betaFeedback

WorksMathematics

Melnikov residuals and stationary amalgamation in PAC fields

Abstract

We construct an invariant symmetric ternary relation with existence and stationarity over models for classes of characteristic-zero Frobenius fields associated with Melnikov formations. For proper formations, the automorphism theories of their canonical root-predicate expansions have amalgamation and no model companion. The underlying fields are then unbounded, and their theories are NSOP1_1 but not simple. In this case, on triples of elementary submodels, the constructed relation implies ordinary nonforking, which implies Kim-independence, which implies linear disjointness; all three implications are strict. We also study p.e.c. fields with group actions, namely fields existentially closed in every regular extension carrying an extension of the action. We prove that the class of p.e.c.\ difference fields of characteristic zero is not elementary, and that the theory of all characteristic-zero difference fields with canonical root predicates has no model companion.

Introduction

The theory of algebraically closed fields with an automorphism has a model companion, ACFA. This raises the question of which other theories of fields admit model companions after adding an automorphism. The present paper is a companion to [5], which studies this question for bounded PAC fields. Here we consider characteristic-zero Frobenius fields associated with Melnikov formations, with particular emphasis on proper formations. We construct a ternary relation on elementary submodels and use it to prove amalgamation for

2020 Mathematics Subject Classification. 03C60, 12E30, 20E18.

Key words and phrases. PAC fields, Frobenius fields, Melnikov formations, stationary amalgamation, generic automorphisms.

The first author acknowledges support from the Tübitak 1001 grant no. 124F359.

The second author acknowledges support from the Narodowe Centrum Nauki grant no. 2018/31/B/ST1/00357 and the Tübitak 1001 grant no. 124F359. the automorphism theories of the canonical root-predicate expansions. For proper formations, we prove that these automorphism theories have no model companion. The formation of all finite groups gives the familiar algebraically closed case, whose model companion is ACFA0_0 up to the canonical root-predicate expansion.

The construction uses the residual theory of Fried–Haran [15, 16] for Melnikov formations of finite groups. The relation combines linear disjointness with a prescribed relative algebraic closure of the compositum. We prove that it is invariant and symmetric and has existence and stationarity over models.

A related question concerns existential closure in regular extensions of fields with group actions. A field extension is regular if it is separable and the base field is relatively algebraically closed in it. Motivated by the study of group actions in [2, 4], we ask whether this restricted closure condition is axiomatizable. We show that this condition is not elementary for a single automorphism in characteristic zero, whereas for every finite group it is elementary in arbitrary characteristic. These ordinary-language results are proved directly in § 8, where we also deduce the nonexistence of a model companion in the root-predicate language using Corollary 5.3, applied to the canonical root-predicate expansions of ω\omega-free PAC fields.

For the amalgamation argument, we use the ring language expanded by predicates expressing that monic polynomials have a root. For a field KK, write KSK_S for its canonical definitional expansion by these predicates, taken from the Galois language of Cherlin–van den Dries–Macintyre [11]; see § 3. In characteristic zero, embeddings between these expansions are precisely regular field embeddings. For the Frobenius fields considered here, the complete theories of these expansions eliminate quantifiers and are therefore model complete. For proper formations, these theories also have the independence property and are therefore unstable. Our amalgamation theorem then supplies the remaining hypothesis of Kikyo’s theorem [23 Theorem 3.3], which rules out a model companion for each corresponding automorphism theory.

We isolate a general criterion for the amalgamation property for automorphisms (PAPA): invariance, existence, and stationarity of a ternary relation on small elementary submodels of a complete theory suffice (Lemma 5.1). The new ternary relation constructed on the PAC fields considered here provides instances of this criterion in nonsimple theories. For proper formations, it is strictly stronger than ordinary nonforking and Kim-independence (§ 7). We also recall the general ultraproduct criterion for elementary axiomatizability of existentially closed models: for any first-order theory, the class of all its e.c. models is elementary if and only if it is closed under ultraproducts; for inductive theories, these conditions are equivalent to the existence of a model companion (Fact 8.3; see also [19 Introduction]). For the inductive root-predicate theory, this connects nonexistence of a model companion with failure of ultraproduct closure of its e.c. models.

We introduce the notation needed to state the main result. We call a nonempty class D\mathcal{D} of finite groups a Melnikov formation if it is closed under isomorphisms, normal subgroups, quotients, and extensions. We call D\mathcal{D} proper if it is different from the class of all finite groups.

For a field EE, let EalgE^{\mathrm{alg}} denote an algebraic closure of EE, and let EsepE^{\mathrm{sep}} denote its subfield consisting of the elements separable over EE. Throughout the paper, these closures are chosen compatibly: for subfields of a common field LL, we take their closures inside a fixed algebraic closure LalgL^{\mathrm{alg}}. We write Gal⁡(E):=Gal⁡(Esep/E)\operatorname{Gal}(E):=\operatorname{Gal}(E^{\mathrm{sep}}/E) for the absolute Galois group of EE. When EE has characteristic zero, Esep=EalgE^{\mathrm{sep}}=E^{\mathrm{alg}}; in arguments under this assumption we use EalgE^{\mathrm{alg}}, so that Gal⁡(E)=Gal⁡(Ealg/E)\operatorname{Gal}(E)=\operatorname{Gal}(E^{\mathrm{alg}}/E).

Throughout this paper, by a homomorphism between profinite groups we mean a continuous homomorphism, and by a subgroup of a profinite group we mean a closed subgroup.

For a profinite group GG, let Im⁡(G)\operatorname{Im}(G) denote the class of its finite continuous quotients, up to isomorphism. We call Im⁡(Gal⁡(E))\operatorname{Im}(\operatorname{Gal}(E)) the finite Galois-group spectrum of EE; equivalently, it is the class of groups Gal⁡(L/E)\operatorname{Gal}(L/E), up to isomorphism, where L/EL/E ranges over the finite Galois extensions of EE.

A field is pseudo-algebraically closed (PAC) if every nonempty geometrically integral variety over it has a rational point. A Frobenius field is a PAC field whose absolute Galois group has the embedding property [13 Section 24.1]. The embedding property is defined in Subsection 2.1.

For the fixed Melnikov formation D\mathcal{D}, put

D⊥={H finite∣H has no nontrivial quotient belonging to D}.\mathcal{D}^{\perp}=\{H\text{ finite}\mid H\text{ has no nontrivial quotient belonging to }\mathcal{D}\}.

This is the class denoted by C\mathcal{C} in [15 Definition 1.4], with C=D\mathcal{C}=\mathcal{D}. Let KD\mathcal{K}_{\mathcal{D}} be the class of characteristic-zero Frobenius fields FF such that Im⁡(Gal⁡(F))=D⊥\operatorname{Im}(\operatorname{Gal}(F))=\mathcal{D}^{\perp}.

For a field EE, let EDE^{\mathcal{D}} be the compositum, inside EalgE^{\mathrm{alg}}, of all finite Galois extensions of EE whose Galois groups belong to D\mathcal{D}. For subfields A,BA,B of a common field, ABAB denotes their compositum. For a subfield EE of a field LL, write ELalgE^{\mathrm{alg}}_{L} for the relative algebraic closure of EE in LL. A field is bounded if, up to isomorphism over that field, it has only finitely many finite separable extensions of each degree.

Let F∈KDF\in\mathcal{K}_{\mathcal{D}} and let U\mathbb{U} be a monster model of Th⁡(F)\operatorname{Th}(F). For small elementary submodels M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U}, define

A⊥MDB⟺{A,B are linearly disjoint over M,(AB)Ualg=(AB)D.A\mathrel{\mathop{\perp}\limits^{\mathcal{D}}_{M}}B \Longleftrightarrow \left\{ \begin{array}{l} A,B\text{ are linearly disjoint over }M,\\ (AB)^{\mathrm{alg}}_{\mathbb{U}}=(AB)^{\mathcal{D}}. \end{array} \right.

Theorem 1.1 (Main result). With this notation, the relation defined above is invariant and symmetric and has existence and stationarity over models. Put T=Th⁡(FD)T=\operatorname{Th}(F^{\mathcal{D}}). Then:

(i) TσT_{\sigma}, the theory of models of TT with an automorphism, has amalgamation.

(ii) If D\mathcal{D} is proper, then FF is unbounded and TT is NSOP1_1 and not simple.

(iii) If D\mathcal{D} is proper, then TσT_{\sigma} has no model companion. For the formation of all finite groups, the model companion is the canonical root-predicate expansion of ACFA⁡0\operatorname{ACFA}_{0}.

(iv) On triples of elementary submodels, this relation implies ordinary nonforking, which implies Kim-independence, which in turn implies linear disjointness over the base model. If D\mathcal{D} is proper, each of these three implications is strict.

For proper D\mathcal{D}, unboundedness has a concrete Galois-theoretic form: for every nontrivial finite quotient HH of Gal⁡(F)\operatorname{Gal}(F), there are infinitely many finite Galois extensions of FF with Galois group isomorphic to HH. In particular, Gal⁡(F)\operatorname{Gal}(F) is not topologically finitely generated. The NSOP1_1 assertion in Theorem 1.1(ii) is the known result for Frobenius fields [8 Introduction], transferred to their canonical root-predicate expansions; see § 3. The non-simplicity assertion follows from our unboundedness argument in § 6.

For example, for D={1}\mathcal{D}=\{1\}, the class D⊥\mathcal{D}^{\perp} consists of all finite groups and ED=EE^{\mathcal{D}}=E for every field EE. Consequently, KD\mathcal{K}_{\mathcal{D}} consists precisely of the characteristic-zero PAC fields whose absolute Galois groups are Z^\widehat{\mathbb{Z}}-free; see Subsection 2.1. For small elementary submodels M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U} as in Theorem 1.1, the independence relation becomes

A⊥MDB⟺{A,B are linearly disjoint over M,(AB)Ualg=AB.A\mathrel{\mathop{\perp}\limits^{\mathcal{D}}_{M}}B \Longleftrightarrow \left\{ \begin{array}{l} A,B\text{ are linearly disjoint over }M,\\ (AB)^{\mathrm{alg}}_{\mathbb{U}}=AB. \end{array} \right.

Our second main result concerns difference fields with a distinguished automorphism, independently of the classes KD\mathcal{K}_{\mathcal{D}}. Such a field is pseudo-existentially closed, or p.e.c., if it is existentially closed in the ordinary difference-field language in every difference-field extension whose underlying field extension is regular.

Theorem 1.2 (Difference fields).

(i) The class of p.e.c. characteristic-zero difference fields is not closed under ultrapowers. In particular, it is not elementary.

(ii) The theory of all characteristic-zero fields with an automorphism and canonical root predicates has no model companion.

Assertion (i) is proved in Theorem 8.10 and (ii) in Theorem 8.6. The root predicates make embeddings correspond to regular extensions. Since every ordinary existential formula is also existential in the expanded language, the reducts of its e.c. models are p.e.c. difference fields. The converse fails: models of ACFA0\mathrm{ACFA}_0 are p.e.c., but their canonical root-predicate expansions are not e.c. (Remark 8.4).

For finite GG, we prove that a GG-field is p.e.c. if and only if the action is faithful and its fixed field is PAC (Theorem 8.12). This gives an explicit characterization of the elementary class recorded in [17 Remark 6.25]. A GG-field is algebraically GG-closed if it has no proper algebraic field extension to which the GG-action extends. For every group GG, existential closure is equivalent to the conjunction of p.e.c. and algebraic GG-closedness (Fact 8.2). We compare the axiomatizability of these three classes for finite GG and for G=ZG = \mathbb{Z}.

This paper is organized as follows. § 2 separates notation and definitions from the group-theoretic facts used in the proofs. § 3 records the algebraic-closure and elementarity criteria, proves the regular-realization criterion, and recalls quantifier elimination for the canonical root-predicate expansions. It also identifies embeddings in that language with regular field embeddings and records the existence of existentially closed extensions after naming an automorphism. § 4 and § 5 construct the relation, prove existence and stationarity, and deduce amalgamation and the model-companion dichotomy. § 6 proves unboundedness and non-simplicity. § 7 compares the relation with ordinary nonforking, Kim-independence, and linear disjointness, and proves that all three implications are strict for proper DD, completing the proof of Theorem 1.1. § 8 begins with the closure notions for fields with group actions and the ultraproduct criterion for e.c. models. Subsection 8.1 proves the nonexistence of a model companion in the root-predicate language. Subsection 8.2 proves non-elementarity of p.e.c. difference fields, while Subsection 8.3 gives the finite-group characterization. Subsection 8.4 compares the axiomatizability of the resulting classes.

Preliminaries

Throughout this section, D\mathcal{D} is the Melnikov formation fixed in § 1.

Notation and definitions

2.1. Notation and definitions. We follow Fried–Haran [15] for formations and residuals, and Fried–Jarden [13] for the remaining profinite-group and field-arithmetic terminology.

We write N⊴oGN \trianglelefteq_{\mathrm{o}} G to mean that NN is an open normal subgroup of the profinite group GG. A profinite group is small if it has only finitely many open subgroups of each finite index. For a field EE, boundedness is equivalent to smallness of Gal⁡(E)\operatorname{Gal}(E). A profinite group is pro-D\mathcal{D} if all its finite continuous quotients belong to D\mathcal{D}. We retain the notation Gal⁡(E)\operatorname{Gal}(E), Im⁡(G)\operatorname{Im}(G), D⊥\mathcal{D}^{\perp}, and EDE^{\mathcal{D}} introduced in § 1.

For a profinite group GG, define its D\mathcal{D}-residual by

RD(G)=⋂N⊴oGG/N∈DN.R_{\mathcal{D}}(G)=\bigcap_{\substack{N\trianglelefteq_{\mathrm{o}}G\\G/N\in\mathcal{D}}}N.

