LOGIC COLLOQUIUM '82 Proceedings of the Colloquium held in Florence 23-28 August, 1982
Edited by
G. LOLL1 Dipartimento di lnformatica Univer.sitadi Torino Torino Italy
G. LONG0 Dipartimenio di lnformatica Universita di Pisa
Pisn Itnly
and
A. MARCJA Dipartimento di Matematica Lihern Universita degli Stud di Trento Trento Italy
1984
NORTH-HOLLAN'D AMSTERDAM 0 NEW YORK. OXFORD
QELSEVIER SCIENCE PUBLISHERS B . V . , 1984 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner.
ISBN: 0 444 86876 3
Published by: Elsevier Science Publishers B . V P.O. Box 1991 1000 BZ Amsterdam The Netherlands Sole distributors for the U . S .A . and Canada. Elsevier Science Publishing Company, Inq; 52 Vanderbilt Avenue New York, N . Y . 10017 U.S.A.
Library of Congress Cataloging in Publication Data
Logic Colloquium (1982 : Florence, Italy) Logic colloquium '82. (Studies in logic and the foundations of mathematics ; v. 112)
Bibliography: p. 1. Model theory--Congresses. matics)--Congresses. 3. Lambda I. Lolli, Gabriele, 1942111. Marcja, A. (Annalisa) IV.
.
@9*7*L64 1982 511' .8 ISBN 0-444-86876-3 ( U . S . )
2. Categories (Mathecalculus--Congresses. 11. Longo, Giuseppe. Title. V. Series.
84-1630
PRINTED IN T H E NETHERLANDS
V
PREFACE
The Logic Colloquium ‘82 has been held in Florence (Italy), from 23 to 28 August, 1982. The date had been so chosen in order to allow a combined participation to the Warsaw ICM. Notwithstanding the postponement of the latter, more than 200 logicians were able to attend the Logic Colloquium and the Summer Meeting of the ASL. The organization and the program of the conference were undertaken by a Committee consisting of E. Casari (Florence), J.E. Fenstad (Oslo), G . Lou (Turin), G . Longo (Pisa), A. Marcja (Trento), and D. van Dalen (Utrecht). The organization has been made possible by the financial support of many institutions: a substantial help came through the Comitato per la Matematica of the Italian Consiglio Nazionale delle Ricerche; further contributions were made accessible from the University of Florence, the city of Florence and the Regione Toscana, and the Ente Provinciale per il Turismo di Firenze. The conference was sponsored by the International Union of History and Philosophy of Science, Division of Logic, Methodology and Philosophy of Science. The main topics chosen for the Colloquium were: Model Theory, with particular emphasis on models of Peano arithmetic (a small if not exactly pertinent contribution to the celebrations for the 50th anniversary of Peano’s death); Categorical Logic; Lambda-calculus. Much care and effort have been put by the invited speakers into investigating and further stretching the overlapping and crossbreeding of these areas. We regret that the present volume does not faithfully reflect the balance of the three topics at the Logic Colloquium, since some of the speakers in categorical logic and in lambda-calculus were unable to submit the written text of their lectures (while the editors were guilty against the readers of too much waiting). We hope that their contributions will however be made accessible in other ways to all interested logicians. We are grateful to Elsevier Science Publishers B.V. for inserting this volume in their glorious series in Logic and the Foundations of Mathematics. November 1983
G . Lolli (Torino) G . Longo (Pisa) A. Marja (Trento)
LOGIC COLLOQUIUM '82 G. Lolli, G.Long0 and A. Marqa (editors) @Elsevier Science Publishers B. V. (North-Holland), 1984
1
LECTURES ON NONSTANDARD MODELS OF ARITHMETIC
Commemorating Guiseppe Peano C. Smoryikki Department of Mathematics The Ohio State University Columbus, OH 43210
USA
Contents § 0.
1. 2. 3.
4. 5. 6. 7. 8. 9. 10.
Preface The Beginnings (The 1950s and Earlier) The R6le of the Infinite Integer (The 1960s. I) Extensions of Models (The 196Os, 11) Saturation Properties (The 1970s. I) Recursively Saturated Models of Rich Theories (A Digression) The Arithmetised Completeness Theorem (The 197Os, 11) Powerful Arithmetisations (The 1970s, 111) Diversity (The 1980s) A Dead End Summer Reading List
0. PREFACE Guiseppe Peano died in 1932; nonstandard models of arithmetic were born the following year. Thus, Peano never studied models of arithmetic and it may seem odd to commemorate him with a series of lectures on such a topic. However, one of the more active areas in the study of what has come to be called Peano arithmetic, o r P A , is the study of its nonstandard models and this subject seems thus appropriate. Besides, with the subject beginning just after his death, we might say it picked up where he left off. It is my intention subject. so.
to
give a partly historical account of the development of the
Since these are commemorative lectures, it seems most appropriate to do
Moreover, there are now two collections of papers edited by Kenneth McAloon
e t aZ. with expositions of recent work, while there is nothing about the pioneering efforts of the 1950s and 1960s.
Progress was much greater in this period
than most people realise.
I will depart from the historical line in two important respects.
First, much
early work concerned only "strong" nonstandard models of arithmetic rather than arbitrary models of P A .
A nonstandard model was called "strong" if it elemen-
tarily extended the standard model No= (w;+,.,',O)
and was of particular interest
only because of a lack of familiarity with the strength of P A :
One needed to
know, say, that all integers (standard and nonstandard) had certain coding prop-
c. S M O R ~ S K I erties that ordinary integers had and, unaware that such properties were provable outright in P A , one simply assumed them to hold by assuming the model to be strong. A second departure from the historical line must occur when several independent trends develop simultaneously. This first happened around 1960, when nonstandard model theory reached its first level of maturity. Because of time and space limitations, I shall have to assume the reader familiar with a great deal of what can be done within PA.
This means that, with respect to,
say, the hypothesis of strength mentioned above, I can discuss the full results available without the irrelevant hypothesis-- even though the full result came a decade later.
It also means that some important developments, which preceded
formalisation, will look more trivial than they should. My notation will be fairly standard-- at least by my standards-- and I hope everything will be fairly self-explanatory. 1.
THE BEGINNINGS (THE 1950s AND EARLIER)
It was in a paper published in 1929 that Thoralf Skolem first suggested the possibility that the standard numbers might not be alone, that there might be poor In SkoZem 1 9 3 3 , he was able to
imitations satisfying the same first-order laws.
psove this relative to any finite set of axioms. This, of course, could have been established by appeal to Gb'del's Incompleteness and Completeness Theorems; or, it could have been done by the now-common appeal to the Compactness Theorem.
The
former approach would seem to have little bearing on the larger problem of obtaining nonstandard models satisfying a l l the true sentences of arithmetic; and neither of these abstract approaches would very likely have been to Skolem's taste. In any event, in the following year in SkoZem 1934 he published a proof of the existence of strong nonstandard models of arithmetic, i.e. structures 1'4 =
(M;+,-,',O)
not isomorphic to No but, nonetheless, making true the same sentences
as N o . Skolem's original papers were in German.
He republished the proof in English in
SkoZem 1955 in the proceedings of a conference-- the same proceedings in which bos published his theorem on ultraproducts. Although this was merely a repetition
of his earlier proof, it is this later paper to which most people refer.
Perhaps
this is because this paper is in English, a much easier language than German for many of us; perhaps it is simply that this paper, appearing in a slim little volume that has been reprinted, is in more private libraries and hence more accessible; or perhaps it is the juxtaposition with bos' paper that strikes one's fancy-- for, Skolem's construction of a nonstandard model is something of an ultrapower construction.
Sketch of Skolem's construction:
Given No
=
(u;+,*,',O),
let F consist of
3
Lectures on Nonstandard Models of Arithmetic all functions F:w
+
w
definable in No.
... of F it is
From an enumeration FO,F1,
not difficult to construct, by diagonalisation, a function G:w all F ,F. i J
E
+
w
such that, for
F,
< F.G(x) eventually F G(x) i J or eventually F.G(x) > F . G ( r ) .
or
eventually F.G(x) = F.G(x) J
J
Using G, one can define an equivalence relation :by F. 1
5
F. iff eventually F.G(x) = F.G(r) J
J
and shew the structure F/: to be an elementary extension of N 0 '
It should be noted that, although Skolem's construction resembles the ultrapower construction so much that one feels like calling it such, an important element is missing.
Skolem's use of the diagonalising G in place of the now-usual ultrafil-
ter relies heavily on the countability of the arithmetic language; his method does not yield the existence of nonstandard models when, say, a continuum of predicates naming all sets of natural numbers is added to the language. For this latter, one must use one of the standard abstract existence theorems of logic. It is also worth mentioning that Skolem's goal in constructing nonstandard models was philosophical: He aimed to shew that first-order logic could not characterise the number series; he did not care to start a new subject.
Until the 1960s, this
was generally the case-- nonstandard models of arithmetic were either objects of philosophical interest or tools, not objects of mathematical interest in their own right.
The major counterexample to this was an observation made by Leon Hen-
kin in his paper Henkin 1950 on the Completeness Theorem for Type Theory. announced the order type of a nonstandard model of arithmetic to be w where 8 is a dense linear order.
( Exercise: If, in particular,
+
He
(w*+u)8,
0 is countable,
it must be the order type of the rationals. A related, but trickier, exercise due (I believe) to Klaus Potthoff is this: not have the order type of the reals.
Shew, in the uncountable case, 0 can
)
The important, i.e. useful, fact about the order type of a model M = (M;+,*,*,O) of PA is that M begins with w and then follows this with the nonstandard or infi-
nite integers. The significance of the infinite size of the nonstandard integers is manifold: They code various paths to infinity and, particularly in Nonstandard Analysis, simulste limit processes; they code infinite sets; and they code nonprincipal ultrafilters in the algebra of definable sets of natural numbers. all this, their potential usefulness is clear.
With
In the early 1950s, however, not
all of this was known: The first application of infinite integers depended merely on their size. In RyZZ-Nardzewski 1952Czesaaw Ryll-Nardzewski essentially proved the following theorem: 1.1.
Theorem.
PA is not finitely axiomatisable.
c. S M O R ~ S K I
4
Ryll-Nardzewski actually proved something a bit stronger: If T is any finitely axiomatised theory in a language extending that of PA and T is true in some expansion of No to accommodate the extended language, then T does not prove some instance of induction in this language. This result is still rather weak.
As I
mentioned in the Preface, many early results were proven in weak form because it was necessary to assume certain arithmetic truths held in a nonstandard model and, through lack of familiarity with the power of PA, the best guarantee of this assumption was that the nonstandard model was a strong one.
Dropping this assump-
tion, the final result along these lines is the following: 1.2.
Theorem.
PA is essentially unbounded, i.e. no consistent extension T of PA
can be given by axioms of a fixed bounded complexity. This theorem was first published by Michael Rabin in Rubin 1962 and is thus often referred to as Rabin's Theorem, although he explicitly announced it to have been known to others-- most probably including Solomon Feferman, Georg Kreisel, Dana Scott, Stanley Tennenbaum, and Hao Wang.
A s the reader might guess from the long
list of names, the road from Theorem 1.1 to Theorem 1.2 was a long one.
Indeed,
it cap now be recognised to have been one of the two major themes in nonstandard model theory in the 1950s, if one may refer to the then-primitive development as "nonstandard model theory".
Today, we have many proofs of Theorems 1.1 and 1.2,
among which is a modification of Ryll-Nardzewski's proof of Theorem 1.1. Let M be an arbitrary model of T ? P A and let MI be an elem-
Proof of 1.2:
0
entary extension of M which contains at least one nonstandard integer a which is 0
infinite relative to M ~ i.e. , a > b for every element b
E
J M ~ J .Suppose n is a
finite positive integer and U is a set of C -sentences true in Mo. struct a model Mz
b
U but such that M2
y
The construction is very simple: Let IN
We will con-
PA. 2
I
be the closure of IN
I
{ a } under all M 2 is automatic-
0 "
total functions definable in Ml by parameter-free Zn-formulae. ally a Cn-elementary substructure of M I , whence it satisfies U.
The reason that M 2 is not a model of PA is also simple: There is a Cn+I-definable function F which can schematically be proven in PA to eventually dominate each C definable function, i.e. for each parameter-free C -formula natural number
I such
@VoVl
-
there is a
that
where Ji defines F. ( F is obtained by a simple diagonalisation on the uniform C enumeration of Zn-definable partial functions. ) In Mz, F is not total; in fact,
-
Fa is not defined.
To establish this last claim, suppose Fa exists, i.e. M 2 k 3 V 1V 2$'av102, where I )' is Il
and 3v2$'defines F .
Let b be the image of a under F in M2 and let c witness
5
Lectures on Nonstandard Models of Arithmetic this fact, i.e. M2 Fa = b in MI.
I=
$ 'abc.
function G such that b
=
Since $ ' E
M2
k $'&
E
calling that a > d for every d E lMo 1 , function H.
nn,
1M21, there is some d G(a,d) in Ml. Letting Hvo
Now, since b
=> Ml
IW 0 I
E
k I)'&?
, whence
and some C -definable
supv
,V ) , and re0 1 we have b < Ha in MI for some 1 -definable =
G(V
I- 0
But this readily yields a contradiction: For, there is a finite x
such that P.4l-V
vo
> HV ) and a is quite infinite, whence b
> :(FVO
0
=
Fa > Ha in
&ED I wish to emphasise here the r61e played by the size of a relative to Mo: The integer a is so large that a Zn+l-function eventualzy dominating all C -functions already dominates them at a.
Thus, the mere closure under Z -functions, which
yields a C -elementary substructure, does not yield closure under the provably total majorising function.
Through such an application, the mere size of an in-
finite integer presents itself as a useful tool, one which could still be exploited unaided in the 1950s in the work of Feferman, Scott, and Tennenbaum, and even as recently as 1975 in a paper of Alex Wilkie. As I remarked just before proving Theorem 1.2, the passage from Theorem 1.1 to Theorem 1.2 was one of two main themes in the study of nonstandard models.
The
other, which nowadays one can see to partially merge with this one in the work of Feferman, Scott, and Tennenbaum, is a couple of decades older and more technical. This theme is the complexity of models and goes back to the second volume of Hilbert and Bernays, specifically to a result Paul Bernays proved and included in the volume: 1.3.
Theorem.
Let
0 be
any sentence of a given language.
If $I is consistent,
i.e. if $I has a model, then $I has an arithmetically definable model, i.e. a model whose domain is an arithmetically definable set of natural numbers and whose primitive relations are also arithmetically definable. Actually, the entire satisfaction relation for the model can be taken to be arithmetically definable-- the satisfaction relation for the language of @ is simply a new predicate describable by a new axiom @ and one can apply the result to $ A $ . The fact that the description could be given by a single axiom @ was not immediately recognised.
This was the decade* in which Stephen Kleene (KZeene 1952.4)
(and later William Craig and Robert Vaught) proved one could finitely axiomatise r.e. theories by the addition of new predicates.
Thus, when the early researchers
became interested in extending Bernays' result to r.e. theories, the direct reduction to the finite case by finitising the theory in telrms of a satisfaction relation did not occur to them and they proved the result anew. Bernays had proved Theorem 1.3 by arithmetically analysing the proof o f Godel's
*In fact, Robinson's finitely axiomatised Q was not published until 1953.
c. S M O R Y ~ K I
6
Completeness Theorem for the predicate calculus.
The new result was established
(by Gisbert Hasenjsger, Stephen Kleene, and Hao Wang) by doing the same to modifications of Henkin's proof. We will discuss this later in 96, where we will see that this final completion of Bernays' Theorem is a useful tool. This realisation came, however, around 1970; in the 1950s the question was simply one of the complexity of models. The upper bound on the complexity of a model of an r.e. theory is readily established by inspection: If T is a consistent r.e. theory, then the arithmetic encoding of the proof of the Completeness Theorem shews T to have a model on an initial segment of the set of natural numbers and the definition of satisfaction for this model (and consequently each of its primitive relations) to be A2.
[ Recursion theoretically, this means the model is recursive in
-
0'.
Later
improvements were made by Joseph Shoenfield and, ultimately, by Carl Jockusch and Robert Soare; cf. ShoenfieZd 1960 and Jockusch and Soare 2972A8B. ) For lower bounds on the complexities of the interpretations, one can again ask two questions: How complicated must the primitive relations be; and, how complicated must the satisfaction relation be? Well, one r.e. theory's satisfaction relation is another finitely axiomatised theory's primitive relation and, globally, the questions conflate.
For specific theories (like P A ) or fixed languages (e.g.
that of one binary relation symbol), however, the equivalence is nontrivial and one usually asks for large laver bounds on the complexities of the primitive relations. In the 1950s, Kreisel and Andrzej Mostowski alternated in a series of papers, the
main outcomes of which were the existence of r.e. and finitely axiomatised theories with no recursive models.
Mostowski even shewed that, whereas the set
of sentences true in all models is r.e., that of all sentences true in all recur-
sive models is not even arithmetically definable. In 1960, Vaught offered yet another generalisation. In the late 1950s, the problem of the difficulty of constructing nonstandard models of PA, i.e. the complexity problem for models of P A , was considered by Feferman, Scott, and Tennenbaum, who published a short series of abstracts. Among other things, they reconsidered Skolem's ultrapower-like construction and shewed, e.g., that no homomorphic image of the Xn-functions could model PA.
This was
largely an application of size, 6 Za Ryll-Nardzewski (or, at least, my exposition of the latte&
proof, above).
However, their work led Tennenbaum to an important
discovery: Infinite integers not only have infinite size, but they also code infinite sets.
Looking at the encoding yields:
1.4. Theorem (Tennenbaum's Theorem). Let M = (w;+,X,',O) be a nonstandard model of PA and let 4m be any formula of the language of arithmetic (with parameters
Lectures on Nonstandard Models of Arithmetic from M allowed).
X
=
Then In fact, X is recursive in each of
is recursive in i , x .
( Notes: i. Note that
M k G }
{ x E w :
+,x.
denotes the successor function of the model; 0 the zero element. is recursive in
isince X I
=
x
iI
, where 1 is the unit of M.
Without
loss of generality, we can assume 0 to coincide with 0 and use the simpler notation. and x,was not
In the first published proof in Ehrenfeuckt and KreiseZ
1966 it is noted that X is recursive in
credits the observation on
+.
to Feferman.
X
Kenneth McAloon, in M c A l o o n 1 9 8 2 ,
)
First, we recall that an integer x can serve as a code for a
Proof of 1 . 4 :
finite set of integers.
There are many ways in which this can be done, my least For an integer X, let Dz
favourite being one expecially suited for this proof. consist of all 9 ' s such that the (y +l)-th where p o , p l ,
+
The extra flourish, that X is recursive in each of
ii.
stated by Tennenbaum.
prime divides x :
... is the sequence 2 , 3 , ... of primes.
y
E
iff p (z,
Dx
Y
A simple induction on v 0 proves the Aussonderungsaxiom, PA
1 vVo3VIv
V2(
V 2 E
Du 1
E V2
VoA@V2
),
for any formula @ v 2 . Let a be an infinite integer in M, let $ be given, and let E w: M 1. Applying Aussonderung to @ and a, we obtain b E lMl such
6
X = { x
that X = { x
E
w: M
13: E
iff
This shews X to be r.e. in shews o
-
X to be r.e. in
Recursivity in
X
3 1.
But then
M I G I iff E 3 c E w ( ~ k rX * c = E ) - 3 , E w ( M k C i... ic b ) ( p , times).
iff
EX
f.
Since w - X corresponds to
i , whence
X is recursive in
is a similar affair:
the same proof
l@,
f.
From the equivalence between z E X and the
representability of b as a p -fold sum of c's, we get the equivalence with the X
representability of Zb as a p -fold product: x X b some d , where e = 2
E
X iff d
.
x
...
x
d
e , for
&ED
Tennenbaum's Theorem has some immediate corollaries:
1.5. Then
Corollary. +,X
Let M = (w;i,x,',O)
be a strong nonstandard model of arithmetic.
are not arithmetical.
This can, of course, be hierarchically refined. 1.6.
Corollary.
If M = ( w ; f , ~ , ~ , O )
The main such refinement is:
is a nonstandard model of P A , then +,x are
not recursive.
Proof:
It suffices to choose @ so that
This is a simple recursion theoretic trick:
x
E
o: M k
61
is not recursive.
If $,J, define two provably disjoint
effectively inseparable r.e. sets, say A,B, respectively, then A 5 X =
c. S M O R ~ S K I
8 { J:
E
61
o: M
and B n X = @, whence X is not recursive.
I have one more important corollary to cite.
BED
Before giving this, however, I must
remark that the proof of Corollary 1.6 did not depend on the full power of Tennenbaum's Theorem. fic formula
It really only depended on establishing Theorem 1.4 for one speci-
@.For this, the proof needs:
i. some minimal arithmetic-- say,
Robinson's Q; ii. the totalities of the exponential and prime enumerating functions; and iii. two instances of induction to establish Aussonderung for
0,I$.
If we
conjoin these to obtain a single axiom 9, Corollary 1.6 can be rewritten as
1.6'.
Corollary. If M
= (w;+,x,',O)
b
$ is nonstandard, then
+,X
are not recur-
sive. Conjoining with ii, any axiom forcing the model to be nonstandard yields a sentence having no recursive models.
In fact, with ii, we can easily prove Mostowski's
result:
1.7.
Corollary. The set of sentences valid in all recursive models is not
arithmetical.
Proof:
The only recursive model of $ is the standard one.
Hence, if fJ is any
arithmetic sentence, $ 4 0 is valid in all recursive models iff 0 is true in the standard one.
Since arithmetical truth is not arithmetical, the Corollary follows. QED
The sentence ii, consisting mainly of two odd instances of induction conjoined with a few natural axioms is a bit odd and one can ask for an aesthetically pleasing finite theory to which Corollary 1.6 can be applied.
It is not hard to see that
(over &) Z1 -induction suffices to establish the totalities of the exponential and prime enumerating functions and C
-
1
and Il -Aussonderung. 1
One can also ask for the weakest theory which can be used in place of $.
Z -1nduc1 tion is rather strong and one can use weaker theories, as shewn in McAZoon 1982.
Tennenbaum's Theorem can also be generalised in another direction, as we shall see in 55, below. I might also add that the question of the complexity of a model has of late resumed an interest; cf., e.g., Marker 1982. A
2.
THE ROLE OF THE INFINITE INTEGER (THE 1960s I)
The year 1960 was something of a watershed in the development of nonstandard model theory. Prior to 1960 most papers on nonstandard models of arithmetic were aimed at metamathematical targets, the nonstandard models themselves not being of central interest.
Then, suddenly, this changed: In the early part of the decade any num-
ber of papers appeared in which the models themselves were studied. There were papers on the additive group of integers (negative ones added) of a nonstandard
9
Lectures on Nonstandard Models of Arithmetic model, on the nonstandard model as a semi-ring, on the algebra of definable sets in a nonstandard model, and on the forms of embeddability on nonstandard models into one another. The variety is such that a straight historical discussion of the period is pointless.
Instead, I shall outline a few central themes.
First, let me hark back to the r8les played by the infinite integers. While mere size considerations remained important (cf. for example Ehrenfeucht and KreiseZ 1 9 6 6 ) , greater sophistication arose.
The following theorem relating the infinite
to the finite integers is a cornerstone of Abraham Robinson's Nonstandard Analysis; in the proper language, it is the lemma upon which the equivalence of the standard and nonstandard definitions of the limits of sequences depends.
2.1.
Theorem. Let M be a strong nonstandard model of arithmetic, i.e. a proper
elementary extension of No. i. ii.
Proof: <=.
Let
3 infinite a( M infinite a( M
$V
have only v free.
6)
b vv0 3vl
iff N~ iff N~
b Q;; )
+ xuo vvl
> v o @vl
> vo $vl.
Because i and ii are duals, I shall only prove i. Observe: No
b
V V o 3v1 > v 0 @vl =>
M
k
Vv0 3Vl >
Vo
@v1. Picking u
infinite, the right-hand-side of this implication produces an infinite v
=>.
Let a be infinite such that M
MI=
a > ;*&
->
M1=3 v
Since cc was arbitrary, we conclude N 0
>
b 6.Let x E o and 3: @vl => No b 3v1
0
1'
observe
>
3: @ v l .
V V 0 3v1 > V 0 @ v l *
QED
This theorem is slightly disappointing in that it assumes M to eZementariZy extend
No.
There are hierarchical variants which the reader can work out for himself and
also something called Overspill to handle the non-elementary case.
I shall shortly
discuss the latter; but, before doing s o , I would like to remark that the present theorem, however much it assumes, is rather useful.
It is, for example, the key
to one of the corollaries of the main theorem of Jeff Paris and Leo Harrington: Paris and Harrington proved the independence of various true 112 sentences @ = V V 0 3 v 1 ~ 0 V 1 , with ii, E
the least y such that No tions.
A,.
The (recursive) function-- call it G-- taking x to
I= wT
eventually dominates all provably recursive func-
These two facts are equivalent, the latter trivially implying the former.
One can, with some effort, prove directly the majorisation result and thereby conclude the underivability of $; conversely, and more easily, one can appeal to Theorem 2.1.11 and the model theoretic construction underlying the independence proof:
Most expositions of the Paris-Harrington Theorem demonstrate the underiv-
ability of $ by starting with an elementary extension M of No in which @ is true. For each infinite integer a
E
/MI and its image b = Ga, they shew the existence of
an initial segment I 5 IWI such that i. a the inherited functions. Since
E
I and b t! I,and ii. I I= PA when given
I is a model of
P A , it is closed under all provably
recursive functions. For such a function F, from the facts that Fa
E
I,Ga t! I,
c. S M O R Y ~ K I
10
and I is an initial segment of M, it follows that Fa < Ga. Theorem 2.l.ii yields: N o
b
Bvo vvl > v o (FVl >
Since a was arbitrary,
GOl).
I have omitted a few things here, such as an absoluteness of the provably recursive functions, and I have used much stronger conditions on M than are necessary for the above argument; but I have illustrated the use of the Theorem. The reader familiar with Ramsey’s Theorem may wish to use the other half of Theorem 2.1 to
( Similarly, the reader familiar
reduce the finite form to the infinite form.
with the work of Paris and Harrington may wish
to
reduce their variant to the
infinite Ramsey Theorem. The lazy reader may wish to wait for 56, below.
)
In the non-elementary case there is, as I said, a replacement for Theorem 2.1: 2.2.
(Weak Overspill).
v
5
0 1
Let
$v0V2
( M I= & )
iff
(Strong Overspill).
Let
x ii. $71
Let M
Theorem (Overspill Principle). i.
E
w
defines a function F : I M ( Vz
w
E
I= PA
have only
be nonstandard and b
Vo,VI
3 infinite a ( M @v0VIv2
+
E
(MI.
free. Then
I=
vv0 <
ii $ v o b ) .
have only vo,V1,v2 free and suppose
IM(. Then 3 infinite a
( Fx is infinite ) iff
v
c < a (Fc is infinite ).
To prove this we first need a small lemma: Lemma. Let M
2.3.
I= PA
be nonstandard. Then: w is not parametrically definable
in M.
Proof: Let b
E
(MI and @ v ~ with u ~ only vO,vl free be given and suppose w
lMl: M I= @I 1 . Obviously, M I= $Ob and M k V v o ( $vob -+ $(vo’,b) ), whence M Vv 4V 5 and w = (MI, contrary to the nonstandard hypothesis. a
=
E
0
&ED
0
( The reader who has not yet solved the exercise in 51 on the order type w + (w*
+ w)8,
namely that of shewing 8 not to have the order type of the reals will
now get a hint:
Let a
E
Proof of OverspiZZ:
Assume
Vx
E
Vx
w M
infinite a ii.
E
I=
w
(M
@ccb
IMI be nonstandard and consider
m:
z
E
w
1. )
i.
The right-to-left implication is trivial.
1.
Let
JrvoV1
be the formula V v 2 <
Vo~V2Vl.
Clearly
I@. Since w is not parametrically definable in M, there must be some
E
IM( such that M
I=
Vv0 <
a
Again the one direction is trivial.
To establish the other, let F
denote the function defined and apply Weak Overspill to the formula
FVO
> vo:
For
all finite I, Fx is infinite, whence Fx > x, Thus, for a few infinite a ’ s following w, Fa > a , i.e. Fa is infinite.
&ED
The Overspill Principle is, in its weak form, apparently due independently to Rabin and Abraham Robinson and, in its strong form, to Robinson.
Weak Overspill
is used a great deal in the study of nonstandard models of arithmetic, as we shall
Lectures on Nonstandard Models of Arithmetic
11
shortly see.
For now, however, a simple application will suffice: Suppose a E 1 1 1
is infinite.
Then there is an infinite b
finite, so is ZX; whence 2" < a .
E
IMI such that 2b < a .
For, if x is
By Overspill there is also an infinite b such
b < a.
that 2
In its strong form, Overspill is extremely useful in Nonstandard Analysis.
Applied
in that language, with F replaced by l/F, it becomes the Infinitesimal Prolongation Theorem:
... is an internal sequence of infinitesimals, then for a few
If r o , r l ,
infinite integers a , r
remains infinitesimal. One can consult, e.g. Martin Davis'
book for examples of the usefulness of this principle.
In the study of nonstandard
models of arithmetic, however, Strong Overspill has not made much of an appearance until recently, and then only tangentially as an insight behind the Indicator Theory of Laurie Kirby and Jeff Paris.
Even here, unless there is some recent
work I am unaware o f , there are no results which cannot as easily be proven by other means.
( Hence,
I shall not discuss indicators in the sequel. )
Both Overspill and the preceding Theorem 2.1 are manifestations of the r81e of an infinite integer as a path to infinity: An infinite integer behaves somehow like the integers it encounters in its path.
In an ultrapower, an infinite integer is
literally such a path; in the more general case, my words are mystical but of some heuristic value.
In any event, I view these last two theorems as sophisticated
views of the sizes of infinite integers. in $ 1 was their coding power.
The other property of integers we used
Tennenbaum's Theorem was a crude mixture of this
coding power and size. We now take a more sophisticated look at this. 2.4.
Definitions. Let M be a model of arithmetic. A set X c w is standard on M
if there are a formula @ J ~ with u ~ only u O , V l free and an element b
E IMI such that x E w: M I= I@ 1. The collection of standard sets of M is called the standard system of M, written SSyIMl. A number a E IMI codes X E SSy(MI iff X =
X
=
I
.c E w:
MI=;
E
D;; 3 .
Here, by Dx I mean the finite set canonically indexed by
X.
We already used the
prime decomposition of numbers as a means of encoding finite sets.
Generally,
however, it is more convenient (i.e. I prefer) to use the following: Do
=
{
1;
D
=
,..., xn- I
x0
1,
+ ... + ZXn-I ... > xn-I'
if x = 2"O
x* >
and
Before discussing standard sets and their codes, I should warn the reader that the terminology presented here is not universally used.
Following the deplorable
logical tradition of referring to sets of natural numbers as real numbers, the standard sets are often called the reazs of the model, SSyfMI being thus denoted
RM.
Moreover, for reasons soon to be evident, SSyfMI = R
Scott set.
M is often called a
(Well, actually SSy(MI is proven to be a Scott set and then called
the Scott set of the model.
)
c. S M O R Y ~ S K I
12
Tennenbaum's Theorem depended on the fact that in a nonstandard model every standard set possessed a code.
2.5.
A bit more is true:
Let M I = PA be nonstandard, X
Lemma.
E
SSyfMl.
X has arbitrarily small
infinite codes.
Proof:
Let X be defined through $vv,F, a an arbitrary infinite integer, c
infinite such that 2c < a , and F defined by FV
=
Then
{
1,
@vF
0,
+vF.
C-2
d
c
=
~v
+ zC-'
< 2'
<
a
i=l
is a code for X. The lemma yields, in the nonstandard case, a characterisation of standard sets: They are precisely those which have codes. The culmination of the work of Feferman, Scott, and Tennenbaum was to have been a joint paper of Scott and Tennenbaum. Their manuscript was, unfortunately, lost; there was, however, a paper by Scott in which a characterisation of standard systems was given.
x
x is completion
2.6. Definition. Let be a class of sets of natural numbers. c l o s ed, or c-ctosed, if it satisfies: i.
x is an algebra of sets, i.e. x is closed under the boolean operations
ii.
x is closed under relative recursiveness, i.e. imply Y
iii.
x
E
X
E
X
satisfies a weak form of Kgnig's Lemma:
If X
E
x and Y
X
x codes an infin-
ite binary tree (in the strong sense that the elements of X code finite sequences of 0 ' s and 1 ' s ) .
then some Y
E
x codes an infinite
path through X.
( A s remarked above, this terminology is not universal; the term "Scott set" is often applied. )
The choice of the term "completion closed" will be evident shortly.
The lost Scott-Tennenbaum paper included a proof of the following.
2.7.
Let M I=PA.
Theorem.
Then SSyM is c-closed.
Proof sketch (nonstandard case: for the standard case cf. 56): nonstandard and
i.
That
x = SSy(M/.
Let M be
x is closed under boolean operations is obvious as the formulae
defining the standard sets are closed under the logical connectives. ii.
That
x is closed under relative recursiveness follows from the charac-
13
Lectures on Nonstandard Models of Arithmetic
The details are routine,
terisation of relative recursiveness as being "A -in". 1 but grubby, and I omit this part of the proof.
Suppose X E
iii. This is the interesting part of the proof.
x codes an in-
finite binary tree. By K6nig's Lemma, there is an infinite path Yo through X; but Y
0
might not lie in
x.
Write X
3: E
=
to:
M k @
I= 38 ( $uz,lh(u) 0 lh(v) denotes the length of the sequence coded by V . to conclude M
the existence of Y
a,c
f
IMI, c infinite, such that M k @,,lh(a)
=
x codes an infinite path through X.
} for some
)
=
Then Y
E
o, where
{ (a)3::
=
w )
3: E
E
QED He actually proved
To state this, recall that a set
sentable in a theory T if there is a formula $G 0 with only X = { Z E O :
and b and use
Apply Overspill to obtain
c.
In Scott 2962, Scott proved a converse in the countable case. a relatively strong converse.
I$
for each y
Vo
Xs
w is repre-
free such that
?'!-&I T C l & l .
w - x = i 3 : E w :
The collection of sets representable in T we denote R e p i T l .
With all this we can
now state Scott's Theorem: 2.8.
Theorem (Characterisation of Standard Systems).
of sets of natural numbers. i. ii. iii. Proof:
Let
x be a countable family
The following are equivalent:
x is c-closed x R e p @ ) for some complete T Z PA in the language of arithmetic =
X
= S S ~ ( M for )
some M
PA. Implication ii => iii follows from
We have already proven iii => i.
the existence of a minimum model M
T
to prove i => ii.
of T: SSy(MT) = R e p ( T ) .
Thus, it suffices
For the sake of convenience, I will only shew i=> iii. This
requires a lemma:
2.9.
Lemma.
Let
x be c-closed and let X
E
sistent theory T (in a recursive language). is also coded in
x.
x code a set of axioms for some conThen some consistent completion of P
roof: Let, for some axiomatisation of T , E
x.
x= I
'$':
is an axiom of T
A l s o , let I $ ~ , @ ~be , .a. .recursive enumeration of all sentences in the lan-
guage of T .
A completion of
T is essentially an infinite path through the binary
tree
more exactly, it is an infinite such path consistent with 2'. is Ill in T, not recursive in T, and the fact that T
E
Consistency with P
x does n o t
guarantee
x to
c. S M O R Y ~ ~ S K I
14
include the tree of paths consistent with T .
eo
el
;!:$I,..., those paths $o ',I$]
However, it does include the set of
which have no proofs of their inconsistency with T
.~
of code at most k .
x+
=
I
/x\
where I) =
'
yclhkc) predicate calculus of The
+ set X
V].
+
coding an infinite path
+ through X .
E
Let T' have axioms I) for x X
leave to the reader to shew, a consistent, complete theory. it is recursive in its axiomatisation and T' = Before returning to the proof of Theorem 2.8,
'$':
x
x.
Choose Y f Y. T' is, as I Since T' is complete,
is an infinite binary tree recursive in X, whence X
T'b $
1
E
E
x.
&ED
let me pause to remark that this
lemma not only explains the terminology adopted, but it also provides a very Not only will we use it immediately,
useful property of completion-closed algebras. bur also we will apply it later.
Proof vf 2.8, continued:
x be a countable c-closed family of sets of ... . We construct a modelM with
Let
natural numbers with an enumeration Xo,X1, SSyIM) = Let C =
x in stages by a Henkin argument.
{ Co, Cl,...
be a new set of constants and let L
,...,cn-Ij.
arithmetic augmented by { go
S t a g e 0. S t a g e 2n
E
x
Let T0 in Lo be any completion in of PA. i1. Let TZni3 be TZn u { E % : x E Xn U{
Note that, if T2n
S t a g e 2n
be the language of
f
i2.
x,
then T2n+l
Let T2ni2
E
1 ;
f
n
x.
be any completion in LniI
%
:
x ?!
