— Let (λα)α∈A be a family of positive measures on T, directed for the relation ≤ and having a supremum λ in M(T). Then, for every numerical function f≥0,
λ∙(f)=α∈Asupλα∙(f).
When f belongs to K(T), this relation reduces to the definition of the supremum of a directed set in M(T) (Ch. II, §2, No. 2, Lemma). Suppose next that f≤g for some function g∈K+ (in other words, that f is bounded and is zero outside a compact set K); let α be an index such that λα(g)≥λ(g)−ε, where ε is a number >0; since the measure ν=λ−λα is positive, we have ν∗(f)≤ν(g)≤ε, or λα∗(f)≥λ∗(f)−ε (Ch. IV, §1, No. 3, Prop. 15). It follows (since ε is arbitrary) that the second member of (5) is ≥ the first; the reverse inequality being obvious, (5) is established for the special case under consideration. Next, suppose that f is zero outside K but is not necessarily bounded, and set fn=inf(f,n) for every integer n. Then
λ∙(f)=n∈Nsupλ∙(fn)=n∈Nsupα∈Asupλα∙(fn)=α∈Asupn∈Nsupλα∙(fn)=α∈Asupλα∙(f).
Finally, with no restriction made on f, denoting by K the set of compact subsets of T we have
λ∙(f)=K∈Ksupλ∙(fφK)=K∈Ksupα∈Asupλα∙(fφK) =α∈AsupK∈Ksupλα∙(fφK)=α∈Asupλα∙(f).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.