Remark 2.1. For every profinite group GG, one has

RD(G)=G⟺Im⁡(G)⊆D⊥.R_{\mathcal{D}}(G)=G \quad\Longleftrightarrow\quad\operatorname{Im}(G)\subseteq\mathcal{D}^{\perp}.

Both conditions are equivalent to saying that GG has no nontrivial finite continuous quotient belonging to D\mathcal{D}. Indeed, RD(G)=GR_{\mathcal{D}}(G)=G has this meaning directly from the definition of the residual. If GG has no such quotient, then every finite quotient of GG belongs to D⊥\mathcal{D}^{\perp}, since a quotient of a finite quotient of GG is again a finite quotient of GG. Conversely, no nontrivial group in D⊥\mathcal{D}^{\perp} belongs to D\mathcal{D}, because it is its own nontrivial D\mathcal{D}-quotient.

A finite embedding problem for a profinite group GG is a pair of continuous epimorphisms

G→αH←βB,H,B finite.G \xrightarrow{\alpha} H \xleftarrow{\beta} B,\qquad H,B\ \text{finite}.

A weak solution of this embedding problem is a continuous homomorphism γ:G→B\gamma:G\to B such that β∘γ=α\beta\circ\gamma=\alpha. It is called a proper solution if γ\gamma is surjective.

A profinite group is projective if every finite embedding problem for it has a weak solution, and is ω\omega-free if every finite embedding problem for it has a proper solution. It has the embedding property if every finite embedding problem as above with B∈Im⁡(G)B\in\operatorname{Im}(G) has a proper solution. We write F^ω\widehat{F}_{\omega} for the free profinite group of countably infinite rank.

Standard facts and cited results

We first collect the residual properties used below. They are recorded in [15 Lemma 1.3(a),(b),(d)].

Fact 2.2. Let GG be a profinite group.

(a) The quotient G/RD(G)G/R_{\mathcal{D}}(G) is the maximal pro-D\mathcal{D} quotient of GG.

(b) For every epimorphism π:G↠H\pi:G\twoheadrightarrow H of profinite groups, we have

π(RD(G))=RD(H).\pi(R_{\mathcal{D}}(G))=R_{\mathcal{D}}(H).

(c) If RD(G)≤V⊴GR_{\mathcal{D}}(G)\leq V\trianglelefteq G, then we have

RD(V)=RD(G).R_{\mathcal{D}}(V)=R_{\mathcal{D}}(G).

(d) In particular, we obtain

RD(RD(G))=RD(G).R_{\mathcal{D}}(R_{\mathcal{D}}(G))=R_{\mathcal{D}}(G).

Proof of Fact 2.2(c) and (d). For the reader’s convenience, we recall the short proof. Suppose that RD(G)≤V⊴GR_{\mathcal{D}}(G)\leq V\trianglelefteq G. The group V/RD(G)V/R_{\mathcal{D}}(G) is pro-D\mathcal{D}, since it is a closed normal subgroup of the pro-D\mathcal{D} group G/RD(G)G/R_{\mathcal{D}}(G). By Fact 2.2(a), applied to VV, we have

RD(V)≤RD(G).R_{\mathcal{D}}(V)\leq R_{\mathcal{D}}(G).

Conversely, RD(V)R_{\mathcal{D}}(V) is characteristic in VV and hence normal in GG. The group G/RD(V)G/R_{\mathcal{D}}(V) is an extension of the pro-D\mathcal{D} group G/VG/V by the pro-D\mathcal{D} group V/RD(V)V/R_{\mathcal{D}}(V). Since D\mathcal{D} is a Melnikov formation, this extension is again pro-D\mathcal{D}. Fact 2.2(a), applied to GG, therefore gives

RD(G)≤RD(V).R_{\mathcal{D}}(G)\leq R_{\mathcal{D}}(V).

This proves part (c). Taking V=RD(G)V=R_{\mathcal{D}}(G) gives part (d). □

We next recall the embedding-property input. By [21 Lemma 4.2], every open subgroup of an ω\omega-free profinite group is ω\omega-free, and every closed normal subgroup has the embedding property. Together with the residual identities in [15 Lemmas 1.3 and 1.5], this gives the following standard consequence; compare [15 Theorem 5.5 and Remark 5.6(b)].

Lemma 2.3. If GG is an ω\omega-free profinite group, then RD(G)R_{\mathcal{D}}(G) has the embedding property and

Im⁡(RD(G))=D⊥.\operatorname{Im}(R_{\mathcal{D}}(G))=\mathcal{D}^{\perp}.

Proof. Put N:=RD(G)N:=R_{\mathcal{D}}(G). By definition, NN is an intersection of open normal subgroups of GG, so it is closed and normal in GG. Hence NN has the embedding property by [21 Lemma 4.2], as recalled above.

Let BB be a finite quotient of NN. By Fact 2.2(d) and (b), we have RD(B)=BR_{\mathcal{D}}(B)=B. Equivalently, BB has no nontrivial quotient in D\mathcal{D}, so B∈D⊥B\in\mathcal{D}^{\perp}.

Conversely, suppose that B∈D⊥B\in\mathcal{D}^{\perp}. Since GG is ω\omega-free, the embedding problem G→1←BG\to1\leftarrow B has a proper solution q:G↠Bq:G\twoheadrightarrow B. Fact 2.2(b) gives

q(N)=q(RD(G))=RD(B)=B.q(N)=q(R_{\mathcal{D}}(G))=R_{\mathcal{D}}(B)=B.

Thus Im⁡(N)=D⊥\operatorname{Im}(N)=\mathcal{D}^{\perp}. □

We also record the field-arithmetic existence result used in the regular-realization argument. A field EE of characteristic zero is Hilbertian if, for every polynomial f(T,X)∈E[T,X]f(T,X)\in E[T,X] that is irreducible in E(T)[X]E(T)[X] and has positive degree in XX, there are infinitely many a∈Ea\in E such that f(a,X)f(a,X) is irreducible in E[X]E[X] and has the same degree in XX as f(T,X)f(T,X).

Fact 2.4. Every characteristic-zero field EE has a regular extension L/EL/E such that LL is PAC and Gal⁡(L)\operatorname{Gal}(L) is ω\omega-free.

Proof. This follows by combining the regular PAC–Hilbertian extension construction of [13 Proposition 13.4.6] with the Fried–Völklein theorem [14] that characteristic-zero Hilbertian PAC fields are ω\omega-free; see also [21 Section 3, discussion preceding Example 3.2]. □

Regular realization and elementary embeddings

We first record the algebraic-closure and elementarity criteria used below. We then prove the regular-realization criterion and deduce the nonemptiness of KD\mathcal{K}_{D}. Finally, we record two general facts about the root-predicate language after naming an automorphism.

For a subset XX of a field FF, we write acl⁡F(X)\operatorname{acl}_{F}(X) for its model-theoretic algebraic closure in FF.

We use the root-predicate reduct of the Galois language of Cherlin–van den Dries–Macintyre [11 Section 4.2, p. 60]. Let us set

LS:=Lrings∪{Sn:n≥2}.\mathcal{L}_{S}:=\mathcal{L}_{\mathrm{rings}}\cup\{S_{n}:n\ge2\}.

where SnS_{n} is an nn-ary predicate symbol. For a field KK, write KSK_{S} for its canonical expansion, defined by

Sn(c1,…,cn)⟺∃x(xn+c1xn−1+⋯+cn=0).S_{n}(c_{1},\ldots,c_{n})\Longleftrightarrow\exists x\left(x^{n}+c_{1}x^{n-1}+\cdots+c_{n}=0\right).

Thus KSK_{S} is a parameter-free definitional expansion of KK. In particular, the expansion does not change the definable sets or model-theoretic algebraic closure. Every field automorphism preserves this expansion.

Proposition 3.1. The following assertions hold in characteristic zero.

(a) If FF is PAC and A⊆FA\subseteq F is a subfield, then acl⁡F(A)=AFalg\operatorname{acl}_{F}(A)=A^{\mathrm{alg}}_{F}.

(b) If F⊆KF\subseteq K are Frobenius fields, K/FK/F is regular, and Im⁡(Gal⁡(F))=Im⁡(Gal⁡(K))\operatorname{Im}(\operatorname{Gal}(F))=\operatorname{Im}(\operatorname{Gal}(K)), then F≼KF\preccurlyeq K.

(c) If FF is a Frobenius field and AA, BB are relatively algebraically closed subfields of FF, then every field isomorphism f:A→Bf:A\to B is a partial elementary map of FF.

The same elementarity conclusions hold for the canonical root-predicate expansions.

Proof. Assertion (a) is the algebraic-closure description recalled in [8 Section 1.8].

For (b) and (c), we recall the Galois formalism of Cherlin–van den Dries–Macintyre [11 Section 4.2, p. 60]:

LGal=LS∪{IG:G a finite group up to isomorphism}.\mathcal{L}_{\mathrm{Gal}}=\mathcal{L}_{S}\cup\{I_{G}:G\text{ a finite group up to isomorphism}\}.

The predicates SnS_{n} have the root interpretations given above, and IGI_{G} is a nullary predicate interpreted by

K⊨IG⟺G∈Im⁡(Gal⁡(K)).K\models I_{G}\Longleftrightarrow G\in\operatorname{Im}(\operatorname{Gal}(K)).

In the terminology of [11], perfect Iwasawa RC-fields are precisely perfect Frobenius fields. Their theory pIRC, with these interpretations, eliminates quantifiers in LGal\mathcal{L}_{\mathrm{Gal}} by [11 Theorem 41, p. 60]. This theorem does not require a fixed characteristic or finite Galois-group spectrum. In particular, it applies to characteristic-zero Frobenius fields.

For (b), regularity implies that the inclusion F⊆KF\subseteq K preserves and reflects every root predicate. The equality of the finite Galois-group spectra ensures that the nullary predicates agree. Thus the inclusion is an LGal\mathcal{L}_{\mathrm{Gal}}-embedding, and quantifier elimination makes it elementary.

For (c), a polynomial over AA has a root in FF if and only if it has a root in AA, because AA is relatively algebraically closed in FF. The same holds for BB. Therefore ff preserves and reflects the root predicates inherited from FF. The substructures on AA and BB also inherit the same nullary predicates from FF. Thus ff is a partial LGal\mathcal{L}_{\mathrm{Gal}}-isomorphism, and quantifier elimination makes it partial elementary. Taking reducts gives the conclusions in both the ring language and LS\mathcal{L}_{\mathsf{S}}. □\square

By Proposition 3.1(a), every subfield E⊆UE \subseteq\mathbb{U} satisfies

EUalg=acl⁡U(E),E^{\mathrm{alg}}_{\mathbb{U}}=\operatorname{acl}_{\mathbb{U}}(E),

since U\mathbb{U} is PAC. Thus the closure condition in the introductory definition of ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}} can equivalently be written as

acl⁡U(AB)=(AB)D.\operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal{D}}.

We use this equivalent formulation in the model-theoretic arguments below. The same identification holds in the canonical root-predicate expansion, which is definitional.

We next record the Galois-theoretic interpretation of the field construction EDE^{\mathcal{D}}.

Fact 3.2. Let EE be a characteristic-zero field.

(a) The extension ED/EE^{\mathcal{D}}/E is Galois, and

ED=(Ealg)RD(Gal⁡(E)).E^{\mathcal{D}}=(E^{\mathrm{alg}})^{R_{\mathcal{D}}(\operatorname{Gal}(E))}.

The restriction map induces the following canonical isomorphism

Gal⁡(E)/RD(Gal⁡(E))≅Gal⁡(ED/E).\operatorname{Gal}(E)/R_{\mathcal{D}}(\operatorname{Gal}(E))\cong\operatorname{Gal}(E^{\mathcal{D}}/E).

(b) We have

Gal⁡(ED)=RD(Gal⁡(E)),ED=E⟺RD(Gal⁡(E))=Gal⁡(E).\operatorname{Gal}(E^{\mathcal{D}})=R_{\mathcal{D}}(\operatorname{Gal}(E)),\qquad E^{\mathcal{D}}=E\Longleftrightarrow R_{\mathcal{D}}(\operatorname{Gal}(E))=\operatorname{Gal}(E).

Proof. By definition, EDE^{\mathcal{D}} is the compositum in EalgE^{\mathrm{alg}} of the finite Galois extensions L/EL/E with Gal⁡(L/E)∈D\operatorname{Gal}(L/E)\in\mathcal{D}. An element of Gal⁡(E)\operatorname{Gal}(E) fixes this compositum if and only if it fixes every such LL. Therefore

Gal⁡(Ealg/ED)=⋂N⊴Gal⁡(E)Gal⁡(E)/N∈DN=RD(Gal⁡(E)).\operatorname{Gal}(E^{\mathrm{alg}}/E^{\mathcal{D}})=\bigcap_{\substack{N\trianglelefteq\operatorname{Gal}(E)\\ \operatorname{Gal}(E)/N\in\mathcal{D}}}N=R_{\mathcal{D}}(\operatorname{Gal}(E)).

Galois correspondence gives part (a). Since ED/EE^{\mathcal{D}}/E is algebraic, EalgE^{\mathrm{alg}} is also an algebraic closure of EDE^{\mathcal{D}}, proving the equality in part (b). □\square

The same description gives the following compatibility with regular extensions.

Lemma 3.3. If E′/EE'/E is a regular extension of characteristic-zero fields, then we have

(E′)D∩Ealg=ED.(E')^{\mathcal{D}}\cap E^{\mathrm{alg}}=E^{\mathcal{D}}.

Proof. Regularity makes the restriction map

ρ:Gal⁡(E′)⟶Gal⁡(E)\rho:\operatorname{Gal}(E')\longrightarrow\operatorname{Gal}(E)

surjective. By Fact 2.2(b), we obtain

ρ(RD(Gal⁡(E′)))=RD(Gal⁡(E)).\rho(R_{\mathcal{D}}(\operatorname{Gal}(E')))=R_{\mathcal{D}}(\operatorname{Gal}(E)).