Xn
n
of TZnil
such that TZnil
x.
Let T = U Tn, M the term model of T (i.e. the model generated by all definable
n functions, using the constants of C).
Since M
b T 2 PA, M
is a model of arith-
metic. The fact that
x ESSy@l) is clear:
Xn = {
X E 0:
TI
E
% 1
= { X E W:
; E % 1
Mk
n The converse follows also by construction: Let
b
E
IMI is explicitly definable from some c. MI= v u ( $ W E
c-+
$*Gi ...c. v
for some $*.
E
SSy(M).
n @U0VI
,...,-ci
"0
)
"k-1 Thus, we can represent X f SSy(M) as
be any formula. Since any
k-I
, we
have
1.
Lectures on Nonstandard Models of Arithmetic
for all m > 2i
Tm
E
x.
T~
[ z E w:
=
+ 2, i
Thus X
=
I-
L
max{iO,
x.
E
...ci k - l 0 -
xci
15
1
...,ik-l 1 .
Thus, X is recursive in Tm, where
@ED
( Remark: The proof that i => ii is trickier. One does not have new codes to make elements of
x standard.
Thus, one alternates p a r t i a l z y completing the
theory-- first the C -sentences, then I: -sentences, etc.-- and representing 1 2 by using independent formulae of increasing complexity. ) elements of
x
Theorem 2.8 can be generalised. For example, one can replace PA by any consistent r.e. extension To (using Craig's observation that an r.e. theory has a recursive axiomatisation to handle stage 0).
A more important generalisation due indepen-
dently to Don Jensen and Andrzej Ehrenfeucht (Jensen and Ekrenfeuckt 1 9 7 6 ) and David Guaspari (Guaspari 1 9 7 9 ) is the following: 2.10.
Theorem.
Let P be a consistent completion of PA in a language extending
that of arithmetic by a countable set of individual constants. Let able c-closed family of sets. i. Rep(!?) 5 ii.
x be a count-
The following are equivalent:
X
There is a model M
b T with SSy(M) =
x.
The proof of Theorem 2.10 is a bit mare delicate. The extra constants offer no real problem-- they can be handled finitely many at a time. one only has R e p ( T )
c
x and not T x. E
The problem is that
However, for each finite n,m the Il
consequences of T referring only to the first rn constants is a set in
x.
-
Thus,
in the construction underlying the proof that i=> ii, one simply handles progressively larger chunks of T as well as progressively larger sets of constants.
I omit the details. Theorem 2.10 is, of course, of interest in its own right, but it is also a lemma. The theories T to which one wishes to apply the result are the complete diagrams of models.
As apparently first noticed by Guaspari, this allows an iteration of
Theorem 2.8 which, by a direct limit argument, yields: 2.11.
Theorem.
Let
x be a family of at most ,ql
sets of natural numbers.
The
following are equivalent: i. ii. 2.12.
x is c-closed X=
S S ~ ( M )for some M
Corollary (Assuming CH).
I=
PA.
x 5 P(w)
is c-closed iff
x is the standard sys-
tem of a model of PA. 2.13.
Open Problem.
Can CH be eliminated as an assumption in Corollary 2.12?
The derivation of Theorem 2.11 from 2.10 is entirely routine and I omit it. is time to move on to another topic.
It
We will not leave standard systems for very
c. S M O R Y ~ K I
16
long:
they are of extreme usefulness.
EXTENSIONS OF MODELS (THE 1960s 11)
3.
Till now I have largely considered single nonstandard models. Another development of the 1960s was the study of the embeddability of one nonstandard model into another, i.e. of generalising the r8le of N o as the standard model to allow nonstandard models to be standard relative to other models.
There is a big
difference in the relation between a nonstandard model M and an extension N and the relation between N (NI
-
a
IN1 -
E
0
and its nonstandard extensions: The new integers of
[MI could fail to be larger than all the integers of /MI. Thus we say
]MI is M-infinite if a > b for all b
E
IMI, and a is M-finite otherwise.
There are three possible types of extensions of models (relative to the crude criterion of the existence of M-finite and M-infinite integers): 3.1.
Definition. Let M _ C N be models of PA. i.N is an end extension of M, or M is an initial segment of N , E J N J - J M Jis M-infinite. N is a c o f i n u l extension of M, written M c N, if every a
written M c N, if every a ii.
E
IN1 -
IMI is M-finite. iii.
N is a mixed extensionofM if IN1 - \MI contains both M-finite and
M-infinite integers. All three possibilities are realised: 3.2.
Theorem.
Let M be a nonstandard model of PA. M has elementary mixed
extensions. 3.3.
Theorem. Let M
3.4.
Theorem.
b PA.
M has proper elementary end extensions.
Let M be a nonstandard model of PA. M has proper elementary
cofinal extensions. Theorem 3 . 2 is a triviality: Two new constants and compactness or any proper ultrapower provide the extension.
Theorem 3 . 3 is nontrivial.
It was first proven
by R. MacDowell and Ernst Specker in MacLbwelZ and Specker 1961.
I postpone a
proof of this Theorem until 57, when I will give a stronger result. section I will also give some related results for the countable case. is from Rubin 2962 and is relatively easy:
In the next Theorem 3 . 4
One doesn't have to construct the
cofinal extension directly; one merely extracts it from a mixed extension. The modern descendent of Rabin's extraction is the following.
3.5.
Theorem (Splitting Theorem).
Let
MzN
be models of PA. There is a unique
model k f f k PA such that M
Cc
?8* Ce N.
Moreover, the cofinal portion of the extension is elementary.
17
Lectures on Nonstandard Models of Arithmetic The Splitting Theorem is due to Haim Gaifman (Gaifian 1 9 7 1 ) and was the end product of a development begun by Rabin and partially furthered by A. Adler and Grigorii Chudnovskil.
It is a fundamental result in that it shews the crude
order theoretic trichotomy given by Definition 3.1 to be model theoretically significant:
Cofinal and end extensions are the, so-to-speak, building blocks of
all extensions and it suffices to study these two extreme cases.
Moreover, it
points to a fundamental difference in the r6les of the two kinds of extensions:
3.6.
Corollary.
Cofinal extensions are elementary.
The reader can easily deduce 3.6 from 3.5. If M
elementary case of 3.5:
5
Another easy exercise is given by the
N, there is a unique model
4f such that M Sc1'8~
The full Theorem 3.5 is nontrivial-- it is equivalent to the formalisa-
4 e N. bility within PA of the solution to Hilbert's 10th Problem.
Because of this, the
result is language-dependent; if one adds a few new predicates to the language, L the result can fail. If we let PA denote the extension of PA to a language L L L containing that of arithmetic-- assuming full induction in L-- and let A o , z I , etc. denote the quantifier classes of the extended language, the Splitting Theorem becomes 3.7.
L L Let M E N be models of PA and assume the extension is A -elemen0 Then there is a unique fif such that M
Theorem.
tary.
Of 3.5 and 3.7 I will only prove 3.5. 3.8.
Lemma.
Let M 5 N be models of PA.
,...,anb1
shewn, be given and let a. i.
If
M ii.
is a Zl-formula,
)I
If @ is a
r[
1
-formula,
$zo...&-l=>
If
M
Let $ n ~ ~ . . . v ~with - ~ , free variables as
E IMI.
- b $Fo...zn-l==> N b $ a o . ..an-l
N iii.
To do this I need a lemma.
c$
M
is provably A,
k $zo..
$zo...-an- 1
in P A ,
iff N
b $Zo..
Parts ii and iii are immediate consequences of part i, which follows from the provable equivalence in PA of all C -formulae with existential ones and the per1 sistence of existential formulae under extensions.
Proof
Of
3.5:
obvious:
Let M 5 N.
We want M
=
{ a
E
3b
=
a
E
INI: INI:
Moreover, it is easy to see that
Q
A -function. 1
Ifif I
fif INI(a < b )
7
N.
The choice of
fif
is
is M-finite }.
!ffis a
i.e. it contains 0 and is closed under segment of N,
E
structure for the arithmetical language,
I,+,*.
In fact, since
fif
is an initial
is closed under any PA-provably polynomially bounded total
c.S M O R Y ~ S K I
18
One A -function that can be proven total and polynomially bounded in PA is GEdel's 7 @-function: There is a A -definable term @ for which the following are satisfied: 1
for each n there is a polynomial &(VO,...,un-l)
i.
1 V U ~ . . . U ~ -3~V n 1 B(uo,vl) 5 V o
PA
ii. PA
c vvO...un-l
iii. PA
&(u,,
$vo...u
3 V ~ . . . U ~ - $~ u o...u
PA
iv.
<
-
18fl is
,..., R ( v , z ) ) ,..., ~ ( u , n - I ) )
@ ( V o ,B ( U 2 ,u o ) ) .
First, expand the language to make
8 is a provably A 1-function,
( N , @ ) and, by the bound ii,
vi )
=
(B(v,O)
< ul
aU2vVo
Now let us prove the Theorem in earnest. Since
3 u@
such that
B(V,,i)
u $ (B(v,o)
t-f
n-1 n- 1
v. PA 1 v v o < U 1 3 U 2 $ u o U 2 + I omit the routine, but boring, proofs.
primitive.
...,u ~ - ~ nc\ ) ( i
8
( M , B ) is still a substructure of
closed under
8. We shew (M,B) to be an
elementary substructure of (Mef,@) by induction on the length of the prefix of a prenex formula,
@ = QOVO* * * q - l u n - l J, 9 where we suppress mention of the free variables and J, is quantifier-free. By iii and iv, we can assume the quantifiers alternate in kind. We call $ 3 if and V
Qo
is 3
otherwise.
Basis: Suppose @
E
Let a
b
(M,8)
E
1~8~ and 1a
< b
1Ml.
E
b v v o < Wo => (N,B)I= v 8) b 6=' (4f,B) I= G,
V v o W o => W,B)
=> ( N ,
6) b
For n = 1 , the subcase $
The case n = 0 is trivial. Q l , say $ = ' f V o $ v o .
(!@,
(M,8) b $ =>
Formally: We prove by induction on n that
since J, is quantifier-free. Since, moreover, a
E
Igfl was
$.
is trivial.
E
Observe -
v0 < b Wo,
by 3.8
arbitrary, we have:
(M"f,B) b v ~ o ~ o . Induction s t e p :
If n > I, there are again two cases:
Case 1.
$
E
3n. This is trivial.
Case 2.
$
E
Vn.
Using v above,
(M,B)
Write $
=
Vvo3
I= V U o 3 U l J , = >
ulJ, and let a
(M,f3)
I=
3VlVV 0
E
<
Igfl and a
< b
E
IMI.
~J,(Uo,B(Vl,~O)).
But the right-hand side of this is basically 3n, whence Case 1 applies to yield (Mef ,B) b ~ U l V U 0 < ~ J , ( U , , B ( U l , V O ) ) => (4 f.B) 3VlJ,(Z,B(Vl,Z))
I=
and, since a
E
=> (4f,B) I= a u1 $Civ arbitrary, (4 ,@) -/- v U o 3 U l
lffl was
8'
$VoVl.
( N.B.
The essen-
t i a l ~- n e w of the formula in question was based on contraction of quantifiers, which 2n
!ffdepends on the bound
i-- or a different n-tupling function.
) &ED
The interested reader is referred to MZ&k 1982B for an axiomatic analysis of the proof of the Splitting Theorem (in the form 3.7) and to Motohashi A&B for additional observations on this result. I also refer to my survey Smoryn'ski 1 9 8 1 A
Lectures on Nonstandard Models of Arithmetic
19
for more on cofinal extensions; there is much of interest in them, but I do not have the space to devote to them here. I have given more space to the Splitting Theorem and given a more detailed proof of it than many expositors would because it is rather more important than it appears.
It does, as I said above, establish the significance of the crude order-
theoretic trichotomy of extensions into cofinal, mixed, and end extensions, and it justifies focussing one's attention on the two extremes. Moreover, as first observed by Wilkie, it has its applications.
(Cf. Theorem 4 . 8 , below, or, for a
simpler application in the style of Wilkie, Smoryriski 1 9 8 1 A . )
I shall close this
section with an amusing application. The Tarskian attempt of the 1950s to turn model theory into algebra resulted in a number of preservation theorems-- theorems characterising those sentences preserved under various kinds of restrictions, extensions, and homomorphic images. With the emergence of the notion of an end extension came also the problem of preservation under end extensions. This was settled in Feferman and XreiseZ 1 9 6 6 by appeal to a many-sorted interpolation theorem.
Straight model theory also
suffices to prove the characterisation, and, in the arithmetic case, the result is
an easy consequence of the Splitting Theorem: 3.9.
Theorem. Let L be a language extending that of arithmetic. Let
I@~...V
n- 1
be an arithmetical formula. The following are equivalent: i. ii.
Proof:
for all models M c
Mb
&o...Fn-l
PAL
c Qi
1
L
and a.
Nel= Go..., n-1
=> for some ?-formula
-
N of PA
,...,an-l
IMI,
E
JI,
JI.
The proof that ii => i is routine.
i => ii. Suppose Qi is upward preserved under all end extensions. By Theorem L 3 . 7 , it follows that @ is preserved under all A -elementary extensions. Letting 0 L L' extend L by a stock of primitives sufficient to make all A -formulae equivalent 0
to quantifier-free ones, we see that @ is preserved under all extensions M 5 N of models of P A L ' . By a standard result of model theory, @ is existential in PAL' , L whence 1; in PA &ED
.
4.
SATURATION PROPERTIES (THE 1970s I)
The model theoretic notions of types and of saturated and special models crystallised in the 1960s.
Although a crude mixture of recursion theory and model theory
can be found already in the late 1930s with Bernays' arithmetisation of the completeness theorem and one can find further discussion of the complexity of models in the 1950s and 1960s, the refined mixture of the two branches of logic-- namely, recursive saturation-- only arrived in the 1970s.
As one might predict on the
c. S M O R Y ~ S K I
20
basis of the fact that recursive saturation is a marriage of model theory and arithmetic definability, this notion is of great importance in the study of nonstandard models of arithmetic.
In fact, recursive saturation-- of a sort-- was
being applied in nonstandard model theory before it was realised that such a concept existed.
In the present section, I will discuss these early applications
and one or two minor issues related t o recursive saturation; in the next section,
I will digress to discuss recursive saturation in a more general, but arithmetically relevant, context. First, of course, come some definitions: 4.1.
Definitions. Let M be a model for some (not necessarily arithmetic) lan-
,...,
bmwl E 1Ml given parameters, and ~ U ~ . . . U ~ - ~ U ~ . .nim-1 .U a set of guage, bo formulae with only the free variables shewn. -
i. set T 5
M
7,
-
. ba~type - ~ over M if, for every finite The set ? ~ ~ . . . ~ ~ - ~ b ~ . .is
k
(M
$v o...v
S U ~ . . . U ~ - ~
-
b
-
n-1 o..*bm-l
1
tJET The type 7 ~ ~ . . . v ~ - j b ~ . . . b ~is - ~recursive if the set TU o...v
ii.
(or, rather, the set of its numerical codes rtJ1) iii.
-
-
The type ~ v ~ . . . ~ ~ - ~ b ~ . . .isb ~a -r-type, ~ for
(e.g. all formulae, Z -formulae, TI -formulae), if every
n
iv.
M b M k
B V o...v
-
The type ?v0.. .v n-l 5o...bm-l
n- 1
$
r
E 7
a set of formulae belongs to
r
is realised in M if
( ~ ~ o . . . v n ~ 1 5 0 . . . b ), i.e. for some
~o...Zn-lbo...Fm-l, for all
nim-1
is recursive
m-1
a.
,...,an-l IM\ E
E 7
V. M is recursively r-saturated if every r-type over M is realised in M; M is recursivezy saturated if M is recursively r-saturated for the set r of all formulae of the language of M.
N.B.
In the arithmetic language, we can restrict our attention to 1-types ‘ruz
with only one parameter because of the arithmetic ability to code finite sequences. We will be interested, however, in models of theories in languages not admitting such coding. One of the keys to the arithmetic importance of these notions is the following notion: 4.2.
Definition. Let
r
be a set of formulae. We say
r
a h i t s a truth definition
(more accurately: a satisfaction relation) in PA if there is a formula Tr (v such that, for all PA
where ( v 0 ,
$ V ~ . . . V ~ -E~
r
r
b@U~...V~-~
...,Unm1)
v ) 03
I
with free variables as shewn,
tf Tr (r$’,(vo,...,V r n-1 ) ) , is the code of the sequence V o ,
...,V n-1‘
The bearing of this notion on recursive saturation is the following lemma implicit in Robinson 1963 and very nearly explicit in Friedman 1 9 7 3 : 4.3.
Lemma (Robinson-Friedman Lemma).
Let M be a nonstandard model of PA and
21
Lectures on Nonstandard Models of Arithmetic suppose of
T,
M
r
admits a truth definition. Then: M is recursively r-saturated.
~ a recursive type over M. By the recursiveness Proof: Let b E IWI and T U be there is a formula Tau(u ) such that, for any formula @ u U o , @VV E T iff
0
0
Let, further, Trr be the truth definition for I-.
Tau('$').
From the fact that
M I=
T
is a type, we have, for each
3 u 1 V r@'
<
1 ( Tau('$uUi)
BY Overspill, for some infinite c
M
I= 3
Lc E 0 ,
+
Trr(r@',(U,E))
).
-+
Trr(r$l,(U,E))
).
]MI,
E
c ( Tau('@Uu;)
v1 V r @ ' <
lM1 witness this last, we see for each realises W E in M. Letting a
E
@UUo E T
that M
I=
a,i.e.
a &ED
The application Robinson made of this lemma is the arithmetic undefinability of arithmetic truth: 4.4.
Corollary.
The class of arithmetic formulae does not admit a truth defini-
tion in PA.
Proof sketch:
Then every non-
Suppose Tr(Uo,VI) gave a truth definition.
standard model of PA would be recursively saturated.
But we can easily construct
nonstandard models which are not recursively saturated:
(6 la
integer b in a nonstandard model N and let
Take, e.g., any infinite
Ryll-Nardzewski) M be the sub-
model of N generated from b by the parameter-free Skolem functions. does not realise the recursive type
x u D $7) OE
+
TUF
Clearly M
of formulae of the form
3 u o < V@V$.
A small remark:
&ED
Corollary 4.4 is not quite as strong as GEdel's result (usually
attributed to Tarski, who rediscovered and first published it) on the undefinability of the truth of sentences. The application Friedman made of the Robinson-Friedman Lemma is a bit deeper: 4.5.
Let M , N be countable nonstandard models of PA.
Theorem. i.
ii.
The following are equivalent: a.
M is embeddable in N
b.
SSy(M) 5 SSy(N) and
T h , (M) 5 f i x ( N )
The following are equivalent: a.
M is isomorphic to an initial segment of N
b.
SSy(M) = S S y ( N )
and
Thz (M) 5 T h z l ( B ) . 1
A few words of explanation are in order:
For
r
=
TI or E l
(the existential or,
respectively, Cl-formulae), T h r ( K ) is the set of sentences in
r
Matijaseviz's result, we could restrict our attention to !Z'k3
in both parts.
explicit reference to Th
true in K.
By The
in ii is, however, the form holding in expanded lan=1
guages as well as the one reflecting more directly the closure properties needed
c. S M O R Y ~ K I
22 in the proof.
In each equivalence, the implication a => b is fairly direct:
(1) The inclusion
of the standard system of M in that of N follows from i. the rigidness of w in the A, nature of the coding of standard sets, and
all nonstandard models, ii. iii. L e m a 3.8:
For a
(MI and z
E
u, MI=
E
3: E
W
iff N ) =
M
c
E
N are both nonstandard, one can choose a code for X in M.
That !7‘h3(M)
sions of the theories are similarly easy.
IY.
If X
equality of standard systems in ii.b follows from Lemma 2.5:
E
(2)
The S S y ( N ) and
The inclu-
5 T h 3 ( N ) is a familiar
fact of model theory; if the embedding is initial, Theorem 3 . 9 can be invoked. Indeed, ignoring the r6le of the standard systems, Theorem 4.5 can be viewed as a generalisation of the model-theoretic preservation theorems.
(I shall, in fact,
derive the hard part of Theorem 3.9 as a corollary to Theorem 4.5 below.) The proofs of the converse implications are facilitated by the following lemma: 4.6.
Lemma. Let K be nonstandard, a.
definition Try. The r-types
Let, for any a
Proof:
-
I= 6 O...an-l,
the parameters a,,,
x= Clearly X n u
SSy(K).
TUZo..
To see that
9l
IKI, and let T admit a truth
T
-a
.an- 1
-
a
E
is the r-type of a with respect to
K
-_
I= Trr(‘$’,(a,a0
,..., -an-I ) )
be a r-type over K, with X =
I
r$’:
}-
-
@ao..
E T
1
is realised, either repeat the proof of Lemma 4 . 3 or
T
-
T‘VU
D j ; + TrT(rI$l,(u,~o
E
be the set of all I$vVO...un-l
uVO...Vn-l
Let b be any infinite integer in IKl and consider
replace T by the recursive type r
E
over K realised in K are precisely those
T U~~...U,-~
-I < b: n- 1 SSy(K1 codes T ~ .
Conversely, let E
.. -
IK1,
E
i.e.
{ FI$~uo...u E
,...,an-,
O...an-,
.
coded in S S y ( K )
such that K
TU;
o...a
- -
b consisting of all formulae n- 1
,..., -an-l)),
where X = Db nu.
&ED
Proof of fieoran 4.5:
We have but to prove the implications b => a.
To do
this, we let M , N be given with convenient enumerations of their elements. We
,... of
construct such enumerations a O,al
model of N that the function mapping a A bit of notation:
r
we let T~
o.
.
i
\MI and bo,bl
For K a model of PA, co
denote the r-type of c o ,
.c
,... of an appropriate sub-
to bi is the desired embedding.
,...,em-,
E
IKI, and T either 3 or Zl,
...,cm- 1 over K.
m-1 i.b. => a. Let M,N be countable with SSy(M) Moreover, let a o , a l , enumeration b ,b Stczge
-
n.
...
5 SSy(N) and Th3 (M) 5 T h , (N). be the given enumeration of IMI. The construction of the
,,... proceeds by stages.
Suppose bo,...,b,-,
-
.an-l ‘to. ..bn-l
Tao.. LL
have already been chosen so that (*)
23
Lectures on Nonstandard Models of Arithmetic ( M ) 5 T h R (N) ) . The claim is that ~b~...b,_~ is a type over N and is in SSy(N).
(if n = 0,this reads: 3
anao...an-1
T =
By the Lemma, T is coded in S S y ( M ) , whence the inclusion of standard systems It thus suffices to shew T to be a type over N.
places T in S S y ( N ) . For any @ ,,
easy:
-
* *
9
T>
@k-l-E
.
M I= mbiZnC0. .an-l => i
I=
M
existence of b
Let b
E
3v
[m+ivC,..
I=
3v
1
.“,-l
z
=> N by the assumption (*).
But this is
-
3,
[ fi@;vb0.. .bn-l i
IN1 be any element of N realising
T.
( The
is, of course, guaranteed by the Lemma. )
I leave to the reader the verification that the map ai to bi constitutes the desired embedding. ii.b. => a.
This half of the proof is a bit more complicated in that we must
map back to guarantee the function constructed maps M onto an initial segment of N . such that, for each n,
A s in part i, we construct ao,...,an-l,b,,...,bn-l
c.
z1 h,. ..an-l
(*)
c ‘b0.. 1 .bn-l’
The w-stage construction alternates in direction.
Even-numbered stages.
Given a,,
...Jan-Z,boJ.. .,bn-l
satisfying (*), let a
be the first element of the given enumeration of I M ( not already listed and choose
bn exactly as in part i such that
Odd-numbered stages,
I1 ‘a,...a
c
5
(**)
- ‘b0...bn‘
satisfying (*), let bn be
Given a,,...Jan-I,bO,...Jbn-l
the first element of the given enumeration of IN1 such that br, < bk for some k
n-1.
(If no such bk exists, move on to the next stage.)
5
Choose any convenient
k for which bn < bk and fix it. We want to find an E ILY such that (**) holds. By the Lemma and the To this end, we look at the n -type T of b ,bo,...,bn-l. 2 assumed equality of standard systems, T E S S y ( M ) .
We thus need only shew that T
is a type over M . For any $o,.
N
.., ok I= En
<
E
T,
.
-
Kk A m$iKnEo.. bn-? i
=> N => M
b
3v < 3v <
Zk
[fi@iVbg.
Zk
(fi$iva,.
..%n-2 ) ..an-l ),
?.
since the sentence in question is TIl and since (*) yields
-
“1
Thus,
T
n- 1 Tbo...bn-l‘ is a type over M and, by the Lemma, it is realised in M by some an’
Again, I leave to the reader the routine task of verifying that the map ai to bi
c. S M O R Y ~ S K I
24
&ED
has the desired properties. Corollary. Let M be a countable nonstandard model of PA.
4.7.
Then: M is iso-
morphic to a proper initial seEment of itself. One proof proceeds by repeating the second of the above constructions-- being careful not to map M above a certain fixed element of itself. be had:
A slicker proof can
By the MacDowell-Specker Theorem (Theorem 3 . 3 , to be proven in § 7 ) , M has
a countable proper elementary end extension N which, by Theorem 4.5, can be initially embedded in M.
Composing the maps yields an isomorphism of M onto a proper
initial segment of itself. Theorem 4.5 is hardly the last word in back-and-forth constructions between nonstandard models of arithmetic. A fairly exhaustive list of basic embeddability results can be found (each with “constructive” and ”slick” proofs) in my survey
Smory6ski 1981E. 4.8.
Here let me cite merely the following theorem of WiZkie 1977B: Let M , N be countable nonstandard models of PA.
Theorem.
The following are
equivalent: i. M is arbitrarily highly initially embeddable in N , i.e. M is isomorphic to arbitrarily large initial segments of N ii. SSy(M) = SSy(N)
and T h
%
(M) -
= *% (N)
-
For a proof cf. WiZkie 1977B or ShroryGski 1981E.
There are, of course, many other such results deserving mention.
I like Wilkie’s
result because it has an interesting application to the construction of end extensions.
For that matter, Friedman’s Theorem has its applications to the study of
end extensions.
I cite one application of each.
First, there is the arithmetic case of the Feferman-Kreisel result: Theorem. Let Q, be a sentence of the language of arithmetic (plus a few
4.9.
constants if one wants a result about formulae).
The following are equivalent:
Q, is preserved under end extension: If M M I= @ => N I= Q, ii. for some $ E zlJ PA Q, ++ 9. i.
Proof:
CeN
are models of PA,
This is Theorem 3 . 9 , which we have already proven.
Theorem 4.5.ii, we assumed 4.9.ii=> 4.9.1.
In proving half of
Thus, I really intend only to use
4.5.11 to give a new proof of the more difficult implication i = > ii.
I prove the equivalent dual result: If I$ is preserved under restriction to initial segment, then Q, is equivalent to a I$-sentence. Let
I$
be given and T = PA
countable model of T .
+
{
TI E
111:
PA
1 Q, + TI
1.
Suppose M is an arbitrary
If we can shew MI= Q,, it will follow thatPAC I$++
TI~~...,,T%
25
Lectures on Nonstandard Models of Arithmetic for some IIl-sentences vl,...,v
E
T , whence
9 is
equivalent to a 17 -sentence. 1
Clearly it suffices to embedd M as an initial segment of some model N Now PA
T h z (M) is consistent; else PA f 9 7(UlA A5mm-1) for some 0 I’ 1 Th (M), contradicting the fact that the II -consequence I( U l ...fi ~
E
El
is true in M
I=
1
T. Now, the truth definition for X 1-formulae shews 7 !h ‘
+ 9 + Thz
Let T’ be any completion of PA
coded in SSy(M).
(which exists by the completion-closure of SSy(M)). model of T’ with S S y ( N )
=
embeddable in N, whence M
( Query:
I$.
...
+ 9+
...,um-1
I= PA +
SSy(M) (Theorem 2.10).
I=
5
(M) to be
(M) coded in SSy(M) 1 Finally, let N be a countable
By Theorem 4.5.ii M is initially
I$.
QED
Is the present proof model theoretically more pleasing than that of the
preceding section? ) As an application of Wilkie’s Theorem 4 . 8 , I cite the following application of
Wilkie 1 9 7 7 B : 4.10.
Theorem. Let M be a nonstandard model of PA and let T be any consistent
completion of PA. The following are equivalent: i. ii.
~N(M
c~NI=T)
R e p ( T ) C_ SSy(M)
and
Thz (M)
5 T.
1
The remarkable thing here is that M is not assumed to be countable--
Proof:
the countable case is nearly trivial.
Theorem 4 . 8 , in conjunction with the
Splitting Theorem and the Arithmetised Completeness Theorem (56, below), yields the uncountable case. i = > ii.
Left to the reader.
ii => i.
Countable ease.
Let M be given and countable. If Rep@) 5 SSy(M)
and T is complete, there is (by Theorem 2.10) a countable model N =
SSy(M).
k
T with SSy(N)
If T 33% (M), Theorem 4.5.ii applies and M is isomorphic to an init-
ial segment of N:
M
C
N
I= T.
General ease (countable and uncountable): Let M be given, T I Th plete theory with Rep(T) 5 SSy(M).
By the Arithmetised Completeness Theorem,
there is an end extension K of M modelling PA
+ TrI12,
i.e. the 17 -theorems of T. 2
By a Lgwenheim-Skolem argument, one can find a countable K
b
E
lKol is M-infinite, and ii. S S y ( K o ) 2 R e p ( T ) .
countable model Mo
(M) a com-
T with SSy(Mo) = S S y ( K o ) .
such that i. some
Apply Theorem 2.10 to obtain a
By Theorem 4.8, we can initially
in K above b , i.e. the image of Mo contains b . Composing the initial 0 0 embedding of Ma in K and the embedding of KO into K, we can assume M K. 0 0 -
embedd M
Apply the Splitting Theorem to find N
=
A$f
satisfying Mo
CcN
Ce
K.
Since M,N
c. S M O R Y ~ ~ S K I
26
are initial segments of K , either M the latter inclusion fails and M
C
C
N
N or N
C
M. As b
E
lMol 5 I N I , b
T.
E! IMI, &ED
Now for an abrupt change of topics: Note that we have thus far only explored the consequences of the saturation inherent in every nonstandard model of arithmetic-recursive r-saturation for
r
admitting a truth definition-- and not full recursive
saturation. Among countable recursively saturated models, the ability to realise full types allows the construction of elementary maps:
4.11.
Theorem. i.
Let M , N be countable recursively saturated models of PA.
The following are equivalent:
M is elementarily embeddable in N b. M is (properly) cofinally embeddable in N a.
C.
ii.
Th(M) = Z'h(N)
and SSy(M)
5 SSy(ilT)
The following are equivalent: b.
M is isomorphic to N M is isomorphic to a (proper) elementary initial segment of N
C.
Th(M)
a.
=
Th(N)
and SSy(M) = S S y ( N ) .
I refer the reader to my survey Smoryn'ski 1981B for detailed proofs and credits. The basic results-- namely, the criteria for elementary embeddability and isomorphism-- are exercises for the reader. a minor point:
I would even say "easy exercises", but for
In proving Theorem 4 . 5 , we relied on Lemma 4 . 6
made of the existence of a truth definition for
r
where use was
in shewing the r-types over M
realised in M were precisely those coded in S S y ( M ) .
This appeal to the truth def-
inition can be replaced by appeal to an appropriate recursive type-- which will be done in greater generality in the next section. Friedman's 1973 paper had a tremendous effect on the study of nonstandard models of arithmetic.
It was largely responsible for the renaissance of nonstandard
model theory in the 1970s.
Among those papers largely inspired by his paper were
WiZkie A , Jensen a d Ehrenfeucht 1 9 7 6 , Kirby and Paris 1 9 7 7 , WiZkie 197?B, Guasp a r i 1 9 7 9 , L i p s h i t z 1 9 7 9 , H&ek and Pudlak 1 9 8 0 , and Srnoryn'ski 1981B.
Moreover,
several of these have had sequels themselves. Perhaps the most interesting of these is the paper of Don Jensen and Andrzej Ehrenfeucht, which will be the point of departure of 55, below.
The marriage of recursive saturation with nonstandard model theory begat a number of additional developments which I cannot discuss here.
I refer the reader to
H6jek and B d Z a k 1 9 8 0 , KotZarski A , and Smoryriski 1982B for more advanced backand-forth arguments, and to K o t Z a r s k i , Krajewski and LachZan 1 9 8 1 , Lachlan 1 9 8 1 ,
Kaufman 1 9 7 7 , Kossak A , Shelah 1978B, Schmerl 1978 and Schmerl A for further interesting properties of recursively saturated models of arithmetic.
27
Lectures on Nonstandard Models of Arithmetic 5.
RECURSIVELY SATURATED MODELS OF RICH THEORIES (A DIGRESSION)
Donald Jensen wrote up some joint work with Andrzej Ehrenfeucht in 1973. When he died in 1974, the task of editing the manuscript fell to Harry Simmons. The final paper, Jensen and Ehrenfeucht 1 9 7 6 , appeared in 1976.
The notion of recursive
saturation, due to J.P. Ressayre and J. Schlipf, was only first mentioned in print in 1975.
It is historically false to say that the Jensen-Ehrenfeucht paper had
any results on recursive saturation.
However, the paper falls short of proving
many basic results on recursively saturated models only in the non-appearance of the words "recursively saturated".
The present section is a sort of elaboration
of ideas implicit in the Jensen-Ehrenfeucht paper.
I worked the details out in
December of 1979 during a visit to Heidelberg, but never published them because Julia Knight and Mark Nadel, in Knight and Nude2 1 9 8 2 A , went a bit further in a key direction.
However, I've never been particularly pleased with their exposi-
tion and its emphasis on the useless necessary and sufficient conditions; to my mind, it would have been more appropriate to emphasise the recognisable sufficient condition of richness.
Nonetheless, I refer the reader to their paper for more
general results on expandability. Our point of departure is the following pair of results of Jensen and Ehrenfeucht: 5.1.
Theorem. i. ii.
5.2.
Theorem.
Let M = (lMl;+,*,',O) be a nonstandard model of PA. Then: (lMl;+,<.a) is recursively saturated ([MI - (01;.,1) is recursively saturated. Let M , N be countable nonstandard models of PA.
(lNI;+,<,O) are isomorphic, then
so are
(IMl
-
If (IMI ;+,<,a)
and
{Ol;*,l) and ( I N 1 - (Ol;-,l); and
conversely, if the models of the latter pair are isomorphic, so are those of the
.
former
Closely related (and of some interest in the model theoretic assault on complexity theory) is the following result due (I believe) t o Wilkie. 5.3.
Theorem. i.
Let M be a nonstandard model of PA and a , b
and @,@ are the restrictions of ii.
E
(
+,'
to
IMI. c
E
IMI: c < a
if M is countable and Th(
is isomorphic to (
2975.
)
If one looks carefully at Jensen and Ehrenfeucht's proof of Theorem 5.2, one sees that the model M is needed only to invoke Theorem 5.1 and thereafter it is properties of recursively saturated models of Presburger-Skolem Arithmetic, PSA
=
Th(w;+,.,O),
and of Skolem's Multiplicative Arithmetic,
c. S M O R Y ~ ~ S K I
28
that are used.
Moreover, these properties are guaranteed by the richness of the
class of recursive types of the given theories. Without further ado, let me introduce the class of rich theories and the study of Following this, Theorems 5.2 and 5.3.ii will
their recursively saturated models.
( As for Theorems 5.1 and 5.3.i, let me note that
be recognised as trivialities.
the Robinson-Friedman Lemma applies:
The full truth definitions for the reducts
of 5.1 exist by the provable quantifier-eliminability within PA of PSA and SMA (Cf. Presburger 1 9 3 0 and Skolem 1931 for PSA, CegieZski 1 9 8 0 and Nude2 1 9 8 0 for SMA, and CegieZski, McAzoon and W i h e r s 1982 for both) and the Ao-nature in M of
all questions about
)
A theory T in a recursive (preferably finite) language L is
Definition.
}new, of formulae in one free variable
r i c h if there is a recursive sequence,
such that, for any disjoint finite sets X,Y of natural numbers, we have
+@jvl.
C R V cM\@ivn.
iEX 5.5.
JEY
The obvious examples are PA and ZF. Other examples, in increas-
Examples.
are:
ing order of the complexity of @ i.
of w independent sets.
The theory P
sequence of monadic predicates P,,,P,, "
A
This theory has a countable
... (no equality) and axioms
)
3v (/?(\PjVAh 1 P . V ZEX .icy J
for any disjoint finite sets X,Y of natural numbers.
Evidently, the choice
P,"'
will do. ii. Atoms.
The theory BA
-
an
_ _
where a c , a l , iii.
5v,
... are constants naming the designated atoms. PSA
=
Th(w;+).
3vo(v where p
=
Simply choose
pnvo).
denotes the n-th prime and mu the m-fold sum v iv.
