*There exists an isomorphism θ from the vector space F⊗E′ onto the space Lf(E;F) of all finite rank continuous linear mappings from E into F, characterized by θ(b⊗a∗)=ba∗ for a∈E,b∈F.
By A, II, § 4, No. 2, there exists an injective linear mapping θ from F⊗E′ into L(E;F) and only one such, which transforms b⊗a′ into the linear mapping x↦ba′(x) for a′∈E′,b∈F. Evidently θ(b⊗a∗)=ba∗, and the image of θ is contained in Lf(E;F). However, let u∈Lf(E;F) and let (e1,...,en) be an orthonormal basis of the image of u in F. Let fi=u∗(ei) for 1≤i≤n. For every x∈E, we have
u(x)=i=1∑n⟨ei∣u(x)⟩⋅ei=i=1∑n⟨fi∣x⟩⋅ei,
hence u=∑i=1neifi∗=θ(∑i=1nei⊗fi∗). Therefore the image of θ is equal to Lf(E;F).
We shall henceforth assume that E=F, and we set Lf(E)=Lf(E;E). By lemma 1, there exists a unique linear form τ on Lf(E), such that τ(θ(a⊗a′))=a′(a) for a∈E,a′∈E′; in other words, we have
(22)
τ(ba∗)=⟨a∣b⟩for a,b in E.
When E is finite dimensional, we have Lf(E)=L(E) and τ(u) is the trace of the endomorphism u of E (A, II, § 4, No. 3).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.