1 Introduction
In 1973, Harvey Friedman proved a striking result for countable nonstandard models of finite set theory, and consequently for countable models of Peano arithmetic $ \mathrm {(PA)} $ stating that every countable nonstandard model of $\mathrm {PA}$ carries a proper initial self-embedding; here an initial self-embedding is a self-embedding whose image is an initial segment of the ground model [Reference Friedman, Mathias and Rogers5]. Afterward, many versions of Friedman’s style theorem appeared in the literature of model theory of arithmetic (e.g., see [Reference Dimitracopoulos and Paris3] or [Reference Wilkie, Lachlan, Srebrny and Zarach16]). In [Reference Bahrami and Enayat1], it is shown that some results on the set of fixed points of automorphism of countable recursively saturated models of $\mathrm {PA}$ can be generalized for initial self-embeddings of countable nonstandard models of $ \mathrm {I}\Sigma _{1} $ (see Theorem 2.4). In this paper, inspired by results about automorphisms of models of $\mathrm {PA}$ , we will investigate some more properties of countable models of $ \mathrm {I}\Sigma _{1} $ through initial self-embeddings.
In [Reference Enayat4], Enayat generalized the notion of a small submodel from [Reference Lascar, Kaye and Macpherson14], to I-small Footnote 1 for a given cut I of a model of $\mathrm {PA}$ (see Definition 1), and proved that:
Theorem 1.1 (Enayat).
Suppose $ \mathcal {M}\models \mathrm {PA} $ is countable, recursively saturated, and I is a strong cut of $\mathcal {M}$ . Moreover, let $ \mathcal {M}_{0}$ be an $ I$ -small elementary submodel of $\mathcal {M}$ . Then there exists some automorphism $ j $ of $\mathcal {M}$ such that $ M_{0} $ is equal to the set of fixed points of $ j $ .
In Section 3 of this paper, after investigating some basic properties of $ I$ -small $ \Sigma _{1}$ -elementary submodels of a countable model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ for some cut I of $\mathcal {M}$ , we will refine the above theorem for initial self-embeddings; i.e., we will show that I is strong in $\mathcal {M}$ iff every $ I$ -small $ \Sigma _{1}$ -elementary submodel of $\mathcal {M}$ is equal to the set of fixed points of some proper initial self-embedding of $\mathcal {M}$ . This result also generalizes one of the main theorems of [Reference Bahrami and Enayat1] (see Corollary 4.3).
Section 4 of this paper is devoted to the investigation of equivalent conditions to strongness of the standard cut, denoted by $ \mathbb {N} $ , in a countable model of $ \mathrm {I}\Sigma _{1} $ , through the set of fixed points of initial self-embeddings. In [Reference Kossak and Schmerl12], it is shown that:
Theorem 1.2 (Kossak–Schmerl).
Suppose $\mathcal {M}$ is a countable recursively saturated model of $\mathrm {PA}$ . If $ \mathbb {N} $ is not strong in $\mathcal {M}$ , then for every automorphism $ j $ of $\mathcal {M}$ the set of fixed points of $ j $ is isomorphic to $\mathcal {M}$ .
In Corollary 4.2, we will show that for every countable nonstandard model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ , if $ \mathbb {N} $ is not strong in $ \mathcal {M}$ , then the set of fixed points of any initial self-embedding $ j $ of $\mathcal {M}$ is either a model of $ \neg \mathrm {B}\Sigma _{1} $ , or is isomorphic to some proper initial segment of $\mathcal {M}$ . Then, we conclude that $ \mathbb {N} $ is strong in a countable recursively saturated model $\mathcal {M}$ of $\mathrm {PA}$ iff there exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that the set of fixed points of $ j $ is small in $\mathcal {M}$ and consequently it is not isomorphic to any proper initial segment of $\mathcal {M}$ .
In Section 5, we will study the extendability of initial embeddings of models of $ \mathrm {I}\Sigma _{1} $ to larger models. In particular, we will prove that any isomorphism between two $ \Sigma _{1}$ -elementary initial segment of a countable nonstandard model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ is extendable to some initial self-embedding of $\mathcal {M}$ iff it preserves coded subsets (for the case of automorphisms of countable recursively saturated models of $\mathrm {PA}$ this condition is only a necessary condition for extendability to larger models [Reference Kossak and Kotlarski10]).
2 Preliminaries
In this section we will review some definitions and results which are used through this paper. All unexplained notions can be found in [Reference Hájek and Pudlák6, Reference Kaye7].
-
• Through this paper, we will work in the language of arithmetic $ {\mathcal {L}_{ A}:=\{+,.,<,0,1\}} $ . For a given class $ \Gamma $ of $ \mathcal {L}$ -formulas (where $ \mathcal {L}\supseteq \mathcal {L}_{ A} $ ), $ \mathrm {I}\Gamma $ is the fragment of $\mathrm {PA}^{*}:=\mathrm {PA}(\mathcal {L})$ with the induction scheme limited to formulas of $ \Gamma $ . The $\Gamma $ -Collection scheme, denoted by $ \mathrm {B}\Gamma $ , consists of the formulas of the following form for every $\varphi \in \Gamma $ :
$$ \begin{align*} \forall \bar{z},u \ ((\forall x<u \ \exists y \ \varphi(x,y,\bar{z}))\rightarrow \exists v \ (\forall x<u \ \exists y<v \ \varphi(x,y,\bar{z})) ). \end{align*} $$Moreover, the strong $\Gamma $ -Collection scheme, denoted by $ \mathrm {B}^{+}\Gamma $ , consists of the formulas of the following form for every $\varphi \in \Gamma $ :$$ \begin{align*} \forall \bar{z},u \ \exists v \ \forall x<u \ (\exists y \ \varphi(x,y,\bar{z})\rightarrow \exists y<v \ \varphi(x,y,\bar{z})). \end{align*} $$It is folklore that $ \mathrm {I}\Sigma _{n+1}\vdash \mathrm {B}^{+}\Sigma _{n+1}\vdash \mathrm {B}\Sigma _{n+1} $ for all $ n\in \omega $ ; moreover, for every $ n\in \omega $ , $ \mathrm {I}\Sigma _{n}\nvdash \mathrm {B}\Sigma _{n+1} $ and $ \mathrm {I}\Sigma _{n}\nvdash \neg \mathrm {B}\Sigma _{n+1} $ (see [Reference Hájek and Pudlák6, Chapter I]). -
• Within $ \mathrm {I}\Delta _{0}+\mathrm {Exp}$ , the $ \Delta _{0}$ -formula $ x\mathrm {E} y $ denotes the Ackermann’s membership relation, asserting that “the x-th bit of the binary expansion of y is 1.” For every ${\mathcal {M}\models \mathrm {I}\Delta _{0}+\mathrm {Exp}} $ and each $ a\in M $ , $ a_{\mathrm {E}} $ denotes the set of $ \mathrm {E}$ -members of $ a $ in $\mathcal {M}$ . Moreover, the $ \Delta _{0}$ -formulas $ \mathrm {Card}(x)=y $ , $ \langle \bar {x}\rangle =y $ , $ \mathrm {Len}(x)=y $ , $ (x)_{y}=z $ , and $ x\upharpoonright _{y}=z, $ respectively, express that “there exists some bijection between y and the set coded by $x,$ ” “the sequence number of $\bar {x}$ is y,” “length of the sequence coded by x is $y,$ ” “the yth element of the sequence number x is z,” and “the restriction of the sequence number x to y is z.” In addition, for every formula $ \varphi (x) $ , by the formula $ y=\mu _{x}\ \varphi (x) $ we mean “y is the least element such that $\varphi (y)$ holds.”
Furthermore, for every $ n\in \omega $ there exist $ \mathcal {L}_{ A}$ -formulas $ \mathrm {Sat}_{\Sigma _{n}} $ and $ \mathrm {Sat}_{\Pi _{n}} $ which define the satisfaction predicate for $\Sigma _{n}$ -formulas and $ \Pi _{n}$ -formulas, respectively, in an ambient model. For every natural number $ n>0 $ , it can be shown that $ \mathrm {Sat}_{\Sigma _{n}} $ and $ \mathrm {Sat}_{\Pi _{n}} $ are $ \Sigma _{n} $ and $ \Pi _{n,} $ respectively, in $ \mathrm {I}\Sigma _{1} $ . Moreover, $ \mathrm {Sat}_{\Delta _{0}}\in \Delta _{1}^{\mathrm {I}\Sigma _{1}} $ [Reference Hájek and Pudlák6, Chapter I, Theorem 1.75]. If $\mathcal {M}$ is a nonstandard model of $ \mathrm {I}\Sigma _{n} $ , the aforementioned feature along with $ \Sigma _{n}$ -Overspill in $\mathcal {M}$ imply that every coded $ \Sigma _{n}$ -type and every coded bounded $ \Pi _{n}$ -type is realized in $\mathcal {M}$ .
-
• $ \Sigma _{n}$ -Pigeonhole Principle. For every $n>0$ , if $ \mathcal {M}\models \mathrm {I}\Sigma _{n}$ , $a\in M $ , and $\varphi $ is a $ \Sigma _{1}$ -formula which defines a function from $ a+1 $ into a in $\mathcal {M}$ , then $ \varphi $ is not one-to-one [Reference Hájek and Pudlák6].
-
• Given $ \mathcal {L}_{A}$ -structure $\mathcal {M}$ and subset X of M, for every $n>0$ , we define:
-
– $ \mathrm {K}^{n}(\mathcal {M};X):=$ the set of all $\Sigma _{n}$ -definable element of $\mathcal {M}$ with parameters from $ X $ .
-
– $ \mathrm {I}^{n}(\mathcal {M};X):=\{x: \ x\leq a \text { for some } a\in \mathrm {K}^{n}(\mathcal {M};X)\} $ .
-
– $ \mathrm {H}^{n}(\mathcal {M};X):=\bigcup _{k\in \omega }\mathrm {H}_{k}^{n}(\mathcal {M};X) $ , where:
$$ \begin{align*} &\quad \mathrm{H}_{0}^{n}(\mathcal{M};X):=\mathrm{I}^{n}(\mathcal{M};X), \text{ and } \\ &\mathrm{H}_{k+1}^{n}(\mathcal{M};X):=\mathrm{I}^{n}(\mathcal{M};\mathrm{H}_{k}^{n}(\mathcal{M};X)). \end{align*} $$ -
– ${\mathrm {K}(\mathcal {M};X):=\cup _{n\in \omega }\mathrm {K}^{n}(\mathcal {M};X)}$ .
(When $ X=\emptyset $ , we omit $ X $ from the notations.) Clearly, $\mathrm {I}^{n}(\mathcal {M};X) $ and $ \mathrm {H}^{n}(\mathcal {M};X) $ are initial segments of $\mathcal {M}$ . The following properties of these submodels of $\mathcal {M}$ are well-known (see, e.g., [Reference Hájek and Pudlák6, Chapter IV, Theorem 1.33]):
Theorem 2.1. Suppose $ n>0 $ , and $ \mathcal {M}\models \mathrm {I}\Sigma _{n} $ and $ X\subseteq M $ , then the following hold:
-
(1) $\mathrm {K}^{n}(\mathcal {M};X)\prec _{\Sigma _{n}}\mathcal {M} $ , and if $ \mathrm {K}^{n}(\mathcal {M}) $ is nonstandard, then $\mathrm {K}^{n}(\mathcal {M})\models \neg \mathrm {B}\Sigma _{n} $ .
-
(2) $\mathrm {I}^{n}(\mathcal {M};X)\prec _{\Sigma _{n-1}}\mathcal {M} $ and $\mathrm {I}^{n}(\mathcal {M};X)\models \mathrm {B}\Sigma _{n} $ .
-
(3) $\mathrm {H}^{n}(\mathcal {M};X)\prec _{\Sigma _{n}}\mathcal {M} $ and $\mathrm {H}^{n}(\mathcal {M};X)\models \mathrm {B}\Sigma _{n+1} $ .
-
-
• A given structure $\mathcal {M}$ is called recursively saturated if it realizes every recursive type with finite number of parameters in $ M $ . In [Reference Barwise, Schlipf, Saracino and Weispfenning2], Barwise and Shilipf showed that any countable model $\mathcal {M}$ of $\mathrm {PA}$ is recursively saturated iff it carries an inductive satisfaction class; here an inductive satisfaction class $ S $ of $\mathcal {M}$ is a subset of $ M $ which contains $ \langle \varphi ,a\rangle $ such that (1) $ \mathcal {M}\models \mathrm {Form}(\varphi ) $ , (2) $ (\mathcal {M}; S)\models \mathrm {PA}^{*} $ , and (3) $ (\mathcal {M}; S) $ satisfies Tarski’s inductive conditions for satisfaction (for a more precise definition see [Reference Kaye7]). Smoryński in [Reference Smoryński15], by generalizing Barwise–Ressayre expandability result, proved that for every countable recursively saturated model $\mathcal {M}$ of $\mathrm {PA}$ there exists some inductive satisfaction class S such that $(\mathcal {M}; S)$ is also recursively saturated.
-
• For every cut I of $\mathcal {M}$ the I-Standard System of $\mathcal {M}$ , denoted by $ \mathrm {SSy}_{I}(\mathcal {M}) $ , is the family of subsets of I of the form $ I\cap a_{\mathrm {E}} $ for some $ a\in M $ . By $ \mathrm {SSy}(\mathcal {M}) $ we mean $ \mathrm {SSy}_{\mathbb {N}}(\mathcal {M}) $ . It is well-known that for every model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{n} $ (for $ n>0 $ ), $ \mathrm {SSy}_{I}(\mathcal {M}) $ is equal to the family of intersections of $ \Sigma _{n}$ -definable (with parameters) subsets of $\mathcal {M}$ with I (see [Reference Hájek and Pudlák6, Chapter I]). Moreover, it is easy to check that if $ \mathcal {N}$ is an initial segment and a submodel of $\mathcal {M}$ containing I, then $\mathrm {SSy}_{I}(\mathcal {M})=\mathrm {SSy}_{I}(\mathcal {N}) $ (see [Reference Kaye7]).
-
• A given model $\mathcal {M}$ of $ \mathrm {I}\Delta _{0} $ is called 1-tall if $ \mathrm {K}^{1}(\mathcal {M};a) $ is cofinal in $\mathcal {M}$ for no $ a\in M $ ; and it is called 1-extendable if it possesses some end extension $ \mathcal {N}\models \mathrm {I}\Delta _{0} $ such that $ \mathrm {Th}_{\Sigma _{1}}(\mathcal {M})=\mathrm {Th}_{\Sigma _{1}}(\mathcal {N}) $ . Dimitracopoulos and Paris [Reference Dimitracopoulos and Paris3] showed that:
Theorem 2.2 (Dimitracopoulos–Paris).
