d 38,

it suffices to p ro v e th e theorem for G2 assu m in g th e g iv e n proof a lre a d y
to h a v e th e pure v a ria b le p ro p erty . W e do so b y in d u ctio n on th e n u m b er
m of m ix es in th is 'g iv e n p r o o f. I f m > 0, there m u st o ccur in it a m ix
w h ich has no o th er m ix o ve r it. Consider th e p a rt of th e g iv e n proof
w h ich term in ates w ith th e con clusion II, S M ®M, O of th is m ix ; c a ll
it th e 'g iv e n p a rt'. S u p p o se th a t w e ca n tran sfo rm th is g iv e n p a rt so as
to o b ta in a n o th er p roof in G2 o f II, Z M O m, Q w ith o u t m ix ; ca ll it th e
're su ltin g p a r t ’ . T h e n th e replacem en t of th e g iv e n p a rt b y th e re su ltin g
p a r t in th e g iv e n proof g iv e s us a n ew proof in G 2 o f th e sam e seq u en t
w ith o n ly m—1 m ixes. S u p p o se fu rth er th a t th e resu ltin g p a rt ca n be
co n stru cte d so th a t it has th e pure v a ria b le p ro p e r ty b y itself, a n d also
co n ta in s no v a ria b le free (b o u n d ; as th e b of an V or 3 - * ) t h a t d id n o t
so o ccu r in th e g iv e n part. T h e n th e n ew p roof as a w h o le w ill h a v e th e
p ure v a ria b le p ro p erty . H e n ce w e ca n a p p ly th e h y p o th e sis o f th e in
d u c tio n to co n clu d e th a t there is a pure v a ria b le p roof w ith no m ix. T o
p ro v e th e th eorem it th u s rem ains to estab lish th e fo llo w in g lem m a.
E x am ple 1 (continued). The g iv e n p a rt for (a) (restated w ith a

m ix in s te a d of a cut) is as follow s.
^ l( b)^ a{b) 3x^(x) - » lx^(x)
£2(b) -*3x<Zl(x) -i3xH(x)t 3xCl(x)
<3(6), —i3jc<C3T(jc) ->.
L e m m a 39. Given a proof in G2 of II, S VI 0 M, O with a m ix as
the fin al step , and no other m ix , aw^ with the pure variable property , a
proof in G2 of H, S M - > O m, Q caw 6c found with no m ix , with the pure
variable property , a w i with no variable occurring free (bound ; as b o/ aw
—^ V or 3 - * ) which did not so occur in the given proof . Ow/y logical rules
are applied in the resulting proof which are applied in the given proof.
(P rin cipal lem m a.)
P roof o f t h e p r in c ip a l le m m a . W e define th e left rank a of a m ix

as th e greatest n um ber o f sequents, lo cated c o n se c u tiv e ly one a b o v e
an oth er at th e b o tto m of a n y b ran ch term in atin g w ith th e left prem ise of
th e m ix, w h ich co n ta in th e m ix form ula M in th e succed ent. T h e right
rank b is d efined sim ilarly. T h e rank r = a + 6 . (The least possible ran k
is 2.) T h e grade g of th e m ix is th e num ber ( > 0) of occurrences of lo gical
sym b o ls ( D , & , V, - i , V , 3) in th e m ix form ula M.
E x a m pl e 1 (concluded). T h e left ran k is 1, th e righ t ran k 2, th e

ran k 3, and th e grade 1.
T h e lem m a is p ro v ed b y cou rse-of-valu es in d u ctio n on th e grade g

of the m ix. W ith in b o th th e basis an d th e in d u ctio n step of th is in d u ction ,
a co u rse-o f-valu es in d u ctio n is used on th e rank r. W e g iv e a trea tm en t
b y cases, so th a t b y d raw in g upon th e results of th e cases th e bases an d
in d u ctio n steps of th e in d u ctio n s can all be carried through.
W e w rite th e m ix to be elim in ated thus,
S i____ S 2
or b riefly
S 3,
w here M £ (S> an d MSS.
T h e letters " A " , " B ” , “ C ” , “ D ” , T " , “ 0 ” , " x ” , “ A ( x ) ” , " t ” , “ b ” w ill
refer to th e sta tem e n ts of th e other p o stu lates in question.
A. P r e l im in a r y c a se s .
C a se la : S x has M in th e an teced en t, i.e. M £ II. T h e n th e co n
clusion II, S M —> ®M, Cl of the m ix com es from its second prem ise S - > Di
by T C I, an d is hence p ro v ab le w ith o u t m ix. T h e in tu itio n istic restriction
on the —> T 's can be satisfied , as the TCI steps term in ate w ith th e
origin al endsequent, w h ich b y L em m a 32a can h a v e in tu itio n istic a lly
a t m ost one form ula in its succedent.
§78 gentzen ’s normal form theorem 455
C ase 2 a : th e left ra n k is 1, an d S x is b y a stru ctu ra l rule. Sin ce
M 8 $ , b u t n o t to th e su cced en t of th e prem ise for th e inference of S 1,
th e inference can o n ly b e a T w ith M as th e C. T h u s th e b o tto m of th e
g iv e n proof is as a t th e left b elo w w ith M t 0 . W e alter th is to th e figure
a t th e righ t, to o b ta in a p roof of th e original en d seq u en t w ith o u t m ix.
M ix
C ases lb , 2 b : sim ilarly, read in g " S 2” , " s u c c e d e n t” , " Q ” , " r ig h t ” in

p la ce of " S j ” , " a n te c e d e n t” , " I I ” , " le f t ” , re sp e ctiv e ly . T r e a te d s y m
m e tr ic a lly to Cases la , 2a.
B. O th er cases. F o r each of th ese cases, it is p a rt of th e case h y

po th esis th a t none o f th e four p relim in ary cases applies.
B l : th e ra n k is 2 . Sin ce Cases la , l b are ex clu d ed , S x an d S 2 m u st

b o th b e conclusions of inferences. Sin ce th e ra n k is 2 an d Cases 2 a an d 2 b
are ex clu d ed , b o th inferences m u st b e logical. M oreover, since th e ra n k
is o n ly 2, M m u st b e th e p rin cip al form ula of b o th inferences. T h erefo re
M co n tain s a lo gical sy m b o l, an d th e m ix is of gra d e > 1 . T h u s th ese
cases ca n arise o n ly under th e in d u ctio n step for th e in d u ctio n on grade,
an d th e h yp o th esis of th e in d u ctio n on th e gra d e is a v a ila b le in tre a tin g
th em . T h e rules used in inferring S x an d S 2 ca n o n ly b e, re sp e c tiv e ly , th e
rule in tro d u cin g th e ou term ost lo gical sy m b o l of M in su cced en t, an d th e
rule in tro d u cin g th e sam e sy m b o l in an teced en t. T h u s under B l we
h a v e o n ly th e fo llo w in g six cases.
C a s e 3: S x is b y D , an d S 2 b y D -*, w ith M as th e prin cip al

form u la A D B . T h e n th e b o tto m of th e g iv e n proof is as follow s, w here
A d B t 0 , T since th e ra n k is o n ly 2.
W e a lte r th is to th e follow ing.
M ix
T h e upper m ix is of grade low er th a n th e original, so b y th e h yp o th e sis
of th e in d u ctio n on th e grade, it can be elim inated , i.e. a proof w ith o u t
m ix of its conclusion can be found. T h e n th e lower m ix w ill h a v e no m ix
o ver it, an d can likew ise be elim inated. T h u s w e o b ta in a proof w ith o u t
m ix of th e original endsequent n , F 0 , Q.
C ase 4: S x is b y & , an d S 2 b y & w ith M as th e prin cipal

form ula A & B .
C ase 5: S x is b y V, an d S 2 b y V— w ith M as th e p rin cipal

form ula A V B . T r e a tm e n t is d u al to th a t of Case 4.
C ase 6 : S j is b y - i , an d S 2 b y - » wi t h M as th e p rin cip al

form ula —i A .
C ase 7: St is b y ~ ^ V , an d S 2 b y V — wi t h M as th e prin cipal

form ula V x A (x ). W e h a v e as th e g iv e n figure, w here V x A (x ) t 0 , T,
In co n stru ctin g th e altered figure, w e shall need a proof of II 0 , A (t).

I f A (x ) does n ot co n ta in x free, th en A (b) an d A (t) are th e sam e form ula,
an d w e a lrea d y h a v e it. Suppose th en th a t A (x ) does co n ta in x free. T h e n
A (t) con tain s t, an d so, since t is free for x in A (x ), th e va riab les of t occur
free in A (t). N o w w e can con clude from th e pure v a ria b le p ro p e rty of th e
g iv e n proof of II, T —> 0 , Q th a t (i) no va ria b le of t is th e b of an - > V
or 3 in th e proof of II —> 0 , A (b) in th e left b ran ch of our figure,
(ii) our b is not th e b of an - * V or 3 in th e sam e, an d (iii) t is
free for b in all form ulas of th e sam e (since th e pure v a ria b le p ro p e rty
allow s no va ria b le to occur b o th free an d bound). M oreover, in v ie w
of th e restriction on v a riab les for th e ex h ib ite d V , (iv) b does n o t
occur in II, 0 . Therefore, usin g L e m m a 36, b y su b stitu tin g t for b
th ro u gh o u t th e proof of II 0 , A (b) w e o b ta in a proof of II -*■ 0 , A (t).
W h en th is proof is used in th e altered figure as follow s, th e new proof of
II, T 0 , Q th u s o b ta in ed w ill h a v e th e pure va ria b le p ro p e rty an d
w ill co n tain free (b o u n d ; as th e b of an V or 3 -> ) no va ria b le n o t
p re v io u sly so occurring.
Case 8 : Sx is b y -> 3 , and S2 by 1 -* , with M as the principal
form u la I x A ( x ) . D u a l to Case 7.
B2: th e ra n k is > 2. T h ese cases ca n o n ly occur in th e in d u ctio n step

o f an in d u ctio n on th e ra n k (either w ith in th e basis, or th e in d u ctio n
step , o f th e in d u ctio n on th e grade). H en ce th e h yp o th e sis of th e in
d u ctio n on ra n k is a v a ila b le in tre a tin g them .
B 2 . 1: th e le ft ran k is > 2. T h e n M occurs in th e su cced en t o f a t

least one o f th e prem ises for th e inference o f Sv
Case 9 a: S x is b y a su cced en t stru ctu ral rule S w ith M n o t th e C
a n d n o t th e D , or S x is b y an a n teced en t stru ctu ral rule S. L e t th e g iv e n
figu re b e as show n a t th e left. W e alter th is as sh ow n a t th e rig h t (ex
p la n a tio n follows).
B eca u se M £ ® x (b y B 2 .1 ), w e ca n ta k e -*< !>! as first prem ise for

th e new m ix. T h e n from th e case h yp o th e sis an d th e form o f th e on e-
prem ise stru ctu ral rules, w e v e r ify th a t th e new S is correct. In th e altered
figure, th e ran k of th e m ix is one less th a n in th e original. H en ce b y th e
h y p o th e sis of th e in d u ctio n on th e rank, w e ca n fin d a proof of its c o n
clusion, an d hence of th e original en dsequen t, w ith o u t m ix .
Case 10 a : S x is b y a su cced en t stru ctu ra l rule S w ith M as th e C or

th e D . L e t th e prem ise b e II O x. F ro m th e form o f th e su c ced en t
stru ctu ral rules, w e v e r ify th a t th en ® 1M an d ®M are id en tica l.
H ®_____ li-+ Q.
------------ L 5
n, ZM1 ®M, Q.
------------ ---------------— M ix
n, sM-> ®M, o.
Case 1 1 a: Sx is b y a one-prem ise lo gical rule L. T h e inference b y a n y
one of these rules w h ich g iv e s S x has th e form
Av r->e, l a 2
slf r-*©, e2
w here each o f A ^ A 2 is either a side form ula or e m p ty , an d one o f Ev S2
is th e p rin cip al form u la w h ile th e oth er is e m p ty .
458 CONSISTENCY CH. XY
S ubcase 1 : S 2 is not M. T h e n M 8 0 , since M 8 0 , S 2. W e w rite th e
g iv e n figure thus.
W e alter to th is (exp lan atio n follows).

M ix
T h e new L has F , S M as its T an d 0 M, Q as its 0 . T h e conclusion of th e

n ew L is th e original en d sequ en t, e x c e p t for th e order of form ulas w ith in
th e succed ent. T h is enables us to infer th a t, if L is V or 3 th e
restriction on va riab les is satisfied for th e new a p p licatio n , since b y th e
pure v a ria b le p ro p e rty of th e g iv e n proof, th e b ca n n o t occur in th e
origin al endsequent. I f th e g iv e n proof is in tu itio n istic, th en (b y L e m m a
32a) 0 , A 2 consists of a t m ost one form ula. B u t M S 0 ; h ence A 2 is
e m p ty . T h erefo re th e TCI step s p reced in g th e new L require no v io la tio n
of th e in tu itio n istic restrictio n on T. (The p o ssib ility th a t th e new L
is a - i v io la tin g th e in tu itio n istic restriction is ruled o u t a priori, b y
com parison of its con clusion w ith th e g iv e n en d se q u en t; b u t in fa c t L
ca n n o t b e — - i .) T h e new m ix is of ra n k one less th a n th e original.
S ubcase 2: S 2 is M. T h e n E ± is e m p ty , an d w e o m it it in w ritin g th e
g iv e n figure thus.
M ix
S in ce S 2 is M , th e n A 2 is n o t M (and hence M 8 0 , since M 8 0 , A 2 b y B 2 . 1).

B eca u se C ase l b is exclu d ed , M £ Q . W e use these fa cts in w r itin g th e
m ixes in th e altered figure thus.
M ix
M ix
F, > 0 m>
§78 g e n t z e n ’s no rm al fo r m t h e o r e m 459
If L is V or 3 th e restriction on va riab les is satisfied b eca u se
I \ E M, ©M, Q , M all occur in th e g iv e n pure va ria b le proof elsew here
th a n a b o v e th e con clusion of th e original L. (B u t L ca n n o t b e 3 -> .)
T h is subcase ca n n o t occur in tu itio n istic a lly , since th e seq uen t T 0, M
o ccurrin g in th e g iv e n proof has m ore th a n one M in its su cced en t. N o w
th e u pper mix in th e altered figure is of rank one low er th a n th e o rigin al
m ix ; so b y th e h yp o th esis of th e in d u ctio n on rank, it can b e elim in ated .
T h e n th e low er m ix w ill h a v e no m ix o ver it. Sin ce M t Q, A 2, th e le ft
ra n k w ill b e o n ly 1 , w hile th e righ t ran k w ill be th e sam e as in th e original
m ix . Sin ce b y h yp o th esis th e left ran k w as o rigin a lly > 2, th e ra n k of
th is m ix is less th a n th a t of th e o r ig in a l; an d hence b y th e h y p o th e sis of
th e in d u ctio n on rank, it too can b e elim inated .
Case 12 a: S x is b y a tw o -p rem ise lo gical rule L, except th e in -

tu itio n istic D T h e inference w h ich giv e s S x m a y b e w ritte n
An, ^ ©> A -12 a 21, r —► ©, A 22 ^

s x, r ©,s2.
T r e a tm e n t is sim ilar to C ase 1 la . B o th prem ises are m ix e d w ith S —> Q
(noting th a t M 8 0 ), prior to th e L in th e altered figure.
Case 13: S x is b y th e in tu itio n istic D S i n c e MS©, and 0 is

a t m ost one form ula, 0 is M. O n ly th e second prem ise B , T -*> M is
m ix e d w ith S O in th e altered figure.
B 2 .2 . Cases 9 b — 12b. S ta te m e n t an d tre a tm e n t s y m m e tric to

C ases 9 a — 12a, e x c e p t as follow s. T h e in tu itio n istic v e rifica tio n s for
C ase 1 1 b are im m ed ia te. (Subcase 2 can arise in tu itio n istica lly.) T h e
tre a tm e n t of th e in tu itio n istic D ca n b e in clu d ed in C ase 12b by
w ritin g 0 ° in th e first p re m ise ; th e v e rifica tio n th a t th e new D —► h as
an e m p ty seq u en t for its 0 ° is o b ta in e d b y o b servin g th a t in tu itio n is
tic a lly th e $ of th e g iv e n m ix is s im p ly M.
T h e fo llo w in g a p p lica tio n of T h eo rem 48 w as g iv e n for th e in tu i

tio n istic sy ste m b y C u rry 19 3 9 .
T heorem I f a form ula E is provable in the in tuition istic (<classical)

49.
system H , it is provable using only the "D-postulates {the D - an d-\-postu lates)
and the postulates for the logical sym bols occurring in the form ula , provided
that in case V occurs but not & the V -postulates include A xiom Schem a
9a of Lem m a 11 § 24.
P roof . U s in g L e m m a 33a, no lo gical rules ca n b e a p p lied in a p ro o f

in G1 without cut, except rules for logical symbols occurring in the
endsequent. The theorem follows by Theorems 46 (using the first part
only), 48 and 47 (both parts), if no variable occurs in E both free and
bound; otherwise, use also Remark 1 (c) §33.
Curry’s book 1 9 5 0 (not available during the writing of the present
§§ 77—80) contains contributions to the theory of Gentzen-type systems,
and a full bibliography. Curry uses the name “elimination theorem” for
the Hauptsatz. Our name “normal form theorem” is intended to suggest
the intent of the theorem, when it is merely explained, without giving
details of Gentzen’s system, that the normal form is one for proofs;
and we have kept “elimination theorem” for cases when a symbol or
notation is being eliminated as in § 74. (But “normal form” has the similar
disadvantage that it commonly refers to one for formulas as in §§29, 76.)
The next two sections, in which two applications of the theorem are
presented, may be read independently of each other.
*§ 79. Consistency proofs. For Gl or G2, we call an application
of one of the structural rules a structural inference; of one of the logical
rules of the propositional calculus, a propositional inference; of one of
the additional logical rules of the predicate calculus, a predicate inference.
Under special assumptions as to the form of the endsequent, a still
further normalization of proofs can be achieved, as in the following
“extended Hauptsatz” given by Gentzen 1 9 3 4 - 5 for his classical
system. (It is closely related to Herbrand’s theorem, 1 9 3 0 .) As Gentzen
remarked, the proof is an illustration of the possibilities for permuting
inferences in his systems. These possibilities are discussed further in
Curry 1 9 5 2 for the classical system of his 1 9 5 0 , and in Kleene 1 9 5 2
for the classical and intuitionistic systems described here. For example:
T h e o r e m 50. Given a sequent containing only prenex formulas and
in which no variable occurs both free and bound, and given a proof of this
sequent in the classical system G\ (the intuitionistic system Gl without
V ~>), another proof in the same system of the same sequent can be found
which contains no cut and in which there is an (occurrence of) a sequent S
(called the midsequent) such that no quantifiers occur in S, and the part of
the proof from S to the endsequent consists solely of predicate and structural
inferences. The new proof is a pure variable proof. Similarly reading “G2 ”,
“mix” for “Gl”, “cut”, respectively.
P r o o f , reducing the theorem to a lemma. We suppose Theorem 48
§79 CONSISTENCY PROOFS 461
a lr e a d y applied, so th a t w e are d ealin g w ith a giv e n pure va ria b le p roof

in G 1 w ith o u t cu t (or G2 w ith o u t m ix). B y the subform ula p ro p erty, o n ly
p ren ex form ulas can occur in seq uen ts of this proof.
Consider a n y axio m in th e g iv e n proof co n ta in in g quantifiers, sa y
th a t V com es first. W e can replace th e ax io m (left) b y th e fo llo w in g
figure (right),
V x A ( x ) - > V x A (x ), A (b) A (b)
VxA(x)-»A(bjV^ v
V x A (x ) V x A (x ),
w here b is a v a ria b le n o t p rev io u sly occurring in th e proof. T h e form u la
A (b ) o f th e new axio m has one less q u an tifier th a n th e form u la V x A ( x )
o f th e original axiom . S im ilarly, if 3 com es first. B y in d u ctio n on th e sum
of th e n um bers of qu an tifiers in th e form ulas of all th e axiom s, w e can
th u s elim in ate axio m s c o n ta in in g quantifiers a lto ge th er from th e g iv e n
p roof (retaining its other m en tio n ed features).
N o w suppose, as w e sta te in th e follow ing lem m a, th a t w e ca n rear
range th e lo gical inferences in th is proof (retaining its other features)
so th a t each p red icate inference follow s all the propositional inferences.
T h e n there w ill be (an occurrence of) a sequent S i b elo w w h ich th e p ro o f
con tain s o n ly p red icate an d stru ctu ral inferences (w ith no b ran ch in gs,
as th e rules considered are all one prem ise rules) an d a b o v e w h ich o n ly
propositional an d stru ctu ral inferences. T h ere m a y b e occurrences o f
form ulas w ith q uan tifiers in S x an d above. H o w ev er none of th e m is as
side form ula for a propositional inference, since th e resu ltin g p rin cip al
form ula w ou ld not be prenex. N o n e of th em is as prin cipal form u la,
since there are no p red icate inferences in this p art of th e proof. N o n e of
th em is in an axiom , as p re v io u sly arranged. H en ce each p o stu la te
a p p lica tio n in th e p art of the proof dow n to S x in clu sive w ill rem ain
correct, if w e alter this p art of th e proof b y suppressing e v e ry occurrence
in it of a form ula co n ta in in g a quantifier, e x c e p t th a t som e of th e
stru ctu ra l inferences m ay th e r e b y becom e id en tica l inferences (w ith
prem ise an d conclusion th e same) an d can be o m itted . T h is alteratio n
w ill replace S x b y a sequent S co n ta in in g no quantifiers, b u t from S w e
ca n pass to S x b y zero or m ore T and I steps, and then ce b y th e u n altered
p a rt of th e proof usin g o n ly p red icate and stru ctu ral inferences to th e
original endsequent. T h u s w e o b ta in th e new proof as described in th e
theorem , w ith S as th e m idsequen t.
L em m a 40. Given a pure variable proof in the classical system G\ (the

intuition istic system G 1 without V —>) without cut of a sequent containing
462 C O N S IS T E N C Y CH . X V
only prenex form ulas , another such proof of the sam e sequent can be found in
which each predicate inference follows every propositional inference. The
sam e sequents occur as axiom s in the new proof as in the given proof. Sirni-
larly reading “G 2” , “m ix ’* for “G l” , “cut” , respectively.
P r o o f o f th e lem m a . C onsider a n y p red icate inference in th e g iv e n
proof, an d co u n t th e n u m b er of propositional inferences occurring in
th e b ra n ch lead in g from its con clusion d ow n to th e en d sequ en t of th e
entire proof. T h e sum o f these num bers, for all th e p red icate inferences
in th e proof, w e call th e order of th e proof. W e estab lish th e lem m a b y
in d u ctio n on th e order.
I f th e order is n o t 0, there is som e p red icate inference such t h a t a
p ro p o sitio n al inference occurs b elo w it w ith no in terven in g lo gical
inference.
Case 1 : th e p red icate inference is an V , an d th e first propositional

inference b elo w it is a one-prem ise inference L. A s n o ted a b o ve , usin g th e
su b fo rm u la p ro p erty , th e V x A (x ) of th e V can n o t b e th e side fo rm u
la o f th e inference L . H en ce th e g iv e n figure is as show n a t th e left. C lassi
c a lly w e alter th is to th e figure sh ow n a t th e right.
The T is to ta k e care of th e p o ssib ility th a t th e TCI steps in th e

g iv e n figure include a C w ith V x A (x ) as th e C. T h e pure v a ria b le
p ro p e r ty of th e g iv e n proof ensures th a t th e restriction on va riab les is
satisfied for th e —> V in th e altered figure. In tu itio n istic a lly ®x, ®, 0 2
an d 0 3 are all e m p ty , an d th en w e o m it th e — T in co n stru ctin g th e
altered figure. T h e a lte ra tio n decreases th e order of th e proof b y 1.
Case 2: —> V follow ed by a tw o-prem ise p roposition al inference

L. C la s sica lly th e tre a tm e n t has o n ly to b e ch an ge d to show an e x tr a
prem ise (either left or right) a b o v e th e L , preceded in th e altered figure
(except w h en L is D - ^ ) by a In tu itio n istic a lly L can o n ly be
w ith th e e x tr a prem ise on th e left.
Cases 3 — 8. The classical and intuitionistic treatment of the cases

for V —> followed b y either a one- or a two-premise propositional
inference is essen tially sy m m etric to th e classical tre a tm e n t for —> V.

T h e four cases for - » 1 an d 3 follow d u ally.
E xam ple 1. T h eo rem 50 does not h old for th e in tu itio n istic sy ste m
G1 w ith V -*► . F o r consider th e follow in g proof.
a { a ) - » c3(a) ^ cH{b) H{b)

c ^3
cH{a) 3xc3{x)
^ a ) V ^ ( b ) ^ 3 x ^ (x ).
S u p p o se there w ere an oth er proof as described in th e theorem . T h e n
th e m idseq u en t S w o u ld b e of th e form II <D w here II consists of zero
or m ore occurrences of CZ(a) V CA(b) an d O either is e m p ty or is <3(t)
for som e term t, since read ing u p w ard from th e end sequent to S, th e
TCI steps (if an y) can o n ly suppress, d u p licate or perm u te form ulas
b u t n o t in troduce new form ulas, an d th e p red icate inferences (if any)
ca n o n ly b e an 3 w h ich (reading upw ard) ch anges 3 *c A(x) to <3(t).
B u t th en if II or O were e m p ty , H —> O could not be p ro v ab le in G\
w ith o u t cu t, b y L em m a s 33a an d 33b used as in E x a m p le 4 § 78. So consider
e.g. th e case II — <E> is > d {a ) V CT(b) —> «C2(t). T h e term t m a y be a or b
or neither. Consider e.g. th e case t is a . Then b y C o ro llary T h eo rem 47,
d (a ) V CZ{b) D a (a ) w o u ld be p ro v ab le in th e propositional calculus H\
an d b y T h eorem 4 § 25, so w o u ld be d VB D <£?, co n tra d ictin g T h eo rem 9
§28.
T h e c o n s is t e n c y t h e o r e m . G en tzen illu stra ted th e use of his

ex te n d e d H a u p ts a tz in estab lish in g co n sisten cy results such as h a d
p re v io u sly been o b ta in ed b y A ck erm an n , v o n N eu m an n an d H erbran d.
B e rn a y s 1936 an d H ilb ert an d B e rn a y s 1939 s ta te d a general co n sisten cy
th eo rem for a x io m a tic theories, w h ich th e y b ased on th e A c k e rm a n n
tre a tm e n t. W e sta te such a theorem now.
T h e theorem can be used to infer m e ta m a th e m a tic a lly th e co n sisten cy
of an a x io m a tic th e o ry from n um ber th eo ry, g e o m e try or algebra,
w h en a m odel for th e th eo ry, i.e. m ore p recisely for its notions an d
axio m s (§ 14), can be estab lish ed c o n stru ctiv e ly , i.e. in fin ita r y term s.
T h e th e o ry is to b e form alized as a form al sy ste m , th e term s an d
form ulas of w h ich are co n stru cted usin g th e lo gica l sym b o lism of th e
p red icate calcu lu s H w ith certain in d iv id u a l sym b o ls ev ..., eQ, fu n ctio n
sym b o ls flt . . . , fr an d p red icate sym b o ls P x, . . . , P s. T h e syste m m a y
h a v e a fin ite or in fin ite n um ber o f axiom s in ad d itio n to th e p o stu lates
of th e p red icate calculus H.
T o establish a co n stru ctive m odel, w e m u st sta rt w ith a d om ain D of
o b je c ts w h ich is either fin ite (and n o n -em p ty) or e n u m era b ly in fin ite.
W e co u ld ta k e D to b e th e n a tu ra l num bers N in e v e r y case, b y selectin g
som e fix e d e ffe c tiv e en u m eration of it w hen it is n o t a lre a d y (w ith rep
etitio n s if it is finite), an d th ereafter d ealin g w ith th e îndices in th is
en u m eration in p la ce of th e original o b jects. B u t in p ractice it is co n
v e n ie n t to d eal w ith oth er d om ains d irectly. T h e case of a fin ite d o m ain
a d m its a sim pler tre a tm e n t from T h eo rem 20 § 36 e x te n d e d to in clud e
in d iv id u a l an d fu n ctio n sy m b o ls in th e v a lu a tio n procedure.
In estab lish in g a m odel, th e n e x t step is to in terpret th e in d iv id u a l,
fu n ctio n an d p red icate sy m b o ls in th e d om ain D , i.e. to chose o b jects
ev ..., eq from D as v a lu e s of ev ..., eQt fu n ctio n s fv . . . , fr w ith
in d ep en d en t va ria b le s ra n gin g o ve r D an d v a lu e s in D as va lu e s of
flt . . . , fr, an d lo gical fu n ctio n s or pred icates Pv ..., P 8 w ith ind epen dent
v a ria b les ran gin g o ver D as v a lu e s of P 1, . . . , P s. F o r th e m odel to be
co n stru ctiv e , fv ..., fr m u st be e ffe c tiv e ly calcu lab le fu n ction s an d
P lf . . . , P s e ffe c tiv e ly d ecid ab le predicates. W e s a y in th is case t h a t
ev • * •» zq>f v • • •, fr> P v •••» P » is an effective interpretation of e2, . . . , eQ,
fv . . . , fr, P 1# . . . , P s in D . (W h en D is N , b y C h u rch 's thesis §§ 60, 62
w e ca n e x p e c t fv . . . , /r, P v . . . , P s to b e gen eral recursive, in w h ic h case
w e h a v e a general recursive interpretation . H o w e v e r th e co n sisten cy
th e o r y does n o t need to b e co n n ected w ith C h u rch 's thesis, since w e
m e re ly need to recognize a t each ap p lica tio n of th e c o n sisten cy theorem
t h a t th e p a rticu la r flf . . . , /r, Pv ..., Ps w h ich w e use are effective.)
G iv e n an e ffe c tiv e in terp retatio n , an d usin g th e 2 -v a lu e d tr u th ta b le s
for D , & , V, - i (§ 28), w e h a v e a v a lu a tio n procedure b y w h ich , g iv e n
a n y form u la A (x lt ..., x n) c o n ta in in g no qu an tifiers an d o n ly th e d istin ct
v a r ia b le s xv . . . , x n, w e can, for each w -tuple x v . . . , x n o f o b je c ts from
D as v a lu e s of x lt . . . , x n, e ffe c tiv e ly d eterm ine th e v a lu e o f A ( x 2, . . . , x n)
to b e t (true) or f (false). (W h en D is N, by # # A , C , D § 45, th e p red icate
A (x v . . . , x n) w h ich th e form u la th u s expresses is p rim itiv e recursive
in fv • • • >fr* Pi* . . . , i V )
F o r N as th e d o m ain D , in stea d of sp eak in g of th e v a lu e o f A ( x x, . . . , x n)
w hen x v . . . , x n ta k e th e n a tu ra l num bers x lt . . . , xn as va lu es, it is
u s u a lly m ore co n ven ien t to sp ea k of th e v a lu e of A ^ , ..., xn) (where
xv ..., xn are th e num erals for th e n a tu ra l num bers xv . . x n, re
sp e c tiv e ly ). W h e n w e do th is, A(x1# ..., xn) is a form u la in th e sym b o lism
e x te n d e d if'n e c e s s a r y to in clu d e 0 as an in d iv id u a l sy m b o l an d ' as a
fu n ctio n sy m b o l, an d th e in terp reta tio n is e x te n d e d to g iv e 0 an d ' their
u su al va lu es. W e ca n suppose th e original sy m b o ls n o t to h a v e in clu d ed
0 or ', unless in th e original in terp retatio n it receives th e u su a l v a lu e .
W e ca n do sim ilarly w hen D is not N. T h ere m u st b e p ro vid ed a p articu lar
va riab le-less term x (the an alo g of a num eral) for each o b je ct x of D;
an d th e sym b o lism an d in terp retatio n are e x ten d ed (if necessary) to
include these terms (consistently w ith th e original interpretation). In
b o th cases, w e shall now s a y sim p ly “ n u m b er” for m em ber x of D ,
an d “ n u m eral” for th e variab le-less term x w h ic h expresses x. The
exten sio n of th e sym b o lism (when necessary) is to assist in d escribin g th e
v a lu a tio n process, an d need n o t a p p ly to th e form al sy ste m under
consideration, e x c e p t w hen w e s a y so.
In th e m odel, i.e. after selectin g th e dom ain an d th e in terp retatio n
of th e n on -logical co n stan ts, th e axio m s m u st b e true. A s th e axio m s
w ill (in general) co n tain variab les, w e need a m ore general sense of ‘tr u e ’
th a n th a t g iv e n b y th e v a lu a tio n procedure. W e shall fo rm u late it here
o n ly for p ren ex form ulas (Theorem 19 § 35). Consider first a closed
p ren ex form ula, e.g. for illu stratio n 3 y 1V x 13 y 2V x 23 y 3A ( y 1, xv y 2, x 2, y 3)
w here A ( y x, xv y 2, x 2, y 3) con tain s no quantifiers an d o n ly th e d istin ct
v a riab les sh ow n; call this form ula “ G ” . U n d er th e in terp retatio n of
ev ..., eqf iv . . ., fr, P x, . . ., P s an d th e usual m eanings of the q u a n ti
fiers, G is true if an d o n ly if there is a num ber yv such th a t for each
n um ber xv there is a n um ber y 2 d ep en d in g on xx (write it “ y 2( * i) ” )>
such th a t for each num ber x 2, there is a num ber y 3 d ep en d in g on x x an d
x2 (write it “ y 3(*i> x 2)” ), suc^
(I) A ( y lf X ,, y . t a ) , Xj, y s{xv x 2))

