Sei sulla pagina 1di 8

Inconsistent Nonstandard Arithmetic

Author(s): Chris Mortensen


Source: The Journal of Symbolic Logic, Vol. 52, No. 2 (Jun., 1987), pp. 512-518
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/2274397
Accessed: 24-08-2016 10:52 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Association for Symbolic Logic, Cambridge University Press are collaborating with JSTOR to
digitize, preserve and extend access to The Journal of Symbolic Logic

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

THE JOURNAL OF SYMBOLIC Looic


Volume 52, Number 2, June 1987

INCONSISTENT NONSTANDARD ARITHMETIC

CHRIS MORTENSEN

Abstract. This paper continues the investigation of inconsistent arithmetical structure

?2 the basic notion of a model with identity is defined, and results needed from elsewher

cited. In ?3 several nonisomorphic inconsistent models with identity which extend the (
theory of the usual classical denumerable nonstandard model of arithmetic are exhibited.
inconsistent nonstandard models of the classical theory of finite rings and fields modulo m

Zm, are briefly considered. In ?5 two models modulo an infinite nonstandard number

considered. In the first, it is shown how to model inconsistently the arithmetic of the ratio

with all names included, a strengthening of earlier results. In the second, all inconsisten
confined to the nonstandard integers, and the effects on Fermat's Last Theorem are
considered. It is concluded that the prospects for a good inconsistent theory of fields may be
limited.

?1. Introduction. This paper is a sequel to [1] and [2], and continues the
investigation of inconsistent structures therein. In ?2 the basic notion of a model
with identity is defined, and results needed from [2] are cited. In ?3 several
nonisomorphic inconsistent models with identity which extend the (=, <) theory of
the usual classical denumerable nonstandard model of arithmetic are exhibited. In
?4 inconsistent nonstandard models of the classical theory of finite rings and fields

modulo m, i.e. Zm, are briefly considered. In ?5 two models modulo an infinite

nonstandard number are considered. In the first of these, we obtain a result stronger
than the finite methods of [2] could obtain, namely that the full theory of the field of

rationals Q, equipped with names, can be modelled in an inconsistent extension of


the nonstandard model of (+, x) arithmetic. In the second of these, a model which
bears an interesting relationship to classical nonstandard arithmetic, namely one in

which all inconsistency is confined to the nonstandard part of the diagram, is


displayed. One interest in the model is that it is possible to do better with Fermat's

Last Theorem than the results on that score in [1] suggested. It is claimed on the
basis of these results and those of [1] and [2] that the potential for inconsistent
mathematics to be a rich source of structures, problems, and results looks good.
?2. Summary of definitions and results. We consider various sublanguages of the

language Y consisting of simple terms (names) for elements of various classical


number systems, e.g. N, Z, Q and nonstandard extensions thereof; function symbols
+, x,-, ?,; atomic binary predicates =, <,; variables x, y, z,. ..; and operators -,
&, V, the latter also written ( ). Complex terms, wffs and sentences are defined in the
Received March 5, 1985; revised April 16, 1986.
C 1987, Association for Symbolic Logic
0022-4812/87/5202-0014/$01 .70

512

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

INCONSISTENT NONSTANDARD ARITHMETIC 513

usual way, as are D, v, _ and 3. We regard sentences of the form


as atomic, irrespective of whether t1, t2 contain function symbol
which contain no free variables are considered, and, for simplicity, no term is a
variable. An (RM3-) assignment is a function I assigning to the appropriate
sublanguage of Y under consideration values from { T, N, F} in accordance with: (1
For any atomic wff with terms t1 and t2, I(t1 = t2) and I(t1 < t2) both e {T, N, F}
(2) I(-i A) and I(A & B) are given by the RM3-matrices
& T N F m

T T N F F
N N N F N

(3)