Using the fixed-field descriptions from Fact 3.2(a) for both E′E' and EE, we obtain

(E′)D∩Ealg=(Ealg)ρ(RD(Gal⁡(E′)))=(Ealg)RD(Gal⁡(E))=ED.(E')^{\mathcal{D}}\cap E^{\mathrm{alg}}=(E^{\mathrm{alg}})^{\rho(R_{\mathcal{D}}(\operatorname{Gal}(E')))}=(E^{\mathrm{alg}})^{R_{\mathcal{D}}(\operatorname{Gal}(E))}=E^{\mathcal{D}}.

The following regular-realization criterion follows from Fact 2.4, Lemma 2.3, and the residual identities in Fact 2.2. We include the short argument for completeness.

Lemma 3.4. Let EE be a characteristic-zero field. The following are equivalent:

(i) EE has a regular extension K∈KDK\in\mathcal{K}_{\mathcal{D}};

(ii) RD(Gal⁡(E))=Gal⁡(E)R_{\mathcal{D}}(\operatorname{Gal}(E))=\operatorname{Gal}(E);

(iii) Im⁡(Gal⁡(E))⊆D⊥\operatorname{Im}(\operatorname{Gal}(E))\subseteq\mathcal{D}^{\perp}.

Proof. The equivalence of (ii) and (iii) follows from Remark 2.1. If (i) holds, restriction Gal⁡(K)↠Gal⁡(E)\operatorname{Gal}(K) \twoheadrightarrow\operatorname{Gal}(E) is surjective, so every finite quotient of Gal⁡(E)\operatorname{Gal}(E) belongs to Im⁡(Gal⁡(K))=D⊥\operatorname{Im}(\operatorname{Gal}(K)) = \mathcal{D}^{\perp}. Thus (i) implies (iii).

Assume (ii). By Fact 2.4, choose a regular PAC extension L/EL/E with Gal⁡(L)\operatorname{Gal}(L) ω\omega-free. Put K=LDK = L^{\mathcal{D}}. As an algebraic extension of a PAC field, KK is PAC [13 Corollary 11.2.5]. By Fact 3.2(b),

Gal⁡(K)=RD(Gal⁡(L)).\operatorname{Gal}(K) = R_{\mathcal{D}}(\operatorname{Gal}(L)).

Lemma 2.3 therefore gives the embedding property and Im⁡(Gal⁡(K))=D⊥\operatorname{Im}(\operatorname{Gal}(K)) = \mathcal{D}^{\perp}, hence K∈KDK \in\mathcal{K}_{\mathcal{D}}. Finally, assumption (ii) and Fact 3.2(b) give ED=EE^{\mathcal{D}} = E. Since L/EL/E is regular, Lemma 3.3 gives

K∩Ealg=LD∩Ealg=ED=E.K \cap E^{\mathrm{alg}} = L^{\mathcal{D}} \cap E^{\mathrm{alg}} = E^{\mathcal{D}} = E.

Hence K/EK/E is regular. □\square

Taking E=QalgE = \mathbb{Q}^{\mathrm{alg}}, we have

Gal⁡(E)=1=RD(Gal⁡(E)).\operatorname{Gal}(E) = 1 = R_{\mathcal{D}}(\operatorname{Gal}(E)).

Lemma 3.4 therefore supplies a regular extension K/EK/E belonging to KD\mathcal{K}_{\mathcal{D}}. In particular, KD\mathcal{K}_{\mathcal{D}} is nonempty.

Remark 3.5. By Fact 3.2(b), condition (ii) of Lemma 3.4 is equivalent to ED=EE^{\mathcal{D}} = E. Thus a characteristic-zero field EE satisfies ED=EE^{\mathcal{D}} = E if and only if it occurs as a relatively algebraically closed subfield of some K∈KDK \in\mathcal{K}_{\mathcal{D}}. By Proposition 3.1(a), this is equivalently the condition acl⁡K(E)=E\operatorname{acl}_{K}(E) = E, where acl⁡K\operatorname{acl}_{K} is model-theoretic algebraic closure. The same characterization holds in the canonical root-predicate expansion of KK, since that expansion is definitional.

We now axiomatize the classes of fields under consideration and spell out the quantifier-elimination result used above.

For each finite group GG, let φG\varphi_{G} be a sentence in the ring language expressing that the field has a finite Galois extension with Galois group isomorphic to GG. The requirement Im⁡(Gal⁡(K))=D⊥\operatorname{Im}(\operatorname{Gal}(K)) = \mathcal{D}^{\perp} is expressed by the axiom scheme

ΣD={φG:G∈D⊥}∪{¬φG:G∉D⊥},\Sigma_{\mathcal{D}} = \{\varphi_{G} : G \in\mathcal{D}^{\perp}\} \cup\{\neg\varphi_{G} : G \notin\mathcal{D}^{\perp}\},

where GG ranges over finite groups up to isomorphism. Let TDT_{\mathcal{D}} be the theory in the ring language consisting of the axioms for characteristic-zero Frobenius fields [8 Section 1.9] together with ΣD\Sigma_{\mathcal{D}}. Then

Mod⁡(TD)=KD.\operatorname{Mod}(T_{\mathcal{D}}) = \mathcal{K}_{\mathcal{D}}.

In particular, KD\mathcal{K}_{\mathcal{D}} is an elementary class. Let TD,ST_{\mathcal{D},S} denote the definitional expansion of TDT_{\mathcal{D}} by the root predicates defined above. Its models are precisely the structures KSK_{S} with K∈KDK \in\mathcal{K}_{\mathcal{D}}.

Fact 3.6. Let C\mathcal{C} be a class of finite groups, up to isomorphism, realized as the finite Galois-group spectrum of a characteristic-zero Frobenius field. The theory of characteristic-zero Frobenius fields with spectrum C\mathcal{C}, in their canonical LS\mathcal{L}_{S}-expansions, is consistent and eliminates quantifiers. In particular, TD,ST_{\mathcal{D},S} is consistent, eliminates quantifiers, and is model complete.

Proof. Use the language LGal\mathcal{L}_{\mathrm{Gal}} and the quantifier-elimination theorem recalled in the proof of Proposition 3.1. Over the canonical expansions of characteristic-zero Frobenius fields, every LS\mathcal{L}_{S}-formula is equivalent to a quantifier-free LGal\mathcal{L}_{\mathrm{Gal}}-formula. When the finite Galois-group spectrum is C\mathcal{C}, replace each nullary predicate IGI_{G} in this formula by 0=00 = 0 if G∈CG \in\mathcal{C}, and by 0≠00 \ne0 otherwise. The resulting formula is quantifier-free in LS\mathcal{L}_{S}. Consistency follows from the assumed realization of C\mathcal{C}.

Lemma 3.4, applied to E=QalgE = \mathbb{Q}^{\mathrm{alg}}, shows that D⊥\mathcal{D}^{\perp} is realized by a field in KD\mathcal{K}_{\mathcal{D}}. The first assertion therefore applies to TD,ST_{\mathcal{D},S}. Model completeness follows from quantifier elimination. □\square

By Fact 3.6, every LS\mathcal{L}_S-embedding between models of TD,ST_{D,S} is elementary. Since regular field embeddings preserve and reflect the root predicates, every regular embedding between fields in KD\mathcal{K}_D is elementary, both in LS\mathcal{L}_S and in the ring language.

The theory TD,ST_{D,S} need not be complete. For a fixed F∈KDF \in\mathcal{K}_D, we work with its completion

T=Th⁡(FS).T = \operatorname{Th}(F_S).

This is a parameter-free definitional expansion of Th⁡(F)\operatorname{Th}(F). It inherits quantifier elimination and model completeness from TD,ST_{D,S}.

The theory of every Frobenius field is NSOP1_1; see [8 Introduction and Comments 2.13], which recalls the relevant results of Chernikov–Ramsey. Consequently, for every F∈KDF \in\mathcal{K}_D, the theory T=Th⁡(FS)T = \operatorname{Th}(F_S) is NSOP1_1, since definitional expansions preserve this property. This proves the NSOP1_1 assertion in Theorem 1.1(ii). We also use this result in the comparison with Kim-independence in § 7.

We finish with two facts about fields with an automorphism in the root-predicate language. For later use, put

LS,σ=LS∪{σ},\mathcal{L}_{S,\sigma} = \mathcal{L}_S \cup\{\sigma\},

where σ\sigma is a unary function symbol. Let TS,σT_{S,\sigma} be the theory of all characteristic-zero fields with an automorphism and the canonical interpretations of the root predicates. We write (K,σ)S(K,\sigma)_S for the canonical expansion of a difference field (K,σ)(K,\sigma) to this language. No PAC condition or prescribed finite Galois-group spectrum is imposed on models of TS,σT_{S,\sigma}.

Proposition 3.7. For an inclusion K⊆LK \subseteq L of characteristic-zero fields,

KS⊆LSLS⟺K is relatively algebraically closed in L⟺L/K is regular.K_S \subseteq_{\mathcal{L}_S} L_S \quad\Longleftrightarrow\quad K\text{ is relatively algebraically closed in }L \quad\Longleftrightarrow\quad L/K\text{ is regular}.

Consequently, embeddings between models of TS,σT_{S,\sigma} are exactly the difference-field embeddings whose underlying field extensions are regular.

Proof. If a∈L∖Ka \in L \setminus K is algebraic over KK, its monic minimal polynomial has degree n≥2n \ge2. Its coefficient tuple satisfies SnS_n in LL but not in KK, contradicting substructure inclusion. Conversely, if KK is relatively algebraically closed in LL, every root in LL of a polynomial over KK already lies in KK. Thus all root predicates are preserved and reflected. In characteristic zero this condition is equivalent to regularity. The difference-field assertion follows by additionally requiring the embedding to commute with the automorphisms. □\square

Fact 3.8. The theory TS,σT_{S,\sigma} is inductive. Every model (K,σ)S(K,\sigma)_S has an existentially closed TS,σT_{S,\sigma}-extension (M,τ)S(M,\tau)_S. The underlying field extension M/KM/K is regular.

Proof. The field, characteristic-zero, automorphism, and root-predicate axioms admit a universal–existential axiomatization. Hence TS,σT_{S,\sigma} is inductive. Every model of an inductive theory has an existentially closed extension. The underlying field extension is regular by Proposition 3.7. □\square

Existence and stationarity over models

In this section, we prove the assertion about the relation in Theorem 1.1. Recall that D\mathcal{D} is a Melnikov formation: a nonempty class of finite groups closed under isomorphisms, normal subgroups, quotients, and extensions. Fix F∈KDF \in\mathcal{K}_D; thus FF is a characteristic-zero Frobenius field with Im⁡(Gal⁡(F′))=D⊥\operatorname{Im}(\operatorname{Gal}(F')) = \mathcal{D}^{\perp}, where D⊥\mathcal{D}^{\perp} consists of the finite groups having no nontrivial quotient in D\mathcal{D}. We work in a sufficiently saturated and homogeneous monster U⊨Th⁡(F)\mathbb{U} \models\operatorname{Th}(F). For small elementary submodels M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U}, recall that

A⊥MDB⟺{A,B are linearly disjoint over M,acl⁡U(AB)=(AB)D.A \mathrel{\mathop{\perp}\limits^{\mathcal{D}}_M} B \quad\Longleftrightarrow\quad \left\{ \begin{array}{l} A,B\text{ are linearly disjoint over }M,\\ \operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal{D}}. \end{array} \right.

Proposition 4.1 (Existence). For small models M≼A,B≼UM \preccurlyeq A, B \preccurlyeq\mathbb{U}, there is A′≼UA' \preccurlyeq\mathbb{U} such that

A′≅MA,A′↓MDB.A' \cong_{M} A,\qquad A' \mathrel{\downarrow^{\mathcal{D}}_{M}} B.

Proof. Since A/MA/M and B/MB/M are regular, A⊗MBA \otimes_{M} B is a domain. Put

E=Frac⁡(A⊗MB).E = \operatorname{Frac}(A \otimes_{M} B).

We identify AA and BB with their canonical images in EE. These images are linearly disjoint over MM, and the extensions E/AE/A and E/BE/B are regular. By Fact 3.2(b),

Gal⁡(ED)=RD(Gal⁡(E)).\operatorname{Gal}(E^{\mathcal{D}}) = R_{\mathcal{D}}(\operatorname{Gal}(E)).

Since A,B∈KDA, B \in\mathcal{K}_{\mathcal{D}}, we have AD=AA^{\mathcal{D}} = A and BD=BB^{\mathcal{D}} = B. Applying Lemma 3.3 to the regular extensions E/AE/A and E/BE/B, we obtain

ED∩Aalg⁡=AD=A,ED∩Balg⁡=BD=B.E^{\mathcal{D}} \cap A^{\operatorname{alg}} = A^{\mathcal{D}} = A,\qquad E^{\mathcal{D}} \cap B^{\operatorname{alg}} = B^{\mathcal{D}} = B.

Thus ED/AE^{\mathcal{D}}/A and ED/BE^{\mathcal{D}}/B are regular. Moreover, Fact 2.2(d) gives

RD(Gal⁡(ED))=RD(RD(Gal⁡(E)))=RD(Gal⁡(E))=Gal⁡(ED).R_{\mathcal{D}}(\operatorname{Gal}(E^{\mathcal{D}})) = R_{\mathcal{D}}(R_{\mathcal{D}}(\operatorname{Gal}(E))) = R_{\mathcal{D}}(\operatorname{Gal}(E)) = \operatorname{Gal}(E^{\mathcal{D}}).