(where y 2(x x) is th e num eral for th e num ber y 2(x x), etc.) h as th e v a lu e t.
F o r m e ta m a th e m a tica l purposes, m oreover w e w ish th e existen ce to b e
u n d erstoo d c o n stru ctiv e ly , i.e. to m ean th a t th e y v th e y 2{xx) ( f°r a n y
g iv e n % ) an d th e y 3(xv x2) (for a n y giv e n xlf x 2) can b e found. T h is
im plies th a t y 2(xx) an <3 yz(x v xz) are e ffe c tiv e ly calcu lab le function s. W e
s a y th en th a t G is effectively true , if there are a num ber y x an d e ffe c tiv e ly
ca lcu lab le fu n ction s y 2(xx) an d y 3(x1, x 2) such th a t for e v e r y x x an d x 2,
(I) is t. (W hen D is N , b y C h u rch ’s thesis w e can e x p e c t y 2(x i) an d
y z(x i, x 2) to be gen eral recursive, in w h ich case w e sa y th a t G is general
recursively true.) A n open p ren ex form ula shall be effectively {general
recursively) true, if its closure is. T h e nam e verifiable is giv e n to a form u la
w ith o u t quantifiers, w hen for each su b stitu tio n of num erals for its
va riab les th e resultin g form ula ta k e s the va lu e t b y th e v a lu a tio n pro
cedure, i.e. in th e present term in o lo gy, to an e ffe c tiv e ly true form u la
w ith o u t quantifiers.
A c t u a lly our requirem ent th a t y 2{x1) an d y z{xv x 2) be e ffe c tiv e ly
ca lcu la b le o n ly em ph asizes w h a t w o u ld in a n y case b e im p lied in th e
co n str u c tiv e use of th e e x iste n tia l q u a n tifie r; a n d in oth er respects also
(left im p licit) th e h yp o th e sis of th e theorem m u st b e satisfied in a co n
s tr u c tiv e w ay to co n clu de co n sisten cy m e ta m a th e m a tic a lly , e.g. th e
n u m b er yx an d th e d escrip tion of th e fu n ctio n s y 2(x1) a n d y 8(#i, x 2) n iu st
b e g iv e n e ffe c tiv e ly , an d th e d em o n stratio n th a t (I) is t for all x 1 a n d x2
m u st b e b y fin ita r y reasonin g (cf. th e rem arks on th e con verse of C h u rch 's
thesis § 62).
A p ren ex fo rm u la in w h ich no V -q u a n tifie r follow s an 3 -q u a n tifie r w e
sh all ca ll an V 3 -prenex form ula.
T heorem 5 1 . L et the term s and form ulas be constructed using the
logical sym bolism of the predicate calculus H w ith ev . . . , eq, flt . . . , fr,
P v . . . , P 8 as non-logical constants. F or any given dom ain D and effective
evaluation of these constants in D :
(a) I f T are effectively true closed prenex form ulas, E is a closed
V3-prenex form ula, and T b E in the predicate calculus H , then E is
effectively true.
(b) F or any form al system S, the postulates of which are those of the
predicate calculus H and axiom s each of which is {or is equivalent in H to)
an effectively true prenex form u la : If E is an V3-prenex form ula, and
b E in S, then E is effectively true. In p a r tic u la r : S is sim p ly consistent.
P roof of (b) from (a). {b E in S} {T b E in H, w here T is
som e fin ite list of n o n -lo gical ax io m s of S } - > {T 1 b E x in H, w here
are th e closures of th e e ffe c tiv e ly true p ren ex eq u iv a le n ts of T, w h ich
are likew ise e ffe c tiv e ly true, an d E j is th e closure of E } -> (E x is ef
fe c tiv e ly true} (b y (a) w ith Tv E x as its V, E ) -> { E is e ffe c tiv e ly true}.
In p articu lar, 1 = 0 (in th e case of th e n u m b er-th eo retic sy m b o lism w ith
th e u su al in te rp re ta tio n ), or A & - i A where A con tain s no q u a n tifiers, is
n o t e ffe c tiv e ly true, an d hence is u n p ro vab le in S . So S is sim p ly con
sisten t, b y th e second form o f th e d efin ition in § 28.
P roof of (a). I f it is in th e in tu itio n istic H th a t F b E is giv e n ,

th e n a fortiori F b E i n th e classical H, since th e in tu itio n istic p o stu la te
81 is p ro v a b le cla ssica lly (§ 23).
Sin ce T , E are closed, all v a riab les are h eld co n sta n t in th e d ed u ctio n
T b E , an d no v a ria b le occurs b o th free an d b o u n d in th e seq uent
F —► E . So b y T h eorem s 46 an d 50 there is in th e classical p red icate
ca lcu lu s G1 a proof o f F —> E w ith th e features described in T h eorem 50.
Let h b e th e level a t w h ich th e m idsequen t occu rs.
Continuing the illustration, say that T is the single formula G above
and E is Vv1Vv23wB(v1, v2, w) where B(v1, v2, w) contains no quantifiers
and only the distinct variables shown.
Select any pair vv v2 of numbers.
Now we shall carry out a series of acts, one to each predicate inference
in this Gl proof, beginning with the lowest and working upward. Each
act will consist in substituting a certain numeral (specified below) for
every occurrence throughout the tree of each variable occurring free in
the premise of the inference. When the act is performed on an inference
by — or 3 the inference is of course destroyed. The given proof is
a pure variable proof, so the b of an application of —> V or 3 —> does not
occur below its premise. Hence by Lemma 36, at the stage (call it stage g)
when the acts have been performed for the inferences the premises of
which are at or below level g (< h), each of the original postulate applica
tions the premises of which are above level g will have been transformed
into an application of the same postulate, but in the system G1 with
the symbolism extended (if necessary) to include the numerals.
We now specify the choice of the numerals for the substitutions, at
the same timeVproving by induction on g that, if g < h, then at the g-th
stage all sequents up to level g inclusive have the following property
(call it P ): the antecedents (succedents) contain only formulas of the
forms shown in the left (right) column
3 yiVx1 3y2Vx23y3A(y1> x1( y2, x2, y3) V v^v^w B K , v2, w)
Vx^yaVx^yaA^,, x1( y2, x2, y3) Vv23wB(u1,v2, w)
3 y 2Vx23 y 3A(yi> ti. y2>x2>y3) 3wB(u1)v 2, w)
Vx23 y3A(yi. ti. y 2(0 >x2>y3) B(vv v 2> s)
3 y 3A(yi> t,, y 2(h)> t2, y3)
A ( y i . î, y 2{ti)> t 2, y 3 (î. ^2))
where tlf t 2 and s are variable-less terms, ix and t2 are the numbers
expressed under the given effective interpretation by tx and t2, respective
ly, and y lf y2(î)» y ^ i * ^)> v i an<^ are the numerals for the numbers
v 2
Vi> VM> y$i> vi and v2 respectively, where yv y2(%) and y3(xlf x2)
are the number and effectively calculable functions given by the hy
pothesis that G is effectively true.
B a s i s : g = 1 . Only the endsequent is at or below level g, and it
has property P.
I nd. step : g > 1 . For g > h the induction proposition holds vac
uously. For g < h by the hypothesis of the induction, as the tree figure
now stands, the sequents below level g have property P. We distinguish
cases acco rd in g to th e k in d of inference w h ic h ap p lies from le v e l g to
g —1.
le v e l
Case 1: —> V . In our illu stra tio n , th e p rin cip al form u la m u st h a v e

one of tw o form s, e.g. (using th e second) \/v 23 w B (v lf v 2, w ). T h e side
fo rm u la is th e n l w B ( i ; 1, b , w) w here b is th e b o f th e origin al V. W e
su b s titu te v 2 for b w h erever it occurs, i.e. o n ly in th e side fo rm u la o f
th e prem ise (at le v e l g) an d in seq u en ts a b o v e le v e l g. B y th is a c t, p ro p e r ty
P is estab lish ed a t le v e l g, since th e sid e form u la b ecom es 1 w B ( i;1, v 2, w)
w h ich is one of th e allo w ed su cced en t form s for p ro p e r ty P , a n d all th e
oth er form ulas in th e seq u en t are d u p lic a te s of re sp e c tiv e form u las in
th e conclusion, w h ic h b y h y p . ind. w ere a lre a d y of allo w ed form s.
Case 2: V - * . In our illu stra tio n , th e p rin cip al form u la m u st be

of one o f tw o form s, e.g. (ta k in g th e second) V x g B y g A ^ , tv y 2(ti), x 2, y 3) ;
an d th e side fo rm u la is th e n of th e form 3 y 3A (yv tlf y2(^), t * , y 3) w here
t * is a term resu ltin g b y p rev io u s su b stitu tio n s from th e t of th e origin al
inference. B y th e h y p . in d ., no va ria b le occurs free in seq u en ts b elo w
le v e l g, an d so o n ly th e v a ria b le s of t * o ccur free in th e prem ise. W e
su b stitu te 0 for each of th ese (if an y), if th e d o m a in D is N, an d other
w ise som e sp ecified num eral, w h erever t h e y occur, i.e. o n ly in th e side
fo rm u la (at le v e l g) an d above le v e l g. The sid e form u la b ecom es
i y g A ^ , t x, y 2{h)> t 2>y 3) w here t 2 is th e result o f th is s u b stitu tio n on
t * , an d th e oth er form ulas a t le v e l g are u n c h a n ged a n d w ere a lr e a d y of
allo w ed form s for p ro p e r ty P .
Case 3 : —* 3 . S im ilar to C ase 2 (w ith one p o ssib ility for th e form

o f th e p rin cip al fo rm u la in our illu stratio n ).
Case 4: 3 - > . I n our illu stra tio n th e p rin cip al fo rm u la m a y h a v e

one of three form s, e.g. 3 y 2V x 23 y 3A ( y 1, tlf y 2, x 2, y 3) ; an d th e side
fo rm u la is th e n o f th e form y /x ^ y 3A (y lf t 1# b, x 2, y 3) w ith th e origin al b.
W e su b s titu te th e n u m eral y ^ ) for b.
Case 5 : T , C or I. T h e prem ise a lre a d y h as p ro p e r ty P , b ecau se

it co n ta in s no fo rm u la n o t occurring in th e co n clu sio n ; an d no a c t is
perform ed to ch an ge th e situ atio n .
T h is c o m p le tes th e sp ecificatio n of th e acts, an d th e proof t h a t a t

sta ge g ( < A) seq u en ts u p to le v e l g in clu sive h a v e p ro p e rty P . So a t
sta g e A, w h en th e w h ole alte ra tio n of th e g iv e n proof in th e classical
p red icate ca lcu lu s G\ h as been co m p leted , th e o rigin al m id seq u en t,
since it co n ta in ed no q u a n tifie rs, has becom e of th e form
a (xl >^n, y ^ v iv i t 12, y S u > ^12))»• • • * A(yi, t ?1, y%{tn)> t Z2»y-&{tn> ^2))
-> 6(1?!, 1;,, Si), . . . , B(|?!, i?2) s j .
T h e tree figure dow n to th e m id seq u en t has becom e a proof of this

seq u en t in th e propositional ca lcu lu s G\ w ith num erals ad d ed to the
sym b o lism . So b y C o ro llary T h eo rem 47,
A (yi> ^n> y ^ id * î2> ^ 3(^11^^12)) & * • • & A ( y 1? t zl, y 2(^11)> t Z2>y ^ n * tft))
D B(|>!, i?2, sx) V . . . V B(|?!, i?2, s ml
is p ro v a b le in th e p ro p o sitio n al ca lcu lu s H, an d hence b y T h eo rem 9

§ 28 (and T h eo rem 4 § 25) is id e n tic a lly true w hen its d istin ct prim e
p a rts are tre a te d as d istin ct proposition letters for th e v a lu a tio n procedure
of th e p roposition al calculus. In p articu lar, it ta k e s th e v a lu e t w h en
w e assign to th e d istin ct prim e p arts (w hich co n ta in no variables) th e
v a lu e s t or f w h ich th e y ta k e und er th e g iv e n e ffe c tiv e in terp reta tio n of
th e n o n -lo gica l con stan ts. B y h yp o th esis, (I) ta k e s th e v a lu e t for e v e r y
p air % , %2 of num bers. H en ce b y th e v a lu a tio n ta b les for & , D an d V,
one of B (v v v 2> sx), . . . , B (i;1, v 2, s m) m u st b e t. L e t th e first w h ich
is b e B (v lt v 2, s 0). T h e va riab le-less term s a th en expresses a n u m b er s a
such th a t B (v{, v 2, s a) is t.
T h e w hole process b y w h ich , after ch oosin g th e n um bers vlf v2, w e
o b ta in th is n um ber s a is e ffe c tiv e ; so s a = w (vv v2), w here w (vlt v2) is
an e ffe c tiv e ly ca lcu lab le fu n ctio n . T h e n for e v e r y vlf v2,
( ii) v 2, w(vv v2))

is t, i.e. E is e ffe c tiv e ly true, as w as to b e shown.
R emark 1. T h e a b o v e co n stru ctio n en ables us to s a y so m eth in g a b o u t

h o w th e fu n ctio n w (vlt v2) is re lated to th e fu n ctio n s an d p red icates
fv ..., fri P lf . . P S) y 2(%)> > irresp ectiv e of th e n a tu re of th e
la tte r. T h u s, w ith or w ith o u t th e h yp o th e se s of effectiven ess in th e theorem ,
w e h a v e w h en D is N : (a) w is prim itive recursive in flt ..., fr) P lf ..., P Si
y 2> y 3. F o r w e can sh ow b y in d u ctio n on g th a t, if g < A, a t sta g e g
each term t occu rrin g free in th e tree an d c o n ta in in g p
e x a c tly
d istin ct v a ria b les expresses, as vlt v2 v a r y , a fu n ctio n t(vv v2i u lt . . . , u p)
e x p lic it in flt ..., f rt y 2, y 3 an d co n stan ts. H en ce (for i = 1, . . . , m) s t-
expresses a fu n ctio n $*•(%, v^ e x p lic it in th e sam e; an d each prim e
p a rt o f B (v lf v 2i st) expresses a p red icate e x p lic it in th e sam e an d
P lf . . . , P s, w hen ce usin g # # A , C, D § 45 B ^ , v 2, s,-) expresses a p red ic a te
B i(v i, v2) p rim itiv e recursive in flt . . . , fri P lt . . . , P s, y 2, y 3. T h e n c e (a)
follow s b y a p p ly in g # F w ith as th e an d s{ as th e (no <pm+1 b ein g
required). M ore s p e c ific a lly : (b) w is explicit in fv
.. ., f r, y2, y 3, ~b, •, sg,
the representing functions of P lf P a$ and constants ( # # 1, 2, 9).
...,
A p p l ic a t io n s . T h e o r e m 52. F or the prim itive recursive predicate

R (x, y), form al system S, and form ulas A (x ) (x — 0, 1 , 2 , . . . ) described in
E xam ple 2 § 7 3 : A (x ) is provable in S , only if (E y)R (x, y). (In fa c t : E very
'i'i-prenex form ula provable in S is effectively true under the intended
interpretation.)
P ro o f. T h e axio m s of S are all verifiab le (hence e ffe c tiv e ly true),
w h en th e dom ain is th e n a tu ra l num bers an d th e n o n -lo gical constants
0, ', fv ..., 4, = are in terpreted in th e in ten d ed w a y (w hich is e ffe ctiv e ),
1. e. b y 0, ', <pv ...,cp*, = , re sp e ctiv e ly . So (b) applies. U sin g it w ith
A ( x ) , i.e. 3 y f fc(x, y ) = 0 , as th e E : I f b A (x ) in S , th e n A (x ) is effectively
tru e under th is in terp retatio n , i.e. there is a y for w h ich f k(x, y ) = 0 is t
(i.e. for w h ich <pk{x, y) (E y)R (xt y).
= 0 ) , i.e.
T heorem 53. (a) Robinson's system {Lemma 18b § 49), call it S , is

sim p ly consistent {and every 'i'i-prenex form ula provable in S is effectively
true), (b) In 5 , for a given prim itive recursive predicate R {x, y), and a
form ula R ( x , y) which num eralwise expresses it obtained by the method
of proof of Corollary Theorem 27 § 4 9 : b 3 y R ( x , y) only if (E y)R (x t y).
(c) In S , if the form ulas A {a, h) and B {a, c) of E xam ple 1 § 61 are obtained
by the method of proof of Corollary Theorem 27 , then {conversely to (52)):
{ b B (x )} -> {E y)W 0{x, y).
P ro o f . I t w ill suffice to p ro ve th e theorem for the classical systems,

as 1 = 0 or 3 y R ( x , y) or B (x ) is p ro v a b le in th e intuitionistic system only
if p ro v ab le in th e classical.
(a) E a c h of th e th irteen axio m s e x ce p t * 1 3 7 (or *13 6 ) is v e rifia b le

under th e usual (effective) in terpretatio n . E m p lo y in g *90 § 35, we
3 b { a = 0 V a = b ') as
o b ta in an V 3 -p re n ex eq u iv a le n t o f * 1 3 7 . T h is is
b{a) =
e ffe c tiv e ly true, w ith a - M , since th en for each n a tu ral n u m ber a,
a = 0 V a = {b {a ))' is t.
(b) T h e co n sisten cy theorem does n o t a p p ly im m e d ia te ly , becau se
3 y R ( x , y) is n o t an V 3 -p re n ex form ula. H o w ev er w e sh all show (follow ing
th e lem m a) th a t w e ca n ad d new axio m s to S co n ta in in g new p red icate
sym b o ls, an d e x te n d th e in terp retatio n e ffe c tiv e ly to th e la tte r, in such
a w a y th a t each of th e new axio m s is e q u iva le n t in th e p red icate calcu lu s
to an e ffe c tiv e ly true p ren ex form ula, an d b S (x, y) ^ R (x , y) in th e
resu ltin g sy ste m S ' w here S(x, y) con tains no qu an tifiers a n d expresses
R {x ,y ) under th e in terpretatio n . T h e n (b) w ill follow th u s :
§79 C O N S IS T E N C Y PROOFS 471
{I- 3 y R ( x , y ) in S} -> { b 3 y R ( x , y) in S'}->{b 3 y S ( x , y) in S '}
E y)R {x , y).-> (■
Lemma 41°. Let A ( x , y) be a form ula containing no quantifiers and

only the distinct variables shown, suppose given an effective interpretation
of its non-logical constants, and let A [x, y) be the predicate which it expresses
under this interpretation. Let A (x) be a new predicate letter, (a) S u ppose
that
(1) (Ey)A (*. y) -* (Ey)v<»(x)A(x>y)
where \j(x) is an effectively calculable function. Let
(2) A ( x ) ^ ( E y ) A ( x ,y ) .
Let A (x) be interpreted by A(x) [which by ( 1) and (2) is effective). Then the
form ula
(i) A (x) ~ 3 y A ( x , y)
is equivalent in the predicate calculus to an effectively true prenex form ula.

S im ila rly w ith “x v . . . , x n, y v . . . , y m” in place of “x, y ” . (b) S im ila rly
w ith universal instead of existential quantifiers [and “ in place of
in (1)).
P roof of L emma 41. (a) U n a b b r e v ia tin g ~ in (i), a n d u sin g *96,
*89, *9 7 an d * 9 1 , w e o b ta in as an V 3 -p re n e x eq u iv a le n t
(ii) V y 3 z [{ A (x ) D A ( x , z)} & { A (x , y) D A (x)}].
T h is fo rm u la is e ffe c tiv e ly tru e w ith z(x, y) = yLZz<^lx)A (x, z).

(b) S im ila rly u sin g *95, *89, *98 an d * 91 , w e o b ta in
V y 3 z [{ A (x ) D A ( x , y )} & {A (x , z) D A (x)}],
w h ich is e ffe c tiv e ly tru e w ith z(x, y) = [lzz < uM A ( x , z).

P roof of T heorem 53 (b) (concluded). B y th e p roof of C o ro lla ry
T h eo rem 27 § 49 from T h eo rem 27, it w ill suffice n o w to show , for th e
p roof of T h eo rem 27, t h a t S ca n b e e x te n d e d so t h a t h P ^ , . . . , x B, w) ~
P ( x 1( . . . , x „ , w) w ith P ( x 1, . . . , x „ , w) in terpreted b y cp(xlt ..., x n) = w .
Case (Vb). B y h y p . in d ., S is a lre a d y e x te n d e d to in clud e a p re d ic a te
sy m b o l Q (x 2, . . . , x „ , w) in terp reted by ^{xt , ..., x n) = w such th a t
h Q ( x 2, . . . , x B, w) ~ Q ( x 2, ..., x b, w ), a n d , R (y , z, x 2, . . . , x B, w) in
terp re ted b y x(y, z ,x t , x n) = w such t h a t h R (y , z, x 2, . . . . x B, w) ~
R (y, z, x 2, . . . , x „ , w). We in tro d u ce (if w e have not already) a<b,
to be in terpreted b y a <b, w ith th e a x io m a < b ~ a < b , i.e. a < tb ~
3 c ( c '+ a = £ ) , u sin g c < 6 as th e b o u n d y <, u(x1( x,) for L e m m a 4 1. W e
fu rth e r e x te n d S (if w e h a v e n o t already) to in clu d e a p red icate sy m b o l
B (c, d, i, w), to b e in terpreted b y [3(c, d, i) = w , such th a t \- B (c, d, i, w)
/-^B(c, d, i, w ) ; referring to th e d efinition of B (c, d, i, w) a c c o m p a n y in g
*(180) § 4 1, th is ca n b e done b y several steps (left to th e reader) sim ilar
to those w e perform n ex t. N o w we in trod uce D (c, d, i, x 2, . . . , x n), in
terp reted b y (Eu)(Ev)[$(c, d, i') = u & p(c, d, i ) = v & %(i, v, x 2, . . ., x n) = u ] ,
w ith th e ax io m D (c, d, i, x 2, . . . , x w) ~ 3 u 3 v [B (c , d, i', u) & B(c, d, i, v) &
R (i, v, x 2, .. x n, u)], usin g th e bou n d s u < (3(c, d } i'), v < p(c, d, i).
N e x t w e introduce E (c, d, y , x 2, . . . , x n), in terpreted b y
-> (Eu)(Ev)[${c, d } i') = u & p(c, d, i) = v & v, x 2, . . x n) = u } ] ,
w ith th e axio m E (c, d, y , x 2, . . x n) ~ V i [ i < y D D (c, d, i, x 2, . . x n)],
usin g th e b o u n d i < y. T h e n w e introd uce F to elim in ate th e first 3u
u < (3(c, d } 0); and fin a lly P to elim in ate th e
of P (y , x 2> . . . , x n, w), usin g
3c3d, usin g as bou n ds c < C an d d < D w ith C an d D as described in
E x a m p le 1 (A) § 57 w hen yj(y, x 2, . . . , x n) = cp(y, x2, . . . , x n). B y T h eorem
I Case (Vb) (C) § 49, th e p red icate P(y, x2, . . . , x n, w) assigned to in terpret
P (y , x 2, . . x n) w) is indeed cp(y, x 2, . . x n) = w ; an d b y th e eq u iva len c es
in trod u ced as axio m s or a lre a d y established,
h P (y , X2, . . . , x n , w) ^ P (y , x 2, . . . , x n, w).
P roof of T heorem 53 (c). Sim ilarly, w ith a q u an tifier elim in ation

of V c follow ing those for T h eorem 27.
T h eo rem 53 (a) giv e s th e co n sisten cy p ro p e rty required for th e first

proof of T h eorem 54 in § 76 (and T h eorem s 53 (b) an d 52 for th e proofs
in R e m a rk 2 § 76).
T heorem The form al number-theoretic system of Chapter I V ,

55.
under the restriction on the induction schema that in the A (x) the x should
not occur free within the scope of a quantifier , is sim ply consistent. (A cker-
m an n 1924-5, v o n N eu m a n n 1 9 2 7 ; cf. H ilb ert an d B e rn a y s 1939 pp. 121,
122 an d 127.)
M oreover: E very V l-pren ex form ula provable in it is effectively true ,
and this rem ains the case upon adjoining to it Robinson's axiom s {Lemma
18b) and the axiom s used in the proof of P arts (b) and (c) of Theorem 53,
which parts apply to this system also.
P roof , reducin g th e theorem to a lem m a. I t suffices to pro ve th e
theorem for th e classical system . W e shall first trea t th e case th e A (x)
is restricted to co n ta in no q u an tifier a t all. T h e n L e m m a 42 w ill co m p lete
th e proof.
Consider an axio m b y P o stu la te 13 in this system , sa y its A (x) is
A (x, xv . . . , x n) w ith e x a c tly th e d istin ct va riab les sh ow n an d no
quantifiers. Let A (x , xv . . . , x n) be the (primitive recursive) predicate
expressed by A(x, x„ .... x„) under the usual interpretation. Using
*89 and *98 , the axiom is equivalent to
3 y [A (0 , x1( . . . , x„) & (A (y, x1; . . . . x„) D A ( y ', x v . . . . x n))
D A(x, x1( ..., xn)].
This formula is effectively true, with
J if •••>*«) V ^ (?* *1* ••■.*«).
’ ’
X WysJ- (y, x v . . . , *„) & A (/, * „ . . . , x n) otherwise.
v ( x X X ) = ° A (X >
yi lf • " n>
L emma 42 °. The class of the provable form ulas (or of those deducible
from given assum ption form ulas) in the system of Theorem 55 is not d i
m inished by further restricting the A(x) for the induction schema to contain
no quantifier at all .
P roof . A n y form ula A (x ) n o t co n ta in in g x free w ith in th e scope
o f a q u a n tifie r m u st b e com posed b y operatio n s of th e p rop osition al
ca lcu lu s from form ulas Av . . . , A m , each of w h ich co n tain s no q u a n tifiers
b u t m a y co n ta in x , an d form ulas A mi+1, . . . , A mrl_m2, each of w h ic h
co n ta in s q u an tifiers b u t does not co n ta in x free (mv m 2 ^ 0; m =
m 1+ m 2 ;> l).J B y T h eo rem 11 § 2 9 on p rin cipal d isju n c tiv e norm al form
(w ith T h eorem 3 § 25), A (x ) is eq u iv a le n t to a d isju n ctio n o f form ulas
An & ... & A im (i = 1, . . . , n) w here each A iy is eith er A ?. or A ;.
d ep en d in g on i , if th e first case o f th e norm al form applies. F o r th e
m o m en t suppose m v m 2 ^ 1. W r ite “ B t-(x)” for An & . . . & A imi,
“ C »•” for A *,mi+1 & . . . & A im, an d ' ^ ( x ) ' ' for som e re fu tab le form u la
w ith o u t quantifiers. T h e n u sin g *48 an d *3 4 § 27, A (x) is e q u iv a le n t to
(a) B (x ) V (B x(x) & C ,) V . . . V (B „(x) & C n).
N o w consider a n y form ula of th e form (a) w ith n ^ 0 w here (as ab o ve)

th e B ’s co n ta in no qu an tifiers an d th e C s co n ta in q u an tifiers b u t no
free x ’s; call n its degree. W e p ro ve b y in d u ctio n on th e degree t h a t a n y
a x io m b y P o s tu la te 13 w ith A (x ) of th is form is p ro v a b le u sin g P o s tu la te
13 o n ly w ith A(x)*s co n ta in in g no quantifiers.
Ind. step: n > 0. W rite (a) as
(b) D (x) V (B „(x) & C„).
T h e ax io m under con sideration is then
[D(0) V ( B n(0) & Cn)] & V x [D (x ) V (B n(x) & C n) 3 D (x ') V (B n(x') & C n)]
(c) 3 D (x) V ( B n(x) & C n).
W e sh all sh ow th a t (c) is d ed u cib le in th e p red icate calcu lu s from th e

tw o fo llo w in g axiom s,
(d) [ B n(0) V D(0)] & V x [ B n(x) V D (x) 3 B »(x') V D (x')] 3 B ,( x ) V D (x ),
(e) D(0) & V x [D (x ) 3 D (x')] 3 D ( x ) ;
i.e. w e are to sh ow : (d), (e) (c). B y T h eorem 14 § 33 w ith *4 5 an d

*34 § 2 7 : C n b (c) ~ (d ); an d w ith *4 7 an d *4 8 : ~ iC n |- (c) ^ (e). N o
va ria b le s are v a ried since C n does not con tain free the v a ria b le x of th e
q u a n tifie r Vx w ith in w h ich replacem en ts are perform ed. So, usin g
V -e lim .: (d), (e), C „ V i C n |- (c); an d hence usin g * 5 1 : (d), (e) b (c).
B u t in (e) th e A (x ) is of th e form (a) w ith degree n — \, an d likew ise in
(d), w hen B n(x) V B (x ) is ta k e n as a new B (x).
W hen m 1 = 0, th e m e th o d applies con stru in g “ B t-(x)” an d “ B*(x) & ” as
th e e m p ty expression, w ith th e fo llo w in g differences. Consider
(f) Cn & V x[C n D C J D C n.
U sin g *4 6 : C n h (c) ~ (f); an d u sin g *4 8 : - i C n h (c) ~ (e). B u t u sin g

*1 § 26, * 7 5 § 35, *4 5, an d *1 a g a in : (-■ (f).
W hen m2 — 0, th e A (x ) a lre a d y lack s quantifiers.
I f th e second case of th e norm al form applies, th en u sin g *5 3 , *3 4 an d
*5 3 , A (x ) is e q u iv a le n t to B (x) & - i B ( x ) .
R emark 2. In th e sy ste m s of T h eorem s 52, 53 an d 55, e v e r y p ro v a b le

V 3 -p re n ex form ula is prim itive recursively true. For R (x, y) an d R ( x , y)
as in T h eo rem 53 (b), { h 3 y R ( x , y )} {there is a p rim itiv e recursive
fu n ctio n cp such th a t (x)R(x, cp(x))}. P roofs . By R e m a rk 1 , a n d our
co n stru ctio n s of th e fu n ctio n s an d p red icates for th e e ffe c tiv e in te r
p re ta tio n an d e ffe ctiv e tr u th in T h eorem s 52, 53 an d 55.
E xample 2. L e t S b e th e n u m b er-th eo retic form al sy ste m w ith

o m itted . W h a t th e a d a p ta tio n of Presburger 1930 cite d a t th e b e gin n in g
o f § 42 g iv e s d ire c tly is an e ffe ctiv e correlation, to each closed form u la A
of S, o f an oth er B , w ith properties ( 1 ) — (4) as follow s. (To consider an
open form ula, let A be its closure.) ( 1 ) f- A ^ B in S . (2 ) B is either
true or false, under an o b vio u s exten sio n of th e v a lu a tio n procedure
(preceding T h eo rem 51). N o w s a y A is true {false), if B is. (This special
d efin itio n o f tr u th for closed form ulas of S is e q u iv a le n t to th e one ca lled
for und er th e gen eral d efin ition o f tru th w h ich w ill b e ta k e n u p a t th e en d of
§ 8 1, b y (3).) (3) T r u th an d fa ls ity in th is sense o b e y th e 2 -v a lu e d tr u th
tables.; { 3 x A (x ) is true} = {E x){ A(x) is tru e}; an d { V x A (x ) is true} ==
(x){A (x) is true}. (4) A c c o rd in g as B is true or false, B is p ro v a b le or
re fu ta b le in 5 . T h u s : (5) S is sim p ly co m p lete, an d co m p le te w ith respect
to th e in terp retatio n ( §§41, 29). (6 ) T h ere is a decision procedure for th e
q u estio n w h e th er A is true. (7) I f S is sim p ly co n sisten t, th ere is a d e
cision procedure for S , i.e. for th e question w h eth er b A in 5 . N o w w e
sh o w : (8 ) I f b A in S , then A is tru e; so S is sim p ly consistent. Su p p ose
bA in S . T h e n for som e finite collection T of n o n -lo gical axio m s o f S ,
r b A in H. W e can use L e m m a 41 to elim in ate th e q u an tifiers from
A an d from th e in d u ctio n form ula A (x ) of each in d u ctio n axio m am o n g T,
ta k in g e.g. for a p art 3 y A ( x , y)
\j(x) = |xy[({3yA (x, y) is false} & y = 0 ) V A (x, y)].

W e th u s o b ta in T ', 0 b A' where 0 are th e form ulas (i) of L e m m a 41
w ith 3 or V for th e p red icate sy m b o ls in tro d u ced in th e elim ination.
T h e n A ' (and hence A) is true, b y T h eo rem 55 (or th e fir s t p a rt o f its
proof) for th e sy ste m w ith 0 ad jo in ed as axiom s.
D is c u s s io n . In these a p p licatio n s of th e co n sisten cy theorem , th e

d o m a in D is th e n a tu ra l num bers, th e num erals are a lre a d y a p art of
th e sym b o lism , an d th e e ffe c tiv e in terp retatio n is th e in ten d ed one.
H ilb e rt an d B e rn a y s 1939 p p . 38 — 48 g iv e tw o a p p licatio n s of th eir
co n sisten cy theorem to g e o m etric a l ax io m system s. T h ere certain enu m er
a b le sy ste m s of co m p le x num bers are used as th e dom ains D.
T h ese co n sisten cy proofs all d ep en d on h a v in g a m odel for th e axio m s,
as did those g iv e n before th e a d v e n t of H ilb e rt's proof th e o r y (cf. § 14).
B u t g iv in g a m odel for th e axio m s in in tu itiv e arith m e tic a l term s does
n o t estab lish b e y o n d all d o u b t th a t no co n trad ictio n ca n arise in th e
th e o r y d ed u ced from th e axiom s, unless it can also b e d em o n strated
t h a t th e reasonings in th e th e o ry can be tran slated in to in tu itiv e a r ith m e t
ica l reasonings in term s of th e o b jects used in th e m odel. T h is d em o n
stra tio n is w h a t th e present co n sisten cy theorem (Theorem 51) a d d s to
th e earlier treatm en t. (Cf. B e rn a y s 1936 pp. 1 1 5 — 1 1 6 an d H ilb e rt-
B e r n a y s 1939 p. 48.)
W e d id ta k e th e d ed u ctio n of th e th e o ry from its axio m s in to ac co u n t
in th e discussion of co n sisten cy in th e first p a rt of § 75 , b u t there w e w ere
u sin g n o n -fin ita r y set-th e o retic m ethods.
T h e co n sisten cy theorem depends, v ia th e e x te n d e d H a u p ts a tz (T h eo
rem 50), on th e H a u p ts a tz (Theorem 48) an d th e redu ctio n of H to G\
(Theorem 46). B y these, g iv e n a proof of a form al theorem in a s y s te m
b a se d on th e p red icate calcu lu s w ith m a th e m a tic a l axio m s, it is possible
to alter th e sy ste m an d proof u n til w e h a v e a proof w h ic h co n ta in s no
form u las m ore c o m p lic a te d th a n th e ax io m s an d theorem th em selves,
i.e. o n ly su bform u las of them . In th is situ atio n there is no excu rsion
th ro u g h “ id e a l” s ta te m e n ts in p ro v in g a “ real” theorem from “ real”
a x io m s (§ 14).
Our metamathematical consistency result of Theorem 55 for a part
of number theory of course holds good upon extending the system
by further effectively true prenex axioms, e.g. one could add symbols
for further primitive recursive functions with their recursion equations.
These were eliminable in the full system (Example 9 §74), but presum
ably are not in general in the system with the restricted induction
schema.
One may also consider some new induction postulates, which would
hold as derived rules in the full system, but presumably do not when
Postulate 13 is restricted as for Theorem 55, and which are susceptible
of like treatment under corresponding restrictions. Examples of such are
given in Hilbert and Bernays 1 9 3 4 pp. 343 — 346, which however appear
from Skolem 1 9 3 9 and Peter 1 9 4 0 to be derivable when suitable primitive
recursive functions with their recursion equations are adjoined.
All such consistency results obtained by strictly elementary methods
must stop short of giving the consistency of the number-theoretic
formalism with the unrestricted induction schema, as we know from the
famous second Godel 1 9 3 1 theorem (Theorem 30 § 42), according to which
the consistency of that system cannot be established by methods
formalizable in the system itself.
G e n t z e n ’s We shall now
c o n sist e n cy proof for num ber t h e o r y .
give a brief heuristic account of the method used by Gentzen (1 9 3 6 ,
1938) in a proof of the consistency of classical pure number theory with
the unrestricted induction postulate.
In the proof of Gentzen's 1 9 3 4 - 5 Hauptsatz (Theorem 48 § 78),
we used a triple induction, consisting of an induction on the number of
mixes, within the induction step of which (in establishing the principal
lemma) we used an induction on the grade, within the basis and induction
step of which we used an induction on the rank. This triple induction can
be regarded as a single “transfinite induction”, if our system of ordinal
numbers, hitherto consisting simply of the natural numbers, is extended
sufficiently into the transfinite.
Beyond the natural numbers or “finite ordinals” a next number or
“first transfinite ordinal” called co is supplied. Then new numbers
g> + 1, c*>+2, ... are obtained by use of the successor operation, after the
infinity of which still another called 200 shall follow. Repeating this
process, after all these infinite sequences of numbers, each isomorphic to
the natural number sequence, and starting respectively with 0 , ca, 2 co,
still another number called co2 shall follow; and so on, thus.
0, 1 ,2 , . . .; w, co-j~L o )+ 2 , . . . ; 2co, 2co-f-l, 2co-f-2, .. .; . . . ; ;
co2, co2+ l , co2+ 2 , . . . ; co2-fco, co2+ c o + 1, co2-fc o -f 2, . . . ;
co2+2co, co2+ 2 c o -f-1, co2+ 2 c o + 2 , . . . ; . . . ; ;
2co2, 2co2+ l , 2co2+ 2 , . .. ; 2co2+co, 2co2+ c o + l, 2co2+ c o + 2 , . . .;
2co2+2co, 2co2-f-2 c o -H , 2co2-j-2co-f-2, • •• • • •^
T h is figure is in ten d ed o n ly to suggest th e m anner of gen eratio n an d
th e n o tatio n s, u p to a certain p cin t. T h e gen eral th e o r y of tran sfin ite
ordinals form s a p art of C a n to r’s a b stra c t set th e o r y (cf. 18 9 7 ). A set
lin early ordered (§ 8 ) so th a t each n o n -e m p ty subset has a first elem ent is
called well-ordered. T w o ordered sets M an d N are sim ilar (M ~ N ), if
th e y can b e p u t in to a 1-1 correspondence p reservin g th e order. C a n to r ’s
ordin al num bers arise b y a b stra c tin g from w ell-ordered sets w ith respect
to sim ila rity, in th e sam e w a y as his cardinals arise b y a b stra c tin g from
sets w ith respect to eq u ivalen ce (§ 3). H o w ever, w hile th e w hole sy ste m of
C a n to r ’s tran sfin ite ordinals requires a set-th e o retic approach , th e th e o ry
of in itia l segm en ts (at least of n o t too gre at ones) can be h a n d le d in a
fin ita r y w a y .
F o r ex am p le, the sy ste m of th e ordinals < co3 can be represented as th e
triples of n a tu ral num bers, in a certain ordering. L e t a = [a, b, c) =
ao)2+ 6o> +c be a n y such trip le ; th e last n o ta tio n is th e one cu s to m a ry
in th e th e o ry of ordin al num bers < co3. T h e order relation b etw ee n tw o
such triples (as ordinals < to3) is defined th u s:
04 < a 2 = (ci1< a 2) V (a1= a 2 & b1< b 2) V (ax= a 2 & b1= b 2 & cx< c2).
In other w ords, th e ordering of th e triples (a, b, c) is a lp h a b e tica l, w ith
an in fin ite a lp h a b e t con sistin g of th e n a tu ral num bers.
N o w th e trip le o rd in ary in d u ctio n on num ber of m ixes a, grade b an d
c can be
ra n k considered as a single tran sfin ite in d u ctio n w ith (a , b, c)
— ao>2+bu>-\-c as th e in d u ctio n num ber, thus. In a tran sfin ite in d u c tio n
u p to co3, to p rove th a t all ordinals < co3 h a v e a p ro p erty , one show s
th a t, for a n y ordinal a < co3, if all ordinals (3 < a h a v e th e p ro p erty ,
th en a has the p ro p e rty also. W e are usin g th e m ore co m p a ct sta te m e n t,
in w h ich basis an d in d u ctio n step are co m b in ed (cf. * 1 6 2 b § 40 for th is
form of th e sta te m e n t of ord in ary induction ). T h e case a = 0, for w h ich
th e set of the p’s is e m p ty , can be treated se p a ra te ly as the basis, if one
wishes. T h e in d u ctio n is of th e cou rse-of-valu es ty p e , w here now w h e n eve r
a > co, there are an in fin itu d e of preceding p ’s. F o r the G en tz en H a u p ts a tz ,
th e reasoning is th a t, if the theorem is true for all proofs w ith in d u ctio n
n um ber (a, b, c) = p < oc ( < co3), it is true for a proof w ith (a, b, c) = a.
The induction number a = (a , bt c) now well-orders the cases of the
theorem in an order in which they are being proved, just as the natural
number used as induction number does in an ordinary induction.
Conversely, using the definition of ordinals < c*>3 as triples (a, b, c),
every transfinite induction up to c*>3 can be accomplished by ordinary
inductions.
One can define in a finitary manner somewhat higher ordinals. Clearly
we can go up to cow for any finite n. After all these ordinals as next we
take eventually co^; and so on.
Gentzen’s discovery is that the Godel obstacle to proving the con
sistency of number theory can be overcome by using transfinite in
duction up to a sufficiently great ordinal. His transfinite induction is
up to the ordinal called s0 by Cantor, which is the first ordinal greater
than all the ordinals in the infinite sequence co, ___It figures
in Cantor's theory as the least of the solutions (called e-numbers) for \ of
the equation = £.
Gentzen works with systems of his type with sequents, but they have
H ilb e r t-ty p e e q u iva len ts. In th e 1938 version of his c o n sisten cy proof,
the simple consistency of the system is identified with the unprovability
of the sequent For from any sequent can be deduced by thinnings,
while conversely from ->A & -iA and the provable sequent A&~iA—*
one can infer by a cut. He begins by correlating to each proof in
his system an ordinal number < e0. Taking the induction in the form of
an infinite descent (cf. *163 § 40), he shows that, given any proof of the
sequent another proof of with a lesser ordinal can be found. So
his system (and hence ours) is simply consistent.
In the 19 3 6 * version, his sequents each have exactly one succedent
formula. He shows, by transfinite induction on the ordinal (<e0) °f a
proof, that a certain kind of reduction can be carried out on any provable
sequent. It is absurd that this reduction should be performable on
-*• 1 = 0 . The performability of this reduction is offered as a finitary
meaning which can be attributed to the ideal statements of classical
number theory (§ 14).
Ackermann 1940 uses transfinite induction up to e0 to carry through
a proof of the consistency of elementary number theory in another
manner using Hilbert’s e-symbol (originally proposed by Hilbert, and
carried by Ackermann in 1924-5 as far as showing the consistency
with the restricted induction schema).
Just as transfinite induction up to 003 can be reduced to ordinary in
duction, so can induction up to e0, as is done formally by Hilbert
§80 DECISION PROCEDURE, UNPROVABILITY 479
a n d B e rn a y s 1939 pp. 360 ff. B u t there is th e difference th a t in redu cin g
t h e la tte r a p red icate is used as th e in d u ctio n p red icate (the A (x) of (I)
§ 7 5 , or th e P(n) of § 7) in to w h ich there enters e x p lic itly a p red icate Q
defin ed b y an in d u ctio n sim ilar in n atu re to th a t d efin in g M in T h eo rem
V I I I § 57 an d hence p ro b a b ly not arith m e tic a l (§ 48) or e q u iv a le n tly n ot
“ e le m e n ta ry ” (Theorem V I I § 57). I t can in fa c t be a n tic ip a te d from the
secon d G odel theorem (Theorem 30) th a t tran sfin ite in d u ctio n up
to £q can n o t be reduced to ord in ary in d u ction w ith in th e sy ste m , since
th e reasonings used in G e n tz e n 's co n sisten cy proof oth er th a n this tran s
fin ite in d u ctio n are of sorts th a t are form alizable in th e sy ste m (w hence
in p a rticu la r no form ula of th e sy ste m can sa tisfy th e eq u ivalen ces
d efin in g Q). In his last paper 19 4 3 , G en tzen proves th e n o n -re d u c ib ility
of in d u ctio n u p to e0 d irectly, in stead of in d ire ctly from G o d e l’s theorem
w ith his co n sisten cy proof.
T h e original proposals of th e form alists to m ake classical m a th e m a tics
secure b y a co n sisten cy proof (§§ 14, 15) did not c o n tem p la te th a t such
a m e th o d as tran sfin ite in d u ctio n u p to s 0 w o u ld h a v e to be used. T o
w h a t e x te n t th e G en tzen proof can be a ccep te d as securing classical
n u m b er th e o ry in th e sense of th a t problem form u latio n is in th e present
sta te of affairs a m a tte r for in d iv id u a l ju d g em en t, d epen d in g on h ow re a d y
one is to a c cep t in du ction u p to e0 as a fin ita ry m eth od . (Cf. end § 81.)
G e n tz e n in 1938a speculates th a t th e use of tran sfin ite in d u ctio n up
to som e ordinal greater th a n s0 m a y enable th e co n sisten cy of an alysis
to be pro ved . B y a result of S c h iitte 1 9 5 1 , stronger form s of in d u ctio n
are o b ta in a b le th u s; in fa c t for a n y ordinal a, in d u ctio n u p to th e least
C a n to r s-num ber greater th a n a can n o t be reduced to in d u ctio n u p to a
(but in d u ctio n u p to a n y in term ed ia te ordinal can be).
§ 80. Decision procedure, intuitionistic unprovability. G iv e n