SMA
$lnv:
=
T h ( w - {O};.).
3vo(v
with p , as above and V.
reads
of Boolean Algebras with Countably Many Distinguished
One chooses,
=
+
... + v .
Here one chooses
%,
v0
vm the m-fold product.
Skolem's Divisibility Arithmetic, SDA = T h ( w - {Ol; We want to define 3 v o ( v o is primeAvon+l I v h y v on+2 I v ) .
"v0 divides V I " .
To do this, we define a few auxiliary predicates: vo = I : Prim(vo):
"Vl(V0~Vl)
T V 0= I ~
V
(V V
IV
+ v l = v0V V 1
= 1)
I),
where
V 0 IV 1
Lectures on Nonstandard Models of Arithmetic
29
Primpow(v ) : 0 n-Chain(U ) : 0
n-Chain(u) uo
=
asserts that
1 and all divisors distinct, i.e. that v has at least n prime factors.
Primn(vo) : Primpow(Uo)A Primn(v) asserts u =
$nu:
where
.
1 (n+l)-Chain(vo).
for some prime V We now define ) p , Primn'l(ul)h Primn+2 ( v , ) 0
A
~
~
3vou1v2 (Prim(v A u
nated
V;
n-Chain(vo),,
~
'
~
~
u
1- ~
~
~
~
A
u
~
u
A
1
~
~
+ vi. The theory ODAG of Ordered Divisible Abelian Groups with a Desig+ + Positive Element, i.e. ODAG = Th(Q;+,<,O,l)= Th(R;+,<,O,l)= Th(R ;+,<,l,e), + Q,R,R denote the sets of rational numbers, real numbers, and positive real
numbers, respectively.
If we let
element, mt the rn-fold sum
t+
denote the constant designating the positive
... + t for any term t, and m the particular
instance mi, we can choose
$nu: \x/{ K 1.P+'v < 2m+1: o 5 m < 2n 1 . ( Explanation: Consider the iterated bisection of the "unit interval" based on 0 $nu asserts that u lies in the left half of the result of the n-th bisec-
and 1. tion.
) vii.
Any extension (by language or axioms) of a rich theory is rich.
Of
+.
particular interest is the theory RCF of Real Closed Fields, which contains ODAG
( For later reference, note that in RCF
-
uo < v l :
a u 2 ( l V 2 = O A U0
< is definable in terms of
+ v 22
o,+,.,=
by:
= V l ) .
Thus, the choice of $ U of vi can, for RCF, be translated into an existential
) formula in the language with primitives o,T,+,*,=. By point vii and the weakness of the theories cited, there are evidently a great many rich theories.
5.6.
Nonexamples.
There are also quite a few non-rich theories: Any theory T with a countable saturated model fails to be rich.
These include: i.
h'o-categorical a. b.
theories, e.g.
the theory of Equality on an Infinite Domain, Em = Th(w;=) the theory of the Order of the Rationals or Reals, OR = Th(Q;<) = Th(R;<)
ii.
sa-categorical theories for a > 0, e.g. a.
the theory of Algebraically Closed Fields of Characteristic 0, ACFO =
b.
UDAG iii. iv.
Th(C;+,*,O,l),where C is the complex plane
the theory of Uniquely Divisible (Torsion-Free) Abelian Groups, =
!i%(Q;+,O)
=
Th(R;+,O) = Th(R+;*,l)
w-stable theories occasional other theories, e.g. Th(w;<,O).
~
~
c. SMORYI~SKI
30
The reason a theory possessing a countable saturated model is not rich is simply that this model has only countably many types, while every model of a rich theory If T is rich, with
possesses at least continuum many types:
InEW witnessing
@nV
this richness, then, for each set XC w, the set
I
T (v) = X
$,v:
n
x 1
E
x I
n L
U{
is a type over every model of T. And clearly, if X # Y, then T ( v ) and -rY(V) are X
incompatible, whence distinct. The types
T (0)
X
If a
are the key to the mystery:
E
lMl realises
T (1))
X
in a model
M of T, then a c o d e s the set X in M. To emphasise this point, we will use a different notation: n E v : @nU. ( N.B. n is a subscript of the formula n
E
v,
@zv in an arithmetic lann E w: M k n E a 1 is coded
not a participant (unless @ u is uniformly defined as guage). by a .
) Thus, if a
E
lMl realises T XV , then X
k
The family of sets coded in a model M
=
T is analogous to the standard system of a
nonstandard model of arithmetic: 5.7.
Let T be a rich theory, { n
Definitions.
E V
InEw, a recursive sequence of
unary formulae witnessing the richness of T, and M a model of T. An element a
E
\MI codes a set X 5 w iff the condition cited above holds. A set X 5 w is s t a n dard on M (or, a r e a l of M) if X has a code in M. The standard system, SSy(M), of M is the collection of all standard sets of M. There is a serious ambiguity in this definition: not merely on M, but on the choice of for example, to construct a model M when we choose n choose n
E
E
E,
k Pw
in which SSy(M) =
v to be PZnU ; and SSy(M)
v to be P
The standard system of M depends,
i.e. of the formulae n = { X
5 w:
E
v.
XC w:
It is easy,
X is finite
X is cofinite
1,
I,
when we
2n+lV'
For recursively saturated models, however, there is no ambiguity:
5.8.
Let T be a rich theory and M a recursively saturated model of T.
Lemma.
The standard system of M is independent of the choice of
Proof: iff M
Let
Let X
given.
k
n
E~
Since
E,
and E ~ ,i.e. the sequences { n E v InEw and { n E~ v InEo, be 1 be standard with respect to El, i.e. let a E IMI satisfy n E X
E
5w
a,
-
TUU:
E.
for all n YZ Ez
U
E
++
w.
Consider the recursive set of formulae,
a:
YZ E
n
E
w
I.
witnesses the richness of T, this is a type over M.
we see that b codes X relative to
E
2'
The symmetric argument yields the converse: are standard with respect to
Letting b realise it,
All sets standard with respect to
E
1'
E
2
&ED
Since we are only interested in recursively saturated models of T; the lemma justifies our not mentioning
E
in "SSy(M)", i.e. we need not refer to "SS~(M,E)".
31
Lectures on Nonstandard Models of Arithmetic The significance of the standard systems of recursively saturated models of rich theories is the same as that of standard systems of (recursively saturated) nonstandard models of arithmetic. 5.9.
Let T be a rich theory and M a recursively saturated model of T.
Lemma.
The types realised in M are precisely those coded in S S y ( M ) .
Proof:
-
-
mbo...bm-l Suppose
Let M be a recursively saturated model of a rich theory T and let be a type over M . Let b
is coded in S S y ( M ) :
T
E
[ M I be such that, for $wv0...
(the set of formulae with free variables v , v o ,
M
!=
'$'
z. -Simulate - -
E
T
..
by the set
..., m-1 V
),
V m-1 @vbo...bm-l E T
&m+l iff
-
o...
@vz bm-l: @ E km+l I. This is easily seen to be a recursive type over M, whence realised in M ; but any
~ ' v b ~b,-*b: .
a
E
/MI
realising
T'
Conversely, if a.bo
{ r@l
E
%
-+
automatically realises
,...,bmml
E
T.
\ M I , the type of a over M in parameters bo
,...,bm-l
(i.e. the (m+l)-type of ~ , b ~ , . . . , b ~ -is ~ ) seen to be coded by appeal to the recursive type -
-
@abo...Emm-lt+
-r'vabo...bm-l:
r$l
E 0:
@ E
R",+l 1.
&ED
An immediate corollary is: Theorem. Let M ,M be countable recursively saturated models of a rich 1 2 theory T with equality. The following are equivalent:
5.10.
M1 is isomorphic to M z
i.
and SSy(MI) = S S y ( M z ) .
Th(Mz) = Th(M2)
ii.
I leave the proof to the reader. reduces to: S S y ( M l )
=
SSy(Mz).
I also note that, if T is complete, condition ii
( Exercise: Prove Theorems 5.2 and 5.3.ii. )
The key to the parenthetical exercise is that, if M is a countable nonstandard model of PA and Mo is its
+-
or --reduct, then SSy(M) = S S y ( M o ) .
In fact, the
standard systems of countable recursively saturated models of rich theories are precisely those of nonstandard models of arithmetic. 5.11.
Theorem.
Let M be a recursively saturated model of a rich theory.
Then:
SSy(M) is c-closed.
Proof: n
E
i. ly.
Let M be a recursively saturated model of a rich theory and let
v InEw witness this richness. S S y ( M ) is an a2gebra.
Let X,Y
Consider TTX(V):
Tx n Y ( V ) :
{ n YZ
E V
++
1E E
-
a:
E V t-t YZ E ; A n
E
SSy(M) with codes a,b
?' Ei W E
z:
Evidently, these are recursive types over M.
E
]MI, respective-
1 n
E W
1.
Any element of M realising
T
,
~
c. SMORYI~SKI
32
codes the complement of X; any element realising ii. X
E
T~~~
codes X n Y .
SSyiM) is ctosed under r e k t i v e recursiveness. To shew Y
SSy(M).
Let Y be recursive in
SSy((M), we first recall a convenient characterisation of
E
relative recursiveness: Y is recursive in X iff there is a partial recursive function f such that, for all n
Let b
n
E
Y
iff
n
L
Y
iff
w,
E
3 km ( D~ 5 X , , D ~ n x 3 k m ( Dk C X , , D m r l X
=
0 hf(k,m,n)
=
=
BAf(k,m,n)
=l).
0)
-
1M1 code X and f reduce Y to X and consider, i E FA/xz 7 i E 7 n E v : f(k,m,n) = T V ~ : iEDk i"Dm ~ { r ) c \i E /x\ I i E 7 + --, n E v : f ( k , m , n )
E
{m
One easily sees
TVF
to be an r.e. type over M.
recursive ones,
TVZ
is realised by some a
Let X
E
=
11.
iED,
iEDk
iii.
0 } U
E
Since r.e. types are equivalent to
lMI. Evidently, a codes Y.
SSy(M) code an infinite binary tree and let b
E
(MI code X.
To
conclude that an infinite path through X is coded in SSy(M), we will again realise an appropriate recursive type. tion:
To describe this type, we must recall some nota-
We assume an identification of w with the set of all finite sequences of
0 ' s and 1 ' s under which the usual operations and relations on finite sequences are
In particular, < > is the (number identified with) the empty sequence,
recursive.
(for i
=
0 , l ) is the sequence of length 1 with sole entry
catenation of the sequences m , n ,
TVEbe the set of all formulae of the forms
With this notation, let >
v,
E
i, m*n is the con-
and m _c n asserts that the sequence n extends m.
n
E
v
+
n*
E
vvn*<~> E v,
n
E
v
-+
n
E
6
for all n , and also all formulae of the form E
v),
for all m,n such that m
$n
-(m
us that
E
vAn
T V ~ ,which
and n
m.
Since X is infinite, Kb'nig's Lemma tells
asserts v to code an infinite path through X, is a type over M.
&ED
Recursive saturation does the rest. Shortly we shall have an interesting application of Theorem 5.11.
First, however,
we want to prove the converse and one of its applications. 5.12.
Theorem.
Let T be a consistent, complete rich theory and
c-closed family of sets. i. ii.
a countable
The following are equivalent:
TEX There is a recursively saturated model M
I=
5" with SSy(M) =
x.
The proof is sufficiently similar to that of Theorems 2 . 8 and 2.10 that we can safely skip it. Note that Theorem 5.12 refers only to complete theories.
In the incomplete case,
Lectures on Nonstandard Models of Arithmetic
33
one has: 5.13.
Corollary. Let T be a consistent rich theory and If some axiomatisation of T is coded in
family of sets.
sively saturated model with standard system 5.14.
Corollary.
Let T be a consistent r.e. rich theory and
x a countable c-
Then T has a recursively saturated model with standard
closed family of sets. system
x.
x a countable c-closed x, then T has a recur-
X.
( What about the uncountable analogue to Theorem
5.12?
An easy application of the
Arithmetised Completeness Theorem shews that uncountability is no greater obstacle in the rich case than in the arithmetic case:
If a complete rich theory T and a
x with T x are given, and if x = SSy(M) for some model M of arithSSy(M ) for some recursively saturated model M b T . In particumetic, then x 0 0 lar, Theorem 5.12 extends to the case where x is of cardinality ,q,. However, it x for simple 2'. The characterisation may be easier to construct M with SSy(M) c-closed set
E
=
=
trivially holds without restriction for P and Knight and Nude2 19824 shew it holds without restriction for PSA.
)
The immediate corollary to the existence and uniqueness theorems (5.10 and 5.12) and the independence of SSy(M) of the choice of
E
is expandability:
5.15. Theorem. Let T c T be complete rich theories with respective finite 0- 1 languages Lo 5 Ll with equality and let M be a countable recursively saturated
model of T o .
The following are equivalent:
TI E SSy(M) ii. M can be expanded to a recursively saturated model of T 1 , i.e. there i.
is a recursively saturated modelMl of T pretation on /MI of the primitives of L
Proof:
i => ii.
1
obtained from M by an appropriate inter-
l-Lo'
Let M be a countable recursively saturated model of T o .
Construct a recursively saturated model MI of Tl with SSy(Ml) = S S y ( M ) .
Letting
M
be the L -reduct of M and appealing to Lemma 5.8 inside M we have SSy(M ) = 0 0 1 1 0 SSy(Ml) = S S y ( M ) . By Theorem 5.10, M is isomorphic to Mo. Pulling the structure of MI back to M yields the expansion desired. ii => i.
&ED
By Lemma 5.9.
Refinements: Obviously, To need not be assumed complete as T h ( M ) is both rich and complete. If T is not assumed complete, but some axiomatisation of T is coded 1 1 in S S y ( M ) , it has a completion so coded and the existence o f an expansion to this completion yields that for T I .
In particular, a countable recursively saturated
model M of a rich theory T is expandable to a recursively saturated model of
!2%(M) U T l , for any recursive theory T I consistent with T h ( M ) . This latter version of expandability-- to recursive theories-- has nothing to do
c. S M O R Y ~ K I
34
with richness; it is a basic expandability result called respzendence and due independently to J.P. Ressayre and to Jon Barwise and John Schlipf.
The present
sharp form for rich theories was also proven in Knight and Nude2 1 9 8 2 4 , where a generalisation to arbitrary theories in finite languages in terms of "ideals" is given. Resplendence-- the expandability of countable recursively saturated models t o recursive theories-- has many applications to models of arithmetic, particularly to recursively saturated models of arithmetic. Many of these applications, however, are not as powerful as those obtainable by more direct methods; I refer the reader to the papers cited at the end of the previous section for such applications. What I really want to do now is digress even further from the subject of nonstandard models of arithmetic and prove an analogue to Tennenbaum's Theorem. Recall that Tennenbaum's Theorem (actually, Corollary 1.6)
told us that no non-
standard model of PA could have recursive operations of addition and multiplication.
By Theorem 5.1,
the
+-
and --reducts of a nonstandard model of ?A are
recursively saturated models of PSA and SMA, respectively; by Theorem 5.15, converse is countably true:
the
A countable recursively saturated model of ?SA
is the +-reduct (--reduct) of a nonstandard model of PA.
(SMA)
Hence, there are no
recursive, recursively saturated models of ?SA or SMA. What we shall see is that this is a better way of looking at Tennenbaum's Theorem-- it asserts the nonexistence of recursive, recursively saturated models of decent rich theories. Before proving the analogue of Tennenbaum's Theorem, we need two definitions. 5.16.
Definition.
Let I MI be M as identity. We say M
Let M be a model for a given recursive language L .
w and let equality, if it occurs in L , be interpreted in
is recursive7y presented if there is a recursive satisfaction definition for atomic formulae of L in M:
MI=
...,xn- 1
The relation, for xo,
Go...Zn-1 '
is a recursive relation of
r$Uo...V
' and
n- 1
E
w and formulae I$,
(X~,...,Z~-~) (the usual recursively
obtained n-tuple). If L is finite, this amounts to each primitive relation and function being recursively interpreted in
W.
The notion of recursive presentation is the appropriate version of model theoretic complexity. 5.17.
The notion of decency of richness is the following:
Definition. A rich theory T is e x h t e n t i a Z Z y r i c h if the recursive sequence
of formulae witnessing the richness of 7' can be chosen to be purely existential. Examples will follow shortly. First, let me state the main result:
Lectures on Nonstandard Models of Arithmetic 5.18.
Let M be a recursively saturated
Theorem (Tennenbaum's Theorem Revisited).
Then: M is not recursively presented.
model of an existentially rich theory. Proof:
Observe first that S S y ( M ) , being completion-closed, contains a nonLet X be such a set and
recursive set-- e.g. some consistent completion of PA. let k M
E
35
( M I = u code X.
I= $nzo., .Kr where
Then: n
avo..
E
X
iff M
n
E
3 m0
iff
is the formula n
.V
n
k
E
,...,m p -I
s.t.
v with$ quan'tzifier-
n
free. Evidently, X is r.e. in any presentation of M . Since SSy(M) is closed under complementation, T X is also r.e. in any presentation, whence X is recursive in such.
Since X is non-recursive, the representation is
&ED
not recursive. 5.19.
Examples.
The following theories are existentially rich and, hence, have
no recursively presented, recursively saturated models: i.
PSA; cf. 5.5.iii
ii. SMA; cf. 5.5.iv iii.
ODAG+; cf. 5.5.vi
iv.
RCF; cf. 5.5.vii.
RCF, at least two of +,.,<
I n fact, in any recursively saturated model of
must be nonrecursive; but any one of these can be
recursive. For further information on variants of Tennenbaum's Theorem for recursively saturated models, cf. Macintyre 1 9 8 1 .
Therein Angus Macintyre goes much further in
ruling out the recursive presentability of recursively saturated fields of various kinds.
He gives an analysis analogous to mine of Tennenbaum's Theorem and cites
Lipshitz' citation of Tennenbaum as another source. If I may add: Mark Nadel informs me he has also proven a generalisation of Tennenbaum's Theorem. Let me finish with a minor problem: 2.20.
Open Problem. Is SDA (cf. 5.5.v) existentially rich?
6. THE ARITHMETISED COMPLETENESS THEOREM (THE 1970s 11)
I have already made several references to the Arithmetised Completeness Theorem; indeed, I have twice referred to its applications. In the present section, I propose to consider it in detail. 6.1.
Theorem.
In its weakest form, it is simply the following:
SSy(N ) is c-closed. 0
Proof sketch:
Recall that S S y ( N o )
=
Def(No), the family of subsets of u
definable in N o , i.e. the class of arithmetical sets. The closure of the family of arithmetically definable sets under boolean operations is trivial; that under relative recursiveness follows from Post's Theorem. The crucial point is the existence of an arithmetically definable infinite path
c. S M O R Y ~ K I
36
through any arithmetically definable infinite binary tree.
By an argument analo-
gous to that in the proof of Theorem 5.11, it suffices to find an arithmetical To each infinite binary tree X we can
completion of any arithmetical theory:
associate recursively in X a theory T any completion of which recursively codes an infinite path Y through X. To complete a theory T with an arithmetically definable axiomatisation, simply
... of all sentences of the language and look
take a recursive enumeration @,,Q,,
for the left-most consistent path through the usual tree (cf. the illustration in T the proof of Lemma 2 . 9 , above). The consistency question is n2, whence the path T T is recursive in II?, i.e. it is A2. &ED
( Remark: The reduction is not absolutely necessary. Given an infinite binary tree X, one can search for the left-most infinite path through X:
Start at < >.
Given u, if there are infinitely many elements of X beyond U* list wise list u*.
other-
O*;
The path Y so generated is recursive in the question, "Are there
X
infinitely many elements greater than u?", which question is II2-- whence Y can be X chosen A3. The reduction reduces this degree. ) The Arithmetised Completeness Theorem generalises Theorem 6.1 in two ways.
First,
by adding witnessing constants, it produces a model, not merely a completion; second, it is formalised within P A , whence valid over any model thereof.
Before
going on to the full result, I should first like to give some sort of application of the result at hand.
It is not, perhaps, the most convincing of applications;
but it allows me to illustrate the relation between infinite integers and ultrafilters:
I shall translate the ultrafilter proof of Ramsey's Theorem into a
nonstandard one and use Theorem 6.1 to obtain sQme extra information. First, some background on Ramsey's Theorem:
6.2.
Definition.
Let X be a set of natural numbers and let 2.y be positive
integers. i.
(,") is the set of subsets of X of cardinality
X,
i.e. the set of
s-element subsets of X ii. A function ~:(jcx) colouring, of
(,XI
P; o r :
6.3.
+
y (
I
O , I , ...,y-
iii. A subset Y 5 X is P-homogeneous ( o r : Y monochromatic ) if P is constant on (s). Theorem (Ramsey's Theorem).
subset of w , and ) : ( P
Y 5 X.
=
+
I 1
) is a
homogeneous with respect t o
Let s,y be positive integers, X an infinite
y a partition.
There is an infinite P-homogeneous set
Moreover, Y can be chosen arithmetically definable
Proof:
By induction on x > 0.
it will remain relatively fixed.
p a r t i t i o n , o r y-
€1
X,P.
We shall denote x by n in the following since
Lectures on Nonstandard Models of Arithmetic
Basis.
For n
31
1 , this is merely the assertion that a finite union of finite
=
In this case, Y is clearly recursive in X,P.
sets is finite.
Induction step.
Without l o s s of generality, we may assume X
=
w.
Let
(w;+,*,',O,P). By Theorem 6.1, SSy(N 0 ,P) is c-closed. 0 Choose (M,P), by Theorem 2.10, to be any nonstandard model of T h ( N ,P) with 0 SSy(M,P) = SSy(NO,P). The canonical embedding of ( N ,P) into (M,P) is clearly P:(nyl)
y and ( N ,P)
-+
elementary.
=
0 Moreover, the restriction to w of any definable subset of or a rela-
tion on (M,P) is definable in (No,P). Let a
E
Using a, deviously define the rela-
\(M,P)I be any nonstandard integer.
tion R on wn+'
by
Rxo...xn-lx:
P(xo,...,xn~l,x)
=
P(xo,...,xn-l'a).
R is codable in (M,P), whence definable in (NO,P). Let @Vo...Vn-lV
Proof:
Suppose P ( X ~ , . . . , X ~ - ~ ,=~ )i < y .
(M,P) I=
Go....: n- I x
-
-
whence (No,P) b Go...Fnn-lx
(No,P)I= v v ( Since p(x0
&o...Fn-Iv
,...,
=
++
For any x E w, P(X0 Xn-,,X) = P(xo Xn-l,a) tf P(F0,. 'xn-l ,x) = T , P(xo X ~ - ~ ~ = X i.) Since x was arbitrary, P; o...xE-lv i whence (M,P) satisfies the same.
,...,
f-t
-
define R.
.. ,...,
i, (M,P)I=
Proof of 6.3, continued:
1
,...,
,,
QED
I$LZ~...X~-~U.
We enumerate a prehomogeneous set Z as follows:
... < znm1 be arbitrary natural numbers. n, such that for any subsequence Suppose we have generated z o < ... < zk-Il k < z i < ... < zi , we have (No,P) b @zi .,.zi By the Claim, for each z zi . i0 0 n -0 - n-1 n ... < z i , we have (M,P) c. ...zi a. From this it follows by Theorem 2.1.i n- 1 '0 n-1 z for all that there are infinitely many z ' s in w such that (N ,P) $yi ...z 0 o in-?
Let z o <
-
such subsequences.
Choose z k to be the least such.
After w steps, we have enumerated Z arithmetically in P,X.
=
zo,zl,
... 1
recursively in R-- hence
Now define a new partition Q: (
Z
)
+
y by
Q is well-defined and arithmetical in P,X. P(X~,...,X~-~,~). hypothesis, there is an infinite Q-homogeneous Y 5 Z. Let yo <
... < y,
P(yo, ...,y
and wo <
n)
= =
... < w
be elements of Y.
Q(Xo,
...,Wn-1 ) ,
Q(wo,
=
P(wo
Observe:
=
P(wO,...,wn-l,~
by Q-homogeneity
,...,wn-l,a), by
=
We claim Y is P-homogeneous:
P(yO,...,yn-Z,a), since the yi's are in Z Q ( y o , ...,y,- I ) , by definition
=
...,xn-l)
By induction
definition
),since the w.'s.are in Z .
Since Y is arithmetical in Q, it is arithmetical in P,X.
c. S M O R Y ~ K I
38 6.4.
Remarks.
As I mentioned earlier, this is a modification of a proof
i.
using ultrafilters (cf. Chang and Keisler's model theory text or Bell and Machover's logic text).
A more faithful representation of such a proof proceeds by
replacing the model constructed by appeal to Theorem 6.1 by any proper elementary This proof, which I leave to the reader as an exercise,
extension of ( N , P ( W ) , E ) . 0
does not give the extra information on the relative arithmetical definability of the homogeneous set. ii.
As first shewn in Jockusch 1 9 7 2 , this relative arithmetical definability
can also be established by analysing a direct proof of Ramsey's Theorem; in fact, this latter gives us explicit arithmetical bounds and the schematic provability of Ramsey's Theorem within PA.
This doesn't seem possible with the above proof or
via the more general Arithmetised Completeness Theorem.
We will see in 87, below,
however, that the above proof can be extended in the face of a yet stronger arithmetisation. To illustrate yet once more the usefulness of infinite integers, let me offer the following reduction of the Finite Ramsey Theorem to the Infinite Ramsey Theorem, as promised in
6.5;
Corollary
following the proof of 2.1.
82,
(Finite Ramsey Theorem).
Let x,y,W be positive integers.
There
is a positive integer z o > x such that, for all z > z u and all partitions P:(")
X
-f
y, there is a P-homogeneous set X 5 z ( = 10,
Proof:
Fix x,y,
...,2-1)
) of cardinality w .
and W and assume the resulting assertion to be false.
This
assertion has the form ~ v O v v >l v o $ ; its negation is thus V u u 3 vI > u o 7 $. Apply Theorem 2.l.i: Let M be a nonstandard elementary extension of Nu and let u
E
IWI be nonstandard and P:Ca) X
homogeneous set exists. partition on
(-), which
+
y any coded partition for which no w-element
A contradiction follows by noting that P restricts to a
has an infiniEe P-homogeneous set Y, any w-element
&ED
subset of which is also P-homogeneous.
I first learned the present reduction from Joram Hirschfeld; I find it a bit easier than the direct appeal to Kb'nig's Lemma. the Paris-Harrington Variant:
Both proofs also quickly yield
The Theorem remains true if the cardinality con-
dition on X (that X have cardinality W) is replaced by one that X be relatively large (i.e. card(X)
2 min(X) , x + l ) .
( Exercise: Carry out this proof. ) The
Finite Ramsey Theorem also has a reasonably direct proof with fairly explicit bounds; the Paris-Harrington Variant does not (cf. e.g. Paris and Hurrington 1577 or Ketonen and S o ~ o u a y1581 ) . Getting back on track, recall Theorem 1.3 and the ensuing discussion by which any consistent r.e. theory has, not only an arithmetically definable completion, but also an arithmetically definable (in fact, A,)
model.
construction can routinely be formalised within PA.
The fact is that the whole The problem is that, although
39
Lectures on Nonstandard Models of Arithmetic the construction is routine, it is a notational morass.
I turn now to the unin-
spiring discussion of syntactic preliminaries.
6.6. Notational Conventions. i. ProvL (v ,v ) is a A -representation of provaPC 0 1 1 bility within the predicate calculus PC based on the language L. That is, we assume, for
Ji,x in
PCL$ iff 3 2 E w P A L ProvpC(uOJ L rJi’)
a. b. ii.
the language L,
( PAlProvgC(;,‘jl’) A ProvpC(ul, L ‘Ji *
If $uvo is any formula withu free and
other free variables) such that PAC$vuo
L
formula Prov($)(v
‘U 0 ’ 1’ $(r$7,uo) }, by
$:
+
ProvpC(uO~V1*(rX”), L
‘-x-).
vo either free or vacuous (and no
”U is a sentence of L”, we define a
v 2 ) , supposedly representing proofs overPC from axioms
L(uO;ul Prov($) iii.
+
)
xT)
,u2) :
3 finite
Q5
I U : $UU
0
, r m ~ U;
I(ProvpC(ul L
+
)).
We also define :
Pr($)(uo;uz)
ulProv($)(VL 0
;v ,u ) , 1 2
suppressing mention of the language. There are two contexts in which we want to use this notation, the syntactic and Semantically, we choose a model M k P A and a parameter a
semantic.
define a theory T with axioms X =
1
$:
k $(rJi7,z) 1. The
M
IMI and M
extends T. Syntactically, one wants to choose a fixed numeral ; 0
and look at the theory T given by the axiomatisation X In this case, the set of theorems of T is
Ji:
contained in
E
set { J i :
T ~PI-($](F~;~J~’) 1.
I ) :
Ji: -
=
M
PA bPr($)(~~;~$~)
I= $J(~J~’,F~C,, 1. 1
and is
The free variable v o in $uu0 allows us
to handle everything unifbrmly.
6.7.
More Notation.
i.
Con($)vo
is the assertion -Pr($)(uo;y),
where A is
a convenient refutable sentence of the language L . ii.
If T is a theory with a more-or-less canonical representation @ of its
($1.
axiomatisation, we write Pry and ConT for Pr($)
and Con
assume such a canonical representation exists.
If T1 = T
If T is r.e., we
+ $,
and $ defines the
axiomatisation of T, we assume TI canonically given, relative to T, by $IUu0 :
$vvo
V
v =‘I)?.
Finally, we sometimes write PrT and ConT, when no canonical
representation is given, to denote the predicates used when some specific representation is in mind. iii.
If xuu
0
has u free, possibly
v0 free, and no other variable free, and
x satisfies PA C p u 0 “U is a sentence of L“, we x defines truth for some structure in the language I,.
if, for a given language L , let Mod(dVo Mod(&
0
assert that
+.
is the conjunction of the following: a.
v
Sentences ~ i ~ , X(~VJ,YJ,~ ~ i ~ ( ,uo)
t-t
xf$;
, u , ) ~ x ( ‘ $ ; ,vo) )
c. S M O R Y ~ K I
40 for
0
E
I
A , V,
1
+
V sentences J, ( x('7 $',V0) C f ix('-$',Vo) ) V ' I @ with ' only v free ( X ( ~ Q U J N ' , V , ) ++ qVx('I$"v0) ) where Q E c V ,2 1 and u" is a name for the object represented by 0. ( If one doesn't like u", rephrase this in terms of satisfaction of J, by V . ] b.
C.
With all this notation we can state the Arithmetised Completeness Theorem: 6.8.
Theorem (Arithmetised Completeness Theorem; Syntactic Form).
Let
@V0
be
an arithmetic formula with u free, v o free or vacuous, and no other free variables. Suppose, for some language L , P A F @VUo "V is a sentence of L". a formula p u g with the same free variables as +VVo such that
Then: There is
-+
i. PA ii. PA iii.
1pvo I-
+
"u is a sentence of L"
Con($)vo
+
Con
-+
PA
C+lV'0
.@v0 pv0 Mod(JI)vo. -+
The proof just consists of arithmetising the proof of the Completeness Theorem and is routine; nonetheless, or therefore, I shall not repeat it here.
Eventually,
the details will be published.
The Syntactic Form of the Arithmetised Completeness Theorem is rather powerful. I n proof theory (in the Orey Compactness Theorem-- cf. Fefemun 1960-- and in Smory6skC 1982C) this power is occasionally necessary; in model theory it-particularly the uniformity-- is generally not necessary. To state the semantic form, we need a definition:
6.9. Definition. Let M be a model of PA and N a structure for some primitive recursively encodable language L . We say that N is s t r o n g l y definable in M iff i.
IN(
=
IM[
ii. the satisfaction relation for N is parametrically definable in M, i.e. there is a formula Tr(uo;ul,t)Z) and some b E /MI such that for all $u0... V n- 1 and a . a E IN(, N !=Go a iff M l=Tr(F;r$',(zo n-1 n- 1
...
,...,
6.10.
,..., an-l)).
in $) b
M
I=
Let M ( , P A ,
Theorem (Arithmetised Completeness Theorem; Semantic Form).
+vva a formula always defining a set of sentences, and (if u E
Con($)),
1M(. Let
JI:
M
occurs nonvacuously
0
@(r$',F)1 axiomatise a theory
2'.
then there is a model N 1=T strongly definable in M.
I N / should be infinite, we can interpret equality in
If M
ConT (i.e.
Moreover, if
N by identity.
Theorem 6.10 is a fairly immediate corollary to Theorem 6.8 and I omit the details. We shall shortly see some good conditions on T guaranteeing M to recognise its consistency. First, let us note an immediate consequence of strong definability. 6.11.
Theorem.
definable in M.
Lst M be a nonstandard model of PA and suppose N is strongly Then:
N is recursively saturated.
Lectureson Nonstandard Models of Arithmetic
41
The proof has, more-or-less, already been given: The result is a variant of the Robinson-Friedman Lemma ( 4 . 3 ; but recall also 5.1 and 5.3). If the theory T contains PA, the connexion between M and a strongly definable N
k2'
6.12.
is much stronger: Lemma.
Let M,N be models of PA with N strongly definable in M.
There is a
canonically M-definable embedding of M onto an initial segment of N .
Proof: Define F:IMl -+ IN1 ( = IMI ) by recursion: F ( O ) = O N , F(S#) = M ( F a ) , where OM (ON), SM (S,) are the zero element and successor function of M N (respectively, N ) . The definability of SN and ON in M, together with recursion in M guarantee that F is well-defined and M-definable.
S
That F is a homomorphism, i.e. that it preserves atomic structure is proven as follows: F preserves the successor function by definition. Preservation of +,* follow by induction in M and the recursion equations in M and N .
+ 5) =
addition, weprove V V ( F ( V
Since
+
F(a
+ 0)
F(a
+ b')
=
Fa
=
FF) by induction on b Fa + 0 = Fa + FO N N N FV +N
+ b ) ' ) = S$(a + b ) = +NSgb = Fa +N F ( b ' ) .
=
F((a
=
Fa
vo
is preserved, so is <:
< vl
t+
E
E.G.
for
IMl :
SN(Fa +N Fb)
3v2(vo
+ v2'
=
vI). Hence F is one-
one, i.e. an embedding. Finally, to see that F maps onto an initial segment, we shew M
I=
VV <
N
by induction on b : note N
I=
V v ( v
z3v0(FV0
=
Fv)
For b = 0, the result is vacuous. For the induction step,
sNE
++
v cN Z v v
=
Z ).
&ED
Lemma 6.12 was multiply discovered: S l a t e r 1972, H&jek 1972, Sinoryn'ski 1973A&B, MeAloon 19784, and Guaspari 1979. Lema 6.12 has its applications; but these depend on our model M's recognising the consistency of a given theory 2'.
Gsdel's Second Incompleteness Theorem tells
us this cannot always be done-- for the obvious definitions of many T ' s .
A
theorem of Mostowski allows us to get around the gb'delian difficulty:
6.13. Theorem (Reflexiveness). Let L be a language extending that of arithmetic and let To be a finite subtheory of PAL axiomatised by ,$nn-l. Then,
PA
b Con , i.e.
PA
1 Con(+),
where +v is
TO
Wv
=
'$2.
. ..
The proof of Reflexiveness, that PAb proves the consistency of each of its finite subtheories, lies beyond the scope of the present lectures: It is a matter of proof theory, not model theory. However, I can remark that it is not particularly difficult-- one proves Cut-Elimination or Herbrand's Theorem for P?, proofs in P 8 are rendered direct.
A direct proof of
-I($~A
...
A$~-,)
by which all uses only
c. S M O R Y ~ S K I
42
L
formulae of bounded complexity and one can use the partial truth definition in PA to shew by induction that all formulae involved in the direct proof are true. Since
7 ( w 0 ~ . . .
A $ ~ - ~is)
n o t true, the given direct proof did not exist--
whence 6.13. 6.14.
Corollary.
Let M be a nonstandard model of PA.
There is a model N
PA,
with M c N and N strongly definable in M. Woof:
It suffices to note that M
I=
for some $ defining PA.
Con($)
The easy
Let $ v be the “canonical” definition of a set of 0 By Theorem 6.13, M Con($lF for each finite V5 V o .
exhibition of I$ is as follows: axioms of PA and ?vv *: z.
By Overspill, M
usual axioms of PA. 6.15.
Remark.
$ o V ~
-
Con($)a
E
(MI. But I$va defines the
Apply 6.10 and 6.12.
QED
A s I mentioned in the above proof, the formula $vvo exhibited is
With a little work, one gets a uniform definition
the easy one.
of PA for which PA kcon(?). Theorem:
for some infinite a
$0
of the axioms
The uniform method yields the Orey Compactness
For an r.e. theory T to be interpretable in P A , it is necessary and
sufficient that every finite subtheory of 7’ be so interpretable (whence its consistency provable in PA).
I refer the reader to F e f e r m a n 1960 and Grey 1961 for
more on this matter. A s one might gather from this Remark, Corollary 6.14 is a 1970s version of 1960s
This version is due to H6jek 1972 and S m o r y f i s k i 1982C (but cf. S m o r y f i s k i
work.
