Bourbaki

Lemma 5

Tag 0821, permanent. Cite it as [Bourbaki, Tag 0821].

Topological Vector Spaces, chapter V, § 4. Some classes of operators in hilbertian spaces

Let E1E_1 be a closed vector subspace of E, stable under u∗uu^*u. If E1≠{0}E_1 \neq \{0\}, then there exists a vector of norm 1 in E1∩PE_1 \cap P.

Since B is weakly compact, so is the weakly closed subspace B∩E1B \cap E_1 of B. Hence there exists (GT, IV, § 6, No. 1, th. 1) a point x0x_0 in B∩E1B \cap E_1 such that Q(x0)≥Q(x)Q(x_0) \geq Q(x) for all x∈B∩E1x \in B \cap E_1. If Q(x0)=0Q(x_0) = 0, we have Q(x)=0Q(x) = 0 and so u(x)=0u(x) = 0 for all x∈B∩E1x \in B \cap E_1. Thus E1⊂PE_1 \subset P and lemma 5 follows in this case.

Suppose now that Q(x0)>0Q(x_0) > 0, then x0≠0x_0 \neq 0. Since the vector ∥x0∥−1⋅x0\|x_0\|^{-1} \cdot x_0 belongs to B∩E1B \cap E_1, we have

Q(x0)≥Q(∥x0∥−1⋅x0)=Q(x0)/∥x0∥2 Q(x_0) \geq Q(\|x_0\|^{-1} \cdot x_0) = Q(x_0)/\|x_0\|^2
i.e. ∥x0∥=1\|x_0\| = 1. We shall prove that x0x_0 belongs to P; let y∈Ey \in E be orthogonal to x0x_0. It is enough to prove that u(y)u(y) is orthogonal to u(x0)u(x_0). But since yy is the sum of a vector of E1E_1 and a vector orthogonal to E1E_1, and both orthogonal to x0x_0 (since x0∈E1x_0 \in E_1), it is enough to consider the following two cases :

a) yy is orthogonal to E1E_1: since E1E_1 is stable under u∗uu^*u, u∗u(x0)∈E1u^*u(x_0) \in E_1, hence 0=⟨y∣u∗u(x0)⟩=⟨u(y)∣u(x0)⟩0 = \langle y|u^*u(x_0)\rangle = \langle u(y)|u(x_0)\rangle.

b) yy belongs to E1E_1: for all t∈Rt \in \mathbf{R}, the vector x(t)=(x0+ty)/∥x0+ty∥x(t) = (x_0 + ty)/\|x_0 + ty\| belongs to B∩E1B \cap E_1. We have Q(xt)=f(t)/g(t)Q(xt) = f(t)/g(t) with

f(t)=∥u(x0)∥2+2tℜ⟨u(x0)∣u(y)⟩+t2∥u(y)∥2g(t)=1+t2∥y∥2. f(t) = \|u(x_0)\|^2 + 2t \Re \langle u(x_0)|u(y)\rangle + t^2 \|u(y)\|^2 \\ g(t) = 1 + t^2 \|y\|^2 .
In view of the definition of x0x_0, we have Q(x(0))≥Q(x(t))Q(x(0)) \geq Q(x(t)) for all real tt, hence ddtQ(x(t))\frac{d}{dt} Q(x(t)) is zero for t=0t = 0. But f(0)=∥u(x0)∥2f(0) = \|u(x_0)\|^2, g(0)=1g(0) = 1, f′(0)=2ℜ⟨u(x0)∣u(y)⟩f'(0) = 2 \Re \langle u(x_0)|u(y)\rangle, g′(0)=0g'(0) = 0. Since
ddtQ(x(t))=f′(t) g(t)−f(t) g′(t)g(t)2, \frac{d}{dt} Q(x(t)) = \frac{f'(t)\,g(t) - f(t)\,g'(t)}{g(t)^2},
we conclude that f′(0)=0f'(0) = 0, that is, ℜ⟨u(x0)∣u(y)⟩=0\Re \langle u(x_0)|u(y)\rangle = 0. When K=RK = \mathbf{R}, u(x0)u(x_0) is orthogonal to u(y)u(y), when K=CK = \mathbf{C}, the vector iyiy belongs to E1E_1 and is orthogonal to x0x_0, hence I⟨u(x0)∣u(y)⟩=−ℜ⟨u(x0)∣u(iy)⟩=0\mathscr{I} \langle u(x_0)|u(y)\rangle = -\Re \langle u(x_0)|u(iy)\rangle = 0, and finally u(x0)u(x_0) is orthogonal to u(y)u(y). This proves lemma 5.

Now we prove th. 2. Applying th. 1 of S, III, § 4, No. 5 we see, as in V, p. 23, that there exists a set SS which is maximal among the orthonormal subsets of EE contained in PP. Let E1E_1 be the set of all vectors orthogonal to SS. Let y∈E1y \in E_1; if x∈Sx \in S, the vectors xx and yy are orthogonal, and since S⊂PS \subset P, we conclude that u(x)u(x) and u(y)u(y) are orthogonal; then

⟨x∣u∗u(y)⟩=⟨u(x)∣u(y)⟩=0 \langle x|u^*u(y)\rangle = \langle u(x)|u(y)\rangle = 0
and u∗u(y)u^*u(y) is orthogonal to SS. Hence E1E_1 is stable under u∗uu^*u. If we had E1≠{0}E_1 \neq \{0\}, there would exist a vector xx of norm 1 in E1∩PE_1 \cap P (lemma 5) and S∪{x}S \cup \{x\} would be an orthonormal subset of EE contained in PP. This would contradict the maximal character of SS. Hence E1={0}E_1 = \{0\} and SS is an orthonormal basis of EE. Q.E.D.

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.