th e conclusion B of an inference b y th e m odus ponens rule (R ule 2) o f
th e form al sy ste m H, w e can n o t determ ine the prem ises A an d A 3 B ,
because th e A w ill be unknow n. S im ilarly, g iv e n th e conclusion
A ,T A , 0 of a c u t in G l , an d th e an alysis of th e conclusion sp e cify in g
h o w its an teced en t is sep arated in to the A an d th e F an d its succed ent
in to the A an d th e 0 , w e can n o t determ ine th e prem ises A A , C an d
C, T 0 , because the C w ill be unknow n.
H o w ev er, for each of the rules of th e propositional calculus G l e x c e p t
th e c u t (or of G 2 e x c e p t the m ix), g iv e n th e conclusion of an inference
b y th e rule an d th e an alysis of th e conclusion, th e premise(s) for th e
inference are ascertain able. U sin g th is fa ct w ith G e n tz e n ’s norm al form
theorem (Theorem 48), we shall obtain a decision procedure for the prop
ositional calculus, which unlike the truth-table procedure (§§ 28—30,
Theorem 12) works also for the intuitionistic system as well as for the
classical.
The steps in the procedure will consist in listing the choices of the
premise(s) for the inference of a given conclusion. In doing this, it is
tedious to have to distinguish all the ways of applying the structural
rules TCI, Therefore, for use in our version of Gentzen's decision proce
dure, we shall introduce a new Gentzen-type system G3t in which the
structural alterations TCI are not counted as separate inferences. We
define G3 for the predicate calculus also, although it is only for the
propositional calculus that we shall have a decision procedure.
In order in G3 to dispense with the TCI rules, we must construe the
postulates of G3 to apply irrespective of the order and number of repeti
tions of formulas in the antecedents, and classically in the succedents.
In other words, for G3 any postulate application shall remain an
application of the same postulate when any sequent is replaced by a
sequent ‘cognate* to it in the following sense: Two sequents T -^ 0
and r '- > 0 ' are cognate, if exactly the same formulas occur in T
(in 0 ) as in V (in 0 '), provided intuitionistically that © and 0 ' neither
consist of more than one occurrence of a formula and are hence the same.
E xample 1 . The sequents C, Cl, 2 & d , 2 and 2 & d , d , C-* 2, 2
are cognate classically, but the latter sequent is not employed intu
itionistically.
For the classical system G3, the postulate list differs from the list given
for G2 thus. The axiom schema is replaced by
C, T -* 0 , C.
There are no structural rules of inference; and each logical rule of
inference is modified by retaining the principal formula in the premise(s).
For example, —► -1 , 3 ► V and -i —^ become:
A, T 0, -i A A 3 B, T 0, A and B, A 3 B, T 0
T -* 0 ,-iA. A 3 B ,T -> 0 .
r - * 0 , AVB, A or r- * 0 , A V B,B - iA, T -* 0 ,A
^ 0 , AVB. -i A, r 0 .
(For 3 — both premises are to be used; for V, written now combining
the two rules into one statement, one or the other.)
For the intuitionistic G3 the postulate list is as follows.
P o stu la te s fo r th e in t u it io n is t ic form al s y s t e m G 3
A xiom schem a.
C, T C.
R u les of inference for the propositional calculus.
A, r-*B A d B ,T - > A and B , A 3 B , T 0

r-> A 3 B . a db , r~>0.
r -> A and T B A, A & B, r 0 or B , A & B , T -*• ©
r - * A & B . A & B, r - > 0 .
T - > A or r-»B A , A V B, T 0 and B , A V B , T - ^ 0

r->AVB. A V B , T - » .0 .
a , r-> -iA, r-*A

r-> -iA . -iA , T -*-© ,
w ith 0 e m p ty or
consisting of one form ula.
A d d ition al rules of inference for the predicate calculus.
r -» A(b) A (t), V x A (x ), r - * 0
r -*► V xA (x), V x A (x ), r - i © 7
subject to the
r-»A(t) A (b ),3 x A (x ), r - ^ e
r -> ix A (x ). 3 xA (x), T - * 0 ,
subject to the
W e also define classical and intuitionistic system s G3a. These differ

from the system s G3 in th a t we perm it a rb itra ry om issions of form ulas
in the antecedent and succedent of the prem ise (s) for an inference b y
a n y one of the rules.
T h e system G3 is designed to m inim ize the num ber of choices of prem
ise^) for a given conclusion, when we are a ttem p tin g to exh au st the
possibilities for proving a given endsequent, especially in show ing the
endsequent to be unprovable. W hen the endsequent is p rovable, th e
use of G3a usually perm its shortening the sequents used in the proof.
A proof in G3 is ir r e d u n d a n t , if it co n tain s no pair of c o g n a te seq uen ts
one occurring a b o v e th e o th er in th e sam e branch.
T heorem 56. (a) I f b T —> 0 in G 3a {a fo r tio r i if b T 0 in G3),

th e n |- T 0 in G 2 (o r G l) , u s in g e x a c tly th e sa m e -n a m ed lo g ic a l r u le s
a s in th e g iv e n p ro o f in G 3a a n d n o m ix {or cu t).
(b) C o n v e r s e ly , if b T —> 0 in G 2 {or G l) , a n d n o v a r ia b le o ccu rs b o th
fr e e a n d b o u n d in 0, th e n b T Q in G 3 {a fo r tio r i in G3a), u s in g
o n ly r u le s th e sa m e -n a m ed a s lo g ic a l r u le s u sed in th e g iv e n p r o o f in G 2
{or G 1). L e m m a s 32a— 33 b { sta te d a b o ve fo r Gl a n d th e cu t, a n d fo r G 2 a n d
th e m ix ) h o ld a ls o fo r G3 {a n d G3a).
(c) A fo r m u la E c o n ta in in g n o v a r ia b le b o th fr e e a n d b o u n d is p r o v a b le
in H , if a n d o n ly if th e re is a n ir r e d u n d a n t p r o o f in G 3 o f th e seq u en t —> E .
(d) A d e c is io n p ro ced u re {or a lg o r ith m ) fo r d e te r m in in g w h e th e r or n o t a
p r o p o s itio n le tte r fo r m u la E is p r o v a b le in th e p r o p o s itio n a l c a lc u lu s H is
a ffo r d e d by th e p ro cess o f a tte m p tin g to co n str u c t a n ir r e d u n d a n t p ro o f o f
E in G3. A c c o r d in g a s su ch a p r o o f is fo u n d or is d e te r m in e d n o t to e x is t ,
E is p r o v a b le in H or is n ot p r o v a b le in H .
P roofs , (a) A n axio m of G 3a is p ro v ab le by T I steps from an axio m

of G2. G iv e n a n y inference in G3a, b y T C I steps w e can b rin g its prem ises
to th e stan d ard form show n in th e p o stu la te list for G3. T h e n th e cor
respon din g rule of G2 applies, w ith th e g iv e n con clusion or a con clusion
w h ich leads to it b y T C I steps.
E xam ple 2. T h e inference in th e in tu itio n istic G3a sh ow n a t th e

le ft is th en acco m plish ed in G2 as show n a t th e right.
3 ,3 - * GT _ 3,3 -+ C 2
2, -1<3 _Tc7 _hla a
—itv?, —\^ i, “B —> 2
B ,-I Z I - + C .
(b) B y T h eorem 48 (and L e m m a 34), w e can ta k e th e g iv e n proof
to b e in G 2 w ith o u t m ix. A n ax io m of G2 is an axio m of G 3 ; an d w e e a s ily
v e r ify th a t a n y inference in G2 w ith o u t m ix can b e perform ed (b y one
or more* steps) in th e sy ste m o b ta in ed from G3 b y a d d in g th e six T C I
rules. H en ce it w ill suffice to show th a t these ad d itio n s to G3 do not

increase its class of p ro v a b le sequents. F o r th is purpose, w e first show b y
in d u ctio n th a t, if b T 0 in G3, th en b T - > 0 , C in G3, p ro v id e d
in tu itio n istic a lly th a t 0 is e m p ty , i.e. w e establish th a t T holds as a
§80 D E C IS IO N PROCEDURE, U N P R O V A B IL IT Y 483
d e riv e d rule for G3. In th is in d u ctio n , th e restriction on v a riab les for an
—> V or 3 -*► is m et b y first u sin g L e m m a 35 (w hich h olds for G 3 as w ell
as for G1 an d G2) to ch an ge th e b in its free occurrences in th e g iv e n
proof o f th e o ld prem ise to a new v a ria b le n o t occurring free in C. T h e
rule r - > is h an d le d s im ila r ly ; an d C, C— /, I im m e d ia te ly
b y th e co n v e n tio n u n d er w h ich th e rules o f G3 are used.
(c) B y (a), (b) a n d T h eo rem s 46 an d 47, E is p ro v a b le in Ht if an d

o n ly if there is a proof o f E in G3. B u t g iv e n a n y proof of E in
G3, w e ca n fin d an irred u n d an t one, as w e sh ow b y in d u ctio n on th e
n u m b er of pairs of c o g n a te seq uents one a b o v e th e o th er in th e sam e
b ran ch . G iv e n such a pair, th e lower seq u en t of th e pair an d th e in
te r v e n in g sequents, to g e th e r w ith all bran ch es c o n trib u tin g to either,
can b e suppressed.
P ro o f of (d) is p o stp o n ed u n til a fter th e fo llo w in g e x a m p le illu stra tin g

th e d ecision procedure.
E xample 3. (a) Is < 3V nprovable in th e in tu itio n istic p ro p

osition al ca lcu lu s HI B y (c), it is if an d o n ly if there is an irred u n d an t
proof of - » <^? V - i e3f in th e in tu itio n istic G3. W e a tte m p t to find such a
p roof thus. T h e seq uent —* V -i< 3 , as w e see b y inspection, is n o t an
a x io m of G3. T h e o n ly rule of inference of G3 ap p licab le w ith —► c3 V
as con clusion is V. In tu itio n is tic a lly th e inference o f —
b y th is rule ca n h a v e as prem ise o n ly <3? or N e ith e r is an axio m ,
th e first —► <*2f can be th e conclusion of no inference in G3, w hile th e second
—► cC? can o n ly com e b y —> - i from th e prem ise or a prem ise
such as <£?, co g n a te to H o w e v e r since tw o co gn ate seq u en ts
are in terch a n gib le for proofs in G3, it suffices to consider c77 —>. T h is
seq u en t is n o t an axio m , an d no inference of G3 is possible w ith it as
con clusion. T h e entire co n stru ctio n is show n b elo w in th e figure, on three
lines or levels num bered u p w ard from th e g iv e n endsequent
B rie fly , w e a p p ly th e rules of G3 u p w ard from conclusion to prem ises,
in all possible w a y s not d istin gu ish in g b etw ee n co g n a te sequents.
3. e T -*
2. —> or —> —i c7
1 c~j y -T c Y
In th is con stru ction , w e h a v e ex h au sted all th e possibilities for fin d in g

a proof of —> V -ic^f in th e in tu ition istic Go, wi t hout finding one.
H en ce < £ ? V -ic 3 is n o t p ro v ab le in the in tu itio n istic //.
(b) We a lre a d y know th a t <C7 V -1 «2f is p ro v a b le in th e classical
p ro p o sitio n a l ca lcu lu s H (*51 § 27). H o w e v e r it is of in terest to see
h o w th e decision procedure lead s us to a proof of 1 in th e
cla ssica l s y ste m G3, w h en ce if w e follow ed o u t th e proofs of T h eo rem 56
(a), T h eo rem 47, an d o f results in C h a p te r V on which th e la tte r d epen d s,
w e sh o u ld b e led to one of V - i <£7 in th e classical H. A t L in e 2, th e
fo rm u la J7 V - i J ? is n o w retain ed in th e succeed ent (ow ing to th e d if
ference b e tw e e n th e classical an d in tu tio n istic V rule o f G3). T h is g iv e s
us more possib ilities a t L in e 3. T h e (t2 ” written there signifies th a t th e
seq u en t liste d b e lo w it a t L in e 2 is one o f th e possible prem ises a t L in e 3
(b y —* V). A s w e are seekin g an irredu n dan t proof, w e need consider
fu rth er o n ly th e n ew prem ise <^7 V - i cSf, <3f, - i <^7 or <C7 cC? V - i <^7,
- i <£7. T h e co n stru ctio n is sh ow n b elo w u p to L in e 4.
4. 3 o ro ^^ e^ V -1 ^ ^-le7 3orc^-*e7V-ie^,eT,-ieT 3 or a V-1 cC7, a , -1 a

3. 2 or —► cT V—i<C7, <C7, 2or-»e7 V n c~7, eT, —1<^ or cv?—> d7 V—icT, <C7
2. cT V-1 c~7, cC7 or —► eT V—icC7, —i<v?
1. —^ cC7V—i<v7.
A t L in e 4, three of our series of ch oices h a v e ended a b o v e in an a x io m ; i.e. w e
h a v e d isco vered proofs of <C? V - i <3 in G3. S a y u sin g th e left ch oice a t
L in e 2 (and th e n ew choices a b o v e t h a t ) , w e h a v e in p a rticu la r th e fo llo w in g
p roof in G3 (left). T h e seq u en ts of th is ca n b e sim plified in G 3a (right).
4. ^ ^ a v - i c 3, a ,-.c 2 r
_^ 4. < 3 7-* <37
—y ^
3. c3 V i c C 7 , cC?, - i c 7 3. ^ 3?,-n<37
■ > V
2. —> cv7 V ” i <v7, <v7
_ ^ \/ 2. - * <37 V-i<37,
1. 1. > V
^ < 3 ? V - i J?. -+av-u 37.
In term s of G2 (or G l) th e la tte r b ecom es:
etc.
3. <C7 V - i e 2 ,
------------------------- !---------------* V
2.
-------------------------------------—^ G
1. —> tv7 V *ni <v7.
(c) W e a lre a d y k n o w th a t - i - i (<3f V - i c3) is p ro v ab le in tu itio n istic a lly

in H ( * 5 la). A proof of -v-i(<C7 V -i<£7) in G 3a follow s w h ich is d is
co v e re d b y use of th e decision procedure. In term s of G2, th e co n tractio n
w h ich w e co u ld not perform in (a), becau se of th e in tu itio n istic restriction
to n ot m ore th a n one form u la in th e su cced en t, is possible n ow as it is
perform ed in th e an teced en t.
§80 D E C IS IO N PROCEDURE, U N P R O V A B IL IT Y 485
7. <C7-».<3r
6. “ ^_v
5. <3, -« (<3 V “ I <3) - * "1 ” >
4. “i (e? V «3) -^n<3

3. “1 (<£f V “Ie?) —^ e? V “1 e3f
1. —► —i - i (<£f V - i o3).
P roof of T heorem 56 (d). It is clear t h a t th e procedure, as il

lu str a te d in E x a m p le 3, ca n b e co m p le ted u p to a n y desired level, n o tin g
th e fo llo w in g fa c t. G iv e n a n y seq u en t T —* 0 com p osed o f p ro p o sitio n
le tte r form ulas, to each ch oice from T or from 0 o f a form u la c o n ta in in g
a lo gica l s y m b o l to serve as th e p rin cip al form ula, th ere are e x a c t ly one
or tw o in co gn ate ch oices of prem ise(s) for th e inference of T 0 in G3.
I t rem ains to p ro v e t h a t th e w h o le procedure m u st term in ate. B y th e
su b fo rm u la p ro p e r ty of G3, e v e r y seq u en t in a p roof of E in G3 m u st
b e com p osed o f su bform u las of E . B u t a p roposition le tte r fo rm u la E h as
o n ly a fin ite class of su b fo rm u la s ; an d from th ese there are o n ly a fin ite
n u m b er k o f w a y s of ch oosin g form ulas to o ccu r in a n te c e d e n t a n d to
occur in su cced en t, i.e. a t m o st k in c o gn a te seq u en ts ca n b e w r itte n d o w n
form ed of su bform u las of E . T h erefo re an irred u n d an t p roof of —> E
ca n n o t e x ist h a v in g m ore th a n k le v e ls ; so w e ca n e x h a u st th e possib ilities
for fin d in g such a proof b y co m p le tin g th e procedure u p to (at m ost)
th e &-th level.
E xample 4. Is -i-i {p lx V <C?2) 3 , V - i - i <£?2 p ro v a b le in th e

in tu itio n istic pro p o sitio n al ca lcu lu s? W e sta rt o u t as follow s.
3. - 1-1 (c%1 V <C?2) —> -1 (e?xV e ?2) Or- 1-1 (e ^ V c ^ y —► “H Or —1—1( c ^ V <£?2) —* - | - 1 <£?2
2. - 1-1 (c^fj V <v?2) —^ - 1-1 V - 1-1 <v?2
1. —> —1—1(<*3^ V C?2) 3 - 1-1 V —1—1c3f2.
W ith o u t c a rry in g th is procedure further, w e ca n n o w answ er th e q u estio n

in th e n e g a tiv e . F o r w e ca n q u ic k ly v e r ify b y 2 -v a lu e d tr u th ta b le m eth o d s
t h a t none o f th e seq u en ts a t L in e 3 is p ro v ab le. T h u s th e first o f th e m
lx V
- i - i (<£ <C?2) - * - i (<3^ V <C?2) is o n ly p ro v a b le in G3 an d hence (Theorem
56 (a)) in G 1 , if - i - i (<3^ V <C?2) 3 - i (<C?XV <C?2) is p ro v a b le in H (C orollary
T h eo rem 47). B u t - i - i (< ZAXV <C72) 3 -i V <C?2) is n o t p ro v a b le in H , since
it assum es th e v a lu e f w h en <^?x, <C?2 ta k e th e v a lu e s t, t (Theorem 9 § 28).
T heorem 57. (a) In the intuitionisiic propositional calculus H , for
any form ulas A and B : h A V B , only if b A or h B . (Godel 19 3 2 .)
(b) Each of the numbered results * 1 4 , * 1 5 , *49, * 5 1 , *5 2, *5 5 — *62,
which were established in Chapter V I only for the classical propositional
calculus, does actually fail to hold for the intuitionistic propositional cal
culus {and likewise the converse of each im plication among *4 9 a— *62a).
P roofs . P a r t (a) is im m e d ia te from th e form of V in G3. F o r (b)
w e can sh ow b y th e decision procedure th a t each form ula in q u estio n
is u n p ro va b le w hen A an d B are sim ple proposition letters d an d S,
as w e h a v e a lre a d y done for *5 1 in E x a m p le 3 (a). (For *5 1 th e u n p r o v
a b ility also follow s from P a r t (a) w ith T h eorem 9.) H o w e v e r for th e
others it is m ore e xp ed itio u s to ta k e a d v a n ta g e of d ed u ction s in th e
in tu itio n istic propositional calculus. F o r exam ple, if *4 9 held, b y R e m a rk
1 § 27 so w o u ld * 5 1 , c o n tra d ictin g our result for * 5 1 . T h e others w e
consider in their n um erical order.
*14 . Suppose -i A d B h “ ^B D A d id hold in th e in tu itio n istic

p rop osition al calculus for all form ulas A an d B . T h e n in p articu lar w e
w o u ld have d ' "D ~ \ d |- —1—1 D d, w hence usin g *1 we w o u ld
get f- m < 3 D d, co n tra d ictin g our result for *49.
*56. (Cf. *56a.) If h n ( n A & n B ) D A V B in tu itio n istic a lly for

all A an d B , th en in p articu lar |- -1 (-1 d & -1 d ) D d W d , w hence
by *3 7 an d *38 again b “i“ i d D d . A fortiori by & -e lim ., not
b A V B ~ - i ( - iA & - i B) in tu itio n istic a lly for all A an d B .
*62. (Cf. *62a.) I f b “ i ( A & B ) D - i A V - i B in tu itio n istic a lly for

all A an d B , th en in p articu lar b (d l & “ * <^0 3 ~i d V -v -i d f w hen ce
b y *50, b -id V -\-\d . B y (a) o f th e theorem , th en either - \ d or
b -^ -\d , co n tra d ictin g T h eorem 9.
R emark 1. L ik ew ise th e theorem s an d corollaries of C h a p ter V I

m ark ed w ith 0 as b ein g estab lish ed o n ly classically can (w ith one e x
ception) be inferred b y th e present m eth od s to fail in tu itio n istica lly.
T h u s T h eo rem 8 w o u ld g iv e *4 9 ; its corollary w o u ld g iv e *5 5 from *5 4 ;
T h eo rem 11 w o u ld g iv e th e eq u ivalen ce of -1 —i d to one of d V —1 d ,
d , -\d an d d &-id ] etc. E xception : T h eo rem 12 is o b v io u s ly false

for th e in tu itio n istic p roposition al ca lcu lu s in th e sense th a t th e in
d ica te d procedure does n o t a p p ly . T h a t no other tru th ta b le procedure
w ith fin ite ly m a n y va lu es applies w as show n b y G o d el 19 3 2 . T h ere does
ex ist a decision procedure of an oth er kin d (Theorem 56 (d)).
P redicate calculus . B y Theorem 54 § 76, there is no decision pro
cedure for the predicate calculus. (A t w h at point does the proof of
Theorem 56 (d) fail, when w e attem p t to a p p ly it to the predicate ca l
culus ? Contrast E xam p les 2 and 3 (a) § 78.)
A lth ou gh Theorem 56 (c)
thus does not afford a decision procedure in the case of the p redicate
calculus, nevertheless it is useful in in vestigatin g p ro v a b ility in the pred
icate calculus. G iven a predicate letter form ula E , b y a ttem p tin g to
find an irredundant proof in G3 of E , w e m a y a ctu a lly find one, or
w e m a y discover some feature of the situation w h ich shows th a t there
cannot be any.
T heorem 58. In the intuition istic predicate calculus H:

(a) “ i“ i Vx(c3f(x) V is unprovable. (H eyting 1930 a ; K leen e 1945
w ith Nelson 1947 .)
(b) (i) Vx(<£? V S{x)) 3 V V xS(x) is unprovable,
(ii) i “i{Vx(<v7 V B(x)) 3 V V xS(x)} is provable , but
(iii) V 2(x)) 3 *%(y) V V xS(x)} is unprovable.

(c) In each table of Corollary Theorem 17 § 35, when A (x) is the sim ple
predicate letter Cl{x), the im plication {and hence the equivalence) of a form ula
above a line by {to) one below the line is unprovable, and likew ise the double
negation of that im plication {and hence , b y *25, of that equivalence) when a
double line separates the form ulas. (H eyting 1946 .)
(d) Each of the numbered results *83— *85, *92, *97— *99 {Theorem 17),
which were established in Chapter V I I only for the classical predicate cal
culus, does fail to hold. Of them, *83, *92 and *97 hold, but *84, *85, *98
and *99 fa il to hold , when double negation is applied to the form ula.
P roofs, (a) W e attem p t to construct an irredundant proof of
—► "i-iV x(cH (x) V -i<3f(x)) in th e in tuition istic system G3 as follows.
F o r abb reviation, we let “ B ” stan d for V x(e^?(x)V - \d {x )) a t certain
places. From Line 3 to 4, either n B or B can be the principal form ula
of the inference. If - i B (using -1 - * ) , the prem ise is w h at is alread y ob
tained a t 3 (the original B disappearing as the 0 and a new occurrence
of it appearing as the side, form ula). F or the definition of w h at co nstitutes
a proof of a given sequent in G3,
all variab les not occurring in th e sequent
are on a p ar; hence at Line 4 it suffices to list the prem ise for the V
w ith the particular variab le b v T hen sim ilarly at L in e 8 w e choose
another p articu lar variab le b 2, w hich m ust be distinct from b x to sa tisfy
th e restriction on variables for the V ; etc.
48 8 CONSISTENCY CH. XV
*11. -.B , Z U fo , Z U fo B
10. 7 7or-.B, c
9. 7 or - i B , c ^ ) ^ ( b 2) or -iB , Z ifa ) -+ ~ y ^ (b 2)
8. 7 or -i B, -► cZ {b2) V - i a ( b 2)
*7. - i B , < 0 7 ^ )- * B
6. 3 3 or —iB , —►
5. 3 or —i B — or n B - » - i 31(b^)
4. 3 o r - i B - ^ a ^ V - i ^ , )
*3. -iB -* > B
2. - i B -*•
1. —>• —i—i B .
F ro m th e portion of th e stru ctu re show n, th e gen eral p a tte r n is clear.

W e fail to h a v e an axio m in th e first n ew seq u en t - i B , 31 (,b x) —> 31 (b2)
of L in e 9 b ecau se b x an d b2 are d istin ct va ria b le s; sim ilarly in th e first
n ew seq u en t - t B , 31 (brf, 3 l(b 2) —* 31 (bz) of L in e 13 ; an d so on. A ll other
seq uen ts occurring are ev e n m ore o b v io u s ly n o t axio m s. N o te th e form
of L in es 3, 7, 11, . . . ; a t L in e 3 + 4 ^ for n = 0, 1 , 2 , . . . w e h a v e closed
o u t all th e possibilities for fin d in g an irredu n dan t proof of —* - i - i B
oth er th a n b y fin d in g one of i B , 31{b^)t ..., 31(bn) B w ith d istin ct
va ria b le s b lf ..., b n. T h u s th e search for an irredu n d an t proof of -i-iB
in 6 3 w ill n ever term in ate su ccessfu lly, b u t w ill o n ly lead us u p w ard
th ro u gh an in fin ite regression of in crea sin gly co m p lic a ted sequents.
O th erw ise expressed, w e estab lish th a t there is an irred u n d an t proof
of - i - i B in G 3 , o n ly if there is a shorter one of - i B B , a still shorter
one of - i B , 31 (b^ B , a still shorter one again of - i B , 3 }(b x)t 3 t(b 2) B ,
ad in fin itu m . Sin ce it is absu rd t h a t there sh ould ex ist an in fin ite su c
cession of su cce ssiv e ly shorter proofs sta rtin g w ith a first proof, w e
con clu de t h a t there is no proof of - » - i - i B in th e in tu itio n istic sy s te m
G 3 ; an d hence b y T h eo rem 56 (c), none of - n B in th e in tu itio n istic
p red icate calcu lu s H.
(b) (i) W e a tte m p t to co n stru ct a proof in G3, w ith th e first step
u n iq u e ly d eterm in ed th u s:
2. V*(<07 V »(*)) -► C2 V VxB(x)

1. Vx(<07 V S(x)) V xS(x).
F ro m th e form of th e seq uent a t L in e 2, an d of th e rules of G3, w e see

th a t in a n y seq uen t a b o v e th e b o tto m one o n ly form ulas o f th e fo u r
form s Vjc(c^f V B{x)), 3 1 V B(t) (t a term , i.e. for th e pure p red icate c a l
culus, a va riab le), 31 an d B(t) ca n occur in th e a n te c e d e n t, an d o n ly a
form u la of one of th e four form s 3 1 V yi x <
B {x)t <31, V xB (x) an d B(b) (b a
va riab le) as the succed ent. T h e o n ly ch ances for such a seq u en t to b e an
a x io m are for it to h a v e th e form 311 T -► 31 or th e form B(t), T - * B(b)
w h ere t is b. W h e n e ve r th e tw o -p rem ise rule V —^ is ap p lied , th e tree w e
are co n stru ctin g w ill bran ch . T h e o th er rules w h ich there is a p o ssib ility
of a p p ly in g after th e first step are th e one-prem ise rules V V an d
—► V . In order to h a v e fou n d a proof, for som e succession of choices,
e v e r y b ran ch m u st b e term in ated a b o v e in an axiom . W e sh all n o w
sh o w th a t, no m a tte r w h a t succession of steps h as been perform ed, an
a x io m w ill n o t h a v e been reach ed alo n g one o f th e branches. F o r th is
purpose, w e define th e designated branch (along w h ich w e are to sh ow
t h a t an ax io m ca n n o t b e reached) b y sp e cify in g w h ich prem ise b elo n gs
to it a t each V thus. L e t th e p rin cipal form ula for th e V be 3 1 V B (t).
31 as w h ole or p art, th e prem ise
I f th e form ula of th e su cced en t co n tain s
in th e d esign ated b ran ch (or designated prem ise) sh all b e th e one w ith
B(t) as th e side fo rm u la; otherw ise th e one w ith 31 as th e side form ula.
N o w consider th e follow in g property P of a seq uen t, n a m e ly th a t (1) <£?
does n o t occur as one of th e a n teced en t form ulas, if th e su cced en t fo rm u la
co n ta in s 31 as w hole or p art, an d (2) for e v e r y v a ria b le b , B(b) does n o t
o ccu r as one o f th e an teced en t form ulas, if B(b) (for th e sam e b) is th e
su cced en t form ula. N e ith er k in d of a x io m d escribed a b o v e has p ro p e r ty
P. H en ce, to p ro v e th a t an ax io m ca n n o t b e reach ed alo n g th e d esign ate d
b ran ch , it suffices to show t h a t e v e r y seq uent in th e d esign ate d b ra n ch
h as p ro p e r ty P. T h is w e do b y in d u ction . F o r, first, th e seq u en ts a t L in es
1 an d 2 h a v e p ro p e rty P . T h ere afte r, as w e sh all v e r ify n e x t, each in
ference w ill preserve p ro p e rty P alo n g th e d esign ate d b ran ch , i.e. if th e
con clusion of th e inference h as p ro p e rty P , so does th e prem ise, or in
th e case of an V - > th e d esign ate d prem ise. T o v e r ify th is, w e m u st
ex a m in e four cases, acco rd in g to th e form of th e p rin cip al fo rm u la o f th e
inference. Case 1: V x(31 V B(x)) in th e an teced en t. P r o p e r ty P is o b v io u s ly
p reserved, since th e side form u la in tro d u ced in to th e an te c e d e n t b y th e
V ^ is o f th e form 3 1 V B(t) an d th e su cced en t is u n ch an ged . Case 2:
3 1 V B(t) in th e a n teced en t. I f th e su cced en t co n ta in s 31 as w hole or p a rt,
th e in tro d u ctio n of B(t) as side fo rm u la for th e d esign ate d prem ise o f th e
V —► preserves p ro p e rty P . I f th e su cced en t does n o t co n ta in 31 as w h o le
or p a rt, th e in tro d u ctio n of 31 as side form ula for th e d esign ate d prem ise
preserves p ro p e rty P . Case 3: 3 1 V yi x eB {x) as th e su cced en t. Sin ce th e
490 CONSISTENCY CH.
con clusion has p ro p e rty P , th e form ula does n o t occur am o n g th e
an te c e d e n t form ulas, b y (1). Then th e in tro d u ctio n of either or
V xB (r) in to th e su cced en t as th e side form ula of th e —► V preserves
p ro p e r ty P . Case 4: VxB(x) as th e succedent. B y th e restriction on
va ria b le s for th e V , th e va ria b le b of th e side form ula B(b) m u st b e a
v a ria b le n o t occurring in th e a n teced en t, w h ich assures th a t (2) for prop
e r ty P rem ains satisfied.
(ii) N ow consider in stea d - i - i {V jc(<£? V B(x)) 3 V 'ixB {x)}\ call
it “ - v - i C ” . W e m a y n o w h a v e in an teced en ts of seq uen ts a b o v e th e
b o tto m line also th e form ula - i C , as succed en t also C. T h e in d u ctio n for
p ro p e r ty P fails in th e case for - i C in th e an teced en t, as th e - i can
in tro d u ce C as th e su cced en t after has p re v io u sly been in tro d u ced
in to th e an teced en t, lead in g to a vio la tio n of (1). B y fo llo w in g o u t th is
looph ole in th e p reviou s d em o n stratio n of u n p r o v a b ility , w e are led
to th e fo llo w in g proof of - > - n { V ^ ( d f V S ( x ) ) D £ 1 W x B(x )}, w h ich
w e s ta te in G3a. T h e d esign ate d bran ch , w h ich e x c e p t b y u sin g i C
as th e p rin cip al form ula for an - i w e w o u ld b e u n a b le to term in ate in
a n ax io m , is th e left one.
(iii) I f w e n o w ch an ge th e form ula to

—i —1V j y { V B(#)) 3 <C?(y) V yix B {x )}i th e loophole is closed. For
th e <3 o b ta in e d in th e a n tec ed en t before th e - i becom es c3(c) for som e
v a r ia b le c. T h e n th e <C? o b ta in e d in th e su cced en t after - i —* fo llo w ed n o w
by V w ith resp ect to y w ill b ecom e <5f(d) w here d is a v a ria b le d istin ct
from c b y th e restriction on v a ria b les for th e —*■ V . T h e reader m a y w o rk
o u t th e m o d ificatio n s in th e d em o n stratio n g iv e n in (i) to estab lish th e
u n p r o v a b ility rigorously.
(c) F o r a g iv e n ta b le, consider form ulas A , B , C, D, w here B is A
o r b elo w A , C is im m e d ia te ly b elo w B an d sep arated from it b y a line,
and D is C or b elo w C. U sin g * 2 , if f- A 3 B a n d |- C 3 D, b u t n o t
h COB, th en n o t |- D o A . S im ila rly, usin g *2 4 an d *4 9 a, if b A 3 B
a n d |- C 3 D, b u t not b —i- 1 (C 3 B ), th en n o t f- - i- i (D 3 A) an d
n o t b D 3 A . H en ce it suffices to tre a t th e u p w ard im p licatio n s b e tw e e n
t h e six pairs of form ulas im m e d ia te ly sep arated b y a line, under d o u b le
n e g a tio n w h en th e line is double.
Ib 3 l a , l i b 3 I la . I f either w ere p ro vab le, b y su b stitu tin g Zl for
Z l(x) (Theorem 15 § 34) an d u sin g * 7 5 or *7 6 , or b y u sin g T h eo rem 22
§ 3 7 for k = 1 , —i—i 3 Zl w o u ld b e p ro vab le. H en ce n o t b Ib 3 l a
and not b H b 3 H a.
- 1- 1 (10 ! 3 Ib). L e t “ B ( x ) ” a b b re v ia te Zl{x) V -i Z l(x). B y *51a an d
V -in tro d ., b V x - i - i B ( x ) . H en ce V x - i - i B ( x ) 3 - i - i V x B ( x ) b t i V x B (x ).
Thence b y 3 -in tr o d . a n d co n trap o sitio n tw ice (*13 , *12 ),
b —i —i { V x —i - i B ( x ) 3 —i —iV x B (x )} 3 —i—iV x B (x ). Thus if
—i - i { V x - i- iB(x) 3 - i - i V x B ( x ) } w ere p ro vab le, - i- i Vx B(x) w o u ld be.
B u t b y (a), n o t b T " i V x B ( x ) ; hence not b - i“ i { V x _i- iB ( x ) 3
- i - i V x B ( x ) } ; an d hence b y th e su b stitu tio n rule (Theorem 15) n o t
b —»—i { V x - i —i<^7(x)3 —i-» V x o ?(x )}, i.e. n o t b “ i“ i (Ict 3 Ib).
—i—i ( I I Ic 3 I I I b 2). E a s ily reduces to - 1- 1 (10! 3 Ib).
H e! 3 lib . B y E x a m p le 3 § 3 7 an d E x a m p le 4 th is section.
I llb ! 3 I lia . E a s ily reduces to I l c j 3 l i b .
(d) *8 3— *8 5, *92. In clu d ed u nder (b) an d (c) (w ith *2 5 , *92a.)