1973A&B) and was rediscovered by McAZoon 1978A and Gzcaspri 1979. McAloon actually offers the model theoretically nicest version, but I defer this till later.
Right now I want to give an application of the results already at hand.
The following is the uncountable generalisation of Theorem 5.12: 6.16.
Application.
x a c-closed x is the standard system of some model of PA, then
Let T be a consistent complete rich theory,
family of sets, and T
x.
E
If
x is the standard system of some recursively saturated model of 2’. In particular, there is a recursively saturated model of T with if x has cardinality at most s,, standard system x. Proof: Let M be a nonstandard model of PA with S S y ( M ) = x. Case 1 .
M
I= ConT
for some designation of T. By Mostowski’s Theorem (6.13)
we also have M (= ConpA for some encoding of PA. are consistent theories
so
Now, the argument that, if T1,TZ
is the theory of disjoint pairs of structures
modelling T I , T 2 , respectively, can be formalised within PA. Thus, M I= Con be a strongly definable model of produced by means of Theorem 6.10.
By Theorem 6.11,
is recursively saturated.
SSy() = SSy(A) = SSy(N).
By Lemma 6.12,
But, by Lemma 5.8,
we can regard N as an end extension
43
Lectureson Nonstandard Models of Arithmetic
.
of M, whence SSy(N) = SSy (M)
Thus, SSy ( A ) = SSy(M), as was to be proven.
M b ConT for any designation of T. I claim that some nonstandard
Case 2.
initial segment of M believes T to be- consistent. To find this, first apply the Lzwenheim-Skolem Theorem to obtain a countable nonstandard elementary submodel Mo
< M with T E SSy(Mo)
coded by, say, a E \ M o \ .
Expand the language of Mo to include a new predicate Tu defined by Tv: V Note that (M ,T) is merely a definitional expansion of Mo. 0
E
D;;.
Let L denote the
extension of the arithmetical language by the predicate Tu.
-
L
Now consider the theory T given by Th (M ) , PA , Con(Tu,V_
c
0
holds in ( N ,T) under an appropriate numerical interpretation of 0 is coded in S S y ( M o ) , whence it has a consistent completion TI
E
TuAV <
c allows us
Moreover,
TO
SSy(Mo).
Let (Ml,T) be a countable model of Tl with SSy(MI,T) = S S y ( M o ) . Friedman's embeddability criterion, observe that the bound
c.
Before applying
c in the formula
to replace this formula by a T-free formula @uZ, with new
parameter b , say, for which MI
b
Con
(@UZ)Z.
Now map MI isomorphically onto an initial segment of Mo and compose embeddings:
MI
ceM o <
M.
Apply the Splitting Theorem to get M
is a nonstandard initial segment of M, whence SSy($) Con(+vq,
i.e.
4fI=ConT for some definition
+UE
cc$f =
ceM.
S S y( M ) .
The model $f But $f
of the axioms of T, whence
&ED
Case 1 applies.
Application 6.16 is as much an application of the Splitting Theorem as it is one of the Arithmetised Completeness Theorem. There is a nice application of the Arithmetised Completeness Theorem which allows u s to avoid an appeal to the Splitting Theorem. This is Hamid Lesan's proof (Lesan 2 9 7 8 ) of Wilkie's Theorem 4.10
characterising the theories of end extensions of nonstandard models.
First, let me offer McAloon's statement of the Arithmetised Completeness Theorem: 6.17.
Theorem. Let M be a nonstandard model of PA and 2'2 PA a theory consistent
with the Infl-diagram of M, i.e. T is consistent with Th ( ( M ; u ) , E: 1 ~ 1 ) . If 'n+1 some axiomatisation of T is coded in S S y ( M ) , there is a strongly definable C elementary end extension N of M modelling T .
Proof sketch:
If T is consistent with the Xnfl-diagram of M, then, by a
hierarchical generalisation of Mostowski's Reflexiveness Theorem and the trick of Corollary 6 . 1 4 , M recognises this consistency. One now applies the Arithmetised Completeness Theorem to construct N modelling M and the I: -diagram of M. nf1 of course, satisfy more Zn+l-sentences, but on all I: -sentences, the two structures agree.
N may,
&ED
c. SMORYI~SKI
44
For full details, cf. McAZoon 1 9 7 8 4 . Now for Lesan's application: Let M be a nonstandard model of PA and T a consistent com-
6.18. Application.
plete extension of PA. ~
i.
I
~
The following are equivalent: (
M
C
~
N
~
T
)
R e p ( T ) E SSy(M).
ii.
The implication i => ii is immediate.
Proof:
To prove ii=> i, define a sequence M = M
C
M
C
... of models M
satisfying:
I= PA + nn+,(T) Mn ' x n Mn+l'
a. M
b. where n,+](T)
of T. This is done by noting TIn+z(T') + that
is the set of n,+]-theorerns
+ Th'n+1 ((M,;a) , . I I ) is consistent: If not, there is x E Cn+l such M, b x ~ ~ . . .while ~ ~ -nn+2(T) ~ , + PA 1 V V o . . . ~ m - l i po...V ,,,-I* But, v uo.. 1 pa.. E nn+l (T),whence Mn b B u g . . .urn-, l p o . . a PA
contra-
diction. Chqose N =
u
&ED
M,.
n The Arithmetised Completeness Theorem has other applications.
For some of these
cf. Smoryn'ski 1 9 7 3 A&B, Srnoryn'ski 1 9 7 7 , M c A t o o n 1 9 7 8 4 , and Srnoryn'ski 1982C.
7.
POWERFUL ARITHMETISATIONS (THE 1970s 111)
The Arithmetised Completeness Theorem is a fairly powerful tool.
As we saw, it
has many applications and it gives a variety of end extensions-- particularly, C -elementary ones.
But it does not seem to yield fully elementary end extensions.
Moreover, I see no way of obtaining, e.g., from it.
the fully formalised Ramsey Theorem
However, Ramsey's Theorem can be proven in PA and elementary end
extensions exist.
Moreover, these facts are intimately related.
The existence of an elementary end extension of any given nonstandard model M of PA was first proven in MacDowelZ and Specker 1 9 6 1 by an ultrapower construction:
Like Skolem before them, R. MacDowell and Ernst Specker took definable functions over M and made appropriate identifications by means of a carefully constructed ultrafilter on the algebra of definable sets. not a great deal of) care was needed:
In the countable case, some (but
Most ultrafilters would yield mixed
extensions and some will yield cofinal ones.
Also, in the countable case one can
simply appeal to a type-omitting argument (which uses Dirichlet's S c h u b f a e h p m h i p
--
i.e. the trivial case of Ramsey's Theorem).
The uncountable case requires a
great deal of uniformity because of the essentially countable nature of the construction.
45
Lectures on Nonstandard Models of Arithmetic In the early 1 9 7 0 s nonstandard modellists took a closer look at this uniformity.
Robert Phillips introduced the notion of a c o n s e r v a t i v e elementary end extension, his student Alan Cantor (in Cantor 1 9 7 2 ) proved No to have conservative proper elementary end extensions, Phillips (in PhiZZips 1 9 7 4 ) proved the existence in the general case, and Gaifman (in Gaifman 1 9 7 6 ) provided a nearly definitive treatment of the subject by introducing the notion of a d e f i n a b t e t y p e . 7.1.
Definition.
Let
Mc
N be models of PA.
We say N is a c o n s e r v a t i v e e x t e n -
sion of M if the restriction to 1M1 of every subset of IN1 parametrically definable in N is so definable in M. In the special case in which M is the standard model N o , an extension is conservative just in case its standard system consists of the arithmetic sets (or, for an extended language, those arithmetic in the new primitives).
If we generalise
from standardness to M-standardness, we can redefine conservativeness: Definitions. Let M 5 N be models of PA. A subset X 5 IMI is called an 7.1'. M-standard s e t of N, or more simply an M-set of N , if X is the restriction to lMl of a parametrically definable subset of N . The collection of M-sets of N is the M-standard s y s t e m , SSyM(N), of N . N is a c o n s e r v a t i v e e x t e n s i o n of M i f f SSYM(M) = SSYM(N).
We will not make full use of this conceptual breakthrough.
I merely wish to use
the notion of an M-set to explain conservativism. 7.2.
Theorem. i. ii.
Let M 5 N
be models of PA.
N is an end extension of M:
The following are equivalent:
M ce N
M and N have the same bounded (i.e. bounded in M) M-sets.
Moreover, each of these conditions implies: iii. lMI
7.3.
M and N have the same parametrically definable bounded subsets of
Corollary.
Let
MsN
be models of PA.
If the extension is conservative,
then it is an end extension. The corollary is immediate.
Proof of Theorem 7 . 2 :
i = > ii.
The proof is almost entirely analogous to
that slewing the standard systems of nonstandard models related by end extension to coincide:
If X
it has a code < 2c
E
SSyM(K), where K = M or N , is bounded by some c E (MI. I leave the details to the reader.
E
IMl , then
( This part is surprisingly easy and the reader might prefer to attempt the proof before reading further. ) Suppose M and N have the same bounded M-sets. Let b E I N ( - (MI and consider X = { a E (MI: N b a < % 1 . ii => i.
X
5 IMl
contains 0 and is closed under successor.
If b were M-finite, X would be
c. SMORYI~SKI
46
a bounded M-set of N , whence one of M--
whence X would be parametrically definable
in M.
But then induction in M would yield X = 1 M 1 , contradicting the boundedness
of X.
Hence, b is M-infinite. iii.
Since b
E
IN1 - IMI was arbitrary, M c
N.
&ED
Obvious.
I'm not sure whom to credit 7.2 and 7.3 to:
The proof of the elernentary case can
be found in Gaifman 2 9 7 6 , but, presumably, the result was also known to Phillips.
( As usual, I am separated from my library and am relying
on my shaky memory.
)
Theorem 7.2 seems not t o have been stated in print prior to 3noryfiski 1 9 8 1 A , even though it requires nothing new in the matter of proof.
Then again, the gain
of 7.2 over the elementary case of 7.3 is primarily heuristic:
It illustrates
nicely the difference in the r8les of cofinal and end extensions: The former add new definable sets; the latter can change truths.
( Remark: Using 7.2, one can see that the last paragraph of the proof of Lemma )
6.12 was unnecessary.
Anyway, for o u r present purposes, the importance of this digression is this: To construct an elementary end extension, it suffices to construct an elementary conservative extension.
I will present here Gaifman's method of doing this: Simply put, if M < K is given, and c E 1x1 - ] M I , there is a minimal extension of M in
K containing c-- namely, the model N generated over IMI
free Skolem functions.
U I c ] by the parameter-
Thus, it suffices to construct such an extension.
More-
over, because PA has built-in Skolem functions, constructing the model N amounts to constructing the type realised by c over M .
Guaranteeing the extension N will
be conservative is done by constructing a (locally) definab2.e type for c . For the sake of applications, it is handy to proceed in some generality:
7.4.
Definition. Let L be a countable language extending that of PA and let M be L Let Lf be the extension of L. by constants naming the various
a model of PA
.
elements of \MI. An L+-type (i.e. one naming all the elements of M )
Tv
over M is
a definubte type (perhaps better: a %ocalZy definabZe t y p e ) over M iff: F o r each L-formula
@ V ~ . . . V ~ -with ~
with only v o
u v O...Vn-Z
-
-
$vao.. .an-I E
only
v , v ~ , . . . , V ~ - ~ free,
,...,V n-1
TV
iff M
Thinking in terms of an element c
there is an L-formula
free such that, for all a.
,...,an-l
E
\MI,
I= u$Zo.. .Zn-3. E
IN1 realising
TU
in N
> M, TV
a
1M1 : N
is definable iff
(locally) each property of c is determined in M: N
k
iff MI=o$
$Cao...zn-2
o-*.a*-1. -
Switching roles of parameters and variables, if X X = { a
E
IMI : M
b
0
@
1
=
E
b
1 , then
and, if N is generated by M over c, the extksion is
readily seen to be conservative. So we see that the problem of producing elementary end extensions reduces to
47
Lectureson Nonstandard Models of Arithmetic producing definable types.
One more point:
We want the extension to be proper,
a
to include the formula V # for each a E /MI. Since T v is to be the type of an element c, TV is complete; since c will be M-infinite, it follows that
i.e.
TV
a
TV also includes v > for each a E IMI. Now, the one thing we know about M-infinite elements of elementary extensions is that they satisfy all properties
held by "almost all" elements of M and any property they satisfy are satisfied cofinally through M-- cf. e.g. Theorem 2.1.
-
Thus, in deciding the truth of
-
-
@ao.. .an-I ' the first question to ask is if M b B v o 3 v I > vo$vl~o.. .an-I. If - the answer is yes, we are free to put $Cao...an-I into the type TU as we are generating it.
There are only two problems:
avoiding clashes and being uniform.
But these obstacles are surmountable.
7.5.
Lemma.
Definable types exist.
Sketch of the construction: within PA
L
.
The idea is to ignore M and L
i
by working
Since M believes lMl to be w , a formal induction will work.
Since there are arithmetical pairing functions, it suffices to construct U $VV
0
with only
V,vo
V
$ 0 for free. To this end, let $ovvo,$IVvo,... be an enumeration of
L-formulae in these variables. The idea o f the proof is made clearer if we switch from formulae to sets and declare ourselves to be constructing a "definable ultrafilter" on the algebra of parametrically definable sets instead of constructing a type. $Vvo is taken to define, for each parameter
xa
=
c
b
E
(MI: M
a
E
Thus, a formula
[MI, a set
l='$E 1.
Since M believes itself to be standard and since we wish to work within P A , we
... of isuch. sets. ... .
think of '$ as defining a sequence Xo,XI,
sponding to the i-th formula ${, we denote by Xo,X?;, The crucial lemma is this:
The sequence corre-
. .
...
If X is a given unbounded definable set and Xi,X';,
is a given definable sequence of sets, then there is a definable unbounded.set
Y
5X
such that, for each n, Y is eventually inside X"
or disjoint from X i
,
i.e.
i is unbounded. Moreover, this is provable: . n i P A L L V nn(ltY n Xi is bounded" V "Y - Xn is bounded").
exactly one of Y
n X"
and Y - X
The provability in PAL follows from the fact that Y is defined via a recursion and the proof of the basic properties is inductive. So much for talking about the lemma! Let us now see how it is proven: Given X ; xi xi we define a sequetfceX.Yo ?YI 2 such that each Y n is unbounded
...
o,
and either Y c X: n-
and
or 'f,
n X; , if
=
@ by recursion:
this is unbounded
c. S M O R Y ~ K I
48
[ 1
if this is unbounded
Ynn X t + l ,
yn+l
=
Y
- x n+l,
otherwise.
L Although it is not immediately apparent, this recursion is admissible in PA
The desired set Y is simply the diagonal of this sequence:
I Yni: i
E
1, we have
w
segment { 0, ...,ynn
Y
ynn:
=
n
E
w }.
Writing Y
Note that, except for the initial
}, Y is either contained in X
i
or disjoint from it.
0 Now for the actual construction: Start with the enumerations X o , 0
,... and the set Z-l
I$,,$,
given by 0
and X o , X l ,
... to obtain Y
PAL
1Vn
("Z
n
= Zo
X
= w
bounded" v "Z
is 1
1
... .
Any definable set is of the form X = X i.e. define
TU
by:
OI$Vo:
(i.e. 1M1 in M).
Apply the lemma to Zm1
i
-X
is bounded").
Etc.
for some i , n ( n can be nonstandard when
Put X into the ultrafilter iff Z. n X" Z
If @Vv0is an L-formula, say
v u 1 3 v2 >
...; X1o , ...;...
satisfying
Then apply the lemma to Zo and X o , X l ,
interpreting the result in M).
.
=
0=
I$.VUo,
n
is unbounded,
then
Ul ( $ i ~ z A $ i V 2 ~ o ) ,
where Qi defines Z
i'
I fear I must leave the miserable verification that everything I claim holds to
QED
the energetic reader. Theorem. Let M ( = P A .
7.6.
Proof:
Then M has a proper elementary end extension.
QED
Immediate.
A few comments might be in order before we discuss applications: MacDowell and Specker's original construction was a sort of Skolem ultrapower-- one takes all parametrically definable functions d Zu Skolem, but carefully chooses one's ultrafilter, as in the proof of Theorem 7.5.
It turns out, after all the extra
work, that it is considerably harder to construct nonconservative elementary end extensions than conservative ones-
Kaufman's "rather classless model" (Kauftnan
2977)
provides an example of an uncountable model with no nonconservative exten-
sion.
For the countable case, cf. BZass 1 9 7 4 , PhiZZips 1974B, and Potthoff 1 9 7 6 .
( Exercise: There is a "natural" class of nonconservative elementary end extensions. Shew:
If M and iV are recursively saturated and N is an elementary
end extension of M, then the extension is not conservative. (Kotlarski)
)
As for applications, Gaifman, in Guifman 1 9 7 6 , shewed how the Ehrenfeucht-Mostowski Theorem on the existence of models generated by indiscernibles followed, without appeal to Ramsey's Theorem, from Lemma 7.5.
This is not too surprising:
The schematic derivability of Ramsey's Theorem in PAL is equivalent to the L In the last section, we saw existence of definable types over all models of PA
.
how the truth of Ramsey's Theorem followed from the mere existence of a conserva-
49
Lectures on Nonstandard Models of Arithmetic tive elementary end extension of No.
A minor variation of the proof (Exercise.)
shews that the schematic provability follows from the existence of such extensions of arbitrary nonstandard models.
For the converse, i.e. the construction of
definable types on the basis of the formalised Ramsey Theorem, cf. P i Z z a y 1981C.
( The schematic derivability of Ramsey's Theorem in PAL was established in Jockusch 1 9 7 2 .
Incidentally, Jockusch gives exact bounds on the arithmetic
complexities of the homogeneous sets in terms of those of the given partitions. I note that the proof via conservation yields explicit bounds, but confess I McAloon, in McAZoon 1978B,
haven't checked how close they come to Jockusch's. derives upper bounds model theoretically.
)
Another obvious application is Theorem 6.1:
SSy(No) is c-closed.
any proper conservative extension of No, SSy(M)
=
For, if M is
SSy(No) and we know from
Theorem 2.7 that SSy(M) is c-closed for nonstandard M. Other applications include generalisations to nonstandard models of anything we proved for No by appeal to appropriate infinite integers. The existence of
)I/-
1
like models, of course, follows as in ordinary model theory from the existence of elementary end extensions of countable models... and certain two-cardinal
I shall close this section with a delightful deriva-
theorems follow from this.
tion of a two-cardinal theorem.
7.7.
Let T be a theory in a language L possessing a unary predicate
Definition.
symbol U
.
If M is a model of T , we denote the interpretation of U in M by
For cardinal numbers of IMI and
7.8.
ff are
K
1. A,
K
A, respectively.
and
Application. Let T be a theory in a countable language L .
(Ho,n)-models for each n
Proof:
E
ff = {
w.
Then:
For any
K
2
A
O,...,n-I 1 .
If Ma is an (/!-'o,n)-model We can also expand M
of T, we can assume IM
model will satisfy Tn
=
T
+ PALi + v U(UV
By compactness, we get an (,~'o,h'o)-model: and add the axioms
c#
for all z
E W.
++
I
= w
to include the arithmetic :elations
and satisfy full induction in the augmented language, say L
n). Replace n in T
+.
In fact, the new
v <
by a new constant 7
Let M be a model of this new theory Tc.
/ ' Since M is an (,L)0,,50)-model, M expands to ( M ; F ) where F: 11'41
onto.
Suppose T has
.)I/,, T has a (K,A)-model.
Without loss of generality, we can assupe L disjoint from the
language of arithmetic. and
ff.
we say M is a (~,A)-modezof T if the cardinalities
+
ff is one-one and
Applying the Upward Lowenheim-Skolem Theorem toget an elementary extension
( N ; F ) of cardinality A, one also gets, by dropping F a (A,A)-model
Now iterate
K
of T,.
times the construction of elementary end extensions to get a
(K,A)-
model (Since U is interpreted by a bounded initial segment, no new elements occur at this time.) of T,.
Restricting this new model to the language L produces a
c. S M O R Y ~ S K I
so (K,X)-model of T.
&ED
The above proof is due to David Slater ( S l a t e r 1 9 7 2 ) .
The success of the argu-
ment may be a fluke; at least it is not clear how to obtain the more standard two-cardinal theorems by such an argument. 8.
DIVERSITY (THE 1980s)
It is still too early to attempt a discussion of the nonstandard model theory of the 1980s.
The only clear thing is the diversity the subject will display;
already one can discern several distinct directions: "hard" model theory, the relation to complexity theory, and the invasion of proof theory.
In addition to
these, there are, of course, continuations of programmes begun in the 1970s, when diversification began in earnest. It would exceed the scope (actually: the page limit) of this paper to go down any of these diverse paths any respectable distance.
In sections 1-7 I have presented
what I consider a minimal background and introduction to the study of nonstandard models of arithmetic.
I strongly recommend the reader next turn to the two
collections of expositions edited by Ken McAloon for further introduction to more recent trends.
Here, I wish merely to say a few unorganised words about some of
the things I've not discussed. My major omission has been the study of initial segments by Laurie Kirby and Jeff Paris (Kirby 2 9 7 7 , Kirby and Paris 1 9 7 7 , P a r i s 1 9 8 0 , and P a r i s 1 9 8 2 ; but cf. also P i Z z a y 2 9 8 2 B ) .
This study of combinatorially determined initial segments
modelling mere fragments of PA has had unexpected consequences-- most notably independence results in combinatorics (cf. e.g. P a r i s 1 9 7 8 , MeAZoon 1 9 7 9 , MeAZoon 1980A&B, Kirby 1 9 8 0 , Zbierski 1 9 8 0 , Tverskoy 1 9 8 0 , CZote 2 9 8 1 , von der Twer 1 9 8 2 ,
Friedman, MeAZoon and Simpson 1 9 8 2 , Ratajczyk 1 9 8 2 , Clote and M c A l o o n A , and
h d l a ' k A ) and in analysing the complexities of solutions to problems (cf. MeAZoon 2978B and CZote 1 9 8 1 ) .
Their work has even shed light on the study of elementary
end extensions (cf. Smoryiiski 1 9 8 1 C ) .
Fortunately, this development has been
amply documented and exposited, whence, though it is a major omission, it is not a serious one. A serious omission is the Kochen-Kripke notion of f u Z f i Z l a b i Z i t y .
In response to
Paris and Harrington's application of the work of Kirby and Paris to the problem of the combinatorial incompleteness of P A , Simon Kochen and Saul Kripke, in
Koehen and Kripke 1 9 8 2 , introduced an ultrapower-like construction useful for obtaining further such independence results.
Moreover, Kripke has derived from
this the ultrapower-free notion of fulfillability and he and Joseph Quinsey (the latter in his excellent thesis &u<nsey 1 9 8 0 ) demonstrated the general utility of fulfillability in the study of PA and its nonstandard models.
(I might note, for
example, that Quinsey has used fulfillability to detour the use of the Splitting
51
Lectures on Nonstandard Models of Arithmetic Theorem in establishing results in the uncountable case.)
When this material
becomes more widely exposited, it should find itself frequently applied. Very fashionable and used in the major omitted work of Kirby and Paris is their method of indicators.
Indicators are, as I announced back in section 2 , the
arithmetic manifestation of Strong Overspill.
Despite their fashionability and
an intuitive appeal, I am not much taken with indicators.
To my mind, i. they
fail as explanations of, say, the combinatorial incompleteness results because their whole machinery does not help construct the crucial indicators or prove them to be indicators, and ii. where they are successfully applied, they obscure more than they reveal.
The reader will have to either take my word for this or
dismiss my remark out of hand:
I have a lovely example, but am afraid to give it
because of the inevitable rumours about my personal vendettas; already I have heard it said that I "fight my battles fiercely" because I did not cite a modal logician in an algebra paper I wrote. Of course, I could be wrong about indicators in one area-- the study of models of subtheories of PA.
Back in the Preface I announced that the nonstandard pioneers
of the 1950s and 1960s proved theorems about strong models of arithmetic, i.e. elementary extensions of No, because they were not familiar with the extent to which such things could be formalised in PA.
In this complexity-conscious age,
the drift has been toward studying models of fragments ofPA (cf. e.g. Paris 1981,
Wilkie and Paris 1981, McAloon 1982, Gaifian and Lhitricopou1ou.s 1982, and
Paris and Dimitricopoulous 1 9 8 2 ) .
I am not familiar enough with this work to
determine if the use of indicators can be avoided. Another topic I've neglected that has assumed an increased interest is that of
nonstandard satisfafaction classes.
This notion originated with Abraham Robinson
(Robinson 1963) and was rejuvenated by S. Krajewski (Krajewski 1974A&B, m j e w s k i 1 9 7 6 ) and has begun to be energetically studied in Poland (cf. also KotZarski,
Krajewski and LachZan 1981, Lachlan 1981, and Ratajczyk 1 9 8 2 ) . I think I've said enough.
In the next section, I close by following a line that
does n o t go very far. 9.
A DEAD END
In contrast to initial segments, end segments of models of arithmetic have not been much studied.
In fact, they have not been studied at all.
Jensen and
Ekrenfeucht 1 9 7 6 speculates briefly on the idea that an end segment is somehow the "ultimate" arithmetic untouched by early irregularities and propose two problems (numbers 3 and 4 ) on end segments.
In February 1977 I solved these
problems in a simple and disappointing manner:
A nonstandard model of arithmetic
is uniquely determined by any of its end segments.
c. S M O R ~ S K I
52 9.1.
Theorem. Let M be a model of PA and suppose ]MI = I UE, where I is an
initial segment of M and E the end segment lMI recoverable from the structure E
-
I. Then: M is completely
(E;+,-).
=
The recovery is a variation of the construction of a field of quotients of an integral domain. The only subtle point is the means of singling out the allowable quotients. For, the formula, v0I1v1: 3 V Z ( V l = V0V2)’ only defines divisibility in the domain of quantification: For a , b
EbalIb
3 c
iff
E
E
E,
~(~bb=&).
For a proper definition of divisibility, one needs some intermediate notions: vo
((
a v (v
vl:
2 2
vo < v l :
svo
vo
((
v0I7J1: v
+ v2
=
vl)
I
vo <
v
=
B v
< u 1 (v2 - v 0 )
VlAVV2
I v (v + 0 1 1 0
I).
Note that < defines in E the restriction to E of the ordering of M, S the suc-
I
cessor function of M, and
the restriction to E of the divisibility relation of
M. With these, we can now “interpret“ M in E by:
M ( v o , v I ) : vl I v o (u
0’
v ) E (v
1
(Vo,V1)
(V,,V,)
2’
v ):
‘0’3
3
0 ( u 2 ,v 3 ) 0(v2,v3)
=
vlvZ
+ vlvz,~1v3) (vo”2,vIv3).
= (VOV3 =
The interpretation of M is: 9.2.
Lemma.
Let M
=
(M/Z;@/5,@/-)
via the map ( a , b ) / !
-+
{ (a,b) .-
E
E X E:
E
M(Z,F) 3 .
Then:
(IMl;+,*)
a / b , where ( )/E denotes the obvious quotient structure
modulo E. The proof of 9.2 is routine. From Lemma 9 . 2 , one quickly concludes Theorem 9.1.
Another corollary is the
following: 9.3.
Corollary. Let M ,M be models of PA with isomorphic end segments El.E2, 1 2 MI,M2 are isomorphic.
respectively. Then:
A closer look at the construction yields a few extra conclusions.
Combining the
interpretation of M in E with the coding properties of the standard systems of nonstandard models, one can prove the following: 9.4. Corollary. I plete Ill.
On the other hand:
The theory of nonstandard parts of nonstandard models is com-
Lectures on Nonstandard Models of Arithmetic 9.5.
53
Corollary. The theory of end segments of nonstandard models of PA is recur-
sively axiomatisable.
10. SUMMER READING LIST Before I begin, let me cite two books of particular value: Kenneth McAloon, Editor 1980
Mo&Zes de 11Arithm6tique, Aste'risque 73.
Chantal Berline, Kenneth McAloon and J.-P. Ressayre, Editors 1981
Model Theory and A r i t h m e t i c , Springer, Heidelberg.
These contain a number of expositions of more recent and advanced material than
I have covered in these lectures. As for the rest of the bibliography, a few words of explanation are in order:
I
have not listed abstracts unless the papers they abstracted did not appear. Unpublished papers that got absorbed into larger published works or which I doubted would ever get published or which I have not seen I have conservatively decided not to list.
But for this, I have been fairly liberal and have included
some papers cited in the body of my lectures which are not truly papers on models of arithmetic. There is no hope that the list is up-to-date, but it offers a decent introduction to the literature. Fred Abramson and Leo Harrington 1978
Models without indiscernibles, JSL 43, pp. 572-600.
A. Adler 1969
Extensions of nonstandard models of number theory, ZML 15, pp. 289-290.
Norman L. Alling 1965
Rings of continuous integer-valued functions and nonstandard arithmetic, Trans. AMS 118, pp. 498-525.
Jon Barwise and John Schlipf 1975
On recursively saturated models of arithmetic, in: D.H. Saracino and V.B. Weispfenning, eds., Mode2 Theory and Algebra; A Memorial !Tribute t o Abraham Robinson, Springer-Verlag, Heidelberg.
Chantal Berline 1981
Iddaux des anneaux de Pdano (d'apr'es CHERLIN), in: and Ressayre, cited above.
Berline, McAloon
Andreas Blass 1972
The intersection of nonstandard models of arithmetic, JSL 37, pp. 103106.
c. S M O R Y ~ K I
54 Andreas Blass (cont.) 1974
On certain types and models for arithmetic, J S L 39, pp. 151-162.
1977A
End extensions, conservative extensions, and the Rudin-Frolik ordering, Trans. AMS 225, pp. 325-340.
1977B
Amalgamation of nonstandard models of arithmetic, JSL 42, pp. 372-386.
1978
A model-theoretic view of some special ultrafilters, in: A. Macintyre, L. Pacholski and J. Paris, eds., Logic Colloquiwn '77, North-Holland, Amsterdam.
1980
Conservative extensions of models of arithmetic, Arch. f. math. Logik 20, pp. 85-94.
George Boolos 1973
A note on Beth's
theorem, Bull. Acad. Polon. Sci. 21, pp. 1-2.
Alan Cantor 1972
Minimal proper elementary extensions of N with respect to definable relations, Dissertation, Univ. S. Carolina.
Hans Georg Carstens 1972
uber die Kompliziertheit numerischer Modelle, Dissertation, Minster.
1975
Reducing hyperarithmetic sequences, Fund. Math. 89, pp. 5-11.
1976
0 A -Mengen, Arch. f. math. Logik 18, pp. 55-65. 2
Patrick Cegielski 1981
Theorie Elementaire de la multiplication des entiers naturels, in: Berline, McAloon and Ressayre, cited above.
Patrick Cegielski, Kenneth McAloon and George Wilmers 1982
Zoe'
Modsles re'cursivement saturgs de l'addition et de la multiplication des entiers naturels, in: D. van Dalen, D. Lascar and T.J. Smiley, eds., Logic Cozloquiwn '80, North-Holland, Amsterdam.
Chatzidakis
1981
La reprgsentation en termes de faisceaux des modiles de la th6orie glgmentaire de la multiplication des entiers naturels, in: Berline, McAloon and Ressayre, cited above.
Greg Cherlin 19758
Ideals of integers in nonstandard number fields, in: D.H. Saracino and V.B. Weispfenning, eds., Mode2 Theory and Algebra; A Memorial Tribute to Abrahmn Robinson, Springer-Verlag, Heidelberg.
1975B
Ideals in some nonstandard Dedekind rings, Logique et Analyse 18, pp. 349-406.
Lectures on Nonstandard Models of Arithmetic
55
Grigory Chudnovsky 1971
Certain arithmetical problems, (Russian), Kiev Univ. preprint.
Peter Clote 1980
Weak partition relations, finite games, and independence results in Peano arithmetic, in: L. Pacholski, J. Wierzejewski and A.J. Wilkie, eds., Model Theory of Algebra and Arithmetic, Springer-Verlag, Heidelberg.
1981
Anti-basis theorems and their relation to independence results in Peano arithmetic, in: Berline, McAloon and Ressayre, cited above.
Peter Clote and Kenneth McAloon A
Two further combinatorial theorems equivalent to the 1-consistency of Peano arithmetic,
Maryvonne Daguenet 1972
Un modsle non standard de l'arithmgtique, Compte Rendus 274, Serie A, pp. 685-688.
A. M. Dawes 1972
First-order hierarchies in general models and in models of Peano arithmetic, Dissertation, Toronto.
1977
End extensions which are models of a given theory, ZML 23, pp. 463-467.
S.D. Denisov 1972
Models of noncontradictory formulas and the Ershov hierarchy, (Russian), Alg. i Logika 11, pp. 648-655; English translation: Alg. and Logic 11 (1974), pp. 359-362.
Max Dickmann 1980
Types remarquables et extensions de modsles dans 1'arithm;tique Peano, I, Asterisque 73, pp. 59-117.
de
Lou van den Dries 1980
Some number theory and model theory of weak systems of arithmetic, in: L. Pacholski, J. Wierzejewski and A.J. Wilkie, eds., Mode2 Theory of AZgebra and Arithmetic, Springer-Verlag, Heidelberg.
Andrzej Ehrenfeucht 1973
Discernible elements in models for Peano arithmetic, JSL 38, pp. 291292.
Andrzej Ehrenfeucht and Georg Kreisel 1966
Strong models for arithmetic, Bull. Acad. Polon. Sci. 14, pp. 107-110.
Solomon Feferman 1958
Arithmetically definable models of formalized arithmetic, Notices AMS 5 , pp. 679-680.
c. SMORYE~SKI
56 Solomon Feferman (cont.) 1959
Models of arithmetic obtained from various sets of functions, Notices AMS 6, pp. 382-383.
1960
Arithmetization of metamathematics in a general setting, Fund. Math. 49, pp. 35-92.
Solomon Feferman and Georg Kreisel 1966
Persistent and invariant formulas relative to theories of higher order, B u l l . AMS 72, pp. 480-485.
Solomon Feferman, Georg Kreisel and Steven Orey 1962
1-consistency and faithful interpretations, Arch. f. math. Logik 6, pp. 52-63.
Solomon Feferman, Dana Scott and Stanley Tennenbaum 1959
Models of arithmetic through function rings, Notices AMS, pp. 173-174.
Harvey Friedman 1970
Iterated inductive definitions and $-AC,
in:
A . Kino, J. Myhill and
R. Vesley, eds., I n t u i t i o n i s m and Proof Theory, North-Holland, Amsterdam. 1973
Countable models of set theories, in: A.R.D. Mathias and H. Rogers, eds., Cambridge S m e r School i n Mathematical Logic, Springer-Verlag, Heidelberg.
1975
One hundred and two problems in mathematical logic, JSL 4 0 , pp. 113129.
Haim Gaifman 1967
Uniform extension operators for models and their applications, in: Crossley, ed., S e t s , Models and Recursion Theory, North-Holland, Amsterdam.
1970
On local arithmetical functions, and thier applications for constructing types of Peano's arithmetic, in: Y. Bar-Hillel, ed., Mathematical Logic and Foundations of S e t Tkeory, North-Holland, Amsterdam.
1972
A note on models and submodels of arithmetic, in: W. Hodges, ed., Conference i n Mathematical Logic-- London '70, Springer-Verlag, Heidelberg.
1976
Models and types of Peano's arithmetic, Annals math. Logic 9, pp. 223-
J.
306.
Haim Gaifman and Constantine Dimitracopoulous 1982
Fragments of Peano's arithmetic and the MRDP theorem, in: E. Engeler, H. Lzuchli and V. Strassen, eds., Logic and Algorithmic, L'Enseignement Mathgmatique, Geneva.
Lectures on Nonstandard Models of Arithmetic Robin 0. Gandy 1958
Note on a paper of Kemeny's, Math. Ann. 136, p. 466.
D.C. Goldrei, Angus Macintyre and Harry Simmons 1973
The forcing companions of number theories, Israel 3. Math. 1 4 , pp. 317337.
Andrzej Grzegorczyk 1971
An unfinitizability proof by means of restricted reduced power, Fund. Math. 73, p p . 37-49.
David Guaspari 1979
Partially conservative extensions of arithmetic, Trans. AMS 254, p p . 47-68.
Petr Hgjek 1971/ 1972
On interpretability in set theories I, 11, Comment. Math. Univ. Carol. 1 2 , pp. 73-79; 13, pp. 445-455.
1981A
Completion closed algebras and models of Peano arithmetic, Comment. Math. Univ. Carol. 22, pp. 585-594.
1981B
On interpretability in theories containing arithmetic 11, Comment. Math. Univ. Carol. 22, pp. 667-688.
