Model Theory Becomes Mathematics

Models first entered this series as interpretations used to understand truth, consistency, and the limits of axiomatization. That role is important but incomplete. Model theory also became a way to prove mathematical theorems about algebraic structures and to classify the behavior of whole theories.

The transition was visible in work by Anatoly Mal'tsev, Alfred Tarski, and Abraham Robinson. General logical results could be applied to groups, fields, and other structures already studied by mathematicians. Later tools associated with Henkin, Łoś, Fraïssé, Morley, and Shelah deepened the subject.

The central question shifted from whether a formal system had an interpretation to what its language could define and what those definitions revealed about its models.

Elementary equivalence and elementary embeddings

Two structures are elementarily equivalent if they satisfy the same first-order sentences in a specified language:

This need not make them isomorphic. First-order sentences may be unable to distinguish structures of different cardinalities.

An elementary embedding preserves more detailed information. An embedding $j:\mathcal M\to\mathcal N$ is elementary if, for every formula $\varphi$ and every tuple $\vec a$ from $M$,

The language matters. A relation or function omitted from the signature is not automatically controlled by this equivalence. Adding an exponential function to an ordered-field language, for example, changes the theory under investigation.

This sensitivity to language is a source of mathematical information. It asks which properties are visible through a selected collection of operations and relations.

Definable sets

A formula with free variables defines a set of tuples in a structure. In the real ordered field,

defines the unit circle. The formula

defines the nonnegative reals.

The second example already suggests an operation on definable sets: quantification projects a relation onto some of its coordinates. A relation between $x$ and $y$ yields the set of $x$'s for which an appropriate $y$ exists.

Logical connectives correspond to other set operations. Conjunction gives intersection, disjunction gives union, and negation gives complement. The study of definability therefore connects formal syntax with geometric and algebraic structure.

Parameters are allowed when specified. For instance, $x<a$ defines an initial segment once a particular element $a$ is supplied. Definability with parameters and without parameters are different conditions.

A complete example: dense linear orders

Consider the first-order theory of dense linear orders without endpoints, in a language containing only $<$.

Density says that between any two distinct ordered elements there is another:

The absence of endpoints says that every element has something below it and something above it. Both $(\mathbb Q,<)$ and $(\mathbb R,<)$ satisfy these axioms.

For example,

is equivalent, in this theory, to $a<b$. The existential quantifier has been eliminated.

More generally, the theory admits quantifier elimination. Every formula is equivalent, modulo the theory, to a quantifier-free formula. To see the mechanism, organize conditions on a prospective witness into lower bounds, upper bounds, equalities, and exclusions. Density supplies a witness between compatible bounds, and the endpoint axioms handle unbounded cases.

The construction must examine all Boolean combinations, but its mathematical idea is simple: the finite order relationships among the parameters determine whether a witness can exist.

As a consequence, subsets of a model definable in one variable with finitely many parameters are finite unions of intervals and points. The language cannot define an arbitrary scattered subset of the order.

Categoricity within a cardinality

Any two countable dense linear orders without endpoints are isomorphic. A back-and-forth construction proves this.

Enumerate the elements of both orders. Extend a finite order-preserving partial map by first including the next unused element from the first order, then including the next unused element from the second. Density and the absence of endpoints provide an element in the required interval at each stage.

The union of the finite maps is an order isomorphism.

This does not contradict Löwenheim–Skolem. The theory also has uncountable models, including the real order, which cannot be isomorphic to a countable model. Its categoricity is confined to the countable cardinality.

Fraïssé's work developed a broader method for constructing countable structures from compatible finite configurations. The finite patterns and their extension properties can determine a highly homogeneous infinite structure.

These results show why “first-order logic cannot specify an infinite structure” is too crude. One must state which cardinalities and which form of specification are being considered.

Ultraproducts and Łoś's theorem

An ultraproduct combines a family of structures into one structure while controlling the truth of first-order formulas.

An ultrafilter $U$ on an index set chooses which subsets of indices count as large. It is closed under finite intersection and upward inclusion, and for every subset it selects that subset or its complement. The empty set is not large.

In a product of structures, two sequences are identified when they agree on a $U$-large set of indices. Functions and relations are interpreted coordinatewise, modulo this identification.