*97. W e show {Z l 3 3 x 2 (x)) 3 3x(<C? 3 2(x)) u n p ro vab le, sim ila rly
to I lc i 3 lib . (A fter a p p ly in g T h eo rem 22 w ith k = 2, w e o b ta in a t
L in e 3: {Z l 3 2 XV 2 2 Z l an d 2 i V 2 2, Z l 3 2 XV S2- > {Z l 3 2 X) V
{Z l 3 2 2)} or Z l 3 2 XV 2 2 - > Z l 3 2 X or Z l 3 2 XV 2 2 - > Z1 3 2 2. I n
tr e a tin g th e first alte rn a tiv e , it suffices of course to sh ow t h a t th e first
prem ise Zl
3 2 X V 2 2 ->• Z l o f th e tw o is un p ro vab le.) W e p ro ve
- i - i {(Z l 3 3 x 2 (x)) 3 3x {Z l 3 2(x))}, sim ilarly to (b) (ii). (In L in e s 4 — 9,
u se su cce ssiv e ly 3, 3, 3 3 —>.) Cf. *9 7a .
*98. S u b s titu tin g Z l & -i Z f for 2 in -i-i{(Vx<C7(x) 3 2 ) 3 3x(<C7(x) 3 2 )}
a n d u sin g *5 0 a n d *44, w e g e t - i - i ( I I I c 3 I l i a ) .
* 99 . R ed u ces to *98 (cf. th e first m eth o d for I b 3 la ).
H e y t in g ( 1930 a p. 65) infers th e u n p r o v a b ility o f th e form ula o f T h e o

rem 58 (a) an d of I c i 3 I b from th e in terp retatio n o f th e in tu itio n istic
p re d ic a te calcu lu s in term s of B ro u w e r’s th e o r y o f sets (end § 13).
The proof of Theorem58 (a) by Kleene andNelson is from 1945 1947
results to be taken upin §82.

The present treatment of (a) and (b) was reported in Kleene ; 1948
and some like applications of Gentzen's theorem are in Curry 1950
(alreadyinpress in 1948). de Iongh uses the methodinestablishing 1948
an intuitionistic classification of thoseformulas formed from< ^ ( x , y , z )

by quantifying x , y , and z and possibly applying negation which are
classically equivalent to V#3jVze^(ar,y , z), analogous toeachof the four
tables of Heyting for one quantifier (Corollary Theorem 17 §35
1946
andTheorem58(c)). Hetakesthissequenceof quantifiers forillustration,

because of its roleinformulatingthe notionof convergence of asequence
to alimit (cf. §35 (i) or (ii), omitting the x s ) .
Mostowski demonstrates the unprovability of -v~i(Ic D lb),

1948 1
and (b) (i), by aninterpretationof theintuitionistic predicate calculus in

terms of "complete Brouwerian” lattices. Henkin a extends M os- 1950
towski’s results to obtain an algebraic characterization of quantifiers

both for intuitionistic and for classical logic.
§ 81. Reductions of classical to intuitionistic system s. For the

rest of this chapter, we use the Hilbert-type systems . When is a H T
sequence of zeroormore formulas, nT, t -i T, r°, etc. shall be the result

of applying- , - - , (as definedbelow), etc., respectively, to eachof the
1 1 1 0
formulas of T .
The main result of the first part of this section is given in several
versions, thoughits significancecanbeseenfromone. Thereaderdesiring
a simplified treatment may accordingly select: Theorem59 and Proofs,
DefinitionandDiscussionof °, Theorem60 (a) for onlyand(c), Lemma 0
43a and Proof, Proof of Theorem60 (c), Corollary 2 (omitting the other
material upto that point).
T h e o r e m 59. (al) bE I f T ,
in th e c la s s ic a l p r o p o s itio n a l c a lc u lu s
th e n—r — 1—1 i—E

\- 1 in th e . (a )
in tu itio n is tic p r o p o s itio n a l c a lc u lu s 2 I f
-1T, A b "iE in , “iT,nnA b “iE

th e c la s s ic a l p r o p o s itio n a l c a lc u lu s th e n
in th e in tu itio n is tic (Glivenko .)

p r o p o s itio n a l c a lc u lu s . 1929
(b) L ik e w is e fo r th e 9 , and
p r e d ic a te the
c a lc u lu s w ith R u le o m itte d fo r
fo r m a l n u m b e r -th e o r e tic sy ste m w ith R u le 9 o m itte d .
P r o o f s . ' (al) By induction on the length of the given classical de

duction r b E (i.e. the deduction of Efrom which ‘T b E” asserts V
to exist, cf. § ), usingthefollowingobservations. If Eis anaxiomof the

22
classical propositional calculus by any axiomschema except , then E 8

§81 REDUCTIONS TO INTUITIONISTIC SYSTEMS 493
is also an axiomintuitionistically, and by *49a §27, b t i E in the
intuitionistic system. If Eis anaxiomby the classical AxiomSchema 8,
thenby *5lb, again |- -v-iE intuitionistically. Moreover, corresponding
to Rule 2, -i-iA, -i-i (A3 B ) b “i~«B intuitionistically, using * 2 3 § 26.
(a2) From(al) using*49b.
(b) Because the additional axiom schemata and particular axioms
belong to the intuitionistic as well as to the classical system, we need
only add a treatment of the additional rale of inference 12. Using *23,
Rule 12and*49a: (i) -r-i(A(x) DC) b "viA(x) 3 -t-iC bx 3x - th A(x)
3 -v-iC b ~i~i (3x - i- i A(x) 3 —»—»C). Using *49a, *70 § 32 and
*49a: (ii) bt »(3xA(x) 3 3x t - iA(x)). By*5lb: (iii) b -v-i(-r-iC 3 C).
Combining (ii), (i) and ( i i i ) by *24 , -r-i(A(x) 3 C) bx ■“i “ i(3 x A (x ) 3 C).
Example 1. By (al), each of the numbered results which were
established in Chapter VI only for the classical propositional calculus
(see Theorem 57 (b)) holds intuitionistically under double negation
(appliedin *14 and *15 to both formulas).
E xample 2. That *97 holds intuitionistically under double negation
(which we proved by another method for Theorem58 (d)) nowfollows
from(b) and Theorem49 §78.
Corollary (to (a2)). I f E is a proposition letter form ula containing
no logical sym bols except & and - i , and b Ein the classical propositional
calculus, then b E in the intuition istic propositional calculus. (Godel
I932-3-)
P roof . Consider E as a conjunction of n formulas (n ;> 1) each

of whichis not aconjunction, andis therefore either apropositionletter
orbegins withthe symbol -i. By &-elim., eachof these n components is
provable classically. But nopropositionletter is provable (byTheorem9
§28). Soeachcomponent is anegation, andby Glivenko’s theorem((a2))
is also provable intuitionistically. Hence by &-introd., so is E.
D efinition of °. For the rest of this section, the formulas T, the
formula E, etc., shall be proposition letter formulas, predicate letter
formulas, or number-theoretic formulas, according as we are considering
propositional calculus, predicate calculus, or formal number theory.
By a prim e part of a formula, we mean a (consecutive) part which is
a prime formula, i.e. one containing no logical symbol.
For anyformulaE, we define E° by the followingrecursion. 1. If Pis
a prime formula, P° is P. 2—5. If Aand Bare formulas, (A3 B)° is
A° 3B°, (A&B)° is A° &B°, (AVB)° is -i(-iA° &-iB°), and (- A 1 )0
is - 1A 0. 6 — 7. I f x is a va ria b le, an d A (x ) is a form ula, (V x A (x ))° is

V x A ° ( x ) (where A °(x ) is (A (x))°), an d (3 x A (x ))° is - i V x n A ° ( x ) .
B rie fly , E ° com es from E b y replacin g (or “ tra n sla tin g ” ) each p a rt o f
E of th e form show n b elo w in th e first line b y th e resp ective expression
show n in th e second.
A D B A & B A V B lA V x A (x ) 3xA (x)

“ “ - i ( “i A & - i B ) “ “ -iV x -iA (x )
E xample 3. Let A(x) and B be prime (and B not contain x free).

If E is [V xA (x ) 3 B] 3 3 x [A (x ) 3 B] (cf. *98), then
E° is [V xA (x ) 3 B] 3 n V x - i [A(x) 3 B ].
D is c u s s io n o f °. In th e n e x t theorem it is show n th a t th e classical

sy ste m s can be defined w ith in th e in tu itio n istic. In particu lar, for th e
n u m b er-th eo retic sy ste m , if h E classically, th en \- E° in tu itio n isti-
ca lly . T h e con verse holds o b v io u s ly (since h E ^ E ° classically), as do
th e con verses for T h eo rem 59 an d th e other p arts of T h eo rem 60. So a
form u la E is p ro v a b le in th e classical syste m , if an d o n ly if th e correlated
form u la E ° is p ro v a b le in th e in tu itio n istic system . W e can th in k of E °
as resu ltin g from E b y ch a n g in g th e logical sy m b o ls 3, & , V, - i , V , 3
to 3°, & °, V °, - i ° , V °, 3°, re sp e ctiv e ly , where “ A 3° B ” is an ab b re v ia tio n
for A 3 B , “ A V ° B ” for n ( i A & n B ) , etc. T h e sense in w h ich th e
cla ssica l form ulas are th u s “ tra n sla te d ” in to in tu itio n istic ones ca n b e
em p h asized b y usin g differen t lo gical sym b ols (say 3 C, <£c , Vc , - i c, V c, 3C)
for th e classical sy ste m (upper row in th e a b o v e tran slatio n table).
D e f in it io n o f ', etc. F o r th e propositional an d p red icate calculi,

w e use other correlations. L e t E ' b e o b ta in ed from E like E ° e x c e p t th a t
A 3B is tran slated as - i ( A & - i B ). L e t E * b e o b ta in ed from E b y re
p la cin g each prim e p a rt P b y - v - i P ; an d E t likew ise e x c e p t th a t P is
re placed b y n n P o n ly w here it is alone (i.e. w h en E itself is P), or im
m e d ia te ly w ith in th e scope of an & or an V , or is th e second p art of th e
sco pe of an 3 ; an d E * like E t e x c e p t w ith o u t th e replacem ents in th e
second p art of th e scope of 3.
E' is - r { - i [V xA (x ) & n B ] & - r i V x - i - i [A(x) & i B ] } ,
E°t is [ V x - i - i A ( x ) 3 - i - i B] 3 n V x n [ n n A ( x ) 3 - n B ] ,
E °t is [V x - 1 - 1 A (x) 3 - v - i B] 3 - i V x - i [A(x) 3 n - i B ] , and
E *' is i { - i [ V x - i - i A (x) & - i B ] & - i - i V x - i - i [A (xj & —i B ] }.
F o r th e propositional calculus, there can be no inverse theorem , g iv in g
a sim ilar redu ctio n of th e in tu itio n istic to th e classical sy ste m , in w h ich
th e in tu itio n istic propositional co n n ectives are d efined e x p lic itly from
th e classical. F o r th a t w o u ld g iv e a tru th ta b le decision procedure for the'
in tu itio n istic propositional calculus, co n tr a d ic tin g G o d el 19 3 2 .
T heorem 60.F or any form ula E , in the propositional calculus,

(a)
predicate calculus, or number-theoretic form al system , b E ^ E ° ~ E ' ^
E ° t ^ E ° t ^ E * ' classically (b y *56, *83, *58, *49).
(bl ) F or the propositional calculus, if |- E classically, then b E '

in tu ition istically . (b2 ) For the number-theoretic form al system , if Y b E
classically, then T ' b E ' in tu ition istically. (Godel 1932 - 3 .)
(c) F or the number-theoretic form al system , if Y b E classically, then
r ° b E ° in tu ition istically. (G en tzen 1936 p. 532 an d B ern ays.)
(d) F or the propositional calculus, predicate calculus, or number-theoretic

form al system , if Y b E classically, then r ° t b E ° f (also Y °t b E ° t
and T * ' b E * ') intuitionistically.
P roofs , (b l) U sin g (a), if b E c la ssica lly , th en b E ' classically.
B u t E ' con tain s as operators o n ly & an d —1. T h u s (bl ) follow s from
C o ro lla ry T h eo rem 59. (C onversely, C orollary T h eo rem 59 is im p lied b y
(bi).)
W e sh all p ro ve (c) after th e first lem m a, an d th en infer (b2).
L emma 43 a. For the number-theoretic form al system , if F contains no

logical sym bols except D , & , - 1 , V {in particular, if F is E ° for some form ula
E ), then b ~ i~ iF D F [and hence b —*—*F ^ F) intuition istically. (A fter
G o d el 19 3 2 - 3 .)
P roof of L emma 43a, b y in d u ctio n on th e num ber of (occurrences of)

lo gica l sy m b o ls in F .
B a s is : F is of th e form s = t w here s an d t are term s. B y * 1 5 8 § 40,

b s = t V -is = t, w h en ce by *49c, b m s = t Ds=t.
Ind. step. C ase 1: F is A D B . B y h y p . in d . : (i) b ""i~*B D B . B y

* 6 0 g ,h : (ii) - i - i ( A D B ) b A D n n B . F ro m (ii) an d (i) b y chain
inference (*2 ), -1-1 (A D B) b A D B , an d b y D -in trod.,
b -1-1 (A D B) D (A D B ). C ase 2 : F is A & B. B y h y p . ind.,
b m A 3 A an d b " n B D B . U se * 25 . C ase 3 : F /s i A . B y
*49b. C ase 4: F is V x A ( x ) . U se h y p . ind., *69 an d b l b D I c x from
C o ro llary T h eorem 17.
P roof of T heorem 60 (c). B y in d u ctio n on th e le n g th of th e g iv e n
classical d ed u ctio n F b E , w ith cases as follows.
Case 1 : E is one of th e p a rticu la r axiom s 1 4 — 2 1, or an a x io m b y

a n y sch em a e x c e p t 11. T h e n E ° is E , or is an a x io m b y th e sam e schem a,
or is d ed u cib le in th e classical propositional calcu lu s (using *5 6 w ith
T h eo rem 6 § 26) from an a x io m b y th e sam e schem a* so E ° is p ro v a b le
in th e sy ste m o f th e classical p rop osition al calculus w ith th e oth er axio m s
an d ax io m sch em a ta ad ded . (For ex am p le, if E is an axio m A D (B D A)
b y S ch em a la , th en E ° is A ° 3 (B ° 3 A °), w h ich is an axio m b y th e sam e
schem a. If E is an a x io m A D A V B by S ch em a 5a, then E° is
A ° D n ( - i A ° & - i B ° ) , w h ich is d ed u cib le from A ° D A ° V B ° b y *5 6 an d
T h eorem 6, since th e p a rt to b e replaced does n o t sta n d w ith in th e scope
of a quantifier.) H en ce b y T h eo rem 59 (b), b “ i~"*E° in th e in tu itio n istic
n u m b er-th eo retic s y s te m ; an d hence b y L e m m a 43a, b E ° in th e sam e.
Case 2 : A x io m S ch em a 11. T h e n E is A (t) 3 3 x A (x ), an d E ° is
A °(t) 3 - l V x - i A ° ( x ) , w h ich is p ro v ab le in tu itio n istic a lly b y co n tra
position (*13) from th e a x io m V x ~ iA ° ( x ) 3 n A ° ( t ) .
C ase 3 : R u le 2 . W e m u st show th a t A ° , (A 3 B )° b B ° in tu
itio n istica lly. B u t (A 3 B )° is A 0 3 B °. Case 4: R u le 9. S im ilarly.
C ase 5 : R u le 12 . W ith th e help of *12 an d L e m m a 43a.
P roof of T heorem 60 (b2). B y L e m m a 43a w ith *5 8 f § 2 7 , a n y p a rt

of r ° , E ° of th e form A 3 B is eq u iv a le n t to - i ( A & - i B).
L e m m a 43b. For the propositional or predicate calculus, if F contains

no logical sym bols except 3 , &, - i , V {in particular , if F is E ° for some
form ula E ), then b - i - i F t 3 F t {and hence b ~ v " iF t ^ F t) in tu ition isti
cally.
P roof of L emma 43b. S im ila rly to L e m m a 43a, usin g in th e basis
*4 9 b in stea d of * 1 5 8 an d *49c.
P roof of T heorem 60 (d). F o r th e n u m b er-th eo retic sy ste m an d °t,

from (c) b y L e m m a 43a. F o r th e propositional or p red icate calculus an d °t,
from L e m m a 43b in th e sam e m ann er as (c) from L e m m a 43a. (U sin g
Lem m a 43a or 43b, *58e an d *49b, th e result can be m odified to
r°t hE°t; usin g also *58f, to T *' b E *\)
R emark L T o show th a t T h eorem 59 does not h old for th e p red icate

ca lcu lu s w ith o u t th e exclusion of R u le 9, an d th a t Th eorem 60 (b) does
n o t h old for th e p red icate calculus, consider as an exam p le V x n n « 3 ( x ) 3
-i - iV x c ^ ( x ) . C a ll this form ula “ E ” . T h e n b E classically, but in tu -
itio n istic a lly neith er b E nor b " i “Ê (Corollary T h eorem 17 § 35 an d
T h eo rem 58 (c)), an d m oreover not b E ', as from E ' b y *4 9 b an d *5 8 b
w e can deduce E in tu itio n istica lly. T h e ex am p le D ZT show s th a t
T h eo rem 60 (c) does not h old for th e propositional or p red icate c a lc u lu s ;
an d th e exam p le —i—i b ZT th a t C o ro llary T h eorem 59 an d T h eo rem
60 (bl ) do n ot h old w ith assu m ption form ulas T. A n exam p le to sh ow
th a t T h eorem 59 does n o t hold for th e nu m b er-th eo retic syste m w ith o u t
th e exclu sion of R u le 9 w ill h a v e to w a it u n til th e n e x t section (Theorem
63 (iii)), as w e h a v e y e t no m eth o d of d em o n stratin g an exam p le of a
cla ssica lly p ro v ab le b u t in tu itio n istica lly u n p ro vab le n u m b er-th eo retic
form ula.
C o ro llar y 1 F or the number-theoretic form al system , if T, E

(to (c)).
contain no logical sym bols except D , & , - i , V, and T b E classically, then
T b E intu ition istically . (to (d) for °t). Likew ise for the propositional
or predicate calculus, provided also that F, E contain no letter unnegated
other than as antecedent of an im plication.
C o ro llary 2 (to (b2 ), (c) or (d)). The classical number-theoretic form al
system is sim ply consistent, if the intuitionistic is.
P roof of C o r o llar y 2 . I f 1 = 0 were provable in the classical
system, it would also be in the intuitionistic.
D isc u ssio n . G o d el rem arks, “ T h e theorem [60 (b 2 ), or n ow (c)] . . .
show s th a t th e in tu itio n istic arith m etic an d num ber th e o ry is o n ly
a p p a r e n tly narrow er th a n th e classical; in fa ct [it] includes th e entire
classical [num ber th eo ry], m erely w ith a som ew hat differing in te r
p r e ta tio n / ' H e y tin g adds, “ H ow ever for th e in tu itio n ists this in ter
p retatio n is the essential t h in g / ' (1934 * p. 18.)
de Io n gh says, “ In our sign ificist opinion the m ost im p o rtan t a d v a n ta g e
of in tu itio n istic m a th e m a tics is, th a t it d istinguish es in e v e ry in stan ce
b etw een d ire c tly an d in d ir e c tly p ro ved propositions an d an alyses th e
m a th e m a tic a l con cepts in to sequences of co n cep ts w ith different degree
of in d ir e c tn e ss /’ (1948 , p. 746.)
v a n D a n tz ig 1947 proposes to in v e stig a te h ow m uch further th e d e
v e lo p m e n t of classical m a th e m a tics can be carried w ith in th e in tu
ition istic, in the m anner ju st show n to be possible for all of the usual
e lem en ta ry num ber th eory. F o r this purpose, the classical form ulas E
are tran slated into classical eq u ivalen ts F w hich are stable in tu itio n isti
ca lly , i.e. such th a t b “i ~ i F ^ F (cf. L em m a 43a). va n D a n tz ig su ggests
th a t it m a y be possible to interpret p ra c tic a lly the w hole of classical
m a th em a tics w ith in this sta b le part of the in tu itio n istic system .
For the consistency problem, the present results can be regarded as
showing that the intuitionistic number theory is equally in need of a
metamathematical consistency proof with the classical, or, if oneaccepts
the consistency of the intuitionistic system on the basis of its inter
pretation, as securing the consistency of the classicalsystem.
Someformalistspoint out that themethodsof intuitionistic elementary
number theory go beyond what they consider as finitary (cf. Hilbert-
Bemays 1934 p. 43 and Bernays ). It is said that the intuitionistic
1935
use of negations of complicated formulas, and of implications having in

the antecedent a complicated formula (e.g. a generality formula, or
another implication) involves the general logical notion of what is an
intuitionistic proof. It is by such use of negation and implication that
Brouwer and his followers are enabled to go much further in the de
velopment of a constructivistic mathematics than Brouwer’s forerunner
Kronecker.
The intuitionists do not attempt to give anexact description of their
notion of a proof in general, and they say that inprinciple no such de
scription is possible.
The intuitionists’use of negationandimplicationmust thenbe under
stood as only requiring us to recognize, e.g., that a particular given
proof is intuitionistically acceptable, or (whenthey prove astatement of
the form (A -> that if one should produce an intuitionistically
B ) C )
acceptable deduction of one statement fromanother then on the

B A ,
basis of it one couldby agiven methodsurely construct anintuitionisti

cally acceptable proof of a third C.
Anattempt is madebyBernays todefendthe Gentzentransfinite
1938
induction up to (end §79), as constituting less of an extension of the

t Q
narrower finitary standpoint than the whole body of intuitionistic

methods in number theory.
de Iongh 1948 touches briefly upon current discussions regarding
intuitionism and related trends (particularly , represented by
s ig n ifie s
Mannoury 1909 , 1925, ).

1934
Nowlet us examine the way in which Corollary 2 Theorem60 gives

a consistency proof for classical elementary number theory fromthe
intuitionistic standpoint. The second part of this proof is a tacit or
explicit verification that the intuitionistic formal system for number
theory is correct intuitionistically.
Since the proof of Corollary Theorem60 is entirely elementary, by
2
Godel’s theorem on consistency proofs (Theorem 30 §42) the second

part cannot be.
It is in terestin g to n o te th a t, ju st as in th e case of G e n tz e n ’s
co n sisten cy proof usin g tran sfin ite in d u ctio n u p to s0, th e present co n
siste n c y proof can b e a n a ly z e d as d ep en d in g for its sole n o n -elem en tary
step on th e use of a p red icate defined b y an in d u ctio n w ith qu an tifiers
of b o th sorts en terin g in th e in d u ctio n step, n a m e ly (here) th e tru th
p red icate for n u m b er-th eo retic form ulas. W e shall define this pred icate
n ex t.
U n d er th e usual in terp retatio n of th e sym b o ls 0, *, of the va riab les
as n a tu ral n um ber variables, an d of the operations of b u ild in g term s
frcm th em as corresponding to inform al operations of e x p licit d efinition,
a n y term t ( x 1, . . ., x n) co n ta in in g o n ly th e d istin ct variab les x 1? . . . , x n
expresses a p rim itive recursive fu n ctio n t(x1, . . ., x n), or for n = 0
a n um ber t. U n d er the usual in terp retatio n of — , then e v e ry prim e

form u la P ( x 1, . . ., x n) co n ta in in g o n ly x v . . . , x n expresses a p rim itiv e
recursive p red icate P (x 1} . . ., x n), or for n = 0 a proposition P . For
a n y closed prim e form ula P , th e tru th or fa ls ity of P is determ ined (and
e ffe c tiv e ly decidable) in our th e o ry of p rim itive recursive fu n ction s,
so w e shall not elaborate u pon this p art of th e tru th definition. (Indeed,
t h a t th e o r y w o u ld ca rry us som ew h at fu r th e r ; cf. E x a m p le 4 below .)
V
(A) F ro m this as basis, w e define "true’ as ap p lied to a n y closed

n u m b er-fh eo retic form ula E , b y in d u ctio n on th e num ber of (occurrences
of) lo gical sym b o ls in E . In this d efinition, of course “ i f ” m eans “ if an d
o n ly i f ” , as is com m on in definitions.
1 . A closed prim e form ula P is tru e , if P , i.e. if P is a true proposition

in th e th e o r y of recursive function s.
F o r Clauses 2 — 5, A an d B are a n y closed form ulas.
2. A & B is tru e , if A is tru e an d B is tr u e .
3. A V B is tru e , if A is tr u e or B is tru e .
4. A D B is tru e , if A is tr u e im plies B is tr u e (i.e. w hen A is tru e
o n ly if B is tru e ).
5. - i A is tru e , if A is n o t tr u e .
F o r Clauses 6 an d 7, x is a va riab le, an d A (x ) is a form ula co n ta in in g

o n ly x free. (Then w hen x is a n a tu ral num ber, x is th e corresponding
num eral, § 41 . )
6 . 3 x A (x ) is tru e , if, for som e n a tu ral num ber A (x ) is tru e .
7. V x A (x ) is tru e , if, for e v e ry n atu ral num ber x , A (x ) is tr u e .

(B) A n u m b er-th eo retic fo rm u la A ( y 1, . . . , y m) co n ta in in g free o n ly
th e d is tin c t va ria b le s y v . . . , y m is true , if, for each w -tu p le y v . . . , y m
of n a tu r a l num bers, A (y lf . . . , y m) is true. (W e need n o t stip u la te here
t h a t y lt . . . , y m all occur free in A ( y 1, . . . , y m) or th e order of occurrence,
since if a fo rm u la is tru e for a n y one ch oice o f th e list y lt . . . , y m, it is
true for e v e r y other.)
E xample 4. W h e th e r a form u la E w ith o u t va ria b le s is tru e or false

(i.e. n o t true) ca n b e d e cid e d usin g th e 2 -v a lu e d tr u th ta b le s; an d a n y
fo rm u la A ( x 1# . . . , x n) w ith o u t q u an tifiers an d ju s t xv . . . , x n as va ria b le s
expresses a p rim itiv e recursive p red icate A ( % , . . . , x n) su ch th a t
A (x v ..., x n) S3 { A ( x x, . . . , x n) is true}. (Cf. before T h eo rem 51 § 79.)
B y (A), (C) an d (D) § 4 1 : In the number-theoretic form al system every true
form ula without variables is provable , and every form ula A ( x x, . . . , x n)
without quantifiers num eralwise expresses the predicate A (xv . . . , x n)
which it expresses under the interpretation.
U s in g th is defin ition , w e ca n estab lish th e fo llo w in g theorem , in m u ch
th e sam e m ann er as T h eo rem 21 § 3 7 , w h ich corresponds to it for th e
p red icate calculus.
T heorem 61. (a)N I f T f- in the intuition istic form al system of number

E
theory, and the form ulas Y are true, then E is true. (b)c S im ila rly in the
classical form al system of num ber theory.
T h e o n ly difference in th e proofs of P a r ts (a) an d (b) is th a t for (b)
w e need to use classical m eth o d s in th e tre a tm e n t of an ax io m b y th e
classical A x io m S ch em a 8. W e lab el P a r t (a) w ith to in d ic a te th a t,
a lth o u g h th e reasoning is in tu itio n istic, n o n -elem en tary m eth o d s are
u sed ; an d P a r t (b) w ith to in d ic a te th a t n o n -in tu itio n istic classical
m eth o d s are e m p lo y e d (cf. § 3 7 ) .
Sin ce A an d - i A ca n n o t b o th be true, T h eo rem 61 (a) (for Y em p ty )
im plies th e sim ple c o n sisten cy of th e in tu itio n istic n um ber th eo ry, an d
th en ce b y C o ro llary 2 T h eo rem 60 of th e classical n u m b er th eo ry, as an
“ N” result. T h e g a in b y T h eo rem 60 is th a t w e do n o t h a v e to c a ll the
la tte r a <<c” result, as w e w o u ld in inferring it d ir e c tly from T h eorem
61 (b).
E xample 5. (a) A n V i-pren ex form ula, if true, is general recursively

true (§ 79). F o r exam p le , if V v 3 w 03 w 1C (v, w 0, w x), w here C (v, w 0, w x) con
tain s no qu an tifiers an d o n ly th e d istin ct variab les show n, is true th en
{v){C (v, w 0(v), w t (v)) is t) w hen w^v) = (\lw C{ v , (w)0, (w)1))i, w h ich is
gen eral recu rsive using # 1 9 § 4 5 an d T h eorem I I I § 5 7 . (b)N or c H en ce
§82 RECURSIVE REALIZABILITY 501
b y T h eo rem 61 (a) or (b): In the number-theoretic form al system , every prov
able V l-pren ex form ula is general recursively true. (Cf. R e m a rk 2 § 79.)
U n d er our G o d el n u m b erin g o f th e form ulas, th e p red icate *A is tru e 1
becom es a n u m b er-th eo retic p red ic a te T(a), th e va lu e s of w h ich can be
g iv e n as propositions co n stru cted from p rim itiv e recursive p red icates
by th e operations of th e p ro p o sitio n al calcu lu s an d qu an tifiers, th e
num ber of th e la tte r used b ein g unbound ed. W e see from T h eo rem 30
th a t th is p red icate T(a) ca n n o t b e expressed, an d its essential properties
p ro ved , in th e sy ste m , as th en w e could form alize th e a b o v e co n siste n cy
proof in th e syste m . (Cf. H ilb e rt-B e r n a y s 1939 pp. 329— 340.)
In fa c t, each of th e pred icates (E x)T 1(a, a, x), (x )(E y)T 2(a, a, x, y),
(E x)(y)(E z)T z(a} a, x, y t z), . . . (cf. T h eo rem V P a r t I I (b) § 57} is e x
pressible in th e form T(^(a)) w ith a p rim itiv e recursive as can be seen
b y C o ro llary T h eo rem I § 49 (b y w h ich th e form ulas g iv e n b y C o ro llary
T h eo rem 27 to num eralw ise express th e predicates T ly T 2, T Z) . . . express
th e m also und er th e interpretatio n) w ith E x a m p le 2 § 52. Th erefore
b y T h eo rem s V I I (d) an d V , T(a) is not arith m etical.
T r u th d efinitions for form al sy ste m s were o rigin a lly in v e stig a te d b y
T a r s k i (19 3 2 ,^19 3 3 ). H e estab lish ed th a t, if an (effective) form al sy ste m
in c lu d in g th e usual n u m ber th e o r y is con sistent, it m u st be im p ossible
to express th e p red icate T(a) for th e sy ste m b y a form ula T(a) so th a t
T (« ) ~ A 0 is p ro v a b le in th e sy ste m w h en ever a is th e G o d el n u m ber of
a closed form u la A a. F o r th e n th e reasoning of th e E p im e n id e s p a ra d o x
( § 1 1 ) cou ld b e carried ou t in th e system . (For m ore d eta il, see H ilb e rt
an d B e r n a y s 1939 pp. 254— 269.)
The n otions of tru th for form ulas in tu itio n istic a lly an d cla ssica lly
sh ould differ. T h e a b o v e d efin itio n of tru th h o w e ver is p h rased alike for
th e tw o , an d a n y difference in th e notions has to be m ad e in our read in g
of th e w ords used in th e definition. In § 82, w e shall g iv e an o th er tru th
d efinition, w ith a theorem for it corresponding to T h eorem 61, w h ich
w ill a p p ly s e le c tiv e ly to th e in tu itio n istic system . T h e first results, like
T h eo rem 61 (a), w ill b e in tu itio n istic th o u g h n o n -elem en tary
b u t t h e y lead to results w h ich are m e ta m a th e m a tic a l in th e narrow er
sense.
§ 82. R e c u r s i v e r e a liz a b ility . Our problem is to express th e

in terp reta tio n of th e in tu itio n istic n u m ber th e o ry in a w a y w h ich m akes
e x p lic it som e featu re in w h ich it differs from th e classical.
T h e m ean in g of an e x iste n tia l sta te m e n t <<(E x)A (x)>' for th e in tu -
itio n is ts h as been exp la in ed b y sa y in g th a t it co n stitu tes an in co m p lete
co m m u n ica tio n o f a sta te m e n t g iv in g an x such th a t A (x) (H ilb ert-
B e rn a y s 1934 p. 32). B u t “A (x )” itself m a y in tu rn b e an in co m p lete
co m m u n ica tio n . A c c o r d in g ly le t us s a y t h a t tt(E x )A (x y> is an in co m p lete
co m m u n ica tio n , w h ich is co m p le ted b y g iv in g an x such th a t A (x) t o
g e th e r w ith th e further inform ation required to co m p lete th e co m
m u n ica tio n “A (x )” for th a t x.
T h e id ea ca n b e e x te n d e d to th e oth er lo gical operations. F o r exam p le,
w e ca n regard a g e n e ra lity sta te m e n t “ {x)A(x)" in tu itio n istic a lly as an
in co m p lete co m m u n ica tio n , w h ich is co m p le ted b y g iv in g an e ffe c tiv e
gen eral m eth o d for fin d in g, to a n y x, th e inform ation w h ich co m p letes
th e co m m u n ica tio n ttA{x)t$ for th a t x.
S im ila rly, an im p lica tio n “A -> B " can be regarded as an in co m p lete
co m m u n ica tio n , w h ich is co m p leted by g iv in g an e ffe c tiv e gen eral
m e th o d for o b ta in in g th e in form ation w h ich co m p letes w h en ever
t h a t w h ich co m p letes “A ” is given .
N e g a tio n can b e reduced to im p lica tio n (cf. E x a m p le 3 § 74).
N o w e ffe ctiv e gen eral m eth o d s are recursive ones, w h en it is a n a tu ral
n u m b er th a t is b ein g g iv e n (§§ 60, 62, 63). M oreover, b y th e d evic e o f
G o d el num berin g, in fo rm atio n can b e g iv e n b y a num ber.
C o m b in in g these ideas, w e shall define a p ro p e rty of a n u m b er-
th eo retic form ula w h ich w ill am o u n t to th e fo rm u la's b ein g tru e under
th e in terp retatio n su ggested . H o w ever, in stead of sa y in g 'tru e ', w e sh all
s a y '(recursively) realizab le', to d istin gu ish th e p ro p e rty d efined b elo w
from 'tr u th ' as defined b y usin g d irect tran slation s of th e form al lo gical
sym b o ls b y corresponding in form al w ords (end §81 ) .
T h e in terp retatio n of a term t ( x x, . . . , x n) co n ta in in g o n ly xv ..., xn
free b y a p rim itiv e recursive fu n ctio n t(xv ..., x n), or for n = 0 by a
n um ber t, an d th e in terp retatio n of a prim e form ula P ( x x, . . . , x n) b y
a p rim itive recursive p red icate P (x v ..., x n), or for n = 0 b y a p rop
osition P (end § 81 ) , do n o t differ in tu itio n istica lly from classically.
W e b u ild u pon th is in se ttin g u p th e d efinition of 're a liz a b ility ' w h ich
in terprets th e lo gical operators in tu itio n istica lly as ap p lied to nu m b er-
th eo retic form ulas.
F irst w e define th e circu m stan ces under w h ich a n a tu ral n u m ber e
'(recursively) realizes' (or is a 'realizatio n n um ber' of) a closed n u m ber-
th eo retic form u la E , b y in d u ctio n on th e n um ber of (occurrences of)
lo gical sym b o ls in E .
(A) 1. e realizes a closed prim e form ula P , if e= 0 an d P is tru e

(in oth er w ords, if e= 0 an d P ).
F o r C lauses 2 — 5, A an d B are a n y closed form ulas.
2.e realizes A & B , if e — 2a*36 w here a realizes A an d b realizes B .

3. e realizes A V B, if e = 2°*3a where a realizes A , or e = 2 1*35 w here
b realizes B .
4. e realizes A 3 B , if e is th e G o d el n u m ber of a p a rtia l recursive
fu n ctio n <p of one v a ria b le such th a t, w h en ever a realizes A , th e n <p(a)
realizes B .
5. e realizes t A, if e realizes A 3 1= 0 .
F o r Clauses 6 an d 7, x is a va ria b le , an d A (x ) a form u la co n ta in in g
free o n ly x.
6. e realizes 3 x A (x ), if e = 2**3a w here a realizes A (x ).