-
(1) For any two countable and nonstandard models $\mathcal {M}$ and $ \mathcal {N} $ of $ \mathrm {I}\Delta _{0}+\mathrm {Exp} $ such that $\mathcal {M}$ is 1-extendable and $ \mathcal {N} $ is 1-tall, there exists a proper initial embedding from $\mathcal {M}$ into $ \mathcal {N} $ iff $ \mathrm {SSy}(\mathcal {M})=\mathrm {SSy}(\mathcal {N}) $ and $ \mathrm {Th}_{\Sigma _{1}}(\mathcal {M})\subseteq \mathrm {Th}_{\Sigma _{1}}(\mathcal {N}) $ .
-
(2) Any 1-tall countable model $\mathcal {M}$ of $ \mathrm {B}\Sigma _{1}+\mathrm {Exp} $ in which $ \mathbb {N} $ is not $ \Pi _{1} $ -definable (without parameters), is 1-extendable.
-
-
• A cut I of a model $\mathcal {M}$ is called strong if for every coded function $ f $ of $\mathcal {M}$ whose domain contains I, there exists some $ e>I $ such that $ f(i)\in I $ iff $ f(i)<e $ for all $ i\in I $ . Paris and Kirby, in [Reference Kirby and Paris9], proved that I is a strong cut of a model $\mathcal {M}$ of $ \mathrm {I}\Delta _{0}+\mathrm {Exp}$ iff $ (I,\mathrm {SSy}_{I}(\mathcal {M}))\models \mathrm {ACA}_{0} $ (here $ \mathrm {ACA}_{0} $ is the subsystem of second order arithmetic with the comprehension scheme restricted to formulas with no second order quantifiers).
-
• For given $ \mathcal {L}_{A}$ -structures $\mathcal {M}$ and $ \mathcal {N} $ , an (a proper) initial embedding j is an embedding from $\mathcal {M}$ into $ \mathcal {N} $ whose image is an (a proper) initial segment of $ \mathcal {N} $ . To every self-embedding $ j $ of $\mathcal {M}$ , we associate two subsets of M:
$$ \begin{align*} &\mathrm{I}_{\mathrm{fix}}(j):=\{m\in M:\forall x\leq m\ j(x)=x\}, \text{ and} \\ &\qquad \mathrm{Fix}(j):=\{m\in M:j(m)=m\}. \end{align*} $$In [Reference Bahrami and Enayat1], it is shown that for every model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ , and any self-embedding $ j $ of $\mathcal {M}$ , it holds that $K^{1}(\mathcal {M})\prec _{\Sigma _{1}} \mathrm {Fix}(j)\prec _{\Sigma _{1}}\mathcal {M}$ . Consequently, $ \mathrm {Fix}(j)\models \mathrm {I}\Delta _{0}+\mathrm {Exp}. $ The following results on the set of fixed points of initial self-embeddings were also proved in [Reference Bahrami and Enayat1]:Theorem 2.3 (B-Enayat).
Let $\mathcal {M}$ and $ \mathcal {N} $ be countable nonstandard models of $ \mathrm {I}\Sigma _{1} $ , $ c\in M $ and $d,b\in N $ , and let I be a proper cut shared by $\mathcal {M}$ and $ \mathcal {N} $ which is closed under exponentiation. Then the following are equivalent:
-
(1) There exists some proper initial embedding $ j $ from $\mathcal {M}$ into $ \mathcal {N} $ such that $ I\subseteq \mathrm {I}_{\mathrm {fix}}(j) $ , $ j(M)<b $ , and $ j(c)=d $ .
-
(2) $ \mathrm {SSy}_{I}(\mathcal {M})=\mathrm {SSy}_{I}(\mathcal {N}) $ , and for every $ \Delta _{0}$ -formula $ \delta (z,x,y) $ and every $ i\in I $ it holds that
$$ \begin{align*}\mathcal{M}\models\exists z \ \delta(z,c,i) \ \Rightarrow \ \mathcal{N}\models\exists z<b \ \delta(z,d,i) .\end{align*} $$
Remark 1. With the above assumptions, suppose $ a\in M\cap N $ such that for all $ \Delta _{0}$ -formula $ \delta $ and for every $ i\in I $ it holds that
$$ \begin{align*} \mathcal{M}\models\exists z \ \delta(z,c,(a)_{i}) \ \Rightarrow \ \mathcal{N}\models\exists z<b \ \delta(z,d,(a)_{i}). \end{align*} $$Then, by an appropriate modification in the proof of Theorem 2.3, we can manage to construct the above proper initial embedding j with the additional feature that $ {j((a)_{i})=(a)_{i}} $ for every $ i\in I $ .
Theorem 2.4 (B-Enayat).
Suppose $\mathcal {M}\models \mathrm {I}\Sigma _{1}$ is countable and nonstandard and I is a cut of $\mathcal {M}$ . Then the following hold:
-
(1) I is closed under exponentiation iff there exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {I}_{\mathrm {fix}}(j)=I $ .
-
(2) I is strong in $\mathcal {M}$ and $ I\prec _{\Sigma _{1}}\mathcal {M} $ , iff there exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)=I $ .
-
(3) $ \mathbb {N} $ is strong in $\mathcal {M}$ iff there exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)=\mathrm {K}^{1}(\mathcal {M}) $ .
The following lemma from [Reference Bahrami and Enayat1] will be useful in Section 4 of this paper:
Lemma 2.5. Suppose $ \mathcal {M}\models \mathrm {I}\Delta _{0}+\mathrm {Exp} $ in which $\mathbb {N}$ is not a strong cut, then for any self-embedding j of $\mathcal {M}$ , the following hold:
-
(1) The nonstandard fixed points of j are downward cofinal in the nonstandard part of $\mathcal {M}$ .
-
(2) For every element $a\in M$ , and $ m\in \mathrm {Fix}(j) $ there exists an element $b\in \mathrm {Fix} (j)$ such that
$$ \begin{align*} \mathrm{Th}_{\Sigma _{1}}(\mathcal{M};a,m)\subseteq \mathrm{Th}_{\Sigma _{1}}(\mathcal{M};b,m). \end{align*} $$
Convention. Suppose $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ and $ \langle \delta _{r}: \ r\in M\rangle $ is a canonical enumeration of all $ \Delta _{0}$ -formulas within $\mathcal {M}$ as in [Reference Hájek and Pudlák6, Chapter I]. To be more precise, $ \langle \delta _{r}: \ r\in M\rangle $ is an enumeration of all $ \Delta _{0} $ -formulas of $\mathcal {M}$ (containing nonstandard formulas) such that for every $ r\in M $ it holds that
-
• $ \delta _{r} $ is a standard $ \Delta _{0} $ -formula for all standard $ r\in M $ , and
-
• for every $ \Delta _{0} $ -formula $ \delta $ there exists some standard $ r\in M $ such that $\mathcal {M}$ believes that $ r $ is the Gödel number of $ \delta $ .
(For more details see [Reference Hájek and Pudlák6] or [Reference Kaye7].)
Now, for every $ r\in M $ , we define the following notations:
-
• $ f_{r}(\diamondsuit )=\blacklozenge $ denotes the following partial $ \Sigma _{1}$ -function in $\mathcal {M}$ :
$$ \begin{align*} \exists z ((z)_{0}=\blacklozenge \ \wedge \ z=\mu_{y}\mathrm{Sat}_{\Delta_{0}}(\delta_{r}(\diamondsuit,(y)_{0},(y)_{1})). \end{align*} $$ -
• The notation $ [f_{r}(\bar {x})\downarrow ] $ denotes the $ \Sigma _{1}$ -formula $ \exists z,y \ \mathrm {Sat}_{\Delta _{0}}(\delta _{r}(\bar {x},y,z)) $ , and ${ [f_{r}(\bar {x})\downarrow ]^{<w}} $ stands for the formula $ \exists z,y<w \ \mathrm {Sat}_{\Delta _{0}}(\delta _{r}(\bar {x},y,z)) $ .
-
• Let $ \mathcal {F}(\mathcal {M}) $ to be the collection of all $ \emptyset $ -definable partial $ \Sigma _{1}$ -functions in $\mathcal {M}$ .
First note that, it is shown in [Reference Bahrami and Enayat1] that
$$ \begin{align*} \mathrm{K}^{1}(\mathcal{M};a)=\{f(a): \ f\in\mathcal{F}(\mathcal{M}) \text{ and } \mathcal{M}\models[f(a)\downarrow] \}. \end{align*} $$Clearly, $ f_{r}\in \mathcal {F}(\mathcal {M}) $ for all standard $ r\in M $ . Moreover, by a little bit of effort we can show that for every $ f\in \mathcal {F}(\mathcal {M}) $ there exists some standard $ r\in M $ such that $ f=f_{r}$ (for details see [Reference Bahrami and Enayat1, Lemma 3.1.2]).As a result, if $\mathcal {M}$ and $ \mathcal {N} $ are two models of $ \mathrm {I}\Delta _{0} $ such that $ \mathrm {Th}_{\Sigma _{1}}(\mathcal {M})=\mathrm {Th}_{\Sigma _{1}}(\mathcal {N}) $ , then $\mathcal {F}(\mathcal {M})=\mathcal {F}(\mathcal {N})= \mathcal {F}:=\{f_{n}: n\in \mathbb {N}\} $ .
-
3 I-small $ \Sigma _{1}$ -elementary submodels
In [Reference Lascar, Kaye and Macpherson14], Lascar introduced a class of submodels of models of arithmetic, namely small submodels, which resemble those submodels of a model of set theory whose cardinality is less than the cardinality of the ground model. Then, Enayat inspired by a result of Schmerl (stated without proof as Theorem 5.7 in [Reference Kaye, Kossak and Kotlarski8]), generalized this notion in [Reference Enayat4]. In this section we will prove some results about these submodels.
Definition 1. For a given proper cut I of a model $\mathcal {M}$ of $ \mathrm {I}\Delta _{0}+\mathrm {Exp} $ , subset $ X $ of $ M $ is called I-small in $\mathcal {M}$ if there exists some $ a\in M $ such that $ X=\{(a)_{i} : \ i\in I\} $ , and $ (a)_{i}\neq (a)_{j} $ for all distinct $ i,j\in I $ . When $ I=\mathbb {N} $ , we simply use small for $ \mathbb {N}$ -small.
It is easy to see that for every model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ , each proper cut I of $\mathcal {M}$ is $ I$ -small. Moreover, for every $ a\in M $ , $ \mathrm {K}^{1}(\mathcal {M};a) $ is small in $\mathcal {M}$ . In [Reference Kossak and Schmerl12], it is shown that every recursively saturated model $ \mathcal {M}$ of $\mathrm {PA} $ possesses some small submodel which is not finitely generated. This result can be generalized for $ I$ -small submodels, when I is a strong cut of $\mathcal {M}$ (see Theorem 3.2). Furthermore, By using compactness arguments, for every model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ , we can find some elementary extension of $\mathcal {M}$ in which it is small. And finally, in [Reference Kossak and Kotlarski11] it is shown that every nonstandard small submodel is a mixed submodel (i.e., neither cofinal, nor initial segment). In a similar manner, for every cut I of a model $\mathcal {M}$ of $ \mathrm {I}\Sigma _{1} $ , and each $ I$ -small submodel $ \mathcal {M}_{0} $ of $\mathcal {M}$ , if $ I\subsetneq M_{0} $ then $ M_{0} $ is mixed in $ \mathcal {M.} $ (Since if $ M_{0}:=\{(a)_{i}: \ i\in I\} $ , and $ A:=\{i\in I: \ \mathcal {M}\models \neg i\mathrm {E}(a)_{i} \} $ , then $A\in \mathrm {SSy}_{I}(\mathcal {M})\setminus \mathrm {SSy}_{I}(\mathcal {M}_{0}) $ . So $ \mathcal {M}_{0} $ cannot be an initial segment of $ \mathcal {M.} $ )
In the following lemma we will show that in the definition of I-small, if I is a strong cut or it is equal to $ \mathbb {N} $ , then the condition $ (a)_{i}\neq (a)_{j} $ for all distinct $ i,j\in I $ , can be eliminated:
Lemma 3.1. Suppose $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ is nonstandard, $ I\subsetneq _{e}\mathcal {M} $ , $ \mathcal {M}_{0} $ is a submodel of $\mathcal {M}$ such that $ M_{0}=\lbrace (a)_{i}:\ i\in I\rbrace $ for some $ a\in M $ . Then the following hold:
-
(1) If $ I=\mathbb {N} $ , then $ \mathcal {M}_{0} $ is small.
-
(2) If I is strong in $\mathcal {M}$ , then $ \mathcal {M}_{0}$ is $ I$ -small.
Proof First, we will inductively define the following $ \Delta _{0}$ -function (with parameters) in $\mathcal {M}$ :
Note that by the way we defined $ g $ , its domain is an initial segment of $\mathcal {M}$ , and $ {\mathrm {Dom}(g)\leq \mathrm {Len}(a)} $ . Moreover, since I and $ M_{0} $ are not $ \Delta _{0}$ -definable in $\mathcal {M}$ , then $ I\subsetneq \mathrm {Dom}(g) $ . So by $ \Sigma _{1}$ -induction in $\mathcal {M}$ , we can find some $ d\in M $ such that $ (d)_{i}=g(i) $ for every $ i\in I $ . Clearly, $ (d)_{i}\neq (d)_{j} $ for every distinct $ i,j\in I $ , and $ {M_{0}\subseteq \lbrace (d)_{i}: \ i\in I\rbrace } $ . Now, in each case of the statement of the theorem we will prove that $ \lbrace (d)_{i}: \ i\in I\rbrace \subseteq M_{0} $ :
-
(1) Suppose $ I=\mathbb {N} $ . If $ \lbrace (d)_{n}: \ n\in \mathbb {N}\rbrace \nsubseteq M_{0} $ , then there exists the least number $ n\in \mathbb {N} $ such that $ (d)_{n}\notin M_{0} $ . So by the definition of $ g $ , there exist some $ m\in \mathbb {N} $ and some $ r\in M\setminus \mathbb {N} $ such that ${ (d)_{n-1}=(a)_{m}} $ and $ (d)_{n}=(a)_{r} $ . Therefore, by the definition of $ g $ , it holds that $ M_{0}=\lbrace (a)_{0},\ldots ,(a)_{m}\rbrace $ , which is a contradiction.
