The transfer theorem stating that a first-order formula holds in an ultraproduct (or ultrapower) precisely when the set of indices at which the formula holds in the factors belongs to the chosen ultrafilter.

Principle

First-order truth is preserved and reflected coordinatewise through ultraproducts: evaluation reduces to checking measure-one (ultrafilter-large) sets of coordinates, yielding an elementary equivalence/extension relation between factors and the ultraproduct.

Demonstration

If each structure M_i satisfies a sentence φ for all i in a set U belonging to an ultrafilter, then the ultraproduct ∏M_i/U satisfies φ; in particular, a constant sequence embedding into its ultrapower satisfies the same first-order sentences, producing elementary extensions.

Misapplication

Attempting to apply Łoś's theorem to higher-order statements, to non-ultrafilter filters (e.g., cofinite filter), or neglecting that languages must be uniform across factors; such uses break the coordinatewise transfer principle.

Consequence

Łoś's theorem underpins ultraproduct techniques: it yields new models preserving first-order theories, proves compactness-like results, constructs nonstandard models, and shows that ultrapowers are elementary extensions of the original when using a nonprincipal ultrafilter.

Reversal

If one drops maximality of the filter, transfer fails: a formula may hold in almost all coordinates without the ultrapower satisfying it. The reversal highlights dependence on the ultrafilter's decisiveness.

Boundary

Applies to first-order languages with the same signature in each factor and to ultrafilters; it does not extend to logics that quantify over sets of elements (second-order) or to filters lacking ultrafilter maximality.

Semantic Tension

Tension with the compactness theorem and other model-theoretic constructions: Łoś gives a pointwise transfer via ultrafilters, whereas compactness arguments use syntactic finitary consistency — both produce extensions but by different mechanisms.

Synthesis

Łoś's theorem is the coordinatewise transfer law for ultraproducts: first-order formulas hold in the ultraproduct exactly on ultrafilter-large sets of indices, enabling ultrapower-based constructions of elementary extensions and nonstandard models.