I((x)A

min{y:

for

some

term

lattice ordering F < N < T. A senten


An assignment is consistent (compl
consistent (complete). The set of sentences holding in any RM3-assignment I is a
complete and possibly inconsistent RM3-theory. I is an assignment with identity iff
for all terms t1, t2, if t1 = t2 holds, then for all predicates F, Ft1 holds iff Ft2 holds,
where Ft2 is like Ft1, except that t2 replaces t1 in one or more places. An RM3-model
is a pair <D, I> where D is a domain and I is an RM3-assignment, and such that: (1) I
assigns to every simple term of the sublanguage a member of D, and I is onto D, so
that every object is named; (2) I assigns to every n-ary functional expression an n-ary
partial function on D; (3) the assignment to complex terms is given by I(f(t I ... tJ))

-I(f)(I(t1). I(tJ)), provided that these are defined; and (4) I satisfies: t1 = t2
holds iff I(t1) = I(t2). A model is infinite iff D ? No, otherwise finite. If <D, I> is a
model and I an assignment with identity, then <D, I> is a model with identity. Finally,

two models <D,I> and <D1,PI> are isomorphic iff there is a 1-1 correspondence
f: D -* D1 such that for all atomic terms t1, ..., t, ti,. ... t1,if I1(tl) =
I'(tl) = f(I(tn)), then for all atomic F, Ft, tn holds in I iff Ft1... t1 holds in I1.
Then from [2], we have the following. A necessary and sufficient condition for a
model to be a model with identity is that for all terms t1, t2 and all atomic F, if t1 = t2

holds, then I(Ft1) = I(Ft2). In the special case where - is the only predicate of the

language, <D, I> is a model with identity iff for all t1, t2, if t1 = t2 holds then for all
t3, I(t1 = t3) = I(t2 = t3) and I(t3 = t1) = I(t3 = t2).
EXTENDABILITY LEMMA. Let I, I1 be assignments to the same sets of wffs. If th

atomic sentences holding in I are a subset of those holding in IP and the negations
atomic sentences holding in I are a subset of those holding in I1, then the set of a
sentences holding in I is a subset of that holding in IP.
TERM ELIMINATION LEMMA. If <D, I> is a model with identity with at least one
element of D assigned by I to more than one simple term, then there is a model <D, I
which assigns elements to only a proper subset of those in <D, I> such that there is o
one name per element, and the sentences holding in the cut-down language are identi
in the two models.

In [2], it is argued that theories determined by models with identity represent


extensional mathematics in a quite standard sense of "extensional", whether
consistent or not.

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

514

CHRIS

?3.

MORTENSEN

The

theory

of

order.

In

thi

Extendability Lemma, by displaying two inconsistent models which extend the

theory of the classical consistent denumerable nonstandard model of arithmetic, of


order type co + q(w* + co). The inconsistent models are nonisomorphic to that
model.

In all models, including the classical denumerable nonstandard model, take one
simple term for every element of the classical nonstandard model. It is convenient in

this section to adopt a naming system which reflects their classical order type, so we
name them by pairs <q, z> where q is any rational number, 0 < q < 1, and z is any
integer, with the proviso that if q = 0 then z E N only. If we now set I(<q, z>
= <q',z'>)= Tiffq=q' andz=z',andFotherwise;andI(<q,z> < <q',z'>)= T

iff q < qI or (q = q' and z < z'), and F otherwise; and evaluate as pe
matrices for RM3-models, we have the (=, <) fragment with names of the classical
consistent denumerable nonstandard model of arithmetic, since the {T, F} subal-

gebra of the RM3-matrices { T, N, F} is classical logic. Now consider the following


inconsistent models with identity.

1. As elements of the domain, take all names <q, z> where q = 0, together with, for
each q : 0, just one representative, say <q, 1>. For every name <q, z> of the original

language re that of the classical nonstandard model, we set I(<q, z>) = <q, z> if q
= 0, and I(<q, z>) = <q, 1> for q : 0. Then atomic sentences are evaluated as
follows.

