Bourbaki

Proposition 2

Tag 080C, permanent. Cite it as [Bourbaki, Tag 080C].

Topological Vector Spaces, V, p. 26, § 3. Tensor product of hilbertian spaces

For i=1,2i = 1, 2, let EiE_i and FiF_i be two Hausdorff prehilbertian spaces and let ui∈L(Ei;Fi)u_i \in \mathscr{L}(E_i; F_i). The linear mapping u1⊗u2u_1 \otimes u_2 from E1⊗2E2E_1 \otimes_2 E_2 into F1⊗2F2F_1 \otimes_2 F_2 is continuous and we have

∥u1⊗u2∥=∥u1∥⋅∥u2∥. \|u_1 \otimes u_2\| = \|u_1\| \cdot \|u_2\|.

Consider the positive hermitian form on E1E_1 given by

f(x1,y1)=∥u1∥2⟨x1∣y1⟩−⟨u1(x1)∣u1(y1)⟩. f(x_1, y_1) = \|u_1\|^2 \langle x_1 | y_1 \rangle - \langle u_1(x_1) | u_1(y_1) \rangle.

By prop. 1 (V, p. 25), there exists a positive hermitian form Φ\Phi on E1⊗E2E_1 \otimes E_2 such that

Φ(x1⊗x2,y1⊗y2)=f(x1,y1)⟨x2∣y2⟩= \Phi(x_1 \otimes x_2, y_1 \otimes y_2) = f(x_1, y_1) \langle x_2 | y_2 \rangle =
=∥u1∥2⟨x1⊗x2∣y1⊗y2⟩−⟨(u1⊗1)(x1⊗x2)∣(u1⊗1)(y1⊗y2)⟩ = \|u_1\|^2 \langle x_1 \otimes x_2 | y_1 \otimes y_2 \rangle - \langle (u_1 \otimes 1)(x_1 \otimes x_2) | (u_1 \otimes 1)(y_1 \otimes y_2) \rangle
for x1,y1x_1, y_1 in E1E_1 and x2,y2x_2, y_2 in E2E_2. By linearity we have
Φ(z,t)=∥u1∥2⟨z∣t⟩−⟨(u1⊗1)(z)∣(u1⊗1)(t)⟩ \Phi(z, t) = \|u_1\|^2 \langle z|t \rangle - \langle (u_1 \otimes 1)(z) | (u_1 \otimes 1)(t) \rangle
for z,tz, t in E1⊗E2E_1 \otimes E_2. Since Φ\Phi is positive, we get Φ(z,z)≥0\Phi(z, z) \geq 0, that is, ∥(u1⊗1).z∥2\|(u_1 \otimes 1).z\|_2

≤∥u1∥⋅∥z∥2 for z∈E1⊗2E2, or ∥u1⊗1∥≤∥u1∥. Similarly we prove the inequality ∥1⊗u2∥≤∥u2∥, and since u1⊗u2=(u1⊗1)∘(1⊗u2), we get \leq \|u_1\| \cdot \|z\|_2 \text{ for } z \in E_1 \otimes_2 E_2, \text{ or } \|u_1 \otimes 1\| \leq \|u_1\|. \text{ Similarly we prove the inequality } \|1 \otimes u_2\| \leq \|u_2\|, \text{ and since } u_1 \otimes u_2 = (u_1 \otimes 1) \circ (1 \otimes u_2), \text{ we get}
∥u1⊗u2∥≤∥u1∥⋅∥u2∥. \|u_1 \otimes u_2\| \leq \|u_1\| \cdot \|u_2\|.
On the other hand,
∥u1∥⋅∥u2∥=sup⁡∥x1∥≤1,∥x2∥≤1∥u1(x1)∥⋅∥u2(x2)∥ \|u_1\| \cdot \|u_2\| = \sup_{\|x_1\| \leq 1, \|x_2\| \leq 1} \|u_1(x_1)\| \cdot \|u_2(x_2)\|
=sup⁡∥x1∥≤1,∥x2∥≤1∥(u1⊗u2)(x1⊗x2)∥2≤∥u1⊗u2∥. = \sup_{\|x_1\| \leq 1, \|x_2\| \leq 1} \|(u_1 \otimes u_2)(x_1 \otimes x_2)\|_2 \leq \|u_1 \otimes u_2\|.
This completes the proof of prop. 2. Q.E.D.

