The mapping N↦uN is an isometric isomorphism of L2(X×Y) onto the space L2(L2(X);L2(Y)) of mappings
of Hilbert–Schmidt deThe linear mapping of LL2(X)2into(Y)⊗LL2(Y)2(X) into. L2(L2(X);L2(Y)) which associates to g⊗f the Hilbert–Schmidt mapping h↦⟨f∣h⟩g extends to an isometric isomorphism θ1 of L2(Y)⊗2L2(X) onto L2(L2(X);L2(Y)) (EVT, V, p. 52, th. 1).
Let us denote moreover by θ2 the isometric isomorphism of L2(X)⊗2L2(Y) onto L2(Y)⊗2L2(X) which to f⊗g associates g⊗f for every f∈L2(X) and every g∈L2(Y).
The linear mapping θ3=θ1∘θ2 is identified with an isometric isomorphism of L2(X×Y) onto L2(L2(X);L2(Y)).
Let f∈L2(X) and g∈L2(Y), and let N be the element of L2(X×Y) identified with f⊗g. From INT, V, p. 95, § 8, no3, cor. 2 and formula (12) one has
⟨g1∣uN(f1)⟩=⟨f∣f1⟩⟨g1∣g⟩
for all f1∈L2(X) and g1∈L2(Y). The mapping u=θ3(N) is the linear mapping θ1(g⊗f); it therefore satisfies u(h)=⟨f∣h⟩g for all h∈L2(X). Consequently, for all f1∈L2(X) and g1∈L2(Y), one has
⟨g1∣u(h1)⟩=⟨f∣h1⟩⟨g1∣g⟩=⟨g1∣uN(h1)⟩
whence θ3(N)=uN. Since θ3 and N↦uN are continuous, it follows that θ3(N)=uN for all N∈L2(X×Y), which completes the proof.
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.