Petr Hdjek and Pave1 Pudlsk 1980
Two orderings of the class of all countable models of Peano arithmetic,
in: L. Pacholski, J. Wierzejewski and A.J. Wilkie, eds., Model Theory of Algebra and A r i t h m e t i c , Springer-Verlag, Heidelberg. Gisbert Hasenjsger 1953
Eine Bemerkung zu Henkin's Beweis fur die Vollstzndigkeit des Prgdikatenkalkcls der ersten Stufe, JSL 18, pp. 42-48.
Kurt Hauschild 1963
ffber die Charakterisierbarkeit der Zahlenreihe in gewissen Nichtstandardmodellen der Arithrnetik, ZML. 9 , pp, 113-116.
1971
Nichtaxiomatisierbarkeit von Satzmengen durch Ausdrzcke spezieller Gestalt, Fund. Math. 72, pp. 245-253.
Leon Henkin 1950
Completeness in the theory of types, J S L 15, pp. 81-91.
G. Hensel and Hilary Putnam 1969
Normal models and the field
Z;, Fund. Math. 64, pp. 231-240.
Joram Hirschfeld 1972
Existentially complete and generic structures in arithmetic, Disserta-
c. S M O R Y ~ K I
58 Joram Hirschfeld (cont.) tion, Yale. 1974
Models of arithmetic and the semi-ring of recursive functions, in: A . Hurd and P. Loeb, eds., Vietoria Symposim on Nonstandard Analysis, Springer-Verlag, Heidelberg.
19758
Finite forcing and generic filters in arithmetic, in: D.H. Saracino and V.B. Weispfenning, eds., Model Theory and Algebra; A Memorial Tribute t o Abraham Robinson, Springer-Verlag, Heidelberg.
1975B
Models of arithmetic and recursive functions, Israel J. Math. 20, pp. 111-126.
Joram Hirschfeld and William Wheeler 1975
Forcing, Arithmetic and Division Rings, Springer-Verlag, Heidelberg.
Paul Howard 1972
A proof of a theorem of Tennenbaum, ZML 18, pp. 111-112.
Don Jensen and Andrzej Ehrenfeucht 1976
Some problem in elementary arithmetics, Fund. Math. 92, pp. 223-245.
R.G. Jeroslow 1971
Non-effectiveness in S. Orey's arithmetical compactness theorem, ZML 1 7 , pp. 285-289.
Carl G. Jockusch 1972
Ramsey's theorem and recursion theory, JSL 37, pp. 268-280.
1974
Il1 classes and boolean combinations of recursively enumerable sets,
0
JSL 39, pp. 95-96.
Carl G. Jockusch and Robert I. Soare 0
19728
II1 classes and degrees of theories, Trans. AMS 173, pp. 33-56.
1972B
Degrees of members of JIo classes, Pac. J. Math. 40, pp. 605-616. 1
Matt Kaufman 1977
A rather classless model, Proc. AMS 6 2 , pp. 330-333.
Jean Pierre Keller and Denis Richard 1978
Remarques sur les structures additives des modzles de l'arithmgtique, Compte Rendus 287, Serie A , pp. 101-104.
J.G. Kemeny 1958
Undecidable problems of elementary number theory, Math. Ann. 135, pp. 160-169.
Lectures on Nonstandard Models of Arithmetic
59
Jussi Ketonen and Robert M. Solovay 1981
Rapidly growing Ramsey functions, Annals of Math. 113, pp. 267-314.
Laurie Kirby 1977
Initial segments of models of arithmetic, Dissertation, Manchester.
1980
La mgthod des indicatrices et le th6orsme d'incomplgtude, Astdrisque 73, pp. 5-18.
1982
Flipping properties in arithmetic, JSL 47, pp. 416-422.
Laurie Kirby, Kenneth McAloon and Roman Murawski 1981
Indicators, recursive saturation and expandability, Fund. Math. 114, pp. 127-139.
Laurie Kirby and Jeff Paris 1977
Initial segments of models of Peano's axioms, in: A.H. Lachlan, M. Srebrny and A. Zarach, eds., S e t Theory and Hierarchy Theory V, Springer-Verlag, Heidelberg.
Steven C. Kleene 19528
Two papers on the predicate calculus, Memoirs AMS 10.
1952B
Introduction t o Metmathematics, van Nostrand, Princeton.
Julia Knight 1973
Complete types and the natural numbers, JSL 38, pp. 413-415.
1975
Types omitted in uncountable models of arithmetic, JSL 40, pp. 317-320.
19768
Omitting types in set theory and arithmetic, JSL 41, pp. 25-32.
1976B
Hanf numbers for omitting types over particular theories, JSL 41, pp. 583-588.
A
Additive structure in uncountable models for a fixed completion of P,
Julia Knight and Mark Nadel 19828
Expansions of models and Turing degrees, JSL 47, pp. 587-604.
1982B
Models of arithmetic and closed ideals, JSL 4 7 , pp. 833-840.
Simon Kochen and Saul Kripke 1982
Nonstandard models of Peano arithmetic, in: E. Engeler, H. Lguchli and V. Strassen, eds., Logic and AZgorithmic, L'Enseignement MathGmatique, Geneva; and: L'Enseignement Mathgmatique 28 ( 1 9 8 2 ) , pp. 211-231.
Roman Kossak 1980
An application of definable types of Peano's arithmetic, Bull. Acad. Polon. Sci. 28, pp. 213-217.
A
A certain class of models of Peano arithmetic,
c. SMORYI~SKI
60 Roman Kossak (cont.) B
L
C
A note on satisfaction classes,
-,q-elementary
equivalence of w -like models of PA, 1
Henryk Kotlarski 1980
On Skolem ultrapowers and their non-standard variant, ZML 26, pp. 227-
236.
A
On elementary recursively saturated cuts in models of PA, I-IV,
B
On cofinal extensions of models of arithmetic,
Henryk Kotlarski, S. Krajewski and Alisdair H. Lachlan 1981
A construction of satisfaction classes for nonstandard models, Canad. Math. Bull. 24, pp. 295-297.
S . Krajewski
19748
Predicative expansions of axiomatic theories, ZML 20, pp. 435-452.
1974B
Mutually inconsistent satisfaction classes, B u l l . Acad. Polon. Sci. 22, pp. 883-887.
1976
Nonstandard satisfaction classes, in: W. Marek, M. Srebrny and A . Zarach, eds., S e t Theory and. Hierarchy Theory, Springer-Verlag, Heidelberg.
Georg Kreisel 1950/ 1951
Note on arithmetic models for consistent formulae of the predicate calculus, Fund. Math. 37, pp. 265-285.
1953
Note on arithmetic models for consistent formulae of the predicate calculus, 11, in: Actes du XIemeCongrBsInternational de PhiZosophie, North-Holland, Amsterdam.
1968
Survey of proof theory, JSL 33, pp. 321-388.
Alisdair H. Lachlan 1981
Full satisfaction classes and recursive saturation, Canad. Math. Bull: 24, pp. 271-292.
D. Lascar 1980
Une indicatrice de type "Ramsey" pour l'arithmgtique de Peano et le formule de Paris-Harrington, Astgrisque 73, pp. 19-30.
Hamid Lesan 1978
Models of arithmetic, Dissertation, Manchester.
Leonard Lipshitz 1979
Diophantine correct models o f arithmetic, Proc. AMS 73, pp. 107-108.
61
Lectures on Nonstandard Models of Arithmetic Leonard Lipshitz and Mark Nadel 1978
The additive structure of models of arithmetic, Proc. AMS 68, pp. 331336.
R. MacDowell and Ernst Specker 1961
Modelle der Arithmetik, in: I n f i n i t i s t i c Methods, Pergamon, London.
Angus Macintyre 1980
Ramsey quantifiers in arithmetic, in: L. Pacholski, J. Wierzejewskj and A.J. Wilkie, eds., Model Theory of Algebra and Arithmetic, Springer-Verlag, Heidelberg.
1981
The complexity of types in field theory, in: M. Leman, J. Schmerl and R. Soare, eds., Logic Y e a r 1979-80, Springer-Verlag, Heidelberg.
1982
Residue fields of models of P, in: L.J. Cohen, J. Lo;, H. Pfeiffer and K. -P. Podewski, eds. , Logic, Methodo Zogy and Phi Zosophy of Science V I , North-Holland, Amsterdam.
Angus Macintyre and David Marker A
Degrees of recursively saturated models,
Angus Macintyre and Harry Simmons 1975
Algebraic properties of number theories, Israel J. Math. 22, pp. 7-27.
Attila Mate' A
Nondeterministic polynomial time computations and nonstandard models of arithmetic,
Larry Manevitz 1976
Internal end-extensions of Peano arithmetic and a problem of Gaifman, J. London Math. SOC. (2) 13, pp. 80-82.
Larry Manevitz and Jonathan Stavi 1980
0 operators and alternating sentences in arithmetic, JSL 45, pp. 1442 154.
A
Wiktor Marek and Marian Srebrny 1976
Urelements and extendability, in: W. Marek, M. Srebrny and A. Zarach, eds., Set Theory and Hierarchy Theory, Springer-Verlag, Heidelberg.
Kenneth McAloon 1977
Consistency statements and number theories, in: CoZZoque International de Logique, CNRS, Paris.
19788
Completeness theorems, incompleteness theorems and models of arithmetic, Trans. AMS 239, pp. 253-277.
1978B
Diagonal methods and strong cuts in models of arithmetic, in: A . Macintyre, L. Pacholski and J. Paris, eds., Logic Cozzoquim ' 7 7 , North-
M. Guillaume, ed.,
c. S M O R Y ~ K I
62 Kenneth McAloon (cont.) Holland, Amsterdam. 1979
Paris et the'or'eme d'incomplgtude (d'apres Forms combinatoires d'autres), in: Semimire Bourbaki V O Z . ' 1 9 7 7 / 1 9 7 8 E h o s e s 507-524, Springer-Verlag, Heidelberg.
1980A
La rapport entre la mgthode de Ggdel et la mgthode des indicatrices pour obtenir des rgsultats d'independence, Astgrisque 73, pp. 31-39.
1980B
Progressions transfinies de thgories axiomatiques, formes combinatoires du thgor'eme d'incomplgtude et fonctions recursives ?I croissance rapide, Astgrisque 73, pp. 41-58.
1982
On the complexity of models of arithmetic, JSI. 47, pp. 403-415.
Elliot Mendelson 1961
On non-standard models for number theory, in: Y. Bar-Hillel, ed., Essays on the Foundations o f Mathematics, Magnes Press, Jerusalem.
iarko Mijajlovi6 A
Submodels and definable points in models of Peano arithmetic,
George Mills 1977 1978
Extensions of models of Peano arithmetic, Dissertation, Berkeley. A model of Peano arithmetic with no elementary end extension, JSL 43, p p . 563-567.
1979
Substructure lattices of models of arithmetic, Annals Math. Logic 16, pp. 145-180.
Didier Misercque 1980
Sur le trellis des V 1-formules fermges de l'arithmgtique de Peano, Compte Rendus 290, Serie A, pp. 571-573.
1982
The nonhomogeneity of the E-tree-- answer to a problem raised by D. Jensen and A. Ehrenfeucht, Proc. AMS 84, pp. 573-575.
Josef Mlzek 1976
Twin prime problem in arithmetic without induction, Comment. Math. Univ. Carol. 17, pp. 543-555.
1978A
End extensions of countable structures and the induction schema, Comment. Math. Univ. Carol. 19, pp. 291-308.
1978B
A note on cofinal extensions and segments, Comment. Math. Univ. Carol. 19, p p . 727-742.
Carl Morgenstern 1982
On generalized quantifiers in arithmetic, JSL 47, p p . 187-190.
Lectures on Nonstandard Models of Arithmetic
63
Andrzej Mostowski 1952A
On models of axiomatic systems, Fund. Math. 39, pp. 133-158.
1952B
On direct products of theories, JSL 1 7 , pp. 1-31.
1957
On recursive models of formalized arithmetic, Bull. Acad. Polon. Sci. 5 , pp. 705-710.
Nobuyoshi Motohashi A
Preservation theorem and relativization theorem for cofinal extensions,
B
A normal form theorem for first order formulas and its application to
Gaifman's splitting theorem, Gert H. Msller 1961
Nicht-Standardmodelle der Zahlentheorie, Math. Zeitschr. 77, pp. 414438.
Roman Murawski 19761 1977
On expandability of models of Peano arithmetic I, 11, 111, Studia Logica 35, pp. 409-419; 35, pp. 421-431; 36, pp. 181-188; Correction: Studia Logica 36.
1978
Indicators, satisfaction classes and expandability, preprint, Poznan.
1979
Indicators and the structure of expansions, preprint, Poznan.
1980
Some remarks on the structure of expansions, ZML 26, pp. 537-546.
1981
Incompleteness of Z -definable theories via indicators, Func. et Approx. 11, pp. 57-63.
A
A simple remark on satisfaction classes, indiscernibles and recursive
0
saturation, Mark Nadel 1980
On a problem of MacDowell and Specker, JSL 45, pp. 612-622.
A
The completeness of Peano multiplication,
Steven Orey 1961
Relative interpretations, ZML 7 , pp. 156-153.
F. Pabion 1982
Saturated models of Peano arithmetic, JSL 47, pp. 625-637.
Jeff Paris 1972
On models of arithmetic, in: W. Hodges, ed., Conference in Mathematic a l Logic-- London '70, Springer-Verlag, Heidelberg.
1977
Models of arithmetic and the 1-3-1 199.
lattice, Fund. Math. 95, pp. 195-
c. SMORYI~SKI
64
Jeff Paris (cont.) 1978A
Note on an induction axiom, J S L 43, pp. 113-117.
1978B
Some independence results for Peano arithmetic, JSL 43, pp. 725-731.
1980
A hierearchy of cuts in models of arithmetic, in: L. Pacholski, J. Wierzejewski and A.J. Wilkie, eds., Model Theory of AZgebra and Arithmetic, Springer-Verlag, Heidelberg.
1981
Some conservation results for fragments of arithmetic, in: McAloon and Ressayre, cited above.
Berline,
Jeff Paris and Constantine Dimitracopoulous A
Truth definitions for A.
formulae,
Jeff Paris and Leo Harrington 1977
A mathematical incompleteness in Peano arithmetic, in: J. Barwise, ed., Handbook of Mathematical Logic, North-Holland, Amsterdam.
Jeff Paris and Laurie Kirby 1978
X
-
collection schemas in arithmetic, in: A. Macintyre, L. Pacholski
and J. Paris, eds., Logic Colloquium '77, North-Holland, Amsterdam. Jeff Paris and George Mills 1979
Closure properties of countable non-standard integers, Fund. Math. 103, pp. 205-215.
Robert G. Phillips 1971
On the structure of nonstandard models of arithmetic, Proc. AMS 27, pp. 359-363.
1972
Addition in nonstandard models of arithmetic, JSL 37, pp. 483-486.
19748
Omitting types in arithmetic and conservative extensions, in: A. Hurd and P. Loeh, eds., V i c t o r i a Symposium on Nonstandard Analysis, Springer-Verlag, Heidelberg.
1974B
A minimal extension that is not conservative, Mich. Math. J. 21, pp. 27-32.
Anand Pillay 1981A
Models of Peano arithmetic (a survey of basic results), in: McAloon and Ressayre, cited above.
Berline,
1981B
Cuts in models of arithmetic, in: above.
1981C
Partition properties and definable types in Peano arithmetic, in: Berline, McAloon and Ressayre, cited above.
Berline, McAloon and Ressayre, cited
Klaus Potthoff 1967
Untersuchungen Eber Nichtstandardmodelle, Dissertation, Hannover.
Lectureson Nonstandard Models of Arithmetic
65
Klaus Potthoff (cont.) 1969
Uber Nichtstandardmodelle der Arithmetik und der rationalen Zahlen, ZML 15, pp. 223-236.
1970
Ideale in Nichtstandardmodellen der ganzen Zahlen, ZML 16, pp. 321326.
1972
Ordnungseigenschaften von Nichtstandardmodellen, in: G. Asser, J. Flachsmeyer and W. Rinow, eds., Theory of S e t s and Topology; I n Honour of Felix Hausdorff, VEB, Berlin.
1976
A simple tree lemma and its application to a counterexample of Phillips, Arch. f. math. Logik 18, pp. 67-71.
1978
Orderings of types of countable arithmetic, ZML 24, pp. 97-108.
M. Presburger 1930
uber die Vollsthdigkeit eines gewissen Systems der Arithmetik ganzer Zahlen, in welchem die Addition als einzige Operation hervortritt, in: Comptes Rendus I Congrzs des Maththaticiens des Pays Slaves, Warsaw.
Pavel PudlCk A
Another combinatorial sentence independent of Peano's axioms,
Pavel Pudlhk and Antonin Sochor A
Models of the Alternative Set Theory,
C. Puritz 1971
Ultrafilters and standard functions in non-standard arithmetic, Proc. London Math. SOC. ( 3 ( 22, pp. 705-733.
Hilary Putnam 1965
Trial and error predicates and the solution to a problem of Mostowski, JSL 30, pp. 49-57.
Joseph Quinsey 1980
Some problems in logic, Dissertation, Oxford.
Michael 0. Rabin 1961
Non-standard models and the independence of the induction axiom. in: Y. Bar-Hillel, ed., Essays on t h e Foundations of bkzthernatics, Magnes Press, Jerusalem.
1962
Diophantine equations and non-standard models of arithmetic, in: E. Nagel, P. Suppes and A . Tarski, eds., Logic, Methodology and Philosophy of Science, Stanford.
2. Ratajczyk
1982
Satisfaction classes and combinatorial sentences independent from arithmetic, ZML 28, pp. 149-165.
c. S M O R Y ~ S K I
66 J.-P. Ressayre 1980
Types remarquable et extensions de modsles dans l'arithmgtique de Peano 11, (with appendix by M. Dickmann), Astgrisque 73, pp. 119-154.
Denis Richard 1977
On external properties of nonstandard models of arithmetic, Publ. Dep. Math. (Lyon) 14, no. 4, pp. 57-75.
1980
Saturation des modsles de Peano, Compte Rendus 290, Serie A, pp. 351353.
1981
De la structure additive 'a la saturation des modsles de Pgano et 'a une classification des sous-langages de l'Arithmectique, in: Berline, McAloon and Ressayre, cited above.
Abraham Robinson
C.
1963
On languages which are based on nonstandard arithmetic, Nagoya Math. J. 22, pp. 83-117.
1971
Infinite forcing in model theory, in: J.E. Fenstad, ed., Proceedings of t h e Second Scandinavian Logic Symposim, North-Holland, Amsterdam.
1973
Nonstandard arithmetic and generic arithmetic, in: P. Suppes, L. Henkin, G . C . Moisil and A. Joja, eds., Logic, Methodology and PhiZosophy o f Science I V , North-Holland, Amsterdam.
1961
Model theory and non-standard arithmetic, in: Pergamon, London.
Infinitistic Methods,
Ryll-Nardzewski 1952
The role of the axiom of induction in elementary arithmetic, Fund. Math. 39, pp. 239-263.
James Schmerl 1973
Peano models with many generic classes, Pac. J. Math. 46, pp. 523-536; Errata: Pac. J. Math 92 (1981), pp. 195-198.
1978
Extending models of arithmetic, Ann. Math. Logic 14, pp. 89-109.
1981
Recursively saturated rather classless models of Peano arithmetic, in:
M. Lerman, J. Schmerl and R. Soare, eds., Logic Year 1979-80, Springer-Verlag, Heidelberg. A
Recursively saturated models generated by indiscernibles,
James Schmerl and Steven G . Simpson 1982
On the role of Ramsey quantifiers in first order arithmetic, JSL 47, pp. 423-235.
Dana Scott 1961
On constructing models for arithmetic, in: Pergamon, London.
Infinitistic Methods,
Lectures on Nonstandard Models of Arithmetic
67
Dana Scott (cont.) 1962
Algebras of sets binumerable in complete extensions of arithmetic, in: J. Dekker, ed., Recursive Function IPheory, A M S , Providence.
Saharon Shelah 1978A
End extensions and numbers of countable models, JSL 43, pp. 550-562.
1978B
Models with second order properties 11. trees with no undefined branches, Ann. Math. Logic 14, pp. 73-87.
John Shepherdson 1964
A non-standard model for a free-variable fragment of number theory, Bull, Acad. Polon. Sci. 12, pp. 79-86.
1965
Non-standard models for fragments of number theory, in: L. Henkin, P. Suppes and A. Tarski, eds., The Theory of Models, North-Holland, Amsterdam.
Joseph Shoenfield 1960
Degrees of models, JSL 25, pp. 233-237.
Harry Simmons 1976
Each regular number structure is biregular, Israel J. Math. 23, pp. 347-352; Correction: Israel J. Math. 26 (1977), p. 95.
1977
Existentially closed models of basic number theory, in: R. Gandy and M. Hyland, eds., Logic CoZZoquizim ‘76, North-Holland, Amsterdam.
Steven G. Simpson 1974
Forcing and models of arithmetic, Proc. AMS 43, pp. 193-194.
Thoralf Skolem 1929
uber einige Grundlagenfragen der Mathematik, Skrifter Vitensk. i Oslo I, no. 4 , pp. 1-49.
1930
uber einige Satzfunktionen in der Arithmetik, Skrifter Vitensk. Oslo I, no. 7, pp. 1-28.
1933
uber die UnmEglichkeit einer vollstzndigen Charakterisierung der Zahlenreihe mittels eines endlichen Axiomensystems, Norsk Mat. Forenings Skrifter, Ser. 11, nr. 1-12, pp. 73-82.
1934
Uber die Nichtcharakterisierbarkeit der Zahlenreihe mittels endlich oder abzghlbar unendlich vielen Aussagen mit ausschliesslich Zahlvariablen, Fund. Math. 23, pp. 150-161.
1955
Peano’s axioms and models of arithmetic, in: Mathematical Interpretat i o n s of Formal Systems, North-Holland, Amsterdam.
David Slater 1972
Non-standard languages and thier applications, Dissertation, Rutgers.
c. S M O R Y ~ S K I
68 C. Smoryikki 1973A
Investigation of intuitionistic formal systems by means of Kripke models, Dissertation, Chicago Circle.
1973B
Applications of Kripke models, in: A. Troelstra, ed., Metamathematical Investigation of I n t u i t i o n i s t i c Arithmetic and Analysis, SpringerVerlag, Heidelberg.
1977
The incompleteness theorems, in: J. Barwise, ed., Handbook of Mathematical Logic, North-Holland, Amsterdam.
1981A
Cofinal extensions of nonstandard models of arithmetic, Notre Dame J. Formal Logic 22, pp. 133-144.
1981B
Recursively saturated nonstandard models of arithmetic, J S L 46, pp. 259-286; Addendum: J S L 47 (1982), pp. 493-494.
1981C
Elementary extensions of recursively saturated models of arithmetic, Notre Dame J. Formal Logic 22, pp. 193-203.
19828
A note on initial segment constructions in recursively saturated models of arithmetic, Notre Dame J. Formal Logic 23, pp. 393-408.
1982B
Back-and-forth inside a recursively saturated model of arithmetic, in: D. van Dalen, D. Lascar and T. Smiley, eds., Logic Colloquiwn '80, North-Holland, Amsterdam.
1982C
Nonstandard models and constructivity, in: A. Troelstra and D. van Dalen, eds., L.E.J. Brouwer Centenary Symposium, North-Holland, Amsterdam.
C. Smory6ski and Jonathan Stavi 1980
Cofinal extension preserves recursive saturation, in:
L. Pacholski,
J. wierzejewski and A. Wilkie, eds., Mode2 Theory of Algebra and
Arithmetic, Springer-Verlag, Heidelberg. Robert M. Solovay A S.
On interpretability in set theories,
F. Soprunov 19758
Strong nonstandard models of arithmetic, (Russian), Dokl. AN SSSR 220, pp. 293-296; English translation: Sov. Math. Doklady 16 (1975), pp. 80-84.
1975B
Initial segments of nonstandard arithmetics, (Russian), Dokl. AN SSSR 223, pp. 576-577; English translation: Sov. Math. Doklady 16 (1975). pp. 968-970.
1976
Countable nonstandard models of arithmetic, (Russian), in: D.A. Bocvar and V.N. Grisin, eds., Studies i n S e t Theory and Nonclassical Logics, (Russian), Izdat. "Nauka", Moscow; English translation: Selecta Math. Sovietica 1 (1981), pp. 303-315.
1979
Lattices of nonstandard arithmetic, (Russian), in: Studies in NoncZassica1 Logic and S e t Theory, (Russian), Izdat. "Nauka", Moscow.
Lectureson Nonstandard Models of Arithmetic
69
Ernst Specker 1957
Eine Verschzrfung des Unvollst&digkeitssatzes Bull. Acad. Polon. Sci. 5 , pp. 1041-1045.
der Zahlentheorie,
Teodor Stihi 1976
Forcing and Peano arithmetic, Dissertation.
1978
Induction and complete sequences of conditions in the sense of Cohen, Rev. Roum. Math. Pures et Appl. 23, pp. 129-133.
Stanley Tennenbaum 1959
Non-archimedean models for arithmetic, Notices AMS 6 , p. 270.
Akito Tsuboi 1982
On M-recursively saturated models of arithmetic, Tsukuba J. Math. 6 , pp. 305-318.
A.A. Tverskoy 1980
A sequence of combinatorial judgements which are independent of Peano
arithmetic, Vestnik Moskov. Univ. Matem. 35, no. 5, pp. 7-13; English translation: Mosc. Univ. Math. Bull. 35, no. 5 (1980), pp. 6-13. Tassilo von der Twer 1981
Some remarks on the mathematical incompleteness of Peano's arithmetic found by Paris and Harrington, in: R. Jensen and A. Prestel, eds., S e t Theory and Model Theoi-y, Springer-Verlag, Heidelberg.
Jouko Vyingnen 1979
A new incompleteness in arithmetic, (Finnish, English summary), Arkhimedes 31, pp. 30-37.
Hao Wang 1951A
Arithmetic translations of axiom systems, Trans. AMS 71, pp. 283-293.
1951B
Arithmetic models for formal systems, Methodos 3, pp. 217-232.
Alec J. Wilkie 1975
On models of arithmetic-- answers to two problems raised by H. Gaifman, J S L 40, pp. 41-47.
19778
On models of arithmetic with non-modular substructure lattices, Fund. Math. 95, pp- 223-238.
1977B
On the theories of end-extensions of models of arithmetic, in: A. Lachlan, M. Srebrny and A. Zarach, eds., S e t Theory and Model Theory V, Springer-Verlag, Heidelberg.
1978
Some results and problems on weak systems of arithmetic, in: A. Macintyre, L. Pacholski and J. Paris, eds., Logic ColZoquhm '77, North-Holland, Amsterdam.
c. S M O R Y ~ ~ S K I
70
Alec J . Wilkie (cont.) 1980
Applications of complexity theory to 1 -definability problems in
0
arithmetic, in: L. Pacholski, J. Wierzejewski and A. Wilkie, eds., Model Theory of Algebra and A r i t h e t i c , Springer-Verlag, Heidelberg. 1981
On discretely ordered rings in which every definable ideal is principal, in: Berline, McAloon and Ressayre, cited above.
1982
On core structures for Peano arithmetic, in: D. van Dalen, D. Lascar and T. Smiley, eds., Logic Colloquiwn '80, North-Holland, Amsterdam.
A
On the arithmetical parts of strong theories, unpublished.
Alec 3. Wilkie and Jeff Paris 1981
Models of arithmetic and the rudimentary sets, Bull. SOC. Math. Belg. Ser. B 33, pp. 157-169.
George Wilmers 1980
Minimally saturated models, in: L. Pacholski, 3. Wierzejewski and A. Wilkie, eds., Model Theory of Algebra and Arithmetic, SpringerVerlag, Heidelberg.
George Zahn 1971
On the additive structure of non-standard models of arithmetic, Dissertation, Univ. S. Carolina.
Pawel Zbierski 1980
Indicators and incompleteness of Peano arithmetic, Acta. Cient. Venezolana 31, pp. 487-495.
After typing the above, I had a chat with David Marker, who gave me the following additional items: Julia Knight, Alisdair Lachlan and Robert I. Soare A
Two theorems on degrees of models of true arithmetic,
Henryk Kotlarski A
Automorphisms of recursively saturated models of PA,
David Marker 1982
Degrees of models of true arithmetic, in: loquiwn ' 8 1 , North-Holland, Amsterdam.
Jeff Paris and Alec J. Wilkie A
A -sets and induction,
0
George Wilmers A
Bounded existential induction,
J . Stern, ed., Logic Coz-
LOGIC COLLOQUIUM '82 G. Lolli, G. Long0 and A . Marcia (editors] 0 Elsevier Science Publishers B. V. (North-Holland), I984
71
STRONG SATURATION A N D THE FOUNDATIONS OF STABILITY THEOPY
John T. B a l d w i n Department o f Kathematics, S t a t i s t i c s and Computer Science U n i v e r s i t y o f I l l i n o i s a t Chicago
The n o t i o n o f n o n - f o r k i n g i s one o f t h e p r i n c i p a l t o o l s o f s t a b i l i t y t h e o r y . Since b e i n g i n t r o d u c e d b y Shelah L197E1,
t h e r e have been a number o f e o u i v a l e n t
d e f i n i t i o n s o f t h e concept: i n t e r m s o f t h e fundamental o r d e r i n Lascar-Poizat w i t h v a r i a n t s stemming from an i d e a o f Z i e g l e r i n Balduin-198x and and i n t e r m s o f r a n k i n Shelah-197E and Lascar C19761.
Harnik-Harrington-19?2
We
b e g i n h e r e b y g i v i n g an o p e r a t i o n a l d e f i n i t i o n o f n o n - f o r k i n g as a r e l a t i o n s a t i s f y i n g c e r t a i n axioms.
Then a l l t h e o t h e r d e f i n i t i o n s can be viewed as p r o o f s we w i l l show i n
o f t h e e x i s t e n c e o f a r e l a t i o n s a t i s f y i n g t h e axioms.
Poreover,
t h e s p i r i t o f Lascar ( c f .
t h a t c e r t a i n o f these
a l s o Harnik-Harrington-19?2)
axioms a r e c a t e g o r i c a l and t h u s c o n c l u d e t h e i d e n t i t y o f t h e v a r i o u s n o t i o n s .
Our
axioms a r e a n o t h e r approach t o t h e problem o f a s s i g n i n g a d i m e n s i o n t o a mathematical s t r u c t u r e .
A key consequence o f s t a b i l i t y t h e o r y i s t h e a b i l i t y t o
d e f i n e such a s t r u c t u r e t h e o r y f o r a wide c l a s s o f s t r u c t u r e s .
A f t e r d e s c r i b i n g t h e axioms we want t h e n o n - f o r k i n g r e l a t i o n t o obey we g i v e another p r o o f o f t h e e x i s t e n c e o f such a r e l a t i o n .
T h i s p r o o f i s a v a r i - a n t on
t h a t i n L a s c a r - P o i z a t w i t h some o f t h e more c o m p l i c a t e d symmetry arguments s i m p l i f i e d by t h e use o f s t r o n g l y s a t u r a t e d models ( d i s c u s s e d i n 4 ) .
T h i s paper
i s p r i m a r i l y i n t e n d e d t o i n c u l c a t e t h e v i e w o f n o n - f o r k i n g i n d i c a t e d above. v i e w p o i n t i s e x p l o r e d i n g r e a t e r d e t a i l i n Paldwin-1982. i d e a s i n n o v e l ways.
However,
This
we d e v e l o p two
I n s e c t i o n 2 we suggest s y s t e m a t i c usage o f t h e t e r m i n o l o g y :
t h e t y p e p a l m o s t b e a r s a c e r t a i n r e l a t i o n (e.g. b e a r s t h a t r e l a t i o n t o e v e r y model
definable over) t o a set A i f p
M c o n t a i n i n g A.
I n section
4 we e x p l o r e a new
and s i m p l e r d e f i n i t i o n o f S h e l a h ' s concept o f an FaX ( T ) - s a t u r a t e d model. m a t e r i a l h e r e i s a p e r m u t a t i o n o f a subset o f B a l d w i n - 1 9 8 ~ .
The
Details o f proofs
o c c u r o n l y where t h e y a r e new; o t h e r w i s e we r e f e r t o e a r l i e r work.
J.T. BALDWIN
72
Our n o t a t i o n i s t h a t g e n e r a l l y used i n model t h e o r y and p a r t i c u l a r y i n s t a b i l i t y theory. model
C
A l l models and s e t s d i s c u s s e d a r e c o n t a i n e d i n an e x t r e m e l y s a t u r a t e d
( t h e monster model).
types w i t h a circumflex,
e.g.
We c a l l a t y p e o v e r
C
a g l o b a l t y p e and w r i t e g l o b a l
6.
1. ABSTRACT DEPENDENCE RELATIONS
I n t h e 1 9 7 0 ' s Van d e r Waerden
and Whitney
generalized the notions o f linear
independence i n v e c t o r spaces and a l g e b r a i c independence i n a l g e b r a i c a l l y c l o s e d f i e l d s t o o b t a i n t h e n o t i o n o f a dependence system.
VAN DER WAERDEN'S A X I O M S
T h i s r e l a t i o n i s most e a s i l y d e s c r i b e d i n t e r m s o f a
i)a depends on (a).
int
dependin
on a s e t X.
(Reflexivity)
ii)I f a depends on X and X
CY
t h e n a depends on Y.
(Monotonicfty)
iii)I f a depends on X and each x f X depends o n Y t h e n a depends on Y. ( T r a n s i t i v i t y o f dependence) iv)
I f a depends o n X U b u t a does n o t depend on X t h e n b depends
on X U €a).
(Exchange axiom)
v) I f a depends on X t h e n a depends o n a f i n i t e subset o f X.
(Finite
character)
These axioms a r e n o t so c a t e g o r i c a l as we m i g h t n a i v e l y t h i n k .
[ X I t o denote t h e subspace g e n e r a t e d b y X, dim(CA U E l ) = dimCAI f a i l s f o r a l g e b r a i c a l l y closed f i e l d . i m p o r t a n t "hidden axiom":
For example,
using
t h e law i n v e c t o r spaces:
+ dim(B)
Moreover,
- dim(A 1 B) i t t u r n s o u t t h a t t h e r e i s an
i f X i s an independent s e t ( i n a v e c t o r space o r f i e l d )
13
Strong Saturation and the Foundations of Stability Theory then,
up t o isomorphism,
t h i s independence.
t h e r e i s o n l y one way t o add a neu element x p r e s e r v i n g
Finally,
t h e most i m p o r t a n t consequence o f t h e s e axioms,
that
a l l bases f o r a space have t h e same c a r ' d i n a l i t y t u r n s o u t n o t t o r e q u i r e t h e f u l l s t r e n g t h o f axiom iii). We d e s c r i b e here,
n o t a g e n e r a l i z a t i o n o f t h e Van d e r
Waerden axioms as i t i s ueaker i n some senses and s t r o n g e r i n o t h e r s b u t r a t h e r another d e f i n i t i o n o f independence which s p e c i a l i z e s t o t h e v e c t o r space and f i e l d cases.
We d i f f e r from t h e t r a d i t i o n a l a l g e b r a i c approach i n s e v e r a l ways.
The a l g e b r a i c
approach c o n s i d e r s t h e r e l a t i o n : t h e p o i n t a depends o n t h e s e t B.
This r e l a t i o n
can be t r i v i a l l y extended t o : a depends o n E o v e r C. B i f some n o n - t r i v i a l
(e.g.
I n f i e l d s a depends on
p o l y n o m i a l i s s a t i s f i e d b y a and elements o f B;
a depends on
B o v e r C i f some n o n - t r i v i a l p o l y n o m i a l u i t h c o e f f i c i e n t s from C i s s a t i s f i e d b y a and some e l e m e n t s o f 8.) We u o r k d i r e c t l y u i t h a t r i p l e , (and u l t i m a t e l y ) b y a s e t A. B.
a,P,C.
We r e p l a c e t h e element a b y a seguence 3
Further,
we r e p l a c e t h e sequence 3 b y i t s t y p e o v e r
we f o r m u l a t e o u r n o t i o n as a r e l a t i o n between a t y p e p o v e r a s e t B and
Thus,
Since ue a r e u o r k i n g i n t h e monster model u h i c h i s s a t u r a t e d model t h e r e
a s e t A.
i s a c t u a l l y l i t t l e d i f f e r e n c e b e t u e e n d e a l i n g u i t h a complete type, t h e p a i r (F,B).
For,
t(2;B),
and
a l l r e a l i z a t i o n s o f t h e t y p e c a n b e mapped t o c b y an
automorphism u h i c h f i x e s B.
However,
i)
d e a l i n g u i t h t y p e s has t h r e e advantages.
I t a l l o w s us t o d e a l u n i f o r m l y u i t h f i n i t e sequences g i v i n g no p r e f e r e n c e t o
ii) I t a l l o u s us t o compare t u o p a i r s (T;B)
singletons. t(F;B),
t(a;E).
p a i r (F;B)
Thus,
and (a;E)
by considering
iii) I t makes i t c o n v e n i e n t t o speak o f a p p r o x i m a t i o n s t o t h e
b y d i s c u s s i n g s u b t y p e s o f t(F;B).
i n o u r d i s c u s s i o n h e r e ue u i l l f o r m a l i z e t h e n o t i o n
over A b y d e f i n i n g :
t(B;B)
i s f r e e o v e r A.