We apply Lemma 3.4 to obtain a regular extension K/EDK/E^{\mathcal{D}} such that K∈KDK \in\mathcal{K}_{\mathcal{D}}. The inclusions A,B↪KA, B \hookrightarrow K are regular, hence elementary by Proposition 3.1(b). As ED/EE^{\mathcal{D}}/E is algebraic and K/EDK/E^{\mathcal{D}} is regular, Proposition 3.1(a) gives

acl⁡K(E)=ED.\operatorname{acl}_{K}(E) = E^{\mathcal{D}}.

Replacing KK by an elementary submodel containing EDE^{\mathcal{D}}, we may assume that KK is small; the displayed equality and regularity over EDE^{\mathcal{D}} are preserved. Choose an elementary embedding j:K→Uj : K \to\mathbb{U} over BB, and put A′=j(A)A' = j(A) and f=j∣Af = j|_{A}. Then A′≼UA' \preccurlyeq\mathbb{U}, and f:A→A′f : A \to A' is an elementary isomorphism fixing MM pointwise. The fields A′A' and BB are linearly disjoint over MM. Since algebraic closure of a subset of an elementary submodel is unchanged in the monster, we have

acl⁡U(A′B)=j(ED)=(A′B)D.\operatorname{acl}_{\mathbb{U}}(A'B) = j(E^{\mathcal{D}}) = (A'B)^{\mathcal{D}}.

Thus A′↓MDBA' \mathrel{\downarrow^{\mathcal{D}}_{M}} B, as required. □\square

Proposition 4.2 (Stationarity). Suppose A↓MDBA \mathrel{\downarrow^{\mathcal{D}}_{M}} B and A′↓MDBA' \mathrel{\downarrow^{\mathcal{D}}_{M}} B. If f:A→A′f : A \to A' is an elementary isomorphism over MM, then

f∪id⁡B:A∪B→A′∪Bf \cup\operatorname{id}_{B} : A \cup B \to A' \cup B

is a partial elementary map.

Proof. Linear disjointness extends f∪id⁡Bf \cup\operatorname{id}_{B} uniquely to a field isomorphism

f0:E:=AB⟶E′:=A′B.f_{0} : E := AB \longrightarrow E' := A'B.

We extend f0f_{0} further to an isomorphism f^:Ealg⁡→(E′)alg⁡\widehat{f} : E^{\operatorname{alg}} \to(E')^{\operatorname{alg}}. Conjugation by f^\widehat{f} induces an isomorphism Gal⁡(E)→Gal⁡(E′)\operatorname{Gal}(E) \to\operatorname{Gal}(E'), which sends RD(Gal⁡(E))R_{\mathcal{D}}(\operatorname{Gal}(E)) onto RD(Gal⁡(E′))R_{\mathcal{D}}(\operatorname{Gal}(E')) by Fact 2.2(b). Taking fixed fields and applying Fact 3.2(a), we obtain

f^(ED)=(E′)D.\widehat{f}(E^{\mathcal{D}}) = (E')^{\mathcal{D}}.

Since A↓MDBA \mathrel{\downarrow^{\mathcal{D}}_{M}} B and A′↓MDBA' \mathrel{\downarrow^{\mathcal{D}}_{M}} B, we obtain

ED=acl⁡U(E)=acl⁡U(AB),(E′)D=acl⁡U(E′)=acl⁡U(A′B).E^{\mathcal{D}} = \operatorname{acl}_{\mathbb{U}}(E) = \operatorname{acl}_{\mathbb{U}}(AB),\qquad(E')^{\mathcal{D}} = \operatorname{acl}_{\mathbb{U}}(E') = \operatorname{acl}_{\mathbb{U}}(A'B).

The map f^∣ED\widehat{f}|_{E^{\mathcal{D}}} is therefore an isomorphism between relatively algebraically closed subfields of U\mathbb{U}, so it is partial elementary by Proposition 3.1(c). Since we have

(f^∣ED)∣A∪B=f∪id⁡B,\left(\widehat{f}|_{E^{\mathcal{D}}}\right)|_{A\cup B} = f \cup\operatorname{id}_{B},

the map f∪id⁡Bf \cup\operatorname{id}_{B} is partial elementary as well. □\square

Invariance and symmetry follow directly from the definition of the relation. This proves the invariance, symmetry, existence and stationarity asserted in Theorem 1.1.

No model companion after adding an automorphism

For a complete theory TT in a language LL, let TσT_{\sigma} be the theory of models of TT equipped with an LL-automorphism. We say that TT has the amalgamation property for automorphisms, abbreviated PAPA, if such structures have amalgamation with respect to maps that are elementary on the underlying LL-structures and commute with the named automorphisms; see [24 Definition 3.1]. If TT is model complete, PAPA is equivalent to the ordinary amalgamation property of TσT_{\sigma}.

The following criterion isolates the independence properties used in the standard proof of PAPA for stable theories; compare [24 Fact 3.5].

Lemma 5.1. Let TT be a complete theory with monster model U\mathbb{U}. Suppose that a relation A⊥MBA \mathrel{\perp_M} B is defined for small elementary submodels M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U} and satisfies the following properties.

(i) Invariance. For every g∈Aut⁡(U)g \in\operatorname{Aut}(\mathbb{U}),

A⊥MB⟹g(A)⊥g(M)g(B).A \mathrel{\perp_M} B \quad\Longrightarrow\quad g(A) \mathrel{\perp_{g(M)}} g(B).

(ii) Existence. For every M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U}, there is an elementary copy A′A' of AA over MM such that A′⊥MBA' \mathrel{\perp_M} B.

(iii) Stationarity. If A⊥MBA \mathrel{\perp_M} B and A′⊥MBA' \mathrel{\perp_M} B, then every elementary isomorphism f:A⟶A′f:A \longrightarrow A' fixing MM pointwise has the property that f∪id⁡Bf \cup\operatorname{id}_B is a partial elementary map.

Then TT has PAPA.

Proof. We consider models (M,σM)(M,\sigma_M), (A,σA)(A,\sigma_A) and (B,σB)(B,\sigma_B) of TσT_{\sigma} such that

M≼A,B,σA∣M=σB∣M=σM;M \preccurlyeq A,B,\qquad\sigma_A|_M=\sigma_B|_M=\sigma_M;

and we work in a sufficiently large monster model U\mathbb{U} of TT. By existence, we may replace AA by an elementary copy over MM, transporting σA\sigma_A, so that A⊥MBA \mathrel{\perp_M} B.

We extend σB\sigma_B to an automorphism τ∈Aut⁡(U)\tau\in\operatorname{Aut}(\mathbb{U}), and define

A′:=τ−1(A),f:=τ−1∘σA:A⟶A′.A' := \tau^{-1}(A),\qquad f := \tau^{-1}\circ\sigma_A:A\longrightarrow A'.

Since τ(M)=M\tau(M)=M and τ(B)=B\tau(B)=B, invariance gives A′⊥MBA' \mathrel{\perp_M} B. Moreover, ff is an elementary isomorphism fixing MM pointwise, because τ∣M=σB∣M=σA∣M\tau|_M=\sigma_B|_M=\sigma_A|_M. Stationarity therefore implies that f∪id⁡Bf\cup\operatorname{id}_B is a partial elementary map. Composing with τ\tau, we obtain the partial elementary map

τ∘(f∪id⁡B)=σA∪σB.\tau\circ(f\cup\operatorname{id}_B)=\sigma_A\cup\sigma_B.

By homogeneity, this map extends to an automorphism ss of U\mathbb{U}. Thus (U,s)(\mathbb{U},s) is an amalgam. □\square

The criterion requires only invariance, existence and stationarity over models. Symmetry, monotonicity, transitivity, finite character and local character are not used. In particular, the relation need not be identified with forking or Kim-independence. For a complete stable theory, nonforking over models satisfies the three hypotheses, recovering stable PAPA.

We now return to the fields considered in this paper. Let T=Th⁡(FS)T=\operatorname{Th}(F_S) for F∈KDF\in\mathcal{K}_D, with the canonical root-predicate expansion defined in § 3. By quantifier elimination, TT is model complete. Its embeddings between models are exactly the regular field embeddings, and these are elementary on the field reducts.

Proposition 5.2. The theory TT has PAPA. Equivalently, TσT_{\sigma} has the amalgamation property.

Proof. The relation ⊥\mathrel{\perp} is invariant and has existence and stationarity over models by Propositions 4.1 and 4.2. These properties also hold in the canonical root-predicate expansion, since it is definitional. Lemma 5.1 therefore gives PAPA, and model completeness gives the amalgamation property of TσT_\sigma.

Corollary 5.3. If D\mathcal{D} is proper, then TT has the independence property and TσT_\sigma has no model companion.

Proof. A proper Melnikov formation omits some finite simple group SS. Such an SS belongs to D⊥\mathcal{D}^{\perp}, so Gal⁡(F)\operatorname{Gal}(F) is nontrivial and FF is not algebraically closed. A PAC field of characteristic zero which is not algebraically closed has the independence property [8 Introduction]. This property is preserved by the canonical root-predicate expansion, which is definitional. Thus TT is unstable.

The theory TT is model complete by Fact 3.6, and TσT_\sigma has amalgamation by Proposition 5.2. Kikyo’s theorem states that if a complete theory is unstable and model complete and its automorphism theory has amalgamation, then that automorphism theory has no model companion [23 Theorem 3.3]. Applying this theorem to TT proves the conclusion.

Unboundedness and non-simplicity

In this section, we prove the unboundedness and non-simplicity assertions in Theorem 1.1.

Lemma 6.1. Let GG be a profinite group, and let HH be a nontrivial finite group. Suppose that HnH^n is a continuous quotient of GG for every positive integer nn. Then there are infinitely many open normal subgroups N⊴oGN \trianglelefteq_{\mathrm{o}} G such that G/N≅HG/N \cong H. In particular, GG is not small.

Proof. For every n≥1n \ge1, choose an epimorphism qn:G↠Hnq_n:G \twoheadrightarrow H^n. Composing with the nn coordinate projections gives epimorphisms from GG onto HH with pairwise distinct kernels. Thus, for every nn, there are at least nn open normal subgroups NN with G/N≅HG/N \cong H.

Theorem 6.2. Let GG be a nontrivial profinite group such that

Im⁡(G)=D⊥\operatorname{Im}(G)=\mathcal{D}^{\perp}

for a Melnikov formation D\mathcal{D}. Then, for every nontrivial finite quotient HH of GG, there are infinitely many open normal subgroups N⊴oGN \trianglelefteq_{\mathrm{o}} G such that G/N≅HG/N \cong H. In particular, GG is not small.

Proof. Let us fix a nontrivial finite quotient HH of GG. By Lemma 6.1, it is enough to show that Hn∈D⊥H^n \in\mathcal{D}^{\perp} for every positive integer nn. Suppose that

q:Hn↠Q,Q∈D.q:H^n \twoheadrightarrow Q,\qquad Q \in\mathcal{D}.

We aim to show that QQ is trivial. For each coordinate subgroup HiH_i of HnH^n, its image q(Hi)q(H_i) is normal in QQ. A Melnikov formation is closed under normal subgroups, so q(Hi)∈Dq(H_i) \in\mathcal{D}. On the other hand, q(Hi)q(H_i) is a quotient of HH, and H∈D⊥H \in\mathcal{D}^{\perp}. Hence q(Hi)=1q(H_i)=1 for every ii. The coordinate subgroups generate HnH^n, so Q=1Q=1 as desired.

Corollary 6.3. If D\mathcal{D} is proper and F∈KDF \in\mathcal{K}_{\mathcal{D}}, then Gal⁡(F)\operatorname{Gal}(F) is not small and Th⁡(F)\operatorname{Th}(F) is not simple.

Proof. By the definition of the class KD\mathcal{K}_{\mathcal{D}}, we have

Im⁡(Gal⁡(F))=D⊥.\operatorname{Im}(\operatorname{Gal}(F))=\mathcal{D}^{\perp}.

Since D\mathcal{D} is proper, Gal⁡(F)\operatorname{Gal}(F) is non-trivial and Theorem 6.2 implies that Gal⁡(F)\operatorname{Gal}(F) is not small. In particular, FF is unbounded. Every simple PAC field is bounded [9 Theorem 3.9], so Th⁡(F)\operatorname{Th}(F) is not simple. The same holds for T=Th⁡(FS)T=\operatorname{Th}(F_S), since the root-predicate expansion is definitional.

Corollary 6.3, together with the NSOP1_1 result recalled in § 3, proves part (ii) of Theorem 1.1. Sections 4 and 5 establish parts (i) and (iii), as well as the assertions concerning invariance, symmetry, existence and stationarity. Part (iv) is proved in § 7.

Comparison of D-independence with forking and Kim-independence

Let TT be the complete theory of the canonical root-predicate expansion of a field F∈KDF \in\mathcal{K}_{D}, and work in a monster model U⊨T\mathbb{U} \models T. Write ⊥f\mathrel{\perp^{\mathrm{f}}} for ordinary nonforking and ⊥K\mathrel{\perp^{K}} for Kim-independence. Since the root-predicate expansion is definitional, these relations agree with their counterparts in the field reduct. For subfields M⊆A,BM \subseteq A,B of a common field, write A⊥MldBA \mathrel{\perp^{\mathrm{ld}}_{M}} B if AA and BB are linearly disjoint over MM. We often write tp⁡(A/M)\operatorname{tp}(A/M) implicitly meaning some enumeration of AA.

For elementary submodels M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U}, the definition of our relation becomes

A⊥MDB⟺A⊥MldB and acl⁡U(AB)=(AB)D.A \mathrel{\perp^{\mathcal D}_{M}} B \Longleftrightarrow A \mathrel{\perp^{\mathrm{ld}}_{M}} B \text{ and } \operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal D}.

