Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
BIBLIOTHECA MATHEMATICA
A Series of Monographs on Pure and
Applied Mathematics
Volume I
Editors
N. G. de B ruijn
J. de Geoot
A. C. Zaanen
INTRODUCTION TO
METAMATHEMATICS
BY
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.
Publishers:
WOLTERS-NOORDHOFF PUBLISHING - GRONINGEN
NORTH-HOLLAND PUBLISHING COMPANY - AMSTERDAM • OXFORD
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
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 ..•
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.
8 THE THEORY OF SETS CH. I
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
12 THE THEORY OF SETS CH. I
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
*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 * • •
\ 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 *
y 0 i
X 0 1 * 0 0 0
i 0 i 0 i
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).
(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
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 ).
§ 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
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.
§15 FORMALIZATION OF A THEORY 63
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)
64 A CRITIQUE OF MATHEMATICAL REASONING CH. Ill
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
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
72 A FORMAL SYSTEM CH. IV
(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
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).
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
(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
not
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 .
(H) f(z) = lim /(*, z), not b y f{z) = lim f(z, z).
x— >0 z— > 0
x, A (x ), b, A (b )
m a y b e re sp e c tiv e ly ,
A, A d B
B.
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.
II
u
&
14. 15.
a = b 3 (a = c 3 b = c ).
if
II
u
&
16. 17.
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.
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.
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.
1. A D ( B D C ) , A & B b C — (6 ).
' 2. AD(BDC) h A & B d C — T h eorem 1, 1.
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.
(Im plication ) I f T, A b B , A, A D B b B.
then T h A D B . (M odus ponens.)
(G en erality) A (x ) bx V x A ( x ) . V x A ( x ) b A (t).
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.
100 FORMAL DEDUCTION CH. V
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.
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
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.
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.
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 .
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. 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
§24 DEPENDENCE AND VARIATION 107
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.
(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
c3, B, C, ...,
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 .
(a') B h ^ D B .
§25 PROPOSITION LETTER FORMULAS 111
(b') -t e 2 & C b ®
(10') B b ADB.
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.
*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.
*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
116 THE PROPOSITIONAL CALCULUS CH. VI
dfDS nc3VS
3® ) V-I<3
( a (-|J? V V-i<3 B )
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.
b C0 ~ Cx ~ ~ Cw ~ C n,
118 THE PROPOSITIONAL CALCULUS CH. VI
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.
*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.)
♦ 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 ) .
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.
120 THE PROPOSITIONAL CALCULUS CH. VI
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.
*5 2 . By *4 5 an d * 5 1 .
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 ).
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 ,
E xample 3 (concluded).
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.
e2~2
2 t f
a t t f
f f t
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
128 THE PROPOSITIONAL CALCULUS CH. VI
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 .
§28 VALUATION, CONSISTENCY 129
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.
§ 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
132 THE PROPOSITIONAL CALCULUS CH. VI
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.
P i, P 2 V - i P 2 b E .
(b)
- i P j, P 2 V - i P 2 b E.
(c) P iV -iP !, P 2 V - i P 2 b E,
as w as to b e shown.
(b) P ^ -iP * . . . , P mV - i P w b E V - i E .
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).
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.)
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
E *.
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
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.
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.
*70 . A (x ) D B (x) bx 3 x A (x ) D Ix B (x ).
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).
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 b^O 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.
3. b V b (b ^ 0 D a + b ^ a ) — V -in tro d ., 2.
*7 2 . A(x)~B(x) bx 3 x A (x ) ~ 3 x B (x ).
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.)
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 3. Let m=
1 ; n = nx = 2 ; a 2, a 2 b e c, d P ( a x, a 2) b e \
p lic itly [an onym ously] free (bound) in A ( a x, . . . , a n) are explicit anony [
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.
§34 SUBSTITUTION 157
E xam ple 4. In considering <C?(a, b) & ^aS(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.
158 THE PREDICATE CALCULUS CH. VII
*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.
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
166 THE PREDICATE CALCULUS CH. VII
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)
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 )
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.
a liM U4 U(a)
i t t f f
2 t f t f
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
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
a t 31 2(a)
i t
2 f
t ^ I,(a) t 3 1 2(a)
t =3 1.(1) t3lg(2)
t 3 t t 3 f
t f
f V ( - it & t)
f v (f & t)
fv f
f
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 .
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.
178 THE PREDICATE CALCULUS CH. VII
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
§ 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.
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.)
*113. r= s b r t= s t. *114. r= s b tr = ts .
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 .
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
* 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 ),
* 13 7 ,. b a = 0 V a = I V a = 2 V 3 b {a = b ' " ) .
* 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 .)
