Extension lemmas state that partial elementary maps from finitely generated substructures extend to larger structures or automorphisms of homogeneous models. The back-and-forth construction iteratively extends partial maps by alternating extension in forward and backward directions, producing isomorphisms between structures or embeddings into homogeneous models.
You have studied elementary substructures and elementary maps — structure-preserving injections that respect all first-order formulas. You may also have encountered Ehrenfeucht-Fraïssé games, where two players test whether structures are elementarily equivalent by building a partial isomorphism in stages. The back-and-forth method is the algebraic counterpart of that game: a constructive procedure for building isomorphisms between structures by extending partial elementary maps in alternating rounds.
The basic setup: you have two structures 𝔄 and 𝔅 and a partial elementary map p₀: A₀ → B₀ between finite subsets. The forth step extends p to include a new element a ∈ 𝔄 by finding a matching element b ∈ 𝔅 that realizes the same complete type over p(A₀). The extension lemma guarantees such a b exists whenever 𝔅 is sufficiently saturated or homogeneous: any type realized in 𝔄 over the image of A₀ can be matched in 𝔅. The back step then handles a new element from 𝔅 symmetrically, finding a matching element in 𝔄. Alternating forth and back across all elements of both structures eventually produces a bijection that is elementary in both directions — a full isomorphism.
The back step is what elevates the method beyond mere embedding. Without it, you might map 𝔄 injectively into 𝔅 while leaving elements of 𝔅 without preimages — an embedding, not an isomorphism. The back step ensures every element of 𝔅 eventually acquires a preimage in 𝔄. The name comes directly from this alternation: you go *forth* to cover 𝔄 and *back* to cover 𝔅, guaranteeing surjectivity alongside injectivity.
The method's most celebrated application proves that the dense linear order without endpoints (ℚ, <) is the unique countable model of its theory up to isomorphism — a categorical theory. Given any two countable dense linear orders without endpoints, enumerate their elements and build an isomorphism by back-and-forth: at each stage, match one element from each structure, using density to guarantee that a matching element always exists. This proof is a template: for any countable homogeneous structure — one where every partial elementary map between finite subsets extends to an automorphism — back-and-forth produces an isomorphism from elementary equivalence, and the extension lemma is precisely the engine that makes each step possible.
No topics depend on this one yet.