Let E1,...,EnE_1, ..., E_n be prehilbertian spaces (n≥2n \geq 2). We define the tensor product E1⊗2...⊗2EnE_1 \otimes_2 ... \otimes_2 E_n (also denoted by ⨂i=1nEi\bigotimes_{i=1}^n E_i) by induction, by

E1⊗2...⊗2En=(E1⊗2...⊗2En−1)⊗2En. E_1 \otimes_2 ... \otimes_2 E_n = (E_1 \otimes_2 ... \otimes_2 E_{n-1}) \otimes_2 E_n.
Hence, by the definition of scalar product, we have
⟨x1⊗...⊗xn∣y1⊗...⊗yn⟩=∏i=1n⟨xi∣yi⟩, \langle x_1 \otimes ... \otimes x_n | y_1 \otimes ... \otimes y_n \rangle = \prod_{i=1}^n \langle x_i | y_i \rangle,
and in particular 1^1
∥x1⊗...⊗xn∥2=∥x1∥⋯∥xn∥, \|x_1 \otimes ... \otimes x_n\|_2 = \|x_1\| \cdots \|x_n\|,
for xi,yix_i, y_i in EiE_i (1≤i≤n1 \leq i \leq n). If the EiE_i are Hausdorff, then so is E1⊗2...⊗2EnE_1 \otimes_2 ... \otimes_2 E_n.

Let F1,...,FnF_1, ..., F_n be prehilbertian spaces and ui∈L(Ei;Fi)u_i \in \mathscr{L}(E_i; F_i) for 1≤i≤n1 \leq i \leq n. By induction on nn, prop. 2 implies that u1⊗...⊗unu_1 \otimes ... \otimes u_n is a continuous linear mapping from E1⊗2...⊗2EnE_1 \otimes_2 ... \otimes_2 E_n into F1⊗2...⊗2FnF_1 \otimes_2 ... \otimes_2 F_n and that

∥u1⊗...⊗un∥=∥u1∥⋯∥un∥. \|u_1 \otimes ... \otimes u_n\| = \|u_1\| \cdots \|u_n\|.
Let σ∈Sn\sigma \in S_n be a permutation of the set {1,2,...,n}\{1, 2, ..., n\}. Because of (6), the linear mapping pσp_\sigma from E1⊗2...⊗2EnE_1 \otimes_2 ... \otimes_2 E_n onto Eσ−1(1)⊗2...⊗2Eσ−1(n)E_{\sigma^{-1}(1)} \otimes_2 ... \otimes_2 E_{\sigma^{-1}(n)} characterized by
pσ(x1⊗...⊗xn)=xσ−1(1)⊗...⊗xσ−1(n) p_\sigma(x_1 \otimes ... \otimes x_n) = x_{\sigma^{-1}(1)} \otimes ... \otimes x_{\sigma^{-1}(n)}
is a prehilbertian space isomorphism ("commutativity of tensor product").

Similarly, consider a partition of {1,2,...,n}\{1, 2, ..., n\} into mm consecutive intervals I1,...,InI_1, ..., I_n with Ik=[ak,ak+1−1]I_k = [a_k, a_{k+1} - 1] for 1≤k≤m1 \leq k \leq m. Put

Fk=⨂i=akak+1−1Ei(1≤k≤m). F_k = \bigotimes_{i=a_k}^{a_{k+1}-1} E_i \quad (1 \leq k \leq m).

1^1 Here again we put ∥z∥2=⟨z∣z⟩1/2\|z\|_2 = \langle z|z \rangle^{1/2} for zz in E1⊗2...⊗2EnE_1 \otimes_2 ... \otimes_2 E_n.

The canonical isomorphism from F1⊗...⊗FmF_1 \otimes ... \otimes F_m onto E1⊗...⊗EnE_1 \otimes ... \otimes E_n which transforms ⨂k=1m⨂i=akak+1−1xi\bigotimes_{k=1}^{m} \bigotimes_{i=a_k}^{a_{k+1}-1} x_i into x1⊗...⊗xnx_1 \otimes ... \otimes x_n (A, II, § 3, No. 9) is a prehilbertian space isomorphism (« associativity of the tensor product »).

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Nothing.

Cites

Transcribed from the printed volume. Not translated.