We obtain the implication from this relation to ordinary nonforking by constructing global invariant extensions. Lemma 7.2 records the abstract argument for relations on elementary submodels, and Proposition 7.3 applies it to our relation. Kim-independence implies linear disjointness and, in our characteristic-zero setting, a further condition on the algebraic closures, recalled in Fact 7.4. Among Kim-independent triples of elementary models, our relation is characterized by the additional equality acl⁡U(AB)=(AB)D\operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal D}. Theorem 7.7 proves that every implication in this comparison chain is strict when D\mathcal D is proper, whereas all four conditions coincide in the algebraically closed case.

Lemma 7.1. If M≼B≼C≼UM \preccurlyeq B \preccurlyeq C \preccurlyeq\mathbb{U}, M≼A≼UM \preccurlyeq A \preccurlyeq\mathbb{U}, and A⊥MDCA \mathrel{\perp^{\mathcal D}_{M}} C, then A⊥MDBA \mathrel{\perp^{\mathcal D}_{M}} B.

Proof. The assumption gives A⊥MldCA \mathrel{\perp^{\mathrm{ld}}_{M}} C, hence A⊥MldBA \mathrel{\perp^{\mathrm{ld}}_{M}} B.

Moreover, AC/ABAC/AB is regular. Indeed, C/BC/B is regular because B≼CB \preccurlyeq C. Linear disjointness identifies AB⊗BCAB \otimes_{B} C with a localization of A⊗MCA \otimes_{M} C and identifies its fraction field with ACAC. Thus AC/ABAC/AB is obtained by regular base change from C/BC/B.

Using Proposition 3.1(a), the assumption A⊥MDCA \mathrel{\perp^{\mathcal D}_{M}} C, and Lemma 3.3, we obtain

acl⁡U(AB)=acl⁡U(AC)∩(AB)alg=(AC)D∩(AB)alg=(AB)D.\begin{aligned} \operatorname{acl}_{\mathbb{U}}(AB)=\operatorname{acl}_{\mathbb{U}}(AC)\cap(AB)^{\mathrm{alg}} \\ &=(AC)^{\mathcal D}\cap(AB)^{\mathrm{alg}} \\ &=(AB)^{\mathcal D}. \end{aligned}

Together with A⊥MldBA \mathrel{\perp^{\mathrm{ld}}_{M}} B, this proves A⊥MDBA \mathrel{\perp^{\mathcal D}_{M}} B. □\square

The next lemma is the version for elementary submodels of the stationary-extension construction in [26 Fact 3.2(1) in the arXiv version]. We include the proof to specify the hypotheses needed here.

Lemma 7.2. Let T0T_{0} be a complete theory with monster model U\mathbb{U}. Suppose that a relation A⊥MBA \mathrel{\perp_{M}} B, defined for small elementary submodels M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U}, satisfies invariance, existence and stationarity over models in the sense of Lemma 5.1. Assume in addition right monotonicity:

for all M≼N≼N′≼UA⊥MN′⟹A⊥MN.\text{for all }M \preccurlyeq N \preccurlyeq N' \preccurlyeq\mathbb{U}\qquad A \mathrel{\perp_{M}} N' \Longrightarrow A \mathrel{\perp_{M}} N.

If A⊥MBA \mathrel{\perp_{M}} B, then the type tp⁡(A/B)\operatorname{tp}(A/B) has a global MM-invariant extension and A⊥MfBA \mathrel{\perp^{\mathrm{f}}_{M}} B.

Proof. For every small elementary model M≼N≼UM \preccurlyeq N \preccurlyeq\mathbb{U}, existence supplies a model AN≅MAA_{N}\cong_{M} A (the isomorphism is a partial elementary map) such that AN⊥MNA_{N}\mathrel{\perp_{M}}N. Stationarity implies that

pN:=tp⁡(AN/N)p_{N}:=\operatorname{tp}(A_{N}/N)

does not depend on the choice of ANA_{N}.

If M≼N≼N′≼UM \preccurlyeq N \preccurlyeq N' \preccurlyeq\mathbb{U}, right monotonicity gives AN′⊥MNA_{N'}\mathrel{\perp_{M}}N. Stationarity therefore yields

pN′∣N=pN.p_{N'|N}=p_{N}.

Thus the types pNp_N form a coherent family, and their union

q:=⋃M≼N≼UN smallpNq := \bigcup_{\substack{M\preccurlyeq N\preccurlyeq\mathbb{U}\\N\ \text{small}}} p_N

is consistent. Every finite tuple of parameters from U\mathbb{U} is in one such NN, so qq is complete over U\mathbb{U}.

For τ∈Aut⁡(U/M)\tau\in\operatorname{Aut}(\mathbb{U}/M), invariance and stationarity give

τ(pN)=pτ(N).\tau(p_N)=p_{\tau(N)}.

Consequently τ(q)=q\tau(q)=q, and qq is MM-invariant. Its restrictions have the stated independence property by construction and invariance. Any other global extension with this property must restrict to pNp_N over every NN, so it equals qq. Since A↓MBA \mathrel{\mathop{\downarrow}\limits_M} B, stationarity gives

q∣B=tp⁡(A/B).q|_B=\operatorname{tp}(A/B).

A type admitting a global MM-invariant extension does not fork over MM; see [22 Fact 7.2(3)].

Proposition 7.3. Let M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U} and suppose that A↓MDBA \mathrel{\mathop{\downarrow}\limits_M^{\mathcal D}} B. Then, for every small tuple aˉ\bar a from AA, the type tp⁡(aˉ/B)\operatorname{tp}(\bar a/B) extends to a global MM-invariant type. Consequently,

A↓MDB⟹A↓MfB⟹A↓MKB.A \mathrel{\mathop{\downarrow}\limits_M^{\mathcal D}} B \quad\Longrightarrow\quad A \mathrel{\mathop{\downarrow}\limits_M^{\mathrm f}} B \quad\Longrightarrow\quad A \mathrel{\mathop{\downarrow}\limits_M^{\mathrm K}} B.

Proof. Apply Lemma 7.2 to the relation ↓D\mathrel{\mathop{\downarrow}\limits^{\mathcal D}}. Invariance follows from its definition. Existence and stationarity are Propositions 4.1 and 4.2, and right monotonicity is Lemma 7.1. The lemma gives the required invariant extensions and ordinary nonforking. Ordinary nonforking implies Kim-independence over models by [22 Fact 7.2(3)].

Fact 7.4. In any complete theory expanding the theory of fields, let M≼A,BM\preccurlyeq A,B be elementary submodels of a common monster model U\mathbb{U}. If A↓MKBA \mathrel{\mathop{\downarrow}\limits_M^{\mathrm K}} B, then A↓MldBA \mathrel{\mathop{\downarrow}\limits_M^{\mathrm{ld}}} B and

acl⁡U(AB)∩AsepBsep=AB.\operatorname{acl}_{\mathbb{U}}(AB)\cap A^{\mathrm{sep}}B^{\mathrm{sep}}=AB.

Proof. Both assertions follow from [22 Proposition 9.28(1),(3)].

Corollary 7.5. For the theory T=Th⁡(FS)T=\operatorname{Th}(F_S) considered in this section, let M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U}. The following conditions are equivalent:

  1. A↓MDBA \mathrel{\mathop{\downarrow}\limits_M^{\mathcal D}} B;

  2. A↓MfBA \mathrel{\mathop{\downarrow}\limits_M^{\mathrm f}} B and acl⁡U(AB)=(AB)D\operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal D};

  3. A↓MKBA \mathrel{\mathop{\downarrow}\limits_M^{\mathrm K}} B and acl⁡U(AB)=(AB)D\operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal D}.

Proof. Proposition 7.3 and the definition of ↓D\mathrel{\mathop{\downarrow}\limits^{\mathcal D}} give (i)⇒\Rightarrow(ii). Ordinary nonforking implies Kim-independence, giving (ii)⇒\Rightarrow(iii). If (iii) holds, Fact 7.4 gives A↓MldBA \mathrel{\mathop{\downarrow}\limits_M^{\mathrm{ld}}} B. Together with the assumed closure equality, this is precisely the definition of (i).

Consequently, on triples M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U}, the relations satisfy

↓D⟹↓f⟹↓K⟹↓ld.\mathrel{\mathop{\downarrow}\limits^{\mathcal D}} \Longrightarrow \mathrel{\mathop{\downarrow}\limits^{\mathrm f}} \Longrightarrow \mathrel{\mathop{\downarrow}\limits^{\mathrm K}} \Longrightarrow \mathrel{\mathop{\downarrow}\limits^{\mathrm{ld}}}.

We next show that none of the three reverse implications holds when D\mathcal D is proper. The first non-implication uses the following standard consequence of the classical uniqueness-of-coheirs characterization of stability; see [1 Theorem 3.10].

Fact 7.6. Let T0T_0 be a complete theory with monster model U\mathbb{U}, and let ⊥\mathrel{\perp} be a relation on triples M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U} of small elementary submodels. Assume stationarity over models in the sense of Lemma 5.1. Suppose also that

tp⁡(A/B) is finitely satisfiable in M⟹A⊥MB.\operatorname{tp}(A/B)\text{ is finitely satisfiable in }M \quad\Longrightarrow\quad A \mathrel{\perp}_{M} B.

Then T0T_0 is stable.

Proof. Let M≼BM \preccurlyeq B and let q0,q1∈S(B)q_0,q_1 \in S(B) be finitely satisfiable in MM with the same restriction to MM. By extension of coheirs in additional variables, extend them to types over BB, finitely satisfiable in MM, whose common restriction to MM is the type of an elementary model containing MM. The hypothesis and stationarity imply that these extended types coincide, so q0=q1q_0=q_1. Since coheirs extend over larger parameter sets, this also gives uniqueness over arbitrary parameter sets containing MM. The characterization of stability cited above completes the proof. □\square

Theorem 7.7. If DD is proper, the following reverse implications fail on triples M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U}:

⊥D⇍⊥f⇍⊥K⇍⊥ld.\mathrel{\perp}^{D} \not\Leftarrow\mathrel{\perp}^{f} \not\Leftarrow\mathrel{\perp}^{K} \not\Leftarrow\mathrel{\perp}^{\mathrm{ld}}.

Proof. The theory TT is NSOP1_1 by the Frobenius-field result recalled in § 3. Since DD is proper, it is not simple by Corollary 6.3. In particular, TT is unstable.

The first converse. The relation ⊥D\mathrel{\perp}^{D} has stationarity over models by Proposition 4.2. Since TT is unstable, the contrapositive of Fact 7.6 gives M≼A,B≼UM \preccurlyeq A,B \preccurlyeq\mathbb{U} such that tp⁡(A/B)\operatorname{tp}(A/B) is finitely satisfiable in MM, but A⊥̸MDBA \mathrel{\not\perp}^{D}_{M} B. A type finitely satisfiable in MM extends to a global coheir, hence to a global MM-invariant type. It therefore does not fork over MM [22 Fact 7.2(3)]. Thus A⊥MfBA \mathrel{\perp}^{f}_{M} B but A⊥̸MDBA \mathrel{\not\perp}^{D}_{M} B, with tp⁡(A/B)\operatorname{tp}(A/B) finitely satisfiable in MM.

The middle converse. Since TT is NSOP1_1 and not simple, Kaplan–Ramsey’s criterion [22 Proposition 8.4] gives a model MM and tuples a,ba,b such that

a⊥MKb,a⊥̸Mfb.a \mathrel{\perp}^{K}_{M} b,\qquad a \mathrel{\not\perp}^{f}_{M} b.

We include the passage from tuples to elementary models. Choose B0≼UB_0 \preccurlyeq\mathbb{U} containing MbMb. By extension for Kim-independence, choose a′≡Mbaa' \equiv_{Mb} a with a′⊥MKB0a' \mathrel{\perp}^{K}_{M} B_0. An automorphism over MbMb taking a′a' to aa sends B0B_0 to a model BB containing MbMb, with a⊥MKBa \mathrel{\perp}^{K}_{M} B.

Choose A0≼UA_0 \preccurlyeq\mathbb{U} containing MaMa. By symmetry and extension, an enumerated copy B′≡MaBB' \equiv_{Ma} B can be chosen with B′⊥MKA0B' \mathrel{\perp}^{K}_{M} A_0. An automorphism over MaMa taking this enumeration of B′B' to the corresponding enumeration of BB sends A0A_0 to a model AA containing MaMa. Thus B⊥MKAB \mathrel{\perp}^{K}_{M} A, and symmetry gives A⊥MKBA \mathrel{\perp}^{K}_{M} B. If A⊥MfBA \mathrel{\perp}^{f}_{M} B, monotonicity of ordinary nonforking would give a⊥Mfba \mathrel{\perp}^{f}_{M} b, a contradiction.

The last converse. Let us choose a nontrivial finite simple group H∉DH \notin D. Since HH is simple, H∈D⊥H \in D^{\perp}. We fix a small model M⊨TM \models T. By Theorem 6.2, MM has infinitely many HH-Galois extensions (i.e. Galois extensions with Galois group isomorphic to HH) in MalgM^{\mathrm{alg}}.

Claim 1. Some elementary extension AA of MM has an HH-Galois extension L/AL/A such that L/ML/M is regular.

Proof of Claim 1. Choose pairwise distinct HH-Galois extensions Ki/MK_i/M, i∈Ni \in\mathbb{N}, inside MalgM^{\mathrm{alg}}. Since HH is simple, Ki∩Kj=MK_i \cap K_j=M for i≠ji \ne j. Let V\mathcal{V} be a nonprincipal ultrafilter on N\mathbb{N}, and put

A=MN/V,L=(∏i∈NKi)/V.(1)A=M^{\mathbb{N}}/\mathcal{V},\qquad L=\left(\prod_{i\in\mathbb{N}}K_i\right)/\mathcal{V}. \tag*{(1)}

Then M≼AM \preccurlyeq A via the diagonal embedding. After choosing isomorphisms H≅Gal⁡(Ki/M)H \cong\operatorname{Gal}(K_i/M), the induced coordinatewise action of HH on LL is faithful, and LH=AL^H=A, since HH is finite. Hence we obtain Gal⁡(L/A)≅H\operatorname{Gal}(L/A) \cong H.

