— The sesquilinear form Φ on E1⊗E2 is hermitian and positive, hence assigns the structure of a prehilbertian space to E1⊗E2. This space is Hausdorff if E1 and E2 are Hausdorff.
The formula Φ(z,t)=Φ(t,z) follows from (3) when z=x1⊗x2 and t=y1⊗y2. The general case is obtained by linearity, hence Φ is hermitian.
Suppose E1 and E2 are Hausdorff; we shall prove that the hermitian form Φ is positive and separating. Let z=∑i=1nxi⊗yi be a non-zero element of F=E1⊗E2. Let (e1,...,en) be an orthonormal basis of the subspace of E1 generated by x1,...,xn (V, p. 23, cor. 1). There exist elements f1,...,fm in E2, not all null, such that z=∑i=1nei⊗fi, hence
Φ(z,z)=i,j=1∑mΦ(ei⊗fi,ej⊗fj) =i,j∑⟨ei∣ej⟩⟨fi∣fj⟩=i=1∑m∥fi∥2>0.
For the general case, we shall now prove that Φ is positive. Let E~i be the Hausdorff prehilbertian space associated with Ei and let πi be the canonical mapping from Ei onto E~i (i=1,2). Put π=π1⊗π2. Let Φ~ be the hermitian form on E~1⊗E~2 constructed in the same way as Φ. We have
Φ(z,t)=Φ(π(z),π(t))(z∈F, t∈F), and since
Φ~ is positive, so is
Φ.
The prehilbertian space defined in prop. 1 is called the tensor product of the prehilbertian spaces E1 and E2 and is written as E1⊗E2. Henceforth we shall write ⟨z∣t⟩ for Φ(z,t), and therefore by definition
⟨x1⊗x2∣y1⊗y2⟩=⟨x1∣y1⟩⟨x2∣y2⟩; we shall also write
∥z∥2 or
∥z∥ for
⟨z∣z⟩1/2. From (4), we get
∥x1⊗x2∥2=∥x1∥⋅∥x2∥, then the bilinear mapping
(x1,x2)↦x1⊗x2 from
E1×E2 into
E1⊗E2 is continuous.
For i=1,2 let Fi be a vector subspace of Ei, endowed with the induced prehilbertian structure. Then F1⊗F2 can be identified with a vector subspace of E1⊗E2 (A, II, § 7, No. 7). Formula (4) shows that F1⊗F2 with the prehilbertian space structure induced by that of E1⊗2E2, is exactly F1⊗2F2. We shall henceforth identify F1⊗2F2 with a prehilbertian subspace of E1⊗2E2.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.