7. £ realizes V x A ( x ) , if e is th e G o d el n u m ber of a gen eral recursive
fu n ctio n cp of one v a ria b le such th a t, for e v e r y x, <p(x) realizes A (x ).
N ow we define '(recursive) re a liz a b ility ' for a n y n u m b er-th eo retic

form ula, thus.
(B) A form u la A co n ta in in g no free v a ria b les is realizable , if th ere

e x ists a num ber p w h ich realizes A . A form u la A t y ^ . . . , y m) co n ta in in g
free o n ly th e d istin ct v a ria b les y x, . . . , y m (m > 0) is realizable, if th ere
e x ists a gen eral recursive fu n ctio n 9 of m va ria b le s (called a realization
function for A ; y „ . . . . y m)) su ch th a t, for e v e r y y 1( . . . , y m, cp(y1, . . . . y m)
realizes A ( y x, . . . , y m). (U sin g §44, if a g iv e n form u la is realizable
for one choice of th e yv . . . , y w, it is for e v e r y other.)
T h e h an d lin g of th e free va ria b le s in th e present d efin itio n of real

iz a b ility differs from t h a t in K le e n e 19 4 5 . I t sim plifies th e proof o f th e
first theorem (Theorem 62), a fte r w h ich th e e q u iva len c e of th e tw o
d efin itio n s w ill follow (b y C o ro llary 1 ).
T h e a b o v e defin ition of re a liz a b ility refers o n ly to our n o tio n o f n u m b er-
th eo retic form ula, i.e. to th e fo rm atio n rules of our form al system .
A m odified n o tio n of re a liz a b ility , referring to th e p o stu la te list of th e
sy ste m , an d to assu m p tio n form ulas T if desired, is o b ta in e d b y alte rin g
th ree clauses, as follow s. C lau se 3 : replace “ a realizes A ” b y “ a realizes
A an d T f- A " , an d “ b realizes B ” b y “ b realizes B an d T b B ” .
C lau se 4: replace “a realizes A " b y “ a realizes A an d V b A " . C lause 6 :
replace “a realizes A ( x ) ” b y “ a realizes A (x ) an d T b A ( x ) '\ F o r 'real
izes' ['realizable'] in th is m o dified sense w e s a y realizes -( T b) [realizable-
(r K>].
T heorem 62n. (a) If T b E in the intuitionistic number-theoretic
formal system, and the formulas T are realizable, then E is realizable.
(David N elso n 1947 Part I.)
(b) S im ila rly reading “ realizable-(T b)” in place of “realizable” .
L emma 44n. If x is a variable, A(x) is a formula without free variables
other than x, and t is a term without variables which hence expresses a number
t, then e realizes A(t) if and only if e realizes A(t).
P roof of L emma 44. If A(x) is prime, then whether A(t) is true is
equivalent to whether A(t) is true. Hence by Clause , the lemma holds
1
for a prime A(x). The lemma for any other A(x) follows fromthis basis
by induction on the number of logical symbols in A(x), with cases cor
responding to the other clauses in the definition of ‘realizes'.
L emma 45n. If E is a closed formula, then e realizes E if and only
if e realizes the result of replacing each part of Eof the form Awhere Ais
-1
a formula Jy A3 1= 0.
Lemmas 44 and 45 also hold reading ‘T b”or “e realizes-(r in\-)”
place of “e realizes”, when brefers totheintuitionisticnumber-theoretic

system, and T are any formulas. (For Lemma 44 we then use (A) §41
withTheorem24 (b) §38.)
P roof of T heorem 62. We state the proof for (a), and (optionally)
the reader, by taking slight extra care, can verify that the additional
conditions aremet for (b). The proof is byinductiononthe lengthof the
given deduction T b E, with cases corresponding to the postulates of
our formal system.
First we consider axioms . If A(yv ..., ym ) is an axiomcontaining as
its only free variables y t, ..., y m, then by (B) to establish its realiz
abilitywemust give ageneral recursivefunction<?(yv ..., y m) suchthat,
for every m-tuple of natural numbers yv ..., y m, the number
9( . ..,yw) realizes A
y lt (y v However, for each of the axiom
. . . , y m ).
schemataof the propositional calculus, weshall be able to findanumber

whichrealizesA ...,( y x, y m)foranyaxiomA(yx, ..., y m) bythe schema.
It will sufficetogivethis number (whichrealizestheclosedaxiomsbythe
schema), becausewhenfreevariables ylf ..., ymarepresent, wecantake
as 9( ..., y m) the constant function of variables with this number
y i, m
as value (§44). Similarly for the particular number-theoretic axioms,

we shall merely give a number which realizes the result of any sub
stitution of numerals for the free variables of the axiom. Similarly for
AxiomSchema 13, we can give a realization number, as a general re
cursive function of x, which depends only on the numeral x substituted
for x; and for each of Axiom Schemata 10 and 11 one can be given, as
a general recursive function of x v ..., x n, which depends only on the
t and on the numerals xx........xn substituted for its variables xlf ..., x„.
Then when the y1( . . ym include other variables, the <p{yx, ..., y m)
can be obtained by expanding that function into a function of the
required additional variables by use of identity functions (§ 44).
For each of the axiom schemata and particular axioms (§§ 19, 23),
we shall express our realization number or function using the notations of
§ 65. The proof that it is a realization number or function, and the
necessary verifications of recursiveness, are left to the reader in cases not
discussed in detail.
la. In accordance with the preliminary remarks, consider an axiom
A D (B D A) by this schema containing no free variables. We show
that AaAba, i.e. AaAb U\(a, b) (§44), realizes A 3 (B 3 A). For let
a realize A; by Clause 4, we must show that {AaA6 a}(a), i.e. Ab a (by
(71) § 65), realizes B 3 A. To show this, let b realize B; we must show that
{A6 a}(b), i.e. a, realizes A. But a does realize A, by hypothesis.
lb. (A 3 B ) 3 ((A 3 (B 3 C)) 3 (A 3 C)) is realized by
ApAqAa {q(a)}(p(a)). For let p realize A 3 B ; we must show that
{ApAqAa{q(a)}(p(a))}(p), i.e. AqAa{q(a)}(p(a)), realizes (A 3 (B 3 C))3
(A 3 C). To show this, let q realize A 3 (B 3 C); we must show that
Aa {q{a)}(p{a)) realizes A 3 C. To show this, let a realize A; we must
show that {q(a)}(p(a)) realizes C. Now by hypothesis, p realizes A 3 B
and a realizes A; hence p(a) realizes B. Moreover q realizes A 3 (B 3 C),
and a realizes A; so q(a) realizes B 3 C. But now q(a) realizes B 3 C,
and p(a) realizes B; hence {q(a)}(p(a)) realizes C, as was to be shown.
3. A 3 (B 3 A & B). AaAb 2°-36.
4a. A & B 3 A. Ac (c)0 (cf. #19 § 45). 4b. A & B 3 B. Ac {c)v
5a. A 3 A V B. Aa 2°-3°. 5b. B 3 A V B. Ab 21-36.
6. (A 3 C) 3 ((B 3 C) 3 (A V B 3 C)).
ApAqAr x(P> <?. P) where
X{p, q, r) ~ [ M r ) i) if.(r)0 = ° . ? ( M i) if Wo = >]> usin g T h eo rem X X (c).
Su p p ose p realizes Ad C , q realizes B d C, a n d r realizes A V B ; w e
m u st sh ow th a tx{P> <1> r) realizes G. Case l: r = 2°*3a w here a realizes A .

T h e n (r)0 = 0 an d (r)x = a. Sin ce p realizes A d C an d (r)x realizes A ,
p ((r) j) realizes C. B u t (r)0 = 0; so (and becau se p ((r) i) is defined)
l(fi> r) = p {{r) i), an d so it realizes C, as w as to b e show n. Case 2:
r = 2 1*36 w here h realizes B . S im ilarly.
506 C O N S IS T E N C Y CH . X Y
7. (A 3 B) 3 ((A 3 - i B ) 3 - i A ). U sin g Lem m a 45, th e n u m b er

w h ich realizes th e closed a x io m s b y A x io m S ch em a lb (in p a rticu la r
th o se w ith 1 = 0 as th e C) realizes th o se b y th is schem a.
81. - i A3 (A 3 B ). 0. F o r if p realizes - i A , th en b y C lause 5, p

realizes A 3 1 = 0 . B u t th e n no n u m ber a ca n realize A , since p(a) w o u ld
realize th e false closed prim e fo rm u la 1 = 0 , co n tra d ic tin g C lause 1. T h u s
v a c u o u sly , if p realizes n A a n d a realizes A , th en {0 (p)}{a) realizes B .
(Th e reader m a y fin d it in s tr u c tiv e to v e r ify th a t there is no ap p aren t
w a y to tre a t th e classical A x io m S ch em a 8.)
10. Let the t for the axiom contain exactly the distinct variables
Xp . . . , x B {n ^ 0); denote it as "t(xlf ..., xn)”, and let t(xlt ,.., x n)
be the primitive recursive function (or for n = 0, the number) which
it expresses. By the preliminary remarks, we suppose the axiom contains
free only x1? . . ., xn; if none of . . ., is x, let it be VxA(x,
x v x n x lf
. • X„) D A(t(x„ . . x„), Xj, . . x„). Since t(x1, . . x„) is free for x
in A(x, xlf ..., xn), the result of substituting numerals x v . . . , x n
for (the free occurrences of) x v ..., xn in the axiom is VxA(x, x v ..., x n)
3 A ( t ( x lf ..., xn), x v ..., xn). We shall show that the number
A p p(t(xv ..., x n))t which as x v ..., x n vary is a general (in fact, primi
tive) recursive function of x v ..., x n, realizes this formula. By Clause 4,
for this purpose we must show that, if p realizes VxA(x, x v ..., xn),
then p(t(xv ..., x n)) realizes A(t(xx, ..., x„), x^ ..., xn). But, if p
realizes VxA(x, xlf . . . , x n), then by Clause 7, p{t(xv . . . , x n)) realizes
A(t, x v ..., x n) where t = t(xv ..., x n); and hence by Lemma 44,
p { t(x v. . ., also realizes A(t(xlf ..., xn), xl9 . . . , x n). — If say is x,
x n ))
the axiom is Vx1A(x1, . .., xn) 3 A(t(x1? . . ., x j, x2, . . x j, etc.

11 . A(t(xx, . . . , xn), . . . , xn) 3 3xA(x, x v . . . , xn).
x lf
Aa 2tlxi'->x»]-3a.
13. A(0) & Vx(A(x) 3 A(x')) 3 A(x). We treat the case that the
A(x) contains free only x, as the preliminary remarks will then take care
of the general case. Let a partial recursive function p(x, a) be defined by a
primitive recursion thus,
r p ( 0 , a) = (a)0,
X p ( * ' > « ) - { { ( « ) i K * ) } ( p ( * . « ) ) •
Now we show’that for every x the number Aa p(x, a), which is a primitive
recursive function of x, realizes A(0) & Vx(A(x) 3 A(x')) ^ A(x). To
do so (Clause 4), we prove by induction on x that, if a realizes
A(0) & Vx(A(x) D A(x')), then p(x, a) realizes A(x). B asis. If a realizes
A(0) & V x (A (x ) D A ( x /)), th en b y C lause 2, p(0, a) [= (a)0] realizes A(0).
I n d . step. S im ila rly (a)x realizes V x (A (x ) D A (x ')), an d hence (Clause 7)
{(a)i}(x) realizes A (x ) D A ( x ') . B u t b y h y p . ind., p(x, a) realizes A(x).
H en ce (Clause 4), p (x \ a) [ = {{(^)i }{x)}{pix > a))] realizes A ( x ').
14. A fte r su b stitu tio n of num erals, we have from th is a x io m

a'=b' D a=b. T h is form ula is realized b y 0. F o r suppose ft realizesA ft
a'=b'. W e m u st show th a t th en 0 realizes a=b. Sin ce a'=b' is prim e, it

is o n ly realizable if it is true, i.e. if a ’ = V . T h e n a = b, so a=b is also
true, an d 0 realizes it.
S im ilarly, for th e other p articu lar axiom s, a fter s u b stitu tin g num erals,
w e h a v e realization num bers as follows.
15, 18 — 2 1 : 0. 16: Aft A q 0. 17 : A ft 0.
R u l e s o f in f e r e n c e . 2. W e ta k e a d v a n ta g e of th e rem ark a c c o m p a
n y in g th e defin ition of r e a liz a b ility to regard th e form ulas as each d e
p en d en t on all of th e v a riab les occurring free in a n y of th em . Thus w e
w rite th e rule
A(yi, • •., y„) A(y1, ..., ym) 3 B(y1, . ■ ., ym)

B ( y 1) . . . , y m).
By hypothesis of the induction, the premises A(y1( ..., ym) and

A(yi, ..., vm) 3 B(y1, ..., ym) are realizable, i.e. there are general
recursive functions a and <J>, such that, for every w-tuple of natural
numbers ylt .... y„, A(yv ..., y m) is realized by the number a(y1, ..., y m)
and A(y1; 3 B(y,, . . . , y m) by the number <J>{yv .... y m).
Then the number {^(y1( ..., ym)}(a(y1, ..., ym)) realizes B(yx, . . . . y m).
Moreover, {ty(yv ..., ym)}(a(y1, ..., y m)) is obviously a partial recursive
function of yx, ..., y m. But its value is a realization number for every
Vx> ■ • Vm, so it must be defined for every y1( ..., y m; thus it is general
recursive. Thus the conclusion B(y1( .... ym) is realizable.
9- C ( y x, . . y m) 3 A ( x , y 1, . . . , y m)
C(yi, ..., ym) 3 VxA(x, y1( ..., ym).

By the hypothesis of the induction and the definition of realizability,
there is a general recursive function ^ such that, for every x ,y v ..., y m,
Vi, ■ ■ ■ . Vm) realizes C ^ , ..., y m) 3 A(x, y v ..., y m). We shall
prove that, for every yv . . . , y m, AcAx{^{x, yv ..., y m)}(c) realizes
C(yi, ..., y m) 3 VxA(x, y 1: ..., y m). This will give the realizability of
the conclusion, since Ac Ax {${x,yv ..., y-m)}{c) is a primitive recursive, a
fortiori general recursive, function of yv ..., y m. Accordingly suppose that
c realizes C(yv . . y m); we must show that Ax {<\>(x, yv ..., y m)}(c)
realizes VxA(x, y lt ..., y m). To do this, we must show that, for every x,
{<K*> Vi> • • •> ym)}(c) realizes A (x ,yv .. . , y m). But since c realizes
C(yi, .. •, y m), and by hyp. ind. <]/(*. ylt .... y n) realizes C(ylt . . . , y m) 3
A(x, y v . . . . y m), {<j/(*, yv .... ym)}(c) does realize A(x, y v . . . . y m).
(Note how this treatment would break down, if the C contained x free,
call it “C(x, y v ..., ym)”. Then, we would have to assume that c realizes
C(x ,y v . . . , y m) for some x, and we could conclude only that
{ty(x>yv realizes A(x ,y v . for that x, whereas we
would need to conclude it for every x.)
12. A(x, yx, ..., yTO) => C(ylt .. ■ , y m)
3xA(x, ylf , . . , y j 3 C(yx, .... y m).
Similarly, using Ap {^((^>)0>yv .. •,y m)}((P) 1) as realization function for
the conclusion, given that ^ is for the premise.
The theorem includes the simple consistency of the intuitionistic
formal system of number theory ( by using (a) with T empty and 1 = 0
as the E), as does Theorem 61 (a). The additional interest in Theorem
62 in this connection stems from the different condition on new axioms
T under which it is shown that the simple consistency is preserved (as
we shall discuss further following Theorem 63).
Corollary 1 n. If y v ..., ymare distinct variables, and A (y v ..., ym)
is a formula, then A (y v ..., y m) is realizable, if and only if
vyi • • • vymA(y„ ... i Ym) ^ realizable.
For A(yj........ym) and Vyx ... VymA(ylt ..., ym) are interdeducible
in the intuitionistic formal system.
This corollary (applied to the case y1, ..., ym are the free variables
of the given formula in order of first free occurrence) gives the equivalence
of the present version of the definition of realizability (Kleene 1948 )
to that of Kleene 1945.
Corollary 2n. (a) If T are realizable formulas, A(x1( ..., xn, y) is a
formula containing free only the distinct variables x^ ..., xn, y, and
r I- 3yA(x1, ..., xn, y) in the intuitionistic number-theoretic formal
system, Then there is a general recursive function y = <p(xv such
that, for every xv . . . , x n, A(x1( ..., xn, y) (where y = <p(*1( ..., *„))
is realizable.
(b) Similarly reading in place of "realizable" any one of the following
combinations of properties: (i) ”realizable-(T |-) and deducible from T”,
realizable -(T b)> deducible from F , and true” , (ill) “rea liza b le-^ |-),
(ii) “
deducible from T, and realizable” , (iv) " realizable - ( r \-), deducible from T ,
true and realizable ''.
P roofs , (a) B y (a) o f th e th eorem w ith (B) a n d (A) 6 o f th e d efin itio n s,
(b) (i) U s in g in stea d (b) o f th e theorem , (ii) U s in g fu rth e r T h e o re m 61 (a)
t o infer t h a t A ( x 1# . . . , x n, y ) is true, (iii) U s in g fu rth e r (a) o f th e th eo rem
to infer t h a t A ( x 1# . . x n, y ) is realizable.
R e a liz a b ility is in ten d ed as an in tu itio n istic in te rp re ta tio n o f a fo rm u la ;

a n d to s a y in tu itio n is tic a lly t h a t A (x v . . . , x n, y) is realizab le sh o u ld
im p ly its b e in g in tu itio n is tic a lly true, i.e. th a t th e p ro p o sitio n
A (x lt ..., x n, y) c o n stitu tin g its in tu itio n istic m ean in g holds. T h e fo rm u la
3 y A ( x 1, . . . , x n, y) asserts th e existen ce, for e v e r y xl9 . . . , x n, o f a y
d ep en d in g on x1$ . . . , x n, such th a t A (x v . . . , x n, y ); or in o th er w o rd s,
th e e x isten ce of a fu n ctio n y = <p(xv . . . , x n) su ch th a t, for e v e r y
xv . . . , * n, A (x lf ..., x n, <p(xv By (a) o f th e co ro lla ry for T
e m p ty , t h a t form u la ca n be p ro v e d in th e in tu itio n istic fo rm al sy s te m ,
o n ly w h en there ex ists such a <p w h ich is gen eral recursive. In b rief,
o n ly n u m b er-th eo retic fu n ctio n s w h ich are gen eral recu rsive can be
p ro v e d to ex ist in tu itio n istic a lly . (W e are here con siderin g th e assertion
o f th e existen ce of a fu n ctio n v a lu e <p(xv ..., x n) for all n -tu p le s
xv ..., xn o f argu m en ts, so th is is n o t in co n flict w ith our use in tu
itio n istic a lly o f p a rtia l recursive functions.)
T h is result as inferred from (a) depends on a c c e p tin g th e th esis t h a t
th e re a liz a b ility of A ( x x, . . . , x n, y) im plies its tru th . H ow ever by
u sin g (b) for T e m p ty (in w h ich case, since w e h a v e no h y p o th e sis o n
r to sa tisfy , w e m a y ta k e th e stron gest form (iv) in th e con clusion, i.e.
t h a t A ( x 1# . . . , x n, y) is realizable-(|-), p ro v ab le, tru e an d re alizab le),
w e o b ta in th e sam e result in d e p e n d e n tly of t h a t thesis.
T h e presence o f th e T in th e co ro llary show s t h a t th e resu lt w ill h o ld
g o o d u p o n en largin g th e form al sy ste m b y a n y su ita b le a x io m s F . I f
th e thesis th a t re a liz a b ility im p lies tru th , in tu itio n istic a lly , is a c c e p te d ,
th ese need o n ly b e realizable. O th erw ise t h e y sh o u ld b e r e a liz a b le -( r b)
a n d tru e (d e d u cib ility from F h old s a u to m a tic a lly in th e h y p o th e s is on T ).
T h e result p ro vid es a co n n ectio n b etw ee n B ro u w er's lo gic as fo rm alized
b y H e y tin g an d C h u rch 's thesis (§ 62) th a t o n ly gen eral recu rsive fu n ctio n s
are e ffe c tiv e ly calcu lab le. B o th d e ve lo p m en ts arose from a c o n s tr u c tiv is tic
sta n d p o in t, b u t w ere p re v io u sly u n related in th eir d etails.
The form ula 3 y A ( x 1# . . . , x n, y) does n o t assert th e uniq ueness o f
the function y = (p(xlf ..., xn) such that A(xv . . xnt <p(%, ..., xn));
for this we need 3!yA(x1, ..., xn, y) (§ 41).
Classically, given the existence of some function 9 such that, for all
xt , . . . , x n, A(xv . . xn, cp(xlt ..., #n)), the least number principle
provides formally a method of describing a particular one (*149 § 40,
*174b § 41). While we do not have the least number principle intuition-
istically, we do know by Corollary 2 that, whenever a particular intu
itionistic proof of a formula of the form 3yA(x1? ..., xn, y) is given, we
can on the basis of that proof describe informally a particular genera]
recursive function <p(xv ..., xn) such that, for all xv . . xn, A (xv ..., *n,
?(*!> ••••*»))•
E x am ple 1. (Cf. Example 8 (c) § 74.) Let be the intuitionistic
number-theoretic system. Let A(x, y) be a formula containing free only
x and y. Suppose that for each the formula A(x, y ) is true for exactly
one y. Then when (to obtain S2) we introduce f with the axiom A(x, f(x)),
the axiom characterizes f as expressing a certain function 9 under the
interpretation. By 3-introd. from the new axiom, f-2 3yA(x, y). Now
suppose f with the axiom A(x, f(x)) is eliminable. Then ^ 3yA(x, y).
Then by Theorem 62 Corollary 2 (b) (ii) with F empty, there is a general
recursive function y = ^x(x) such that, for each #, A(x, y) is true. But then
9X= 9. Thus in the intuitionistic number-theoretic system, a new function
symbol f (expressing a function 9) introduced with an axiom of the form
A(x, f(x)), where A(x, y) contains free only x and y, and A(x,y) is true
exactly when y = y(x)t is eliminable only when 9 is general recursive.
E xam ple 2. Let A(x, y) be any formula, containing free only x and y,
such that hi 3yA(x, y). Then as in Example 1, there is a general re
cursive function y = ^(x) such that, for each xt A(x, y) is true. The
demonstration of this (consisting mainly in the proof of Theorem 62 (b))
is constructive; given a proof of 3yA(x, y) (or the Godel number of such
a proof), we can find a system E of equations defining a recursively
(or a Godel number of 9^. Also it is effectively decidable whether a
number a is the Godel number of a proof of a formula of the form
3yA(x, y) where A contains free only x and y (Case 1), or not (Case 2). Let
a Godel number of 9^ in Case 1,
Ax x (i.e. a Godel number of U\), in Case 2,
where ambiguity as to which Godel number of which 9X(or which Godel
number of U{) is chosen is removed by suitable conventions. Then 0(a)
is effectively calculable. So by Church's thesis we may expect that 0(a)
is general recursive. (In fact, it is easy to prove that 0(a) is primitive
recursive, after establishing: ( 1 ) There is a primitive recursive function
£(a) such that, if a is the Godel number of a proof in the intuitionistic number-
theoretic system, then %(a) is a Godel number of a realization-(}-) function
<p(yi, •••Jm ) for the endformula A(yx, . . ym), where ylf . . y m are the
free variables of the endformula in order of occurrence in our list of the
variables.) Let <p(x) = {0(x)}(x)-|-L Then y(x) is general recursive. Now
let A(x, y) be a formula such that A(x, y) is true exactly when y — cp(x)
(e.g. one which numeralwise represents 9 , cf. Theorem 32 (a) §59). If
we now take this formula as the A(x, y) of Example 1 , we are led to a
contradiction by supposing that f with the axiom A(x, f(x)) is eliminable.
Thus: (2) There is a general recursive function 9 such that, in the intu
itionistic number-theoretic system, a new function symbol f expressing 9 with
an axiom of the form A(x, f(x)), where A(x, y) contains free only x and y,
and A(x, y) is true exactly when y — 9 (3;), is not eliminable (and 3yA(x, y)
is not provable for any such A(x, y)).
T h e o r e m 63n. For suitably chosen formulas A(x), B(x) and C(x, y).
the following classically provable formulas are unrealizable and hence
(by Theorem 62 (a)) improvable in the intuitionistic formal system of
number theory. (Specifically, let A(x, z) numeralwise express the predicate
Tt(x, x, z) of § 57, using Corollary Theorem 27 § 49. Let A(x) be 3zA(x, z),
B(x) be A(x) V iA (x) and C(x, y) be y ~ l V (A(x) & y = 0).)
(i) A(x)VnA(x).
(ii) Vx(A(x) V-iA(x)) (the closure of (i)).
(iii) -rnVx(A(x) V -1 A(x)) (the double negation of (ii)).
(iv) V x - n B ( x ) D - i - i V x B ( x ).
(v) nn{V x nnB (x) D -i-iVxB(x)} (the double negation of (iv)).
(vi) 3yC(x, y) D 3y[C(x, y) & Vz(z<y D -iC(x, z))] (cf. *149 § 40).
(vii) 3 y[y<w & C(x, y) & Vz(z<y D -iC(x, z))] V Vy[y<w D iC (x, y)]
(cf. *148).
Also the closure, and the double negation of the closure, of (vi) and of (vii).
((i) — (v): Kleene 19 4 5 with Nelson 1 9 4 7 .)
L e m m a 46n. (a) I f A is realizable, and B is unrealizable, then A d B
is unrealizable. Hence: I f A is realizable, then n A is unrealizable, (b) I f
A is closed and unrealizable, then A D B and (hence) - 1 A are realizable, and
(by (a)) “inA is unrealizable.
P roof of 46. (a) By Theorem 62 (a) or the case for Rule 2
L emma
in its proof, if A and A 3 B are realizable, so is B. (b) For a closed B,
any number, e.g. 0, realizes A 3 B, since vacuously, whenever a realizes
A (i.e. never), 0(a) realizes B.
L emma 47n. If P(xx, ..., xw) numeralwise expresses a general re
cursive predicate P(xv ..., xn) in the intuitionistic formal system of number
theory, then, for every xv ..., xn, P(xlt ..., xn) is realizable if and only if
P(xv . . . , x n).
P r o o f o f L em m a 47. If P(xlt ..., xn), then by § 41 (i), 1- P(xx, ..., x n),
and hence by Theorem 62 (a), P(xx, ..., xn) is realizable. Conversely,
suppose P(x1, ..., xn) is realizable. Because P{xv ..., xn) is a general
recursive predicate, we have (constructively) that, for the given
xt, . . . ,x n, either P{xlf . . . f xn) or P{xlt . . . , x n). In the latter case,
however, by §41 (ii), h ^P (x1, . . . , x n), and hence by Theorem 62
(a), -iP(x1, ..., xn) is realizable, which by Lemma 46 (a) contradicts
our supposition that P(x1, ..., xn) is realizable.
P roof o f T h e o r e m 63. (i) Suppose (i), i.e. 3zA(x, z) V -i3zA(x, z),
were realizable. Let <p(x) be a realization function for it; and set
p(x) = (<p(x))0. Then p(x) is general recursive, and takes only the values
0 and 1 (by (B) and (A) 3 of the definitions). Consider any fixed x. Case 1 :
p{x) = 0. Then (<p(x))x realizes 3zA(x, z); and hence (<p(x))1>;l realizes A(x, z)
where z = (<p(x))1>0, in which case by Lemma 47, Tx{x, x, z). Thus
(Ez)Tx(x, x, z). C ase 2: p(x) = 1. Then (<p(x))1 realizes -i3zA(x, z), i.e.
(<p(*))i realizes 3zA(x, z) 3 1=0. We shall show that then (Ez)Tx(x, x, z).
For if there were a z such that Tx(x, x, z), by Lemma 47 A(x, z) would be
realizable; say k realizes it. Then 2z-3k would realize 3zA(x, z); and
{(?(*))i}(2**3*) would realize 1= 0, which is impossible. The two cases
show that the general recursive function p(x) is the representing function
of (Ez)Tx(x, x, z). But {Ez)Tx{x, x, z) is non-recursive ((15) Theorem
V § 57); hence no such general recursive p{x) can exist. By reductio ad
absurdum, therefore (i) is unrealizable.
(ii), (iii). By V-elim. (i) is deducible intuitionistically from (ii), so
by Theorem 62 (a) also (ii) is unrealizable; and by Lemma 46 (b) so is
(iii), since (ii) is closed.
(iv) Since (iii) can be deduced from (iv), using *5la § 27 and V-introd.
(vi) We show as follows that (i) is deducible from (vi). From 1= 1
(which is provable) by V- and 3-introd., 3yC(x, y). Using this, from (vi) by
3-elim., 3y[C(x, y) & Vz(z<y 3 -iC(x, z))]. Preparatory to &- and
3-elim., assume C(x, y), i.e.
0) y = 1 V (A(x) & y=0)
and Vz(z<y D -iC(x, z)), i.e.
(2) Vz(z<y D -*{z= 1 V (A(x) & z=0)})
We use proof by cases from (1) to deduce (i) with the help of (2).
Case 1: assume y = l. For reductio ad absurdum, assume further
A(x). From this and 0=0, by &- and V-introd., 0=1 V (A(x) &0=0).
But also from y = 1 by *135b, 0<y; and thence from (2) by V-elim.
(with 0 as the t) and D-elim., -i{0=l V (A(x) & 0=0)}. Hence by re
ductio ad absurdum, -iA(x). By V-introd., A(x) V-iA(x), which is (i) and
does not contain free the variable y of our proposed 3-elim. Case 2: assume
A(x)&y=0. By &-elim. and V-introd., A(x)V-iA(x). (This deduction
is related to the intuitive reasoning of Example 6 § 64.)
(vii) From (vii) we can deduce (vi), as in the proof of *149 from *148.
Theorem 63 (i) — (v) imply that <3 V -i cC? is unprovable in the intu-
itionistic propositional calculus, and Vx{^{x) V -i <£?(*)),
“■ v~iVx(<£?(x)V-i<C?(^)), V.r-r-!<3(.r) D -\—C4x£Z{x) and
D -i-iV *^*)} in the intuitionistic predicate calculus,
as we already knew from Theorem 57 (b) and Theorem 58 (a) and (c).
The present proofs are less elementary than those based on Gentzen’s
normal form theorem, but contribute insight into the working of the
intuitionistic logic as an instrument for number-theoretic reasoning.
We succeed in showing AV-iA unprovable in intuitionistic number
theory only in the presence of a free variable x.
Corollary (to (ii))N. The formula m~\Vx(A(x) V-iA(x)) {although the
negation of a classically provable formula) is realizable.
By (ii) and Lemma 46 (b).
The formula Vx(A(x) V nA(x)) is classically provable, and hence under
classical interpretations true. But it is unrealizable. So if realizability
is accepted as a necessary condition for intuitionistic truth, it is untrue
intuitionistically, and therefore unprovable not only in the present
intuitionistic formal system, but by any intuitionistic methods whatsoever.
This incidentally implies that our classical formal system reinforced
by an intuitionistic proof of simple consistency cannot serve as an in
strument of intuitionistic proof, as suggested in § 14, except of formulas
belonging to a very restricted class (including those of the forms B(x)
and VxB(x) end § 42, but not the present formula Vx(A(x) V-iA(x))).
T h e n eg atio n - i V x ( A ( x ) V n A ( x ) ) of th a t fo rm u la is cla ssica lly un tru e,
b u t (b y th e corollary) realizable, an d hence in tu itio n istic a lly true, if w e
a c c e p t re a liz a b ility (in tu itio n istica lly established) as su fficien t for in tu
it ion istic tru th .
So th e p o s sib ility appears of asserting th e form u la - i V x ( A ( x ) V - i A ( x ) )
in tu itio n istic a lly . T h u s w e sh ould o b ta in an ex ten sio n of th e in tu itio n istic
n u m b er th e o ry , w h ich h as p re v io u sly been tre a te d as a su b sy ste m of th e
classical, so th a t th e in tu itio n istic an d classical n u m b er theories d iverge,
w it h " ! V x (A (x ) V - i A (x)) h o ld in g in th e in tu itio n istic a n d V x (A (x ) V -i A ( x ) )
in th e classical.
S u ch d ivergen ces are fam iliar to m a th e m a ticia n s from th e ex a m p le of
E u c lid e a n an d n o n -E u c lid e a n geom etries, an d oth er exam p les, b u t are
a n ew phenom enon in arith m etic. T h e first ex a m p le com es b y ad jo in in g
A „ (p ) or - i A v(p) to th e n u m b er-th eo retic form alism , cf. end §§ 42 an d 75.
N o t o n ly is th e form u la - i V x ( A ( x ) V - i A ( x ) ) itself realizable, b u t by
T h eo rem 62 (a) (ta k in g it as th e T), w hen w e ad d it to th e present in tu
itio n istic form al sy ste m , o n ly realizable form ulas beco m e p ro v a b le in th e
en larged sy ste m . So th en e v e r y p ro v a b le form u la w ill be tru e under th e
r e a liz a b ility in terp retatio n . In p articu lar, th e stren gth en ed in tu itio n istic
sy ste m is th u s show n b y in terp retatio n to be sim p ly con sisten t.
A fu ller discussion is g iv e n in K le en e 19 4 5 , w here th e proposed a d
ju n c tio n s to th e u n stre n gth en ed in tu itio n istic form al sy ste m of n um ber
th e o r y 5 , to o b ta in a stren gth en ed in tu itio n istic sy ste m S' d iv ergin g
from th e classical S c, are in th e form of an id e n tifica tio n of tr u th w ith
re a liz a b ility .
R efin em en ts of th e results w h ich w e are b a sin g here on in terp retatio n
are o b ta in e d b y N elso n 19 4 7 P a r ts I I — I V (w ith K le en e 19 4 5 ). B eca u se
t h e y all in v o lv e th e c o n sisten cy of th e n u m b er-th eo retic form alism , no
co m p le te ly ele m e n ta ry tre a tm e n t ca n b e ex p ected . B u t th e non -elem en-
tarin ess is m in im ized in th e results b ased on th is fu rth er w o rk of N elson
to th e fu ll e x te n t th a t th e results are p ro ved in e lem en ta ry m e ta m a th e
m a tic s u n der th e h yp o th e sis of th e sim ple co n sisten cy of 5 . In p articu lar,
b y these results w ith those of G o d el 1932-3 (cf. C o ro llary 2 T h eo rem 60),
it is d em o n strated m e ta m a th e m a tic a lly th a t b o th S' an d S c are sim p ly
co n sisten t if S is. (Nelson ta k e s as his S not our in tu itio n istic form al
sy s te m b u t one o b ta in ed , a p a rt from an in essential difference in th e
e q u a lity p o stu lates, by a d jo in in g to ours som e a d d itio n a l fu n ctio n
sy m b o ls w ith th eir d efin in g equ ation s. T h ese eq u a tio n s fit our sch em ata
(I) — (V) § 43 or clo sely sim ilar sch em ata, e x c e p t th a t also a certa in
sch em a of co u rse-o f-valu es recursion is allow ed. U sin g N e lso n ’s (i) — (iv)
p. 332, to each application of that schema a pair of equations having the
same form with f, g, h, t* replaced by f', g', h', t[ is provable without the
application; so the course-of-values recursion schema is eliminable. Then
by Example 9 § 74 with the remarks preceding it, the additional function
symbols are eliminable.)
Nelson 1 9 4 9 introduces a notion of 'P-realizability', using Which one
can set up a number-theoretic system diverging from both the strength
ened intuitionistic and the classical.
Gene Rose 1 9 5 2 investigates realizability in relation to the intu
itionistic propositional calculus.
Kleene 1 9 5 0 a plans the use of recursive functions in interpreting
intuitionistic set theory.
E xample 3. The operators D , - i , & , V applied to closed form ulas
(a)
A and B obey the strong 3-valued truth tables (§ 64, re sta ted w ith th e present
sym bols), when t, f, u are read as ‘realizable\ ‘unrealizable’y ‘unknown {or
value im m aterial)’, respectively ; i.e. th e ta b les th en g iv e o n ly correct
in fo rm atio n a b o u t th e re a liz a b ility or u n re a liza b ility of A D B , -1 A ,
A & B , A V B , w hen entered from such inform ation a b o u t A an d B . P roof .
Consider D. i f B is realizable, th en b y *11 § 2 6 w ith T h eo rem 62 (a),
so is A D B , corresponding to th e three t's in C olum n 1 of th e ta b le for D.
I f A is unrealizable, th en b y L e m m a 46 (b), A D B is realizable, corre
sp o n d in g to th e three t's in R o w 2. I f A is realizable an d B i s unrealizable,
th en b y L e m m a 46 (a), A D B is u n realizable, corresponding to th e f in
R o w 1 C olu m n 2. T h e ta b le for -1 is sim p ly th e f colum n of th a t for D ;
an d & an d V are ea sily treated , (b) A form ula without variables is realizable ,
if and only if it is true. I ts re a liz a b ility (and truth) or u n re a liza b ility
(and falsity) is th u s e ffe c tiv e ly d ecid ab le b y th e v a lu a tio n procedure
furnished b y th e usual in terp retatio n of 0, = an d th e classical
2 -v a lu e d tru th ta b les for D, - 1 , & , V (cf. § 79 before T h eorem 51).
P roof b y using E x a m p le 4 § 81, or t h u s : F o r closed prim e form ulas, tr u th
an d re a liz a b ility agree, an d can be decided. In b u ild in g th en ce co m p o site
form ulas b y th e operations of th e propositional calculus, w e a lw a y s
rem ain w ith in th e first tw o rows an d colum ns of th e 3 -v a lu e d tables,
(c) W e call a num ber e an 'Revaluation number of a closed form u la E , if
either e = 2°*3Cl (then ex = (e)x) an d ex realizes E , or e = 2 1*3° an d E is
unrealizable. F o r open form ulas, R -valuation function is d efined in a n a lo g y
to 'rea lizatio n fu n ctio n '. A form ida C (z1? . . . , z w) containing no quanti
fiers and only the distinct variables z v . . ., z m (m > 0) has a prim itive
recursive R -valuation function y (zv . . . , z m). P roof ( o m ittin g “ zv . . . , z m”
to save space). C ase 1 : C is a prime formula P. Then P expresses a prim
itive recursive predicate P, with representing function <p. Let y = 2?-3°.
C ase 2: C is A 3 B, where by hyp. ind. there are primitive recursive
R-valuation functions a and {3 for A and B, respectively. Let
2°-[3 exp Aa (p)J if (|3)0 = 0,
Y = 20,3° if («)0 = (P)0= l .
21-3° otherwise.
C ase 3: -iA. Take (3 = 21-3° in Case 2. Case 4: A & B. Let
f 2°-[3 exp 2^-3< N if (a)0= ([3)0= 0,
* | 21-3° otherwise.
C ase 5: A V B. Similarly, (d) A firenex formula is realizable, if and only
if it is general recursively true. (Cf. Remark 2 § 79 and Example 5 § 81.)
P roof . Consider again the formula G used as illustration in § 79. Let
a(y1; xv yt, xt, y3) be a primitive recursive R-valuation function for
A(y„ Xj, y2, x2, y3). Then if G is recursively true, it is realized by
21'1-[3 exp A% 2Viixl)-[3 exp Ax2 2?*ix'-Xi)-3 exp
(«CVi. xv y 2{ x i), x2, y3(xv *2)))J].
Conversely, if g realizes G, then G is recursively true with y1 == (g)0,
yt(xi) = ({(?)iH*i))o and y3(xv x2) = ({({fe)i}(î))1}W )0 as the required
number and general recursive functions.
E xample 4. Under the thesis that (x)(Ei)B(xf i ) holds intuition-
istically only if there is a general recursive a such that (x)B(x, ol(x)) (cf.
above, or Kleene 1943 p. 69 Thesis III), we show that
(a) (y)(Ei)i<aA(i, (y)f) - (Ei)i<a(y)A(i, y)
does not hold intuitionistically for all A (cf. (20) § 57). Using (51) § 61,
(x)(y)(Ei)i<2Wi{x, (y)().
Thence if we had (a) intuitionistically, we would get (x){Ei)i<2{y)Wi(x, y)
intuitionistically; whence by the thesis, for some recursive a talcing
values < 2 , (*)(y)Wa(x)(*, y ) ; whence (x)(Ey)Wa{x)(x, y); whence for
each x,
(b) (Ey)W0(x ,y )^ a (x )= 1, (Ey)W1(x, y) -> *(x)=0.
But we carl take a(%)= 1 as the R0(x, y) and <x(x)=0 as the R^x, y) of
(57) and (58) §61. Then either a(/) = 1 or a(/) = 0. If a(/) = 1, then
(Ey)R0(f, y), whence it follows as in § 61 that (Ey)W1(f, y), contradicting
(b). Similarly, if «(/) = 0.
BIBLIOGRAPHY
A date in medieval tailciphers (e.g. 1 9 0 8 , not 1908) appearing in conjunction with a
name constitutes a reference to this bibliography (e.g. Brouwer 1 9 0 8 ). The date
is ordinarily the date of publication of the complete volume of the journal in which
the article appears. There are exceptions; mainly, in the case of papers at inter
national congresses, for which the date is usually that of the congress (e.g. Hilbert
1 9 0 4 ). Letters a, b, etc. may be suffixed to the date to distinguish additional titles
with the same date; or a star *, to indicate the presence of a note accompanying
the title in this bibliography.
Only those works are listed which are cited from the text of the book (or in a
few cases, from elsewhere in the bibliography). In quoting above from foreign-
language works, we have translated into English.
Alonzo Church’s A bibliography of symbolic logic , constituting vol. 1 (1936)
no. 4 (pp. 121-218) of The journal of symbolic logic, with the Additions and cor
rections (and an index by subjects) in vol. 3 (1938) no. 4, pp. 178-212, are intended
to give complete coverage of the literature on symbolic logic (including such
related fields as recursive functions) up through 1935. These are available bound
together. (A few additional items have come to light since, and are listed in sub
sequent volumes.)
The literature since 1935 is covered by critical reviews in The journal of symbolic
logic. These are indexed by authors biennially (in odd-numbered years) and by sub
jects quinquennially (in 1940, 1945, etc.).
The present bibliography contains some items which are not widely available.
For publications after 1935, information about their content can almost invariably
be obtained by consulting the review sections of The journal of symbolic logic,
or also for publications after 1938, from Mathematical reviews under the subject
“Foundations” or (vols. 20-) “Logic and foundations” .
The author has made use of Church’s Bibliography and the review sections of
The journal of symbolic logic in preparing the present bibliography, and also of the
bibliographies in Fraenkel 1 9 2 8 and Heyting 1 9 3 4 (especially for items not within
the field of the Bibliography and Journal).
Beginning with 1951 an international series of monographs is appearing under the
title Studies in logic and the foundations of mathematics (North-Holland Pub. Co.,
Amsterdam), which includes research reports and expositions on many topics.
In the original printing of this book (1952), eleven works were cited which had
not yet appeared. Now these eleven bibliographical items have been completed or
corrected, but no new items have been added.
517
518 BIBLIOQRAPHY
A ckerm ann, W il h e l m
192 4 5 - . Begrundung des “terHum non datur” mittels der Hilbertschen Theorie der
Widerspruchsfreiheit. Mathematische Annalen, vol. 93, pp. 1-36.
1 9 2 8 . Zum Hilbertschen Aufbau der reellen Zahlen. Ibid., vol. 99, pp. 118-133.
1 9 4 0 . Zur Widerspruchsfreiheit der Zahlentheorie. Ibid., vol. 117, pp. 162-194.
See Hilbert and Ackermann.