Case I (<q, z> = <q' z' >) = Tif q = q 0 = and I(<q, z>) = I(<q', z'>), while
I(<q, z> <q', z'>) = N if (either q : 0 or q' 0) and I(<q, z>) I(<q', z'>) (note
that by the construction of the model, this latter condition is equivalent to q : 0 and

q 0 and I(<q, z>) = I(<q', z'>)), while I(<q,z'> = <q',z'>) = F otherwise, i.e. if
I(<qz>) : I(<qz1>)Case <. I(t1 < t2)= T if I(t1) < 1(t2), I(t1 < t2)= N if I(t1) -I(t2), and
I(t1 < t2) = F if I(t1) > 1(t2). Now the conditions of the Extendability Lemma
are met, so every true sentence of the classical nonstandard theory continues to
hold in this inconsistent model. Furthermore, it is a model with identity. [Proof.
By ?2 and [2], we have to prove that for all atomic F, if t1 = t2 holds, then I(Ft1)
= I(Ft2). Let t1 = t2 hold, and we have to prove four cases (i) I(t1 = t3) =

I(t2 = t3); (ii) I(t3 = t1) = I(t3 =t2); (iii) I(tl < t3) = W~2 < t3); (iv) 4~3 < t
I(t3 < t2). Ad (i): If I(<q,z> <q",z">) = T then q = = 0 and I(<q,z>)
= I(<q",z">). Now <q,z> = <q',z'> holds, so I(<q,z>) = I(<q',z'>). Hence q' = 0

also. Thus q' = q 0 = O and I(<q',z'>) = I(<q",z">) = 0, i.e. I(<q',z'> = <q",


= T as required. The subcase I(<q,z> = <q",z">) = N is similar. If I(<q,z>
= <q",z">) = F then I(<q,z>) : I(<q",z">), whence I(<q',z'> : I(<q",z">) as

required. Ad (ii): The argument is similar. Ad (iii): If I(<q, z> < <q", z"
I(<q, z>) < I(<q", z">), whence I(<q', z'>) < I(<q", z">), i.e. I(<q', z'> < <q", z">)
= T as required. The N and F subcases are similar. Ad (iv): The argument is similar.
Hence, it is a model with identity.]

Also, the Term Elimination Lemma permits an obvious cut-down of simple terms
to one for each member of the domain of the inconsistent model, with the resultant
inconsistent theory agreeing with the inconsistent theory just constructed in their
common language, and thus also having all sentences of the standard model for
arithmetic holding. The models just constructed essentially collapse blocks of order-

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

INCONSISTENT NONSTANDARD ARITHMETIC 515

type co* + co in the standard model, by identifying elements within a block. The

elements of the inconsistent models have a "natural" order type, namely the natural
order on the names, which determines the assignment to t1 < t2 sentences of order

type co + q. More exactly, since there is no order-preserving 1-1 correspondence


between co + q and co + q(wo* + co), then there is no isomorphism between the
inconsistent model and the classical model, because any attempt to do so produces a
reversal of the order somewhere, and thus by the construction results in some

sentence t1 < t2 going from "holds" to "value F".


2. A more drastic identification and collapse is this. For domain, take the

standard names <0, n> together with one nonstandard name, w. For any standard

term, set I(<O, n>) = itself, and for any nonstandard term, set I(<q, z>) = co. Then
I(t1 = t2) = T if t1 or t2 is standard and I(t1) = 1(t2); I(t1 = t2) = N if t1 and t2 are
nonstandard; and I(t1 = t2) = F otherwise. I(t1 < t2) is evaluated as in the previous

model. The Extendability Lemma applies as before, so all sentences of the classical
standard and nonstandard models hold, as do all sentences of the previous
inconsistent models. It is a model with identity (proof left to reader), and the Term

