--- title: Gaifman Conjecture in Model Theory url: https://www.emergentmind.com/topics/gaifman-conjecture type: topic --- # Gaifman Conjecture in Model Theory Searching arXiv for recent papers directly relevant to the Gaifman Conjecture and closely related uses of the term “Gaifman.” The Gaifman Conjecture is a model-theoretic conjecture about a complete first-order theory \(T\) equipped with a distinguished unary predicate \(P\). In its standard form, it asserts that if \(T\) is relatively categorical over \(P\), then \(T\) has the Gaifman property: every model of the induced theory on the \(P\)-part occurs as the \(P\)-part of some model of \(T\) [2602.05866]. Recent work places this existence problem in the framework of Classification Theory, recasting it as part of a broader structure/non-structure program over \(P\) and proving a strong stability-theoretic sufficient condition for the Gaifman property in countable theories [2507.12631]. ## 1. Origins and historical formulation The conjecture originates in Gaifman’s work on “single-valued operations,” where the basic theme is reconstruction of an ambient structure from a distinguished part. Pillay describes the historical setting through examples such as passing from an integral domain \(R\) to its field of fractions \(\operatorname{Frac}(R)\), where the new structure is explicitly definable from the old, and passing from a field \(F\) to an \(n\)-dimensional vector space \(V\) over \(F\), where \(V\) is not definable from \(F\) alone but is internal to \(F\) after choosing extra data such as a basis [2602.05866]. In the classical formulation, one fixes a complete theory \(T\) with a distinguished unary predicate \(P\), and asks whether uniqueness of reconstruction from the \(P\)-part forces existence of reconstructions for all possible \(P\)-parts. Gaifman conjectured that if a countable theory \(T\) is categorical over a unary predicate \(P\), then \(T\) has the Gaifman property over \(P\). In modern terminology, “categorical over \(P\)” means relative categoricity: whenever two models of \(T\) have isomorphic \(P\)-parts, the isomorphism lifts to an isomorphism of the full models over that \(P\)-part [2507.12631]. The conjecture has important precedents but remains open in general. Gaifman proved it when \(T\) is rigid over \(P\), and Shelah proved an absolute version under absolute categoricity over \(P\). Pillay emphasizes that the conjecture still remains open “in full entirety,” despite later proofs of the Gaifman property under additional stability-over-\(P\) assumptions [2602.05866]. ## 2. Formal framework: \(P\)-parts, existence, and completeness For a model \(M \models T\), the induced structure on the distinguished predicate is written \(M^P\) in one source and \(P^M\) in the other; its theory is \(T^P\). The Gaifman property is the statement that for every \(N \models T^P\), there exists \(M \models T\) such that \(P^M = N\). Equivalently, every model of the induced theory on \(P\) is exactly the \(P\)-part of some model of \(T\) [2507.12631]. A more local notion is the existence property over \(P\). For \(A \subseteq \mathcal C\) in a monster model \(\mathcal C \models T\), one says that \(A\) has the existence property over \(P\) if there exists \(M \models T\) such that \[ A \subseteq M \quad\text{and}\quad P^M = P^A. \] This requires realization of the prescribed \(P\)-part without adding new \(P\)-elements. The Gaifman property is the global version: every \(N \models T^P\) has the existence property [2507.12631]. The fundamental closure notion is completeness. A set \(A \subseteq \mathcal C\) is complete if for every formula \(\psi(\bar x,\bar y)\) and \(\bar b \subseteq A\), \[ \models (\exists \bar x \in P)\,\psi(\bar x,\bar b) \quad\Rightarrow\quad (\exists \bar a \subseteq P \cap A)\,\models \psi(\bar a,\bar b). \] Thus any \(P\)-witness to a formula with parameters from \(A\) must already lie in \(P^A\). If \(M \prec \mathcal C\) and \(P^M \subseteq A \subseteq M\), then \(A\) is complete, so completeness is necessary for the existence property. A central theme of the subject is to determine when completeness is also sufficient [2507.12631]. The stronger classification-theoretic analysis imposes two standing assumptions on \(P\), called “very stable embeddedness”: \(P\) is stably