Tag 03OT, permanent. Cite it as [Bourbaki, Tag 03OT].
Theory of Sets, III, 251
Let (Eα,fαβ) and (Fα,gαβ) be two inverse systems of sets relative to the same index set I. For each α∈I, let uα be a mapping of Eα into Fα, such that the uα form an inverse system of mappings. Let Gα⊂Eα×Fα be the graph of uα. Show that (Gα) is an inverse system of subsets of Eα×Fα and that its inverse limit may be canonically identified with the graph of u=limuα.
Solution
This solution was written by nemotron-3-ultra-free and judged by a machine. It is not Bourbaki's and it has not been checked by a person.
Let (Eα,fαβ) and (Fα,gαβ) be two inverse systems of sets relative to the same index set I. For each α∈I, let uα:Eα→Fα be a mapping such that the family (uα) is an inverse system of mappings, i.e. gαβ∘uβ=uα∘fαβ whenever α⩽β. Let Gα⊂Eα×Fα be the graph of uα, so Gα={(x,uα(x))∣x∈Eα}.
For α⩽β, the mapping fαβ×gαβ:Eβ×Fβ→Eα×Fα sends (x,y) to (fαβ(x),gαβ(y)). If (x,y)∈Gβ, then y=uβ(x), and
where the second equality uses the hypothesis gαβ∘uβ=uα∘fαβ. Hence the restriction hαβ=(fαβ×gαβ)∣Gβ maps Gβ into Gα. We verify that (Gα,hαβ) satisfies the inverse system axioms. - For each α, fαα=idEα and gαα=idFα, so fαα×gαα=idEα×Fα. Restricting to Gα gives hαα=idGα. - If α⩽β⩽γ, then fαγ=fαβ∘fβγ and gαγ=gαβ∘gβγ, hence fαγ×gαγ=(fαβ×gαβ)∘(fβγ×gβγ). Since hβγ maps Gγ into Gβ, the restriction of the composition to Gγ equals hαβ∘hβγ, while it is also hαγ. Thus hαγ=hαβ∘hβγ. Therefore (Gα,hαβ) is an inverse system of subsets of Eα×Fα.
Let E=limEα and F=limFα, with canonical projections fα:E→Eα and gα:F→Fα. By Proposition 1 of § 1 (tag 03OC) applied to the inverse system (Fα,gαβ) and the family of mappings vα=uα∘fα:E→Fα, which satisfies gαβ∘vβ=vα for α⩽β, there exists a unique mapping u:E→F such that vα=gα∘u for all α, i.e.
uα∘fα=gα∘u(α∈I).
(Although Proposition 1 is stated for an ordered index set, its proof only uses the preorder relations — transitivity and reflexivity — and does not require antisymmetry; hence it remains valid for preordered sets.) This u is the inverse limit limuα.
The product inverse system (Eα×Fα,fαβ×gαβ) has inverse limit canonically isomorphic to E×F by Corollary 2 of Proposition 4 (tag 03OH). Under this isomorphism, an element (zα)∈lim(Eα×Fα) corresponds to (x,y)∈E×F with xα=pr1(zα), yα=pr2(zα).
The inverse limit limGα is the subset of lim(Eα×Fα) consisting of families (zα) with zα∈Gα for all α. Under the canonical isomorphism, this corresponds to the set of (x,y)∈E×F such that for every α,
(xα,yα)∈Gα⟺yα=uα(xα).
If (x,y) satisfies this condition, then yα=uα(xα)=uα(fα(x))=gα(u(x)) for all α, so y=u(x) because the canonical map F→∏αFα is injective. Conversely, if y=u(x), then gα(y)=gα(u(x))=uα(fα(x)), so yα=uα(xα) for all α. Hence the image of limGα is exactly the graph of u. This identification is canonical because it is induced by the canonical projections.