Łoś's theorem states that a first-order formula holds in the ultraproduct exactly when the set of indices on which it holds is $U$-large.

This permits local information from the component structures to be assembled into a new model. When a nonprincipal ultrafilter on the natural numbers is used, every cofinite set is large and no finite set is large. The existence of such an ultrafilter uses background set-theoretic principles; it is not supplied by a finite explicit selection algorithm.

The construction became valuable in algebra and in the study of theories. It also provides one route to nonstandard number systems.

Where infinitesimals actually live

Take an elementary extension ${}^\ast\mathbb R$ of the real ordered field that contains an element $H$ greater than every standard real number. Compactness applied to the elementary diagram of $\mathbb R$, together with all conditions $H>r$, can establish such an extension.

Because this structure is a field, $H$ has an inverse. Put

Then $\varepsilon>0$, and $\varepsilon<r$ for every positive standard real $r$. It is an infinitesimal relative to the embedded copy of $\mathbb R$.

This corrects a common shortcut. A nonstandard model of natural-number arithmetic does not itself contain inverses of its positive infinite elements. The ordered-field construction is an additional mathematical step.

For a simple polynomial calculation,

The result differs from $2a$ by an infinitesimal. In a suitable nonstandard-analysis framework, taking the standard part recovers the derivative.

Abraham Robinson's work in the 1960s developed such methods systematically. To treat arbitrary functions and further analytical objects, the language and transfer framework must be expanded appropriately. Elementary equivalence in the language of ordered fields alone does not automatically transfer every assertion involving every real function.

The distinction between standard and nonstandard elements is also external to the simple first-order field structure. These qualifications are part of the method, not defects hidden behind the notation.

Morley and the classification of theories

Michael Morley's 1965 categoricity theorem states that a complete theory in a countable language, if categorical in one uncountable cardinality, is categorical in every uncountable cardinality.

The hypotheses are substantial. The theorem is not a statement that countable categoricity implies categoricity everywhere, as the dense-order example already shows.

One central tool is the study of types: collections of formulas that describe a possible element or tuple, often relative to named parameters. A type records more information than one finite formula. Its realization asks whether a model contains an element satisfying the whole collection.

Counting and organizing types reveals differences among theories. Saharon Shelah's classification theory developed this approach through stability and related dividing lines. The question becomes whether models of a theory admit a manageable structural account or exhibit patterns that generate many different possibilities.

This is a positive research program. Formal language is being used to detect mathematical organization, not merely to demonstrate the inadequacy of axioms.

Applications and the independence of the subject

Tarski's quantifier elimination for real closed fields connects first-order reasoning with polynomial inequalities. Mal'tsev applied logical methods to algebraic problems. Ax–Kochen results related valued fields through model-theoretic methods, and later work by Hrushovski demonstrated further interactions with algebraic geometry and number theory.

These applications require specialized mathematics in addition to general metatheorems. Compactness is a powerful tool, but not a substitute for understanding the structures to which it is applied.

The history also reflects distinct research communities. Mal'tsev's work developed in the Soviet mathematical environment; Tarski's school in the United States became a major center; subsequent classification theory grew through extensive international collaboration. A history that jumps directly from Gödel to programming languages loses these mathematical developments.

Models as instruments of discovery

Model theory investigates what languages define, how structures can be extended or combined, and which theories admit classification. Its methods can produce unfamiliar objects while preserving exact first-order information.

Set theory uses model constructions for a related but different purpose: investigating whether particular axioms settle questions about the universe of sets. The next note explains constructibility and forcing, and why their independence results require more than the general existence of nonstandard models.

Sources and further reading

• Alfred Tarski, A Decision Method for Elementary Algebra and Geometry (1948; later editions); Anatoly Mal'tsev's work on applications of compactness in algebra.
• Jerzy Łoś, “Quelques remarques, théorèmes et problèmes sur les classes définissables d'algèbres” (1955).
• Roland Fraïssé, “Sur l'extension aux relations de quelques propriétés des ordres” (1954).
• Michael Morley, “Categoricity in Power” (1965).
• Abraham Robinson, Non-standard Analysis (1966); Saharon Shelah, Classification Theory and the Number of Nonisomorphic Models (1978; later expanded edition).
• First-order Model Theory, for the relationship between definability, classification, and applications.