An irreducible polynomial f∈M[X]f \in M[X] of degree greater than one has a root in at most one KiK_i: otherwise normality would put its nontrivial splitting field inside two of the KiK_i, contradicting their intersection being MM. By Łoś’s theorem, ff has no root in LL. Thus MM is relatively algebraically closed in LL, and characteristic zero gives that L/ML/M is regular. □\square

We take two MM-isomorphic copies (A1,L1)(A_1,L_1) and (A2,L2)(A_2,L_2) of (A,L)(A,L) from Claim 1 and define

P:=Frac⁡(L1⊗ML2),E:=A1A2.P := \operatorname{Frac}(L_1 \otimes_M L_2), \qquad E := A_1A_2.

Regularity of Li/ML_i/M ensures that the tensor product is a domain and that P/LiP/L_i is regular. We identify the fields with their canonical images in PP; in particular, A1⊥MldA2A_1 \mathrel{\mathop{\perp}\limits_M^{\mathrm{ld}}} A_2. We will embed PP in Ualg\mathbb{U}^{\mathrm{alg}} so that

A1A2≠acl⁡U(A1A2)∩A1algA2alg.A_1A_2 \ne\operatorname{acl}_{\mathbb{U}}(A_1A_2) \cap A_1^{\mathrm{alg}}A_2^{\mathrm{alg}}.

Then Fact 7.4 witnesses the failure of Kim-independence.

The two coordinate actions give ∣H∣2|H|^2 distinct EE-automorphisms of PP, while

[P:E]≤[L1:A1][L2:A2]=∣H∣2.[P:E] \leq[L_1:A_1][L_2:A_2] = |H|^2.

Hence P/EP/E is Galois and

Gal⁡(P/E)≅H×H.\operatorname{Gal}(P/E) \cong H \times H.

Let us define

ΔH:={(h,h):h∈H},C:=PΔH.\Delta_H := \{(h,h):h \in H\}, \qquad C := P^{\Delta_H}.

Then

[C:E]=∣H∣>1.[C:E] = |H| > 1.

Claim 2. The extension C/AiC/A_i is regular for i=1,2i=1,2.

Proof of Claim 2. Fix i∈{1,2}i \in\{1,2\}. Since P/LiP/L_i is regular and Li/AiL_i/A_i is finite, the relative algebraic closure of AiA_i in PP is LiL_i. Since ΔH\Delta_H projects surjectively onto each factor, its restriction to LiL_i is the full group Gal⁡(Li/Ai)\operatorname{Gal}(L_i/A_i). Thus

C∩Li=LiGal⁡(Li/Ai)=Ai.C \cap L_i = L_i^{\operatorname{Gal}(L_i/A_i)} = A_i.

It follows that AiA_i is relatively algebraically closed in CC. Since the characteristic is zero, C/AiC/A_i is regular. □\square

As Ai∈KDA_i \in\mathcal{K}_D, we have AiD=AiA_i^D=A_i. Lemma 3.3 and Claim 2 yield

CD∩Aialg=Ai,C^D \cap A_i^{\mathrm{alg}} = A_i,

so the extension CD/AiC^D/A_i is regular.

By Fact 3.2(b) and Fact 2.2(d),

RD(Gal⁡(CD))=Gal⁡(CD).R_D(\operatorname{Gal}(C^D)) = \operatorname{Gal}(C^D).

Lemma 3.4 therefore supplies a regular extension F′/CDF'/C^D with F′∈KDF' \in\mathcal{K}_D. Then, F′/AiF'/A_i is regular and Proposition 3.1(b) gives Ai≼F′A_i \preccurlyeq F'. Thus F′⊨TF' \models T, and we may embed F′F' into U\mathbb{U} over MM. We extend this embedding to an embedding of (F′)alg(F')^{\mathrm{alg}} into Ualg\mathbb{U}^{\mathrm{alg}}. In particular, PP is embedded into Ualg\mathbb{U}^{\mathrm{alg}} now as required above. Under this embedding, the images of LiL_i lie in AialgA_i^{\mathrm{alg}} and their compositum is the image of PP. Since C/EC/E is finite and C⊆UC \subseteq\mathbb{U}, we obtain

A1A2=E⊊C⊆acl⁡U(A1A2)∩A1algA2alg,A_1A_2 = E \subsetneq C \subseteq\operatorname{acl}_{\mathbb{U}}(A_1A_2) \cap A_1^{\mathrm{alg}}A_2^{\mathrm{alg}},

which gives A1⊥̸MKA2A_1 \mathrel{\mathop{\not\perp}\limits_M^K} A_2 as indicated above. □\square

Proposition 7.3 and Theorem 7.7 establish part (iv) of Theorem 1.1, completing its proof.

We finish this section with several remarks on the ternary relations considered above, their descriptions in the ω\omega-free case, and the failure of local character for ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}}.

Remark 7.8. Suppose that D\mathcal{D} is proper. The failure of the implication from ⊥K\mathrel{\mathop{\perp}\limits^{K}} to ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}} also follows from Fact 7.6: if the two relations coincided on triples of models, every coheir would satisfy ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}}, and stationarity would force TT to be stable.

Remark 7.9. Kaplan–Ramsey identify Kim-independence over models in Frobenius fields with Chatzidakis’s weak independence [22 Definition 9.31 and Theorem 9.33]; see also [8 Definition 2.3]. Their description combines independence in a separably closed overfield with nonforking in the stable theory of the associated Galois systems. In characteristic zero, the field-theoretic condition is algebraic independence in an algebraically closed overfield. Our relation imposes the closure condition acl⁡U(AB)=(AB)D\operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal{D}}. Theorem 7.7 shows that it gives a strict refinement of Kim-independence over models whenever D\mathcal{D} is proper.

Remark 7.10. Suppose that D={1}\mathcal{D}=\{1\}, so that FF is an ω\omega-free PAC field of characteristic zero and ED=EE^{\mathcal{D}}=E for every field EE. Let M≼A,B≼UM\preccurlyeq A,B\preccurlyeq\mathbb{U} be small elementary submodels. All three relations below require A⊥MBA\mathrel{\mathop{\perp}\limits_{M}}B. Under this common assumption, their additional conditions are

A⊥MDB⟺(AB)Ualg⁡=AB,A\mathrel{\mathop{\perp}\limits^{\mathcal{D}}_{M}}B\quad\Longleftrightarrow\quad(AB)^{\operatorname{alg}}_{\mathbb{U}}=AB,
A⊥MfB⟺(AB)Ualg⁡∩(AB0)alg⁡Balg⁡=(AB0)Ualg⁡BA\mathrel{\mathop{\perp}\limits^{f}_{M}}B\quad\Longleftrightarrow\quad(AB)^{\operatorname{alg}}_{\mathbb{U}}\cap(AB_{0})^{\operatorname{alg}}B^{\operatorname{alg}}=(AB_{0})^{\operatorname{alg}}_{\mathbb{U}}B

for every B0⊆BB_{0}\subseteq B with (B0)Ualg⁡=B0(B_{0})^{\operatorname{alg}}_{\mathbb{U}}=B_{0},

A⊥MKB⟺(AB)Ualg⁡∩Aalg⁡Balg⁡=AB.A\mathrel{\mathop{\perp}\limits^{K}_{M}}B\quad\Longleftrightarrow\quad(AB)^{\operatorname{alg}}_{\mathbb{U}}\cap A^{\operatorname{alg}}B^{\operatorname{alg}}=AB.

The first equivalence is our definition. The second is Chatzidakis’s description of ordinary nonforking in ω\omega-free PAC fields [7]; see also [12 Example 5.5] for this explicit formulation. The third follows from Kaplan–Ramsey’s identification of Kim-independence over models with weak independence [22 Theorem 9.33], together with the description of weak independence in [7].

Thus ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}} requires the whole compositum ABAB to be relatively algebraically closed in U\mathbb{U}, whereas ⊥K\mathrel{\mathop{\perp}\limits^{K}} only requires that no element of (AB)Ualg⁡∖AB(AB)^{\operatorname{alg}}_{\mathbb{U}}\setminus AB lie in Aalg⁡Balg⁡A^{\operatorname{alg}}B^{\operatorname{alg}}. Ordinary nonforking imposes the displayed family of intersection identities; taking B0=MB_{0}=M recovers the identity for Kim-independence. By Theorem 7.7, both implications in

⊥D⟹⊥f⟹⊥K\mathrel{\mathop{\perp}\limits^{\mathcal{D}}}\Longrightarrow\mathrel{\mathop{\perp}\limits^{f}}\Longrightarrow\mathrel{\mathop{\perp}\limits^{K}}

are strict, even on triples of elementary submodels.

Remark 7.11 (Failure of local character). The relation ⊥D\mathrel{\mathop{\perp}\limits^{\mathcal{D}}} admits a natural extension beyond elementary submodels. For relatively algebraically closed subfields C⊆A∩BC\subseteq A\cap B of U\mathbb{U}, put

A⊥CDB⟺A⊥Cid⁡B  and  acl⁡U(AB)=(AB)D.A\mathrel{\mathop{\perp}\limits^{\mathcal{D}}_{C}}B\quad\Longleftrightarrow\quad A\mathrel{\mathop{\perp}\limits^{\operatorname{id}}_{C}}B\ \text{ and }\ \operatorname{acl}_{\mathbb{U}}(AB)=(AB)^{\mathcal{D}}.

For arbitrary small sets X,Y,ZX,Y,Z, define X⊥ZDYX\mathrel{\mathop{\perp}\limits^{\mathcal{D}}_{Z}}Y by applying this condition to C=acl⁡U(Z)C=\operatorname{acl}_{\mathbb{U}}(Z), A=acl⁡U(X∪Z)A=\operatorname{acl}_{\mathbb{U}}(X\cup Z), B=acl⁡U(Y∪Z)B=\operatorname{acl}_{\mathbb{U}}(Y\cup Z).

This extended relation satisfies invariance, symmetry, existence, extension, monotonicity, base monotonicity, transitivity, finite character, and stationarity over algebraically closed bases. We briefly indicate the verifications. The proofs of Propositions 4.1 and 4.2 apply to relatively algebraically closed subfields: such a field AA satisfies AD=AA^{D}=A, and Lemma 3.4 and Proposition 3.1(c) give the required realizations and partial elementary maps. Monotonicity follows from Lemma 3.3; base monotonicity and transitivity follow by regular base change and Fact 2.2(c). Finite character follows by descending finite Galois extensions along directed unions of the relevant regular composita. Extension then follows from existence, monotonicity and stationarity.

If DD is proper, however, local character fails. Otherwise, the properties established above, including stationarity over models, would imply that TT is stable by the axiomatic characterization of stability recalled in [25 Section 1, discussion preceding Definition 1.3]. This contradicts Corollary 6.3, since every stable theory is simple.

Existential closedness for fields with group actions

We study existentially closed and pseudo-existentially closed fields with group actions. We begin with the general closure notions and the ultraproduct criterion for e.c. models. Subsection 8.1 applies Corollary 5.3 to rule out a model companion for the theory TS,σT_{S,\sigma} introduced in § 3. We then compare p.e.c. difference fields with p.e.c. fields carrying finite group actions in Subsections 8.2 and 8.3.

Let GG be an arbitrary group, and use the language

LG={0,1,+,−,⋅}∪{σg:g∈G}.L_G=\{0,1,+,-,\cdot\}\cup\{\sigma_g:g\in G\}.

A GG-field is a field with an action of GG by automorphisms; faithfulness is not part of this definition. We initially allow every characteristic, including imperfect fields. A field extension is regular if it is separable and the base field is relatively algebraically closed in it. The following terminology refers to the ordinary language LGL_G; see [3 Definition 2.1 and Remark 2.3]. In particular, p.e.c. always refers to this ordinary-language notion.

Definition 8.1. A GG-field KK is

(a) existentially closed, or e.c., if it is existentially closed in every GG-field extension;

(b) pseudo-existentially closed, or p.e.c., if it is existentially closed in every GG-field extension whose underlying field extension is regular;

(c) algebraically GG-closed, or simply GG-closed, if it has no proper algebraic GG-field extension.

In the last condition, algebraic extensions may be infinite.

For the trivial group, p.e.c. fields are exactly PAC fields [13 Proposition 11.3.5] and e.c. fields are exactly algebraically closed fields. For G=ZG=\mathbb{Z}, a GG-field is a difference field with a distinguished automorphism.

The following observation is recorded in [3 Remark 2.3(0)–(1)]. Neither finiteness nor finite generation of GG is needed.

Fact 8.2. For every group GG and every GG-field KK,

K is e.c.⟺K is p.e.c. and algebraically G-closed.K\text{ is e.c.}\quad\Longleftrightarrow\quad K\text{ is p.e.c. and algebraically }G\text{-closed.}

For a first-order theory T0T_0, let EC⁡(T0)\operatorname{EC}(T_0) denote the class of all models of T0T_0 that are existentially closed in every extension which is also a model of T0T_0. Here elementary means axiomatizable by a possibly incomplete first-order theory. We recall the following standard facts; see also [19 Introduction and discussion following Fact 5.3] for a recent discussion of the ultraproduct criterion.

Fact 8.3. Let T0T_0 be a first-order theory.

(a) The class EC⁡(T0)\operatorname{EC}(T_0) is closed under elementary substructures.

(b) The class EC⁡(T0)\operatorname{EC}(T_0) is elementary if and only if it is closed under ultraproducts.

(c) If T0T_0 is consistent and inductive, the equivalent conditions in (b) are also equivalent to the existence of a model companion of T0T_0.