Elimination Lemma is applicable. The natural order type on the domain is co

+ 1, and since there is no order-preserving 1-1 correspondence between co + 1 and

co + q(wo* + co) or co + q, then as before there is no isomorphism between this model


and the previous models, consistent or inconsistent.
?4. Finite extensions of the nonstandard model with addition, multiplication and

other field properties. In [1] and [2] the finite extensions of the < +, x ) theory of
with names derive from the divisibility properties of numbers: to obtain inconsistent
arithmetic modulo m, identify all numbers divisible by m with zero, those 1 more
than a multiple of m with 1, and so on. But nonstandard numbers can also have finite
factors. Furthermore, for every finite m, every nonstandard number is within m - 1
of some (nonstandard) multiple of m. Multiplication modulo m is uniquely definable
on nonstandard numbers. It should come as no surprise, then, that there are
nonstandard versions of modulo arithmetics, and that there are corresponding finite

structures in which inconsistent extensions of the (+, x ) classical standard and


nonstandard theories hold. For every simple name t (standard or nonstandard)

assign I(t) to be t mod m; I(t1 + t2 ) is I(t1) + (mod m) 1(t2) and correspondingly fo


x; and I(t1 = t2) is N iff I(t1) = 1(t2), and F otherwise. This is a model with identity.
Thus by the Term Elimination Lemma, there are finite structures with finite
numbers of simple names in which every term-free sentence of the nonstandard

model of arithmetic holds. The latter is exactly the standard theory, and the former
structures are of course exactly the finite inconsistent modulo arithmetics of [1] and
[2]. In turn, these inconsistent modulo arithmetics with cut-down sets of names are
obtainable from the finite classical rings Zn by inconsistentizing and applying the

Extendability Lemma. The Term Elimination Lemma thus enables us to close the
circle.

For arbitrary (standard) modulo m, a unique additive inverse - n is definable for


every n, as m - (n mod m) if n mod m : 0, and 0 otherwise; and of course the
nonstandard numbers do not disturb this property. Hence, in the fashion of [2], we
can inconsistently model (with identity) the (+, x, -) arithmetic of the full ring of

integers Z in finite modulo m with domain {1, ..., m - 1}.

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

516

CHRIS

MORTENSEN

The properties of finite primes carry over to the nonstandard numbers, in that for

prime p, a unique multiplicative inverse n-'(modp) is definable for every n

E {0, 1,... 'p - 1}. This gives nonstandard finite fields in the classical case. In the
inconsistent case, it means that we can get the (+, x, -, ) arithmetic for those

rationals whose denominator :A Omodp, and the (+, x, -) arithmetic of the


remainder (since O' is not defined). A point to note is that these inconsistent finite
domain models for nonstandard fields fail to make all sentences of the classical

theory, with names, of the rationals Q hold. The reason is that, as noted in [2], the
construction identifies with zero elements which are classically distinct from zero.
But O1 remains undefined in the inconsistent structures (alas!), so some rational

numbers which have classical inverses fail to do so here. The problem then arises of

whether it is possible to extend inconsistently the full classical theory of Q with


names using some modulo construction on nonstandard integers. The main result of

?5.1 is that the properties of nonstandard primes make this possible.

?5. Modulo infinity. An interesting class of inconsistent nonstandard structures


begins from the observations that infinite numbers can have infinite divisors, that
infinite numbers may have no finite divisors, and that infinite primes with no
divisors save themselves and unity exist.
5.1. In which -i 0 = 0 holds. The first of these allows for infinite rings modulo an

infinite number as follows. As we have seen, for any m, every number is within m - 1
of some multiple of m. So, for classical arithmetic modulo infinite m, set the modulus

of any number n equal to the least x such that x + some multiple of m = n. We know
this x to be a number, finite or infinite, between 0 and m - 1. The set of numbers
{0,1,... m - 1} has a natural addition modulo m, calculated by adding the two
together in normal fashion and if necessary subtracting m. Similarly multiplication
is given on the set by multiplying the two normally and taking the remainder after
repeated division by m leaves a remainder < m - 1. Inconsistentizing, as in the

previous section, this gives infinite inconsistent models with identity of the standard

and nonstandard theories of the natural numbers. Formally, take names for all
standard and nonstandard numbers as before. Domain D = {x: x < m}, where m is
any nonstandard number, the intended modulus. On D, we define + mod m and
x mod m in the standard fashion above. For any name n, set I(n) = n mod m; set
+ +Se
I(+) = + modm, I(x) = x modm, and I(t1 x t2) = I(x)(I(t1), 1(t2)). Set

I(t1 = t2) = N iff I(t1) = 1(t2), and F otherwise.


