Saturated Models

Origin. Jónsson (1956, 1960) built homogeneous universal structures in the algebraic setting, generalizing Hausdorff's η_α orders. Morley and Vaught, "Homogeneous universal models" (Mathematica Scandinavica, 1962), transposed the construction to elementary classes and proved the uniqueness theorem that makes saturation usable; Keisler's contemporaneous work on ultrapowers supplied the standard existence method. Chang and Keisler, Model Theory (1973), consolidated the subject, including the special models that stand in where saturated ones fail to exist.

Models. A model that realizes every type it has room for. M is κ-saturated when every complete type over a parameter set of size less than κ is realized in M; M is saturated when it is |M|-saturated. Such a model is universal — every model of the same theory of size at most |M| embeds elementarily — and homogeneous, so partial elementary maps between small subsets extend to automorphisms. It serves as a universal domain in which every small configuration already occurs.

Formalism.

Type: p(x) = consistent set of formulas with parameter x. Complete: decides all formulas.

Realizes: M realizes p if some a ∈ M satisfies all φ ∈ p. p(a) holds.

κ-saturated: M is κ-saturated iff: for all A ⊆ M with |A| < κ, M realizes all types over A.

ω-saturated: Realizes types over finite sets. Countable saturation.

Existence: Every M has a κ⁺-saturated elementary extension — ZFC, by elementary chains or iterated ultrapowers. A saturated model of a prescribed cardinality κ exists when κ^{<κ} = κ. Under GCH this holds at every successor, so saturated models exist in every κ⁺ ≥ |T|⁺. If T is κ-stable, T has a saturated model of size κ outright. In ZFC alone, existence at a given κ may fail.

Uniqueness (Morley–Vaught): Any two saturated models of a complete T of the same cardinality are isomorphic. Proof: back-and-forth, using saturation on each side to realize the next type.

Countable case (Vaught): A complete countable T has a countable ω-saturated model iff Sₙ(T) is countable for every n.

Homogeneity: κ-saturated → κ-homogeneous. Partial isomorphisms extend.

Special models: Union of an elementary chain of increasingly saturated models. Exist in ZFC at suitable ℶ-limits of cofinality ω, where saturated models may not.

Recursively saturated (Barwise–Schlipf 1976): Realizes recursive types only. Every countable model has a countable recursively saturated elementary extension, in ZFC.

Monster model: Very saturated and strongly homogeneous. "Universal domain." Contains all small models.

Applications: Definability: if definable over all sets, globally definable. Automorphisms: many in saturated. Saturation + categoricity = uniqueness.

Ultrapower: ∏_U M / ultrafilter. ω₁-saturated (with countable index). Łoś theorem.

Symbols.

Metatheory. Morley and Vaught (1962) proved the two facts the notion rests on: κ-saturation implies κ-homogeneity and κ-universality, and any two saturated models of a complete theory of the same cardinality are isomorphic, by a back-and-forth argument that uses saturation on each side. Existence is the delicate half. Every model has a κ⁺-saturated elementary extension in ZFC, but a saturated model of a prescribed cardinality κ requires κ^{<κ} = κ, which holds at successors under GCH and can fail without it; if T is κ-stable a saturated model of size κ exists outright, which is why stability theory can work comfortably inside a monster model. Vaught settled the countable case: a complete countable theory has a countable ω-saturated model iff Sₙ(T) is countable for every n. An ultrapower over a countably incomplete ultrafilter on ω is ℵ₁-saturated, which is where saturation meets the transfer principle of nonstandard analysis, and the Keisler–Shelah theorem — Keisler (1961) under GCH, Shelah (1971) in ZFC — gives M ≡ N iff M and N have isomorphic ultrapowers.

Applies to. Stability theory. Classification. Non-standard analysis. Model constructions.

Limitations. Saturated models of a given cardinality need not exist: outside the stable case existence at κ generally needs κ^{<κ} = κ, so failures of GCH leave theories with no saturated model at the convenient sizes, and special models or merely κ-saturated models are substitutes rather than the thing itself. The monster model is a convention, not an object — it must be fixed relative to a cardinal exceeding everything under discussion, and "small" is small by stipulation. The constructions are non-effective and use choice at every step, through ultrafilters and through unions of chains, so no saturated model is exhibited and none is computable. Saturation is a first-order notion and constrains only the types of the language at hand. It is also not inherited: an elementary substructure of a saturated model need not be saturated, so the property must be re-established rather than assumed downward.

© 2026 Lingenic LLC