embedded, and every definable subset of \(P^\mathcal C\) is already definable in \(T^P\). Under these assumptions, the induced structure on \(P\) fully captures all subsets of \(P\) definable in the ambient theory [2507.12631]. ## 3. Relative categoricity and the first existence theorems Relative categoricity is the uniqueness side of the problem. In Pillay’s formulation, \(T\) is relatively categorical if whenever \(M_1,M_2 \models T\) and \(f:M_1^P \to M_2^P\) is an isomorphism, then \(f\) lifts to an isomorphism \(M_1 \cong M_2\). There are cardinal-restricted variants, notably relative \((\omega,\omega)\)-categoricity, where the lifting property is required only for countable models whose \(P\)-parts are also countable [2602.05866]. A key characterization is that \(T\) is relatively \((\omega,\omega)\)-categorical iff every model \(M \models T\) is atomic over \(P(M)\): for every finite tuple \(a\) from \(M\), the type \(tp_M(a/P(M))\) is isolated. This atomicity yields uniform definability of types over the \(P\)-part, described by Pillay as a form of stable embeddability of \(P\) [2602.05866]. These observations already produce nontrivial existence theorems. Any countable model \(N \models T^P\) is equal to \(M^P\) for some countable \(M \models T\). More substantially, if \(T\) is relatively \((\omega,\omega)\)-categorical, then every \(N \models T^P\) of cardinality at most \(\aleph_1\) is of the form \(M^P\) for some \(M \models T\). The proof is a transfinite construction through a continuous chain of countable elementary submodels, with the induction step supplied by a completeness transfer lemma [2602.05866]. Pillay also isolates a strong sufficient condition for the full conjecture. If, in addition to relative \((\omega,\omega)\)-categoricity, the monster model is \(1\)-co-analyzable in \(P\), equivalently almost internal to \(P\), then \(T\) is relatively categorical and has the Gaifman property. Concretely, almost internality means that every model is algebraic over its \(P\)-part together with a finite tuple. Under this hypothesis, the finite parameters needed to recover the model are controlled by isolated types over the \(P\)-part, so both uniqueness and existence follow [2602.05866]. An important limitation is also explicit: stable embeddedness alone does not imply the Gaifman property. Pillay records a counterexample due to Hrushovski, reported by Kaplan, which shows that definability of types over \(P\) is not by itself sufficient in full generality [2602.05866]. ## 4. The classification-theoretic strengthening A major reformulation replaces the original conjecture by a broader dichotomy. Instead of asking only whether relative categoricity implies existence, one asks whether failure of existence already forces large-scale non-structure over \(P\). The strengthened conjecture states that if \(T\) fails the Gaifman property, then for every regular cardinal \(\lambda\) big enough, and every \(\mu \ge \lambda\), \(T\) has \(2^\lambda\) models of cardinality \(\mu\) that are pairwise non-isomorphic over \(P\) [2507.12631]. This reframes the problem in the language of Classification Theory. The proposed dividing line is a hierarchy of stability notions over \(P\). The generalized program is split into two directions: stability over \(P\) implies the Gaifman property, while instability over \(P\) implies non-structure. The first direction is proved for countable theories; the second is left open [2507.12631]. The basic stability notion is defined using \[ S_*(A)=\{\,\operatorname{tp}(\bar c/A): P\cap (A\cup \bar c)=P\cap A \text{ and } A\cup \bar c \text{ is complete}\,\}. \] These are the complete types over \(A\) whose realizations do not enlarge the \(P\)-part and preserve completeness. For a complete set \(A\), \[ A \text{ is stable over }P \quad\Longleftrightarrow\quad \forall A' \equiv A,\ \ |S_*(A')| \le |A'|^{|T|}. \] Thus stability over \(P\) is not simply stability of \(T\) in the ordinary sense; it is a stability condition for complete sets relative to the distinguished predicate [2507.12631]. The significance of this reformulation is conceptual as well as technical. The Gaifman problem is no longer treated as an isolated existence statement. It becomes part of a classification-theoretic program in which structure over \(P\) is