Model-Theoretic Connections

Preservation Theorems: Horn, Universal and Positive Sentences

Each construction preserves a syntactic class exactly. Substructures preserve universal sentences, products preserve Horn sentences, surjections preserve positive ones — and each correspondence is an if-and-only-if.

Engineering · Mathematics5 min readKV-MATH-0253

Learning objectives

State the Łoś–Tarski theorem for universal sentences.
State Lyndon's theorem for positive sentences.
State the Horn characterisation for reduced products.
State the Chang–Łoś–Suszko theorem for unions of chains.
See Birkhoff's HSP theorem as the intersection of these results.
Recognise which preservation theorems fail in the finite setting.

01The programme

Every preservation theorem has the same shape: a construction on structures, a syntactic class of sentences, and a proof that the class is exactly what the construction preserves.

The preservation theorems
Construction	Preserved class	Theorem
Substructures	universal (∀) sentences	Łoś–Tarski
Extensions	existential (∃) sentences	Łoś–Tarski, dual form
Surjective homomorphisms	positive sentences	Lyndon
Direct and reduced products	Horn sentences	Keisler–Galvin
Unions of chains	∀∃ sentences	Chang–Łoś–Suszko
H, S and P together	identities	Birkhoff

Each row is an if-and-only-if. Preservation is not merely implied by the syntactic form — it characterises it, up to logical equivalence. That is what makes these theorems informative rather than routine.

02Łoś–Tarski

Key resultThe Łoś–Tarski theorem

A first-order sentence is preserved under substructures if and only if it is logically equivalent to a universal sentence — one of the form ∀x⃗ ψ with ψ quantifier-free. Dually, a sentence is preserved under extensions iff it is equivalent to an existential sentence.

Easy direction
A universal sentence quantifies only over elements, and a substructure has fewer elements. If the quantifier-free matrix holds for all tuples in A it holds for all tuples in B ⊆ A.
Hard direction
Given preservation, construct the universal consequences of the sentence and show they imply it. The proof uses compactness and a diagram argument.
Algebraic reading
Being a subgroup-closed property is a universal condition. Being torsion-free is universal; being a torsion group is not, and correspondingly torsion is not preserved by substructures in any useful sense.
Where identities sit
Identities are universal, so they are preserved by substructures — the S in HSP. This is one third of Birkhoff's theorem falling out of Łoś–Tarski.

03Lyndon's theorem

Positive sentences — those built without negation or implication — are exactly the sentences preserved by surjective homomorphisms.

Positive: built from atomic formulas using &, ∨, ∀, ∃ only.
No ¬, no →, no ↔.

Identities are positive: p ≈ q is atomic, universally quantified. Quasi-identities are NOT positive, because the implication is essential.

CautionQuasi-identities are not preserved by homomorphisms

This is exactly why quasivarieties are closed under S and P but not H. Cancellativity is a quasi-identity: xy ≈ xz → y ≈ z. A homomorphic image of a cancellative semigroup need not be cancellative. Lyndon's theorem explains the failure structurally rather than as an accident of the example.

Lyndon's theorem supplies the H of HSP: identities are positive, hence preserved by surjective homomorphisms. With Łoś–Tarski giving S and the Horn theorem giving P, Birkhoff's theorem is the conjunction.

04Birkhoff as the intersection

Identities are simultaneously universal, positive and Horn. Each membership supplies one closure property.

Identities in three classes at once
Class	Membership	Closure supplied
Universal	∀x⃗ (p ≈ q) has all quantifiers universal	S
Positive	no negation or implication appears	H
Horn	a single positive literal per clause	P
All three	identities are exactly this intersection	HSP

Key resultWhy identities and nothing weaker

Birkhoff's theorem holds for identities precisely because they lie in the intersection of the three preserved classes. Weaken to quasi-identities and positivity is lost, so H fails and one gets quasivarieties. Weaken to universal sentences and positivity is lost too. The equational case is the unique point where all three preservation theorems apply at once.

Read this way, Birkhoff's theorem is not an isolated algebraic result but the strongest member of a family of preservation theorems, and its unusual strength has a structural explanation.

05Chang–Łoś–Suszko

Unions of chains preserve ∀∃ sentences — those with all universal quantifiers preceding all existential ones.

∀∃ form: ∀x⃗ ∃y⃗ ψ(x⃗, y⃗) with ψ quantifier-free

Also called inductive sentences, because their model classes are closed under unions of chains — the condition Zorn's lemma needs.

The connection to Zorn is direct: a class closed under unions of chains admits maximal elements by Zorn, so inductive theories are exactly those where maximal-model arguments work. Algebraically closed fields and existentially closed structures are the standard beneficiaries.

Inductive theories

∀∃ axiomatisable

Closed under unions of chains. Zorn applies, so maximal and existentially closed models exist.

Model companions

Built from inductive theories

The model companion of a theory, when it exists, is closely tied to its existentially closed models — the machinery Burris and Werner used for discriminator varieties.

06Failure in the finite setting

Several preservation theorems fail when attention is restricted to finite structures, which matters because universal algebra frequently works with finite algebras.

Preservation over finite structures
Theorem	Holds for finite structures?
Łoś–Tarski (substructures)	No — Tait's counterexample
Lyndon (homomorphisms)	No — Ajtai–Gurevich
Compactness	No — fails outright
Homomorphism preservation	Yes — Rossman, but this postdates the source
Birkhoff for finite algebras	Requires pseudovarieties and profinite identities

CautionFinite model theory is a different subject

The classical preservation theorems rely on compactness, which fails over finite structures. Results proved here should not be assumed to transfer to the finite setting, and several are known to fail. Where the finite case is wanted, consult finite model theory directly — Reiterman's theorem plays the role of Birkhoff's, using profinite rather than ordinary identities.

Frequently asked

Are preservation theorems constructive?

No — the hard directions use compactness, so they assert the existence of a logically equivalent sentence of the right shape without producing it. Finding the universal equivalent of a given preserved sentence is a separate and generally hard problem.

Is every universal sentence preserved by substructures?

Yes, and that is the easy direction. The content of Łoś–Tarski is the converse: anything preserved by substructures must already be equivalent to a universal sentence, so no sentence is 'accidentally' preserved.

How does this relate to quasivarieties?

Quasi-identities are universal and Horn but not positive, so quasivarieties are closed under S and P but not H. The characterisation of quasivarieties — closed under S, P and ultraproducts, containing a trivial algebra — is the corresponding preservation result, and it requires ultraproducts precisely because the class is not equational.

Sources and further reading

S. Burris and H. P. Sankappanavar, A Course in Universal Algebra, Millennium Edition (a corrected re-typesetting of Springer GTM 78, 1981).
G. Grätzer, Universal Algebra, 2nd edition, Springer.
R. McKenzie, G. McNulty and W. Taylor, Algebras, Lattices, Varieties, Volume I.

Original KEVOS® explanatory article. Written from the topic map of the cited works; no text is reproduced from them.