B e h m a n n , H e in r ic h
1922 . Beitrdge zur Algebra der Logik , insbesondere zum Entscheidungsproblem .
Ibid., vol. 8 6 , pp. 163-229.
B e r e c z k i, I l o n a
19 4 9unpublished. Results reported to the author in a letter from Kalmar dated
November 17, 1949. The author obtained (A) — (C) of Example 1 § 57 after
receiving this letter; (D) was (in substance) mentioned (orally) to the author
by Kalmar on August 18, 1948.
1952 .
Nem elemi rekurziv fuggveny letezese {Existenz einer nichtelementaren rekur-
siven Funktion ). Comptes rendus du Premier Congres des Math. Hongrois
27 aout-2 sept. 1950, Budapest 1952, pp. 409-417. Cf. Peter 1951 pp. 61-67.
B ernays, P aul
1935. Sur te platonisme dans les mathdmatiques. L’Enseignement mathematique,
vol. 34, pp. 52-69.
1 9 3 5 a. Hilberts Untersuchungen uber die Grundlagen der Arithm etik. David
Hilbert Gesammelte Abhandlungen, vol. 3, Berlin (Springer), pp. 196-216.
1 9 3 6 . Logical calculus. Notes on lectures at the Institute for Advanced Study
1935-6, prepared with the assistance of F. A. Ficken. Mimeographed. Inst,
for Adv. Study, Princeton, N.J., 1936, 125 pp.
1 9 3 7 -4 8 . A system of axiomatic set theory. The journal of symbolic logic, vol. 2
(1937), pp. 65-77, vol. 6 (1941), pp. 1-17, vol. 7 (1942), pp. 65-89 and 133-145,
vol. 8 (1943), pp. 89-106, vol. 13 (1948), pp. 65-79.
1 9 3 8 . Sur les questions mithodologiques actuelles de la thiorie hilbertienne de la
dimonstration. Les entretiens de Zurich sur les fondements et la methode des
, Exposes et discussions, published
sciences mathematiques, 6 -9 D6cembre 1 9 3 8
by F. Gonseth, Zurich (Leemann) 1941, pp. 144-152. Discussion on pp. 153-161.
See Hilbert and Bernays.
B e r n s t e in , F e l ix
1898 . See p. 104 of Borel 1898 .
B e r r y , G. G.
190 6 . See p. 645 of Russell 190 6 .
B lack, M ax
1933 . The nature of mathematics. A critical survey. London (Kegan Paul, Trench,
Trubner) and New York (Harcourt, Brace), xiv-j-219 pp. Reprinted London
(Routledge and Kegan Paul) and New York (The Humanities Press) 1950.
BIBLIOGRAPHY 519
B oole, G eorge
1847 . The mathematical analysis of logic, being an essay toward a calculus of
deductive reasoning. Cambridge (Macmillan, Barclay & Macmillan) and London
(George Bell), 82 pp. Reprinted Oxford (Basil Blackwell) and New York (Phil
osophical Library, Inc.) 1948.
1 8 5 4 . An investigation of the laws of thought, on which are founded the mathe
matical theories of logic and probabilities. London (Walton and Maberly),
v -h iv + 424 pp. Reprinted as vol. 2 of George Boole’s collected works, edited
by Ph. E. B. Jourdain, Chicago & London 1916. Reprinted New York (Dover
Publications) 1951.
B oone, W il l ia m W.
1951 . abstract. A n extension of a result of Post. The journal of symbolic logic,
vol. 16, pp. 237-238.
1 9 5 2 . Review of Turing 1 9 5 0 . Ibid., vol. 17, pp. 74-76.
B orel, £ m il e
1898 . Lemons sur la theorie des fonctions. Paris (Gauthier-Villars).
B r o u w e r , L . E. J .
1 9 0 8 . De onbetrouwbaarheid der logische principes (The untrustworthiness of the
principles of logic). Tijdschrift voor wijsbegeerte, vol. 2, pp. 152-158. Reprinted
in Wiskunde, waarheid, werkelijkheid, by L. E. J. Brouwer, Groningen (P.
Noordhoff) 1919, 12 pp.
1923. Uber die Bedeutung des Satzes vom ausgeschlossenen Dritten in der Mathe-
matik, insbesondere in der Funktionentheorie. Journal fur die reine und ange-
wandte Mathematik, vol. 154 (1925), pp. 1-7. Original in Dutch 1923.
1 9 2 8 . Intuitionistische Betrachtungen uber den Form alism us . Sitzungsberichte der
Preussischen Akademie der Wissenschaften, Physikalisch-mathematische
Klasse, 1928, pp. 48-52. Also Koninklijke Nederlandsche Akademie van Weten-
schappen, Proceedings of the section of sciences, vol. 31, pp. 374-379.
B u r a l i- F o r t i, C e s a r e
18 9 7 . Una questione sui numeri transfiniti. Rendiconti del Circolo Matematico
di Palermo, vol. 11, pp. 154-164. See also ibid., p. 260. Concerning Cantor’s
discoveries of the Burali-Forti and Cantor paradoxes, see Fraenkel 1 9 3 2 , p. 470.
Ca n t o r , G eo r g
18 74 . Uber eine Eigenschaft des Inbegriffes alter reellen algebraischen Zahlen.
Journal fur die reine und angewandte Mathematik, vol. 77, p. 258-262. Reprint
ed in Georg Cantor Gesammelte Abhandlungen, Berlin (Springer) 1932, pp. 115-
118.
1 8 9 5 -7 . Beitrdge zur Begrundung der transfiniten Mengenlehre. Mathematische
Annalen, vol. 46 (1895), pp. 481-512, and vol. 49 (1897), pp. 207-246. Reprinted
in Georg Cantor Gesammelte Abhandlungen, pp. 282-351. English translation by
Ph. E. B. Jourdain entitled Contributions to the founding of the theory of
transfinite numbers, Chicago and London (Open Court) 1915, ix + 211 pp.
520 BIBLIOGRAPHY
Ca r n a p, R u d o l f
i 9 3 i -2 . D ie logizistische Grundlegung der M athematik.
Erkenntnis, vol. 2, pp.
91-105.
1 9 3 4 . The logical syntax of language. New York (Harcourt, Brace) and London
(Kegan Paul, Trench, Trubner) 1937, xvi+352 pp. Tr. by Amethe Smeaton
from the German original 1934, with additions.
Ch u r c h , A lonzo
19 3 2 . A set of postulates for the foundation of logic. Annals of mathematics,
second series, vol. 33, pp. 346-366.
1 9 3 3 . A set of postulates for the foundation of logic (second paper). Ibid., vol. 34,
pp. 839-864.
1 9 3 5 * A proof of freedom from contradiction . Proceedings of the National Academy
of Sciences, vol. 2 1 , pp. 275-281.
1 9 3 6 . A n unsolvable problem of elementary number theory. American journal of
mathematics, vol. 58, pp. 345-363.
1 9 3 6 a. A note on the Entscheidungsproblem . The journal of symbolic logic,
vol. 1, pp. 40-41. Correction, ibid., pp. 101-102.
1 9 3 8 . The constructive second number class. Bulletin of the American Mathe
matical Society, vol. 44, pp. 224-232.
1 9 4 1 . The calculi of lambda-conversion. Annals of Mathematics studies, no. 6 .
Lithoprinted. Princeton University Press, Princeton, N.J., ii+77 pp. Second
printing 1951, ii+82 pp.
1 9 5 1 . Special cases of the decision problem . Revue philosophique de Louvain,
vol. 49, pp. 203-221. A correction , ibid., vol. 50 (1952), pp. 270-272.
1956 . Introduction to mathematical logic. Princeton University Press, Prince
ton, N.J., vol. I (1956) x + 376 pp.,A vol. II was projected.
See Church and Kleene, Church and Quine.
C h u r c h , A l o n z o a n d K l e e n e , S. C.
1 9 3 6 . Formal definitions in the theory of ordinal numbers. Fundamenta mathe-
maticae, vol. 28, pp. 1 1 -2 1 .
C h u r c h , A l o n z o and Q u i n e , W. V.
1 9 5 2 . Some theorems on definability and decidability. The journal of symbolic
logic, vol. 17, pp. 179-187.
C u r r y , H a s k e i l B.
1 9 2 9 . A n analysis of logical substitution. American journal of mathematics,
vol. 51, pp. 363-384.
1 9 3 0 . Grundlagen der kombinatorischen Logik. Ibid., vol. 52, pp. 509-536, 789-834.
1 9 3 2 . Some additions to the theory of combinators. Ibid., vol. 54, pp. 551-558.
1 9 3 9 * A note on the reduction of Gentzen’s calculus L J . Bulletin of the American

Mathematical Society, vol. 45, pp. 288-293.
1 9 4 8 -9 . A sim plification of the theory of combinators . Synthese, vol. 7, pp. 391-399.
BIBLIOGRAPHY 521
1950 . A theory of formal deducibility. Notre Dame mathematical lectures, no. 6,
University of Notre Dame, Notre Dame, Ind., ix+126 pp.
1952 . The perm u tability of rules in the classical inferen tial calculus. The journal of
symbolic logic, vol. 17, pp. 245-248.
D a n t z ig , D . van
1947 . On the prin ciples of in tu itio n istic and affirm ative m athem atics. Koninklijke
Nederlandsche Akademie van Wetenschappen, Proceedings of the section of
sciences, vol. 50, pp. 918-929, 1092-1103; also Indagationes mathematicae, vol.
9, pp. 429-440, 506-517.
1948. S ig n ifies, and its relation to sem iotics. Library of the Tenth International
Congress of Philosophy (Amsterdam, Aug. z i- 18, 1948), vol. 2 Philosophical
essays, Amsterdam (Yeen) 1948, pp. 176-189.
D a v is , M a r t in
1950 abstract. R elatively recursive functions and the extended K leene hierarchy.
Proceedings of the International Congress of Mathematicians (Cambridge,
Mass., U.S.A., Aug. 30-Sept. 6, 1950), 1952, vol. 1, p. 723.
D e d e k i n d , R ic h a r d
1872. Stetigkeit und irrationale Zahlen. Braunschweig (5th ed. 1927). Also in
Dedekind Gesammelte mathematische Werke, vol. Ill, Braunschweig (Vieweg
& Sohn)*,1932, pp. 315-334. Eng. tr. by Wooster Woodruff Beman entitled Con
tin u ity and irration al num bers, pp. 1-24 of Essays on the theory of numbers,
Chicago (Open Court) 1901, 115 pp.
1888. Was sind und was sollen die Zahlen? Braunschweig (6th ed. 1930). Also in
Werke, vol. Ill, pp. 335-391. Eng. tr. by Beman, The nature and m eaning of
numbers, loc. cit., pp. 29-115.
D e M organ, A ug ustus
1847 . Formal logic: or, the calculus of inference, necessary and probable. Lon
don, xvi+336 pp. Reprinted Chicago and London 1926 (ed. by A. E. Taylor).
1864. On the syllogism , no. I V , an d on the logic of relations (read 23 April 1860).
Transactions of the Cambridge Philosophical Society, vol. 10, pp. 331-358.
D i x o n , A. C.
1906. O n “ well-ordered” aggregates. Proceedings of the London Mathematical
Society, ser. 2, vol. 4, pp. 18-20. Cf. ibid., pp. 317-319.
E in s t e in , A l b e r t
1944 . R em arks on B ertrand R u ssell's theory of knowledge. The philosophy of
Bertrand Russell, ed. by Paul Arthur Schilpp, Northwestern University,
Evanston and Chicago, pp. 277-291. (German with Eng. tr. by Schilpp.)
F e y s, R obert
- . L es logiques nouvelles des m o dalitis. Revue n£oscolastique de philosophie,
1937 8
vol. 40 (1937), pp. 517-553, vol. 41 (1938), pp. 217-252.
522 BIBLIOGRAPHY
1965 . Modal logics. Ed. with some complements by Joseph Dopp, Louvain
(E. Nauwelaerts) and Paris (Gauthier-Villars), xiv + 219 pp. This is the out
come of a plan for a joint work by Feys and J. C. C. McKinsey.
F in s l e r , P a u l
1926 . Formale Beweise und die Entscheidbarkeit. Mathematische Zeitschrift, vol.
25, pp. 676-682.
F raenk el, A dolf
1922. Der Begriff “definit” und die Unabhdngigkeit des Auswahlaxioms. Sit-
zungsberichte der Preussische Akademie der Wissenschaften, Physikalisch-
mathematische Klasse, 1922, pp. 253-257.
1925 . Untersuchungen iiber die Grundlagen der Mengenlehre . Mathematische
Zeitschrift, vol. 22, pp. 250-273.
1928. Einleitung in die Mengenlehre, 3rd ed., Berlin (Springer) 1928, xiii+424 pp.
Reprinted New York (Dover Publications) 1946.
1932. D as Leben Georg Cantors. Georg Cantor Gesammelte Abhandlungen mathe-
matischen und philosophischen Inhalts, ed. by Ernst Zermelo, Berlin (Springer),
pp. 452-483.
1953 . Abstract set theory. Studies in logic and the foundations of mathematics,
Amsterdam (North-Holland Pub. Co.), xii -f 479 pp.
F r eg e, G ottlob
1879 . Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des
reinen Denkens. Halle (Nebert), viii + 88 pp.
1884. Die Grundlagen der Arithmetik, eine logisch-mathematische Unter-
suchung fiber den Begriff der Zahl. Breslau, xix+119 pp. Reprinted Breslau
(M. & H. Marcus) 1934. Eng. tr. by J. L. Austin (with German original):
The foundations of arithmetic. A logico-mathematical enquiry into the concept
of number. Oxford (Basil Blackwell) and New York (Philosophical Library)
1950, (xii-j-XI + 119) X 2 pages.
1893. Grundgesetze der Arithmetik, begriffsschriftlich abgeleitet. Jena (H. Pohle),
vol. 1, xxxii+254 pp.
1903. Ibid., vol. 2, xv+265 pp. Eng. tr. of Sections 86-137 by Max Black entitled
Frege against the formalists in The philosophical review, vol. 59 (1950), pp. 77-93,
202-219, 332-345.
G entzen, Gerhard
: - . Untersuchungen iiber das logische Schliessen. Mathematische Zeitschrift,
3934 5
vol. 39, pp. 176r210, 405-431. Apart from minor differences in the notion of
•formula, and for the Hilbert-type systems in the precise selection of the postu
lates, our classical “formal system H ” for predicate calculus (cf. § 77) is Gent-
zen's “Kalkiil L H K ” , our intuitionistic “H ” is his “L H J ” , our classical “G l”
his “L K ”, and our intuitionistic “G l” his “L J ”.
1936. D ie W iderspruchsfreiheit der reinen Zahlentheorie. Mathematische Annalen,
vol. 112, pp. 493-565. He uses 1, 2, 3, . .. where we use 0, 1, 2, ....
BIBLIOGRAPHY 523
1938 . Die gegenwdrtige Lage in der mathematischen Grundlagenforschung. For-

schungen zur Logik und zur Gmndlegung der exakten Wissenschaften, new
series, no 4, Leipzig (Hirzel), pp. 5-18.
1 9 3 8 a. Neue Fas sung des Widerspruchsfreiheitsbeweises fur die reine Zahlenthe-
orie. Ibid., pp. 19-44.
1 9 4 3 . Beweisbarkeit und Unbeweisbarkeit von Anfangsfalien der transfiniten
Induktion in der reinen Zahlentheorie. Math. Ann., vol. 119 no. 1, pp. 140-161.
G l iv e n k o , V .
192 9. Sur quelques points de la logique de M . Brouwer. Acad6mie Royale de
Belgique, Bulletins de la classe des sciences, ser. 5, vol. 15, pp. 183-188.
G o del, K urt
1930 . Die V ollstandigkeit der Axiome des logischen Funktionenkalkills. Monats-
hefte fur Mathematik und Physik, vol. 37, pp. 349-360.
1 9 3 1 . Vber formal unenhcheidbare Sdtze der P rincipia M athematica und ver-
wandter Systeme I. Ibid., vol. 38, pp. 173-198.
1 9 3 1 -2 . Vber Vollstandigkeit und Widerspruchsfreiheit. Ergebnisse eines mathe
matischen Kolloquiums, Heft 3 (for 1930-1, pub. 1932), pp. 12-13. This paper
lists results without proofs.
19 3 1- 2 a. Remarks contributed to a Diskussion zur Grundlegung der M athematik.
Erkenntnis, vol. 2 , pp. 147-148.
1 9 3 2 . Zum intuitionistischen Aussagenkalkul. Akademie der Wissenschaften in
Wien, Mathematisch-naturwissenschaftliche Klasse, Anzeiger, vol. 69 (1932),
pp. 65-66. Reprinted in Ergebnisse eines mathematischen Kolloquiums, Heft 4
(for 1931-2, pub. 1933), p. 40.
1 9 3 2 - 3 . Z ur intuitionistischen Arithmetik und Zahlentheorie. Ergebnisse eines
math. Koll., Heft 4 (for 1931-2, pub. 1933), pp. 34-38.
1 9 3 4 . On undecidable propositions of formal mathematical systems. Notes by S.
C. Kleene and Barkley Rosser on lectures at the Institute for Advanced Study,
1934. Mimeographed, Princeton, N.J., 30 pp.
1 9 3 6 . Vber die Lange von Beweisen. Ergebnisse eines math. Koll., Heft 7 (for
1934-5, pub. 1936, with note added in press), pp. 23-24.
1 9 3 8 . The consistency of the axiom of choice and of the generalized continuum-
hypothesis. Proceedings of the National Academy of Sciences, vol. 24, pp. 556-
557. A full-length treatment is given in 1 9 4 0 .
1939- Consistency-proof for the generalized continuum-hypothesis. Ibid., vol. 25,
pp. 220-224.
1 9 4 0 . The consistency of the axiom of choice and of the generalized continuum-
hypothesis with the axioms of set theory. Lectures delivered at the Institute for
Advanced Study 1938-9; notes by George W. Brown. Annals of Mathe
matics studies, no. 3. Lithoprinted. Princeton University Press, Princeton 1940,
66 pp. (In Axiom 4 insert “(u)Mafter “(3z)’\ Also cf. Example 13 § 74 above.)
Second printing 1951, v + 69 pp.
524 BIBLIOGRAPHY
1944. Russell's mathematical logic. The philosophy of Bertrand Russell, ed. by
Paul Arthur Schilpp, Northwestern University, Evanston and Chicago, pp.
123-153.
1947. W hat is Cantor's continuum problem ? American mathematical monthly,
vol. 54, pp. 515-525.
G o n s e t h , F e r d in a n d
1933. La vdritd mathematique et la rdalitd. U Enseignement mathematique, vol. 31
(for 1932, pub. 1933), pp. 96-114.
Also: A propos d'un catalogue paradoxical. Rdponse de M . Gonseth d M . W inants .
Ibid., pp. 269-271.
H all, M ar sh a ll, J r .
1949. The word problem for semigroups with two generators. The journal of sym
bolic logic, vol. 14, pp. 115- 118.
H a s e n j a e g e r , G is b e r t
1950 . Vber eine A rt von Unvollstandigkeit des Prddikatenkalkuls der ersten Stufe,
Ibid., vol. 15, pp. 273-276.
H a u s d o r f f , F e l ix
19 14 . Grundziige der Mengenlehre. Leipzig (Viet), viii+476 pp. Reprinted New
York (Chelsea) 1949.
1 9 2 7 . Mengenlehre. Goschens Lehrbucherei, 1 Gruppe Band 7, Berlin and
Leipzig (Gruyter), a 2nd revised ed. of 1 9 1 4 (but less complete in some respects),
285 pp. 3rd ed., 1935, 307 pp. Reprinted New York (Dover Publications) 1944.
H e n k in , L eon
1949. The completeness of the first-order functional calculus. The journal of sym
bolic logic, vol.
14, pp. 159- 166.
1950 . Completeness in the theory of types. Ibid., vol. 15, pp. 81-91.
1950 a. A n algebraic characterization of quantifiers. Fundamenta mathematicae,
vol. 37, pp. 63-74.
H erbrand, Jacques
1928. Sur la thdorie de la ddmonstration. Comptes rendus hebdomadaires des
stances de l’Academie des Sciences (Paris), vol. 186, pp. 1274-1276.
1930. Recherches sur la theorie de la demonstration. Travaux de la Soci6t 6 des
Sciences et des Lettres de Varsovie, Classe III sciences math£matiques et
physiques, no. 33, 128 pp.
19 3 1 - 2 . Sur la non-contradiction de Varithmdtique. Journal fur die reine und
angewandte Mathematik, vol. 166, pp. 1-8.
H erm es, H ans
193 8 . Semiotik. Eine Theorie der Zeichengestalten als Grundlage fur XJnter-
suchungen von formalisierten Sprachen. Forschungen zur Logik und zur Grund-
legung der exakten Wissenschaften, n.s., no. 5, Leipzig (Hirzel), 22 pp.
BIBLIOGRAPHY 52 5
H e y t in g , A rend
1930. Die formalen Regeln der intuitionistischen Logik. Sitzungsberichte der
Preussischen Akademie der Wissenschaften, Physikaliseh-mathematische
Klasse, 1930, pp. 42-56.
1930a. Die formalen Regeln der intuitionistischen M athematik. Ibid., pp. 57-71,
158-169.
1931 2 - . Die intuitionistische Grundlegung der M athematik. Erkenntnis, vol. 2,
pp. 106-115.
1934 . Mathematische Grundlagenforschung. Intuitionismus. Beweistheorie. Er-
gebnisse der Mathematik und ihrer Grenzgebiete, vol. 3, no. 4, Berlin (Springer),
pp. iv + 73. Erratum: The theorem of Godel 1932-3 does not hold for the
predicate calculus quite as stated by Heyting on p. 18. Cf. Remark 1 § 81 above.
1946. On weakened quantification. Jour, symbolic logic, vol. 11, pp. 119-121.
H il b e r t , D a v id
1899. Grundlagen der Geometrie. 7th ed. (1930), Leipzig and Berlin (Teubner),
vii+326 pp. Eng. tr. by E. J. Townsend, The foundations of geometry, Chicago
(Open Court) 1902, iv-f 143 pp.
1900. Vber den Zahlbegriff. Jahresbericht der Deutschen Mathematiker-Vereini-
gung vol. 8, pp. 180-184. Reprinted with an omission in Grundlagen der Geo
metrie, 7th ed., Leipzig and Berlin (Teubner) 1930, pp. 241-246.
1904. Vber die Grundlagen der Logik und der Arithm etik. Verhandlungen des
Dritten Intemationalen Mathematiker-Kongresses in Heidelberg vom 8. bis 1.3.
August 1904, Leipzig 1905, pp. 174-185. Reprinted, loc. cit., pp. 247-261.
1918 . Axiomatisches Denken. Mathematische Annalen, vol. 78, pp. 405-415.
Reprinted in David Hilbert Gesammelte Abhandlungen, vol. 3, Berlin (Sprin
ger) 1935, pp. 146-156.
1926. Vber das Unendliche. Math. Ann., vol. 95, pp. 161-190. Reprinted in ab
breviated form in Jahresb. Deutschen Math.-Verein., vol. 36 (1927), pp. 201-215.
Also with some revisions in Grundlagen der Geometrie, 7th ed., 1930, pp. 262-288.
1928. Die Grundlagen der M athematik. Abhandlungen aus dem Mathematischen
Seminar der Hamburgischen Universitat, vol. 6, pp. 65-85. Reprinted with
abridgements in Grundlagen der Geometrie, 7th ed., pp. 289-312.
See Hilbert and Ackermann, Hilbert and Bernays.
H il b e r t , D a v id a n d A c k e r m a n n , W il h e l m
1928 . Grundzuge der theoretischen Logik. Berlin (Springer), viii-f 120 pp. 2nd ed.
1938, viii-f 133 pp. Reprinted New York (Dover Publications) 1946. 3rd ed.
Berlin, Gottingen, Heidelberg (Springer) 1949, viii-f 155 pp. Eng. tr. of the
2nd ed. by L. M. Hammond, G. G. Leckie and F. Steinhardt, ed. with notes by
R. E. Luce, Principles of mathematical logic, New York (Chelsea Pub. Co.)
1950, xii-f 172 pp.
H il b e r t , D a v id a n d B e r n a y s , P a u l
1934 . Grundlagen der Mathematik, vol. 1, Berlin (Springer), xii-f471 pp.
Reprinted Ann Arbor, Mich. (J. W. Edwards) 1944.
526 BIBLIOGRAPHY
1939 . Ibid., vol. 2, Berlin (Springer), xii-f498 pp. Reprinted Ann Arbor, Mich.
(Edwards) 1944.
I o n g h , J o h a n J. d e
1 9 4 8 . Restricted forms of intuitionistic mathematics. Proceedings of the Tenth
International Congress of Philosophy (Amsterdam, Aug. n - 1 8 , 1 9 4 8 ), Am
sterdam (North-Holland Pub. Co.) 1949, pp. 744-748 (fasc. 2).
J a n k o w s k i, S t a n is l a w
193 4 . On the rules of suppositions in formal logic. Studia logica, no, 1, Warsaw,
32 pp.
1 9 3 6 . Recherches sur le systim e de la logique intuitioniste. Actes du Congr£s
International de Philosophic Scientifique, VI Philosophic des mathematiques,
Actuality scientifiques et industrielles 393, Paris (Hermann & Cie.), pp. 58-61.
Jankowski does not give his proofs in detail; a reconstruction of the proofs is in
Gene Rose 1 9 5 2 Part I.
K a l m Ar , L A sz l 6
: - . Vber die Axiomatisierbarkeit des Aussagenkalkiils. Acta scientiarum
1934 5
mathematicarum (Szeged), vol. 7, pp. 222-243.
1 9 4 3 . Egyszeru pdlda eldonthetetlen aritmetikai probUmdra (Ein einfaches Beispiel
fur ein unentscheidbares arithmetisches Problem ). Matematikai es fizikai lapok,
vol. 50, pp. 1-23. Hungarian with German abstract. Kalmar takes, as his basis
z z
for elementary functions, the variables, 1, + , *, | a—b |, [afb], E , II , but
y= w y —w
remarks that then *and [afb ] are redundant. (Cf. Example 1 § 57 above.)
1 9 4 8 . On unsolvable mathematical problems . Proceedings of the Tenth Inter
national Congress of Philosophy (Amsterdam, Aug. 1 1 - 1 8 , 1 9 4 8 ), Amsterdam
(North-Holland Pub. Co.) 1949, pp. 756-758 (fasc. 2). Preprints 1948, pp. 534-
536.
1 9 5 0 . Eine einfache Konstruktion unentscheidbarer Satze in formalen Systemen.
Methodos, vol. 2, pp. 220-226; Eng. tr. by Ernst \. Glasersfeld, pp. 227-231.
1 9 5 0 a. Another proof of the Godel-Rosser incompletability theorem. Acta scien
tiarum mathematicarum (Szeged), vol. 1 2 , pp. 38-43.
K e m e n y , J o h n G.
1 9 4 8 . Review of M ostowski 1 9 4 7 a. Jour, symbolic logic, vol. 13, pp. 46-48.
K e t o n e n , O iv a
. Untersuchungen zum Prddikatenkalkul. Annales Academiae Scientiarum
19 4 4
Fennicae, ser. A, I. Mathematica-physica 23, Helsinki, 71 pp.
K l e e n e , S t e p h e n C.
193 4 . Proof by cases in formal logic. Ann. of math., 2 s., vol. 35, pp. 529-544.
Relative to § 20 above, cf. p. 534. The use of “ |-” to express derivability by the
rules of inference originated with Rosser; the modification to make |- relative
also to the axioms, with the author.
BIBLIOGRAPHY 527
1935 . A theory Amer. jour, math., vol. 57,
of po s i t i v e integers in formal logic.
pp. 153-173, 219-244.

1936 . Math. Ann., vol. 112, pp. 727-
G e n e r a l r e c ursive f u n c t i o n s of n a t u r a l n u m b e r s .
742. For an erratum and a simplification, cf. Jour, symbolic logic, vol. 3 p. 152,
vol. 2 p. 38 and vol. 4 top p. iv at end.
1936 a. X- definability Duke mathematical journal, vol. 2,
and recursiveness.
pp. 340-353.
1936 b. A Bull. Amer. Math. Soc., vol. 42, pp. 544-546.
note o n recursive functions.
1938 . O n Jour, symbolic logic, vol. 3, pp. 150-155.

n o t a t i o n for o r d i n a l n u m b e r s .
1943 . q u a n t i f i e r s . Transactions of the American Mathe

Rec u r s i v e predicates a n d
matical Society, vol. 53, pp. 41-73. Omit § 15, because the proof of Theorem 1
of 1944 contains an error. — Footnote (21) cites only a function which is partial
but not potentially recursive, though the text mentions predicates also. (This
oversight was brought to the author’s attention by J. C. E. Dekker, March 18,
1952.) For such a predicate, cf. Example 6 § 63 above.
1944 . O n the f o r m s of the p r e d i c a t e s inAmer.
the t h e o r y of c o n s t r u c t i v e ordinals.
jour, math., vol. 66, pp. 41-58. The stars should be om itted from (4) and (11)
p. 43 (cf. *86 and *95 above). — The treatm ent of the example in 8 is not P(a)
complete. For (18) is not simply another way of writing the inductive def
inition of ‘a is provable’, but can have other solutions for besides P(a) = P(a)
{a is provable) (e.g. = is a formula}). However for any solution of (18),

P(a) {a
{a is provable) -> ; and it is easily shown from (22) th at for the particular
P(a)
solution == P(a) ) -> is provable). — Similarly for all ap

( E x ) R ( a , x), P ( a {a
plications of the technique in which the particular solution contains only an

existential quantifier (cf. end § 53). But in the application to 8 Q in 14, where a
there is also a generality quantifier, the treatm ent cannot be completed in this
manner; and so Theorem 1 and the first half of Theorem 2 are not established.
The author plans to discuss the situation in a second paper under the same title.
1945 . O n the i n t e r p r e t a t i o n Jour, symbolic logic,
of intuitionistic num b e r theory.
vol. 10, pp. 109-124.

1948 . O n Proceedings of the Tenth International Congress
t h e i n t u i t i o n i s t i c logic.
of Philosophy (Amsterdam, Aug. 1 1 - 18 , 1948 ), Amsterdam (North-Holland

Pub. Co.) 1949, pp. 741-743 (fasc. 2). Preprints 1948, pp. 185-187.
1930 . A symmetric form Koninklijke Nederlandsche Akademie
of G o d e V s theorem.
van Wetenschappen, Proceedings of the section of sciences, vol. 53, pp. 800-802;
also Indagationes mathematicae, vol. 12, pp. 244-246.
1950 a. Recursive functions and Proceedings of the
intuitionistic mathematics.
International Congress of Mathematicians (Cambridge, Mass., U.S.A., Aug. 3 0 -

Sept. 6 , 1950), 1952, vol. 1, pp. 679-685.
1952 . P e r m u t a b i l i t y o f i n f e r e n c e s i n G e n t z e n ' s c a l c u l i L K and LJ. Memoirs
of the American Mathematical Society, no. 10, pp. 1-26.
See Church and Kleene.
K r e i s e l , G.
1950 . Note on arithmetic models for consistent f o r m u l a e of the predicate calculus.
Fundamenta mathematicae, vol. 37, pp. 265-285.