B does n o t depend on
I n t h e remainder o f t h i s s e c t i o n ue
l i s t t h e axioms ue u a n t t h i s r e l a t i o n t o s a t i s f y .
Much o f t h e m a t e r i a l i n t h e
l a t t e r p a r t o f Shelah-1978 can b e s i m p l i f i e d b y s y s t e m a t i c use o f t h e s e axioms.
B
J.T. BALDWIN
14 1.1 FREE EXTENSIOhlS.
A n o t i o n o f freeness,
1i s
a c o l l e c t i o n o f p a i r s (p,A).
Each p i s a ( n o t n e c e s s a r i l y complete) t y p e which d e t e r m i n e s i t s domain: B. (p,A)
€
I we
w i t h a p o i n t c r e a l i z i n g p and w r i t e t(c;B)
1.2 ISOMORPHISM CONVENTIOF!. isomorphism,
1 A.
say p i s f r e e o v e r A and w r i t e p
i.e.
Alternatively
, we
If
may d e a l
f A.
We r e q u i r e any n o t i o n o f f r e e n e s s t o be c l o s e d under
i f f i s an isomorphism and (p,A)
€
F then
(f(p),f(A))
€
L,
1.3 MONOTONICITY A X I O M S . M.I
M2.
1.4
Ep
If a
If A
C A
1-
and p i s f r e e o v e r A t h e n q fs f r e e o v e r A.
2
and p i s f r e e i s over A1 t h e n p i s f r e e o v e r A
T R A N S I T I V I T Y A X I O M FOP INDEPENDENCE.
Let C
B
C dom
p.
2'
I f p i s f r e e over B
and p J B i s f r e e o v e r C t h e n p i s f r e e o v e r C. Note t h a t t h i s axiom is m i t e d i f f e r e n t from t h e t r a n s i t i v i t y o f dependence e n u c i a t e d as iii)i n t h e Van d e r Uaerden t i s t .
I n particular,
we w i l l e s t a b l i s h
t h i s axiom f o r o u r n o t i o n o f dependence i n t h i s paper w h i l e t h e a t t e m p t t o approximate t r a n s i t i v i t y o f dependence l e a d s t o t h e s u b j e c t o f r e g u l a r t y p e s . However,
i t a150 easy t o o b t a i n t h e analogue o f t h e exchange p r o p e r t y ,
symmetry axiom.
the
We m e n t i o n i t h e r e a l t h o u g h ue do n o t t r e a t i t e x p l i c i t l y l a t e r
i n t h e paper s i n c e s u r p r i s i n g l y i t can be d e r i v e d from t h e c a t e g o r i c i t y o f a s m a l l e r s e t o f axioms ( c f . s e c t i o n 6 ) .
1.5 SYMMETRY A X I O M .
I f t(B;
B U
f6>)
i s f r e e o v e r B t h e n tt6;B
o v e r 8.
1.6
EXISTENCE AND EXTENSION A X I O M S .
El.
t(6;A)
i s f r e e o v e r A f o r any
6 and A.
U fB))
i s free
15
Strong Saturation and the Foundations of Stability Theory
E2.
I f p i s f r e e o v e r A and domp
CB
t h e r e i s a p1 € S(P) which extends
p and such t h a t p1 i s f r e e o v e r A.
As u s u a l we d e n o t e b y S(B)
1.7 DEFINITION.
Let
t h e c o l l e c t i o n o f a l l complete t y p e s o v e r F.
be a n o t i o n o f f r e e n e s s .
A type p i s s t a t i o n a r y over A i f
i f ql, q 2 € S t B ) e x t e n d p and a r e f r e e o v e r A t h e n q1 =
f o r e v e r y 8 c o n t a i n i n g A, 42'
Me say p i s s t a t i o n a r y i f p i s s t a t i o n a r y o v e r domp. The f o l l o w i n g axiom expressed a h i d d e n proper'ty o f v e c t o r spaces; t h e r e i s o n l y one way (up t o isomorphism) t o f r e e l y e x t e n d an independent s e t .
As n o t e d i n 5.14
t h i s axiom imposes a bound on t h e number o f such e x t e n s i o n s and t h i s bound s u f f i c e s f o r o u r purposes.
1.8 BOUNDEDNESS OR NON-TRIVIALITY A X I O M .
Every t y p e o v e r a model i s s t a t i o n a r y .
2. DEFINABILITY AND "ALMOST"
I n t h i s s e c t i o n we d i s c u s s t h e r e l a t i o n between such n o t i o n s as p i s d e f i n a b l e over A and p i s d e f i n a b l e o v e r e v e r y model c o n t a i n i n g A.
We do n o t assume T i s
stable i n t h i s section.
2.1
DEFINITION.
there i s a
i)The t y p e p € S ( B ) i s o v e r A i f f o r each f o r m u l a +(X;6) f4(F;5)
w i t h 5 € A and such t h a t
€ p
4 <-> 3'4.
ii) The t y p e p € S ( B ) i s d e f i n a b l e o v e r A i f f o r each +(F;6) € p t h e r e i s a f o r m u l a d+(g;S)
( o v e r A) such t h a t
+(F;6)
€ p iff
b dbt6;i).
iii) The t y p e p € S ( B )
i s f i n i t e l y s a t i s f i a b l e i n A i f f o r each
i s an 3 € A such t h a t
+(3;6).
+tx;6)
€ p there
J.T. BALDWIN
76
Now we i n t r o d u c e a n a t u r a l a p p r o x i m a t i o n t o each o f t h e s e n o t i o n s . s e c t i o n we f o c u s on almost d e f i n a b i l i t y .
In this
The r o l e o f a l m o s t s a t i s f i a b i l i t y
becomes c l e a r i n s e c t i o n 5 .
2.2
DEFINITION.
p i s over M .
A,
The t y p e p € S t E ) i s almost o v e r A i f f o r each model M c o n t a i n i n g S i m i l a r l y p i s almost d e f i n a b l e over A
o r almost s a t i s f i a b l e i n
A
i f p i s d e f i n a b l e o v e r o r f i n i t e l y s a t i s f i a b l e i n e v e r y model c o n t a i n i n g A.
Note,
t h a t f o r m a l l y p i s a l m o s t o v e r A means t h a t f o r each
M t h e r e i s a f o r m u l a #b(Z;ii)
with
4 <->
f + and t h e
compactness arpument shows t h a t t h e f o r m u l a
++
+b does
+(%;6)
f p and f o r each
may depend on M.
An easy
not depend On M.
We w i l l r e a u i r e t h e f o l l o w i n g n o t i o n l a t e r .
2.3 DEFINITION.
Then p i s s t r o n g l y d e f i n a b l e o v e r B i f
L e t p be a t y p e o v e r A.
t h e r e i s an e x t e n s i o n p ' o f p t o a complete t y p e o v e r A U @ such t h a t p ' i s d e f i n a b l e o v e r 8.
The f o l l o w i n g n o t i o n s a r e from Shelah-1978.
2.4 DEFINITION.
The t y p e p i s d e f i n a b l e almost o v e r A i f f o r each f o r m u l a
t h e r e i s a f o r m u l a d 4 ( 9 ) which i s a l m o s t o v e r A such t h a t f o r any
F b(3i;C)
4(li;9) € p iff
d+(F).
2.5
DEFINITION.
( R € FE(A))
i) The r e l a t i o n R(X;Y)
i s a f i n i t e eauivalence r e l a t i o n over A
i f R i s a n e q u i v a l e n c e r e l a t i o n which h a s o n l y f i n i t e l y many c l a s s e s
and which i s d e f i n a b l e w i t h p a r a m e t e r s f r o m A .
ii)The f o r m u l a + ( X I depends o n t h e f i n i t e e q u i v a l e n c e r e l a t i o n R i f t h e s o l u t i o n set o f
4 is
a u n i o n o f c l a s s e s o f R.
I1
Strong Saturation and the Foundations of Stability Theory Now we can t i e t h e s e i d e a s t o g e t h e r .
2.6 THEOREM.
Let
p
be a g l o b a l t y p e and A a s e t .
The f o l l o u i n g a r e e q u i v a l e n t .
i)p i s d e f i n a b l e almost o v e r A. ii)p i s almost d e f i n a b l e o v e r A.
iii)p has a bounded number o f c o n j u g a t e s o v e r A. ( I f t h e n o t i o n o f c o n j u g a t e i s u n f a m i l i a r see 4 . 3 below.) PROOF.
Clearly,
i)-> ii). I f i f ) h o l d s t h e n f o r any P c o n t a i n i n g A say w i t h
= 4, e v e r y c o n j u g a t e o f p i s d e f i n a b l e o v e r M. conjugates.
Cow,
suppose iii)h o l d s .
number o f c o n j u g a t e s .
Then,
Thus
f o r each
T h i s i m p l i e s as i n Shelah-1978
depends on a f i n i t e e a u i v a l e n c e r e l a t i o n o v e r A. t h e r e a r e lii
< k, e l e m e n t s o f
M such t h a t :
p
has a t most 4
b(F;y),
IAl
IT'
d 4 has a bounded
( c h a p t e r 111) t h a t d 4
Thus f o r any
M containing A
k b ( F ; E ) <-> V<E(F,iii):i
< k>.
3. FREE EXTENSIONS O F TYPES OVER M O D E L S I N STAELE THEORIES.
Now,
a s t a b l e t h e o r y can b e d e f i n e d as one f o r which i f p i s o v e r A t h e n p i s
d e f i n a b l e o v e r A. by H a r n i k .
3.1
We u i l l use a p a r t i c u l a r y s t r o n g form o f t h i s r e s u l t e n u n c i a t e d
H e n c e f o r t h ue d e a l o n l y w i t h s t a b l e t h e o r i e s .
LEMMA ( C H a r n i k - H a r r i n g t o n
19821.
I f T i s stable,
f o r every formula
4
and each
p € S ( A ) t h e r e e x i s t L - f o r m u l a s e(!?;Z) and M 5 ; Z ) such t h a t :
i)
For some 3, e(X;3)
ii)For e v e r y 6 € A,
I=
€ p.
i f e(%;a) € p t h e n
[bt!?;6)
€ p iff
At6; B) 1.
Now we w i l l d e f i n e a f r e e e x t e n s i o n o f a t y p e o v e r a model and show t h a t t y p e s o v e r models a r e s t a t i o n a r y .
We a l s o p r o v i d e some o t h e r i m p o r t a n t
c h a r a c t e r i z a t i o n s o f f r e e e x t e n s i o n s of t y p e s o v e r models.
In the following
s e c t i o n we i n t r o d u c e an i m p o r t a n t concept and t h e n r e t u r n t o t h e t a s k o f d e f i n i n g
J.T. BALDWIN
78 f r e e extensions o f a r b i t r a r y types.
I f p i s a type over a model, M,
then we w i l l choose t h e f r e e extension o f p t o be
one which i s d e f i n a b l e over M .
Such an extension e x i s t s and i n f a c t i t has
several other n i c e p r o p e r t i e s which we record nou.
f o r some
ZI
ii)(THE FUNDAMENTAL ORDER) Then p
,
over A i f
b(ic;5) 8 p.
f A,
2 q,
and q i n S(B).
Let p be i n S(A)
i f every formula which i s represented i n p i s represented i n
-
5
q and q
5
n a t u r a l l y extend t h i s n o t i o n t o p
2
q.
p
i)The formula bCii;l) i s represented i n t h e type
3.2 DEFINITION.
We w r i t e p
q if p
p.
When C q
i s represented i n p i s represented i n o.
domp
fl
domq we can
i f every formula i n L ( C ) )
which
We denote t h e equivalence
c l a s s o f q by Cql.
3.3 D E F I N I T I O N . Let M be a model o f T and an h e i r o f p € S(M) i f q extends p and p coheir o f p on A i f t(A;M
3.4 THEOREM.
M be contained
-,,,
q.
i n A.
Let a r e a l i z e p.
Then q i n S(A)
is
Then t(B;A)
is a
U 3 ) i s an h e i r o f t(3,M).
Let M be a model o f a s t a b l e theory T,
p f S(M)
and M
4 C,
then p has
a unique d i s t i n g u i s h e d extension t o a complete type over C which i s i t s - h e i r , coheir,
and d e f i n a b l e extension.
The proof o f t h i s can be found i n Lascar-Poizat o r i n any o f t h e various p r e p r i n t s o f Baldwin-198x.
4.
STRONGLY SATURATED MODELS
We uant nou t o d e f i n e t h e f r e e extension o f a type p uhose domain may n o t be a
79
Strong Saturation and the Foundations of Stability Theory For t h i s our v a r i o u s n o t i o n s o f a l m o s t w i l l be u s e f u l .
model.
e x p l o r e one f u r t h e r concept.
Eut f i r s t ,
we must
T h i s n o t i o n was i n t r o d u c e d b y Shelah as
F a X - s a t u r a t i o n and p l a y s an i m p o r t a n t r o l e i n t h e c o u n t i n g o f models o f s t a b l e theories.
4.1
However,
The model M i s s t r o n g l y - X - s a t u r a t e d
DEFINITION.
X, any
<
t h i s i s i t s f i r s t use a t such a b a s i c p o i n t i n t h e t h e o r y .
i f f o r each s e t A w i t h ( A ( <
X-type o v e r A which i s f i n i t e l y s a t i s f i e d i n M i s r e a l i z e d i n M.
I f M i s a X - s a t u r a t e d model o f a s t a b l e t h e o r y and X > IT1 t h e n M i s
4.2 LEMMA.
strongly-X-saturated. PROOF.
Suppose
lA( <
such t h a t t(e;A)
X and Let A
< X and
g u a r a n t e e s t h a t t(e;A) L e t d d e f i n e p = t(3;M) o f d.
A
0
4 N 5 M.
< X.
i f fk d+(m,fi)
.
-
L e t e be an element
Ao.
For any f i n i t e sequence 2 from A1
U 2) i n t o an e q u i v a l e n t t y p e q o v e r a subset N
Given t h i s c o n v e r s i o n t h e X - s a t u r a t i o n o f M
i s r e a l i z e d i n M. By 3.1,
o v e r M.
o n l y I T ( p a r a m e t e r s a r e used i n t h e range
L e t N be a submodel o f M w i t h A.
and w i t h I N 1
and A1 = A
i s f i n i t e l y s a t i s f i a b l e i n M.
we w i l l show how t o c o n v e r t t(e;A of M with IN(
M = A.
and a l l t h e s e p a r a m e t e r s c o n t a i n e d i n N
Cow f o r any m € M, 3 E A
- Ao,
6
and
from Ao,
kb(m,Z,fi)
So t h e r e q u i r e d t r a n s l a t i o n t y p e i s Cd+(9;x,p)(x,fi):+(x;a,S)
PTo I. see t h i s t y p e i s c o n s i s t e n t n o t e t h a t f o r any
4
f i n i t e s a t i s f i a b i l i t y t h a t t h e r e i s an e ' E M such t h a t
k4(e,3,6)
I= 4(e',3,6)
€
i m p l i e s by and t h u s
k
d+(e',fi).
4.3 DEFINITION. write p
€ S(M),
PROOF.
CM
and p,a
E S(M).
We say p i s c o n j u g a t e t o q o v e r A and
q i f t h e r e i s an automorphism o f M which f i x e s A and maps p t o q.
4.4 THEOREM. p,q
Let A
Let
IfM i s IMI-saturated,
i f p -A q then p
e
r e a l i z e p and
aA
a
X
>
(TI,
(MI
= X, and ( A 1 < X t h e n f o r any
q.
r e a l i z e q.
Choose f o r each
4,
f o r m u l a s e ( i i ; i ) and
J.T. BALDWIN
80
'(g;r) t
t o d e f i n e t(F;M)
IT1
Cb(X;FI)
U 2 ) = t(B';A
atn;6)
a).
Let
with
a(X;6') Thus a
€ p and
611 f q .
<-> U1';6')1
-Cb(X';B')
CB
and I B I
5 IAI
f o r each e chosen
€ p and f o r every I€ M,
4M
such t h a t
be an automorphism o f M f i x i n g A and t a k i n g E t o
(E
<-> A(iii;a
Now suppose -Cb(F;ii)
q.
M with A
Now b y s t r o n g s a t u r a t i o n choose 8'
p.
U
€ B such t h a t
Now f o r each 0 and each e, A depending on
8'.
'.
<-> u~1;6)1 €
6
4
That is,
c o n t a i n a l l parameters necessary t o d e f i n e p.
according t o 3.1 t h e r e i s a
t(B;A
Let P
i n the manner o f 5.1.
b,
if
etX;6)
Since p * A q,
€ p then
a(R;a
6)
€
t h e r e e x i s t 6 ' and I'
€ p c o n t r a r y t o our choice o f
e and
p = q as r e q u i r e d .
Lemma 4.2 i s i n Shelah-1978 ( u s i n g h i s d e f i n i t i o n ) and Theorem 4.4 i s i m p l i c i t i n Lascar-Poizat
5.
.
FREE EXTENSIONS OF ARBITRARY TYPES
Now we g i v e
our d e f i n i t i o n o f a f r e e e x t e n s i o n o f an a r b i t r a r y type.
5.1
DEFINITION.
M
T with A
Let
4 M,
b
f3
€
S(4)
Then
6
does n o t f o r k over A i f f f o r every
i s d e f i n a b l e over M.
f o r k over A i f f o r some
b f StL)
with p
Let p be a type over 8.
4 6,
f3 does n o t f o r k over
Then p does not A.
Note t h a t t h e f o l l o w i n g two lemmas a r e immediate from t h e d e f i n i t i o n o f d e f i n a b l e almost over and t h e d e f i n i t i o n o f n o n f o r k i n g .
5.2 LEMMA.
For any g l o b a l type
0, b
does n o t f o r k over A i f and o n l y i f
is
s t r o n g l y d e f i n a b l e almost over A.
5.3 LEMMA.
The n o n - f o r k i n g r e l a t i o n s a t i s f i e s t h e f o l l o w i n g axioms. M.I
If
q5
p and p does n o t f o r k over A t h e n q does n o t f o r k over A.
81
Strong Saturation and the Foundations of Stability Theory I f A1
M2.
2 A2
and p does n o t f o r k o v e r A 1 t h e n p does n o t f o r k o v e r
A2'
I f p d.n.f.
E2.
o v e r P and domp
p and such t h a t p,
5
B t h e r e i s a p,
€ S(B)
which e x t e n d s
o v e r A.
d.n.f.
I t i s easy t o e s t a b l i s h t h e f o l l o w i n g r e s u l t b y compactness.
I t corresponds t o
t h e e x i s t e n c e o f a bound i n t h e L a s c a r - P o i z a t t r e a t m e n t .
5.4 THE EXTENSION LEMMA. t h a t f o r every M w i t h A
5.5 THEOREM.
Let p € S(A).
E M,
6
There e x i s t s a
i s the heir o f
p
SCL) w i t h p L b such
€
plM.
Ifp i s a t y p e o v e r A t h e n p does n o t f o r k o v e r A.
Now u s i n g s t r o n g l y s a t u r a t e d models we show t h e s i g n i f i c a n c e o f almost satisfaction.
5.6 LEMMA.
Let A
C@
and p € S ( E 9 ;
p d.n.f.
o v e r A i f f p i s almost s a t i s f i e d i n
A.
PROOF.
I f p d.n.f.
over A then p
5
p f o r some g l o b a l t y p e b which i s d e f i n a b l e
o v e r M f o r each M c o n t a i n i n g A (and t h u s b y 3.4 converse,
f i n i t e l y s a t i s f i e d i n M).
Choose b y t h e E x t e n s i o n Lemma a
suppose p i s almost s a t i s f i e d i n A.
model M which i s I B I + - s a t u r a t e d and such t h a t t(M;B)
d.n.f.
5.7 DEFINITION.
over A.
d.n.f.
f i n i t e l y s a t i s f i e d i n M and t h u s r e a l i z e d i n M b y some F.
-C t(F;B)
For t h e
Then p i s
Then b y m o n o t o n i c i t y , p
o v e r A as r e q u i r e d .
For any A and any p € S(A),
upper bound o f W = CCql: p
5a
t h e bound o f p, p ( p ) i s t h e l e a s t
and q i s a t y p e o v e r a model).
Now we r e g a i n t h e d e f i n i t i o n from L a s c a r - P o i z a t .
We show t h a t f o r k i n g can be
82
J.T. BALDWLN
described i n terms o f t h e bound.
5.8 LEMMA.
5 6,
Let A
p € S(B)
and suppose p extends t o
L e t M be any model c o n t a i n i n g P and plA
f o r k over A.
p
5q
€ SCL) u h i c h does n o t
€ S(M).
&
Then Cql
Cpl M I .
p
Since
N i s s t r o n g l y (PI'-saturated
Choose N such t h a t
PROOF.
d.n.f.
5 l M i s r e a l i z e d i n N by some t.
over A,
be an automorphism o f
E uhich
f i x e s A and maps
extension o f q t o a g l o b a l type. suppose b(X;g)
Since t(P;N)
That is,
4(Y,g) i s
Then,
t o E.
represented i n a
Let
a
Let
4 denote t h e d e f i n a b l e
r e a l i z e a and l e t a
Now
MI.
T h i s i m p l i e s t h a t f o r some
m
€ M,
f o r some iii' € a M, +(t;iii').
4, hence i n 4, hence i n
uith A
over A.
d.n.f.
= (a 6) I (a M) = t(E;a
i s almost s a t i s f i a b l e i n A,
For any A and p € E(B)
5.9 COROLLARY.
a
a t(a;M)
is represented i n [PI.
€ F(A)
+(E;iii).
and t(M;N)
4 P.
q.
I f q = p ( A t h e n p f o r k s over A
i f f D(p) < D(p1A). PROOF.
p and 4 r e a l i z e p(p) and P ( p ( A ) .
Let
ue see Cp1
= C41.
b
and
4
over A t h e n by 5.8
d.n.f.
I f p f o r k s over A f o r every $ extending p t o
c o n t a i n i n g A such t h a t
Cpl < CPIMI.
If
SCL) t h e r e
i s an M
p i s n o t t h e d e f i n a b l e e x t e n s i o n o f P l M so by 3.4
But by C o r o l l a r y 5.8,
CplMl 5 CQI.
We nou can deduce t h e
t r a n s i t i v i t y axiom w i t h o u t e f f o r t .
5.10 COROLLARY. d.n.f.
5.11
Let A
5
over A t h e n p d.n.f.
DEFINITION.
i)
B
4
I f p d.n.f.
C and l e t p € S ( C ) .
over A.
X (T) i s t h e l e a s t i n f i n i t e c a r d i n a l such t h a t f o r each
f i n i t e sequence 3 and ascending sequence o f s e t s
t(Z;Ait,)
over B and p l B
< X
(T)>,
f o r some i
does n o t f o r k over Ai.
i i ) k T ) i s t h e l e a s t i n f i n i t e c a r d i n a l such t h a t f o r every type p t h e r e i s a set A
5 domp
uith
IAl < k T )
and p d.n.f.
over A.
83
Strong Saturation and the Foundations of Stability Theory I t i s r a t h e r easy t o show:
5.12
LEMMA.
k T )
< X
5 k(T)
(T)
Now since p f o r k s over plA, represented i n B(plA);
5.13 THEOREM.
+.
Cb(p)l c o n t a i n s a formula + ( i i ; p ) which i s n o t
thus i t i s immediate t h a t :
For every T,
x
(TI
5
IT!+.
Now ue can e a s i l y conclude t h e f o l l o w i n g r e s u l t .
5.14 LEMMA.
For any t y p e p, p has a t most 2’
contradictory nonforking
extensions. Choose A
PROOF.
-C R.
2 domp
such t h a t p d.n.f.
Now plA has a t most
ZIT’
over A.
Choose M w i t h
IMI 5 IT1
and P
extensions t o t y p e s over models and they are a l l
s t a t i o n a r y so plA has a t most 21T1 p a i r w i s e c o n t r a d i c t o r y n o n f o r k i n g extensions t o g l o b a l types.
@y t r a n s i t i v i t y o f independence and m o n o t o n i c i t y t h i s bound extends
t o p.
6. THE C A T E G O R I C I T Y O F THE DEPENDENCE A X I O M S
Although t h e r e a r e a number o f ways o f d e f i n i n g independence i n s t a b l e t h e o r i e s we can now see t h a t i f they meet a f e w l i m i t e d c r i t e r i a they are e q u i v a l e n t .
THEOREM.
axioms,
Suppose T admits a dependence r e l a t i o n s a t i s f y i n g t h e m o n o t o n i c i t y t h e isomorphism convention,
and each type has o n l y a bounded number o f
p a i r w i s e c o n t r a d i c t o r y f r e e extensions. type
fs i s f r e e over A i f f
PROOF.
Then T i s s t a b l e .
i s d e f i n a b l e almost over A.
L e t A be an a r b i t r a r y subset o f a model o f T. (T)
Moreover,
Then
IS(A))
the global
J.T. BALDWIN
84
i s a t most 2XCr'
x A and t h u s T i s s t a b l e .
For,
each p i n S ( A )
i s f r e e over a
s e t o f power Less t h a n X ( T ) and each t y p e o v e r a s e t o f power X ( T I has a t most A f r e e extensions.
Now suppose t h e g l o b a l t y p e b i s f r e e over A. so
Then a l l c o n j u g a t e s o f b a r e f r e e
b has a t most A c o n j u g a t e s and t h e r e f o r e b y 2.6 i s d e f i n a b l e almost o v e r
A.
REFERENCES
J.
Baldwin,
T.
ms.
C19821
J.T.
Baldwin,
V.
The e l e m e n t a r y t h e o r y o f a b s t r a c t dependence r e l a t i o n s ,
book on s t a b l e t h e o r i e s ,
H a r n i k and L. H a r r i n g t o n ,
vol.23
no.1,(1976),
Shelah,
45 pages i n ms.
"Rank and d e f i n a b i l i t y i n s u p e r s t a b l e t h e o r i e s " ,
D. Lascar and B.
S.
i n preparation.
Fundamentals o f Forking, t o appear Proceedings o f
Model Theory Year i n Jerusalem, D. Lascar,
56 pages i n
I s r a e l J. Math.,
53-87. Poizat,
"An i n t r o d u c t i o n t o f o r k i n g " ,
J.S.L.,
44(1979),
C l a s s i f i c a t i o n Theory and t h e Number o f Non-isomorphic
Models,
330-351. North
Holland(1978). 8. L. Van d e r Waerden, Moderne Algebra,
2nd e d i t i o n ,
Springer-Verlag
(Berlin),
1937. H. Whitney,
57(1935),
On t h e a b s t r a c t p r o p e r t i e s o f l i n e a r dependence,
509-533.
Amer. J . o f Math.,
LOGIC COLLOQUIUM '82 G. Lolli, G. Long0 and A . Marcjg (editors) 0 Elsevier Science Publishers B. V. (North-Hollnnd), 1984
85
UNDECIDABILITY OF RATIONAL FUNCTION FIELDS IN NONZERO CHARACTERISTIC Gregory L. Cherlin* Rutgers University INTRODUCTION The main result to be proved here is the undecidability of the rational function field F(t) over any infinite perfect field F of characteristic p > 0 . Actually I was aiming at something that I would consider a good deal more interesting: the undecidability of fields of formal power series in characteristic p. (I am grateful to Jan Denef for pointing out that my approach was well suited to the case of the rational function field.) The decision problem for power series fields of characteristic p (in the language of fields) is very much open, and nothing in the present paper refutes the possibility that their theories are model complete in a natural language. The plan of the paper is as follows. It is convenient to begin with a discussion of the undecidability of the theory of any infinite field F of nonzero characteristic with respect to certain weak monadic second order logics. 'Ihis is of interest because this theory will be interpretable in the theory of the field F(t) OK F((t)), in (unfortunately) an extended language continuing a predicate for the constant subfield F, as well as the valuation v on F((t)) (or its restriction to F(t)). All of this happens to be quite simple, but a small encoding trick is needed involving the Artin-Schreier polynomial 2 - x . present paper.)
(This is the new point of the
Turning to the case of rational function fields, we can show that the use of the valuation and the predicate for the constant subfield is inessential, though its elimination requires artificial devices. Thus we obtain the main theorem. Naturally we will be unable to resist a concluding discussion of the case of power series fields. For earlier discussion see [ 3 , 5 I . The most clear-cut statement we can make at present is the following. If K = F((t)) is a power series field, let the A-K-E language for K be the language of valued fields enriched by both a predicate for the constant subfield F and the "cross-section" function THEOREM.
x:
2
-+
K defined by
X(n) = tn.
Let F be a perfect field of characteristic p.
Then F((t))
is
undecidable in the A-K-E language. ( ' ) Of course, this formulation of the state of affairs suppresses a lot of relevant information. See SS2.4 for details. I mention some open problems on definability in power series fields in S5. *This report was prepared while the author was visiting Simon Fraser University. The research was supported in part by NSF Grant MCS 8 1 0 2 3 8 3 , and in part by the Alexanderyon-Humboldt Foundation, which funded a visit to the University of Bonn. I am grateful for the hospitality shown by the Seminar for Logic and Foundational Research at Bonn.
G.L. CHERLIN
86
Field theory in weak monadic second order logic
S1.
Notation 1. 2.
F is an infinite field of characteristic p > 0. Q, is the collection of finite subsets of F.
3.
Q+
is the collection of all finite subgroups of the additive group of
For greater clarity we may sometimes write 4.
Tho(F) by Q,
or or
Th+(F) is the theory of Q+ respectively.
QO(F), Q+(F)
F
rather than
Q,,
F. Q+.
in the language of fields extended
More precisely, if we introduce the two-sorted structures: OF = 0
then
Tho(F), Th+(F)
are
Th('F),
Th(+F)
in the usual sense.
We will prove the undecidability of Tho IF) model of arithmetic in a natural way, and I will mention another treatment of
F
Th+(F).
Tho (F)
encodes a
encodes T h O ( F ) . At the end
suggested by Macintyre.
0
The undecidability of If
Th+(F)
and
Th+(F)
Th (F) :
contains a transcendental element u, then the set
u
N
= {l,u,u
2
,...
with the operations u"@un = and urn 43 un = urn provides a natural model of arithmetic, which can be described using Q, and the parameter u. (The introduction of such auxiliary parameters is of course completely harmless.) To define uN we introduce the formula "y Pow x" defined as: or 3 X 6 Q [ I ,f X and x F X and 'da E X (ax f y => ax 6 X I " . y @
Oar
" y = 1 or x This means that
.
We define @ on uN using multiplication in F, and we define is in x from i+) using the set theoretic power of Q,. (One can do the same thing in
any semigroup having an element of infinite order.)
To encode a model of arithmetic in the general case, when F may be algebraic over the prime field, we introduce finite cardinals as equivalence classes of certain sets, and we then derive the arithmetic operations from cardinal arithmetic: Given z € F and X,Y,Z finite subsets of F, let " ( z , Z ) : X + Y" the following relation R(a,b) induces a bijection between X and (1)
a
+
mean that Y:
bz € 2 .
-
Let " X Y" mean that z,Z can be found so that (2.2): X Y. This relation is Q,,-definable, but it is simply the equicardinality relation, for if X,Y are +
finite sets we can find z € F so that all the elements a + bz (a E XI b € y) are distinct, and if X,Y are of the same cardinality we can then select Z satisfying (2.Z): X Y.
-
87
Undecidability of Rational Function Fields in Nonzero Characteristic
Let an F-integer be an equivalence class of finite subsets of F with respect to the equicardinality relation. The ordering on F-integers is induced by settheoretic inclusion (X X' Y) and addition and multiplication may be derived from union and encoded Cartesian product as follows. Let " 2 X + Y" mean that Z is the disjoint union of the sets X ' X and Y ' * Y. This induces the relation C = 5 + q on F-integers. For multiplication, let z: Z XXY mean that Z = {a + bz: a C X, b E Y} and that those elements a + bz are all distinct. Then the relation 32: 2 XxY induces multiplication of F-integers.
-
<
-
-
-
The undecidability of
-
Th+(F)
We will see that the quantifier Q,
can be interpreted in the quantifier Q+,
via the following very natural encoding procedure. Associate to finite additive subgroups A,B set S(A,B,c) given by:
5
(F,
+)
and a parameter c
F
the
{x c A: ( x + c ) - ' C B).
(2)
We need only show that an arbitrary set suitable A,B C Q+(F) Let A
and
be the additive subgroup of
that the elements { (a+c)-': a E A}
QO(F) occurs as S(A,B,c)
for
F generated by
{ (aX1-l : a E S}.
S
and choose c c F
so
are linearly independent over
finitely many elements c will serve). generated by
S €
c E F.
Then
Let
B
F (all but P be the additive subgroup of F
S(A,B,c) =
S.
0
Thus the undecidability of Ih+(F) follows from that of Th (F). Macintyre observes that it is also a consequence of a result of W. Baur I21 and Martyanov [ 1 3 1 , or rather of the proof of that result. They showed that the theory of a vector space (e.g. over Fp) equipped with five distinguished subspaces (as predicates) is undecidable Th+(F) encodes the theory of a finite vector space equipped with n distinguished subspaces (for any fixed n), which for suitable n is also undecidable. The proof of Baur and Martyanov used the word problem for l-generated groups. If we replace this by the word problem for finite semigroups, we get the result in the form needed here. I have recorded this last observation as a point of interest, but we will need the flexibility afforded by the reduction of Q, to Q+ later on.
Other quantifiers Let us define a weak second order of finite subsets of F which is of F is contained in an element Q, and Q+, but it is natural to
quantifier over the set F as a collection Q unbounded in the sense that any finite subset of Q. The only examples needed here will be ask whether every infinite field has an
undecidable theory with respect to any weak second order quantifier. Using an idea of hret [ 7 ] one can show: Proposition 1.
Let
ThQ(F)
F
be a p.a.c.
field which is not separably algebraically closed.
is undecidable for any weak second order quantifier Q on F.
Then
G.L. CHERLIN
88
x
Let Q* be the collection of finite subfields of the algebraic closure F P of the prime field. Then ThQ* (k ) is undecidable. P Nonetheless it seems quite plausible that a weak second Order quantifier with a 2.
-
decidable theory could be constructed on F
P'
though I do not see how to carry
this out. 52. Undecidability of
for
K/F
K =
F(t) or
F( (t)).
is perfect, infinite, and of characteristic p. K will be a field lying between F ( t ) and F ( ( t ) ) . In the present section K will be considered exclusively as a valued field with respect to the valuation v induced by the natural valuation on F((t)). Let K/F denote the structure obtained from K by adjoining a predicate for the constant subfield F of K. We will show that Th(K/F) is undecidable. This suggests that if it should happen that Th(K) is decidable, it is unlikely that this could be proved by a structural analysis of saturated models. F
We proceed by interpreting 'I~I+(F) into Th(K/F), in the terminology of S 1 . This means that we will encode quantification over finite additive subgroups of F into the language of K. We will make use of the function T(X) additive polynomial, and the choice of algebraic considerations. To motivate reader might want to determine the set Associate to x C K the set that the collection:
xp-x. We could actually use any was motivated only by irrelevant the following encoding procedure, the z [Kl for K = F ( (t)) before continuing. =
z
S(x) = {y C F: 32 z ( z ) = xy).
Our claim will be
{S(x): x F K , v(x) < O } is exactly Q + ( F ) .
Hence Q+(F)
Since z ( z +z ) = z ( z l ) 1
2
+
r ( z 2 ) , S(x) will be an additive subgroup of
v(x) < 0, the finiteness of write
u = Cu tn
with
can be encoded into K/F.
u
S(x)
is seen as follows.
If u = z ( z )
F F, then for any negative integer
F.
When
and we
i we have the
relation, whose verification we leave to the reader:
= 0. ipn u = xy, where x is fixed and condition on y: p(y) = 0. To Assuming xi f 0, this relation (If ipn ,t Z
take
u
The sum is finite because
i
is negative.)
If
y F F varies, then ( * ) imposes a polynomial see this raise ( 1 ) to a high power of p. is nontrivial, and hence is satisfied by only
finitely many elements y. We have seen that {S(x): v(x) < 0)
is contained in Q + ( F ) .
For the converse,
if A is a finite additive subgroup of F, we seek an element x of the particular form
89
Undecidability of Rational Function Fields in Nonzero Characteristic
such that S(x) = A. Here n is chosen so that pn According to ( * ) , the elements a F S(x) satisfy: n
is the cardinality of
A.
n-i
1 xiap
= 0.
i=O Conversely any solution of ( 2 ) will lie in S(x). n l/Pi u = 1 uitmpl and C ui = 0 then u C rIF(t)l
(In Other words, if
5
T[KI.)
i=O
Our claim accordingly takes on the following form. = x. € F elements, we seek coefficients c n-i 1
pn
vanishes precisely on A.
