Loading field map
Model theory studies the relationship between formal languages and the mathematical structures that interpret them. At its heart lies a tension: first-order logic is expressive enough to capture deep mathematical content, yet constrained enough to admit powerful metatheorems. The history of model theory is the story of how researchers learned to classify theories by the shapes of their models, and how that classification project forced the subfield to branch into specialized frameworks, each with its own notion of what makes a structure "tame."
Before model theory existed as a named discipline, logicians had already assembled its essential toolkit. In the 1910s and 1920s, Leopold Löwenheim and Thoralf Skolem proved that if a first-order theory has an infinite model, it has models of every infinite cardinality—a result that hinted at the wild variety of structures a single theory can admit. Kurt Gödel's completeness theorem (1929) showed that first-order logic is complete: every logical consequence of a theory is provable. His compactness theorem, a corollary, stated that a theory has a model if and only if every finite subset of it does. Compactness became the workhorse of early model theory, allowing logicians to construct nonstandard models by adding new constants and forcing consistency. Alfred Tarski then gave the subject its semantic foundation: a rigorous definition of truth in a structure (1933) and, with his students, developed the concepts of elementary equivalence (two structures satisfying the same sentences) and elementary embeddings (maps preserving all first-order formulas). By the 1950s, the tools were ready, but the questions were still forming.
Classical model theory turned the pre-history's scattered results into a coherent research program. Its central method was the ultraproduct construction, which builds a new structure from a family of structures by taking a product and quotienting by an ultrafilter. Ultraproducts gave a clean proof of compactness and allowed logicians to transfer properties between structures in ways that felt almost algebraic. The framework's signature achievement was Michael Morley's categoricity theorem (1965): if a countable first-order theory is categorical in some uncountable cardinality—meaning it has exactly one model of that size up to isomorphism—then it is categorical in every uncountable cardinality. Morley's proof introduced the notion of a "totally transcendental" theory and used ranks and indiscernibles, techniques that pointed beyond classical methods.
Yet classical model theory also revealed its own limits. Compactness and ultraproducts work beautifully for infinite structures, but they fail completely when attention shifts to finite ones. Moreover, Morley's theorem raised a question it could not answer: what makes a theory categorical in the first place? The proof gave a sufficient condition (total transcendence) but not a structural classification of all such theories. The field needed a finer-grained analysis of how formulas define sets within models, and that demand drove the next framework.
Saharon Shelah took up Morley's challenge in the 1970s and transformed model theory. Where classical model theory had treated models as whole objects, stability theory looked inside them at the definable sets—the subsets of a model that can be picked out by formulas. Shelah introduced the concept of forking independence, a model-theoretic analogue of linear independence in vector spaces or algebraic independence in fields. A formula "forks" over a small set of parameters if it forces a division into many incompatible cases; a theory is stable if no formula can encode a linear order of infinite length using forking. This gave rise to the stability hierarchy: theories can be classified as stable, superstable, or totally transcendental depending on how complex their definable sets are.
Stability theory was a direct response to the limits of classical model theory. Morley's theorem had shown that categoricity implies a kind of tameness, but Shelah's classification program aimed to characterize all possible degrees of tameness. The framework's crowning achievement was the proof that every uncountably categorical theory is "unidimensional"—its models are determined by a single dimension, like the dimension of a vector space. Stability theory also produced the notion of a "stable group," connecting model theory to algebraic groups and laying groundwork for later interactions with algebra.
Stability theory remains active, but it has evolved. In the 1990s and 2000s, researchers extended its ideas to broader classes: simple theories (where forking still behaves well but the hierarchy relaxes) and NIP theories (which forbid the "independence property" that allows encoding arbitrary bipartite graphs). NIP theories form a particularly important bridge, as they include both stable theories and o-minimal theories (discussed below), showing that these two frameworks are not rivals but special cases of a larger notion of tameness.
While stability theory pursued abstract classification, a parallel tradition focused on applying model theory to specific algebraic structures. The model theory of fields and valued fields grew out of Tarski's work on real closed fields (which he proved admit quantifier elimination: every formula is equivalent to one without quantifiers) and Abraham Robinson's development of nonstandard analysis. In the 1960s, James Ax and Simon Kochen proved a landmark result: for almost all primes p, the field of p-adic numbers Qp has the same first-order properties as the field of formal Laurent series Fp((t)). This used quantifier elimination for valued fields and showed that model theory could settle open problems in number theory. Later, the framework absorbed tools from stability theory: researchers proved that many valued fields are "C-minimal" or have NIP, linking concrete algebraic questions to the abstract classification hierarchy.
The distinctive commitment of this framework is its focus on algebraic structures that arise naturally in number theory and algebraic geometry—fields, rings, and their valued extensions. Its methods center on quantifier elimination, which reduces complex formulas to simple ones, and on understanding definable sets in these structures. Unlike stability theory, which asks "how complex can definable sets be?", the model theory of fields asks "what are the definable sets in this specific field?" The two frameworks coexist and enrich each other: stability theory provides general theorems that apply to many fields, while field-specific results test and refine the abstract classification.
O-minimality, introduced by Lou van den Dries in the 1980s, addresses a different kind of structure: those with a distinguished linear order. A theory is o-minimal if every definable subset of the ordered line is a finite union of intervals and points—the simplest possible description. This framework was a direct response to the limitations of stability theory for ordered structures. Stable theories cannot define a linear order on an infinite set, so they are useless for analyzing the real numbers, the real field, or any structure where order is essential. O-minimality filled that gap.
The framework's key method is cell decomposition: any definable set in an o-minimal structure can be partitioned into finitely many "cells," each homeomorphic to an open hypercube. This gives strong geometric control. O-minimality has produced striking applications, including the Pila-Wilkie theorem (2006), which bounds the number of rational points on definable sets in o-minimal expansions of the real field—a result with consequences for Diophantine geometry. Today, o-minimality and stability theory are largely complementary. O-minimal theories are typically unstable (they define an order), but they are NIP, placing them in the broader tameness hierarchy. Researchers often move between the two frameworks, using stability-theoretic tools for algebraic structures and o-minimal tools for real-geometric ones.
Finite model theory broke away from the mainstream in the 1980s, driven by a simple observation: the central theorems of classical model theory fail when structures are required to be finite. Compactness, completeness, and ultraproducts all rely on infinite processes. For finite structures, a different toolkit is needed. Boris Trakhtenbrot had already shown in 1950 that validity in finite models is not recursively enumerable—a stark contrast to Gödel's completeness theorem. Finite model theory took shape around computational motivations: database theory, complexity theory, and the study of algorithmic properties of finite structures.
Its methods center on Ehrenfeucht-Fraïssé games, which characterize when two finite structures satisfy the same first-order sentences up to a given quantifier depth. These games reveal that first-order logic is weak on finite structures—it cannot count beyond a fixed bound or express connectivity. This weakness led to the study of extensions (fixed-point logics, infinitary logics) and to descriptive complexity, which asks what logical resources are needed to capture complexity classes like P or NP. A landmark result is the 0-1 law: for many classes of finite structures, the probability that a random structure satisfies a given first-order sentence converges to either 0 or 1. This law relies on infinite models (the "generic" structure) to understand finite ones, creating a rare bridge between finite and infinite model theory.
Finite model theory remains largely separate from the other frameworks. Its questions are different (computational rather than structural), its methods are different (games rather than forking or cell decomposition), and its structures are finite. Yet it shares a common origin in first-order logic and a common interest in definability.
Today, the leading frameworks—stability theory, o-minimality, and the model theory of fields and valued fields—agree on a core principle: the tameness of a theory can be measured by the complexity of its definable sets. They disagree on what counts as the right measure. Stability theory privileges the absence of order; o-minimality privileges order but restricts definable sets to simple shapes; the model theory of fields and valued fields privileges quantifier elimination and concrete algebraic descriptions. The NIP property has emerged as a unifying concept: both stable and o-minimal theories are NIP, and much current research explores the structure of NIP theories in general, aiming for a classification that includes both. Finite model theory, by contrast, operates under different constraints and asks different questions, but its results on 0-1 laws and descriptive complexity continue to inform the broader field. The tension that opened model theory—between the expressive power of logic and the variety of mathematical structures—remains unresolved, but each framework has carved out a domain where that tension becomes productive.