-
(2) In the general case with the extra assumption that I is strong in $\mathcal {M}$ , consider the following partial $ \Delta _{0}$ -function in $\mathcal {M}$ :
$$ \begin{align*} h(x):=\mu_{r}( (d)_{x}=(a)_{r}). \end{align*} $$Since I is strong and $ I\subseteq \mathrm {dom}(h) $ (because $ g $ is well-defined on I), there exists some $ e\in M $ such that $ h(i)\in I$ iff $ h(i)<e $ , for all $ i\in I $ . Moreover, by the definition of $ d$ , $g $ and $ h $ , for every $ i\in I $ it holds that $ (d)_{i}=(a)_{h(i)} $ . So it suffices to prove that $ h(i)<e $ for every $ i\in I $ . Suppose not; so there exists some $ i_{0}\in I $ which is the least element of $ M $ such that $ h(i_{0})>e $ . Now, by the way we defined $ g $ and $ h $ , it holds that$$ \begin{align*} \mathcal{M}\models\forall i<h(i_{0}) \ ((a)_{i}\neq(a)_{h(i_{0})}\wedge \exists j\leq h(i_{0}-1) ((a)_{i}=(a)_{j})). \end{align*} $$Therefore, $ M_{0}=\lbrace x\in M: \ \mathcal {M}\models \exists i\leq h(i_{0}-1) \ x=(a)_{i} \rbrace $ . So $ M_{0} $ is $ \Delta _{0}$ -definable in $\mathcal {M}$ , which is a contradiction.
In the following theorem, we will show that when I is strong, the basic properties which hold for small submodels, also hold for $ I$ -small ones.
Theorem 3.2. Let $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ be nonstandard, and let I be a strong cut of $\mathcal {M}$ . Then:
-
(1) For every $ a\in M $ , $ \mathrm {K}^{1}(\mathcal {M};I\cup \lbrace a\rbrace ) $ is $ I$ -small.
-
(2) If $ \mathcal {M}_{0} $ is an $ I$ -small submodel of $\mathcal {M}$ , then $ I\subseteq M_{0} $ .
-
(3) If $ \mathcal {M}\models \mathrm {PA} $ is countable and recursively saturated, then there exists some $ I$ -small elementary submodel of $\mathcal {M}$ which is not of the form of $ \mathrm {K}(\mathcal {M}; I\cup \lbrace a\rbrace ) $ for any $ a\in M $ .
Proof
-
(1) First fix some arbitrary $ s>I $ . So by using strong $ \Sigma _{1}$ -Collection in $\mathcal {M}$ for the formula $ \mathrm {Sat}_{\Delta _{0}}(\delta _{r}(i,a,z)) $ , we will find some $ b\in M $ such that
$$ \begin{align*} \mathcal{M}\models\forall \langle r,i\rangle<s \ ([f_{r}(i,a)\downarrow]\rightarrow[f_{r}(i,a)\downarrow]^{<b}). \end{align*} $$Then, by using $ \Sigma _{1}$ -induction we observe that $ \mathcal {M}\models \exists y \ \forall \langle r,i\rangle <s \ \varphi (y,r,i,a,b) $ , in which $ \varphi (y,r,i,a,b) $ is the following $ \Delta _{0}$ -formula:$$ \begin{align*} \left( \begin{array}{@{}c@{}} ([f_{r}(i,a)\downarrow]^{<b}\rightarrow(y)_{ \langle r,i\rangle}=f_{r}(i,a)) \ \wedge \ (\neg[f_{r}(i,a)\downarrow]^{<b}\rightarrow(y)_{ \langle r,i\rangle}=0) \end{array}\right). \end{align*} $$As a result, if $ d\in M $ is such that $ \mathcal {M}\models \forall \langle r,i\rangle <s \ \varphi (d,r,i,a,b) $ , then$$ \begin{align*} \mathrm{K}^{1}(\mathcal{M}; I\cup\lbrace a\rbrace)=\{(d)_{i}: \ i\in I \}. \end{align*} $$So by Lemma 3.1, $ \mathrm {K}^{1}(\mathcal {M}; I\cup \lbrace a\rbrace ) $ is $ I$ -small in $\mathcal {M}$ . -
(2) The exact argument used in [Reference Enayat4, Theorem 4.5.1] works here: let $ { M_{0}=\lbrace (a)_{i}: \ i\in I\rbrace } $ for some $ a\in M $ such that $ (a)_{i}\neq (a)_{j} $ for all distinct $ i, j\in I $ . Then put
$$ \begin{align*} { Z:=\lbrace \langle y,z\rangle\in M: \ \mathcal{M}\models (a)_{y}=z \rbrace }. \end{align*} $$Since $ Z $ is $ \Delta _{0}$ -definable in $\mathcal {M}$ , then $ X:= I\cap Z \in \mathrm {SSy}_{ I}(\mathcal {M}) $ . As a result, because $ I $ is strong in $\mathcal {M}$ , $ ( I; X)\models \mathrm {PA}^{*} $ . Now, suppose $ I\nsubseteq M_{0} $ . So $ ( I; X)\models \exists x \ (\forall y \ \langle y,x\rangle \notin X) $ . Let $ ( I, X)\models \textbf {x}_{0}:=\mu _{x} ( \forall y \ \langle y,x\rangle \notin X) $ . Therefore, $ \textbf {x}_{0}\notin M_{0} $ . So since $ \textbf {x}_{0}\neq 0 $ , and by the definition of $ \textbf {x}_{0} $ , we conclude that $ \textbf {x}_{0}-1\in M_{0} $ , which contradicts the fact that $ \mathcal {M}_{0} $ is a submodel of $\mathcal {M}$ . -
(3) We will generalize the method used in [Reference Kossak and Schmerl12, Proposition 2.10]: let $ S $ be a nonstandard partial inductive satisfaction class for $\mathcal {M}$ such that $ (\mathcal {M}; S) $ is recursively saturated. Put $ \mathcal {M}^{*}:=(\mathcal {M}; S) $ , and $ \mathcal {N}:=\mathrm {K}(\mathcal {M}^{*};I\cup \{s\}) $ for some $ s>I $ . First, note that $ N $ is $ I$ -small in $\mathcal {M}$ : since $ \mathcal {M}^{*} $ is a countable recursively saturated model of $ \mathrm {PA}^{*}$ , so it also possesses an inductive satisfaction class. Moreover, I is also strong in $ \mathcal {M}^{*} $ . Therefore, by repeating the argument used in the proof of part (1) of this theorem, and Lemma 3.1(2), we can show that $ N $ is $ I$ -small in $\mathcal {M}$ .
Moreover, on one hand, it is easy to see that $ S\cap N $ is a nonstandard satisfaction class for the $ \mathcal {L}_{A}$ -structure $ \mathcal {N} $ . So $ \mathcal {N} $ is also a recursively saturated model of $\mathrm {PA}$ . On the other hand, I is a proper initial segment of $ \mathcal {N} $ (because $ s>I $ ). Therefore, $ \mathcal {N} $ is of the form of $ \mathrm {K}(\mathcal {M};I\cup \{a\}) $ for no $ a\in M $ .
Remark 2. In part (2) of the above theorem, as the anonymous referee suggested, we do not need the whole strength of $ \mathrm {PA}^{*} $ ; it suffices that $ (I;X)\models \mathrm {I}\Sigma _{1} $ . As a result, if I is a semiregular cut of $\mathcal {M}$ , then part (2) of the above theorem holds (for the definition of a semiregular cut and their properties, see [Reference Kossak and Schmerl13]).
The following lemma will be useful in the proof of the main theorem of this section:
Lemma 3.3. Suppose $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ , $ I $ is a strong cut of $\mathcal {M}$ , and $ a\in M\setminus I $ such that $ (a)_{i}\neq (a)_{j} $ for all distinct $ i,j\in I $ . Moreover, let $ { M_{0}=\lbrace (a)_{i}: \ i\in I\rbrace } $ be a $ \Sigma _{1} $ -elementary submodel of $\mathcal {M}$ , $ X\subseteq M_{0}$ be coded in $\mathcal {M}$ , and $ i_{0}\in I $ such that $ i<i_{0}$ for all $ (a)_{i}\in X $ . Then $ X $ is coded in $ \mathcal {M}_{0} $ .
Proof Suppose $ \alpha \in M $ codes $ X $ in $\mathcal {M}$ . So $ \mathcal {M}\models \overset {\delta (\alpha ,\sigma ,i_{1})}{\overbrace {\alpha =\sum _{i<i_{1}}2^{(\sigma )_{i}}} }$ , in which $ i_{1}=\mathrm {Card}( X)\leq i_{0} $ and $ \sigma :=\langle x: \ x\mathrm {E}\alpha \rangle $ (so $ \mathrm {Len}(\sigma )=i_{1} $ ). Since $ \delta (x,y,z) $ is a $ \Delta _{0}$ -formula and $ \mathcal {M}_{0}\prec _{\Sigma _{1}}\mathcal {M} $ , it suffices to prove that $ \sigma \in M_{0} $ . For this purpose let $ Y:=\{i<i_{0}:\ \mathcal {M}\models (a)_{i}\mathrm {E}\alpha \} $ . Then there exists some $ \gamma \in I $ which codes $ Y $ .
Now, we define
Since $ I $ is strong in $\mathcal {M}$ , there exists some $ e $ such that $ h(i)>e $ iff $ h(i)>I $ , for all $ i\in I $ . We claim that $ \mathcal {M}\models \forall x \ \varphi (x,a,\gamma ,e) $ , where $\varphi (x,a,\gamma ,e) $ is the following $ \Delta _{0}$ -formula:
Therefore, $ \mathcal {M}\models \varphi (\sigma ,a,\gamma ,e) $ , which implies that $ \sigma =(a)_{c} $ for some $ c<\min \{e,\mathrm {Len}(a)\} $ . So $ \sigma =(a)_{h(\gamma )} $ and $ h(\gamma )<e $ , which implies that $ \sigma \in M_{_{0}} $ .
In order to prove the above claim, we will use $ \Delta _{0}$ -induction inside $\mathcal {M}$ : let $ \textbf {x}\in M $ such that $ {\mathcal {M}\models \varphi (w,a,\gamma ,e)} $ for every $ w<\textbf {x} $ , and $ \mathcal {M}\models \forall y<\mathrm {Len}(\textbf {x}) \ \exists z\mathrm {E}\gamma \ ((\textbf {x})_{y}=(a)_{z}) $ . So by induction hypothesis ${\mathcal {M}\models \textbf {x}\upharpoonright _{\mathrm {Len}(\textbf {x})-1}=(a)_{\textbf {z}} }$ for some $ \textbf {z}<\min \{e,\mathrm {Len}(a)\} $ . Then, we put $ Z:=\{i<\gamma : \ \mathcal {M}\models \exists y<\mathrm {Len}(\textbf {x})-1 \ (\textbf {x})_{y}=(a)_{i}\} $ , and let $ \textbf {z}_{0}\in I $ code $ Z $ . As a result, $ h(\textbf {z}_{0})\leq \textbf {z}<\min \{e,\mathrm {Len}(a)\} $ , which implies that $ \textbf {x}\upharpoonright _{\mathrm {Len}(\textbf {x})-1}=(a)_{h(\textbf {z}_{0})}\in M_{0} $ . So since $ \mathcal {M}_{0}\prec _{\Sigma _{1}}\mathcal {M} $ , then $ \textbf {x} $ is in $ M_{0} $ . Therefore, $ \textbf {x}=(a)_{i} $ for some $ i\in I<\min \{e,\mathrm {Len}(a)\} $ .
Now we are ready to prove the main theorem and corollary of this section. The method we use for proving Theorem 3.4 is a combination of the back-and-forth method used in [Reference Bahrami and Enayat1, Theorem 6.1] and [Reference Kaye, Kossak and Kotlarski8, Theorem 5.6].
Theorem 3.4. Assume $ \mathcal {N}\models \mathrm {I}\Sigma _{1} $ is countable and nonstandard, $ I $ is a strong cut of $ \mathcal {N} $ , and $ \mathcal {N}_{0} $ is an $ I$ -small $ \Sigma _{1} $ -elementary submodel of $ \mathcal {N} $ such that $ I\neq N_{0} $ . If $ \mathcal {M}:=\mathrm {H}^{1}(\mathcal {N};N_{0}) $ , then there exists some proper initial self-embedding $ j $ of $ \mathcal {M} $ such that $ { N_{0}=\mathrm {Fix}(j)} $ .
In order to prove Theorem 3.4, we will first prove the following lemmas:
Lemma 3.4.1. Suppose $\mathcal {N}$ , $N_{0} $ , $ I, $ and $\mathcal {M}$ are as in Theorem 3.4. Then I is strong in $\mathcal {M}$ and there exists some $ a\in M $ such that $ N_{0}=\{(a)_{i}: i\in I\} $ and $ (a)_{i}\neq (a)_{i} $ for distinct $ i,j\in I $ .
Proof By Theorem 2.1, $\mathcal {M}$ is a $ \Sigma _{1}$ -elementary initial segment of $ \mathcal {N} $ such that $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ . So it is easy to see that I is also strong in $\mathcal {M}$ . Moreover, since $ N_{0}\neq I $ , by using $ \Sigma _{1} $ -Overspill in $\mathcal {M}$ we can find the desired $ a\in M $ .
Lemma 3.4.2. Suppose $ \mathcal {N} $ , $ N_{0} $ , I, $ \mathcal {M,} $ and $ a\in M $ are as in Theorem 3.4 and Lemma 3.4.1. Moreover, let $ b\in M $ , $ \bar {u}:=u_{1},\ldots ,u_{n,} $ and $ \bar {v}:=v_{1},\ldots ,v_{n} <b $ be finite tuples in $ M $ . Then the following holds:
-
(i) For every $ m\in M $ there exists some $ \alpha \in M $ such that $ I\cap \alpha _{\mathrm {E}} $ equals to the following set:
$$ \begin{align*}\kern-27pt C:=\left\lbrace \langle r,i\rangle\in I: \ \mathcal{M}\models[f_{r}(\bar{u},m,(a)_{i})\downarrow] \ \text{and} \ f_{r}(\bar{u},m,(a)_{i})\notin\mathrm{K}^{1}(\mathcal{M}; N_{0}\cup\{\bar{u}\}) \right\rbrace.\end{align*} $$ -
(ii) For every $ m'\in M $ there exists some $ \alpha '\in M $ such that $ I\cap \alpha ^{\prime }_{\mathrm {E}} $ equals to the following set:
$$ \begin{align*}\kern-27pt C':=\lbrace\langle r,i\rangle\in I: \ \mathcal{M}\models[f_{r}(\bar{v},m',(a)_{i})\downarrow]^{<b} \ \text{and} \ f_{r}(\bar{v},m',(a)_{i})\notin\mathrm{K}^{1}(\mathcal{M}; N_{0}\cup\{\bar{v}\})\rbrace.\end{align*} $$
Proof We will prove part (i), and part (ii) will be proved similarly. Let
On one hand, since R is $\Pi _{1}$ -definable in $\mathcal {M}$ , then $ R\in \mathrm {SSy}_{ I}(\mathcal {M})$ . On the other hand, by Lemma 3.2(2), it holds that
Since I is strong in $\mathcal {M}$ , which implies that $ (I,\mathrm {SSy}_{I}(\mathcal {M}))\models \mathrm {ACA}_{0} $ , and because $ B $ is arithmetical in $ R $ and $ R\in \mathrm {SSy}_{ I}(\mathcal {M}) $ , we may deduce that $ B\in \mathrm {SSy}_{ I}(\mathcal {M}) $ , and consequently $ C\in \mathrm {SSy}_{ I}(\mathcal {M}) $ .