The "natural" ordering on the domain has a "last" element, and a "natural" order

type of co + q(wo* + co) + co*. We note that in all these models -IO = 0 holds.
As in the finite case, the addition of a natural division to classical nonstandard
infinite rings depends on the definability of a natural multiplicative inverse, and this
is possible for the infinite primes, since for them the congruence ax -1 (mod p) has a

unique solution x as in the standard case (e.g. [3, p. 41]). The Extendability Lemma
and the Term Elimination Lemma thus give inconsistent infinite fields in which all
truths of the standard model for arithmetic hold as well. An interesting feature of the

consistent theory is that the negative (additive inverse) of a finite number is a


nonstandard number in the same block as the modulus: in mod m (any m), we have

that - 1, -2, ... are m - 1, m - 2, ... etc. All finite numbers in blocks different from
the final block (that of the modulus) have their negatives also infinite. When the

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

INCONSISTENT NONSTANDARD ARITHMETIC 517

nonstandard model with names is inconsistentized, all congruences become


identities and we have, for instance, that 2m - 1 = m - 1, as well as m 2m-1 =
mr-1.

The reciprocal of any finite number is an infinite number; but in general


reciprocals of different finite numbers are in different blocks. The same kinds of
inconsistent extensions and term eliminations are available. Note though that the
reciprocals of all finite standard numbers are defined and are infinite numbers. So

since sums, products and additive inverses are defined for all finite numbers, we have

a field of order type co -+ q(o(* + co) + co* which inconsistently extends the (+, x,
-, .) theory of all the rationals with names for all rationals, which supplies the
promised strengthening of the results for finite fields cited at the end of the previous
section.

Problems. Is a nonintegral rational + n . m ever identical with a standard integer


in these models? Does reciprocation always take one to a different block? Do

different numbers in the same block always have reciprocals in different blocks? Is

arithmetic modulo infinity identical with RM' in their common language? (On
RMV' , see [1].)
5.2. Confining the inconsistency to the nonstandard numbers. We now define a

model NSN in which the standard laws of (+, x ) arithmetic remain consistent. The
inconsistency is confined to the infinite numbers, and does not spread back to --0

= 0 via an identification of 0 with the modulus, as it did in ?5.1. We look at what


happens to Fermat's Last Theorem in this model. We also show the troubles which

arise when - and ? are added to the model are, as in [2], not really a matter of
inconsistency-toleration, but of functionality. This strengthens the suspicion raised
in [2], that a useful inconsistent theory of fields may be hard to come by.
NSN is defined as follows. Names for all standard and nonstandard numbers.

For domain D, select an arbitrary nonstandard number m, and let D


= {0, 1, 2,.. .} u {x: m < x < 2m - 1}. Names are interpreted as follows. If n

E {0, 1, 2,... }, I(n) = n. Otherwise, I(n) = m + n mod m. Addition and multiplication


on the standard part of the domain are the standard operations and if at least one of

n2 E {m, ...,2m- 1}, then n1 + n2 = m + (n1 + n2) mod m. Then set I(t1 + t2)
- I(+)(I(t1), I(t2)). Finally, if at least one of t1, t2 is standard, I(t1 = t2) = T if I(t1)
= 1(t2), and F otherwise; and if both t1 and t2 are nonstandard, then I(t1 = t2) = N

if 1(t1) = 1(t2), and F otherwise. The Extendability Lemma ensures that every
sentence of classical standard and nonstandard (+, x ) arithmetic continues to hold.
[Proof. The only interesting case is the operations + and x on the nonstandard
numbers, so let t1 + t2 = t3 hold classically, and show it holds in NSN; the x case
will be similar. Now I(t1 + t2) = m + (t1 + t2) mod m, but from the previous
modulo construction, if t1 + t2 = t3 then (t1 + t2) mod m = t3 mod m. Hence

