Suppose that I is directed, that the sets Sα satisfy conditions (i) and (ii), and that the inverse system (Eα,fαβ) has the following properties:
(iii) For each pair of indices α, β such that α⩽β, and each xα∈Eα, we have f−1αβ(xα)∈Sβ.
(iv) For each pair of indices α, β such that α⩽β, and each Mβ∈Sβ, we have fαβ(Mβ)∈Sα.
Let E=limEα and let fα:E→Eα be the canonical mapping for each α∈I. Then:
(a) For each α∈I we have
(19)fα(E)=β⩾α⋂fαβ(Eβ).
(b) If Eα is non-empty for each α∈I, then E is non-empty.
Let Σ be the set of families A=(Aα)α∈I which satisfy the following conditions :
(20)Aα=∅andAα∈Sαfor allα∈I;
(21)fαβ(Aβ)⊂Aαwheneverα⩽β.
If A=(Aα) and A′=(Aα′) are any two elements of Σ, let the relation A⩽A′ mean that Aα⊃Aα′ for all α. Clearly Σ is ordered by this relation.
(1) Let us first show that the ordered set Σ is inductive. Let L be a totally ordered set and let λ→Aλ=(Aαλ)α∈I be a strictly increasing mapping of L into Σ. For each α∈I, put Bα=⋂λ∈LAαλ. Then it is immediately seen that the family B=(Bα)α∈I satisfies (21); by reason of (i) and (ii), it also satisfies (20), hence belongs to Σ; and it is clear that B is an upper bound of the set of the Aλ.
(2) Let A=(Aα) be a maximal element of Σ. We shall show that Aα=fαβ(Aβ) whenever α⩽β. Let Aα′=⋂β⩾αfαβ(Aβ) for all α∈I, and let us show that A′=(Aα′) belongs to Σ. Note first that if α⩽β⩽γ, we have fαγ(Aγ)=fαβ(fβγ(Aγ))⊂fαβ(Aβ) by (21); moreover, fαβ(Aβ)∈Sα by (iv), and fαβ(Aβ)=∅ by (20). Hence conditions (i) and (ii) show that A′ satisfies (20). Also, if α⩽β, we have
fαβ(Aβ′)⊂γ⩾β⋂fαβ(fβγ(Aγ))=γ⩾β⋂fαγ(Aγ);
and for each δ⩾α there exists γ∈I such that γ⩾δ and γ⩾β, so that fαγ(Aγ)⊂fαδ(Aδ) and consequently
γ⩾β⋂fαγ(Aγ)=δ⩾α⋂fαδ(Aδ)=Aα′.
Hence A′ satisfies (21) and therefore belongs to Σ. Since Aα′⊂Aα for all α, the maximality of A in Σ implies A′=A, and our assertion is proved.
(3) We shall establish next that if A=(Aα) is a maximal element of Σ, then each of the Aαconsists of a single element. Let xα∈Aα. For each β⩾α, put Bβ=Aβ∩f−1αβ(xα); if β is not ⩾α, put Bβ=Aβ; then we shall see that B=(Bβ) belongs to Σ. If β is not ⩾α, the relation β⩽γ implies fβγ(Bγ)⊂fβγ(Aγ)⊂Aβ=Bβ. If, on the other hand, α⩽β⩽γ, then since
f−1αγ(xα)=f−1βγ(f−1αβ(xα)),
we have
fβγ(f−1αγ(xα))⊂f−1αβ(xα),
and since fβγ(Aγ)⊂Aβ, we again have fβγ(Bγ)⊂Bβ, so that the family B satisfies (21). Since Aα=fαβ(Aβ) whenever α⩽β by part (2) of the proof, it is clear that Bβ=∅ for all β∈I. Finally, by virtue of (i) and (iii), we have Bβ∈Sβ for all β∈I, and hence B∈Σ. Since Bβ⊂Aβ for all β∈I, the maximality of A implies that Bβ=Aβ for all β, and in particular Aα={xα}.
(4) We are now in a position to prove Theorem 1. Let us start with (a). We have
fα(E)⊂β⩾α⋂fαβ(Eβ).
Conversely, let
xα∈β⩾α⋂fαβ(Eβ),
and put
Bβ=f−1αβ(xα)
if β⩾α, and Bβ=Eβ otherwise. By the definition of xα, the Bβ are non-empty, and we have Bβ∈Sβ for all β∈I by virtue of (iii) and (i); moreover, it is evident that fβγ(Bγ)⊂Bβ whenever β⩽γ. Hence B=(Bβ)∈Σ. Let A=(Aα) be a maximal element of Σ such that A⩾B (the existence of A follows from (1) and §2, no. 4, Theorem 2, Corollary 1). Since, by (3), Aβ is of the form {yβ} for all β∈I, it follows that y=(yβ) belongs to E, and fα(y)=yα=xα by definition.
Finally, (a) implies (b). We may as well assume that I is not empty (otherwise there is nothing to prove). The hypothesis that the Eα are non-empty implies that fαβ(Eβ)=∅ for all β⩾α. Since the fαβ(Eβ), for fixed α and variable β⩾α, form a left directed set of subsets of Eα belonging to Sα, condition (ii)′ proves that
β⩾α⋂fαβ(Eβ)=∅.
Hence fα(E)=∅ by (a), and a fortioriE=∅. Q.E.D.
Bourbaki sets the proof directly after the statement, and the corpus does not separate them, so what is above is the statement and its proof together.