Lemma 3.4.3. Suppose $ \mathcal {N} $ , $ \mathrm {I} $ , $ N_{0} $ , $ \mathcal {M,} $ and $ a\in M $ are as in Theorem 3.4 and Lemma 3.4.1. Moreover, let $ b\in M $ , $ \bar {u}:=u_{1},\ldots ,u_{n,} $ and $ \bar {v}:=v_{1},\ldots ,v_{n} $ be finite tuples in $ M $ such that
$ \mathrm{P}(\bar{u},\bar{v})\equiv [f(\bar{u},(a)_{i})\downarrow]\rightarrow[f(\bar{v},(a)_{i})\downarrow]^{<b} , \text{ for all } f\in\mathcal{F} \text{ and } i\in I ; $
and
$ \mathrm{Q}(\bar{u},\bar{v})\equiv \left(\begin{array}{@{}c@{}} [f(\bar{u},(a)_{i})\downarrow]\wedge [f(\bar{v},(a)_{i})\downarrow]^{<b} \wedge\\ f(\bar{u},(a)_{i})\notin N_{0} \end{array}\right) \Rightarrow{f(\bar{u},(a)_{i})\neq f(\bar{v},(a)_{i})},\ \text{for all}\ f\in\mathcal{F} \text{ and } \text{all } i\in I. $
Then the following holds:
-
(i) If $ (a)_{\mathrm {i}}\in N_{0} $ and $ m\in M $ such that $ m\leq t(\bar {u},(a)_{\mathrm {i}}) $ for some $ t\in \mathcal {F} $ and $ \alpha \in M $ as in Lemma 3.4.2(i), then for every natural number $ k>0 $ and any $ (k+1) $ -many elements $ f,f_{n_{1}},\ldots ,f_{n_{k}} $ of $ \mathcal {F} $ and each $ z\in N_{0} $ there exist some $ s>I $ and some $ m'<b $ such that $\mathcal {M}\models \Psi (f,f_{n_{1}},\ldots ,f_{n_{k}},\bar {u},m,\bar {v},m',b,a,s,\alpha ,z,(a)_{\mathrm {i}})$ , where $\Psi $ is the following $\Pi _{1}$ -formula:
$$ \begin{align*} \kern-2pt m'\leq t(\bar{v},(a)_{\mathrm{i}}) \ \wedge \left(\begin{array}{@{}c@{}} \forall i<s([f(\bar{u},m,(a)_{i},z)\downarrow]\rightarrow[f(\bar{v},m',(a)_{i},z)\downarrow]^{<b}) \ \wedge \\ \forall i<s \bigwedge_{t\leq k}\left(\begin{array}{@{}c@{}} ([f_{n_{t}}(\bar{v},m',(a)_{i})\downarrow]^{<b} \wedge \langle n_{t},i\rangle\mathrm{E}\alpha)\rightarrow \\ f_{n_{t}}(\bar{u},m,(a)_{i})\neq f_{n_{t}}(\bar{v},m',(a)_{i}) \end{array}\right)\end{array}\right). \end{align*} $$ -
(ii) If $ m'<\max \{\bar {v}\} $ and $ \alpha '\in M $ as in Lemma 3.4.2(ii), then for every natural number $ {k>0} $ and any $ (k+1) $ -many elements $ f,f_{n_{1}},\ldots ,f_{n_{k}} $ of $ \mathcal {F} $ and each $ z\in N_{0} $ there exist some $ s>I $ and some $ m<\max \{\bar {u}\} $ such that $\mathcal {M}\models \Psi '(f,f_{n_{1}},\ldots ,f_{n_{k}},\bar {u},m,\bar {v},m',b,a,s,\alpha ,z)$ , where $\Psi '$ is the following $\Pi _{1}$ -formula:
$$ \begin{align*} \left(\begin{array}{@{}c@{}} \forall i<s(\neg[f(\bar{v},m',(a)_{i},z)\downarrow]^{<b}\rightarrow\neg[f(\bar{u},m,(a)_{i},z)\downarrow]) \ \wedge \\ \forall i<s \bigwedge_{t\leq k}\left(\begin{array}{@{}c@{}} (\langle n_{t},i\rangle\mathrm{E}\alpha'\wedge[f_{n_{t}}(\bar{u},m,(a)_{i})\downarrow]) \rightarrow \\ f_{n_{t}}(\bar{u},m,(a)_{i})\neq f_{n_{t}}(\bar{v},m',(a)_{i}) \end{array}\right)\end{array}\right). \end{align*} $$
Proof
-
(i) First note that since $ \mathrm {P}(\bar {u},\bar {v}) $ holds in $\mathcal {M}$ , Theorem 2.3 and Remark 1 imply that
-
(1): There exists some initial self-embedding $ j_{0} $ of $\mathcal {M}$ such that $ j_{0}( M)<b $ , $ j_{0}(\bar {u})=\bar {v} $ , and $ N_{0}\subseteq \mathrm {Fix}(j_{0}) $ .
Now, suppose that part (i) of this lemma does not hold; i.e., there is some $ k>0 $ for which there exist $ (k+1) $ -many elements $ \lbrace f,f_{n_{1}},\ldots ,f_{n_{k}}\rbrace $ of $ \mathcal {F} $ , and some $ \textbf {z}\in N_{0} $ such that for all $ s> I $ it holds that
$$ \begin{align*}\mathcal{M}\models\forall y<b \ \neg \Psi(f,f_{n_{1}},\ldots ,f_{n_{k}},\bar{u},m,\bar{v},y,b,a,s,\alpha,\textbf{z},(a)_{\mathrm{i}}). \end{align*} $$Therefore, by $ \Sigma _{1}$ -Underspill in $\mathcal {M}$ , there exists some $ s\in I $ such that
$$ \begin{align*}\mathcal{M}\models \forall y<b \ \neg\Psi(f,f_{n_{1}},\ldots ,f_{n_{k}},\bar{u},m,\bar{v},y,b,a,s,\alpha,\textbf{z},(a)_{\mathrm{i}}).\end{align*} $$ -
(2): Let $ k_{0}\in \mathbb {N} $ be the least natural number, for which there exists a set $\lbrace f, f_{n_{1}},\ldots ,f_{n_{k_{0}}}\rbrace $ of elements of $ \mathcal {F} $ , some $ \textbf {z}_{0}\in N_{0} $ , and some $ s_{0}\in I $ such that
$$ \begin{align*} {\mathcal{M}\models \forall y<b \ \neg\Psi(f,f_{n_{1}},\ldots ,f_{n_{k_{0}}}, \bar{u},m,\bar{v},y,b,a,s_{0},\alpha,\textbf{z}_{0},(a)_{\mathrm{i}})}. \end{align*} $$Put
$$ \begin{align*} &\quad X:=\lbrace x\in M: \ \mathcal{M}\models\exists i<s_{0}(x=(a)_{i}\wedge [f(\bar{u},m,(a)_{i}, \textbf{z}_{0})\downarrow] ) \rbrace \\ &\qquad\qquad\qquad\qquad\qquad\qquad \quad \text{and}\\ & \kern-12pt X':=\lbrace \langle n,x\rangle\in M: \ \mathcal{M}\models\exists i<s_{0}(x=(a)_{i}\wedge \bigvee\nolimits_{t=1}^{\kern2pt k_{0}} n=n_{t} \wedge \langle n,i\rangle\mathrm{E}\alpha) \rbrace. \end{align*} $$By Lemma 3.3, there exist $ (a)_{\xi }\in N_{0} $ and $ (a)_{\zeta }\in N_{0} $ which code $ X $ and $ X', $ respectively. So we can restate statement (2) in the following form: -
(3): Let $ k_{0}\in \mathbb {N} $ be the least natural number, for which there exists a set $\lbrace f, f_{n_{1}},\ldots ,f_{n_{k_{0}}}\rbrace $ of elements of $ \mathcal {F} $ , some $ \textbf {z}_{0},(a)_{\zeta },(a)_{\xi }\in N_{0} $ such that
$$ \begin{align*} \kern-40pt\mathcal{M}\models\forall y\leq t(\bar{v},(a)_{\mathrm{i}}) \left( \begin{array}{@{}c@{}} \forall \epsilon<(a)_{\xi}( \epsilon\mathrm{E}(a)_{\xi}\rightarrow[f(\bar{v},y,\epsilon,\textbf{z}_{0})\downarrow]^{<b} \rightarrow \\ \exists \varepsilon<\mathrm{E}(a)_{\zeta} \bigvee_{t=1}^{k_{0}} \left(\begin{array}{@{}c@{}} \langle n_{t},\varepsilon\rangle\mathrm{E}(a)_{\zeta} \ \wedge [f_{n_{t}}(\bar{v},y,\varepsilon)\downarrow]^{<b} \wedge \\ f_{n_{t}}(\bar{u},m,\varepsilon)= f_{n_{t}}(\bar{v},y,\varepsilon) \end{array}\right) \end{array}\right).\end{align*} $$Our plan is to consider two cases $ {k_{0}=1} $ and $ {k_{0}>1} $ , and in each case obtain a contradiction. But before dividing the cases we will define some $ \Sigma _{1} $ -functions which will help us in achieving the desired contradictions.First, by considering the code of the sequence ${ \langle f_{n_{t}}(\bar {u},m,\varepsilon ): \, \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } \rangle } $ in $\mathcal {M}$ , we may quantify out $ f_{n_{t}}(\bar {u},m,\varepsilon ) $ s from the formula in statement $ (3) $ ; in other words, there exists some $ \textbf {x}\in M $ such that $ (\textbf {x})_{\langle n_{t},\varepsilon \rangle }=f_{n_{t}}(\bar {u},m,\varepsilon ) $ for every $\langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } $ . So we will deduce that
-
(4): $ \mathcal {M}\models \exists x \forall y\leq t(\bar {v},(a)_{\mathrm {i}}) \ \theta (y,b,\bar {v},x,(a)_{\xi },(a)_{\zeta },\textbf {z}_{0}) $ , where $\theta (y,b,\bar {v},x, (a)_{\xi },(a)_{\zeta },\textbf {z}_{0})$ is the following $ \Delta _{0} $ -formula:
$$ \begin{align*} \left( \begin{array}{@{}c@{}} \forall \epsilon<(a)_{\xi}( \epsilon\mathrm{E}(a)_{\xi} \rightarrow[f(\bar{v},y,\epsilon,\textbf{z}_{0})\downarrow]^{<b}) \rightarrow\\ \exists \langle n_{t},\varepsilon\rangle\mathrm{E}(a)_{\zeta} \left(\begin{array}{@{}c@{}} [f_{n_{t}}(\bar{v},y,\varepsilon)\downarrow]^{<b}\wedge (x)_{\langle n_{t},\varepsilon\rangle}= f_{n_{t}}(\bar{v},y,\varepsilon) \end{array}\right)\end{array}\right). \end{align*} $$Then, we will define $ \Sigma _{1}$ -definable partial functions $ b(\diamondsuit ,y,(a)_{\xi },(a)_{\zeta },\textbf {z}_{0}) $ and $g(\diamondsuit ,(a)_{\xi },(a)_{\zeta },\textbf {z}_{0},(a)_{\mathrm {i}}) $ , as follows (we omit the parameters $(a)_{\xi } $ , $ (a)_{\zeta } $ , $ (a)_{\mathrm {i}} $ , and $ \textbf {z}_{0} $ in the presentations of these functions for the sake of simplicity):-
– $b(\diamondsuit ,y):=\min \left \lbrace w: \left (\begin{array}{@{}c@{}} \forall \epsilon \mathrm {E}(a)_{\xi } \ ([f(\diamondsuit ,y,\epsilon ,\textbf {z}_{0})\downarrow ]^{<w})\wedge \\ \forall \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } \ ([f_{n_{t}}(\diamondsuit ,y,\varepsilon )\downarrow ]^{<w}) \end {array}\right ) \right \rbrace. $
-
– $ g(\diamondsuit ):=x $ iff $ \exists z \left (\begin{array}{@{}c@{}}(z)_{0}=x \ \wedge \\ z=\mu _{w} \ \forall y\leq t(\diamondsuit ,(a)_{\mathrm {i}}) \ \left (\begin{array}{@{}c@{}}[b(\diamondsuit ,y)\downarrow ]^{<(w)_{1}} \ \rightarrow \\ \theta (y,b(\diamondsuit ,y),\diamondsuit ,(w)_{0},(a)_{\xi },(a)_{\zeta },\textbf {z}_{0})\end {array}\right )\end {array}\right ); $ and $ g_{t}(\diamondsuit ,\varepsilon ):=(g(\diamondsuit ))_{\langle n_{t},\varepsilon \rangle } $ , for every $ \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } $ .
From the definition of $ g_{t}(\bar {v},\varepsilon ) $ s and statement (4) we may infer that
-
-
(5): $ \mathcal {M}\models \forall y\leq t(\bar {v},(a)_{\mathrm {i}}) \left ( \begin{array}{@{}c@{}} ([b(\bar {v},y)\downarrow ]^{<b} \ \wedge \ \forall \epsilon <(a)_{\xi }( \epsilon \mathrm {E}(a)_{\xi }\rightarrow [f(\bar {v},y,\epsilon ,\textbf {z}_{0})\downarrow ]^{<b(\bar {v},y)})) \rightarrow \\ \exists \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } \left (\begin{array}{@{}c@{}} [f_{n_{t}}(\bar {v},y,\varepsilon )\downarrow ]^{<b(\bar {v},y)} \wedge [g_{t}(\bar {v},\varepsilon )\downarrow ]^{<b(\bar {v},y)} \ \wedge \\ g_{t}(\bar {v},\varepsilon )= f_{n_{t}}(\bar {v},y,\varepsilon ) \end {array}\right ) \end {array}\right )$ . It is not difficult to express the formula in the statement (5) in the form of $ {\forall z<b \ \delta (\bar {v},(a)_{\xi },(a)_{\zeta },\textbf {z}_{0})} $ for some $ \Delta _{0}$ -formula $ \delta $ . Therefore, by the property $ \mathrm {P}(\bar {u},\bar {v}) $ , the definition of function $ s $ , and statement (5) we deduce that
-
(6): $\mathcal {M}\models \forall y\leq t(\bar {u},(a)_{\mathrm {i}}) \left ( \begin{array}{@{}c@{}} ([b(\bar {u},y)\downarrow ] \ \wedge \ \forall \epsilon <(a)_{\xi }( \epsilon \mathrm {E}(a)_{\xi }\rightarrow [f(\bar {u},y,\epsilon ,\textbf {z}_{0})\downarrow ]^{<b(\bar {u},y)}) )\rightarrow \\ \exists \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta } \left (\begin{array}{@{}c@{}} [f_{n_{t}}(\bar {u},y,\varepsilon )\downarrow ]^{<b(\bar {u},y)} \wedge [g_{t}(\bar {u},\varepsilon )\downarrow ]^{<b(\bar {u},y)} ) \ \wedge \\ g_{t}(\bar {u},\varepsilon )= f_{n_{t}}(\bar {u},y,\varepsilon ) \end {array}\right )\end {array}\right )$ .
Finally, we will simultaneously define two more $ \Sigma _{1}$ -definable functions in $\mathcal {M}$ :
$$ \begin{align*} &\langle o(\diamondsuit,y),h(\diamondsuit,y)\rangle\\ & \quad := \min\left\lbrace\langle n_{t},\varepsilon\rangle \mathrm{E}(a)_{\zeta}: \left(\begin{array}{@{}c@{}} [b(\diamondsuit,y)\downarrow]\wedge\\ {}[ f_{n_{t}} ({\diamondsuit}, y, {\varepsilon}){\downarrow}]^{<b({\diamondsuit},y)} \ \wedge [g_{t} ({\diamondsuit},\varepsilon) {\downarrow}]^{< b (\diamondsuit,y)} \wedge\\ g_{t}(\diamondsuit,\varepsilon)= f_{n_{t}}(\diamondsuit,y,\varepsilon)\end{array}\right) \right\rbrace. \end{align*} $$(Note that, similar to the way we defined function g, we can express the above definition by a $\Sigma _{1}$ -formula.) Then, by statement (5) it holds that -
(7): $ \mathcal {M}\models \forall y\leq t(\bar {v},(a)_{\mathrm {i}}) \left (\begin{array}{@{}c@{}} ([b(\bar {v},y)\downarrow ]^{<b} \ \wedge \ \forall \epsilon <(a)_{\xi }( \epsilon \mathrm {E}(a)_{\xi }\rightarrow [f(\bar {v},y,\epsilon , \textbf {z}_{0})\downarrow ]^{<b})) \rightarrow\\ {}[\langle o(\bar {v},y),h(\bar {v},y)\rangle \downarrow ]\end {array}\right ) $ .
Similarly, from statement (6) we may deduce that
-
(8): $\mathcal {M}\models \forall y\leq t(\bar {u},(a)_{\mathrm {i}})\left (\begin{array}{@{}c@{}} ([b(\bar {u},y)\downarrow ] \ \wedge \ \forall \epsilon <(a)_{\xi }( \epsilon \mathrm {E}(a)_{\xi }\rightarrow [f(\bar {u},y,\epsilon ,\textbf {z}_{0})\downarrow]^{<b(\bar {u},y)})) \rightarrow \\ {}[\langle o(\bar {u},y),h(\bar {u},y)\rangle \downarrow] \end {array}\right ) $ .
Now, we are ready to examine the mentioned underlined cases for $ k_{0} $ :
-
– If $ k_{0}>1$ : By using Lemma 3.3, let $ (a)_{\rho }\in N_{0} $ be the code of the following subset of $ N_{0} $ :
$$ \begin{align*} \kern-56pt A:=\left\lbrace\langle o(\bar{v},y),h(\bar{v},y)\rangle: \ \mathcal{M}\models \left(\begin{array}{@{}c@{}} y\leq t(\bar{v},(a)_{\mathrm{i}}) \wedge [\langle o(\bar{v},y),h(\bar{v},y)\rangle\downarrow]^{<b} \wedge\\ \forall \epsilon<(a)_{\xi}( \epsilon\mathrm{E}(a)_{\xi}\rightarrow[f(\bar{v},y,\epsilon,\textbf{z}_{0})\downarrow]^{<b} \wedge\\ \exists \varepsilon<(a)_{\zeta} \left(\begin{array}{@{}c@{}} \langle n_{1},\varepsilon\rangle\mathrm{E}(a)_{\zeta} \wedge [f_{n_{1}}(\bar{v},y,\varepsilon)\downarrow]^{<b} \wedge \\ f_{n_{1}}(\bar{v},y,\varepsilon)=f_{n_{1}}(\bar{u},m,\varepsilon) \end{array}\right) \end{array}\right) \right\rbrace .\end{align*} $$
So, by statements (3), (7), and the definition of $ (a)_{\rho } $ , we conclude that
-
-
(9): $\mathcal {M}\models \forall y\leq t(\bar {v},(a)_{\mathrm {i}}) \left ( \begin{array}{@{}c@{}} \forall \epsilon <(a)_{\xi }\left ( \begin{array}{@{}c@{}} \epsilon \mathrm {E}(a)_{\xi }\rightarrow [ f(\bar {v},y,\epsilon ,\textbf {z}_{0})\downarrow]^{<b} \wedge \\ {}[\langle o(\bar {v},y),h(\bar {v},y)\rangle \downarrow]^{<b} \wedge \\ \neg \langle o(\bar {v},y),h(\bar {v},y)\rangle \mathrm {E}(a)_{\rho } \end {array}\right ) \rightarrow \\ \exists \varepsilon <\mathrm {E}(a)_{\zeta }\bigvee _{t=2}^{k_{0}} \left (\begin{array}{@{}c@{}} \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\zeta }\wedge [f_{n_{t}}(\bar {v},y,\varepsilon )\downarrow ]^{<b}\wedge \\ f_{n_{t}}(\bar {u},m,\varepsilon )= f_{n_{t}}(\bar {v},y,\varepsilon ) \end {array}\right ) \end {array}\right )$ .
Let $ f'\in \mathcal {F} $ such that
$$ \begin{align*} &\qquad\qquad f'(\diamondsuit,y,\epsilon,\langle\textbf{z}_{0},(a)_{\rho},(a)_{\zeta},(a)_{\xi}\rangle)=x \\ &\qquad\qquad\qquad\qquad\text{iff} \\ &x=f(\diamondsuit,y,\epsilon,\textbf{z}_{0})\ \wedge [\langle o(\diamondsuit,y),h(\diamondsuit,y)\rangle\downarrow] \wedge \neg \langle o(\diamondsuit,y),h(\diamondsuit,y)\rangle\mathrm{E}(a)_{\rho}. \end{align*} $$So by considering $ f' $ instead of $ f $ in statement (3) and $ \langle \textbf {z}_{0},(a)_{\rho },(a)_{\zeta },(a)_{\xi }\rangle $ instead of $ \textbf {z}_{0} $ , statement (9) leads to contradiction with the minimality of $ k_{0} $ .-
– If $ k_{0}=1 $ : In this case our plan for obtaining a contradiction is as follows: on one hand, by the definitions of $ h(\bar {u},y) $ and $ (a)_{\zeta } $ , $\mathcal {M}$ thinks that the cardinality of $ \{h(\bar {u},y): \ \mathcal {M}\models (y<t(\bar {u},(a)_{\mathrm {i}})\ \wedge \ [h(\bar {u},y)\downarrow ])\} $ is at most $ s_{0} $ . On the other hand, for every $ i\in I $ we will inductively define some $ \Sigma _{1} $ -functions, namely $ w(\bar {u},i) $ s, such that $ \mathcal {M}\models [w(\bar {u},i)\downarrow ]^{<t(\bar {u},(a)_{\mathrm {i}})} $ and $\mathcal {M}$ believes that there is a bijection between members of $ \{h(\bar {u},w(\bar {u},i)): \ i\in I \text { and } \mathcal {M}\models [h(\bar {u},w(\bar {u},i))\downarrow ]\} $ and the elements of I. So the contradiction is achieved. To be more precise, we define
$$ \begin{align*} \kern-56pt w(\diamondsuit,0):=\min\left\lbrace y\leq t(\diamondsuit,(a)_{\mathrm{i}}): \ \left(\begin{array}{@{}c@{}}[b(\diamondsuit,y)\downarrow]\ \wedge \ [h(\diamondsuit,y)\downarrow]^{<(a)_{\zeta}} \wedge \\ \forall \epsilon<(a)_{\xi}( \epsilon\mathrm{E}(a)_{\xi}\rightarrow[f(\diamondsuit,y,\epsilon,\textbf{z}_{0})\downarrow]^{<b(\diamondsuit,y)} ) \end{array}\right) \right\rbrace, \end{align*} $$and$w(\diamondsuit ,i+1):= \min \left \lbrace y\leq t(\diamondsuit ,(a)_{\mathrm {i}}): \varphi (\diamondsuit ,i,y,(a)_{\zeta },(a)_{\xi },\textbf {z}_{0})\right \rbrace $ , where $ \varphi (\diamondsuit ,i,y,(a)_{\zeta },(a)_{\xi },\textbf {z}_{0}) $ is the following formula:
$$ \begin{align*} \kern-24pt \left(\begin{array}{@{}c@{}} [b(\diamondsuit,y)\downarrow]\ \wedge [h(\diamondsuit,y)\downarrow]^{<(a)_{\zeta}} \wedge \\ \forall \epsilon<(a)_{\xi}( \epsilon\mathrm{E}(a)_{\xi}\rightarrow [f(\diamondsuit,y,\epsilon,\textbf{z}_{0})\downarrow]^{<b(\diamondsuit,y)}) \ \wedge \\ \forall x\leq i \left(\begin{array}{@{}c@{}} \left(\begin{array}{@{}c@{}} [h(\diamondsuit,w(\diamondsuit,x))\downarrow]^{<(a)_{\zeta}}\wedge\\ {}[f_{n_{1}}(\diamondsuit,y,h(\diamondsuit,w(\diamondsuit,x)))\downarrow]^{<b(\diamondsuit,y)}\wedge\\ {}[f_{n_{1}}(\diamondsuit,w(\diamondsuit,x),h(\diamondsuit,w(\diamondsuit,x)))\downarrow]^{<b(\diamondsuit,w(\diamondsuit,x))}\end{array}\right) \rightarrow \\ f_{n_{1}}(\diamondsuit,y,h(\diamondsuit,w(\diamondsuit,x)))\neq f_{n_{1}}(\diamondsuit,w(\diamondsuit,x),h(\diamondsuit,w(\diamondsuit,x))) \end{array}\right)\end{array}\right). \end{align*} $$
First, we will show that $ \mathcal {M}\models [w(\bar {u},i)\downarrow ] $ for all $ i\in I $ . Otherwise, there exists the least $ 0<i_{0}\in I $ such that
-
-
(10): $ \mathcal {M}\models \forall y\leq t(\bar {u},(a)_{\mathrm {i}}) \ \neg \varphi (\bar {u},i_{0},y,(a)_{\zeta },(a)_{\xi },\textbf {z}_{0})$ .
Note that by the definition of $ (a)_{\xi } $ and $ (a)_{\zeta } $ it holds that
-
(11): $\mathcal {M}\models ([b(\bar {u},m)\downarrow ] \ \wedge \ \forall \epsilon <(a)_{\xi }( \epsilon \mathrm {E}(a)_{\xi }\rightarrow [f(\bar {u},m,\epsilon ,\textbf {z}_{0})\downarrow ]^{<b(\bar {u},m)}) $ .
So by statements (8), (10), and (11), there exists some $ i_{1}< i_{0} $ such that
-
(12): $\mathcal {M}\models \left (\begin{array}{@{}c@{}} [h(\bar {u},w(\bar {u},i_{1}))\downarrow] \wedge [f_{n_{1}}(\bar {u},m,h(\bar {u},w(\bar {u},i_{1})))\downarrow] \wedge \\ {}[f_{n_{1}}(\bar {u},w(\bar {u},i_{1}),h(\bar {u},w(\bar {u},i_{1})))\downarrow] \wedge \\ f_{n_{1}}(\bar {u},m,h(\bar {u},w(\bar {u}, i_{1})))= f_{n_{1}}(\bar {u},w(\bar {u},i_{1}),h(\bar {u},w(\bar {u},i_{1}))) \end {array}\right ) $ .
Clearly, $f_{n_{1}}(\bar {u},w(\bar {u},i_{1}),h(\bar {u},w(\bar {u},i_{1})))\in \mathrm {K}^{1}(\mathcal {M};N_{0}\cup \{\bar {u}\}) $ . So by statement (12), $f_{n_{1}}(\bar {u},m,h(\bar {u},w(\bar {u},i_{1}))) \in \mathrm {K}^{1}(\mathcal {M};M_{0}\cup \{\bar {u}\}) $ . So $ \mathcal {M}\models \neg \langle n_{1},h(\bar {u},w(\bar {u},i_{1}))\rangle \mathrm {E}(a)_{\zeta } $ (by the definition of $ (a)_{\zeta } $ ), which is in contradiction with the definition of the function $ h $ .
Then, by the definition of $ w(\bar {u},i) $ s and statement (8), the function $ {i\mapsto h(\bar {u},w(\bar {u},i))} $ from $\{i: \ i\leq s_{0}+1\} $ into $ ((a)_{\zeta })_{\mathrm {E}} $ is well-defined and coded in $\mathcal {M}$ . So, since the cardinality of $ (a)_{\zeta } $ is less than $ s_{0}+1 $ , by $ \Sigma _{1}$ -Pigeonhole Principle in $\mathcal {M}$ , there exists some distinct $ i_{0}<i_{1}\leq s_{0}+1 $ such that
-
(13): $\mathcal {M}\models h(\bar {u},w(\bar {u},i_{0}))=h(\bar {u},w(\bar {u},i_{1}))$ .
Therefore, by statement (13) and the definition of $ h $ we conclude that
-
(14): $ \mathcal {M}\models \left (\begin{array}{@{}c@{}} [g_{1}(\bar {u},h(\bar {u},w(\bar {u},i_{0})))\downarrow ] \wedge [g_{1}(\bar {u},h(\bar {u},w(\bar {u},i_{1})))\downarrow ]\wedge \\ g_{1}(\bar {u},h(\bar {u},w(\bar {u},i_{0})))=g_{1}(\bar {u},h(\bar {u},w(\bar {u},i_{1}))) \end {array}\right ) $ .