measured by stability, and failure of existence is expected to coincide with the maximal proliferation of models over \(P\) in many cardinalities [2507.12631]. ## 5. Good systems, higher amalgamation, and the main theorem The central innovation of the classification-theoretic approach is a hierarchy of higher-dimensional stability and existence properties built from “good systems.” A good system is indexed by a hereditary family \(I \subseteq \mathcal P(n)\), usually \(\mathcal P(n)\) or \(\mathcal P^-(n)=\mathcal P(n)\setminus\{n\}\), and consists of a coherent family \[ \mathcal S=\langle A_s:s\in I\rangle \] satisfying structural clauses such as \[ A_s\cap A_t = A_{s\cap t}, \] together with the requirement that nodes not containing \(0\) are models of \(T^P\), nodes containing \(0\) are models of \(T\), and taking the \(P\)-part corresponds to deleting \(0\) from the index set. A further relation, \[ A \subseteq_t B, \] requires that formulas over parameters from \(A\) realized in \(B\) already have realizations in \(A\); it functions as a weak existential-closure condition adapted to the \(P\)-setting [2507.12631]. Using good systems, the paper defines \(n\)-stability over \(P\): \(T\) is \(n\)-stable over \(P\) if for every good \(\mathcal P^-(n)\)-system, the union of its lower faces is stable over \(P\). The corresponding \(n\)-existence property says that every good \(\mathcal P^-(n)\)-system can be completed to a good \(\mathcal P(n)\)-system. For \(n=1\), \(1\)-existence is exactly the Gaifman property. For larger \(n\), this yields a hierarchy of higher amalgamation properties over \(P\) [2507.12631]. The main theorem states that if \(T\) is countable, \(P\) is very stably embedded, and \(T\) is \(n\)-stable over \(P\) for all \(n<\omega\), then \(T\) has the Gaifman property. The stronger theorem proved is that under the same assumptions, the union of every good \(n^-\)-system has the existence property; in particular, \(T\) has \(n\)-existence for all \(n<\omega\). The case \(n=1\) recovers the Gaifman property, but the theorem is fundamentally a higher stable amalgamation theorem rather than only a \(1\)-dimensional existence statement [2507.12631]. The proof architecture explicitly transfers stable-theoretic tools to the relative setting over \(P\). Stable embeddedness gives definability of types over \(P\); stability over \(P\) yields definability and stationarity for \(S_*\)-types; a stationarization relation \(\bar a \ind_A B\) plays the role of nonforking; unions of good systems are shown to be complete; and locally isolated \(S_*\)-types support locally constructible model constructions. The final existence theorem is proved by cardinal induction using a decomposition of large good systems into continuous chains of smaller good systems, with a higher-dimensional coherence clause supplying the induction step [2507.12631]. A further strengthening is local constructibility. If \(\langle A_s:s\in \mathcal P^-(n)\rangle\) is a good system, then its union has the locally constructible existence property. In particular, every \(N \models T^P\) can be realized as the \(P\)-part of a model of \(T\) built by a local construction over \(N\), where each successive type is locally isolated [2507.12631]. ## 6. Open problems, limitations, and terminological clarifications The original conjecture remains open in its unrestricted form. Neither Pillay’s elementary observations nor the classification-theoretic theorem proves that relative categoricity by itself implies the Gaifman property. What is established is more conditional: relative \((\omega,\omega)\)-categoricity gives existence up to \(\aleph_1\), almost internality yields the full Gaifman property, and \(n\)-stability over \(P\) for all finite \(n\) yields not only the Gaifman property but a hierarchy of higher existence properties [2602.05866]. The major unresolved direction is the instability side of the classification program. The conjectural statement is that if \(T\) is \(n\)-unstable over \(P\) for some \(n\), then for every regular cardinal \(\lambda\) big enough, and every \(\mu \ge \lambda\), \(T\) has \(2^\lambda\) models of cardinality \(\mu\) which are non-isomorphic over \(P\). A weaker conjecture isolates the first unstable level: if \(T\) is \(m\)-stable over \(P\) for all \(m