O th e r consequences o f * 148a a r e :
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.
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 .
*171. b 3 !x (t= x ).
200 FORMAL NUMBER THEORY CH. VIII
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.
(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
202 FORMAL NUMBER THEORY CH. VIII
(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 ).
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).
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).
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.)
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
P (y)-
217
2 18 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
(i)
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 .
(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
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 .
f 9 (0 ,« ) = 171(a) _
»
H ( <p(i', a) = x(6. 9 (*. «). a)
(V b), = 2 ; 2 , 4.
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.
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,
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).
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 .
tm (or j £ ( 0) = t),
1 E (a ’) = E(a),
f e(0) = 0 ,
\ «(«') = sg(e(a))
(cf. # 9 § 4 4 ). T h e n E(a) = e (a )= 0 .
# 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
228 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
'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.
[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
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(^iW)-?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-
236 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
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.
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 (^i> • • •>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
242 PRIMITIVE RECURSIVE FUNCTIONS CH. IX
* 0> ^ 1* • • •>
«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
(vi) b 3 !w P ( x 1, . . . , x n, w).
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.
(3 ) b 3 !w P (y , x 2, . . . , x n, w).
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.
D, &, V, V, 3, =, + , •, ', 0, a, ,.
P p Q
D,
d ; D,
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).
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.
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 )).
C F (a, 25) & P f(y) & (E n)n<y[N u(n, x) & S ((y )0, a, n, 25)].
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,
1. X(0,4) = 7. 2. x (1<7) = 7.
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:
—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 :
h ( b ,c ) = 7, } (bi)
(b) />
(0 )= 4 , 1
f{ b ')= h { b ,f{ b )). /
(b2)
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
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
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.
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
0 if n= 0,
(d) oc(n, a) = 1 if n= 1,
a otherw ise.
( 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-
§55 GENERAL RECURSIVE FUNCTIONS 273
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, < p (» , P(», b, 9K b))), 9 (»', b))
besides applications of Schemata (I) — (V); and similarly (with a scheme
reducing to (g) for k = 2) for each k > 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
274 GENERAL RECURSIVE FUNCTIONS CH. XI
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
§55 GENERAL RECURSIVE FUNCTIONS 275
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, . . . , ^i) (§ 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 :
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
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
(E n u m eratio n theorem , K le e n e 19 4 3 .)
282 GENERAL RECURSIVE FUNCTIONS CH. XI
(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
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
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 .
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).
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
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,
^ (E x)(y)(E z)[{R (a, (y)0, z ) & w = 0} V {*(« , x, (y),) & w = l}] (b y (18)).
= W ( £ y ) [ R ( W o , W x, ( * ) „ y) -+ r } ( ( * ) lf g, a, (*)„)] (b y (18)),
(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).
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,)-
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.
{43) , ( E Y ) 9 i< * ,Y ) s h A (x ).
(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
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
304 GENERAL RECURSIVE FUNCTIONS CH. XI
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:
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.
l. K(x,y). 2. K(x, y)
=3(x).
306 GENERAL RECURSIVE FUNCTIONS CH. XI
§ 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
(56) h B (x ) & |- - i B ( x ) .
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) .
(d) ( y ) T M v f, y).
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).
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
(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)).
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.
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(^i> • • •, 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
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 )).
< ...,
p ( x lt is
x n)) ; for an w-tuple ..., for which has no
d e fin e d x v x n 9
= y, for all = , ..., and in the case of each for those mrtuples
j 1 l j
(for fixed .. .,
n , l, m vthereis suchanEindependent of 4i- ..., 4i*
m t)
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
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
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,
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).
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.)
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'.
W e th en h a v e
CT(0) = 0, t (1,1) = 1,
(A)
I 9(*lt ...,*„) = o(x(*i. •••»*«)).
. • • • ’ x ») = •••>*«))>
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
§64 THE 3-VALUED LOGIC 339
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 <p is not
partial recursive.
We ca n read “x v . . . , # w” , “xn , ..., x ln” in p lace of “x ” , €tx x \
re sp e ctiv e ly (where n ;> 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.
0% , qf x 2, .. . , x n) + l ^ <t>(q, q, x 2, .
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> • • •> *«)•
\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 <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 .
f(x1( ...,x„,c)=x.
We now d em o n strate th e converse. Let ( E ^ ' ^ ) c be th e set o f
(e ?),. E t r f j . e .
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) ~ <p>+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
350 PARTIAL RECURSIVE FUNCTIONS CH. XII
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(*).
(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).
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
§67 TURING MACHINES 359
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
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.
§67 TURING MACHINES 361
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
362 COMPUTABLE FUNCTIONS CH. XIII
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.
2t 1 P0
S ta r tin g from th e in itial situ atio n
T 1 1 0,
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
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
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
1 — EL2
2 PL3 L2
3 R 01 ro2
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
(b) 3 . -
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
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,
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.
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 :
2. , x^f . • ., x nt , x , x nt
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 •
of X l Xl
( > • • • > x n) > * • • > Xm x l
{ > • • •t% n)f 4 K X l( ^ l» • • • t n
% )> • • • > X w (% j • • • >% n ))
r ------ , x n), if y = 0.
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 :
Si Si S3 Si So Si
P a .c (M . 0 Ms) if Ml = « & Ms = C
pM (a = 0, c= 1,
w otherw ise.
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 :
^i{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*
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 :
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 ...,
qi
1 2
q<>
1 2 1 1 1 1 1 1 1 1 1 1 1 1 1
• • • 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 ?)
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 ;
U s eq ds0s ,V , U h s 0q ds0s ,V , U s eq dh V , U h s 0q dh V .
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.
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,
(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 .
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.
392 AXIOM SYSTEMS CH. XIV
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
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),
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
(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 .)
(:
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
where x> 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
(“ .) (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),
(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 ),
(vii) f- G * ~ —i F * .
U sin g (vii) in (vi),
(viii) b G * D —iV y D (y ).
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 .)
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.
398 AXIOM SYSTEMS CH. XIV
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 .)
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>
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 )
402 AXIOM SYSTEMS CH. XIV
(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, . . . ) .
(1) P f { S a v .... Sa„.) ss (£Ci) ... (Ecn,) [Cj € S0l & . . . & Cnj £ Sa„. &
P j( c l> • • • > £»/)] = (^l) • • • (c»y) [^i 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
§73 THE PREDICATE CALCULUS WITH EQUALITY 403
(II) h2 E ~ E ' .
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
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)).
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 ) .
3 v [ F ( t 1, v) & C ( v ) ] .
2. V x { F ^ P ( v 2, f ( g ( x ) ) ) & Q ( x , x ) }
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).
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
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.
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* ) ] ,
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).
414 AXIOM SYSTEMS CH. XIV
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 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 .
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 °} .
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
§74 ELIMINABILITY OF DESCRIPTIONS 419
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 .
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).
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 ,
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 .
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
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
(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
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
undecidable.
But S 5 h as as its constants only those of and possibly individual
S x
S t is undecidable.
R emark 4. For the conditions on S in the definition of consistent
3
abstract), Tarski ( a,
1949 babstracts), JuliaRobinson
1949 (1949 abstract,
1949 ) and Raphael Robinson abstract) obtain in rapid succession
(1949
A x io m schem a.
C -*C .
w ith 0 e m p ty for
th e in tu itio n is tic sy stem .
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.
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.
(a) B B C -*C
A A B 3 C, B C ^
r->E .
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.
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 .
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
M ix
M ix
A D B, r - * 0. D->>2
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
A -*A ,A B, T - * 0
A, r - * A , 0 ° , A B, A, T - * A , 0
A D B A A 0 = 5 -> 2
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
M ix
456 CONSISTENCY CH. XV
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.
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.
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
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 .
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.
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.
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 (yi> ^n> y ^ id * ^i2> ^ 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
V y 3 z [{ A (x ) D A ( x , y )} & {A (x , z) D A (x)}],
h P (y , X2, . . . , x n , w) ^ P (y , x 2, . . . , x n, w).
[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).
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.
478 CONSISTENCY CH. XV
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).
C, T C.
R u les of inference for the propositional calculus.
w ith 0 e m p ty or
consisting of one form ula.
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
restriction on variables.
r-»A(t) A (b ),3 x A (x ), r - ^ e
r -> ix A (x ). 3 xA (x), T - * 0 ,
subject to the
restriction on variables.
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).
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
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 .
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 .
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
3. e T -*
2. —> or —> —i c7
1 c~j y -T c Y
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.
etc.
3. <C7 V - i e 2 ,
------------------------- !---------------* V
2.
-------------------------------------—^ G
1. —> tv7 V *ni <v7.
6. “ ^_v
1. —► —i - i (<£f V - i o3).
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.
*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 ^ , )
2. - i B -*•
1. —>• —i—i B .
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
(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
E xample 3 (concluded).
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 ] }.
§81 REDUCTIONS TO INTUITIONISTIC SYSTEMS 495
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 .
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.
3. A V B is tru e , if A is tr u e or B 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 .
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\-)”
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 ))
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
§82 RECURSIVE REALIZABILITY 507
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 ').
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.
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
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.
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.
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.
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.
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) -> ; and it is easily shown from (22) th at for the particular
P(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.
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.
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.
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