...,an, and
al,
A.
...,c
Determine co,
This is easily done.
If A F Q+(F)
Let
a,,
...,a
so that the corresponding polynomial
is not identically zero.
has
so that the polynomial
be a basis for P(a)
vanishes at
(This just requires us to take a
nontrivial solution of a homogeneous linear system.) Now the form of
P
shows that its set of zeroes is an additive subgroup
of F, hence includes A. Since the degree of P is pn, P vanishes only on A. (Notice also for use in S3 that P must have only simple zeroes, so co f 0.) Thus the element of
x
of the form ( 2 ) with
step we used the hypothesis that F
x. = c
~ encodes - ~
A.
In this
is perfect.
S3. Rational function fields
With minor modifications the method of S2 can be used to prove the undecidability of the rational function field F(t) in the pure language of fields. Here F is infinite, perfect, and of characteristic p. Oddly enough the analogous result has not yet been proved in characteristic zero, with F = C providing the outstanding test problem. (R. Robinson managed to prove the undecidability of F(t) for F formally real [ 1 5 1 . ) Of course if F is finite then F(t) is undecidable for much the same reason that Q is (see 1161 for an elegant version of this argument, due originally to, Ershov and Penzin using a method of J. Robinson). This case can certainly not be handled by the method used here. On the other hand one might also want to include imperfect fields, and it appears quite likely that the present approach can be extended to cover this case - but not without considerable effort, involving a m r e precise analysis of the effect
G.L. CHERLIN
90
of the encoding in SZ in the imperfect case, and some study of particular elliptic curves (see below). We may begin the proof of the undecidability of F(t) in the language of fields by a discussion of the role of the valuation and the predicate for the constant subfield in the argument given in S2. With a change of notation, define S(x) for x C F(t) to be the set:
The point established in S2 was that: {S(x) fl F: v(x) < 0) = Q+(F). The role of the valuation is marginal here. It is used to exclude choices of the parameter x which encode infinite sets, but this can be done in other ways. An examination of the end of ~2 shows that the relevant parameters x may be taken to satisfy: 3y Thus
{S(X)
x =
n F:
yp + t-l.
3y x =
+ t-'}
= Q+(F).
The real problem is to eliminate the use of the predicate for F. If F is definable in F(t), as is often the case, there is of course no problem. In general one can define an infinite subset C F in F(t) using a nonrational curve, following an idea used by R. Mbinson. Given such a set C, the arguments of SSl-2 show that finite subsets of C) is interpretable in F(t). a,b,c E F(t), define S(a,b,c) to be the set:
QO(C) (quantificationover
More explicitly, given elements
{x C F(t): x C S(a), (x+c)-l t S(b)}. Evidently
{S(a,b,c)
n
F: a,b,c E Fft), 3x a = xp + t-'} 1
{S(a,b,c) n C: a,b,c C F(t), 3x a =
2
+ t- 1
is QO(F), and hence
is Qo(C). 0
Thus we need only find an F(t)-definable subset C 5 F such that 'Ih (C) is undecidable. For this we will use a nonrational Curve E defined over F, more specifically an elliptic curve with defining equation p(x,y) = 0 with p quadratic in x (further normalization is superfluous). Let E = {(x,y) C E: x,y t F(t)}, which is the same as EF since E is P(t) nonrational. Let C = {y t F(t): 3x (x,y) C EFft)}. If F is quadratically closed then C = F and hence the undecidability of F(t) follows already from SS1-2. If F happens to be algebraic over F , so that in particular F has P at most one quadratic extension F', then F is still definable in F(t), for similar reasons: F = { y C F(t): 3x € F'(t) (x,y) E El. Since F'(t) is interpretable in F(t), our claim follows. Thus we may concern ourselves exclusively with the case in which F contains an element transcendental over F We may then select the curve E so that it P'
91
Undecidability of Rational Function Fields in Nonzero Characteristic contains a point of infinite order over F. Let E = E C F(t) X F(t) equipped Fwith the usual group operation +, definable over F(t). (Take + to be a partial function, since we work only with the affine points on E.) Since E is undecidable, as noted already in S 1 .
has a point of infinite order, T h o ( € )
Of course a slight gap remains to be filled in, since E F (t) Introduce
.
C = { x F F(t): 3y(x,y) C E or Let C' = {x+yz: x,y,z € C}.
is not a subset of
(y,x) F El.
Given X F C'
and
z C C
let E(X.2)
be
{ (x,y) F E: x+yz C X I .
Using
E(x,z)
we can encode QO(E)
in Qo(C').
Hence Th
0
( )
is interpreted
in Th (F(t)) , and the proof is complete. S4.
Power series fields
In characteristic zero most questions about the model theory of power series are settled by the work of Ax-Kochen and Ershov on unramified Hensel fields in characteristic 0 . In particular the theory of F ( (t)) is Turing-equivalent to the theory of F and this is true even in the highly extended language which includes the language of valued fields together with both a predicate for the constant field and the cross-section defined by x (n) = tn (x: Z + F( (t)) "inverts" the valuation). It was known (but more or less forgotten) that this is certainly not the case in characteristic p, and that even power series over a finite field become undecidable if the cross-section is allowed. In 131 this is attributed to Ax (followed independently by 8 . Jacob). A proof is given in [ 3 1 , and since the proof contains the only really convincing evidence that the case of characteristic p is different from the characteristic 0 cases at the level of the language I will give it in detail (or rather a simple variant ~ of fields, which is noticeably more general).
As a preliminary remark, observe that the valuation is already definable in the language of fields. Indeed if p,n are relatively prime then {x: 3y yn = ( 1 + txn)I is just o = {x: V(X) 1. 01. Of course v canimmediately be reconstructed from 0
.
Thus for our purposes the languages of fields and of valued fields are the same language. It seems to me that the problem of decidability in this language of power series fields F((t)) of characteristic p is best thought of (for the present) as containing three separate cases (always assuming F is perfect): 1. 2.
3.
F
F F
infinite, non-kaplansky (see below). infinite, kaplansky. finite.
For the definition of a kaplansky field, recall that an additive polynomial n i P(x) is one of the form 1 ciXp , the usual example being xp-x. I will call a i =O field F a kaplansky field if P[Fl = F for every additive polynomial in FIxl. This is usually called Kaplansky's residual condition - which seems unnecessarily wordy in OUT context - and it has been studied by Delon [61. who showed that
G.L. CHERLIN
92
it is equivalent to the nonexistence of an extension of degree divisible by
p.
Although there is no sound basis for any prediction at the moment, it would not be terribly surprising if F((t)) should be undecidable for all non-kaplansky fields, but decidable for decidable kaplansky fields. (This is idle speculation.) I will prove below: Theorem. Let F be a (perfect) non-kaplansky field. Then F((t)) undecidable in the language of fields with cross-section.
is
Let m e emphasize that I am just reworking an old result. I find the first part of the proof interesting because it does not use the cross-section. Accordingly we will work in the language of (valued) fields until further notice. Notation. If
x = Exiti E F((t)), let
= c [i mod pn]
x.tJ. jzi mod Pn i
x
is definable from x (and the parameter t).
X
Lemma 1. -
Of course the
ti mod pnl
only values of
i
that come into consideration are i = O , l , ...,pn- 1.
Let yi = x [i mod pnl
.
men
, have
yo,...,y
the following two properties:
pn-1
+... + Ypn-l
(1)
x = y + y1
(2)
vi t-iyi c Kpn.
Conversely (1,2) force yi = x Ii mod pnl
.
m e lemma follows easily.
To digress briefly, notice that we have just defined a "pairing function" (actually, a p-tupling function). Hence it is difficult, if not imposible, to conceive of a quantifier-elimination procedure, since there is no such thing as a "single" quantifitr
.
Lemma 2. Let P(x) E F[Xl be an additive polynomial. Then the following predicate Po(y) on F( (t)) is definable in the language of fields, using the parameter
t:
"yo E PIFI" (This means y = E yiti with yi E F, and yo E P[Fl.) Indeed, the definition of
Po(y)
is simply:
93
Undecidability of Rational Function Fields in Nonzero Characteristic where pn is the degree of P. In one direction, if y' € PIF((t))l then clearly y;, C P[Fl, so (*) implies Po(y). In the other direction, if P o ( y )
, and let v(z) = k-p". If v(z) > 0 then easily [O mod pnl z 6 P[F((t))l, so let y' = z in this case. If v(z) '< 0 find z 1 6 P[F((t))]
holds let z = y
v(z-z 1 ) > v(z) and let
with
property Po(z).) treated,
OK
z' =
Replacing z
by
.
(If v(z) = 0, we use the [O mJd Pnl z', we may conclude by the first case
(2-2,)
by induction.
At this point we introduce the cross-section to get the effect of f u l l secondorder quantification over the value group 2 in a non-kaplansky field. For x C F((t)) let Z(x) be the set:
where
is the predicate corresponding to an additive polynomial P with
Po
P[F] # F. This Z ( x ) is an arbitrary subset of on N , the undecidability follows. S5.
N
.
Since we already have
+
A few open problems
Although we speak in term of decidability and undecidability, we are really just studying the structure of the definable subsets of F( (t)) or F(t). The hard questions concern definability in the pure language of fields, but there are also open problems in enriched languages. 5.1
The language of fields
1.
Is the constant subfield definable?
This for me is a key question. My investigations have produced nothing worth going into here. 2.
Is the predicate 'n is a power of p" definable?
This is a predicate on the value group. One can certainly conceive of this happening, although there is no concrete evidence for it. 2A. Is the theory of
F((t)) undecidable in the language of fields enriched by the above predicate?
This is extremely likely. Van den Dries points out that Presburger arithmetic enriched by the "pk"-predicate more to work with. 3.
decidable, but we of course have considerably
Is the predicate "the leading coefficient of x is 1" definable?
This means x = iL&x)xiti field is finite, using
= 1.) This is trivially so if the base with x v (XI = 1 for a F F*. q
G.L. CHERLIN
94 3A.
If F is infinite (and perfect), is F((t)) undecidable in the language of fields extended by the above predicate?
This also seems extremely likely. 4.
Are there any nontrivial definable predicates on than those of the type catalogued in S4?
This, of course, is the real question. its accompanying comment. 5.2
F((t)), other
The reader may formulate question 4A. and
Ehriched languages
Let L(S), L(F) be the languages of valued fields extended by unary predicates S,F respectively, which will be interpreted as follows in power series fields F((t)) (for F perfect, infinite).
s = Itn: n t
x 1
F = constant subfield It turns out that in this context these languages are equivalent - more precisely L(S) has the same O-definable relations as L(F,t) - and strong: they amount to the language of F enriched by quantification over countable subsets. The method used is a somewhat awkward elaboration of the coding in S2. no light on weaker languages.
It casts
5.3 Wel-completeness
Consider the language of fields extended by predicates interpreted in F( (t)) to mean:
where
R
(and x = (*)
R*(x,y,z,
...
)
varies over some extension of the language of fields xiti,...).
The following is conceivable:
If F is model-complete in the extended language then is model-complete.
This is true in any case in characteristic zero, and it does not contradict anything known in characteristic p. Of course such a statement would not be obtainable from a model-theoretic trick. It is essentially a strong statement about algebraic geometry over F( (t)), in a rather vague form.
Undecidability of Rational Function Fields in Nonzero Characteristic
95
References Kochen, S., "Diophantine problems over local fields I, 11," Amer. J. Math. 187 (19651, 605-630 and 631-648. [2] Baur, W., 'Undecidability of the theory of abelian groups with a subgroup,". PAMS 55 (1976). 125-18. [31 Becker, J., Denef, J., Lipschitz , L., "Further remarks on the elementary theory of formal power series rings," in Model Theory of Algebra and Arithmetic, Pacholski et. al. eds., LNM 834, Springer-Verlag, NY 1980, pp. 1-9. [4] Cohen, P., "Decision problems for real and p-adic fields," Comm. Pure Appl. Math. 22 (19691, 131-153. [5] Delon, F., "Hensel fields in equal characteristic p > 0," in Model Theory of Algebra and Arithmetic, Pacholski et. al. eds., LNM 834, Springer-Verlag, NY 1980, pp. 108-116. [6] Delon, F., "Quelques propri6tgs des corps valugs en thgorie des modGles," Thise d'Etat, Univ. Paris 7, 1982. 171 Duret, J.-L., "Les corps faiblement alggbriquement clos non separablement clos out la propriit; d'indipendence,' in Model Theory of Algebra and Arithmetic, Pacholski et. al. eds., LNM 834, Springer-Verlag, NY 1980, pp. 136-157. (81 Ershov, Yu., "On elementary theories of local fields," Alg. Log. 4 (1965), [l] Ax, J.,
5-30. [91 Ershov, Yu., "Cn the elementary theory of maximal normed fields, "Doklady 165 (1965), 1390-1393. [lo] Greenberg, M., "Rational points in henselian discrete valuation rings," Publ. I.H.E.S. 31 (1966), 59-64. [ll] I. Kaplansky, "Maximal fields with valuation I, 11," Duke Math. J. 9 (1942). 303-321 and 12 (1945), 243-248. [12] Kochen, S . , "The model theory of local fields," in Logic Conference, Kiel 1974, LNM 499, Springer-Verlag, NY 1975, pp. 384-425. 1131 Martyanov, V., "The theory of abelian groups with predicates that
distinguish subgroups and with endomorphism operations (Russian)", Alg. Log. 14 (19751, 536-542. [141 P. Ribenboim, Th6orie des valuations, Presses Univ. Montreal, 1964. [15] Robinson, R., "The undecidability of pure trans. ext. of real fields" Zeitschr. f. math. Logik und Grundl. d. Math. 10 (1964), 275-282. [16] Rumely, R., "Undecidability and definability for the theory of global fields," TAMS (1980), 195-217. [17] Schilling, 0 . . The theory of valuations, AMS Math. Survey, 1980. [181 Serre, J.-P., Corps LOcaux, Hermann, [19] M. Ziegler, "Die elementare Theorie der henselchen K&per," Thesis, K6ln 1972.
1. I have since found more precise information, described without proof in S5.2.
LOGIC COLLOQUIUM '82 G. Lolli, G. Long0 and A. Marcia (editors) 0 Elsevier Science Publishers B. V. (North-Holland), 1,984
91
Remarks on T a r s k i ' s problem concerning
Q
, +,
,
*
exp )
Lou van den Dries
S choo 1 of Mathema tics The I n s t i t u t e f o r Advanced Study Princeton, New J e r s e y 08540 U.S.A.
.INTRODUCTION
,+ , * ) s t r u c t u r e @, +, * , exp)
In h i s monograph on t h e elementary theory of t h e s t r u c t u r e T a r s k i asked whether h i s r e s u l t s could be extended t o t h e ([T,
exp(x) = e x
p. 451). ( I n s t e a d of
,
b u t t h i s makes l i t t l e d i f f e r e n c e s i n c e
as t h e unique f u n c t i o n of t h e form
Tarski suggested the f u n c t i o n exp
is d e f i n a b l e i n
x b> f (ax)
t h e axioms mentioned i n [T, p. 57, n o t e 201 f o r from adequate, see e.g.
(R
(R
f (x) = 2'
,+ , * , f)
which is i t s own d e r i v a t i v e ;
Th@
,+ , , f)
are far
[D-W].)
Before we d i s c u s s T a r s k i ' s question, l e t u s b r i e f l y review some a s p e c t s of
his work on
I, +,
(1) D e c i d a b i l i t y of ( 2 ) Tha,
+ , .) =
a)
Th@,
and see what use has been made of i t :
+ , -) ,
t h e o r y of r e a l closed f i e l d s
(3) Elimination of q u a n t i f i e r s f o r
@t
(4) P r o p e r t i e s of d e f i n a b l e s u b s e t s of (5) P r o p e r t i e s of d e f i n a b l e f u n c t i o n s
,
, < , 0 , 1 , + , .) , lRn ,
.
These a s p e c t s are c l o s e l y r e l a t e d i n T a r s k i ' s work, b u t i t makes sense t o d i s c u s s them s e p a r a t e l y .
(1) is a n i c e r e s u l t i n i t s own r i g h t and q u i t e u s e f u l
i n many t h e o r e t i c a l d e c i d a b i l i t y questions, b u t has otherwise n o t been important
i n s e t t l i n g open problems, a s f a r as I know.
(2) i s sometimes u s e f u l i n proving
p r o p e r t i e s of r e a l closed f i e l d s : i n c e r t a i n cases t h e only known proof c o n s i s t s of f i r s t e s t a b l i s h i n g t h e p r o p e r t y f o r t h e f i e l d of reals by transcendental methods and then invoking (2).
(This i s c a l l e d T a r s k i ' s p r i n c i p l e . )
(2) and (3)
combined g i v e a t r i v i a l and improved s o l u t i o n of H i l b e r t ' s 17th problem,
,
L. VAN DEN DRIES
98
and some important g e n e r a l i z a t i o n s , due t o A. Robinson. The c e n t r a l r e s u l t i n T a r s k i ' s work seems t o me (3) a s I hope t o i n d i c a t e in
t h e d i s c u s s i o n of (4) and (5) below.
(Also, (1) and (2) are easy consequences Concerning (4): t h e s i n g l e most
of T a r s k i ' s method of e s t a b l i s h i n g ( 3 ) . )
f r u i t f u l f a c t i s t h e s o - c a l l e d Tarski-Seidenberg theorem: t h e image of a s e m i -
i s a semialgebraic
a l g e b r a i c s u b s e t of lRm under a semialgebraic map lRm ->Elu s u b s e t of
.
Rn
C l e a r l y t h i s i s t h e same a s t h e e x i s t e n c e of a q u a n t i f i e r
elimination f o r the structure than (3). (R
Q
,< ,
+ , *)
( r ) r ER,
which is s l i g h t l y weaker
(Semialgebraic = q u a n t i f i e r f r e e d e f i n a b l e with parameters i n
,< , + ,
.).)Another important p r o p e r t y of semialgebraic sets i s t h a t they
have only f i n i t e l y many connected components, and t h a t each component i s a l s o semialgebraic'
;
see [ K ] f o r a n i c e use of t h i s r e s u l t .
The b a s i c f a c t about (5) i s t h a t a continuous semialgebraic f u n c t i o n Rn ->R
t h i s follows
is bounded i n a b s o l u t e value by a polynomial function:
e a s i l y from ( 3 ) ; an important a p p l i c a t i o n occurs i n [ H 8 r , p. 2761.
A simple proof, due t o
r e s i s t giving one o t h e r b e a u t i f u l a p p l i c a t i o n :
K. McKenna, t h a t t h e inverse of a b i j e c t i v e polynomial map
a polynomial mal.
From complex a n a l y s i s we know t h a t
c l e a r l y the r e a l valued f u n c t i o n
z
+>
I cannot
p
-1
p:En
->
En
is holomorphic, and
i s continuous and semi-
Ip-'(z)l
a l g e b r a i c , hence bounded by a ( r e a l ) polynomial f u n c t i o n ( i d e n t i f y i n g
E2n). Therefore, by L i o u v i l l e , How t o extend a l l t h i s t o
p Q
-1
i s a polynomial map.
, + , , exp
is also
En
with
Q.E.D.
)?
I t seems t o me t h a t concentrating most a t t e n t i o n on t h e analogue of (l), t h a t i s , d e c i d a b i l i t y of the elementary theory, i s a waste of time:
consider
f o r example t h e perplexing problem of deciding t h e statements p(e, e
with
p EZ[X1,X2,X3,
...I
f r e e p a r t of the theory.)
.
e
e
, ee ,
...) = 0
,
(And t h i s i s j u s t a t i n y p a r t of the q u a n t i f i e r
A n a t u r a l 'exponential'
f i e l d ' does n o t seem l i k e l y (but see [vdD
11
analogue of ' r e a l closed
[D-W]),so I d o n ' t expect
This p r o p e r t y i s n o t an obvious consequence of T a r s k i ' s work; see a l s o the end of t h i s Introduction.
Remarks o n Tarski's Problem Concerning (W,f, . , exp)
99
an a t t r a c t i v e analogue of (2) f o r our exponential s t r u c t u r e . More p l a u s i b l e problems a r i s e i n the attempt t o extend (3),
(4) and (5).
To e x p l a i n t h i s l e t us go back t o t h e r e s u l t t h a t each semialgebraic s e t has only f i n i t e l y connected components, each semialgebraic.
This follows from
2 C o l l i n s [C] i n which a new d e c i s i o n method for the r e a l s is constructed, much more time e f f i c i e n t than T a r s k i ' s .
us h e r e :
But t h i s e f f i c i e n c y aspect does n o t concern
we a r e i n t e r e s t e d i n C o l l i n s ' key geometric idea, which he c a l l s
" c y l i n d r i c a l decomposition"; i t i s p a r t l y an a l t e r n a t i v e to, p a r t l y a considerable sharpening of t h e
notion
of q u a n t i f i e r elimination.
what a c y l i n d r i c a l decomposition of a s e t
X cRn
I n (3.6) we s h a l l define
is.
For the moment, we
i s the d i s j o i n t union of f i n i t e l y many
only mention t h a t such a s e t
X
a c e l l being a s u b s e t of IRn
m bmeomorphic t o a space R
,
m
,< n
.
The following c o n s i d e r a t i o n s i n d i c a t e t h a t a l t e r n a t i v e s t o (naive) quant i f i e r e l i m i n a t i o n a r e q u i t e welcome i n the s i t u a t i o n we a r e facing.
FAILURE OF 'NAIVE ' QUANTIFIER ELIMINATION The example below shows t h a t the elementary theory of
QR
,< ,
+ , , exp
(r)r
) does n o t admit e l i m i n a t i o n of q u a n t i f i e r s .
In
f a c t , much more i s t r u e :
Let
Proposition. functions,
Fi:R
"i
->
(Fi)i R
.
I
be any family of
Then the s t r u c t u r e (R,
admits q u a n t i f i e r e l i m i n a t i o n i f and only i f each (Note: Semialgebraic = d e f i n a b l e i n (p
,< ,
Fi
( r ) r ER,
(total) r e a l <,
analytic
(r)r ER,+,
-,
(Fi)i EI)
i s semialgebraic.
+,
)
, by
Tarski.)
The key t o t h i s n e g a t i v e r e s u l t is t h e following example which we s h a l l t r e a t before proving the proposition. Example. (Osgood, see [A, p. 1331)
.
f(x,y) = y.exp(x/y) {(x,y,z)ly in
@,<,
>
Then i t s graph
Define G(f)
O A 3 t ( z = y*exp(t)A t y = x))
* 9
exp
f a XR>
->R
by
i s the s u b s e t
of lR3
which i s obviously d e f i n a b l e
1.
The r e d u l t is a c t u a l l y due t o H. Whitney. See [& p. 1101 f o r an e l e g a n t proof.
L. VAN DEN DRIES
100 Claim. -
There a r e no ( r e a l ) a n a l y t i c functions
an open b a l l i n IR3
,
0
centered a t
such t h a t
boolean a l g e b r a generated by the s u b s e t s Proof. and
The c r u c i a l f a c t s about
i s n o t an "algebraic" function.
f
i = 1,
...,k ,
must vanish on
+0
Fi
nU
f o r each
d.3)
(Fi = 01 of
U
,
i
G(f)
and
nu
,
,
Fi:U ->IR
for all
Pd(x,y,f(x,y))
= 0
A t l e a s t one of the c E G(f)
Fi
nU
,
,
with
would c o n t a i n a whole neighborhood of
..., Pd a homogeneous polynomial of n U . Then f o r a l l 0 < X < 1 we have
+
d
.
XPl(x,y,z)
Take
d
+
with
2 XP2(x,y,z)
Pd
,
(x,y) E R x l ? '
for a l l
# 0
+
,
and
... ,
hence
and we see t h a t f
would be an a l g e b r a i c
o
We model the proof of the p r o p o s i t i o n on the argument j u s t given. f a n x I?->
IR be a n a l y t i c and d e f i n e
f (XI,.
..,xrrcl)
G(f)
i s quantifier free definable i n
= xrrclF(xl/xrrtl,.
.
. , X ~ / X ~ + ~ )
@
.
Let
1R by
If we assume t h a t i t s graph
, < , (r)r ER, + , * ,
P(x,f(x)) = 0
variables with
s u i t a b l e value
X >0
analytic function
for the
FXSZn ->IR
n+lst
,
s a t i s f i e s i d e n t i c a l l y an equation polynomial i n
n+l
variables.
a l g e b r a i c function, hence
.
3 , We assume here t h a t converges on U
U
F
for a l l
x EXn X R ?
0
.
,
(Fi)i EI)
then w e d e r i v e e x a c t l y as before t h a t t h e r e is a nonzero real polynomial n+2
F1
say
and
= Po(x,y,z)
Pd(x,y,z) = 0
->
,
+ P1 + P2 +
(x,y,z) E G(f)
Let
0 = F1(Xx,Xy,Xz)
F SZn
.
f(Xx,Xy) = M(x,y)(X>O,y>O)
(otherwise t h e r e would be
F1 = P 0
(Xx,Xy,XZ) E G(f)
function.
U
belongs t o t h e
Suppose t h e f u n c t i o n s
i s i d e n t i c a l l y zero.
G(f)
3 c E R ). Write degree
are:
nU
,
have the property we want t o r e f u t e ; we may of course assume
t h a t none of t h e
Fi(c)
f
G(f)
> 01 ,
{Fi
...,Fk :U ->a
Fl,
P
in
Substituting a
v a r i a b l e , we derive from t h i s t h a t the
given by (x,
,...,xn> &>
Q(x F (x)) = 0
' A
,
x Enn
xF(xl/X
,
The l e m below shows t h a t then
Q
,...,
xn / A )
a nonzero r e a l FX i s a semi-
i s a l s o semialgebraic.
i s taken so s m a l l t h a t t h i s Taylor s e r i e s of
F1
Remarks on Tarski's Problem Concerning (lR,+, . , exp)
kuima.
If a continuous function g d ->IR
equation Q(x,g(x))
'= 0
,x
101
satisfies identically an
E Rn , where Q is a nonzero real polynomial in
n+l variables, then g i s semialgebraic. From the results on pp. 106-110 of [El, it follows that Rn can
Proof.
be partitioned into semialgebraic subsets Ao,A1, connected such that if x E .A 1
-< i 5 m
and x E Ai
functions pil(x) < Pij:Ai ->lR
,
,
then Q(x,Y)
...,Am
of x
... Am
vanishes identically, while if
the real roots of Q(x,Y)
... < Pik(i)(x)
with A1,
.
are given by continuous
(Obviously, these functions
are semialgebraic.) By continuity and connectedness, g must
coincide on each Ai (i
> 0) with one of the functions Pij
,
hence g
... U Am is semialgebraic. Since every p .A is the limit of points in A1 u ... U Am , the value of g at p is semialgebraically determined by its values on A1 u ... u Am . It follows that g is semirestricted to A1
algebraic.
u
0
The proposition forces us to look for new ways of solving (the realistic part of) Tarski's problem. A line of attack which seems quite promising to Define a k-manifold, k EN
En , n 2 k
,
to be a real analytic submanifold M of some k equipped with an analytic isomorphism h : M " - I R We introduce
.
an IR-algebra
classes ~
goes roughly as follows:
,
of k-manifolds, and for each manifold
for each k a class Jf(k)
M E &@k)
me
(
strutted in stages.
~
&(M)
of
(real) analytic functions on M'.
for1 each M (N.B.,
introduce a l l of M ( M )
,
The
the algebra d ( M ,) are con-
as soon as we have M
,
we do not necessarily
at once.)
At stage 0 we introduce all semialgebraic k-manifolds and all semialgebraic (analytic) functions on them.
To obtain new manifolds, and functions
on them, as well as new functions on the manifolds already available, we use several constructions of which the following three are the most important.
Let
M
9 k-manifold already available.
102
L. VAN DEN DRIES
rf
(i)
are a l r e a d y a v a i l a b l e , then we introduce &he
f,g:M ->R
k-mani f o 1d graph(f)
t h e k+l-manifold
def (-m,f)M ,
(f,g)M
the k+l-manifolds If
(ii) f
0
fo,
...,f d :M ->R
+ fl Y +...+ ,
graph(f)
d fdY
(f,g)M
,
def
(x,f(x)) : x E MI
{(x,y) : f ( x ) (f,m)M
,
<
&
y
<
,
(if
g(x)]
f
<
g
fl M
),
&
defined s i m i l a r l y .
(-m,m)M,
a r e a v a i l a b l e , then w e introduce t h e f u n c t i o n
0" M XR (-m,f)M
,
,
=
(f,m)M
and a l s o i t s r e s t r i c t i o n s t o
(where
a r e as i n (i)), as w e l l
f,g
a s a l l those a n a l y t i c f u n c t i o n s which a r e a l g e b r a i c over t h e 'polynomial' functions so obtained. (iii) $:M
X1,
.
=IR k
...,\
be t h e coordinate functions on
M
defined by t h e c h a r t
Then we introduce as new functions on
M
those a n a l y t i c
f:M ->R
functions
which s a t i s f y a system of d i f f e r e n t i a l equations -a=f
where each
Fi
Fi(X1,.
..,%,f) ,
,...,k
i = 1
is a f u n c t i o n a l r e a d y a v a i l a b l e , e.g.,
,
v i a constructions ( i )
and ( i i ) . Now t h e d e l i c a t e p a r t is t o show t h a t t h e z e r o s e t of each f u n c t i o n i n
,
i d ( M )
,
M
as w e l l a s i t s complement, i s a d i s j o i n t union of
.
U ./fci) This can only be proved i f the coni l k s t r u c t i o n i s done i n e x a c t l y the r i g h t o r d e r ( t h e ' s t a g e s ' have t o be indexed
f i n i t e l y many manifolds i n
by a s u i t a b l y chosen w e l l o r d e r i n g ) , and i f t h e c o r r e c t induction hypotheses are selected.
(The i n d u c t i o n hypothesis on the "asymptotic behavior of
z e r o s e t s " seems t o be t h e c r u c i a l problem, b u t I have an i d e a as t o what i t should be.
Since p r e c i s e statements are complicated and I expect t o come back
t o i t l a t e r , I ' l l leave i t a t these vague i n d i c a t i o n s . ) Remarks
(1) All f u n c t i o n s b u i l t up from real constants, v a r i a b l e s , and t h e o p e r a t i o n s
+ , * , and
exp
a r e e v e n t u a l l y introduced v i a c o n s t r u c t i o n ( i i i ) , b u t i t i s
Remarks on Tarski's Problem Concerning
(R,+, , exp)
103
'
i n t e r e s t i n g t h a t i t i s the system of d i f f e r e n t i a l equations which is c r u c i a l t o prove t h e r e q u i r e d r e s u l t s f o r t h e s e functions (and t h e i r z e r o s e t s ) , and n o t the d e f i n i n g expression i n terms of
+,
*
, and
exp
.4 )
I f t h e program w e just sketched works out, then one o b t a i n s a Tarski-
(2)
Seidenberg theorem f o r a very l a r g e c l a s s of sets and functions, including a l l those d e f i n a b l e i n o u r exponential s t r u c t u r e , as w e l l as the r e s u l t t h a t each of
these s e t s i s a d i s j o i n t union of f i n i t e l y many k-manifolds.
Quite a p a r t from
i t s i n t e r e s t f o r T a r s k i ' s problem, i t would be very d e s i r a b l e to o b t a i n r e s u l t s
i n t h i s d i r e c t i o n , f o r example, i n connection with H i l b e r t ' s 1 6 t h problem, see [Hovl, [ P I . References [vdD2] and [Hov] g i v e p a r t i a l r e s u l t s i n t h e s p i r i t of t h e pro-
(3)
gram sketched above, and l e d m e t o t h e p r e s e n t formulation. I n t h i s paper w e s h a l l c a r r y o u t a more r e l a x e d i n v e s t i g a t i o n which has t h e advantage of introducing the n o t i o n of c y l i n d r i c a l decomposition i n the most
9 be
n a t u r a l and p a i n l e s s way:
of
lRm
9.
any expansion of
.
@I,< A s) ubset
i s c a l l e d d e f i n a b l e i f it i s d e f i n a b l e by a formula i n the language of
- t o be introduced i n t h e @ , < , + , ') - we s h a l l
Under a convenient assumption on T h ( 9 )
! next s e c t i o n , and obviously s a t i s f i e d i f &= inter alia: (*)
is piecewise continuous, c . f .
Each d e f i n a b l e f u n c t i o n IR ->W
(**) Each d e f i n a b l e s u b s e t of IRm
(2.2).
is the d i s j o i n t union of f i n i t e l y many
c e l l s , each of which i s a l s o d e f i n a b l e . See (3.11).
(Cell = space
k
homeomorphic t o an R .) (It is of some i n t e r e s t t h a t t h i s g i v e s a proof, based on q u i t e g e n e r a l
p r i n c i p l e s , of Whitney's " f i n i t e number of components" r e s u l t on semialgebraic
sets.)
I f we assume moreover that
2 is
an expansion of
(R
, < , +) ,
4 , The treatment i n [vdDL] could have been s i m p l i f i e d and generalized cons i d e r a b l y i f t h i s p o i n t had been c l e a r e r t o me.
then
L. VAN DEN DRIES
104
(*) and (**) can be strengthened i n the sense t h a t i n (*) the function
,
f o r each
n
piecewise
c e l l s as
Cn-submanifolds of
,
Cn
and i n (*),
.
lRm
one can take, given any
n
f
is,
,
the
In an appendix we s h a l l indicate the proof
of t h i s and make some remarks on the connection with the important notion of Whitney s t r a t i f i c a t i o n .
GENERAL CONVENTIONS
d= (A, ...)
In t h i s a r t i c l e we consider L-structures a f i r s t - o r d e r language with equality.
L
is said to define the set
{al,.
..,an)
An L-fomula
1 &F
E An
$
...,vn,vn+,,)
= $(vl,
1 &p
{a E A
i s given and
x
X C Am
,
Y
C
...,xm, yl,. ..,Y,)
A function
f:X
->
E
1
If
denotes the s e t
is called
is
X
if
i n that their
Y) of
.
A*
(A-)definable i f i t s graph
(A-)definable.
vary over elements of R = {0,1,2,,
the number of elements of
A"
in the usual way with the subset
..,xm) E X, (yl,. ..,Y),
(xl,.
A1-
A(X C Am)
k, a,m,n
We l e t
X
is f i n i t e ;
..]
otherwise
-.
and w r i t e
#X =
#X
for
Theories of f i n i t e type
(1.1) L e t
(A,<)
be a nonempty dense l i n e a r order without endpoints.
-- ,
notational purposes we adjoin two 'endpoints'
-- < a < + = subset that
#(x,A)
2
we allow parameters.
we i d e n t i f y X X Y
An
G(f) = {(x,f(x)) Ix E X I C Awl
51.
then
.
An
generalize definable subsets of
An
defining formula a r e LA-fomulas, i.e.,
{(xl,
,
An
..,an) ]
, ( i n d) , n
$(x,a)] ; these s e t s a r e s p e c i a l cases of A-definable s e t s :
A-definable subsets of
If
E
@,
..,vn),
$(vl,.
$(al,.
and a s e t of t h i s form i s c a l l e d a definable subset of
s= ...) ,
and
a
for a l l
(a,b)
&&!
and
b
a
EA
.
Let
(r E Ala < r < b3 of
2=
A
A ,
u
+m
.
E-m,+=J
where
--
a r e uniquely detemined by the interval.
c l e a r l y i n f i n i t e and
(A,<)-def inable.
,
For
and p u t ~n i n t e r v a l i s a
<+m
.
An i n t e r v a l i s
N ote
Remarks on Tarski‘s Problem Concerning (R, +, . , exp)
A subset
(1.2) Definition.
of
X
105
i s c a l l e d of f i n i t e type i f
A
is the
X
union of a f i n i t e s e t and f i n i t e l y many i n t e r v a l s . Example: X
If
(A,<)
then a s e t
= @,<)
has only f i n i t e l y many connected components.
(1.3) D e f i n i t i o n . bd(X) = {a
E
X
For
C
we d e f i n e t h e boundarx of
A
Note t h a t f o r
(A,<) = @,<)
m. (1)
bde)
(2)
if a
0
-OD
,
... < am .
awl
Let
...,am] ,
=
X
i = 1,
..,m
...,m I
A
m is
{-1, + l j
al
<
... < am
each i n t e r v a l
,
X
then, p u t t i n g
,
(ai,ai+l)
N
A
be of f i n i t e type and
m of
i = 0
,...,m ,
. {al,
...,am]
i t s boundary,
i s t h e sequence 7 = < T1,T2,
X
X
...,
T
:
+1 i f
(ai,ai+l)
C
X
-1 i f
(ai,ai+l)
C
A\X
:
ai E X
+1 i f
a f i n i t e sequence
.
where
o r d i s j o i n t from
X
C
Then t h e
i = 0,.
where f o r each
and f o r each
Then:
L e f t t o the reader.
(1.5) D e f i n i t i o n .
<
as
.
i s o f f i n i t e type.
e i t h e r i s p a r t of Proof. -
X
is f i n i t e ;
bd(X) = (al, =
.
XI
t h i s gives t h e u s u a l n o t i o n of boundary of
X C A
Suppose
t o be t h e set
X
c o n t a i n s both a p o i n t
and a p o i n t o u t s i d e
a s u b s e t of t h e t o p o l o g i c a l space IR (1.4)
a
Alevery i n t e r v a l containing
from X
al
i s of f i n i t e type i f and only i f
X CR
7=
<
T1,.
..,72 ,
i s c a l l e d t h e l e n g t h of
7
.
a l l whose terms
‘Ti are i n
~
~
~
L. VAN DEN DRIES
106
(1.6) In the remainder of t h i s section we assume t h a t the language a binary predicate symbol
<
J?f= (A,<,
and t h a t the L-structure
expansion of the dense l i n e a r ordering without endpoints Definition.
We c a l l an L-theory
T
A-definable subset of
d= (A,<, ...)
We c a l l
A
of f i n i t e type i f
(This condition implies t h a t each
contains
...) a
.
(A,<)
(extending the theory of dense l i n e a r
order without endpoints) of f i n i t e type i f each model property t h a t each
L
d= (A,<, ...)
has the
i s of f i n i t e type.
m(&)
is of f i n i t e type.
A-definable subset of
i s of f i n i t e
A
type, but the converse implication i s probably not valid.) (1.7) (bd
=.
For each
$) (vl,
... , v ~ + ~such )
Proof.
(1.8)
for a l l
that for a l l
&f= (A,< ,...)
If
then there i s
a
...,v,+~) E
there is a formula
An :
Proof. 3 vn+l $(vl,
4
With respect t o
...,vn+,)
i s of f i n i t e type and
k = k EN
.
a E A”
m
3 x 3y(x
$(vl,
Clear.
-.
L-formula,
L-formula
such t h a t
,...,v ~ + ~i s) an { O , l , ...,k 3 U {=I
$(vl
# $(a,A) E
the ‘ i n f i n i t a r y formula’
T h ( d )
i s equivalent t o the formula
A V V ~ + ~<( X vn+l
< y ->
$(v,,
we can use a simple ...,v ~ + ~ ) ) ) and ,
model theoretic compactness argument t o e s t a b l i s h the existence of a number
k
as required. (1.9) Proposition. an L-formula. subsets
X1,
each
and
i
Proof. #(bd $(x,A))
Let
d= (A,< ,...)
be of f i n i t e type and
Then there i s a f i n i t e p a r t i t i o n of
...,)(M, x
E
and there a r e d i s t i n c t types Xi
the s e t
There is
-< k ,
k EN
#(x,A)
i s of type
such t h a t f o r each
by (1.4), (1.7) and (1.8).
An
#(vl
,...,v ~ + ~ )
i n t o d i s j o i n t definable
T(l), T(i) x E An
...,
T(M)
such t h a t f o r
.
we have
Now each s e t
$(x,A)
i s of
Remarks on Tarski's Problem Concerning (IR,f , . , exp) f i n i t e type, and t h e type of
$(x,A)
5
i s moreover of length
107
f i n i t e l y many types a c t u a l l y occur among t h e types of t h e sets Let
T(l),,..,T(M)
f o r which
be these types and d e f i n e
$(x,A)
has type
.
T(i)
,
2k+l
so only
,
$(x,A)
as t h e s e t of a l l
Xi
x
E
x
E
An.
An
0
Remarks (1)
E
x
For
let
Xi
T(i)
have length
bd b(x,A) = ( f i , l ( x ) , . . . , f i , m ( i ) ( x ) ~ the functions
fi,
:xi ->
+ 1,
2m(i)
where
<...
fi,l(x)
Then
d.
are d e f i n a b l e i n
A
and l e t
(2)
The p r o p o s i t i o n w i l l be used l a t e r i n an i n d u c t i v e argument (see (3.10).
02.
Piecewise c o n t i n u i t y of d e f i n a b l e f u n c t i o n s
9i s
Throughout t h i s s e c t i o n
t h a t each IR-definable s u b s e t of
R
an expansion of
@,<)
has only f i n i t e l y many connected components.
9 i s of
( This assumption seems t o be somewhat weaker than:
m.
(2.1) f
under
n
f
,
f-'(b)
.
E
b
Then
is
f(J)
,
so t h e r e i s
J
i s c o n s t a n t on
f
f i n i t e type.) IR-definable, then
.
I
J
of
whose image
I
i s the union of the f i n i t e l y many s e t s
J
IR-definable s e t
Then
f : I ->R
Assume f i r s t t h a t there i s a s u b i n t e r v a l
is f i n i t e .
hence t h e J
i s an i n t e r v a l and
i s continuous i n a t l e a s t one p o i n t of Proof, -
J
I cIR
If
with the property
b
such t h a t
n f-'(b) J'
,
J
n f-'(b)
is infinite,
must c o n t a i n a s u b i n t e r v a l
hence
f
J'
of
i s continuous a t each p o i n t of
I n the remainder of t h i s proof we s h a l l assume t h a t each s u b i n t e r v a l of has i n f i n i t e image under of closed segments f[an,bn]
f
.
[an,bn] c I
with
0
< bn-an <
l/n
,n2
Jn of length
1
,
<
l/n
such t h a t
.
[al,bl]
and
length
<1
J1
n [an,bn] n= 1 we take an i n t e r v a l of
i s continuous a t the unique p o i n t i n a r e obtained a s follows:
contained i n the
for
J1
IR-definable i n f i n i t e s e t
i s chosen a s a closed segment contained i n the
It is
.
rn
f
I
I n d u c t i v e l y we s h a l l f i n d a descending sequence
i s contained i n an (open) i n t e r v a l
c l e a r t h a t then
J'.
f(1)
,
and
[al,bl]
IR-definable i n f i n i t e s e t
L. VAN DEN DRIES
108
,
f-l(J,)
Jnfl C f [ a n , b n ]
< b&l-an+l
0
(2.2)
Corollary.
< If
,
obviously
Proof.
.
I
f[a,b]
->
(a,b)
on which
g
,
al
<
a t which
a
,
then
... < a m ,
a o = a ,
.
(ai,ai+l)
f
f
i s c o n s t a n t o r s t r i c t l y monotone.
a
,
and assume t h a t
f
,
c
by:
[c,d]
E
g ( y ) = minCx
<
... < a
of
If
,
e i t h e r c o n s t a n t on (ai,ai+l)
Proof.
.
,
d
,
[c,d]
so
g
->R
f:(a,b)
a. = a (ai,ai+l)
I n particular
,
a
m
i s continuous on
[c',d']
R-definable,
,
[c',d']
under
0
then t h e r e a r e
such t h a t f o r each
l i m f(x) x1a
and
(a,b)
exist i n
l i m f(x) xt b
i
<m
is
f
cannot c o n t a i n a whole s u b i n t e r v a l o f
,...,amml) ,
4
.
which have no neighborhood on which
i s c o n s t a n t , o r continuous and s t r i c t l y monotone, i s
{al
By
o r continuous and s t r i c t l y monotone on
The s e t of p o i n t s i n
must be a f i n i t e s e t
.
is s t r i c t l y monotone on
is = b
and we
[ a , b ] : f ( x ) = y)
i s s t r i c t l y monotone on t h e image of
is
f
is n o t c o n s t a n t
f ( g y ) = y ), and t h i s image c o n t a i n s an i n t e r v a l .
(since
(a,b)
IR - d e f i n a b l e then
is
i s i n j e c t i v e , and by t h e p r e v i o u s c o r o l l a r y g [c',d']
f
i s n o t continuous i s
f
f : I ->R
c o n t a i n s a segment
[a,b]
(2.4) Corollary.
f
EI
Then
and t h e r e f o r e
1<
<
a.
i s a n i n t e r v a l and
I CIR
a,b
a subsegment
aO < a
R -definable,
is
Hence, i t must be a f i n i t e s e t . 0
Take
g:[c,d]
definition
g
R
By the previous c o r o l l a r y we may as w e l l assume t h a t
continuous.
define
(Jn+l)
a - d e f i n a b l e , b u t i t cannot c o n t a i n a whole s u b i n t e r v a l of
e. If
[a,b]
-1
[an+l,bn+l]Cf
i s continuous on each s u b i n t e r v a l
t h e r e i s a s u b i n t e r v a l of
on
we choose a n i n t e r v a l
then we choose
that is, there are
f
by t h e p r e v i o u s lemma.
(2.3)
->
The s e t of p o i n t s i n
Proof.
;
l/n+l
f:(a,b)
such t h a t
[an,b ]
Given
.
l/n+l
i s piecewise continuous,
a = b m
<
of l e n g t h
with
.
< bl-al < 1
0
with
(a,b)
,
with
a = a0 < a l <
by t h e p r e v i o u s lemma.
It i s e a s y t o s e e t h a t on each s u b i n t e r v a l (ai,ai+l) c o n s t a n t , o r continuous and s t r i c t l y monotone.
a-definable,
...
the function
f
and So i t
= b
.
is either
Remarks on Tarski's Problem Concerning (lR,+, . , exp) 13.
Cylindrical
Decomposition
-
(3.1) This s e c t i o n c o n t a i n s the main r e s u l t , Theorem (3.7). Recall t h a t W = R
ventions and d e f i n i t i o n s . a l l an e x p a n s i o n 9
u
(3.2) Given a nonempty d e f i n a b l e s u b s e t
of
X
the s e t of d e f i n a b l e continuous f u n c t i o n s
f(x)
<
g(x)
.
(--,m]
for a l l
For
E
x
f,g
and
-m
G(f) = { ( x , f ( x ) ) : x E X ] CRW1
we write
(f,g)X
I t i s h e l p f u l t o view
parametrized by
X
(f,g)X
.
i s a l s o homeomorphic t o IR
,
and
(3.3) By i n d u c t i o n w e d e f i n e f o r each (il,
...,i n )
X .)
namely n = 1: F(0) F(1)
no= (01
{a] with
of
E F(il, F(0)
F(1) .)
,
f
F(il,
and
f 6 C(X) and i f and
we l e t
E
f,g
,
t(X)
f(x)
,
then
G(f)
.
...,in) of
nonempty de-
F 0 has onl'y one member,
a ER
W
definable i n
9.
whose ( f i n i t e ) endpoints
G(f)
with
F(il,. f E C(X)
.. , i n ) ,
...,i n , l ) t o c o n s i s t of a l l sets ...,i n ) , f , g E 8(X) , f < g . (Note: and
if
9.
More generally, given the c o l l e c t i o n
X
f )
and each f i n i t e sequence
0
0
c o n s i s t s of a l l i n t e r v a l s of
F(il,
>
g
k+l ( f , g ) X i s homeomorphic t o R
2
be
.
c o n s i s t s of a l l
are definable i n
define
C(X)
.
f
c o n s i s t s of a l l graphs
(or
k X i s homeomorphic t o R
there is o n l y t h e empty sequence
n = 0:
we l e t
and we p u t
If
((x,y)lx E X
n
refers
are i d e n t i f i e d with the
m
of zeros and ones a c o l l e c t i o n
f i n a b l e s u b s e t s of
, f
,
20
m
as t h e family of i n t e r v a l s
Note t h a t i f k
,
Wm
be t h e graph of
f o r t h e set
Rm and Rn
.9.
E ?(X) we w r i t e
X . mere
(So the
The term ' d e f i n a b l e ' a s
f:X ->W
corresponding c o n s t a n t f u n c t i o n s defined on
f
(1.6).
and functions between s u b s e t s of
33"
t o d e f i n a b i l i t y by an L - f o m l a i n the s t r u c t u r e
u
We f i x once and f o r
which i s of f i n i t e type, c.f.
@,<)
applied t o s u b s e t s of
F i r s t some con-
.
(-m,m)
r e s u l t s of Sections 1 and 2 a r e a v a i l a b l e t o us.)
?(X) = C(X)
109
X
we define
E
F(il,
(f,g)X
F(il,.
...,i n )
,
.. , i n , O ) and we
where
t h i s g e n e r a l i z e s the d e f i n i t i o n
to
L.VAN DEN DRIES
110 Clearly, disjoint.
k = il
where only i f let
Also, i f
+
X
,
n
f o r given
E
.
E
X
F(n)
.
.,in) a r e mutually
k i s homeomorphic t o R
X CRn
dim(X) = k
collections
2"
a c o l l e c t i o n of d e f i n a b l e s u b s e t s of
(3.4) For
then
.
F(il,.
dim(X) = n
Note:
such t h a t
Rn
n .)
...,i n )
,
We
F ( ~ ) is
so
.
dim(X)
F(il,
,
i f and
( A l l t h i s i s i m e d i a t e by induction on
be the union of the
F(n)
collections
,
We p u t
i s open i n Rn
X
.., i n )
F(il,.
... + in .
2"
the
we s h a l l d e f i n e a d e f i n a b l e homeo-
morphism %:X
If
E
.
where
,
< n-1
dim@) = dim(hX)
h(X)
h(X)
By induction t h i s shows t h a t i f
In c a s e Y = h(X)
i s a graph
X
X = (f,g)y
Suppose
G(f)
$:x ->
and l e t
implies
where
Y
where
E
,
< n-1
dim(Y)
be t h e s e t
(3.5) For each s e t X p a r t i t i o n s of Dec(X)
X
n = 0
we have
Let
n
20
X E F(n)
T:G(f)
X
,
C(Y)
Y
X
F(n-l)
(y,r)
,
we put
+>
y
f , g E e(Y)
, gh;' . Clearly %
E ?(hY)
. . Then
,
and we
l i f t s t o a homeo-
.
we s h a l l d e f i n e a c o l l e c t i o n F(n)
.
DeC(X)
of f i n i t e
We c a l l a member of
. 0
X =W is
.
%'
We g e t
$(y),r)
F(n)
f
i s definably
so by r e c u r s i o n we may assume t h a t
, gh;l)h(y) +>
,
F(n-l)
E
X
{XI
,
since ;
F(O) =
we let
1x3
ho),
and the only
be t h e only member of
.
Dec(X)
X
E
i n t o s e t s a l s o belonging t o
( f i n i t e ) p a r t i t i o n of
of
in
a decomposition of
For
Let
(%'
hx:X Z h(X) : ( y , r )
morphism
f
then
FOCI
Y be t h e p r o j e c t i o n map
Y
h(X)
,
dim X = k
belonging t o
h :Y Z h(Y) has a l r e a d y been defined. let
.
F(n'l)
then one can r e p e a t t h e same c o n s t r u c t i o n with
homeomorphic t o an open s u b s e t of JRk
dim(X)
E
h(X)
and suppose and
f E C(X)
l i f t s t o one of
->
X
G(f)
Dec(X)
. ,
has been defined f o r a l l
Then each f i n i t e p a r t i t i o n -1 namely (TT-'(A1), ...,Tr (Ad)]
i s the p r o j e c t i o n on t h e f i r s t
n
coordinates.
X
E
F(n)
g ={A1, ,
. ...,Ad]
where
Let us c a l l t h i s
Remarks o n Tarski's Problem Concerning (R, +, . , exp)
'lifted' partition
gf.
Then we define Dec(G(f)) for f,g E ?(X)
The definition of Dec((f,g)X) complicated. Let again
9 E DecO() .
Then
=
B={Al,..., Ad
111
(gf 1 9E Dec(X)3
,
f
4 g
,
.
is a bit more
be a decomposition of X
,
i.e.,
9 induces a partition
E(f(Al,g(A
,...,(flAd,gIA )Ad 1
A1
of
(f,g)x
,
which we denote by
A second kind of partition of
,...,fm E c"(x)
fo,fl
partition of
with
(f,g)X
and the sets G(fi)
gfTg . (f,g)X
f = fo < fl <
is obtained as follows: let
... 4 fm = g
consisting of the sets
,
0< i< m
.
on X
(fi,fi+l)X
,
.
'ken we have the
i = O,..,,m-1
We denote this partition by
and call it a proper decomposition of
.
(f,g)X
,
(foIfl~...Ifm)X
("Proper" because it is not
induced by a decomposition of X .)
g =(A1, ...,Ad] E Dec(X) and that for each Ai a is given. Then -9/1 u ... u gd proper decomposition giof (f(Ai,glAi)Ai is a partition of (f,g)X which we denote by . We define Dec((f,g)X) as consisting of all partitions <3; [ sl,. ..,2d) > . We call 9 the base of the decomposition <9;( gl,. ..,3 d 1 > . Suppose now that
(3.6) Let us say that a decomposition A
is the union of a subcollection of
in
9 is contained in
of lRn
9,
partitions a set A CRn
if
in other words, if each set
A or disjoint from A
.
The following terminology will not be used in the remainder of this section, but it might be helpful in comparing with results in [ C ] . decomposition is a decomposition of lRn some
n
.
,
that is, a member of
A cylindrical
Dee@) ,
A cylindrical decomposition which partitions a set A CIRn
called a cylindrical decomposition of A
.
is also
With this terminology the next
theorem states that for any definable sets A1,AZ, cylindrical decomposition.
for
...,Am clRn
there is a c o m n
L.VAN DEN DRIES
112
n-1 X CIR
( a ) For each d e f i n a b l e s e t
X i n t o d e f i n a b l e s e t s on each of which
f i n i t e p a r t i t i o n of
(b) Given any d e f i n a b l e s u b s e t s p a r t i t i o n i n g each of
0
X =
or
...,Am
A1,
...,Am .
Al,
The proof i s by i n d u c t i o n on (since
f:X ->B
and d e f i n a b l e
.
n
of
Rn
t h e r e is a
i s continuous.
f
9€ DecQRn)
there i s
n = 1 statement (a) holds t r i v i a l l y
For
i s a s i n g l e t o n ) , and (b) i s an easy e x e r c i s e i n boolean
X
a l g e b r a u s i n g the f a c t t h a t each
Ai CIR
i s of f i n i t e type.
The induction
s t e p s take p l a c e i n t h e proof of t h e following t h r e e lenrmas.
m.
(3.8) all
for
n
.
N
Suppose
>
1 and statements (a) and (b) of (3.7) hold f o r
Then statement (a) holds f o r
Proof. -
Let
n = N-1
we may r e s t r i c t t o t h e c a s e t h a t
x
be d e f i n a b l e and
cIRN-l
Suppose f i r s t t h a t
dim X
< N-1
homeomorphism h:X 2' X '
where
XI
.
.
n = N f:X X
-> E
R
definable.
Using (b)
.
F(N-l)
Then, by ( 3 . 4 ) , we have a d e f i n a b l e
E F(N'2)
.
The f u n c t i o n
fh-':X'
->a
is d e f i n a b l e , so by the i n d u c t i o n hypothesis t h e r e i s a f i n i t e p a r t i t i o n of i n t o d e f i n a b l e s e t s on each o f which partition via o f which
f
w e g e t a f i n i t e p a r t i t i o n of
h
X
i n t o d e f i n a b l e s e t s on each
The set
Y = {x
dim X = N-1
E Xlf is
.
use (b) f o r
n = N-1
Y
.
X\Y
continuous a t
x)
contained i n
Y
,
so
which a r e contained i n
i s dense i n
Y
t o g e t a decomposition g o f lRN-'
The s e t s i n
9which
a r e contained i n
X
arguments used i n t h e c a s e
dim X
< N-1
and a r e open must be
< N-1
The sets i n
,
9
and so the
apply.
To prove t h e claim we t a k e any nonempty open s e t i s continuous i n a t l e a s t one p o i n t of
.
which p a r t i t i o n s both X
but n o t open a r e of dimension
X
i s definable, we can
i s continuous o n each of those sets.
f
.
Then X i s open i n
I f we accept t h i s claim f o r a m m n t then, s i n c e
and
I f we l i f t t h i s
is continuous.
W e a r e l e f t with the case Claim. -
is continuous.
fh-'
X'
U
U C X
.
To show t h a t
w e use e x a c t l y t h e same arguments
f
Remarks on Tarski's Problem Concerning (lR,+, . , exp) The r o l e of the decreasing segments
a s i n the proof of (2.1).
113
i s now
[an,bn]
U
of course taken over by a decreasing sequence of closed b a l l s contained i n
their diameter tending t o
11 (xl ,...,%-1) 11
by the norm b a l l s are
@ere we l e t the distance on rPN-' = max (lxll
,...,IX+~~) ,
be defined
i n order t h a t the
R-definable.)
m.
(3.9)
.
0
,
Assume t h a t
N
Then any two decompositions
>1
n = N-1
and statement (b) of ( 3 . 7 ) holds f o r
gland g2of
EN have a COrrrmOn refinement,
t h a t is, there is a decomposition of RN p a r t i t i o n i n g each of the sets i n
2B1 u s 2 .
Proof.
Note t h a t RN =
(-=,-& N
s t r u c t u r e of decompositions of R t h a t @) holds f o r a conmmn base
where
n
= N-1
see (3.5).
.
.,Bd]
qi
n-1
is a decomposition of B
Bij
,
fh
,
and
,
g* p a r t i t i o n s
in
(-=,")
,
,
are definable and continuous.
we have:
Bi Bi
= (-"Jfl/...lfpl-)Bi
, gCr:,Bi - 3 3 R
Fix a set
(proper) decompositions of
By the assumption t h a t (b) holds f o r that
gland g2have
.
can be partitioned i n t o f i n i t e l y many definable sets any
This gives us the
So we can use the assumption
to reduce t o the case t h a t
9 and say gli,32ia r e
fh
,
.
X =lRwl
9 ,say
g=[B1,.
where the
where
B
Now c l e a r l y
Bi
such that, given
is
< $(x)
either for a l l
x
E
Bij:
fh(x)
o r for a l l
x
E
Bij:
fX(x) = g,(x)
or for a l l
x
E
Bij:
fh(x)
n = N-1
each of the sets
Bij
there i s
> g,,(x)
g*E DecCpN-')
. Now it should
,
,
.
such
b e c l e a r how to
.
L.VAN DEN DRIES
114
the
gp )
f X and
g* (and using r e s t r i c t i o n s s e t i n glU g2. [I
RN w i t h base
c o n s t r u c t a decomposition of
which p a r t i t i o n s each
of
Note t h a t Lemma (3.9) i m d i a t e l y extends t o a f i n i t e c o l l e c t i o n of de-
.
N
compositions of lR
It i s t h i s s t r o n g e r form t h a t we s h a l l use i n t h e
f i n a l s t e p which follows now.
e. Suppose
(3.10)
statement (b) holds f o r Let
Proof. consider
N-1
Ai
Formula f o r
>
B c n(Ai)
f(B,l)
gB % (--I
n = N
Then statement (b) holds f o r
...,Am
A1,
of IRN
be given.
nRN ->lRN-'
where
n = N
and
.
Fix an Ai
and
i s t h e p r o j e c t i o n on
We a r e going t o apply (1.9) t o t h e d e f i n i n g We a l s o use
These c o n s i d e r a t i o n s give us a decomposition
partitioning
there is
n Ai
.
t a k i n g i n t o account Remark (1) following (1.9).
,...,f(B,k):B
n-l(B)
statement (a) of (3.7) holds f o r
rr(Ai)
t h e hypothesis of t h e lemma. . o f lRN-'
,
d e f i n a b l e sets
coordinates.
,
1
n = N-1
a s l y i n g over
Ai
the f i r s t
N
n(Ai)
and such t h a t f o r each s e t
B E
si
with
k = k(B) and t h e r e a r e d e f i n a b l e continuous f u n c t i o n s with
->lR
f(B,1)
<
... < f(B,k)
on
B
such t h a t
is p a r t i t i o n e d by t h e decomposition f(B, 1)
1.. . I
f(B,k) 1 9
n - l ( ~ ) = (-a,m)B
of
.
It i s
il-l(B)
c l e a r t h a t t h i s g i v e s us a decomposition
B of
WN
which p a r t i t i o n s
9; ,...,
Ai
.
p a r t i t i o n s each of
Now a c o m n refinement
A1
,...,Am .
n il(Ai)
=@
>
@ E Dec@RN)
of
u
This completes t h e proof of Theorem (3.7). (3.11) Corollary.
Each d e f i n a b l e s u b s e t of
Rn has only f i n i t e l y many connected
components, and each component i s a l s o d e f i n a b l e .
Remarks on Tarski's Problem Concerning (IR,+, . , exp)
Proof.
By (3.7) each definable subset
decomposition
9of
.
lRn
A
of
Rn
is partitioned by a
Now each of the f i n i t e l y many s e t s i n
(definable) c e l l , hence connected.
Each component of
of f i n i t e l y many c e l l s belonging t o
9,
115
9i s
a
i s therefore a union
A
and i s therefore definable.
( 3 . 1 2 ) Remarks (1) Theorem ( 3 . 7 ) was derived under the assumption t h a t
9i s
arguments i n ( 3 . 2 ) expansion
2 of
of f i n i t e type.
-
f i n i t e type
T h ( 2 ) ). Conversely, a weak form o f ( 3 . 7 )
(which i s r e a l l y an assumption on implies t h a t
9i s of
To be precise:
a l l of the definitions and
( 3 . 6 ) make sense and go through without change f o r any
@,<)
,
whether of f i n i t e type or not.
Now we have the
following f a c t .
If
2 is
finable subset titioning
A
,
an expansion of A
of lRn
then
,
@,<)
n
9i s of
2
1
,
with the property t h a t f o r each dethere i s a decomposition of IRn
f i n i t e type.
(Hint:
par-
derive f i r s t (1.9) f o r
d= 9-1 (2) The assumption that
9i s
with names f o r r e a l numbers.
of f i n i t e type remains v a l i d upon expanding
9
Therefore, Theorem ( 3 . 7 ) also applies t o
R-definable s e t s and functions where of course the notion of decomposition of Rn
i s s u i t a b l y relativized.
(3) The condition t h a t concrete case.
9i s
of f i n i t e type might be d i f f i c u l t t o prove i n a
The point of t h i s paper i s j u s t t o show what more i s true i f
t h i s condition holds.
For 'exponentiation' one may t r y to use t h i s knowledge
the other way around:
since I expect t h a t
QR,
<,+ ,
, exp)
is of f i n i t e
type, b u t don't see any model theoretic technique t o prove t h i s , I prefer t o s t a y inside W
and use the analytic machinery available there t o e s t a b l i s h
d i r e c t l y the conclusion of Theorem ( 3 . 7 ) . method t o achieve t h i s .
The introduction sketches a (tentative)
L. VAN DEN DRIES
116
(4)
9i s
of f i n i t e type i f and only i f each i n f i n i t e R ' - d e f i n a b l e s u b s e t
,
of R'
where
...) =9, has nonempty i n t e r i o r
9' =
@I,<,
(i.e.,
contains
an i n t e r v a l of IR' ). (Exercise.) The second p a r t of t h i s equivalence a l s o makes s e n s e f o r o t h e r topological-algebraic structures. analogue of t h i s second h a l f f o r
Now Macintyre [Mac] e s t a b l i s h e d t h e p - a d i c a l l y closed f i e l d s .
t o use this f a c t t o prove an analogue of Theorem (3.7) numbers
It would be n i c e
f o r t h e f i e l d of p-adic
5 .
(5) The r e l a t i o n between ' c y l i n d r i c a l decomposition' and ' q u a n t i f i e r e l i m i n a t i o n ' i s the following:
i f each q u a n t i f i e r f r e e d e f i n a b l e s u b s e t of
t i f i e r f r e e d e f i n a b l e c y l i n d r i c a l decomposition, then t i f i e r elimination.
Rm has a quan-
T h ( 2)
This is how C o l l i n s [C] proves t h a t
QR
admits quan-
,< ,0 ,
1
, + , .)
has q u a n t i f i e r e l i m i n a t i o n .
(6) Let
9be
an expansion of
@
.
,< ,+ , -)
P o s t u l a t i n g t h e f i n i t e type
c o n d i t i o n i s of course a r a t h e r d r a s t i c way of avoiding t h e G'ddel phenomena t h a t would appear i f
N were definable. I n t h i s connection I would l i k e t o know t h e
answer t o the following question: i n f i n i t e d i s c r e t e s u b s e t of IR
i s the s e t N
d e f i n a b l e whenever some
is definable?
APPENDIX
Throughout t h i s appendix we assume t h a t
2 is
an expansion of
@,
<,
i s of f i n i t e type, i.e.,
with the p r o p e r t y t h a t each B-definable s u b s e t of B has only f i n i t e l y many connected components. I\
Recall t h e n o t a t i o n IR = IR
(A.l)
w.
R-definable.
I
Suppose Then both 4
s i m i l a r l y ) e x i s t i n IR
f'(a-)
.
U(-OP,OD]
i s an i n t e r v a l , = lim
hf 0
a
E
I
f(a+h)-f(a) h
+)
, and
and
f:I
f'(a+)
->a
is
(defined
Remarks on Tarski's Problem Concerning (lR,+, . , exp)
Proof.
W e s h a l l j u s t treat the case of
f'(a+)
.
117
(The case of
f'(a-)
i s handled s i m i l a r l y . ) 4 = l i m i n f f(a+h)-f(a) hL0 h
Suppose
r
Choose a r a t i o n a l h
>0
that
such t h a t
t h e s e t of a l l
h
contradiction.
0
w.
(A.2)
2
rh
.
>
0
such t h a t
Suppose
i s an i n t e r v a l ,
I
increasing, and i t s inverse
Proof.
the maps
Suppose f ' (a')
continuous on a l l of
Proof.
with
all
I
on (A.4)
J
,
(Note:
a EI
.
Then
f
f(1)
l/f'(a+)
is strictly
has the
if
f(a) = b
.
the c o n t i n u i t y assumption cannot
,
a
.
Then
+>
f '(a-)
E P and an i n t e r v a l
J
I
hence
x E J
g
If
I
for a certain
around
a
a
E
I
a E I
.
g'(x+)
. .
Then
such t h a t
Then t h e continuous function
has t h e property t h a t
>0 ,
g'(x-)
g : J --3P
< 0 for
would be both s t r i c t l y increasing and s t r i c t l y decreasing
by the previous lemma.
m.
.
for all
f ' ( a + ) = f'(a-)
f'(a-)
for all
is continuous, and
i s continuously d i f f e r e n t i a b l e on I
f
>
f'(x-)
f : I ->R
a r e w e l l defined, r e a l valued, and
f'(a+)
g(x) = f(x)-cx
,
i s continuous and
defined on the i n t e r v a l
It s u f f i c e s t o show t h a t
>c>
x
fdl
f : I ->lR
for a l l
I i s an i n t e r v a l ,
Suppose the contrary, say c
such
0
a I->
there is
>0
< r h is IR-definable,
i s defined and equal t o
f")'(b+)
Lemma.
f'(x+)
>0
L e f t t o the reader.
be omitted.) (A.3)
b u t a l s o a r b i t r a r i l y small h
f(a+h)-f(a)
i s defined ( i n IR ), and
property t h a t
Then t h e r e are a r b i t r a r i l y s m a l l
This s i t u a t i o n is incompatible with the f a c t t h a t
A
f'(a+)
,
< rh
f (a+h)-f(a)
f(a+h)-f(a)
.
a< r
with
.
L = l i m sup f(a+h)-f(a) hi0 h
Contradiction.
i s an i n t e r v a l and
t h e r e are only f i n i t e l y many x
E
I
0
f:I --3R
such t h a t
+ f'(x )
i s R-definable, =
&m
.
then
L. VAN DEN DRIES
118 Proof. {x
E
1
I
Suppose f o r example t h a t t h e R-definable s e t
f'(x+) =
is infinite.
m]
So t h i s s e t c o n t a i n s a whole i n t e r v a l ,
and f o r the sake of d e r i v i n g a c o n t r a d i c t i o n , we may a s w e l l assume t h a t f'(x+) = that
for a l l
.
I
By ( 2 . 2 ) we may f u r t h e r r e s t r i c t t o t h e case
i s continuous, which implies t h a t
f
whence
E
x
2
f'(x-)
0
for a l l
EI
x
f
,
I
i s s t r i c t l y i n c r e a s i n g on
.
A f t e r f u r t h e r shrinking t h e i n t e r v a l , we may a l s o assume t h a t w e are i n one of two c a s e s :
(1) f ' ( x - )
= a
for all
E
x
I
i s finite for a l l
(2) f ' ( x - )
I
continuous on
.
b
,
whence
I
,
x C> f'(x-)
and
is
(This uses again ( 2 . 2 ) . )
I n c a s e (1) the i n v e r s e for a l l
E
x
f-'
of
f
satisfies
(f-')'(b+)
is . constant, c o n t r a d i c t i n g i t s i n j e c t i v e n e s s .
f-'
case (2) we can apply t h e same argument a s i n t h e proof of Lemma ( A . 3 )
Then a.
<
Proposition. f
... <
am
Proof. that
functions
Suppose
f : (a,b)
->
R
i s &definable,
a
<
,
a.
,
= a
a = b m
such t h a t
f
f
.
b
i s continuously d i f f e r e n t i a b l e
.
(aiJai+l)
By ( 2.2) we may a s w e l l assume t h a t f'(a+)
and
are finite for a l l
f'(a-)
a b> f'(a+)
and
a
b> f'(a-)
i s continuous, and by
f a
E
I
.
But then t h e
a r e R-definable,
hence
0
piecewise continuous, and t h e r e s u l t follows i m e d i a t e l y from ( A . 3 ) .
(A.6)
to get
i s piecewise continuously d i f f e r e n t i a b l e , t h a t i s , t h e r e a r e
on each s u b i n t e r v a l
(A.4)
In
0
a contradiction.
(A.5)
=0
= (f-')'(b-)
Corollary.
i s piecewise
t o i n c r e a s e with
Proof.
Suppose Cn
,
f : ( a , b ) ->R
f o r each
n
6N
.
i s IR-definable,
a
(The number of 'pieces
'
.
Then
i s allowed
n .)
This follows from P r o p o s i t i o n (A.5) by i n d u c t i o n on
n
.
0
Remarks on Tarski's Problem Concerning (IR,+, . , exp)
(A.7)
119
Let us a l s o mention t h e following i n t e r e s t i n g consequence of the R
hypothesis t h a t each IR-definable s u b s e t of
i s of f i n i t e type; we s h a l l
n o t use i t i n t h e r e s t of t h i s appendix. Proposition.
T h ( 9 ) has d e f i n a b l e Skolem functions.
equivalence r e l a t i o n on a d e f i n a b l e s u b s e t of
Also each d e f i n a b l e
has a d e f i n a b l e s e t of
Rn
representatives. and (4.1)] f o r the theory
The proof i s s i m i l a r t o t h e one given i n [vdD3,(1.2) of r e a l closed f i e l d s .
(A.8)
The c o n s i d e r a t i o n s of ?-submanifold
of
XIn
,
X
'defining'
ha)
?(X)
= {f:X
FM(ilJ
..., i n )
(ilJ...,in)
->1
be a c l a s s of
dim(X)
.
C?
,
Put F
De\(X)
So t h e
X E F$)
for
=
u
b u t we r e q u i r e t h e
.
FM(ilJ...,in)
hO[) E F$-')
then
M C -isomorphism.
g)
C'-submanifolds
the d e f i n i t i o n i s
{0,1] ;
in
and similar t o t h e one i n (3.3),
n
is a
Note t h a t
and the homeomorphism
?-analogue
as i n (3.5),
of (3.4) holds.
and we can s t a t e the
of meorem (3.7).
?-analogue
(~.9) Theorem. n z l
f is a d e f i n a b l e
F u r t h e r we l e t
and
We now d e f i n e following
I
we p u t
functions t o be
X E F$)
h:X
For a d e f i n a b l e
f o r each sequence
by i n d u c t i o n on
if
53 can now be r e f i n e d as follows.
of lRn
.
M C -function]
.
M E IN
F i x a number
suppose
s= , < , + , ...I
i s of f i n i t e type.
. m:
(a) For each d e f i n a b l e s e t f i n i t e p a r t i t i o n of
X
X CIRn-'
into
and d e f i n a b l e
E F$-"
?-manifolds
f:X
->
R
there is a
on each of which
f
$-function. (b) Given any d e f i n a b l e s u b s e t s
A1,
p a r t i t i o n i n g each of
.
AIJ...,Am
...,A
m
of
Rn
t h e r e is
9E
DecM(Rn)
L. VAN DEN DRIES
120
The proof is along the lines of the proof of (3.7)
and w e leave i t to the reader
to supply the e x t r a work needed i n the
of ( 3 . 8 ) .
(A.6),
is
induction on
3
(A.lO)
M
(?-version
(Hint:
use
and the f a c t t h a t a function on an open subset of lRm
i f f i t s p a r t i a l derivatives up to order
M
e x i s t and a r e continuous.)
This d i f f e r e n t i a b l e analogue of Theorem (3.7) could be useful i n a proof
of the following conjecture: L e t the expansion $(vl,
...,v-)
an
2 of
%-fornula.
@
,< ,+ , . )
be of f i n i t e type and
Then the family of =-definable
contains only f i n i t e l y many homeomorphism types.
sets
CIn p a r t i c u l a r , there i s a
uniform bound on the number of connected components of
Xa
g a ranges