Proof. Part (a) follows from the elementary-substructure observation in [27 Section 3.2]. For (b), part (a) gives closure under ultraroots: if an ultrapower of MM belongs to EC⁡(T0)\operatorname{EC}(T_0), then so does MM, since the diagonal embedding is elementary. As EC⁡(T0)\operatorname{EC}(T_0) is also closed under isomorphisms, the assertion follows from the ultraproduct–ultraroot criterion [18 Corollary 16.2.6]. Part (c) is [6 Proposition 3.5.15]. □

Parts (a) and (b) require no completeness, inductiveness, or amalgamation hypothesis on T0T_0. Inductiveness is used only in (c).

For an inductive theory, nonexistence of a model companion therefore already implies failure of closure of its e.c. class under ultraproducts. This gives an ultraproduct of possibly different e.c. models which is not e.c. Producing such a witness with all factors equal is a stronger assertion and requires an additional argument. The orbit construction below provides a witness of this more precise kind.

The criterion concerns existential closure in a fixed first-order language with respect to all embeddings between models of a fixed theory. It cannot be applied to p.e.c. difference fields merely by restricting the permitted extensions to regular ones. Although root predicates make embeddings correspond to regular extensions, they also change which formulas are existential; see Remark 8.4.

We next introduce the languages used for difference fields, that is, for G=ZG = \mathbb{Z}, with a generator acting as σ\sigma. The ordinary difference language is

Lσ={0,1,+,−,⋅,σ}.L_\sigma= \{0,1,+,-,\cdot,\sigma\}.

where σ\sigma is required to be an automorphism. Naming all positive and negative iterates gives LZL_{\mathbb{Z}}. Inverse iterates can be eliminated from existential formulas by introducing variables and equations of the form σ(y)=x\sigma(y) = x, so the p.e.c. notions in the two languages agree.

Existentially closed models in the root-predicate language

We first study e.c. models of the theory TS,σT_{S,\sigma} from § 3, and deduce the nonexistence of a model companion from Corollary 5.3. We begin by distinguishing existential closedness in the expanded language from the ordinary-language p.e.c. condition.

Remark 8.4. The allowable extensions are the same as in the definition of p.e.c., but the expanded language has additional existential formulas. For example,

∃a(σ(a)=a∧¬S2(0,−a))\exists a\left(\sigma(a)=a \wedge\neg S_2(0,-a)\right)

is existential in LS,σL_{S,\sigma}, whereas its defining translation into LσL_\sigma is

∃a(σ(a)=a∧∀x(x2≠a)).\exists a\left(\sigma(a)=a \wedge\forall x\left(x^2 \ne a\right)\right).

For any difference field (K,σ)(K,\sigma), extend σ\sigma to K(t)K(t) by fixing a transcendental element tt. The extension K(t)/KK(t)/K is regular, and the displayed expanded-language formula holds with witness a=ta=t. It follows that every e.c. model of TS,σT_{S,\sigma} contains an element fixed by σ\sigma that is a nonsquare in its underlying field.

On the other hand, a model (K,σ)(K,\sigma) of ACFA⁡0\operatorname{ACFA}_0 is e.c. in the ordinary difference language, and hence p.e.c., but its underlying field is algebraically closed. The displayed formula fails in KK, so (K,σ)(K,\sigma) is not e.c. as a model of TS,σT_{S,\sigma}.

Consequently, the LσL_\sigma-reducts of the e.c. models of TS,σT_{S,\sigma} form a proper subclass of the p.e.c. characteristic-zero difference fields. Inclusion follows from Proposition 3.7, and the ACFA⁡0\operatorname{ACFA}_0 example shows that it is strict. Indeed, none of those reducts is a model of ACFA⁡0\operatorname{ACFA}_0.

Let TωT_\omega be the theory of characteristic-zero ω\omega-free PAC fields in the canonical root-predicate language. It is the model companion of the theory of characteristic-zero fields in that language [13 Theorem 27.2.3].

Proposition 8.5. If (K,σ)S(K,\sigma)_S is an e.c. model of TS,σT_{S,\sigma}, then KK is an ω\omega-free PAC field.

Proof. Let E/KE/K be a regular field extension. Take independent copies EiE_i of EE, indexed by i∈Zi \in\mathbb{Z}, with isomorphisms ιi:E⟶Ei\iota_i:E \longrightarrow E_i satisfying ιi∣K=σi\iota_i|_K=\sigma^i, and take E0=EE_0=E. Let LL be the fraction field of their tensor product over KK. Regularity of E/KE/K implies that this tensor product is a domain and that L/EiL/E_i is regular for every ii. The shift

ιi(e)⟼ιi+1(e)\iota_i(e) \longmapsto\iota_{i+1}(e)

defines an automorphism τ\tau of LL extending σ\sigma. Since L/EL/E is regular, every existential LS\mathcal{L}_S-formula over KK realized in ESE_S is also realized in LSL_S. Existential closedness of (K,σ)S(K,\sigma)_S reflects it back to KSK_S. Thus KSK_S is existentially closed among characteristic-zero fields with their canonical root-predicate expansions. By [13 Theorem 27.2.3], these existentially closed structures are precisely the canonical expansions of ω\omega-free PAC fields. Hence KK is an ω\omega-free PAC field, as required. □

Theorem 8.6. The theory TS,σT_{S,\sigma} has no model companion.

Proof. Suppose that T∗T^* is a model companion of TS,σT_{S,\sigma}, and choose a completion TT of TωT_\omega. Since TS,σT_{S,\sigma} is inductive by Fact 3.8, the models of T∗T^* are precisely its e.c. models. Their LS\mathcal{L}_S-reducts therefore satisfy TωT_\omega by Proposition 8.5.

Every model (K,σ)S(K,\sigma)_S of TσT_\sigma embeds in a model (M,τ)S(M,\tau)_S of T∗T^*. Model completeness of TωT_\omega gives KS≼MSK_S \preccurlyeq M_S, so MS⊨TM_S \models T. Consequently, T∗∪TT^* \cup T is a model companion of TσT_\sigma: it is model complete, its models satisfy TσT_\sigma, and every model of TσT_\sigma embeds in one of its models. For D={1}\mathcal{D}=\{1\}, the class KD\mathcal{K}_{\mathcal{D}} consists precisely of the characteristic-zero ω\omega-free PAC fields. Corollary 5.3 therefore rules out such a model companion of TσT_\sigma, a contradiction. □

Remark 8.7. Since TS,σT_{S,\sigma} is inductive by Fact 3.8, Theorem 8.6 and Fact 8.3(c) imply that its e.c. models are not closed under ultraproducts.

In fact, closure under ultrapowers also fails. We omit the proof, which follows the orbit construction in the proof of Theorem 8.10 below, with the initial regular extension of (Q(z),σ)(\mathbb{Q}(z),\sigma), σ(z)=z+1\sigma(z)=z+1, chosen e.c. in the expanded language using Fact 3.8. Proposition 3.7 ensures that the regular extension used in that construction is admissible in the expanded language.

P.e.c. difference fields

We now work in the ordinary difference language Lσ\mathcal{L}_\sigma. Throughout this subsection the characteristic is zero, so regularity means that the base field is relatively algebraically closed in the extension. We use the preservation of regularity under subextensions over the fixed base, towers, and directed unions. The p.e.c. notion for difference fields is also discussed in [4 Remark 4.9 and Question 4.10].

Taking Lσ\mathcal{L}_\sigma-reducts of the e.c. extensions supplied by Fact 3.8 gives the following consequence.

Fact 8.8. Every characteristic-zero difference field (K,σ)(K,\sigma) has a regular difference-field extension (M,τ)(M,\tau) that is p.e.c.

Proof. Apply Fact 3.8 to the canonical expansion (K,σ)S(K,\sigma)_S, obtaining an e.c. TS,σT_{S,\sigma}-extension (M,τ)S(M,\tau)_S with M/KM/K regular. By Proposition 3.7, every regular difference-field extension of (M,τ)(M,\tau) canonically expands to a TS,σT_{S,\sigma}-extension of (M,τ)S(M,\tau)_S. Every existential formula in the ordinary difference language is also existential in the expanded language. Consequently, existential closure of (M,τ)S(M,\tau)_S implies that (M,τ)(M,\tau) is p.e.c. □

Lemma 8.9. Let (M,σ)(M,\sigma) be a characteristic-zero difference field, and let a,b∈Ma,b \in M have disjoint infinite orbits under σ\sigma. Then there is a regular difference-field extension (L,σ)/(M,σ)(L,\sigma)/(M,\sigma) containing t,u,vt,u,v such that

σ(t)=t,u2=t−a,v2=b−t,uv≠0.\sigma(t)=t,\qquad u^2=t-a,\qquad v^2=b-t,\qquad uv\ne0.

Proof. Write αi=σi(a)\alpha_i=\sigma^i(a) and βi=σi(b)\beta_i=\sigma^i(b) for i∈Zi\in\mathbb{Z}. By hypothesis all elements in these two indexed families are pairwise distinct. Take tt transcendental over MM and form

L=M(t)(ui,vi:i∈Z),ui2=t−αi,vi2=βi−t.L=M(t)(u_i,v_i:i\in\mathbb{Z}),\qquad u_i^2=t-\alpha_i,\qquad v_i^2=\beta_i-t.

This is a standard construction, see e.g. the proof of [5 Lemma 3.1], and the extension L/ML/M is regular. We extend σ\sigma to LL by setting

σ(t)=t,σ(ui)=ui+1,σ(vi)=vi+1.\sigma(t)=t,\qquad\sigma(u_i)=u_{i+1},\qquad\sigma(v_i)=v_{i+1}.

The elements t,u0,v0t,u_0,v_0 have the required properties; the two roots are nonzero because tt is transcendental over MM. □

Theorem 8.10. The class of p.e.c. difference fields of characteristic zero is not closed under ultrapowers. In particular, it is not elementary.

Proof. Let aa be transcendental and σ∈Aut⁡(Q(a))\sigma\in\operatorname{Aut}(\mathbb{Q}(a)) such that σ(a)=a+1\sigma(a)=a+1. By Fact 8.8, there is a p.e.c. regular difference-field extension (K,σ)/(Q(a),σ)(K,\sigma)/(\mathbb{Q}(a),\sigma). Since −1-1 is not a square in Q(a)\mathbb{Q}(a) and Q(a)\mathbb{Q}(a) is relatively algebraically closed in KK, it is not a square in KK either. We consider the following existential formula

ψ(y,z):=∃x∃u∃v σ(x)=x∧u2=x−y∧v2=z−x∧uv≠0.\psi(y,z):=\exists x\exists u\exists v\ \sigma(x)=x\wedge u^2=x-y\wedge v^2=z-x\wedge uv\ne0.

For every positive integer nn, one has

K⊨¬ψ(a,σn(a)).K\models\neg\psi(a,\sigma^n(a)).

Otherwise, applying σn\sigma^n to u2=x−au^2=x-a gives (σn(u))2=x−σn(a)=−v2(\sigma^n(u))^2=x-\sigma^n(a)=-v^2. Since uu and vv are nonzero, this would make −1-1 a square in KK.

Let V\mathcal{V} be a nonprincipal ultrafilter on the positive integers and let M=KN>0/VM=K^{\mathbb{N}_{>0}}/\mathcal{V}. In MM, write aa also for the constant class of aa, and put

b=[(σn(a))n>0]V.b=[(\sigma^n(a))_{n>0}]_{\mathcal{V}}.

By Łoś’s theorem, M⊨¬ψ(a,b)M\models\neg\psi(a,b). Nevertheless, aa and bb have disjoint infinite orbits in MM. Indeed, their iterates are a+ia+i and b+jb+j for i,j∈Zi,j\in\mathbb{Z}. Each orbit has distinct elements, and an equality between its two displayed forms would require

{n>0:a+i=a+n+j}∈V.\{n>0:a+i=a+n+j\}\in\mathcal{V}.

This set contains at most one integer, which is impossible. Lemma 8.9 now supplies a regular difference-field extension of MM satisfying ψ(a,b)\psi(a,b). Thus MM is not p.e.c., although KK is. □

Remark 8.11. The non-elementarity assertion of Theorem 8.10 does not follow formally from either Corollary 5.3 or Theorem 8.6. Those results concern existential closure in the root-predicate language. As Remark 8.4 shows, the ordinary reducts of its e.c. models form a proper subclass of the p.e.c. difference fields. Non-elementarity of that subclass does not imply non-elementarity of the entire p.e.c. class.

Finite group actions

We now assume that GG is finite and again allow arbitrary characteristic, including imperfect fields. This subsection concerns p.e.c. in the ordinary language LGL_G. The elementarity of this class is recorded in [17 Remark 6.25]. We give an explicit characterization in terms of faithfulness and the PAC property of the fixed field.

Theorem 8.12. A GG-field KK is p.e.c. if and only if the action of GG is faithful and the fixed field C=KGC=K^G is PAC.

Proof. Suppose first that KK is p.e.c. Adjoin algebraically independent variables (th)h∈G(t_h)_{h\in G} and extend the action to K(th:h∈G)K(t_h:h\in G) by σg(th)=tgh\sigma_g(t_h)=t_{gh}. This is a regular GG-field extension and, for every g≠1g\ne1, it satisfies ∃x (σg(x)≠x)\exists x\,(\sigma_g(x)\ne x). The same sentence holds in KK, so the action is faithful. Artin’s fixed-field theorem gives Gal⁡(K/C)≅G\operatorname{Gal}(K/C)\cong G and [K:C]=∣G∣[K:C]=|G|. Let E/CE/C be a regular field extension. The tensor product K⊗CEK \otimes_{C} E is a domain and is finite dimensional over EE, so it is a field LL. The extension L/KL/K is regular by base change. Extend the action of GG by fixing EE pointwise. An existential field formula over CC realized in EE is therefore realized in LL by coordinates fixed by every element of GG. Adjoin these finitely many fixed-point equations to the formula. The p.e.c. property gives such a witness in KK, and its coordinates lie in CC. Thus CC is existentially closed in every regular extension, and hence is PAC.