Moreover, by the definition of $ h(\bar {u},w(\bar {u},i)) $ , for $ i=i_{0},i_{1} $ it holds that
-
(15): $\mathcal {M}\models \left (\begin{array}{@{}c@{}} [f_{n_{1}}(\bar {u},w(\bar {u},i),h(\bar {u},w(\bar {u},i)))\downarrow ] \ \wedge [g_{1}(\bar {u},h(\bar {u},w(\bar {u})))\downarrow ] \wedge \\ f_{n_{1}}(\bar {u},w(\bar {u},i),h(\bar {u},w(\bar {u},i)))=g_{1}(\bar {u},h(\bar {u},w(\bar {u}))) \end {array}\right )$ .
So statements (13–15) imply that
-
(16): $\mathcal {M}\models f_{n_{1}}(\bar {u},w(\bar {u},i_{1}),h(\bar {u},w(\bar {u},i_{0})))=f_{n_{1}}(\bar {u}, w(\bar {u},i_{0}),h(\bar {u},w(\bar {u},i_{0}))) $ .
But statement (16) is in contradiction with the definition of $ w(\bar {u},i_{1}) $ .
-
(ii) Part (ii) of this lemma will be proved exactly like part (i), except we need an extra statement between statements (3) and (4): by using $ \Sigma _{1}$ -Collection, we deduce that
-
(3’): $\mathcal {M}\models \exists w \forall x<u_{0} \left (\begin{array}{@{}c@{}} \forall \epsilon <(a)_{\lambda } \ ([f(\bar {u},x,\epsilon ,\textbf {z}_{0})\downarrow ]^{<w}\rightarrow \epsilon \mathrm {E}(a)_{\lambda })\rightarrow \\ \exists \varepsilon <(a)_{\eta } \bigvee _{t\leq k_{1}}\left (\begin{array}{@{}c@{}} \langle n_{t},\varepsilon \rangle \mathrm {E}(a)_{\eta }\wedge [f_{n_{t}}(\bar {u},x,\varepsilon )\downarrow ]^{<w}\wedge \\ f_{n_{t}}(\bar {u},x,\varepsilon )= f_{n_{t}}(\bar {v},m',\varepsilon ) \end {array}\right ) \end {array}\right ) $ ;
in which $(a)_{\lambda }\in N_{0} $ and $ (a)_{\eta }\in N_{0} $ code the following Y and $Y',$ respectively:
$$ \begin{align*} &Y:=\lbrace x\in M: \ \mathcal{M}\models\exists i<s_{0}(x=(a)_{i}\wedge[f(\bar{v},m',(a)_{i},\textbf{z}_{0})\downarrow]^{<b} ) \rbrace, \text{ and } \\ &Y':=\lbrace \langle n,x\rangle\in M: \ \mathcal{M}\models\exists i<s_{0}(x=(a)_{i}\wedge \bigvee_{t=0}^{k_{0}}n=n_{t} \wedge \langle n,i\rangle\mathrm{E}\alpha') \rbrace. \end{align*} $$
-
Now we are ready to prove Theorem 3.4.
Proof of Theorem 3.4
By Lemma 3.4.1, there exists some $ a\in M $ such that $ {N_{0}=\{(a)_{i}: i\in I\}} $ and $ (a)_{i}\neq (a)_{i} $ for distinct $ i,j\in I $ . Now, in order to construct $ j $ , first by using strong $ \Sigma _{1}$ -Collection in $\mathcal {M}$ , we will find some $ b\in M $ such that
Then, by using back-and-forth method we will inductively build finite functions $\bar {u}\mapsto \bar {v}$ such that $ \bar {u},\bar {v}\in M $ , and $\mathcal {M}\models (\bar {v}<b \ \wedge \ \mathrm {P}(\bar {u},\bar {v}) \ \wedge \mathrm {Q}(\bar {u},\bar {v}) )$ , where $ \mathrm {P}(\bar {u},\bar {v}) $ and $ \mathrm {Q}(\bar {u},\bar {v}) $ are properties as stated in Lemma 3.4.3. Through the “forth” stages of back-and-forth we shall make the domain of $ j $ to be equal to $ M $ , and “back” stages are for making the range of $ j $ to be an initial segment of $\mathcal {M}$ . For the first step of induction, we will choose $0\mapsto 0 $ . Then, suppose $\bar {u}\mapsto \bar {v}$ is built such that $\mathcal {M}\models (\bar {v}<b \ \wedge \ \mathrm {P}(\bar {u},\bar {v}) \ \wedge \mathrm {Q}(\bar {u},\bar {v}) )$ .
“Forth” stages: Let $ m\in M\setminus \{\bar {u}\} $ be arbitrary. By the definition of $\mathcal {M}$ , without loss of generality, we can assume that $ m\leq t(\bar {u},(a)_{\mathrm {i}}) $ for some $ t\in \mathcal {F} $ and $ \mathrm {i}\in I $ . Moreover, let $ \alpha \in M $ be as in Lemma 3.4.2. In order to find some image for $ m $ , for every $ s\in M $ we define the following bounded $ \Pi _{1} $ -type:
We claim that there exists some $ s> I $ such that the type $ p_{s}(y) $ is finitely satisfiable in $\mathcal {M}$ . Then since $ p_{s} $ is a $ \Pi _{1} $ , bounded and recursive type in $\mathcal {M}$ , there exists some $ m' $ which realizes $ p_{s}(y) $ in $\mathcal {M}$ . Therefore, $\mathrm {P}((\bar {u},m),(\bar {v},m')) $ and $ \mathrm {Q}((\bar {u},m),(\bar {v},m')) $ hold in $\mathcal {M}$ , and this finishes the “forth” stage.
In order to prove the above claim, let $ d> I $ be an arbitrary and fixed element of $ M $ . Moreover, suppose $i,s\in M$ , and $ \Theta (s,i,\bar {u},m,\bar {v},b,a,\alpha ,\beta ,(a)_{\mathrm {i}}) $ is the following $ \Delta _{0}$ -formula:
where $ \beta $ is the code of the following $ \Sigma _{1}$ -definable set in $\mathcal {M}$ :
Note that the parameter $ i $ in $ \Theta $ is for bounding indexes of functions which appear in $ \Theta $ . Moreover, the parameter $ s $ is for bounding elements of $ N_{0} $ appearing in $ \Theta $ ; i.e., if $ \Theta $ contains some element of the form $ (a)_{w} $ , then $ w<s $ . Now, for every $ i\in M $ , we let $\mathrm {G}(i) $ be an upper bound for parameters $ s $ as above. To be more precise, we define
Clearly $ \mathrm {G} $ is $ \Delta _{0}$ -definable function in $\mathcal {M}$ , and $ I\subseteq \mathrm {Dom}(\mathrm {G}) $ (we assume $ \mathrm {max}(\emptyset )=0 $ ). Therefore, since $ I $ is strong, there exists some $ e> I $ such that for all $ i\in I $ , $ \mathrm {G}(i)> I $ iff $ \mathrm {G}(i)>e $ . We will show that $ p_{e}(y) $ is finitely satisfiable in $\mathcal {M}$ :
First, note that, $ p_{e1}(y) $ is closed under conjunctions. (This holds similar to the way statement (1) in the proof of Lemma 3.4.3 holds.) So let $f_{n}, f_{n_{1}},\ldots ,f_{n_{k}} $ be some finite number of elements of $ \mathcal {F} $ , and let $ n^{*}=\mathrm {max}\lbrace n,n_{0},\ldots ,n_{k}\rbrace $ . Then, use Lemma 3.4.3(i), $ (n^{*}+2)$ -many times; i.e., for every $ t=0,\ldots ,n^{*}+1 $ consider $ f_{t} $ instead of $ f $ in the assertion of Lemma 3.4.3(i), $ 0\in M_{0} $ instead of $ z $ , and $ f_{1},\ldots ,f_{n^{*}} $ . So by Lemma 3.4.3(i), for every $ t=0,\ldots ,n^{*}+1 $ there exists some $ s_{t}> I $ such that
-
(1): $ \mathcal {M}\models \exists y<b\ \Psi (f_{t},f_{1},\ldots ,f_{n^{*}}, \bar {u},m,\bar {v},y,b,a,s_{t},\alpha ,0,(a)_{\mathrm {i}}) $ .
Then, let $ s^{*}:=\mathrm {min}\lbrace s_{t}: \ t<n^{*}+1\rbrace $ . Therefore, by statement (1) and the definitions of $ \Theta $ and $ \Psi $ it holds that
-
(2): $ {\mathcal {M}\models \Theta (s^{*},n^{*},\bar {u},m,\bar {v},b,a,\alpha ,\beta ,(a)_{\mathrm {i}})} $ .
So from statement (2) and the definition of $ \mathrm {G} $ , we infer that $ \mathrm {G}(n^{*})>I$ and it holds that
-
(3): $ {\mathcal {M}\models \Theta (\mathrm {G}(n^{*}),n^{*}, \bar {u},m,\bar {v},b,a,\alpha ,\beta ,(a)_{\mathrm {i}})} $ .
Consequently $ \mathrm {G}(n^{*})>e $ , and again by the definition of $ \Theta $ and statement (3), we deduce that
-
(4): $\mathcal {M}\models \Theta (e,n^{*},\bar {u},m,\bar {v},b,a,\alpha ,\beta ,(a)_{\mathrm {i}})$ .
Statement (4) implies that
-
(4): $ \mathcal {M}\models \exists y\leq t(\bar {v},(a)_{\mathrm {i}}) \left (\begin{array}{@{}c@{}} \forall i<e ([f_{n}(\bar {u},m,(a)_{i})\downarrow ]\rightarrow [f_{n}(\bar {v},y,(a)_{i})\downarrow ]^{<b}) \ \wedge \\ \bigwedge _{t= 1}^{k}\forall i<e \left (\begin{array}{@{}c@{}} ([f_{n_{t}}(\bar {v},y,(a)_{i})\downarrow ]^{<b}\wedge \langle n_{t},i\rangle \mathrm {E}\alpha )\rightarrow \\ f_{n_{t}}(\bar {u},m,(a)_{i})\neq f_{n_{t}}(\bar {v},y,(a)_{i}) \end {array}\right ) \end {array}\right ) $ .
So statement (4) finishes the proof of the claim.
“Back” stages: Let $ m'\in M\setminus \{\bar {v}\} $ such that $ m'<v_{0}:=\max \{\bar {v}\} $ , and $ u_{0}:=\max \{\bar {u}\} $ . In order to find some element of $ M $ whose image is $ m' $ , we modify the proof of the “forth” stage in the following way:
-
• Replace $ p_{s}(y) $ by
$$ \begin{align*} &\qquad\qquad\qquad q_{s}(x):= \lbrace x<u_{0}\rbrace\cup q_{s1}(x)\cup q_{s2}(x) , \text{where,} \\ &{q_{s1}(x):= \lbrace\forall i<s(\neg [f(\bar{v},m', (a)_{i})\downarrow]^{<b}\rightarrow\neg [f(\bar{u},x,(a)_{i})\downarrow] ): f\in\mathcal{F}\rbrace;} \text{ and}\\ &\qquad {q_{s2}(x):=\left\lbrace\begin{array}{@{}c@{}} \forall i<s\left(\begin{array}{@{}c@{}} \left(\begin{array}{@{}c@{}}\langle n,i\rangle\mathrm{E}\alpha'\wedge\\ {}[f_{n}(\bar{u},x,(a)_{i})\downarrow]\end{array}\right)\rightarrow \\ f_{n}(\bar{u},x,(a)_{i})\neq f_{n}(\bar{v},m',(a)_{i}) \end{array}\right): \ n\in\mathbb{N} \end{array}\right\rbrace}. \end{align*} $$ -
• Replace $ \Theta (s,i,\bar {u},m,\bar {v},b,a,\alpha ,\beta ) $ with $ \Theta '(s,i,\bar {u},\bar {v},m',b,a,\alpha ',\beta ') $ :
$$ \begin{align*} \forall r<i \ \exists x<u_{0}\left(\begin{array}{@{}c@{}} \forall w<s(\langle x,r,w\rangle\mathrm{E}\beta'\rightarrow[f_{r}(\bar{v},m',(a)_{w})\downarrow]^{<b})\wedge \\ \forall w<s \forall r'<i \left(\begin{array}{@{}c@{}} (\langle r',w\rangle\mathrm{E}\alpha'\wedge\langle x,r',w\rangle\mathrm{E}\beta')\rightarrow\\ f_{r'}(\bar{u},m,(a)_{w})\neq f_{r'}(\bar{v},y,(a)_{w}))\end{array}\right) \end{array}\right),\end{align*} $$where $ \beta ' $ is the code of the following $ \Sigma _{1}$ -definable set in $\mathcal {M}$ :$$ \begin{align*} L':=\lbrace\langle x,r,w\rangle: \ \mathcal{M}\models (x<u_{0} \wedge w<d \wedge r<d \wedge [f_{r}(\bar{u},x,(a)_{w})\downarrow]) \rbrace. \end{align*} $$
The rest of the argument goes smoothly by modifying the “forth” stage according to the above changes, and this completes the proof.
Corollary 3.5. Assume $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ is countable and nonstandard, $ I $ is a proper cut of $\mathcal {M}$ , and $ \mathcal {M}_{0} $ is an $ I$ -small $ \Sigma _{1} $ -elementary submodel of $\mathcal {M}$ . Then the following are equivalent:
-
(1) $ I $ is strong in $\mathcal {M}$ .
-
(2) There exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ { M_{0}=\mathrm {Fix}(j)} $ .
Proof Suppose ${ M_{0}=\lbrace (a)_{i}: \ i\in I\rbrace } $ , for some $ a\in M $ such that $ (a)_{i}\neq (a)_{j} $ for all distinct $ i, j\in I $ .
