Bourbaki

Corollary 1

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

Topological Vector Spaces, V, p. 35, § 3. Tensor product of hilbertian spaces

a) Let x1,…,xnx_1, \ldots, x_n be elements of the hilbertian space E. We have

∥x1∧…∧xn∥≤∥x1∥…∥xn∥, \| x_1 \wedge \ldots \wedge x_n \| \leq \| x_1 \| \ldots \| x_n \|,

where equality holds only if one of the xix_i is null, or if the sequence (x1,…,xn)(x_1, \ldots, x_n) is orthogonal.

b) Let x1,…,xn,y1,…,ynx_1, \ldots, x_n, y_1, \ldots, y_n be elements of the hilbertian space E. We have

∣det⁡(⟨xi,yj⟩)∣≤∥x1∥…∥xn∥⋅∥y1∥…∥yn∥; |\det(\langle x_i, y_j \rangle)| \leq \| x_1 \| \ldots \| x_n \| \cdot \| y_1 \| \ldots \| y_n \|;

if the vectors xix_i and yiy_i are non-null; equality holds in (36) if and only if (x1,...,xn)(x_1, ..., x_n) and (y1,...,yn)(y_1, ..., y_n) are each an orthogonal basis of the same vector subspace of E.

The inequality (35) follows from prop. 6 by induction on nn; the inequality (36) can be deduced by applying the Cauchy-Schwarz inequality in Λn(E)\Lambda^n(E) and the formula (26) of V, p. 33.

Suppose that the sequence (x1,...,xn)(x_1, ..., x_n) is orthogonal. Then

∥x1∧...∧xn∥2=det⁡(⟨xi∣xj⟩)=∏i=1n∥xi∥2 \| x_1 \wedge ... \wedge x_n \|^2 = \det(\langle x_i | x_j \rangle) = \prod_{i=1}^n \| x_i \|^2

since ⟨xi∣xj⟩=0\langle x_i | x_j \rangle = 0 for i≠ji \neq j.

Now suppose that the vectors x1,...,xnx_1, ..., x_n are not null and do not form an orthogonal sequence. Since ∥x1∧...∧xn∥\| x_1 \wedge ... \wedge x_n \| depends symmetrically on the vectors x1,...,xnx_1, ..., x_n, we can assume that x1x_1 is not orthogonal to the subspace F of E generated by x2,...,xnx_2, ..., x_n, and that F is not simply 0. We can decompose x1x_1 in the form x1′+yx'_1 + y with y≠0y \neq 0 in F and x1′x'_1 orthogonal to F; then ∥x1′∥<∥x1∥\| x'_1 \| < \| x_1 \|. But

x1∧x2∧...∧xn=x1′∧x2∧...∧xn, x_1 \wedge x_2 \wedge ... \wedge x_n = x'_1 \wedge x_2 \wedge ... \wedge x_n,

and so

∥x1∧...∧xn∥≤∥x1′∥∥x2∥...∥xn∥<∥x1∥∥x2∥...∥xn∥. \begin{align*} \| x_1 \wedge ... \wedge x_n \| &\leq \| x'_1 \| \| x_2 \| ... \| x_n \| \\ &< \| x_1 \| \| x_2 \| ... \| x_n \|. \end{align*}

Suppose that the vectors xix_i and the vectors yiy_i are not null. The equality in relation (36) is equivalent to the conjunction of the equalities

(37)

∣⟨x1∧...∧xn∣y1∧...∧yn⟩∣=∥x1∧...∧xn∥⋅∥y1∧...∧yn∥ |\langle x_1 \wedge ... \wedge x_n | y_1 \wedge ... \wedge y_n \rangle| = \| x_1 \wedge ... \wedge x_n \| \cdot \| y_1 \wedge ... \wedge y_n \|

(38)

∥x1∧...∧xn∥=∥x1∥...∥xn∥,∥y1∧...∧yn∥=∥y1∥...∥yn∥. \| x_1 \wedge ... \wedge x_n \| = \| x_1 \| ... \| x_n \| , \quad \| y_1 \wedge ... \wedge y_n \| = \| y_1 \| ... \| y_n \| .

By the first part of the proof, the equalities (38) imply that each of the sequences (x1,...,xn)(x_1, ..., x_n) and (y1,...,yn)(y_1, ..., y_n) is orthogonal, which in turn implies that x1∧...∧xn≠0x_1 \wedge ... \wedge x_n \neq 0 and that y1∧...∧yn≠0y_1 \wedge ... \wedge y_n \neq 0. Under these conditions, relation (37) implies that there exists a scalar λ≠0\lambda \neq 0 such that y1∧...∧yn=λx1∧...∧xny_1 \wedge ... \wedge y_n = \lambda x_1 \wedge ... \wedge x_n (V, p. 3, Remark 1); in other words, that (x1,...,xn)(x_1, ..., x_n) and (y1,...,yn)(y_1, ..., y_n) are bases of the same vector subspace of E (A, III, § 11, No. 13).

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.