I(t1 + t2) = m + t3modm = 1(t3); therefore I(t1 + t2 = t3) = N.] The proof of
the fact that it is a model with identity is left to the reader.
One interest in the model is that t1 = t2 is T in classical standard arithmetic iff it is
T in NSN; the contradictions are confined to the nonstandard part. Another interest
in the model is that the "least" nonstandard number in the domain, m, functions as a

pseudo-zero, in that for any nonzero n, n x m = m holds, and for any nonstandard n,
n + m = n holds. This implies that the denial of Fermat's Last Theorem holds in the

model, since m3 + m3 = m3 holds, while none of m = 0, 1,2 and 3 = 0, 1,2 hold.

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

518

CHRIS

MORTENSEN

Now -iFLT, i.e. (Ix,y,z,n)(ix = 0 & &-in = 2 & x' + yf = zn), is at least
N in this model, but may be T (recall that F < N < T). The discussion at this poin
will be helped by introducing the Routley * function: if S is a theory, then S*

{A: - A 0 S}. Now it is straightforward to prove that if - FLT is N in NSN, then

neither FLT norm FLT E NSN*. But also, if t1 = t2 holds standardly, then it is T in
NSN and belongs to NSN * also. But by an uncontroversial argument, m FLT is true
(in classical standard arithmetic) iff for some x, y, z, n not classically identical with 0
1 or 2, x' + yf = z' is true in standard arithmetic. Therefore, if m FLT is N in NSN,

then FLT is true (in classical standard arithmetic). Furthermore, if m FLT is T i

NSN, then it can only be T in virtue of there being some standard x, y, z, n etc. with

xn + yf = z' being standardly true i.e. m FLT is true. In short, FLT is true iff m FL
is exactly N in NSN, and iff neither i FLT nor FLT is in NSN*; and FLT is false iff

- FLT is T in NSN, and iff - FLT e NSN* and FLT 0 NSN*. This represents an

improvement on the results in [1]. Unfortunately, the job of provingI FLT to be T


in NSN would seem to be just as hard as finding a refuting instance to FLT ever wa
We recall from ?5.1 that the use of an infinite prime modulus identified with zero

permitted negatives and inverses for all the natural numbers among the infinit

numbers, and thus the arithmetic of the rationals to be modelled inconsistently. Bu

NSN with m an infinite prime does not have true negatives and reciprocals. Th

reason is that the nonstandard part of the domain is closed with respect to + and x,

but not - and ?, so these operations could not be functional on the domain
without identifying some infinite number with a finite number, which the
construction prohibits. This situation seems to be endemic, at least to models with

identity. To take another illustration, add to the (=, <) model with domain
{0, 1, 2, ... , w} of ?3 the operations
+ finite w9 x 0 finite wo

finite finite wo 0 0 0 0
w9 w9 w9 finite 0 finite wo
(0 0 w9 w9

But here, as before, - and - canno


identity, because e.g. wo + 1 = wo
identity) = 0 would be catastrophic

point, it is not going to be easy to fin


which all the truths of the theories
REFERENCES

[1] ROBERT K. MEYER and CHRIS MORTENSEN, Inconsistent models for relevant a
JOURNAL, vol. 49 (1984), pp. 917-929.

[2] CHRIS MORTENSEN, Inconsistent number systems, Notre Dame Journal of Formal L

[3] G. BIRKHOFF and S. MACLANE, A survey of modern algebra, 3rd ed., Macmillan
[4] CHRIS MORTENSEN, Inconsistent hyperreals (in preparation).
DEPARTMENT OF PHILOSOPHY
UNIVERSITY OF ADELAIDE

NORTH TERRACE, SOUTH AUSTRALIA 5001


AUSTRALIA

This content downloaded from 212.128.154.82 on Wed, 24 Aug 2016 10:52:38 UTC
All use subject to http://about.jstor.org/terms

Potrebbero piacerti anche