528 BIBLIOGRAPHY
K uznecov, A. V.
1950 . O prim itivno rdkursivnyh funkci&h bol'Sogo razmaha (On primitive recursive
functions of large oscillation). Doklady Akademii Nauk SSSR, n.s., vol. 71,
pp. 233-236.
L a n g f o r d , Co o per H a r o ld
1927 . On inductive relations. Bull. Amer. Math. Soc., vol. 33, pp. 599-607.
See Lewis and Langford.
L e w is , C l a r e n c e I r v in g
1912. Im plication and the algebra of logic. Mind, n.s., vol. 21, pp. 522-531.
1917 . The issues concerning material implication. The journal of philosophy,
psychology and scientific method, vol. 14, pp. 350-356.
See Lewis and Langford.
L e w is , C l a r e n c e I r v in g a n d L a n g f o r d , C o o p e r H a r o l d
1932 . Symbolic logic. New York arid London (The Century Co.), xi-j-506 pp.
Reprinted New York (Dover Publications) 1951.
L o w e n h e im , L e o p o l d
1915 . fiber Moglichkeiten im Relativkalkiil. Math. Ann., vol. 76, pp. 447-470.
L u k a s ie w ic z , J a n
1920. O logice trdjwartoiciowej (On three-valued logic). Ruch filozoficzny (Lw6w),
vol. 5, pp. 169-171.
1925. Dimonstration de la compatibilitd des axioms de la thdorie de la ddduction.
Annales de la Soci6t 6 Polonaise de Mathematique, vol. 3 (for 1924, pub. 1925),
p. 149.
See Lukasiewicz and Tarski.
L u k a s ie w ic z , J a n a n d T a r s k i, A l f r e d
1930 . Untersuchungen uber den Aussagenkalkul. Comptes rendus des seances de
la Societe des Sciences et des Lettres de Varsovie, Classe III, vol. 23, pp. 30-50.
M acL a n e , Sa u n d er s
1934 . Abgekiirzte Beweise im Logikkalkul. Dissertation Gottingen. 61 pp.
M a n n o u r y , G e r r it
1909 .. Methodologisches und Philosophisches zur Elementar-Mathematik. Haar
lem (P. Visser), viii+276 pp.
1925 . Mathesis en mystiek. Amsterdam. French translation: Les deux poles de
I'esprit. Paris, 1933.
1934. signifischen Grundlagen der Mathematik. Erkenntnis, vol. 4, pp. 288-
309, 317-345.
BIBLIOGRAPHY 529
Markov, A. A.
1947 . N dvozm oinosf ndkotoryh algorifmov v tdorii associativnyh sistdm. Doklady
Akademii Nauk SSSR, n.s., vol. 55, pp. 587-590. Eng. tr. On the im possibility of
certain algorithms in the theory of associative systems. Comptes rendus (Doklady)
de TAcademic des Science? de l’URSS, n.s., vol. 55, pp. 583-586.
1947a. O ndkotoryh ndrazrdUmyh probldmah kasaullihsd matric (On some un-
solvable problems concerning matrices). Doklady Akademii Nauk SSSR, n.s.,
vol. 57, pp. 539-542.
1947b. Ndvozmohiost ' ndkotoryh algorifmov v tdorii associativnyh sistdm I I (Im
possibility of certain algorithms in the theory of associative systems II). Ibid.,
vol. 58, pp. 353-356.
1947c. O prddstavlenii rdkursivnyh funkcij (On the representation of recursive
functions). Ibid., pp. 1891-1892.
1949. 0 prddstavldnii rdkursivnyh funkcij (On the representation of recursive
functions). Izv6stiy4 Akademii Nauk SSSR, ser. mat., vol. 13, pp. 417-424.
Eng. tr., On the representation of recursive functions, Amer. Math. Soc., trans
lation no. 54, lithoprinted, New York 1951, 13 pp.
1951 . Ndvozmoinosi' ndkotoryh algoritmov v tdorii associativnyh sistdm (Impos
sibility of certain algorithms in the theory of associative systems). Doklady
Akademii Nauk SSSR, n.s., vol. 77, pp. 19-20.
1951 a. Ndvozmotnost' algorifmov raspoznavanid ndkotoryh svojstv associativnyh
sistdm (Impossibility of algorithms for distinguishing certain properties of
associative systems). Ibid., pp. 953-956.
1951 b. Ob odnoj ndrazrdSimoj probldmd, kasaulcdjsd matric (An unsolvable problem
concerning matrices). Ibid., vol. 78, pp. 1089-1092.
M cK in s e y ,J. C. C.
1939 . Proof of the independence of the prim itive symbols of H ey ting* s calculus of
propositions. Jour, symbolic logic, vol. 4, pp. 155-158.
See McKinsey and Tarski.
M cK in s e y , J. C. C. a n d T a r s k i, A l f r e d
1948 . Some theorems about the sentential calculi of Lewis and H ey ting. Ibid., vol.
13, pp. 1-15.
M o s t o w s k i, A n d r z e j
1947 . On definable sets of positive integers. Fundamenta mathematicae, vol. 34,
pp. 81-112.
1947a. On absolute properties of relations. Jour, symbolic logic, vol. 12, pp. 33-42.
1948. Proofs of non-deducibility in intuitionistic functional calculus. Ibid., vol. 13,
pp. 204-207.
1948a. On a set of integers not definable by means of one-quantifier predicates.
Annales de la Societe Polonaise de Mathematique, vol. 21, pp, 114-119.
1949. A n undecidable arithmetical statement. Fund, math., vol. 36, pp. 143-164.
1951 . A classification of logical systems. Studia philosophica, vol. 4, pp. 237-274.
K leene
530 BIBLIOGRAPHY
1952 . Sentences undecidable in formalized arithmetic. An exposition of the
theory of Kurt Godel. Studies in logic and the foundations of mathematics,
Am sterdam (North-Holland Pub. Co., viii-f 117 pp.
a. Models of axiomatic systems. Fund. Math., vol. 39, pp. 133-158. Cf. Ryll-
1952
Nardzewski 1952 .
See Mostowski and Tarski.
M o s t o w s k i, A n d r z e j a n d T a r s k i, A l f r e d
1949 abstract. Undecidability in the arithmetic of integers and in the theory of rings.
Jour, symbolic logic, vol. 14, p. 76.
N e l s o n , D a v id
. Recursive functions and intuitionistic number theory. Trans. Amer. Math.
19 4 7
Soc., vol. 61, pp. 307-368.
1949 . Constructive falsity. Jour, symbolic logic, vol. 14, pp. 16-26.
N eum ann, John von

1925 . E ine Axiomatisierung der Mengenlehre. Journal fur die reine und ange-
wandte Mathematik, vol. 154, pp. 219-240. Berichtigung, ibid., vol. 155 (1926),
p. 128.
19 2 7 . Z u r Hilbertschen Beweistheorie. Math. Zeit., vol. 26, pp. 1-46.
1928 . D ie Axiomatisierung der Mengenlehre. Ibid., vol. 27, pp. 669-752.
1 9 3 1 - 2 . Die formalistische Grundlegung der Mathematik. Erkenntnis, vol. 2,
pp. 116-121.
1947 . The mathematician. The works of the mind, ed. by Robert B. Heywood,
Chicago (U. of Chicago Press), pp. 180-196.
P a s c h , M o r it z
1882 . Vorlesungen liber neuere Geometrie. Leipzig (Teubner), iv-f-201 pp.
Reprinted in Vorlesungen iiber neuere Geometrie by M. Pasch and Max Dehn,
Berlin (Springer) 1926, viii-f 275 pp.
P e a n o , G iu s e p p e
1889 . Arithmetices principia, nova methodo exposita.Turin (Bocca), xvi-f 20 pp.
18 91 . S u l concetto di numero. Rivista di matematica, vol. 1, pp. 87-102, 256-267.
Peano formulates his axioms for the positive integers. (In fact, some writers
call these the “natural num bers”.)
1894 - 1908 . Formulaire de mathematiques. Introduction and five volumes. Turin.
Edited by Peano and written by him in collaboration with Rodolfo Bettazzi,
Cesare Burali-Forti, F. Castellano, Gino Fano, Francesco Giudice, Giovanni
Vailati, Giulio Vivanti.
P e ir c e , C h a r l e s S a n d e r s
1867 . On an improvement in Boole’s calculus of logic (presented 12 March 1867).
Proceedings of the American Academy of Arts and Sciences, vol. 7 (1865-8),
pp. 250-261. Reprinted in Collected papers of Charles Sanders Peirce, ed. by
BIBLIOGRAPHY 531
Charles Hartshorne & Paul Weiss, Cambridge, Mass. (Harvard University
Press), vol. 3 (1933), pp. 3-15.
1880 . On the algebra of logic. Chapter I. — Syllogistic. Chapter II. — The logic
of non-relative terms. Chapter I I I . — The logic of relatives. Amer. jour, math.,
vol. 3, pp. 15-57. Reprinted with corrections in Collected papers, vol. 3, pp.
104-157.
P e t e r , R 6 zsa
1934 . Vber den Zusammenhang der verschiedenen Begriffe der rekursiven Funktion.
Math. Ann., vol. 110 , pp. 612-632.
1935 . Konstruktion nichtrekursiver Funktionen. Ibid., vol. Ill, pp. 42-60.
1935 a. A rekurziv fuggvdnyek elmdletdhez (Zur Theorie der rekursiven Funktionen).
Hungarian with full German abstract. Matematikai 6s fizikai lapok, vol. 42,
pp. 25-49.
1936. Vber die mehrfache Rekursion. Math. Ann., vol. 113, pp. 489-527.
1940. Review of Skolem 1939. Jour, symbolic Logic, vol. 5, pp. 34-35.
1950. Zusammenhang der mehrfachen und transfiniten Rekursionen. Ibid., vol. 15,
pp. 248-272.
1951 . Rekursive Funktionen. Akaddmiai Kiad 6 (Akademischer Verlag) Budapest,
206 pp.
P o in c a r e , H e n r i
1900. Du role de Vintuition et de la logique en mathematiques. Compte rendu du
Deuxieme Congres International des Mathematiciens, tenu a Paris du 6 au 12
aout 1900, Paris (Gauthier-Villars) 1902, pp. 115-130.
1902 . La science et l’hypothese. Paris, 284 pp. Translated by G. Bruce Halstead
as pp. 27-197 of The foundations of science by H. Poincar 6, New York (The
Science Press) 1913; reprinted 1929.
1905 -6 . Les mathematiques et la logique. Revue de metaphysique et de morale,
vol. 13 (1905), pp. 815-835, vol. 14 (1906), pp. 17-34, 294-317. Reprinted in
1908 with substantial alterations and additions.
1908 . Science et methode, Paris, 311 pp. Translated by Halstead as pp. 359-546
of The foundations of science, New York 1913, reprinted 1929.
P o s t , E m il L.
1921. Introduction to a general theory of elementary propositions. Amer. jour,
math., vol. 43, pp. 163-185.
1936 . Finite combinatory processes— formulation 1. Jour, symbolic logic, vol. 1,
pp. 103-105.
1943 . Formal reductions of the general combinatorial decision problem. Amer.
jour, math., vol. 65, pp. 197-215.
1944 . Recursively enumerable sets of positive integers and their decision problems.
Bull. Amer. Math. Soc., vol. 50, pp. 284-316.
1946 . A variant of a recursively unsolvable problem. Ibid., vol. 52, pp. 264-268.
1946 a. Note on a conjecture of Skolem. Jour, symbolic logic, vol. 11, pp. 73-74.
532 BIBLIOGRAPHY
1947 . Recursive unsolvability of a problem of Thue. Ibid., vol. 12, pp. 1-11.
19 4 8 abstract. Degrees of recursive unsolvability. Preliminary report. Bull. Amer.
Math. Soc. vol. 54, pp. 641-642.
P r a n tl, Carl
18 55. Geschichte der Logik im Abendlande, vol. 1 , Leipzig (S. Hirzel), xii-f734
pp. (O th e r v o lu m e s 1861, 1867, 1870.) Reprinted 1927.
P resbu r g er , M.
193 0 . Vber die Vollstandigkeit eines gewissen Systems der A rithmetik ganzer Zahlen,
in welchem die Addition als einzige Operation hervortritt. Sprawozdanie z I Kon-
gresu Matematykdw Kraj6 w Sfowia&skich (Comptes-rendus du I Congr&s des
MatMmaticiens des Pays Slaves), Warszawa 1 9 2 9 , Warsaw 1930, pp. 92-101,
395.
Q u in e , W il l a r d V an O rm an
1940 . Mathematical logic. New York (Norton), xiii+348 pp. See Rosser 19 4 2
and Quine 1 9 4 1 , concerning the fact that the Burali-Forti paradox arises in
the system of this book (although Cantor’s paradox apparently is avoided),
as was discovered by Rosser and by Roger C. Lyndon. Revised ed.,
Harvard University Press, 1951, xii-f-346 pp.
1 9 4 1 . Element and number. Jour, symbolic logic, vol. 6 , pp. 135-149.
See Church and Quine.

R a m s e y , F. P.
1 9 2 6 . The foundations of mathematics. Proc. London Math. Soc., ser. 2, vol. 25,
pp. 338-384. Reprinted as pp. 1-61 in The foundations of mathematics and

other logical essays by F. P. Ramsey, ed. by R. B. Braithwaite, London (Kegan
Paul, Trench, Trubner) and New York (Harcourt, Brace) 1931. The latter re
printed London (Routledge and Kegan Paul) and New York (Humanities
Press) 1950.
R a s i o w a , H. a n d S i k o r s k i , R .
1 9 5 0 . A proof of the completeness theorem of Gddel. Fund, math., vol. 37, pp. 193-
200. For a simplification by Tarski, cf. Jour, symbolic logic., vol. 17, p. 72.
R ic h a r d , J u l e s
190 5 . Les principes des mathematiques et le probUme des ensembles. Revue ge
nerate des sciences pures et appliquees, vol. 16, pp. 541-543. Also in Acta
mathematica, vol. 30 (1906), pp. 295-296.
R o b in s o n , J u l ia
194 9 abstract. Undecidability in the arithmetic of integers and rationals and in the
theory of fields. Jour, symbolic logic, vol. 14, p. 77.
1 9 4 9 . Definability and decision problems in arithmetic. Ibid., pp. 98-114. For § 48
above, her treatment for the positive integers can be adapted to the natural
numbers.
1 9 5 0 . General recursive functions. Proceedings of the American Mathematical
Society, vol. 1 , pp. 703-718.
BIBLIOGRAPHY 533
R o b in s o n , R a p h a e l M.
1947 . Prim itive recursive functions. Bull. Amer. Math. Soc., vol. 53, pp. 925-942.
1948 . Recursion and double recursion. Ibid., vol. 54, pp. 987-993.
1949 abstract. Undecidable rings. Ibid., vol. 55, p. 1050.
1950 abstract. A n essentially undecidable axiom system. Proceedings of the In
ternational Congress of Mathematicians (Cambridge, Mass., U.S.A., Aug.
30-Sept. 6, 1950 ), 1952, vol. 1, pp. 729-730. Robinson's system is simpler than one
we were using previously (since Mostowski and Tarski 1949 abstract) in §§41,
49 and 76 for the same purpose.
R o s e , G e n e F.
1952 . Jankowski’s truth-tables and realizability. Doctoral dissertation, The
University of Wisconsin.
R o s s e r , B a r k l e y (Rosser, J. B.; Rosser, J. Barkley)
1935. A mathematical logic without variables. Ann. math., 2 s., vol. 36, pp. 127-150
and Duke math, jour., vol. 1, pp. 328-355. Relative to § 20 above, cf. p. 130,
p. 329, and the note accompanying Kleene 1934.
1936. Extensions of some theorems of Gddel and Church. Jour, symbolic logic,
vol. 1, pp. 87-91.
1936a. Review of Gddel 1936. Ibid., vol. 1, p. 116.
1939. Bn the consistency of Quine's “New foundations for mathematical logic".
Ibid., vol. 4, pp. 15-24.
1942. The Burali-Forti paradox. Ibid., vol. 7, pp. 1-17.
1942a. New sets of postulates for combinatory logics. Ibid., pp. 18-27. For a cor
rection, cf. Curry 1948-9.
See Rosser a n d Turquette, Rosser and Wang.
R o s s e r , J . B. a n d T u r q u e t t e , A . R .
1945. Axiom schemes for m-valued propositional calculi. Jour, symbolic logic,
vol. 10, pp. 61-82. Cf. 1950.
1 9 4 8 -5 1 . Axiom schemes for m-valued functional calculi of first order. P art I.
Definition of axiom schemes and proof of plausibility. Ibid., vol. 13 (1948), pp.
177-192. P art II. Deductive completeness. Ibid , vol. 16 (1951), pp. 22-34.
1949. A note on the deductive completeness of m-valued propositional calculi. Ibid.,
vol. 14, pp. 219-225.
1952 . Many-valued logics. Studies in logic and the foundations of mathematics,
Amsterdam (North-Holland Pub. Co.), vii-f 124 pp.
R o sser, J. B ark ley and W ang, H ao
1950 . Non-standard models for formal logics. Jour, symbolic logic, vol. 15, pp.
113-129.
R u s s e l l , B e r t r a n d (Russell, B. A. W.)
1902. On finite and infinite cardinal numbers (Section III of A. X. Whitehead’s
O 11 cardinal numbers). Amer. jour, math., vol. 24, pp. 378-383.
534 BIBLIOGRAPHY
- . The Russell paradox appears in Frege 1903, in a postscript (dated by
1902 3
Frege October 1902), pp. 253-265. Concerning Zermelo’s independent discovery
of this paradox, see Zermelo 1908a p. 119 and Hilbert 1926 p. 169.
1906 . Les paradoxes de la logique. Revue de metaphysique et de morale, vol. 14,
pp. 627-650.
1908 . Mathematical logic as based on the theory types. Amer. jour, math., vol. 30,
pp. 222-262.
1910 . La theorie des types logiques. Rev. metaph. mor., vol. 18, pp. 263-301.
1 9 1 9 . Introduction to mathematical philosophy. London (G. Allen and Unwin)
and New York (Macmillan), viii-f 208 pp.. 2nd ed. 1920.
See W hitehead and Russell.
R usto w , A lex a n d er
1910 . Der Liigner, Theorie, Geschichte und Auflosung. Leipzig (Teubner),
v -f 147 pp.
R y l l -N a r d z e w s k i, Cz e sl a w
1952 . The role of the axiom of induction in elementary arithmetic. Fund. Math.,
vol. 39, pp. 239-263. Subsequently Mostowski obtained further results, 1952a.
S c h m id t , A r n o l d
1938 . Uber deduktive Theorien mit mehreren Sorten von Grunddingen. Math. Ann.,
vol. 115, pp. 485-506.
SCHONFINKEL, MOSES
1924 . Uber die Bausteine der mathematischen Logik. Math. Ann., vol. 92, pp.
305-316.
S c h r o d e r , E r n st
18 77 . Der Operationskreis des Logikkalkuls. Leipzig, v -f 37 pp.
-
189 0 19 0 5 . Vorlesungen liber die Algebra der Logik (exakte Logik). Vol. 1,
Leipzig (Teubner) 1890, xii-f 717 pp. Vol. 2 part 1, Leipzig 1891, xiii+400 pp.
Vol. 3 Algebra und Logik der Relative part 1, Leipzig 1895, viii-f 649 pp. Vol. 2
part 2 appeared posthumously, ed. by Eugen Muller, Leipzig 1905, xxix-f 205
pp. Abriss der Algebra der Logik, ed. by Muller, part 1 Elementarlehre Leipzig
and Berlin 1909, v -f 50 pp., part 2 Aussagentheorie, Funktionen, Gleichungen
und Ungleichungen, Leipzig and Berlin 1910, v i+ 5 1 -f 159 pp.
Schutte, K urt
1951 . Beweistheoretische Erfassung der unendlichen Induktion in der Zahlentheorie.
Math. Ann., vol. 122 , pp. 369-389.
Sheffer, H. M .
19 13 .
A set of five independent postulates for Boolean algebras, with application to
Trans. Amer. Math. Soc., vol. 14, pp. 481-488. According to
logical constants.
Quine 1940 p. 49, the definability of &, V and -1 in terms of one operator was
known to C. S. Peirce in 1880. (Cf. § 30 above.)
BIBLIOGRAPHY 535
S kolem , T horalf
19 19. Untersuchungen iiber die Axiome des Klassenkalkuls und uber Produk-
tations- und Summationsprobleme, welche gewisse Klassen von A ussagen be-
treffen. Skrifter utgit av Videnskapsselskapet i Kristiania, I. Matematisk-natur-
videnskabelig klasse 1919 , no. 3, 37 pp.
1920. Logisch-kombinatorische U ntersuchungen iiber die Erfullbarkeit oder Beweis-
barkeit mathematischer Satze nebst einem Theoreme uber dichte Mengen. Ibid.,
1920 , no. 4, 36 pp.
1922 - 3 . Einige B enter kungen zur axiomatischen Begriindung der Mengenlehre.
Wissenschaftliche Vortrage gehalten auf dem Fiinften Kongress der Skandi-
navischen Mathematiker in Helsingfors vom 4. bis 7 . Juli 1922, Helsingfors
1923, pp. 217-232.
1923. Begriindung der elementaren Arithmetik durch die rekurrierende Denkweise
ohne Anwendung scheinbarer V eranderlichen m il unendlichem Ausdehnungs-
bereich. Skrifter utgit av Videnskapsselskapet i Kristiania, I. Matematisk-
naturvidenskabelig klasse 1923 , no. 6 , 38 pp.
1929 . Uber einige Grundlagenfragen der M athematik. Skrifter utgitt av Det Norske
Videnskaps-Akademi i Oslo, I. Matematisk-naturvidenskapelig klasse 1929 ,
no. 4, 49 pp.
1929 - 30 . Uber die Grundlagendiskussionen in der M athematik. Den Syvende
Skandinaviske Matematikerkongress i Oslo 19-22 August 1929, Oslo (Broggers)
1930, pp. 3-21.
1930 - 1 . Uber einige Satzfunktionen in der Arithmetik. Skrifter utgitt av Det
Norske Videnskaps-Akademi i Oslo, I. Matematisk-naturvidenskapelig klasse
1930 , no. 7, 28 pp. (1931).
1933 . Uber die Unmoglichkeit einer vollstdndigen Charakterisierung der Zahlen-
reihe mittels eines endlichen Axiomensystems. Norsk matematisk forenings
skrifter, ser. 2, no. 10, pp. 73-82.
1934. Uber die Nicht-charakterisierbarkeit der Zahlenreihe mittels endlich oder
abzahlbar unendlich vieler Aussagen m it ausschliesslich Zahlenvariablen. Fund,
math., vol. 23, pp. 150-161.
- . Uber die Zuruckfuhrbarkeit einiger durch Rekursionen definierten Re-
1936 7
lationen auf ‘<arithmetische,>. Acta litterarum ac scientiarum Regiae Universita-
tis Hungaricae Franscisco-Iosephinae, Sectio scientiarum mathematicarum
(Szeged), vol. 8, pp. 73-88.
1938 . Sur la portee du theoreme de Lowenheim-Skolem. Les entretiens de Zurich
sur les fondements et la methode des sciences mathematiques, 6-9 Decembre
1938 , Exposes et discussions, pub. by F. Gonseth, Zurich (Leemann) 1941,
pp. 25-47. Discussion on pp. 47-52.
1939 . Eine Bemerkung iiber die Induktionsschemata in der rekursiven Zahlen -
theorie. Monatshefte Math. Phys., vol. 48, pp. 268-276.
1944 . Some remarks on recursive arithmetic. Det Kongelige Norske Videns-
kabers Selskab, Forhandlinger, vol. 17, pp. 103-106. This is the second of a
series of four notes, the others of which appear in the same volume, pp. 89-92,
pp. 107-109, pp. 126-129.
19 5 1 . Review of Rosser and Wang 1950 . Jour. Symbolic logic, vol. 16, pp. 145-146.
536 BIBLIOGRAPHY
T a r s k i, A l f r e d
1930 . Uber einige fundamentalen Begriffe der M etamathematik . Comptes rendus
des seances de la Societe des Sciences et des Lettres de Varsovie, Classe III, vol.
23, pp. 22-29.
1932 . Der Wahrheitsbegriff in den Sprachen der deduktiven D isziplinen. Akademie
der Wissenschaften in Wien, Mathematisch-naturwissenschaftliche Klasse,
Anzeiger, vol. 69, pp. 23-25. A prospectus for 1933 .
1933 . Der Wahrheitsbegriff in den formalisierten- Sprachen. Studia philosophica,
vol. 1 (1936), pp. 261-405 (offprints dated 1935). Tr. by L. Blaustein from the
Polish original 1933, with a postscript added.
1933 a. Einige Betrachtungen uber die Begriffe der a-W iderspruchsfreiheit und der
<0- V oilstandi gkeit. Monatshefte Math. Phys., vol. 40, pp. 97-112.
1949 abstract. On essential undecidability . Jour, symbolic logic, vol. 14, pp. 75-76.
1949 a abstract. Undecidability of group theory. Ibid., pp. 76-77.
1949 b abstract. Undecidability of the theories of lattices and projective geometries.
Ibid., pp. 77-78.
See Lukasiewicz and Tarski, Me Kinsey and Tarski, Mostowski and Tarski.
T h u e, A xel
19 14 . Problems uber Verdnderungen von Zeichenreihen nach gegebenen Regeln.
Skrifter utgit av Videnskapsselskapet i Kristiania, I. Matematisk-naturvidens-
kabelig klasse 1 9 14 , no. 10, 34 pp.
T rahtenbrot, B. A.
1950 . NdvozmoZnost' algorifma did problimy razriUmosti na konilnyh klassah
(Impossibility of an algorithm for the decision problen in finite classes).
Doklady Akademii Nauk SSSR, n.s., vol. 70, pp. 569-572.
T u r in g , A l a n M a t h is o n
1936 7 - . On computable numbers, with an application to the Entscheidungsproblem.
Proc. London Math. Soc., ser. 2, vol. 42 (1936-7), pp. 230-265. A correction ,
ibid., vol. 43 (1937), pp. 544-546.
1 9 3 7 . Computability and X-definability. Jour, symbolic logic, vol. 2, pp. 153-163.
1 9 3 9 . Systems of logic based on ordinals. Proc. London Math. Soc., ser. 2, vol. 45,
pp. 161-228.
1950 . The word problem in semi-groups with cancellation. Ann. of math., 2 s., vol.
52, pp. 491-505. Some points in the proof require clarification, which can be
given, as pointed out by Boone 1952 .
V a n d i v e r , H. S.
1946 . Fermat*s last theorem. Its history and the nature of the known results con* '
cerning it. Amer. math, monthly, vol. 53, pp. 555-578.
V e b l e n , O sw ald
190 4 . A system of axioms for geometry. Trans. Amer. Math. Soc., vol. 5, pp. 343-384.
See Veblen and Bussey.
BIBLIOGRAPHY 537
V e b l e n , O sw a ld and B u s se y ,W. H .
1 9 0 6 . Finite projective geometries. Ibid., vol. 7, pp. 241-259.
W a jsbe r g , M ordechaj
1938. Untersuchungen uber den Aussagenkalkul von A . Heyting. WiadomoSci
matematyczne, vol. 46, pp. 45-101.
W ang, H ao
1952 . Logic of many-sorted theories. Jour, symbolic logic, vol. 17, pp. 105-116.
1 9 5 3 . Certain predicates defined by induction schemata. Ibid., vol. 18, pp. 49-59.
See Rosser and Wang.

W eyl, H erm ann
1918 . Das Kontinuum. Kritische Untersuchungen iiber die Grundlagen der
Analysis. Leipzig (Gruyter), iv+84 pp. Reprinted 1932.
1 9 1 9 . Der circuius vitiosus in der heutigen Begriindung der Analysis. Jahres-
bericht der Deutschen Mathematiker-Vereinigung, vol. 28, pp. 85-92.
1 9 2 6 . Die heutige Erkenntnislage in der Mathematik. Sonderdrucke des Sympo-
sion, Erlangen (im Weldkreis-Verlag), Heft 3 (1926), 32 pp. Also in Symposion
(Berlin), vol. 1 (1925-7), pp. 1-32.
1 9 2 8 . Diskussionsbemerkungen zu dem zweiten Hilbertschen Vortrag iiber die
Grundlagen der M athematik. Abhandlungen aus dem Mathematischen Seminar
der Hamburgischen Universitat, vol. 6 , pp. 86 -8 8 .
1 9 3 1 , Die Stufen des Unendlichen. Jena (Fischer), 19 pp.
1 9 4 4 . D avid Hilbert and his mathematical work. Bull. Amer. Math. Soc., vol. 50,
pp. 612-654.
1 9 4 6 . Mathematics and logic. A brief survey serving as a preface to a review of “ The
Philosophy of Bertrand Russell”. Amer. math, monthly, vol. 53, pp. 2-13.
1 9 4 9 . Philosophy of mathematics and natural science. Princeton, N.J. (Princeton
University Press), x + 3 1 1 pp. Revised and augmented Eng. ed., based on a tr.
by Olaf Helmer from the German original 1926.
W h it e h e a d , A lfred N orth and R u sse ll, B ertrand
- . Principia mathematica. Vol. 1 1910, xv + 6 6 6 pp. (2 nd ed. 1925). Vol.
1910 13
2 1912, xxiv+772 pp. (2nd ed. 1927). Vol. 3 1913, x + 491 pp. (2nd ed. 1927).
Cambridge, England (University Press).
Y oung, John W esley
1911 . Lectures on fundamental concepts of algebra and geometry. New York
(Macmillan), vii-f-247 pp.
Zer m elo , E r n st
1904 . Beweis , dass jede Menge wohlgeordnet werden kann. Math. Ann., vol. 59,
pp. 514-516- Also cf. 1 9 0 8 a.
1 9 0 8 a. Neuer Beweis fiir die Moglichkeit einer Wohlordnung. Ibid., vol. 65, pp.
107-128.
1 9 0 8 . Untersuchungen uber die Grundlagen der Mengenlehre I. Ibid., pp. 261-281.
SYMBOLS AND NOTATIONS
T heory of sets
1-1 3 = 9, 13 - 10 so 478
9 < 10, 13 0 9 CO 476
c 9 > 10 0 12 «0 13
s 9 < 12 + 1 12 $ 16
$ 9 + 10, 16 {a, b, ...} 9 © 16
) 9 * 10, 16 2 M 15 U 15
Math em atical logic an d formal num ber th e o r y
h 87 = 70, 399 0 70 c 443, 444