Conversely, suppose that the action is faithful and CC is PAC. Let L/KL/K be a regular GG-field extension and put D=LGD = L^{G}. The action on LL is faithful, and Artin’s theorem gives

Gal⁡(K/C)≅G≅Gal⁡(L/D),[K:C]=[L:D]=∣G∣.\operatorname{Gal}(K/C) \cong G \cong\operatorname{Gal}(L/D), \qquad[K:C] = [L:D] = |G|.

The subgroup of Gal⁡(L/D)\operatorname{Gal}(L/D) fixing KDKD is trivial, because the action on KK is faithful. Hence L=KDL = KD. It follows from the equality of degrees that a CC-basis of KK remains a DD-basis of LL. The extension D/CD/C is regular. Indeed, K/CK/C is finite separable and L/KL/K is separable, so L/CL/C, and therefore D/CD/C, is separable. If d∈Dd \in D is algebraic over CC, then it is algebraic over KK. Regularity of L/KL/K gives d∈Kd \in K, and therefore d∈K∩D=Cd \in K \cap D = C.

Choose a CC-basis e1,…,eme_1,\ldots,e_m of KK. Write each element of LL uniquely as ∑ieiyi\sum_i e_i y_i, with yi∈Dy_i \in D. Multiplication and the maps σg\sigma_g have structure constants in CC in this basis; the same formulas describe their actions on LL, since GG fixes DD pointwise. Parameters from KK also have coordinates in CC. Consequently, an existential LGL_G-formula over KK translates into an existential field formula over CC: equality becomes equality of all coordinates, and inequality becomes a disjunction of coordinate inequalities. A witness in LL gives a coordinate witness in DD. Since CC is PAC and D/CD/C is regular, there is a coordinate witness in CC, which gives a witness to the original formula in KK. Thus KK is p.e.c. □

Corollary 8.13. For each fixed finite group GG, each of the following classes of GG-fields is elementary:

  • (a) the p.e.c. GG-fields;

  • (b) the algebraically GG-closed GG-fields;

  • (c) the e.c. GG-fields.

These assertions hold without a restriction on characteristic.

Proof. For the first assertion, add to the field and action axioms the faithfulness sentences

∃x  (σg(x)≠x)(g∈G∖{1}).\exists x\;(\sigma_g(x) \ne x) \qquad(g \in G \setminus\{1\}).

The fixed field is defined by the single formula

Fix⁡(x):=⋀g∈Gσg(x)=x.\operatorname{Fix}(x) := \bigwedge_{g \in G} \sigma_g(x) = x.

Relativize the PAC axiom scheme to this formula. Theorem 8.12 identifies the models of the resulting theory with the p.e.c. GG-fields.

For the second assertion, any proper algebraic GG-field extension L/KL/K contains a proper finite GG-field extension: for a∈L∖Ka \in L \setminus K algebraic over KK, take K(σg(a):g∈G)K(\sigma_g(a):g \in G). For each n≥2n \ge2, the existence of an extension of degree nn carrying the prescribed GG-action is first-order expressible over KK. One codes an nn-dimensional field extension by multiplication constants on KnK^n, taking e1=(1,0,…,0)e_1=(1,0,\ldots,0) as its unit and Ke1Ke_1 as its scalar copy of KK. Each σg\sigma_g is coded by a semilinear matrix, using the prescribed σg\sigma_g on scalar coefficients and fixing e1e_1. The field axioms, compatibility with multiplication, and the finitely many group-action identities are first-order conditions. Negating these sentences for all n≥2n \ge2 axiomatizes algebraic GG-closedness.

The third assertion follows from the first two and Fact 8.2. □

The theory of e.c. GG-fields is the model companion GG-TCF studied in [20].

Remark 8.14 (A p.e.c. field that is not e.c.). Let CC be a pseudofinite field of characteristic zero. Then CC is PAC and Gal⁡(C)≅Z^\operatorname{Gal}(C) \cong\widehat{\mathbb{Z}}. Choose a cyclic extension L/CL/C of degree 6, let K/CK/C be its quadratic subextension, and let τ\tau generate Gal⁡(L/C)\operatorname{Gal}(L/C). Equip KK with the action of G=Z/2ZG = \mathbb{Z}/2\mathbb{Z} whose nonidentity element acts as τ3∣K\tau^{3}|_{K}. This action is faithful and KG=CK^{G} = C, so Theorem 8.12 makes KK p.e.c. The automorphism τ3\tau^{3} on LL has order 2 and extends the given action. Hence L/KL/K is a proper algebraic GG-field extension, and KK is not e.c.

General conclusions

The decomposition in Fact 8.2 holds for every group. Its consequences for axiomatizability differ between finite groups and Z\mathbb{Z}. The following observation about Z\mathbb{Z}-closedness holds in every characteristic.

Proposition 8.15. A difference field (K,σ)(K,\sigma) is algebraically Z\mathbb{Z}-closed if and only if KK is algebraically closed. Consequently, in every characteristic,

(K,σ)⊨ACFA⁡⟺((K,σ) is p.e.c.) and K⊨ACF⁡.(K,\sigma) \models\operatorname{ACFA} \Longleftrightarrow((K,\sigma)\text{ is p.e.c.})\text{ and }K \models\operatorname{ACF}.

Proof. Every automorphism σ\sigma of KK extends to an automorphism σ~\widetilde{\sigma} of an algebraic closure KalgK^{\mathrm{alg}}. Thus (Kalg,σ~)/(K,σ)(K^{\mathrm{alg}},\widetilde{\sigma})/(K,\sigma) is an algebraic Z\mathbb{Z}-field extension, and algebraic Z\mathbb{Z}-closedness forces K=KalgK = K^{\mathrm{alg}}. The converse is immediate. The final equivalence follows from Fact 8.2 and the classical axiomatization of e.c. difference fields by ACFA; see [10]. □

For clarity, write PG,0\mathcal{P}_{G,0}, AG,0\mathcal{A}_{G,0}, and EG,0\mathcal{E}_{G,0} for the classes of characteristic-zero GG-fields that are respectively p.e.c., algebraically GG-closed, and e.c. Fact 8.2 gives, for every group GG,

EG,0=PG,0∩AG,0.\mathcal{E}_{G,0} = \mathcal{P}_{G,0} \cap\mathcal{A}_{G,0}.

The elementary-class behavior differs as follows:

ClassGG finiteG=ZG = \mathbb{Z}
PG,0\mathcal{P}_{G,0}elementarynot elementary
AG,0\mathcal{A}_{G,0}elementaryelementary
EG,0\mathcal{E}_{G,0}elementaryelementary

Table 1.

The finite-group column follows from Corollary 8.13. For G=ZG = \mathbb{Z}, the last two entries follow from Proposition 8.15 and the existence of ACFA⁡0\operatorname{ACFA}_{0}. The first entry is the separate non-elementarity result of Theorem 8.10. In particular,

PZ,0∩{(K,σ):K⊨ACF⁡0}=Mod⁡(ACFA⁡0).\mathcal{P}_{\mathbb{Z},0} \cap\{(K,\sigma) : K \models\operatorname{ACF}_{0}\} = \operatorname{Mod}(\operatorname{ACFA}_{0}).

Thus p.e.c. is axiomatizable on algebraically closed underlying fields, although it is not axiomatizable among all characteristic-zero difference fields. An elementary intersection with an elementary class does not force the other class to be elementary.

Corollary 8.16. Existence of a model companion for the theory of GG-fields, even together with axiomatizability of algebraic GG-closedness, does not imply axiomatizability of the p.e.c. GG-fields. This failure already occurs for G=ZG = \mathbb{Z} in characteristic zero.

Proof. For G=ZG = \mathbb{Z}, the model companion is ACFA⁡0\operatorname{ACFA}_{0}, algebraic GG-closedness is axiomatized by ACF⁡0\operatorname{ACF}_{0} together with the automorphism axioms, and Theorem 8.10 gives non-elementarity of the p.e.c. class. □

Use of AI. The authors used ChatGPT (GPT-6.0 with reasoning effort set to Ultra) to assist in developing proofs and drafting portions of the text. The authors refined and streamlined the arguments and assume full responsibility for the final content. Boğaziçi Üniversitesi

Email address: [email protected]

Instytut Matematyczny, Uniwersytet Wrocławski, Wrocław, Poland

Email address: [email protected]

URL: http://www.math.uni.wroc.pl/~pkowa/

References

  1. [1]I. Ben Yaacov, Stability and stable groups in continuous logic, Journal of Symbolic Logic 75 (2010), no. 3, 1111–1136.
  2. [2]Ö. Beyarslan and P. Kowalski, Model theory with virtually free group actions, Proceedings of the London Mathematical Society 118 (2019), no. 2, 221–256. Corrigendum: Proceedings of the London Mathematical Society 128 (2024), no. 4, e12597.DOI
  3. [3]Ö. Beyarslan and P. Kowalski, Model theory of Galois actions of torsion Abelian groups, Journal of the Institute of Mathematics of Jussieu 22 (2023), no. 6, 2943–2985.arxiv.org/abs/2003.02329
  4. [4]Ö. Beyarslan and P. Kowalski, Galois actions of finitely generated groups rarely have model companions, Bulletin of the London Mathematical Society 56 (2024), no. 2, 847–859.arxiv.org/abs/2210.00800
  5. [5]Ö. Beyarslan and P. Kowalski, PAC fields and difference fields with generic automorphisms, preprint (2026). https://arxiv.org/abs/2609.40251.
  6. [6]C. C. Chang and H. J. Keisler, Model Theory, third edition, Studies in Logic and the Foundations of Mathematics, vol. 73, North-Holland, Amsterdam, 1990.
  7. [7]Z. Chatzidakis, Properties of forking in ω-free pseudo-algebraically closed fields, Journal of Symbolic Logic 67 (2002), no. 3, 957–996.DOI
  8. [8]Z. Chatzidakis, Amalgamation of types in pseudo-algebraically closed fields and applications, Journal of Mathematical Logic 19 (2019), no. 2, Article 1950006.arxiv.org/abs/1704.08132
  9. [9]Z. Chatzidakis, Simplicity and independence for pseudo-algebraically closed fields, in Models and Computability, S. B. Cooper and J. K. Truss (eds.), London Mathematical Society Lecture Note Series, vol. 259, Cambridge University Press, Cambridge, 1999, pp. 41–62.DOI
  10. [10]Z. Chatzidakis and E. Hrushovski, Model theory of difference fields, Transactions of the American Mathematical Society 351 (1999), no. 8, 2997–3071.
  11. [11]G. Cherlin, L. van den Dries and A. Macintyre, The elementary theory of regularly closed fields, unpublished manuscript. https://sites.math.rutgers.edu/~cherlin/Preprint/CDM2.pdf.
  12. [12]J. Dobrowolski and D. M. Hoffmann, On rank not only in NSOP₁ theories, Journal of Symbolic Logic 89 (2024), no. 4, 1669–1702.arxiv.org/abs/2111.02389
  13. [13]M. D. Fried and M. Jarden, Field Arithmetic, third edition, Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge, vol. 11, Springer, Berlin–Heidelberg, 2008.
  14. [14]M. D. Fried and H. Völklein, The embedding problem over a Hilbertian PAC-field, Annals of Mathematics (2) 135 (1992), no. 3, 469–481.DOI
  15. [15]S. Fried and D. Haran, Quasi-formations, Israel Journal of Mathematics 229 (2019), no. 1, 193–217.DOI
  16. [16]S. Fried and D. Haran, Θ-Hilbertianity, Journal of Algebra 555 (2020), 36–51.DOI
  17. [17]J. Gogolok, Existentially closed fields with operators in various categories, preprint (2026). https://arxiv.org/abs/2606.17038.
  18. [18]I. Goldbring, Ultrafilters Throughout Mathematics, Graduate Studies in Mathematics, vol. 220, American Mathematical Society, Providence, RI, 2022.DOI
  19. [19]I. Goldbring, B. Seward and R. Tucker-Drob, Existentially closed measure-preserving actions of approximately treeable groups, preprint (2025). https://arxiv.org/abs/2507.03195v1.
  20. [20]D. M. Hoffmann and P. Kowalski, Existentially closed fields with finite group actions, Journal of Mathematical Logic 18 (2018), no. 1, 1850003.DOI
  21. [21]M. Jarden, On free profinite groups of uncountable rank, in Recent Developments in the Inverse Galois Problem (Seattle, WA, 1993), S. S. Abhyankar, W. Feit, M. D. Fried, Y. Ihara and H. Völklein (eds.), Contemporary Mathematics, vol. 186, American Mathematical Society, Providence, RI, 1995, pp. 371–383.DOI
  22. [22]I. Kaplan and N. Ramsey, On Kim-independence, Journal of the European Mathematical Society 22 (2020), no. 5, 1423–1474.arxiv.org/abs/1702.03894
  23. [23]H. Kikyo, Model companions of theories with an automorphism, Journal of Symbolic Logic 65 (2000), no. 3, 1215–1222.DOI
  24. [24]H. Kikyo, On the amalgamation property for automorphisms, RIMS Kôkyûroku 1602 (2008), 93–102.
  25. [25]A. Martin-Pizarro, Non-forking independence in stable theories, preprint (2024). https://arxiv.org/abs/2410.09887.
  26. [26]S. Mutchnik, Conant-independence and generalized free amalgamation, Journal of Mathematical Logic 26 (2026), no. 1, Article 2450028, 32 pp. arXiv version: https://arxiv.org/abs/2210.07527v6.
  27. [27]K. Tent and M. Ziegler, A Course in Model Theory, Lecture Notes in Logic, vol. 40, Cambridge University Press, Cambridge, 2012.DOI

Paper details

vlnc:W_22ywutt4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents