See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/281936805

STANDARD MODELS OF ARITHMETIC

Conference Paper · September 2014

CITATIONS READS

0 2,101

1 author:

Ali Enayat
University of Gothenburg
58 PUBLICATIONS   210 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Metamathematics of Arithmetic and Set Theory View project

All content following this page was uploaded by Ali Enayat on 17 December 2016.

The user has requested enhancement of the downloaded file.

 STANDARD MODELS OF ARITHMETIC

ALI ENAYAT

Abstract. A ZF-standard model of PA (Peano arithmetic) is a model

of arithmetic that is of the form NM for some model M of ZF (Zermelo-
Fraenkel set theory), where NM is the standard model of arithmetic in
the sense of M. We provide a recursive axiomatization ΦT of the set
of arithmetical consequences of any prescribed recursively axiomatizable
extension T of ZF, and then we establish the following characterization
of ZF-standard models of countable cofinality.
Theorem. Let T be a recursively enumerable extension of ZF. The
following conditions are equivalent for a nonstandard model A of PA
that has countable cofinality:
(a) There is a model M of T such that A = NM .
(b) A is a recursively saturated model of PA + ΦT .

Gödel’s incompleteness theorems demonstrated once and for all that our
knowledge of a sufficiently rich structure is highly dependent on the choice
of higher order formal frameworks within which the structure is viewed
through. This Gödelian insight has come to shape a great deal of research
in the foundations of mathematics, particularly in set theory, where the
higher order frameworks are provided by large cardinal axioms. For example,
consider Gödel’s constructible universe L of set theory. It is consistent with
ZF that the class V of all sets coincides with L, and yet it is well-known [J]
that in the presence of sufficiently large cardinals (e.g., a Ramsey cardinal)
0# exists, and in particular the satisfaction-predicate of L is definable in V,
which in light of Tarski’s undefinability of truth theorem is a strong form of
asserting V 6= L.
Date: December 13, 2016.
It is a pleasure to present this paper in a Festschrift volume for Christian Bennet. The
results presented here arose from a question posed by Joel Hamkins and Victoria Gitman
that asked whether there is a characterization of models of arithmetic that arise as the
standard model of arithmetic in some model of ZF; this question was communicated to
the author by Roman Kossak at the Mittag-Leffler Institute, during September of 2009.
Thanks to Roman, Menachem Magidor, and Jouko Väänänen for inspiring conversations,
to Jim Schmerl for reminding me of the relevance of his remarkable paper [Sch-2] to the
questions dealt with here, to Fredrik Engström for a careful reading of the penultimate
draft which resulted in weeding out many infelicities, and to Martin Kaså for his fine efforts
in putting this volume together. NOTE: IN THIS DRAFT THE PROOF OF LEMMA
11.1 IS IMPROVED.
1
2 ALI ENAYAT

The discovery of Paris-Harrington phenomena in the late 1970s, and the

striking work of Harvey Friedman (as recently summarized in [F]) has fur-
ther accentuated the pervasiveness of the Gödelian insight regarding the
“standard model of arithmetic”, whose existence and uniqueness up to iso-
morphism, is only possible in the context of a background theory of sets.
The proof of existence and uniqueness (up to isomorphism) of the standard
model of arithmetic was originally carried out in the pioneering work of
Dedekind [D]. In the ZF-setting, Dedekind’s proof is amalgamated with von
Neumann’s implementation of ordinal numbers [V], where one defines the set
ω of natural numbers as the intersection of all inductive sets1, and then one
uses appropriate recursion theorems to define addition and multiplication
on ω.2
With the above background in place, we are in a position to probe the
class of models of arithmetic that are “the standard model of arithmetic” of
some model M of ZF, i.e., models of arithmetic that arise as NM (read: N
in the sense of M), where N = (ω, +, ·), ω is the set of finite von Neumann
ordinals, and + and · are respectively ordinal addition and multiplication
on ω. We begin with some definitions and observations.
1. Definition. Suppose T is a recursively axiomatizable extension of ZF
(in the same language).
(a) A T -standard model of arithmetic is a model of the form NM , where
M |= T .
(b) PAT is the theory of the class of T -standard models of arithmetic, i.e.,

PAT := ∩ Th(NM ) : M |= T .


2. Remark. It is easy to see, using the completeness theorem for first

order logic, that
PAT = {ϕ : ϕ is an arithmetical sentence and T ` ϕN }.
The above characterization makes it clear that PAT has a recursively enu-
merable axiomatization, and therefore, by Craig’s trick, PAT has a primitive
recursive axiomatization. This motivates the search for a “natural axioma-
tization” of PAT . Note that PAT far extends PA, since, e.g., PAT includes
the statements Con(PA), Con (PA + Con(PA)), etc. Moreover, since Gödel’s
constructible universe L satisfies GCH (generalized continuum hypothesis)
and AC (axiom of choice), routine absoluteness consideration show that
PAZF = PAZFC+GCH .
1In this setting, a set X is inductive if it satisfies (i) ∅ ∈ X, and (ii) ∀x ∈ X x∪{x} ∈ X.
The axiom of infinity of ZF states that there is at least one inductive set; Dedekind justified
the existence of an infinite set by informal considerations related to the totality of “objects
of thought”.
2Dedekind’s uniqueness proof has been recently investigated in the context of reverse
mathematics in [SY] and [KY].
 STANDARD MODELS OF ARITHMETIC 3

Similarly, since the truth of arithmetical statements is not altered by forcing

extensions, we may conclude that
PAZF = PAZFC+¬CH .
On the other hand, let I denote the set-theoretical sentence that expresses
“there is an inaccessible cardinal”. By coupling the well-known fact that
Con(ZF) is provable in ZFI with Gödel’s second incompleteness theorem we
can conclude:
PAZF ( PAZFI .
Our first result provides a “natural” recursive axiomatization of PAT . In
order to describe it, we need the following definition.
3. Definition. Suppose {θn : n < ω} is a recursive enumeration of the
sentences of some extension T of ZF, and let ΦT be the collection of sentences
of the form
ϕ → Con(Tn + ϕN ),
where Tn = {θi : i < n}, ϕ ranges over sentences of arithmetic, ϕN is
the sentence in the language of set theory that expresses “N |= ϕ”, and
Con(X) is the arithmetical sentence that expresses the formal consistency
of X. Note that ΦT is recursive (recall that T is assumed to be recursively
axiomatizable).3
4. Theorem. For every recursively axiomatizable extension T of ZF,
PA + ΦT axiomatizes PAT .
Proof. Each axiom of ΦT is a theorem of PAT , essentially because of the
reflection theorem of ZF [J], and the fact that ZF can verify that a theory is
consistent if it has a model; hence the set of logical consequences of PA + ΦT
is a subset of the set of logical consequences of PAT . To show the other
direction, suppose θ is an arithmetical sentence such that PAT ` θ. Then
there is some natural number k such that Tk ` θN . Therefore, since PA is
Σ1 -complete,
PA ` ¬Con(Tk + ¬θN ).
On the other hand, ¬θ → Con(Tk + ¬θN ) is one of the axioms of ΦT , which
makes it clear that PA + ΦT ` θ, as desired. 
5. Remark. By putting Gödel’s second incompleteness theorem together
with the fact that the Σn -reflection scheme is provable in the fragment of
ZF in which the replacement scheme is limited to Σn+1 -formulas we can
show that ΦZF is not axiomatizable over PA by any collection of arithmetical
sentences of bounded quantifier complexity; and in particular ΦZF is not
3A similar idea was discovered earlier by Azriel Lévy [L]. I realized this about two
years after the preparation of the first draft of this note, as indicated in my FOM posting:
http://www.cs.nyu.edu/pipermail/fom/2011-June/015532.html
Later, Solovay presented another axiomatization of PAT that is mentioned at the end
of Remark 5; for more detail see his FOM posting:
http://www.cs.nyu.edu/pipermail/fom/2006-May/010549.html
4 ALI ENAYAT

finitely axiomatizable over PA. Also, since for each natural number n there is
an arithmetically definable truth predicate for Σn -arithmetical sentences one
can use the proof strategy of Theorem 4 to show that PAT is also axiomatized
by sentences of the form {ψn : n ∈ ω}, where ψn is the arithmetical sentence
that expresses:
“Con(Tn + ϕN ) holds for all true Σn sentences ϕ”.
6. Proposition. Every ZF-standard model of PA that is nonstandard is
recursively saturated.4
Proof: ZF can define the Tarskian satisfaction predicate for every set struc-
ture; and in particular it can do so for N. In light of this observation,
recursive saturation follows from a routine overspill argument, as in [Ka,
Proposition 15.4]. 
7. Theorem. Every countable recursively saturated model of PA + ΦT is a
T -standard model of PA.
Proof: The quickest way to prove this theorem is by a “resplendence ar-
gument”, as in the proof of Corollary 9 (since every countable recursively
saturated model is resplendent [Ka, Theorem 15.7]), but we take this oppor-
tunity to present a direct proof (indeed, by a varation of this proof, one can
show the resplendence property of countable recursively saturated models).
Given a countable recursively saturated model A |= PA + ΦT , first use re-
cursive saturation to find c in A that codes Th(A) by realizing the recursive
type Σ(v), where
Σ(v) := {ϕ ↔ pϕq ∈Ack v : ϕ ∈ Sent},
where pϕq is the Gödel number for ϕ, ∈Ack is “Ackermann’s ∈”, i.e., a ∈Ack b
holds iff the a-th bit of the binary expansion of b is 1; and Sent is the (re-
cursive) set of sentences in the usual language of arithmetic. Next, let
hϕn : n ∈ ωi be a recursive enumeration of arithmetical sentences, and con-
sider the arithmetical formula γ(i) defined below:
γ(i) := Con (Ti + {ϕj : j < i ∧ ϕj ∈Ack c})
It is easy to see, using the assumption that ΦT holds in A and our choice of c
that A |= γ(n) for each n < ω, and therefore by overspill A |= γ(d) for some
nonstandard element d of A. By invoking the Hilbert-Bernays arithmetized
completeness theorem [KS, Thm 1.12] within A, we can conclude that there
is a model M of T with the following three properties:
(1) Th(A) = Th(NM ).
(2) NM is recursively saturated.
4This proposition is a descendent of a result of Ehrenfeucht and Kreisel [EK] that
states that no model of PA that is expandable to a model of second order arithmetic is
finitely generated (recall that an infinite recursively saturated model cannot be finitely
generated).
 STANDARD MODELS OF ARITHMETIC 5

(3) SSy(A) = SSy(NM ).

Note that (1) holds by design and (2) holds by Proposition 6. The definabil-
ity of M within A combined with an internal recursion within A, implies
that there is an A-definable embedding from A into an initial segment of
NM , which in turn shows that (3) is true since standard systems of models
of arithmetic are invariant under end extensions.
Since the isomorphism type of a countable recursively saturated model of
PA is uniquely determined by (a) its first order theory, and (b) its standard
systems [KS, Prop. 1.8.1], we may conclude that A ∼ = NM , as desired. 
8. Corollary. The following statements are equivalent for a countable
nonstandard model A of arithmetic:
(a) A is a T -standard model of PA.
(b) A is a recursively saturated model of PA + ΦT .
Proof: This follows from Proposition 6 and Theorem 7. 
9. Corollary. Every resplendent model of PA + ΦT is a T -standard model.
Proof: This follows from Corollary 8 from a routine “reduction to the
countable” argument. More specifically, given a resplendent model A of
PA + ΦT , augment the language of PA with a new function symbol f , and
binary relation E, and let Σ be the recursive set of sentences in the expanded
language of arithmetic that says that T holds in the structure (A, E) (where
A is the universe of A), and f is an isomorphism between N(A,E) and A. By
Corollary 8, Σ is consistent, and therefore by resplendence, A is a T -standard
model. 
10. Remark.
(a) The countability condition is crucial for the (b) ⇒ (a) direction of Corol-
lary 8. To see this, recall that, by a theorem of Kaufmann, every completion
of PA has a recursively saturated model A all of whose classes5 are para-
metrically definable in A [KS, Thm 10.1.5]. Since ZF-standard models of
PA always carry a full inductive satisfaction class, and by Tarski’s theorem
no such satisfaction class is first order definable, this shows that A is not a
ZF-standard model (however, Theorem 11 presents an uncountable analogue
of Corollary 8).
(b) As shown in Corollary 16, the converse of Corollary 9 does not hold.
11. Theorem. Corollary 8 remains true if “countable” is weakened to “of
countable cofinality” and T includes DCω (dependent choice of length ω).6
5A subset X of the universe of discourse A of a model A of arithmetic is said to be a class
of A if X is piecewise coded in A, i.e., ∀a ∈ A ∃b ∈ A such that {x ∈ X : A |= x < a} =
{x ∈ A : A |= x ∈Ack b} .
6Indeed this theorem holds even without assuming that T includes DC , but the proof
ω
becomes more involved (one would either need to take advantage of the omitting types
theorem, or bring the Shoenfield absoluteness theorem in to the picture).
6 ALI ENAYAT

Proof: It is sufficient to prove that every cofinal elementary extension of

a T -standard model is also a T -standard model, since using a Löwenheim-
Skolem argument one can easily show that a recursively saturated model of
PA of cofinality ω has a cofinal countable recursively saturated elementray
submodel. More specifically, by a theorem due independently to Kotlarski
and Schmerl [KS, Theorem 1.3.7], we have:
If A |= PA, X ⊆ A with (A, X) |= PA∗ [i.e., the induction
scheme continues to hold for formulae mentioning X], and
A ≺cofinal B, then there is a (unique) Y ⊆ B such that
(A, X) ≺ (B, Y ).
Kotlarski [Ko] used this theorem to show that a cofinal extension of a ZFC-
standard model is also a ZFC-standard model. His argument can be refined
to establish the following result.
11.1. Lemma. Suppose A = NM for some model M of ZF + DCω , and
A ≺cofinal B. Then B = NM for some M ≡ M.
Proof: Use the ZF-reflection theorem within M to find an increasing se-
quence of “ordinals” αn of M such that
M |= “ (Vαn , ∈) ≺Σn (V, ∈) ”.
Then, using an external induction, and an internal repeated application of
the Löwenheim-Skolem theorem7, build an increasing chain hXn : n < ωi
with Xn ⊆ Vαn satisfying the following properties:
(a) M |= “Xn is a countable subset of Vαn ”;
(b) M |= “ (Xn , ∈) ≺ (Vαn , ∈) .
Note that by the choice of αn s, (b) implies
(c) M |= “ (Xn , ∈) ≺Σn (Xn+1 , ∈) .
The M-countability of each Xn , and the fact that within M each (Vαn , ∈)
is an ω-model (since ZF proves that each Vα is transitive) coupled with (c),
shows that within M we can copy each (Xn , ∈) over ω M , thereby getting
hold of relations En , and functions g and fn (all in M) such that for all
n<ω:
(d) (A, En , En+1 , g, fn ) |= Φn (En , En+1 , fn , g),
where Φn (En , En+1 , fn , g) is the theory consisting of PA∗ , along with the
following three statements:
(i) En ⊆ ω 2 , fn : ω → ω, and g is an isomorphism between the ambient
model and N(ω,E1 ) .
(ii) fn : (ω, En ) →Σn (ω, En+1 ), i.e., fn embeds (ω, En ) as a Σn -elementary
submodel of (ω, En+1 ) .
7The Löwenheim-Skolem theorem is provable in ZF + DC . Indeed, it is well known
ω
that over ZF, the Löwenheim-Skolem theorem is equivalent to DCω .
 STANDARD MODELS OF ARITHMETIC 7

(iii) Nfn (ω,En ) = N(ω,En+1 ) , i.e., (ω, En+1 ) has no natural numbers other
than the ones in fn (ω, En+1 ) .
By the aforementioned Kotlarski-Schmerl theorem, (d) implies that there
exist E n , g and fn such that for all n < ω :


(e) B, E n , E n+1 , g, f n |= Φn (E n , E n+1 , f n , g).


The desired model M is the direct limit of the models ω, E n via the

embeddings f n : n < ω . 
In order to present the next result (Theorem 14) we need a definition.
12. Definition. Suppose M is a model of ZF.
(a) NM is M-rather classless if all the classes of NM are parametrically
definable in M (and therefore each class is coded as a real number of M).
(b) RM is the field R of real numbers in the sense of M.
(c) R(M) the external Scott completion (also known as the Cauchy comple-
tion) of the field of rational QM of M. The Scott completion of an ordered
field F is the largest ordered field that contains F as a dense subfield (by a
theorem of Dana Scott [Sco], every ordered field has a Scott completion).
13. Remark. Generally speaking, RM is a proper subfield of R(M);
and when M is countable, RM is also of course countable, but R(M) is of
cardinality 2ℵ0 . Moreover, one can show, using [Sch-1, Proposition 1.4] that
RM ∼= R(M) iff NM is M-rather classless.
14. Theorem. Let κ be an infinite cardinal, T be an extension of ZFC,
and suppose A0 is a T -standard model of cardinality less than κ.
(a) There is an elementary extension A of A0 such that A is κ-like and
end extends A0 .
(b) Moreover, if cf( κ) > ℵ0 , then A can be required further to be M-rather
classless, where A = NM , for some model M of T .
Proof sketch: Suppose A0 = NM0 , for a model M0 of T . Fix a nonprin-
cipal ultrafilter U in M0 , and let M be the κ-th iterated ultrapower of M0
modulo U. The desired model A is NM . The moreover clause uses the same
argument as in the well-known PA-case [KS, Thm 2.2.14]. 
15. Remark. The full force of AC is not needed for the above construction.
What is needed is enough choice to have a workable theory of ultrapowers,
i.e., (i) ACω , and (ii) the existence of a nonprincipal ultrafilter on P(ω).
16. Corollary. ZF-standard models need not be resplendent.
Proof: By Friedman’s self-embedding theorem [Ka, Theorem 12.4] every
countable nonstandard model of PA is isomorphic to a proper initial segment
of itself. This shows, by a “reduction to the countable” argument (as in the
proof of Corollary 9), that every resplendent model of PA is isomorphic to a
 8 ALI ENAYAT

proper initial segment of itself, a property that cannot be true of any κ-like
model, so thanks to Theorem 14(a) the proof is complete. 
17. Remark. Lemma 11.1 and Theorem 14 can also be derived as corol-
laries of general results of Schmerl’s [Sch-2]. Schmerl’s paper deals with
sequential theories (i.e., theories in which a “β-function” that codes finite
sequences is available) and which include the full scheme of induction in the
extended language. It is known that such theories are reflexive [HP, Thm
3.30, Ch.III].

References
[D] R. Dedekind, Essays on the Theory of Numbers, Dover Publications, New
York, 1963
[EF] A. Ehrenfeucht and G. Kreisel, Strong models of arithmetic, Bull. Acad. Polon.
Sci. Sér. Sci. Math. Astronom. Phys. vol. 14 (1966), pp. 107–110.
[F] H. Friedman, Boolean Relation Theory, to appear in the Lecture Notes in
Logic Series of the Association for Symbolic Logic.
[HP] P. Hájek and P. Pudlák, Metamathematics of First-order Arithmetic, Per-
spectives in Mathematical Logic, Springer-Verlag, Berlin, 1998.
[J] T. Jech, Set Theory, Springer Monographs in Mathematics, Springer-Verlag,
Berlin, 2003.
[Ka] R. Kaye, Models of Peano Arithmetic, Oxford University Press, Oxford, 1991.
[KY] L. A. Kolodziejczyk and K. Yokoyama, Categorical characterizations of
the natural numbers require primitive recursion, preprint available at
http://www.mimuw.edu.pl/˜lak/publik.html
[KS] R. Kossak and J.H. Schmerl, The Structure of Models of Peano Arithmetic,
Oxford University Press, Oxford, 2006.
[Ko] H. Kotlarski, On cofinal extensions of models of arithmetic, J. Symbolic Logic
vol. 48 (1983), no. 2, pp. 253–262.
[L] A. Lévy, Axiomization of induced theories, Proc. Amer. Math. Soc. 12 (1961)
pp. 251-253.
[Sch-1] J. Schmerl, On a problem of Sikorski, Israel J. Math 50 (1985), pp.145-159.
[Sch-2] , A reflection principle and its applications to nonstandard models, J.
Symbolic Logic vol. 60 (1995), no. 4, 1137–1152.
[Sco] D. Scott, On completing ordered fields, Applications of Model Theory to Al-
gebra, Analysis, and Probability (Internat. Sympos., Pasadena, Calif., 1967)
Holt, Rinehart and Winston, New York, pp. 274–278.
[SY] S. Simpson and K. Yokoyama, Reverse mathematics and Peano categoricity, Ann.
Pure Appl. Logic 164 (2013), no. 3, pp. 284-293.
[V] J. von Neumann, On the introduction of transfinite numbers, in Jean van Hei-
jenoort, From Frege to Gödel: A Source Book in Mathematical Logic,
1879-1931 (3rd ed.), Harvard University Press, pp. 346-354 (English translation
of von Neumann’s 1923 work).

Department of Philosophy, Linguistics, and Theory of Science, University

of Gothenburg, 405 30 Gothenburg, Sweden
E-mail address: [email protected]

View publication stats