Note that 2t+t′ belongs to D. One knows (No. 6, Example 2) that the family (X(t))t∈D is a basis of the topological dual Ω′ of Ω; therefore there exists a symmetric bilinear form M on Ω′×Ω′ characterized by M(X(t),X(t′))=inf(t,t′). By construction, the variance of the Gaussian measure m on Ω is the quadratic form ξ↦M(ξ,ξ) on Ω′. Set, in particular,
ξ=X(2t+t′)−2X(t)+X(t′);
an easy calculation yields
M(ξ,ξ)=4∣t−t′∣.
(2) Espace mesuré: a locally compact space equipped with a measure (Ch. III, 1st edn., §2, No. 2, p. 52).
By Prop. 6 of No. 5 (formula (16)),
∫Ω∣ξ∣3dm=π−1/223/2Γ(2)M(ξ,ξ)3/2;
the lemma follows immediately from formulas (40) and (41).
B) Construction of a mapping u of Ω into C: For every integer n⩾0, denote by En the subspace of C formed by the functions that are affine on each of the intervals [2nk−1,2nk] for 1⩽k⩽2n. An affine function on a compact interval I of R attains its bounds at the endpoints of I; consequently,
∥f∥=1⩽k⩽2nsupf(2nk)
for f∈En.
For every function g∈Ω and every integer n⩾0, there exists one and only one function un(g) that belongs to En and coincides with g at every point of Dn; we shall write Tng=un+1(g)−un(g). Since Dn is finite, the mapping Tn of Ω into C is continuous, hence m-measurable.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.