|_ X j...X n 98, 102 75 0 101, 141 c 176
441, 443 /“•s > ^ 247 ab 183 E? 266, 326
113 < 75, 154 {x}. 247 Eq 399, 403
D 69 >, <, > 187 A(x), B(x), . . 78 f 126
& 69 < 247 a,b, ... 70 V 411
V 69 1 191 <3, S ... 108, 142 I 443
~i 69 + 70 91, © n , . 252, 277 N 500
V 69, 151 • 70 f >g > • • • 263 Pr 191
3 70 / 70 x,y, . . . 195 T 443, 444
3! 199 I 247, 276 A V(P) 207 t 126
R ecursive functions and informal number theory
-> 225, 334 0/ 19,217 34 h 230
& 225, 334 19,217 hxx ... xn 344 pd 223
V 225, 334 + 222 \w 225, 279, 329 Pr 230
_
225, 334 * 222 vy 347 Ri, R» 221
(y) 225, 336 ! 222 n 224 rm 202, 223
(Ey) 225, 336 223 2 224 S 220
(Ely) 225 | a~b | 223 340 °ncm 342
== 225, 334 1 202, 230 F, G, .. . 234 cn 220
rw 328 [alb] 202, 223 c 283 sg, sg 223
= 20, 227, 327 a*b nn
230 Uq 220 t 226, 332, 335
327 ab 222 exp 222 T n> T'b
* n 281,291
< 21, 29, 229 (#) i 230 f 226, 332, 335 u 278, 288
<, > 12, 29 {z}(xv ...,x„) 340 lh 230 u 326, 332, 335
<P 231,291 p(c, d, i) 240 M(a, k) 287 u? 220
A
307 ey 317 min, max 223 w 0,w ± 308
Notations relating to Turing machines 357, 358, 362, 363, 367
538
INDEX
abbreviations 75, 154, 406. atomic act cf. act.
Abel, N. H. 30. attached variables 142, 143.
abstract set theory 9, 16, 36, 40. autonymous: symbol etc. 71, 250, 265.
abstract system 25. auxiliary function letter 266.
Ackermann, W. 53, 204, 271, 273, 440, axiom 26, cf. axiomatic; — of a formal
463, 472, 478; Hilbert— 69, 135, 174, system 81; — schema 81, 140.
177, 180, 389. axiomatic: arithmetic 427, cf. Peano;
act 356, 357, 362, 366, 377, 379. method 26, 28, 53, 60, 421; set
active : situation, state etc. 357. theory 40, 45, 425; theories 421, 436,
actual infinity 48, 52, 55, 175, 317. 463, cf. formal system.
addition cf. sum.
algebraic: equation, number 5. Bachet 48.
algorithm 136, 301, 317, 322, 323, 324, barber, paradox of the 37.
325, 332, 333, 336, 347, cf. Church’s basis (induction) 2 2 , (inductive defi
thesis, decision problem, nition) 259, (primitive recursive func
alphabet 69, 382. tions) 223, 238, (recursion) 232, (re
alternative denial 139. cursively enumerable set) 346.
ambiguous axioms 27, 29, 430. Behmann, H. 436.
ambiguous value 33, 227. Beltrami, E. 54.
analysis 29, 32, 42, 44, 52, 54, 58, 479; Bereczki, I. 286, 287.
— of a deduction etc. 87, 234. Bernays, P. 40, 53, 63, 69, 425, 437, 463,
analytic number theory 30, 57, 58. 475, 495, 498; Hilbert— 52, 54, 58,
anonymous: variable etc. 154, 156, 160. 63, 69, 98, 107, 175, 176, 178, 179, 204,
antecedent 441; — rules 443. 211, 212, 225, 245, 246, 272, 321, 389,
application 29, 72, 83. 394, 397, 403, 407, 415, 416, 424, 435,
applied predicate calculus etc. cf. pure, 436, 463, 472, 475, 476, 478, 498, 501.
argument of a function 32. Bernstein, F. 11.
Aristotle 46, 55, 61. Berry’s paradox 39.
arithmetic 29, 32, 126, 186, 239; funda Black, M. 46.
mental theorem of 230; generalized blank square (Turing machine) 357.
246, 259, 276; cf. number theory, Bolyai, J. 28.
arithmetical: predicate 239, 241, 285, Boole, G. 61.
291, 415; class 431; cf. non— . Boone, W. W. 386, 536.
arithmetization: of analysis 30, 52; of bound, upper etc. 31; — term 410;
metamathematics 246, 276, 286. — variable 76, 153; cf. bounded, free,
assignment, satisfying 389. bounded quantifiers 197, 202, 225, 228.
associative laws 118, 186. 285, 329, 336, 471, 516.
assumed functions 224, cf. functional, branch 107; principal 267.
assumption formula 87, 181. Brouwer, L. E. J. 43, 46, 53, 56, 57, 58,
asymmetry 188. 318, 491, 498, 509
539
540 INDEX
Burali-Forti’s paradox 36, 40, 45, 532. completed infinity cf. actual,
Bussey, W. H. 54. completely defined function etc. 326.
completeness (concepts) 131, 135, 176,
calculable: cf. effectively, reckonable. 194, 212, 266, 302, 304, 309, 326, 422,
calculation: problem 137, 314, 316, cf. (results) 132, 134, 135, 139, 204, 295,
Church's thesis; procedure cf. algo 304, 389, 393, 395, 397, 400, 422, 424,
rithm. 474, cf. incompleteness,
canonical system (Post) 320. computable function 320, 321, 360, 362;
Cantor, G. 3, 6, 9, 15, 16, 30, 36, 40, 46, 1/1, wjs 361; cf. Turing,
48, 477; —'s diagonal method 6, 14, concatenation 71.
205, 207, 272, 282, 320, 341, 382; —'s conclusion (deduction) 87, 107, (im
theorem 15, 425; —'s paradox 36, plication) 113, 138, (inference) 83.
37, 40, 42, 45, 532. condition (Turing machine) 357.
cardinal number 3, 9, 44. conditional: equation 149; interpreta
Carnap, R. 45, 63, 83, 250. tion 150.
cases, definition by 229; induction 186; configuration (Turing machine) 357.
proof by 98, 100, 182, 190. congruent formulas 153.
categorical axioms 27, 430, 432. conjunction (interpretation) 51, 69, 138,
Cauchy, A. L. 30. 503, (formal logic) 69, 113, 118, 122,
Cayley, A. 54. 134, 166, 177, cf. introduction, prop
Cervantes 40. ositional connectives, truth tables,
chain: of equivalences etc. 117; in conjunctive normal form 135.
ference 113, 114. connexity 188.
change of variables 105, 153, 161. consequence 82, cf. deducibility,
choice, axiom of (Zermelo) 52, 394. consistency (concepts) 53, 124, 131, 207,
Church, A. 34, 98, 180, 241, 279, 300, 212, 266, 297, 302, 304, 305, 309, 314,
301, 319, 320, 322, 324, 325, 382, 432, 326, 391, 422, 440, 475, 479, 498, (re
434, 436, 517; —'s theorem 301, 315, sults) 128, 129, 172, 174, 176, 204,
382; —'s thesis 300, 313, 314, 316, 304, 422, 425, 466, 470, 472, 474, 497,
317, 331, 352, 354, 376, 381, 382, 509, 500, 508, 514, cf. Godel’s second
converse of 300, 319. theorem; — in a system 391; —
class cf. set. theorem 463, 466, 469, 475.
classical vs. intuitionistic (logic, mathe consistent interpretability 437.
matics, methods) 46, 48, 51, 53, 56, 63, constant 150; — function 220, 223;
318, (formal logic, formal systems) logical, non-logical 436; variable held
101, 140, 442, 480, 492, 501, cf. in 96, 102, 140, 148, 179.
tuitionistic. constructive: continuum 43; existence
closed formula 151. proofs 49, 52, 501, 509; method 26,
closure 151. 53; infinity cf. potential; ordinals
cognate sequents 480. 325; cf. intuitionistic.
combinatory definability 321. continuous functions 18, 164.
commutative laws 118, 186. continuum 17, 30; constructive 43; —
comparability theorem 16. problem 16, 41 (Godel); cf. analysis,
compatible formal systems 437. contraction (quantifiers) 285, (Gentzen)
complement of a set 10. 443, 444.
complete: equality, equivalence 328, contradiction 119, cf. consistency,
331, 335, 340; formal system cf. contraposition 113.
completeness; — set (Post) 343. contributory deduction 267.
INDEX 541
convergence 32, 164. dictionary (semi-group) 383.
correctness cf. consistency, difference (numbers) 223, (sets) 10.
correspondence 3, 9, 32, 255, 426. Diophantus 48.
countable set 3, cf. enumerable, direct: clause 20, 259; rule 86, 94.
course-of-values: function 231, 291; in discharge 94, 181.
duction 22,193; recursion 231,237,286. disjoint sets 10.
creative set (Post) 346. disjunction (interpretation) 50, 69, 138,
crocodile, dilemma of the 39. 225, 503, (formal logic) 69, 113, 118,
Curry, H. B. 154, 321, 459, 460, 492. 122, 134, 166, 177, 190, cf. intro
cut (Gentzen) 443, 453; cf. Dedekind. duction, propositional connectives,
truth tables.
Dantzig. D. van 38, 497. disjunctive normal form 134, 136.
Davis, M. 315. distributive laws 118, 186.
decidable: closed formula 194, 201; cf. divisibility 191, 202, 230, 241.
effectively, numeralwise, resolvable, Dixon, A. C. 38.
un—. domain 9, 32; (system, valuation) 24,
decimal notation 6, 19, 30, 381. 28, 168, 172, 174, 389, 400, 427, 463.
decision problem (concepts) 136, 137, double: basis 193, 232; negation 99,
301, 313, 314, 316, 382, 435, 437, 119, cf. intuitionistic formal systems;
(results) 137, 204, 301, 313, 314, 383, recursion 273.
386, 407, 423, 432, 438, 439, 474, dual: fraction etc. 16, 30, 361, 381;
482; special cases of the 435; cf. -converse 124, 167; cf. duality,
Church’s thesis, degree, reducibility. duality (logic) 122, 123, 167, 285, 443,
decision procedure cf. algorithm. (projective geometry) 56.
Dedekind, R. 14, 30, 43, 46, 48, 242;
— cuts 30, 36, 43, 52. effective interpretation 464.
deducibility 87, 88, 264, cf. deduction, effectively: calculable function, decid
deduction (formal) 87, 94, 106, 249, 264, able predicate 299, 300, cf. Church’s
278, 290; given, resulting 94; —- thesis, decision problem; enumerable
schema 87; — theorem 90, 97, 98, set of formulas 398; true formula 465.
140, 179; cf. variation, Einstein, A. 57.
deductive: method 59, 60; rules 80. element of a set 9.
defined: notions 28; value 326. elementary: axiom system 421, 424;
degree of unsolvability 315, 316, 343, axioms 421; conjunction etc. 134;
386, 435. function (Kalmar) 285; number theory
Dekker, J. C. E. 527. 30, 51, 69; predicate 284, 291, (Kal
De Morgan, A. 61, 119. mar) 285.
denumerable set 3, cf. enumerable, eliminability 405, 419, 510, 515.
dependence (deduction) 95, 102, (func elimination: laws 118; of logical sym
tion) 220, 224, 354. bols 98, cf. introduction; relations
depth 115, 151, 185. 405, cf. eliminability; theorem 453,
derived rules 86, 407, cf. introduction. 460, cf. Gentzen.
Descartes, R. 17, 54. empty: conjunction etc. 441; expression
descending induction 193, 478. 70, 78, 382; range 226; set 9.
description 199; eliminability of 407; endformula etc. 87, 107, 264.
indefinite 346. entity 247, 251, 259, 276.
designation 71, 250, 251. enumerable set 3, 427, cf. enumeration,
diagonal method cf. Cantor. recursively.
542 INDEX
enumeration 4, 426; effective 398; — finite: axiomatizability 436, cf. 427;
theorem 281, 341. cardinal 12, 44; domain 168, 178, 400,
Epimenides' paradox 39, 42, 45, 205, 501. 435, 464; extension 437; sequence 5,
equality (notions) 9, 13, 20, 127, 172, 70; set 12, 14; cf. finitary.
247, 328, (formal logic) 183, 197, 399, Finsler, P. 212.
409, 418, 424, (recursiveness) 227, f-less transform 411, 417, 419.
329; axioms for 399, 403, 424; — formal axiomatics 28, 41, 53, 60, 421.
and predicate letter formula 399. formal calculation 194, 241, 262, 295,
equation 264, 277; system of —’s 264, 322.
276, 277. formal deduction 87.
equivalence (notions) 9, 113, 225, 320, formal expression 70.
328, 334, 383, 386, 401, 406, (formal formal implication 138.
logic) 114, 116, 151, 161, (recursive formal induction 181, cf. induction.
ness) 228, 329, 337; — classes 9, formal inference 83.
401; — relation 401, 424; — theorem formal mathematics 53, 62.
11. formal objects 62, 64, 70, 85, 249, 251,
erasure (Turing machine) 358. 265.
essential undecidability 437. formal proof 65, 83, 85, 249, 254, 299.
Eubulides’ paradox 39, cf. Epimenides. formal symbols 69, 249, 265.
Euclid 27, 53, 60, 136, 191; —’s theorem formal system (concept) 62, 69, 246, 265,
on primes 191, 230, 286; cf. geometry, 299, 306, 323, 422, 434, 436, (examples)
excluded middle, law of the (informal) cf. eliminability, Gentzen, number
47, 52, 57, 175, 196, 281, 296, 318, theory, predicate calculus, proposi
333, 394, (formal) 119, 120, 134, 190, tional calculus; — for a number-
191, 192, 483, 487, 511, 513. theoretic predicate (Thesis II and
exclusive: disjunction 138; predicates, converse) 301, 304, 323, 404; cf.
relations 1 1, 229. formalism, formalization,
exhaustive: predicates, relations 1 1 . formal theorem 83, 85.
existence 49, 70, 225, 501, 509; unique formal theory 62, 65.
199, 225, 408; cf. introduction, pred formalism 43, 46, 59, 204, 479; a — cf.
icate calculus, quantifiers, formal system; number-theoretic, cf.
existential: quantifier 73; cf. actual, number theory; — of recursive
formal axiomatics. functions 263, 276.
explicit: definition 220, 406; occurrence formalization 53, 59, 69, 210, 245, 298,
etc. 156, 160. 396, 422; cf. formal system, for
exponents 222, 230. malism.
exportation 113. formation rules 72.
extension (function) 324, 338, (system) formula 61, 72, 137, 248, 252, 441.
89, 94, 130, 310, 405, 428, 436, 514. F-quantifier 412.
external inconsistency 2 1 2 . fraction 4, 56, cf. decimal, dual.
extremal clause 20, 259. Fraenkel, A. 16, 40, 45, 426, 517.
free: 76, 253; for 79, 253; substitution
falsity 125, 127, 500, cf. truth. 80, 156; at substitution positions 79;
Fermat's “last theorem” 47, 50, 138, term 79, 253, 410; variable 76, 148,
435. 161.
Feys, R. 141. Frege, G. 9, 43, 44, 61, 8 8 , 250.
fields 439. f-term 410.
finitary methods 63, 479, 498. function: 32, (notation) 33, 34; as a —
INDEX 543
354; letter 263, 266, 276, 277; symbol form of 302, 308, 430, 431, Rosser’s
70, 263, 403, 407, 417, 464. form of 208, 308, symmetric form of
functional 234, 275, 326, 362. (and W v W t) 308, 309, 316, 332, 470,
fundamental: inductive definition 258; 516; — numbering 206, 246, 254, 276,
theorem of arithmetic 230. 281, 288, 290, 296, 300, 313, 322, 381,
386, 394, 398, 431, 434, 501, 502, 510,
Galileo’s “paradox” 3, 14, 46. cf. recursive functions, Turing, —’s
gap (Turing machine) 364. reduction of classical to intuitionistic
Gauss, C. F. 26, 30, 48, 52, 56, 230. systems 211, 493, 495, 497, 514; —’s
general properties of f- 89, 104, 444. second theorem (on consistency
general recursive: class 307; function proofs) 210, 305, 476, 478, 479, 498,
274, cf. recursive function; functional 501.
275, interpretation 464, predicate 276, Gonseth’s paradox 38.
cf. recursive predicate; scheme 275; group 29, 439; semi- 382, 386.
set 307; truth 465, 500, 516.
generality 49, 69, 225, 502; — inter Hall, M. 386.
pretation 149; — quantifier 73; cf. Hasenjaeger, G. 398.
introduction, predicate calculus, quan Hausdorff, F. 16.
tifiers. height 107.
generalized arithmetic 246, 259, 276, Henkin, L. 389, 432, 492.
290. Herbrand, J. 98, 154, 179, 274, 326, 440,
genetic method 26, 53. 460, 463.
Gentzen, Gl 37, 69, 89, 100, 141, 225, Hermes, H. 246.
440, 445, 453, 460, 463, 476, 478, 479, Heyting, A. 51, 52, 57, 140, 166, 487,
480, 495; — s consistency proof for 491, 492, 497, 509, 517.
number theory 476, 498, 499; —'s Hilbert, D. 26, 28, 43, 53, 55, 57, 58,
Hauptsatz (normal form or elimi 61, 63, 136, 271, 318, 415, 424, 478;
nation theorem) 440, 450, 453, 460, — Ackermann cf. Ackermann; — Ber-
475, 476, 479, 492, 513, extended nays cf. Bernays; — type system 441.
460, 463, 475; — type systems 441, hypothesis of the induction 22.
460, 478, Gl 442, G2 450, G3 480,
G3a 481. ideal: elements 55; statements (Hilbert)
geometry (analytic, Cartesian) 17, 54, 55, 213, 475, 513.
(Euclidean, non-Euclidean) 17, 27, 40, idempotent laws 118.
41, 54, 55, 430, 514, (foundations) 28, identical: act, schema etc. 267, 362, 380,
60, 475, (projective) 55, 439. 461; equation, truth etc. 127, 149,
given function letter 266. 172.
Glivenko, V. 492. identity, cf. equality; — function 220;
Godel, K. 16, 41, 46, 141, 204, 211, 212, principal of 113.
213, 221, 225, 227, 239, 240, 241, 246, imaginary (complex) numbers 56, 475.
274, 317, 320, 321, 326, 389, 393, 397, immediate: consequence 83, 254, 277;
398, 400, 425, 437, 486, 493, 495, 497, dependent 220.
514; —’s p-function 240, 243; —’s implication (interpretation) 51, 69, 138,
completeness theorem 389, 393, 394, 141, 225, 498, 502, (formal logic) 69,
397, 400, 422, 423, 427, 430, 436; —'s 113, 118, 124, 154, 167, cf. intro
(incompleteness, undecidability) the duction, propositional connectives,
orem 204, 207, 211, 258, 274, 287, 304, truth tables.
308, 426, 430, 431, 514, generalized importation 113.
544 INDEX
impredicable (paradox) 38. introduction and elimination of logical
impredicative definition 42, 44. symbols 98, 102, 106, 147, 148, (Gent
improper subset 10 . zen) 442, 451, 480; strong 105; weak
inclusive disjunction 138. negation elimination 1 0 1 ; cf. varia
incompletely defined function 326. tion.
incompleteness (results) 207, 208, 213, intuitionism 43, 46, 53, 56, 59, 63, 318,
274, 302, 309, 416, 427, 430, 511, cf. 497, 513, cf. intuitionistic.
CodeEs theorem, intuitionistic: informal logic, mathema
inconsistency, cf. consistency, tics, meanings 48, 281, 291, 318, 333,
indefinite description 347. 336, 497,501, 516; formal logic, formal
independent variable 32. systems 51, 101, 114, 119, 120, 140,
index in an enumeration 4. 163, 165, 190, 192, 194, 414, 442, 444,
indirect proofs 49, 52, 497. 479, 486, 487, 492, 504; — set theory
individual 29, 44, 145, 180; — symbols 52, 491, 515.
70, 162, 403, 407, 419, 421, 464; intuitive, cf. informal, intuitionism.
— variables 145. inverse laws 186.
induction 21, 44, 46, 93, 181, 193, 247, Iongh, J. J. de 492, 497, 498.
259, 274, 427, 429, 432, 476, 498; irrational numbers 17, 31, 56.
— axiom 20, 429, 432; — cases 186; irredundant proof 482.
definition by 217, 287, cf. recursion; irreflexiveness 188.
— number 22, 478; — postulate (or irrefutable formula 389, 423.
schema) 181, restricted 204, 472, isomorphic systems 25.
generalized 476; — predicate (or
proposition) 22, 430, 479; — rule Jankowski, S. 98, 141.
181; — step 2 2 ; — variable 2 2 . juxtaposition 71.
inductive definition 20, 258, 306.
inequalities 118, 187, 197, 225, 229. Kalmar, L. 135, 285, 286, 287, 518.
inference 83; rules of 60, 83. Kemeny, J. G. 432.
infinite: cardinal 13; descent 193, 478; k -equality 172, 400.
sequence 7, 8 , 17; set 13, 14. /e-fold recursion 273.
infinity, axiom of 46, 425; points at £-identity 172, 178, 400.
56; problem of 46; cf. actual, po Kleene, S. C. 8 8 , 135, 261, 274, 275,
tential. 279, 281, 282, 284, 286, 287, 288, 293,
informal: 41, 62, 64, 69, 85, 181; 302, 304, 306, 307, 312, 320, 322, 324,
axiomatics 28, 41, 53; induction 181; 325, 326, 335, 337, 350, 353, 355, 382,
mathematics 62, 69; presentation 181; 460, 487, 492, 503, 508, 511, 514, 515,
symbolism 225; theory 65. 516.
initial: function 219; state etc. 357; 367. Klein, F. 54.
input (Turing machine) 367. £-predicate calculus 177.
integer 3, 4, 56, 439, 530, 532. £-proposition letter formula 177.
interchange 121, (Gentzen) 443, 444; — &-recursive function 273.
of premises 113. Kreisel, G. 431.
interdeducible formulas 151, 154, 398, Kronecker, L. 19, 46, 320 (8*), 498.
435. ^-transform 177, 178.
interpretation 57, 64, 69, 125, 130, 138, Kuznecov, A. V. 289.
143, 175, 194, 212, 303, 421, 430, 478,
499, 501; — of variables 146, 149, 227. lambda-definability 320, 382.
intersection of sets 10 , 16. Langford, C. H. 45, 141, 335.
INDEX 545
language 61, 69, 249, 265. metamathematical: letters, symbols,
lattices 439, 492. variables 62, 70, 140, 250, 265; def
least number: operator (s) 317, 324, initions, functions, predicates, re
414, (p) 225, 228, 279, 289, 295, 323, cursion 72, 80, 251, 258, 276, 290;
327, 329, 330, 347, 350, 414; principle induction, proofs, theorems 85, i81.
189, 200, 511. metamathematics 55, 59, 62, 63, 69, 80,
least upper bound 31, 43. 85, 175, 176, 423; arithmetization of
Leibnitz, G. W. v. 43, 61. 246, 276, 290, 396; cf. formal system,
letter: formula in = , P v ... P s 399; formalism.
ra-tuple 132; in a semi-group 382. metatheory 62, 65.
Lewis, C. I. 138, 141, 335. midsequent 460.
liar, the (paradox) 39, cf. Epimenides. minor premise 264.
limit 32, 348. mix 450, 453; — formula 450.
linear order 29. modal logic 141.
Liouville, J. 7. model 25, 53, 422, 463, 475.
Lobatchevsky, N. I. 28, 54. modus ponens 98.
logic 46, 51, 60, 69, 80, cf. formal Mostowski, A. 284, 287, 293, 297, 304,
system, formalization, predicate cal 321, 430, 432, 437, 439, 492, 534.
culus, set-theoretic, motion (Turing) 357, 358, 379, 380.
logical: constant 436; function 169, 389; multiplication cf. product,
notion 28, 60; operator 73; rules 442; mutually exclusive relations 11.
symbol 69.
logicism 43, 59, 180. name 71, 250, 265.
logistic method 61, cf. logicism. name form: 142, 146; interpretation
Lowenheim, L. 136, 394, 427, 436; —’s 146, 227; occurrence 157; prime —
(or the — Skolem) theorem 394, 398, 161; — replacement 161; variables
400, 425, 426, 427, 428, 436. 142, 143, 146.
Lukasiewicz, J. 140, 335. n- ary relation 144.
lying Cretan, the 39, cf. Epimenides. natural number 3, 12, 19, 44, 48, 51,
Lyndon, R. C. 532. 217, 427, 432, 530, cf. number theo-
ry.
machine cf. Turing. negation (interpretation) 51, 69, 138,
MacLane, S. 154. 225, 498, 502, (formal logic) 69, 101,
major premise 264. 113, 118, 121, 166, 407, cf. introduc
Mannoury, G. 38, 498. tion, intuitionistic, propositional con
Markov, A. A. 289, 382, 386. nectives, truth tables.
material: axiomatics 28, 41, 53; im Nelson, D. 211, 223, 409, 487, 492, 504,
plication 138. 511, 514, 515.
mathematical: induction cf. induction; nested recursion 271.
logic cf. formal system. Neumann, J. von 9, 40, 53, 57, 140, 204,
McKinsey, J. C. C. 141, 522. 425, 437, 440, 463, 472.
meaning (classical mathematics) 57, 478, non-arithmetical predicates 287, 292,
(functional notation) 33, 227 (logical 315, 479, 501.
symbolism) 225, cf. interpretation, non-constructive : logics 321, 431; proofs
member of a set 9. 49, 52.
memory (Turing machine) 356, 366, 379. non-enumerable sets 6, 16, 175, 423, 425.
Meray, C. 30. non-Euclidean geometry cf. geometry,
meta- 63. non-intuitionistic methods 48, 51.
546 INDEX
non-recursive : functions 324; predicates ordinals 16, 476; constructive 325.
283, 310, 316, 332, 382. ordinary notions 28, 60.
normal form, conjunctive etc. 134, 136; oscillation, function of large 289.
— for proofs (Gentzen) 440, 453, 460; output (Turing machine) 358, 367.
— for recursive functions 288, 292,
330; Skolem 435; cf. prenex. pairing of parentheses 23, 73.
normal systems (Post) 320. paradoxes 36, 40, 42, 45, cf. Epimenides,
notions 28. Galileo, Richard, Skolem, Zeno,
n -tuple 5, 17, 33. parameters 34, 150; — in recursion 218,
number 3, 6, 9; 12, 14, 16, 19, 29, 56, 235, 342, 344.
476. parentheses 23, 70, 73, 250.
number theory 29, 51; formal 69, 82, 97, partial function 325, 332, 341, 351.
98, 181, 241, 252, 258, 295, 305, 310, partial order 29, 106.
314, 395, 406, 407, 414, 419, 427, partial predicate 327.
492, 495, 501; — with restricted in partial recursive: function 326, cf. re
duction schema 204, 472, 476; without cursive function; functional 326; pred
• 204, 407, 474; cf. analytic, Robin icate 328; scheme 326.
son’s system. Pasch, M. 61.
number variable 70, 263. passive: state etc. 357, 367.
number-theoretic: formula 108; function Paul (the apostle) 39.
7, 17, 33, 38, 198, 217, 414, 509; Peano, G. 20, 43, 61; —’s axioms 20,
predicate 226, 389; cf. number theory, 26, 186, 429, 432, (generalized) 247,
numeral 177, 195, 252, 254, 255, 263, cf. induction.
278; generalized 465. Peirce, C. S. 61, 534.
numeralwise: decidable formula 196, Peter, R. 223, 231, 271, 272, 273, 395,
297; expressible predicate 195, 200, 476
206, 244, 296, 298; representable plain f-term 410.
function 200, 243, 295, 298. Poincare, H. 30, 42, 46.
n-valued logic 140, 179, (for n = 3) point set 9, 18.
332, 335, 515. Poncelet, J. V. 55.
positive integer 3, 530, 532.
object 24, 144; function as an 354; Post, E. 135, 140, 284, 289, 293, 301,
— language 63; system of —’s 24; 306, 307, 315, 320, 321, 343, 346, 356,
— theory 62, 65; — variable 144; cf. 361, 379, 382; —’s theorem 293, 395,
formal objects. 402; — word 384.
observation (Turing) 356, 378, 381. postulates 26; — of a formal system 80;
occurrence 70, 155, 441. lists of 82, 442, 481; respective 99,
omega-completeness 212. 106, 459.
omega-consistency 207, 212, 304. postulational method 26, cf. axiomatic,
one-place predicate calculus 398, 436. potential: infinity 48, 62, 70, 357, 363;
one-to-one correspondence 3, 9, 426. recursiveness 324, 331, 332.
open formula 151. Prantl, C. 39.
operation 32, ,358. predecessor 223; generalized 247.
operator, formal 73; logical 73. predicate 143, 144, 226.
order 29, (cardinals) 10, (natural num predicate calculus 82, 97, 98, 142, 389,
bers) 13, 21, 187, (reals) 31; first, 432, 435, 440, 480, 487, 492, 495, 513;
second, higher 180, 421; within a — with equality 399, 408, 417, 424,
type 44. 432; higher etc. 179, 213, 321, 430,
INDEX 547
432; — interpreted in a finite domain propositional calculus 82, 90, 98, 108,
168, 178, 400, 435, 464; one-place 398, 190, 406, 442, 479, 486, 492, 504, 513;
436; — with postulated substitution — with postulated substitution 139;
179, 398, 435; subsystems of 106, subsystems of 106, 154, 459.
154, 459; cf. quantifiers, set-theoretic, propositional connectives 73, 226, 334,
predicate inference 460. (recursiveness) 226, 228, 237, 329,
predicate interpretation 146, 227. 337; cf. propositional calculus, truth
predicate letter: 142, 179; formula 143, tables.
equality and 399, 400, in 156, k- 177, propositional function 144, 226.
with numerals 390; propositional cal propositional inference 460.
culus 109; interpretation of —s 145, provable formula 81, 83, 137, 258, 261,
168, 174, 421. 299, 301, 302, 304, 306, 423, 435.
predicate logic cf. set-theoretic, pure: number theory 30; predicate
predicate symbol 70, 403, 421, 464. calculus 143, with equality 399; prop
predicate variable 179, 180, 398, 435. ositional calculus 109; variable proof
premise 83, 264. etc. 451.
prenex: form, formula 167, 285, 389, Pythagoras 31, 47, 60.
435, 465, 466, 500, 516.
Presburger, M. 204, 407, 474. quantifiers 73, 76, 151, 162, 165, 177;
prime : formula etc. 111, 112, 161, 201, contraction of 285; F- 412; cf. bound
410, 493; number etc. 191, 230, 239, ed, introduction, predicate calculus,
286. recursive predicates.
primitive notions 28, 60. Quine, W. V. 46, 241, 250, 265.
primitive recursion 218, 221. quotation marks 250, 265.
primitive recursive : derivation 224; de quotient 188, 202, 223, 285, 414, 526.
scription 220; function 219, 223, 228,
234; predicate 227, 235; scheme 234, Ramsey, F. P. 45.
238; truth 474. range 32; — of definition 326, 331.
principal: branch 267; disjunctive rank (logical operators) 74.
normal form etc. 134; equation 267; Rasiowa, H. 389.
f-less transform 411, 417, 419; formula rational numbers 4, 29, 31, 439.
443; function letter 266. real: functions 18; numbers 6, 17, 30,
printing (Turing machine) 357, 358. 43, 52, 361; statements (Hilbert) 55,
procedure cf. decision, valuation, 213, 475, 513.
product (natural numbers) 186, 201, realizability 503; P- 515.
204, 222, 239, 285, 407, 439, 526, (sets) realization: function 503; number 502.
10, 16; finite 224, 285; logical 179. reckonable function 295, 298, 320, 321,
projective geometry 55, 439. 322, 323.
proof (formal) 65, 83, 85, 137, 249, 254, recursion 217, 221, 231, 232, 233, 237,
444, (informal) 85; — schema 84; 252, 258, 260, 270, 276, 351, (arithme
— theory 55, cf. metamathematics; tical equivalents) 241, (elementary
— thread 107. equivalents) 286; — theorem 352,
proper : pairing 23; subset 10. first 348, 354, 374; cf. recursive
property 44, 144. function.
proposition letter 108, 139, 177; — recursive class 307.
formula 108, in 109, k- 177, with recursive functions 219, 234, 273, 274,
numerals 390. 320, 324, 326; applications of 217,
proposition variable 139. 258, 298, 308, 313, 314, 325, 382, 395,
548 INDEX
427, 432, 437, 502; — defined re residues 25, 27, cf. remainder,
cursively by a system of equations resolvable predicate 295, 296, 298, 305,
266, 275, 326, by a number {Godel 321.
numbers of) 289, 292, 319, 330, 337, restricted: induction schema 204, 472,
340, 352, 354, 416, 510; formalism 476; predicate calculus 180.
of 263, 276; normal form of 288, restriction on variables 442.
292, 330; — in the number-theoretic resulting deduction 94.
formalism (including eliminability Richard's paradox 38, 42, 45, 341.
theory) 243, 295, 407, 415, 471, 515; Robinson, J. 241, 320, 439.
— of zero variables 223, 238, 274; Robinson, R. M. 197, 223, 272, 297, 437,
cf. Church's thesis, 439; —*s system 197, 204, 244, 296,
recursive functional 234, 275, 326. 305, 310, 315, 321, 419, 433, 437,
recursive predicates 227, 235, 276, 328; 439, 470, 472.
applications of cf. recursive functions; Rose, G. F. 515, 526.
— under quantification 281, 292, 337. Ross, J. D. 204.
recursive realizability cf. realizability, Rosser, J. B. 88 , 140, 179, 208, 282,
recursive scheme 234, 275, 326. 307, 308, 312, 320, 321, 407, 432, 437,
recursive set 307. 526.
recursively enumerable set 306, 311, 320, Russell, B. 9, 37, 38, 42, 43, 44, 45, 46,
343, 346, 398. 61, 407; —*s paradox 37, 40, 42, 45.
reducibility, axiom of 44; — of the Rustow, A. 39.
decision problem 314, 435, 436; 1-1 Ryll-Nardzewski, C. 427.
343; cf. degree,
reductio ad absurdum 15, 8 6 , 99. satisfiability 172, 174, 389, 400, 423,
redundant: axioms, postulates 406. 435; joint 390.
reflexive laws 9, 114, 183, 399, 404. satisfying assignment 389.
refutable formula 194. scanned: square, symbol 357, 360.
regular table 334. schema, axiom etc. 81, 84, 87, 140, 234;
relation 144. recursive 219, 234, 267, 275, 279, 326.
relative recursiveness 223, 227, 234, 235, Schmidt, A. 179.
275, 289, 298, 326, cf. recursive Schonfinkel, M. 34, 321.
function, uniformity; Schroder, E. 61, 123, 136.
relatively prime numbers 239. Schiitte, K. 479.
remainder 188, 202, 223, 239, 414. scope 73, 8 8 , 150.
replaceability of function symbols by sequence 5, 7, 16, 19,32, 70; — form 106.
predicate symbols 417, 419, 424. sequent 441.
replacement 115, 151, 161, 184; — in all set 1, 9, 40; — theory 1, 9, 16, 40, 477,
occurrences 109, 115, 117; — in the axiomatic 40, 45, 425, intuitionistic
formalism of recursive functions 264, 52, 491, 515, relativization of 427.
274, 277; lemmas for 116, 152, 185; set-theoretic predicate logic 175, 389,
— property of equality 185, 399, 404, 400, 421, 435.
of equivalence 117, 152, 161, cf. 154; several-sorted predicate calculus 179,
special — properties 183; — theorem ^20.
116, 151, 184. Sheffer stroke 139.
representation: of a system 25; on a Tur side formula 443.
ing machine tape 359, 364, 377, 381. signifies 498.
representing: function 8 , 227; predicate Sikorski, R. 389.
199, 293, 318. similar sets 477.
INDEX 549
simple: completeness 194, 304, 309; of objects 24, 53, 421, 436, 463; cf.
consistency 124, 297, 304, 309, 314, formal system.
cf. 391.
simultaneous: induction 193; recursion table, function 35; logical function 170;
233, 252, 272; satisfiability 390. predicate letter formula 170; Turing
situation (Turing machine) 357. machine 358; cf. truth,
Skolem, T. 40, 41, 273, 287, 394, 398, tally (Turing machine) 358, 381.
426, 427, 431, 432, 435, 436, 476; tape (Turing machine) 357, 361, 377.
—'s normal form 435; —'s paradox Tarski, A. 98, 140, 141, 436, 438, 439,
426. 501, 532.
square (Turing machine) 357, 380. technical notions 28, 59.
standard position 360. term 72, 143, 177, 198, 248, 264, 390,
state (Turing machine) 357, 362, 377. 399, 403, 420, 441, 464; free 79, 253,
strict implication 141, cf. 138. 410.
strong: introduction etc. 105; senses, terminal: state etc. 357, 358, 367.
tables, etc. 334, 336. theorem, formal 81, 83, 85, 422, 424,
structural: inference 460; rules 443, 444, cf. provable formula; informal, meta-
480. mathematical 85; — schema 84.
subformula 449; — property 450. theory 57, 58, 59, 65, 213, 436; formal
subset 9, 426; set of —'s 8, 15, 425, 426. 62, 65; informal 64, 65; meta- 62, 65;
subsidiary deduction 94, 102; — rule object 62, 65; proof 55, cf. meta
87, 94, 123, cf. dependence, varia mathematics; cf. axiomatic, number
tion. theory.
substitution 78, 115; definition by 220, thinning (Gentzen) 443, 444.
228; — in formalism of recursive three-valued logic 332, 335, 515.
functions 264, 274, 277; free 80, 156; Thue, A. 382; — system etc. 383, 385.
— for individual variables 101, 147, Trahtenbrot, B. A. 435.
161, 253, 258; — notation 78; — for transcendental number 7, 17.
predicate letters 155, 156, 159, 179, transfinite: cardinal 13, 15, 16, 29;
converse 161; — for proposition let induction 476, 478; ordinal 476; re
ters 109, 139, converse 112. cursion 273.
subsystem 89, 436, cf. extension; —s transform 417; principal f-less 411, 417.
of predicate calculus 106, 459. transformation rules 80.
succedent 441; — rules 443. transitive laws 9, 114, 183, 187, 399, 404.
successor 12, 19, 44, 70, 217, 220, 223; translation 257, 265, 338, 494.
generalized 247, 276. tree form 106.
sum (natural numbers) 186, 201, 222, truth 41, 57, 62, 125, 196, 226, 474,
239, 285, 439„ (sets) 10, 16; finite 224, 499; effective 465, general recursive
285; logical 179. 465, 500, 516; identical 127; intui-
symbol, use vs. mention 64, 71,250,265; tionistic 501; primitive recursive 474;
scanned 357; — space 377, 379; — tables 126, 140, 141, 2-valued 126,
Turing machine 357, 358, 361, 363. 138, 139, 202, 226, 500, 3-valued 334,
symbolic language 61. 515, regular, strong, weak 334; —
symbolism, logical 69, 225. values 125, 126, 140, 226, 332, 335; —
symbolization 61. value function 125, 226; cf. validity,
symmetric laws 9, 114, 183, 399, 404. valuation.
syntax language 63. Turing, A. M. 301, 315, 320, 321, 322,
systems: of equations 264, 276, 277; 356, 361, 377, 378, 382, 386, 432; —
550 INDEX
machines 321, 356, 362, 377, 381, 382, valuation (effective) 126, 168, 400, 464,
383, Godel numbers of 374, 382, 474, 500, (non-effective) 174, 389,
with two terminals 367; —’s thesis 400; R— 515.
301, 321, 376, 381, 382, 398, 434, cf. value 32; ambiguous 33, 227; —
Church’s thesis. column 127; cf. truth, valuation.
Turquette, A. R. 140, 179. Vandiver, H. S. 48.
two-valued logic 140, cf. truth tables, variable 32, 62, 70, 139, 140, 150, 179,
types 44, 45; systems using higher 179, 248, 252, 263; —’s held constant 96,
213, 321, 430, 432. 102, 149; interpretation of —’s 146,
149, 150, 227; pure — proof etc. 451 ;
undecidable: formula 194, cf. Godel’s restriction on —’s 442; — varied 95,
theorem; system 437, cf. decision cf. variation,
problem. variation in a deduction 95, 102, 140,
undefined notions 28, 59. 148, 179, 420.
uniform : in assumed functions 234, 275, Veblen, O. 27, 54.
292, 298, 316, 326, 344, 362; conver verifiable formula 465.
gence etc. 164; method 136, 314, 316, verification 57, 58.
318. Vieta, F. 61.
union of sets 10.
unique existence 199, 408. Wajsberg, M. 141.
unit set 9. Wang, H. 179, 287, 432.
universal function 289. weak: equality, senses, tables etc. 327,
unsolvability cf. degree, 334; negation elimination 101.
unsolvable problem 301, 313, 314, 382, Weierstrass, K. W. T. 30, 46, 48.
432, 437, cf. decision problem, non- well-ordering 16, 189, 477.
recursive. Wessel, C. 56.
upper bound 31. Weyl, H. 39, 43, 45, 46, 48, 63, 58.
Whitehead, A. N. 43, 45, 61, 407.
vacuous: axioms 27; inference etc. 24, word, Post 384; — problem 383, 386.
138, set 9.
validity 172, 174, 389, 398, 400, 422; Young, J. W. 54.
in finite domains 172, 178, 400, 435,
464; cf. consistency. Zeno’s first paradox 54.
Zermelo, E. 16, 37, 40, 45, 52, 394.
zero 12, 19, 70, 217, 223; generalized
246, 276.

Kleene - Introduction To Metamathematics (1971) PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Kleene - Introduction To Metamathematics (1971) PDF

Caricato da

Copyright:

Formati disponibili

INTRODUCTION TO METAMATHEMATICS

Edited with the cooperation of

STEPHEN COLE KLEENE

WOLTERS-NOORDHOFF PUBLISHING - GRONINGEN

Library of Congress Catalog Card Number 70-97931

North-Holland ISBN 0 7204 2103 9

First published 1952

Sole distributors for the U.S.A. and Canada:

Printed in The Netherlands

Chapter V. FORMAL D EDUCTION........................................... 86

§ 20 . Formal deduction........................... ... 86

Chapter VI. THE PROPOSITIONAL CALCULUS.................... 108

P art III. RECURSIVE FUNCTIONS

Chapter XII. PARTIAL RECURSIVE FUNCTIONS................ 317

Chapter XIII. COMPUTABLE FUNCTIONS............................... 356

Then let these fractions be enumerated by following the arrows. A rational

*10 *11 *12 *13

Select the diagonal fraction shown by the arrows. In this change

/o(0) /o(l) /o(2) /o(3) ...

§ 3. Cardinal number. Cantor’s theory of ‘abstract sets’ deals with

The relationship M = ffi, i.e. M ~ N, evidently falls under Case (2) by

(2) If M = n (for a natural number n) and M ~ c M, then M t = M.

corresponding to a single real number. Conversely, any single sequence

in which each member ot each of the given sequences has a determinate

we rely on the dots “ . . to suggest the continuation of the sequence

respectively. This is a matter of detail concerning decimal notation.

A proper pairing, indicated by the subscripts, is discovered by the fol­

— r forr £ R — x . (If x is ration al, th en — x corresponds to — x.)

N o tw ith s ta n d in g t h a t in th is th e o r y an alysis is “ a r ith m e tiz e d ” , th e

§10. Functions. In th e m ost gen eral sense, a (single-valued) function

E xample 1. W h e n w e s a y “x + y is s y m m e tric ” w e m ean b y “x + y”

T h is situ a tio n ca n b e a v o id e d b y u sin g in stea d of " / ( * ) ” f ° r th e

T h e re is an o th er n o ta tio n (due to C h u rch 1932) in w h ich th e in d e­

E xample 3 . Let 9 b e a fu n ctio n of tw o num bers. U s in g th e X -notation

F o r a n y w -tu ple tv tn o f a rgu m en ts for /,

{ X % . . . x nf{xv . ..,# „ ) } ( * ! ,. . . , * » ) = * •>*»)•

{kyx 9(x, y)}(0 , 3 ) = 9(3, 0 ), {X*y 9(3, y)}{z, x) = 9(2, x).

E xample 4 . Let X an d Y b o th b e th e residues m od u lo 2, i.e.

T h u s b y the second ta b le, 0 •0 = 0 • 1 = 1-0 = 0 an d 1-1 = 1.

§ 11. T h e p a r a d o x e s . T h is ch ap ter is in ten d ed to present th e problem

(A) The B urali-F orti paradox 18 9 7 *, also kn ow n to C a n to r in 1895 ,

(B) S o m ew h a t sim ilar an tin om ies occur in th e th e o ry of tran sfin ite

(C) The R ussell paradox 19 0 2-3*, d iscovered in d e p e n d e n tly b y Zerm elo,

T h is p a ra d o x is esp e c ia lly in terestin g for its im p lica tio n s con cern in g

(E) T h ese m o d e m p arado xes, w h ich fall m ore or less w ith in th e c o n te x t

A xiomatic set theory. R eco n stru ctio n s of set th e o r y ca n b e g iv e n ,

T he broader problem of foundations. A ssu m in g t h a t th e p ara­

I mpredicative definition . W h e n a set M an d a p articu lar o b je c t m

W eyl 's constructive continuum . T h e im p re d ica tiv e ch aracter of

T h ere h a v e arisen three m ain schools of th o u g h t on th e fo u n d atio n s of

Logicism. T h e lo g icistic thesis is th a t m a th e m a tics is a b ran ch of

L o gicism tre a ts th e ex isten ce of th e n a tu ral num ber series as an

§ 13 . I n t u i t i o n i s m . In th e 1880 ’s, w hen th e m eth o d s o f W eierstrass,

a ll ordered q u adru p les (x, y , z f n) o f p o sitiv e in tegers w ith n > 2, an d

b o th th e classical an d th e in tu itio n istic logic, b y re d u ctio ad ab su rd u m

§ 14. Formalism. B ro u w er has revealed w h a t th e g en etic or co n ­

I f th is oth er th e o r y is con sisten t, th e n th e a x io m a tic th e o r y m u st be.

in tro d u c e d a poin t at in fin ity on each of th e original lines, such th a t p ar­

w h ich m a tte r for th e d ed u ctio n s h a v e been expressed b y axiom s. W h e n

so chosen th a t th e y serve as m odels for p arts of inform al m a th e m a tic s

§ 16 . F o r m a l s y m b o l s . W e sh all n o w in tro d u ce a p a rticu la r fo rm al

expressions (, s , ), *, (, (c )\ ) is w ritte n “ (s)-((c)')” H ere “ (s)-((c)')'' is a

§ 17 . Formation rules. W e sh all n o w d efine certa in su b ca tego ries

10 11 12 13

A proper pairing, indicated by the subscripts, is discovered by the fol

T h e re is an o th er n o ta tio n (due to C h u rch 1932) in w h ich th e in d e

T he broader problem of foundations. A ssu m in g t h a t th e p ara

§ 14. Formalism. B ro u w er has revealed w h a t th e g en etic or co n

in tro d u c e d a poin t at in fin ity on each of th e original lines, such th a t p ar

E xample 4. R esto rin g parentheses to “A 3 B V C & D ” g iv e s suc

E xample 1. L e t A , B an d C b e form ulas. T h e n th e fo llo w in g se

B y an analysis o f a d ed u ctio n or proof A v , . . , A *, w e m ean a sp ec