$ (1)\Rightarrow (2) $ : If $ M_{0}=I $ , then by Theorem 2.4(2), we are done. So suppose $ I\subsetneq M_{0} $ . First, by using Theorem 3.4 let $ h $ be some proper initial self-embedding of $ \mathrm {H}^{1}(\mathcal {M};M_{0}) $ such that $ \mathrm {Fix}(h)=M_{0} $ . Moreover, fix some $b\in \mathrm {H}^{1}(\mathcal {M};M_{0})\setminus M_{0} $ such that $ h( \mathrm {H}^{1}(\mathcal {M};M_{0}))<b $ . Now, by using strong $ \Sigma _{1}$ -Collection in $ \mathrm {H}^{1}(\mathcal {M};M_{0}) $ , and since $ \mathrm {H}^{1}(\mathcal {M};M_{0})\prec _{\Sigma _{1}}\mathcal {M} $ , we can find some $ d\in \mathrm {H}^{1}(\mathcal {M};M_{0}) $ such that
Therefore, by Theorems 2.1 and 2.3 and Remark 1, there exists some proper initial embedding $ k:\mathcal {M}\hookrightarrow \mathrm {H}^{1}(\mathcal {M};M_{0}) $ such that $ M_{0}\subseteq \mathrm {Fix}(k) $ , $ k(M)<d, $ and $ b\in k(M) $ (note that since $ \mathrm {H}^{1}(\mathcal {M};M_{0}) $ is an initial segment of $\mathcal {M}$ , then $ \mathrm {SSy}_{I}(\mathcal {M})=\mathrm {SSy}_{I}(\mathrm {H}^{1}(\mathcal {M};M_{0})) $ ). Finally, we put $ j:=k^{-1}hk $ . It is easy to check that $ j $ is a well-defined proper initial self-embedding of $\mathcal {M}$ such that $ \mathrm {Fix}(j)=M_{0} $ .
$ (2)\Rightarrow (1) $ : We combine the methods used in the proofs of Theorems 5.1 and 6.1 of [Reference Bahrami and Enayat1]. Suppose I is not strong; i.e., there exists some coded function $ f $ in $\mathcal {M}$ such that $ I\subseteq \mathrm {Dom}(f) $ , and the set $ {D:=\lbrace f(i): \ i\in I \wedge I<f(i)\rbrace } $ is downward cofinal in $ M\setminus I $ .
Let $ b\in M\setminus M_{0} $ and $ g:=j(f) $ . For every $ k\in M $ , we put
Since $ A_{k} $ is bounded and $ \Delta _{1}$ -definable, it is coded by some $ s_{k} $ in $\mathcal {M}$ . Moreover, the function $ k\mapsto s_{k} $ is $ \Delta _{1}$ -definable in $ \mathcal {M} $ . Now, we define
So note that:
-
(I) For every $ k> I $ , we have $ \mathrm {Th}_{\Delta _{0}}(\mathcal {M};b,\{(a)_{i}\}_{i\in I})\subseteq \mathrm {Th}_{\Delta _{0}}(\mathcal {M};h(k),\{(a)_{i}\}_{i\in I})$ .
-
(II) For every $ i\in I $ , $ h(i)$ is well-defined and inside $ M_{0}=\mathrm {Fix}(j) $ ; the reason behind this statement is that for every $ i\in I $ we consider the following set:
$$ \begin{align*} B_{i}:=\lbrace\langle r,\epsilon\rangle: \ \mathcal{M}\models \exists y<i ((a)_{y}=\epsilon\wedge \langle r,y\rangle\mathrm{E}s_{i})\rbrace. \end{align*} $$Then, by Lemma 3.3, $ B_{i} $ is coded by some $ \alpha _{i}\in M_{0}=\mathrm {Fix}(j) $ . So it holds that$$ \begin{align*} \mathcal{M}\models h(i)=\mu_{x}(\forall \langle r,\epsilon\rangle\mathrm{E}\alpha_{i} \ (\mathrm{Sat}_{\Delta_{0}}(\delta_{r}(\epsilon,x))). \end{align*} $$As a result, since $ \mathcal {M}_{0}\prec _{\Sigma _{1}}\mathcal {M} $ , statement (II) holds.
Now, let $ h':=j(h) $ . So for all $ i\in I $ , and all $ u<i $ such that $ f(u)<i $ , statement (II) implies that
Therefore, for all $ i\in I $ , $ \mathcal {M}\models \varphi (i,f,g,h,h') $ , where $ \varphi (i,f,g,h,h') $ is the following $ \Delta _{1}$ -formula:
So by $ \Sigma _{1} $ -Overspill in $\mathcal {M}$ , there exists some $ s> I $ such that
Since $ D $ is downward cofinal in $ M\setminus I $ , there is some $ i_{0}\in I $ such that $ I<f(i_{0})<s $ . Let $ c:=h(f(i_{0})) $ . On one hand, by (I), $ \mathrm {Th}_{\Delta _{0}}(\mathcal {M};b,\{(a)_{i}\}_{i\in I})\subseteq \mathrm {Th}_{\Delta _{0}}(\mathcal {M};c,\{(a)_{i}\}_{i\in I})$ . As a result, because $ b\notin M_{0} $ , we have $ c\notin M_{0}=\mathrm {Fix}(j) $ . On the other hand $ (\spadesuit ) $ implies that
As a result, I has to be strong in $\mathcal {M}$ .
4 Strongness of the standard cut and fixed points
In this section, we will show some properties of $ \mathrm {Fix}(j) $ , when $ \mathbb {N} $ is not strong in $\mathcal {M}$ . Then we will derive some criteria for strongness of $ \mathbb {N} $ in a countable nonstandard model of $ \mathrm {I}\Sigma _{1} $ in terms of sets of fixed points of its initial self-embeddings.
Lemma 4.1. Suppose $ \mathcal {M}$ is a nonstandard model of $ \mathrm {I}\Sigma _{1} $ in which $ \mathbb {N} $ is not strong. Then for any self-embedding $ j $ of $\mathcal {M}$ the following hold:
-
(1) $ \mathrm {Fix}(j) $ is 1-tall.
-
(2) If $ \mathrm {Fix}(j)$ is a countable model of $\mathrm {B}\Sigma _{1} $ , then it is 1-extendable.
Proof
-
(1) Let $ a\in \mathrm {Fix}(j) $ be arbitrary and fixed. Since $ \mathrm {Fix}(j)\prec _{\Sigma _{1}}\mathcal {M} $ , it suffices to prove that $ \mathrm {K}^{1}(\mathcal {M};a) $ is not cofinal in $ \mathrm {Fix}(j) $ . Since $ \mathcal {M}\models \mathrm {B}^{+}\Sigma _{1} $ , there exists some $ t_{0}\in M $ such that $\mathrm {K}^{1}(\mathcal {M};a)<t_{0} $ . Moreover, by Lemma 2.5(2) there exists some $ t_{00}\in \mathrm {Fix}(j) $ such that ${\mathrm {Th}_{\Sigma _{1}}(\mathcal {M};t_{0},a)\subseteq \mathrm {Th}_{\Sigma _{1}}(\mathcal {M};t_{00},a)}$ . Therefore, $ \mathrm {K}^{1}(\mathcal {M};a)<t_{00} $ .
-
(2) By Theorem 2.2(2), and part (1) of this lemma, it suffices to prove that $ \mathbb {N} $ is not $ \Pi _{1}$ -definable in $ \mathrm {Fix}(j) $ . Suppose not; i.e., $ \mathbb {N} $ is definable in $ \mathrm {Fix}(j) $ by some $ \Pi _{1}$ -formula $ \pi (x) $ . By Lemma 2.5(1), $ \mathrm {Fix}(j)\cap M\setminus \mathbb {N} $ is downward cofinal in $ M\setminus \mathbb {N} $ . So by $ \Sigma _{1}$ -Underspill in $\mathcal {M}$ , there exists some $ n\in \mathbb {N} $ such that $ \mathcal {M}\models \neg \pi (n) $ , and consequently since $ \mathrm {Fix}(j)\prec _{\Sigma _{1}}\mathcal {M} $ , $ \mathrm {Fix}(j)\models \neg \pi (n) $ , which is a contradiction.
The following corollary generalizes Theorem 1.2:
Corollary 4.2. Let $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ be countable and nonstandard in which $ \mathbb {N} $ is not strong, and $ j $ is an initial self-embedding of $\mathcal {M}$ such that $ \mathrm {Fix}(j)\models \mathrm {B}\Sigma _{1} $ . Then $ \mathrm {Fix}(j) $ is isomorphic to a proper cut of $\mathcal {M}$ .
Proof By Theorem 2.2(1) and the previous lemma, it is enough to prove that $ {\mathrm {SSy}(\mathrm {Fix}(j))=\mathrm {SSy}(\mathcal {M})} $ . So let $ X=\mathbb {N}\cap a_{\mathrm {E}} $ for some $ a\in M $ . Since $ \mathbb {N} $ is not strong in $\mathcal {M}$ , by Lemma 2.5(2) there exists some $ b\in \mathrm {Fix}(j) $ such that $ \mathrm {Th}_{\Sigma _{1}}(\mathcal {M};a)\subseteq \mathrm {Th}_{\Sigma _{1}}(\mathcal {M};b) $ . Therefore, $ X=\mathbb {N}\cap b_{\mathrm {E}} $ , and this finishes the proof.
We conclude this section with a generalization of a similar result about automorphisms of countable recursively saturated models of $\mathrm {PA}$ in [Reference Kossak and Schmerl12]. Moreover, the following corollary defines Theorem 2.4(3).
Corollary 4.3. Let $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ be countable and nonstandard. Then the following are equivalent:
-
(1) $ \mathbb {N} $ is strong in $\mathcal {M}$ .
-
(2) There exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)=\mathrm {K}^{1}(\mathcal {M}). $
-
(3) There exists some proper initial self-embedding $ j $ of $\mathcal {M}$ , and some small $ \mathcal {M}_{0}\prec _{\Sigma _{1}}\mathcal {M} $ , such that $ \mathrm {Fix}(j)= M_{0}. $
-
(4) For every small $ \mathcal {M}_{0}\prec _{\Sigma _{1}}\mathcal {M} $ there exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)= M_{0}. $
-
(5) There exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)\subseteq \mathrm {I}^{1}(\mathcal {M}) $ .
If $ \mathcal {M}\models \mathrm {PA} $ and it is recursively saturated, then the above statements are equivalent to the following:
-
(6) There exists some proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)\models \mathrm {B}\Sigma _{1} $ and it is isomorphic to no proper initial segments of $\mathcal {M}$ .
Proof The equivalences of statements (1)–(5) is a straightforward implication of Corollary 3.5 and Lemma 4.1(1). Moreover, $ (6)\Rightarrow (1) $ holds by Corollary 4.2. In order to prove $ (4)\Rightarrow (6) $ , similar to the proof of Theorem 3.2(3), we will find some small recursively saturated elementary submodel $ \mathcal {M}_{0} $ of $\mathcal {M}$ . So statement (4) will provide us with a proper initial self-embedding $ j $ of $\mathcal {M}$ such that $ \mathrm {Fix}(j)=M_{0} $ . Clearly $ \mathrm {Fix}(j)\models \mathrm {B}\Sigma _{1} $ . Moreover, as we mentioned in the beginning of Section 3, $ \mathrm {SSy}(\mathcal {M}_{0})\neq \mathrm {SSy}(\mathcal {M}) $ . As a result, $ \mathrm {Fix}(j) $ is isomorphic to no proper initial segment of $\mathcal {M}$ .
5 Extendability
In this section, we will study the extendability of initial embeddings. Most of the theorems of this section are generalizations of results about automorphisms of countable recursively saturated models of $\mathrm {PA}$ obtained in [Reference Kossak and Kotlarski10, Reference Kossak and Kotlarski11].
Definition 2. Suppose $ \mathcal {M}$ and $ \mathcal {N} $ are models of $\mathrm {I}\Sigma _{1} $ , $ \mathcal {M}_{0}$ and $\mathcal {N}_{0}$ are bounded submodels (or proper cuts) of $\mathcal {M} $ and $ \mathcal {N,} $ respectively. We call an initial embedding $ j:\mathcal {M}_{0}\hookrightarrow \mathcal {N}_{0} $ an initial $ (\mathcal {M},\mathcal {N})$ -embedding if for every $A\subseteq M_{0} $ it holds that
If $ \mathcal {M}=\mathcal {N} $ , we call such $ j $ an initial $\mathcal {M}$ -embedding.
First, in the next lemma we will show that the condition in the above definition, i.e., preserving coded subsets, is a necessary condition for extendability of an initial embedding.
Lemma 5.1. Suppose $ \mathcal {M}$ and $ \mathcal {N} $ are models of $\mathrm {I}\Sigma _{1} $ , $ \mathcal {M}_{0}\subseteq \mathcal {M}$ and $\mathcal {N}_{0}\subseteq \mathcal {N}$ are bounded submodels (or proper cuts), and $ j:\mathcal {M}_{0}\hookrightarrow \mathcal {N}_{0} $ is an initial embedding. If $ j $ is extendable to some initial embedding $ \hat {j}:\mathcal {M}\hookrightarrow \mathcal {N} $ , then $ j $ is an initial $ (\mathcal {M},\mathcal {N})$ -embedding.
Proof Put $ I:=\mathrm {I}^{1}(\mathcal {M};M_{0}) $ , $ J:=\mathrm {I}^{1}(\mathcal {N};j(M_{0})) $ , and let $ A\subseteq M_{0} $ be arbitrary. If $ {A=I\cap (\alpha _{\mathrm {E}})^{\mathcal {M}}} $ for some $ \alpha $ in $\mathcal {M}$ , then clearly $ j(A)=J\cap ((\hat {j}(\alpha ))_{\mathrm {E}})^{\mathcal {N}}$ . Conversely, suppose $ j( A)\in \mathrm {SSy}_{J}(\mathcal {N}) $ . Since $ M_{0} $ is bounded in $\mathcal {M}$ , we have $ J\subsetneq _{e}\hat {j}( M) $ . As a result, $ j(A)\in \mathrm {SSy}_{J}(\hat {j}( M)) $ , which implies that $A\in \mathrm {SSy}_{I}(\mathcal {M}) $ .
Converse of the above lemma holds, when $ \mathcal {M}_{0} $ and $ j(M_{0}) $ are $ \Sigma _{1} $ -elementary initial segments of $\mathcal {M}$ and $ \mathcal {N} $ .
Theorem 5.2. Suppose $ \mathcal {M}$ and $ \mathcal {N} $ are countable and nonstandard models of $\mathrm {I}\Sigma _{1} $ , and $ I$ and $ J $ are $ \Sigma _{1} $ -elementary initial segments of $\mathcal {M}$ and $ \mathcal {N} $ , respectively. Then for any isomorphism $ {j: I\rightarrow J} $ which is an initial $ (\mathcal {M},\mathcal {N})$ -embedding and each $ b>J $ , there exists some proper initial embedding $ \hat {j}:\mathcal {M}\hookrightarrow \mathcal {N} $ such that $ \hat {j}\upharpoonright _{I}=j $ and $ \hat {j}(M)<b $ .
Sketch of proof.
The proof is conducted by a back-and-forth argument similar to the one used in the proof of [Reference Bahrami and Enayat1, Theorem 3.3]; we will build finite partial functions $ \bar {u}\mapsto \bar {v} $ such that the following induction hypothesis holds:
For the “forth” steps, if $ \bar {u}\mapsto \bar {v} $ is built, for given $ m\in M $ we define
Then, let $ h\in M $ such that $ H=I\cap h_{\mathrm {E}} $ . Since $ j $ is an initial $ (\mathcal {M},\mathcal {N}) $ -embedding, there exists some $ h'\in N $ such that $ j(H)=J\cap h^{\prime }_{\mathrm {E}} $ . Therefore, by induction hypothesis for every $ s\in I $ it holds that
-
(1): $ \mathcal {N}\models \exists x, w<b \ \forall \langle r,i\rangle <j(s) \ (\langle r,i\rangle \mathrm {E}h'\rightarrow [f_{r}(\bar {v},x,i))\downarrow ]^{<w} ) $ .
Since $ j $ is onto, statement (1) implies that for every $ t\in J $ it hold that
-
(2): $ \mathcal {N}\models \exists x, w<b \ \forall \langle r,i\rangle <t \ (\langle r,i\rangle \mathrm {E}h^{'}\rightarrow [f_{r}(\bar {v},x,i))\downarrow ]^{<w}) $ .
Therefore, by using $ \Sigma _{1}$ -Overspill in $ \mathcal {N} $ , we will find some image for $ m $ , for which induction hypothesis holds. The “back” stages can be done similarly.⊣
The proof of the above theorem can also be modified for $ I$ -small submodels.
Theorem 5.3. Suppose $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ is countable and nonstandard, I is a strong cut of $\mathcal {M}$ , $ \mathcal {M}_{0} $ is an $ I$ -small $ \Sigma _{1} $ -elementary submodel of $\mathcal {M}$ such that $ M_{0}:=\{(a)_{i}: i\in I\} $ , and $ j $ is an initial embedding of $ \mathcal {M}_{0} $ such that $ j(I)\subseteq _{e}\mathcal {M} $ . Then the following are equivalent:
-
(1) $ j \upharpoonright _{I}$ is an initial $ \mathcal {M}$ -embedding, and there exists some $ b\in M $ such that $ {\mathcal {M}\models j((a)_{i})=(b)_{j(i)}} $ for all $ i\in I $ .
-
(2) $ j $ extends to some proper initial self-embedding of $\mathcal {M}$ .
Sketch of proof.
$(2) \Rightarrow (1) $ holds by Lemma 5.1. In order to prove $ (1)\Rightarrow (2) $ , we will use a similar argument to the proof of [Reference Bahrami and Enayat1, Theorem 3.3] to obtain an extension $ \hat {j} $ of $ j $ . For this purpose, first we will fix some $ d\in M $ which is an upper bound for $ M_{0} $ . Then, we will build finite partial functions $ \bar {u}\mapsto \bar {v} $ such that the following induction hypothesis holds:
Here, we outline the proof for the “back” steps and the proof of “forth” steps is left to the reader. Suppose $ \bar {u}\mapsto \bar {v} $ is built, and $ m<\max \{\bar {v}\} $ is given. We define
Then, let $ l\in M $ such that $ L=j(I)\cap l_{\mathrm {E}} $ . Since $ j\upharpoonright _{I} $ is an initial $\mathcal {M}$ -embedding, then there exists some $ l'\in M $ such that $ j^{-1}(L)=I\cap l^{\prime }_{\mathrm {E}} $ . Moreover, by using Lemma 3.3, for every $ s\in I $ there exists some $ (a)_{i_{s}}\in M_{0} $ which codes of the following subset of $ M_{0} $ :
By $ \Pi _{1} $ -Overspill, it suffices to prove that for every $ s\in I $ it holds that
Suppose not; i.e., there exists some $ s\in I $ which for statement $ (\star ) $ does not hold. So we have,
-
(i): $ \mathcal {M}\models \forall x<\max \{\bar {u}\} \ \exists \langle r,i\rangle <s \ (\langle r,i\rangle \mathrm {E} l' \ \wedge [f_{r}(\bar {u},x,(a)_{i})\downarrow ] ) $ .
As a result, by using $ \Sigma _{1} $ -Collection in $\mathcal {M}$ , from statement $ (i) $ , induction hypothesis, and the way we chose $(a)_{i_{s}} $ , we may conclude that
-
(ii): $ \mathcal {M}\models \forall x<\max \{\bar {v}\} \ \exists \langle r,\epsilon \rangle \mathrm {E} (b)_{j(i_{s})} \ ( [f_{r}(\bar {v},x,\epsilon )\downarrow ]^{<d} ) $ .
So by statement $ (ii) $ , there exists some $ \langle r,i\rangle <s $ such that
-
(iii): $ \mathcal {M}\models (\langle r,i\rangle \mathrm {E} l' \ \wedge \ [f_{j(r)}(\bar {v},m,(b)_{j(i)})\downarrow ]^{<d} ) $ .
But statement $(iii)$ is in direct contradiction with the way we chose $ l' $ .⊣
In the last theorem, we investigate whether we can control the set of fixed points, while extending an isomorphism to an initial self-embeddings with larger domain:
Theorem 5.4. Suppose $ \mathcal {M}\models \mathrm {I}\Sigma _{1} $ is countable and nonstandard, I is a strong $ \Sigma _{1} $ -elementary initial segment of $\mathcal {M}$ , and $ j: I\rightarrow I $ is an isomorphism and an initial $ \mathcal {M}$ -embedding. Then there exists some proper initial self-embedding $ \hat {j} $ of $\mathcal {M}$ such that $ \hat {j}\upharpoonright _{ I}=j $ , and $ {\mathrm {Fix}(\hat {j})=\mathrm {Fix}(j)} $ .
Sketch of proof.
First, we will fix some arbitrary $ a>I $ . Since I is strong in $\mathcal {M}$ , there exists some $ b>I $ such that:
So by Theorem 5.2, there exists some proper initial self-embedding $ \bar {j} $ of $\mathcal {M}$ such that $ \bar {j}\upharpoonright _{I}=j $ and $ \bar {j}(M)<b $ . If $ \mathrm {Fix}(\bar {j})=\mathrm {Fix}(j) $ , then we are done. Otherwise, by using a similar argument to the proof of Theorem 3.4 (as we will briefly outline the modifications which should be made to the proof below), we will construct some proper initial self-embedding $ h $ of $ \mathcal {N}:=\mathrm {H}^{1}(\mathcal {M};a) $ such that $ h\upharpoonright _{I}=j $ , $ \mathrm {Fix}(h)=\mathrm {Fix}(j) $ , and $ h(N)<b $ . If $ \mathcal {M}=\mathcal {N} $ , then we are done. Otherwise, by using Theorem 2.3 we shall find some proper initial embedding $ k:\mathcal {M}\hookrightarrow \mathcal {N} $ such that $ I\subseteq \mathrm {I}_{\mathrm {fix}}(k) $ and $ b\in k(M) $ . Finally, we put $ \hat {j}:=k^{-1}hk $ .
In order to construct the aforementioned $ h $ , we will inductively build finite functions $ \bar {u}\mapsto \bar {v} $ such that
-
• For the first step of induction, we will take $ a\mapsto \bar {j}(a) $ ; clearly $ \mathrm {P}(a,\bar {j}(a)) $ holds in $\mathcal {M}$ . Moreover, by statement $ (\star ) $ and since $ \mathrm {Fix}(\bar {j})<b $ , the property $ \mathrm {Q}(a,\bar {j}(a)) $ also holds in $\mathcal {M}$ .
-
• Then suppose $ \bar {u}\mapsto \bar {v} $ is built. We will just mention the changes that should be made in the “forth” steps of Theorem 3.4, and “back” steps should be modified similarly:
-
– Suppose $ m\in N\setminus \{\bar {u}\} $ is given. By the definition of $ \mathcal {N} $ , without loss of generality, we may assume that $ m\leq t(\bar {u},a) $ for some $ t\in \mathcal {F} $ . Put
$$ \begin{align*} C:= \{\langle r,i\rangle\in I: \mathcal{N}\models[f_{r}(\bar{u},m,i)\downarrow]\wedge f_{r}(\bar{u},m,i)\notin\mathrm{K}^{1}(\mathcal{N};I\cup\{\bar{u}\}) \}. \end{align*} $$Let $ \alpha ,\alpha '\in N $ such that $ C=I\cap \alpha _{\mathrm {E}} $ and $ j( C)=I\cap \alpha ^{\prime }_{\mathrm {E}} $ (note that since $ \mathcal {N} $ is a $ \Sigma _{1} $ -elementary initial segment of $\mathcal {M}$ containing I, $ j $ is an initial $ \mathcal {N}$ -embedding): -
– Let $L:= \{\langle r,i\rangle \in I: \mathcal {N}\models [f_{r}(\bar {u},m,i)\downarrow ]\} $ , $ L=I\cap \beta _{\mathrm {E}} $ , and $ j(L)=I\cap \beta ^{\prime }_{\mathrm {E}} $ for $ \beta ,\beta '\in N $ .
-
– For every $ s\in \bar {j}(N) $ such that $ \bar {j}(s')=s $ for some $ s'\in N $ , let
$$ \begin{align*} &p_{s}(y):= \lbrace y< t(\bar{v},\bar{j}(a))\rbrace\cup p_{s1}(y)\cup p_{s2}(y), \text{ where} \\ &{p_{s1}(y):=\lbrace\forall i<s(\langle n,i\rangle\mathrm{E}\beta'\rightarrow [f_{n}(\bar{v},y,i)\downarrow]^{<b}): \ n\in\mathbb{N}\rbrace,} \text{ and }\\ &p_{s2}(y):=\left\lbrace \forall w<s'\ \forall i<s \left(\begin{array}{@{}c@{}} \left(\begin{array}{@{}c@{}}\langle n,w\rangle\mathrm{E}\alpha \ \wedge \ \langle n,i\rangle\mathrm{E}\alpha'\ \wedge \\ {}[f_{n}(\bar{v},y,i)\downarrow]^{<b}\end{array}\right) \rightarrow\\ f_{n}(\bar{u},m,w)\neq f_{n}(\bar{v},y,i)\end{array}\right) : \ n\in\mathbb{N} \right\rbrace. \end{align*} $$ -
– In order to find some $ s>I $ such that $ s\in \bar {j}(N) $ and $ p_{s}(y) $ is finitely satisfiable, we will adapt the rest of the proof of Theorem 3.4 accordingly; for instance, we will mention two of these adaptations:
-
(1) Let $ d'\in N $ such that $ d'>I $ and $ d:=\bar {j}(d') $ . Moreover, for every $i,s,s'\in N$ , let $ \Theta (s,s',i,\bar {u},m,\bar {v},b,\bar {j}(a),\alpha ,\alpha ',\beta ') $ be the following $ \Delta _{0}$ -formula:
$$ \begin{align*} & \forall r<i \ \exists y\leq t(\bar{v},\bar{j}(a))\\ & \quad \times \left(\begin{array}{@{}c@{}} \forall w<s (\langle r,w\rangle\mathrm{E}\beta'\rightarrow [f_{r}(\bar{v},y,w)\downarrow]^{<b}) \ \wedge\\ \forall w<s\forall w'<s' \forall r'<i\left(\begin{array}{@{}c@{}} \left(\begin{array}{@{}c@{}}\langle r',w'\rangle\mathrm{E}\alpha'\wedge\\ \langle r',w\rangle\mathrm{E}\alpha\wedge\\ {}[f_{r'}(\bar{v},y,(a)_{w})\downarrow]^{<b}\end{array}\right)\rightarrow \\ f_{r'}(\bar{u},m,w')\neq f_{r'}(\bar{v},y,w)\end{array}\right)\end{array}\!\right). \end{align*} $$Then, for every $ i\in M $ , we define$$ \begin{align*} \mathrm{G}(i):=\mathrm{max}\lbrace w<d': \ \mathcal{M}\models\exists x\leq d \ \Theta(x,w,i,\bar{u},m,\bar{v},b,\bar{j}(a),\alpha,\alpha',\beta') \rbrace. \end{align*} $$Since I is strong, there exists some $ e'>I $ such that $ e'\leq d' $ , and for all $ i\in I $ , $ \mathrm {G}(i)> I $ iff $ \mathrm {G}(i)>e' $ . Then, for every $ i\in M $ put$$ \begin{align*} \kern-12pt l(i):=\mathrm{max}\left\lbrace x<\bar{j}(e'): \ \mathcal{M}\models \left(\begin{array}{@{}c@{}} [\mathrm{G}(i)\downarrow]^{<d'} \ \wedge \ \mathrm{G}(i)>e'\ \wedge \\ \Theta(x,\mathrm{G}(i),i,\bar{u},m,\bar{v},b,\bar{j}(a),\alpha,\alpha',\beta') \end{array}\right) \right\rbrace .\end{align*} $$Again, since $ I $ is strong, there exists some $ e> I $ such that $ e\leq d $ , and for all $ i\in I $ , $ l(i)> I $ iff $ l(i)>e $ . Then $ p_{e}(y) $ is a finitely satisfiable type. -
(2) Instead of the function $ \langle o(\diamondsuit ,y),h(\diamondsuit ,y)\rangle $ we need to define the following function:
$$ \begin{align*} & \langle o(\diamondsuit,y),h(\diamondsuit,y),h'(\diamondsuit,y)\rangle\\ & \quad := \min\left\lbrace\langle n_{t},i,w\rangle\mathrm{E}\alpha_{s_{{}_{0}}}: \left(\begin{array}{@{}c@{}} [b(\diamondsuit,y)\downarrow]\wedge\\ {}[f_{n_{t}}(\diamondsuit,y,i)\downarrow]^{<b(\diamondsuit,y)} \ \wedge\\ {}[g_{t}(\diamondsuit,w)\downarrow]^{<b(\diamondsuit,y)} \wedge\\ g_{t}(\diamondsuit,w)= f_{n_{t}}(\diamondsuit,y,i)\end{array}\right)\! \right\rbrace , \end{align*} $$where $ \alpha _{s_{0}}\in I $ is the code of the following subset of I:$$ \begin{align*} \{\langle n,i,w\rangle: \ \mathcal{M}\models i<s_{0}\wedge w<j^{-1}(s_{0})\wedge\langle n,w\rangle\mathrm{E}\alpha\wedge\langle n,i\rangle\mathrm{E}\alpha'\}. \end{align*} $$The rest of the adaptations should be made similar to statements (1) and (2) in order to construct h.⊣
-
Acknowledgements
I would like to thank Ali Enayat for his comments on the very first draft of the paper. I am also grateful to the anonymous referee for the careful reading of this paper and many insightful comments and suggestions.
Funding
This research was supported by a grant from Iran National Science Foundation (INSF) (No. 99019371).