Bourbaki

Proposition 2

Tag 035D, permanent. Cite it as [Bourbaki, Tag 035D].

Théories spectrales, chapter IV, § 5. Opérateurs partiels normaux et théorème spectral

The mapping u↦b(u)u\mapsto b(u) defines a bijection of the set of closed partial operators with dense domain on E onto the set Ω(E). The inverse bijection is given by v↦B(v)v\mapsto B(v).

The relation 1E−b(u)∗b(u)=W(u)1_E-b(u)^*b(u) = W(u) (formula (3)) implies that b(u)b(u) belongs to Ω(E) since W(u)W(u) is positive and injective. Since moreover b(u)=u∘W(u)1/2b(u) =u\circ W(u)^{1/2} (formula (2)), one obtains

u=b(u)∘(W(u)1/2)−1=b(u)∘((1−b(u)∗b(u))1/2)−1=B(b(u)) u=b(u)\circ (W(u)^{1/2})^{-1}=b(u)\circ ((1-b(u)^*b(u))^{1/2})^{-1}= B(b(u))

Conversely, let v∈Ω(E)v\in \Omega (E). Put w=(1E−v∗v)1/2w= (1_E-v^*v)^{1/2} and let u=B(v)=v∘w−1u= B(v) =v\circ w^{-1}. The domain of uu is the image of ww, which is dense in E since ww is hermitian and injective (EVT, V, p. 41, prop. 4 (i)). For every x∈x\in dom(uu), one has

∥u(x)∥2=⟨(v∗v)(w−1(x))∣w−1(x)⟩ \|u(x)\|^2=\langle (v^*v)(w^{-1}(x))|w^{-1}(x)\rangle
=−⟨(1E−v∗v)(w−1(x))∣w−1(x)⟩+∥w−1(x)∥2 =-\langle (1_E-v^*v)(w^{-1}(x))|w^{-1}(x)\rangle +\|w^{-1}(x)\|^2
=−⟨w(x)∣w−1(x)⟩+∥w−1(x)∥2 =-\langle w(x)|w^{-1}(x)\rangle +\|w^{-1}(x)\|^2

since (1E−v∗v)∘w−1(1_E-v^*v)\circ w^{-1} is the restriction of ww to the domain of w−1w^{-1}. Since ww is self-adjoint, one deduces that

∥w−1(x)∥2=∥x∥2+∥v∘w−1(x)∥2=∥x∥2+∥u(x)∥2 \|w^{-1}(x)\|^2=\|x\|^2+\|v\circ w^{-1}(x)\|^2=\|x\|^2+\|u(x)\|^2

and the partial operator B(v)=u=v∘w−1B(v) =u=v\circ w^{-1} is therefore closed by Lemma 2 of IV, p. 231.

Finally let us prove that b(B(v))=vb(B(v)) =v. Since u=v∘w−1u=v\circ w^{-1}, one has v=u∘wv=u\circ w, and it is therefore enough to verify that w=W(u)1/2w= W(u)^{1/2} (formula (2)), or even that 1E−v∗v=W(u)1_E-v^*v= W(u).

The endomorphism v∗v^* belongs to Ω(E) by Lemma 1. Put w′=(1E−vv∗)1/2w'= (1_E-vv^*)^{1/2}. It is a positive injective endomorphism of E, and one has v∘w=w′∘vv\circ w=w'\circ v (prop. 11 of I, p. 113). The graph of uu is the set of elements of the form (w(x),v(x))(w(x), v(x)) for x∈Ex\in E. By prop. 7 of IV, p. 236, the graph of u∗u^* is s(Γu)∘s(\Gamma_u)^{\circ}. Since, for all xx and y∈Ey\in E, one has

⟨(w′(x),v∗(x))∣(−v(y),w(y))⟩=−⟨x∣w′v(y)⟩+⟨x∣vw(y)⟩=0 \langle (w'(x), v^*(x))|(-v(y), w(y))\rangle =-\langle x|w'v(y)\rangle +\langle x|vw(y)\rangle = 0

the graph of u∗u^* contains elements of the form (w′(x),v∗(x))(w'(x), v^*(x)) for x∈Ex\in E, and one then has u∗(w′(x))=v∗(x)u^*(w'(x)) =v^*(x). In particular, the domain of u∗u^* contains the image of w′w'.

Let x∈Ex\in E and put y=w2(x)=(1E−v∗v)(x)y=w^2(x) = (1_E-v^*v)(x), so that x=y+v∗v(x)x=y+v^*v(x). One has y∈y\in dom(uu) and u(y)=v(w−1(y))=v(w(x))=w′(v(x))u(y) =v(w^{-1}(y)) =v(w(x)) =w'(v(x)). In particular, u(y)∈u(y)\in Im(ww)⊂\subset dom(u∗u^*), and u∗(u(y))=v∗(v(x))u^*(u(y)) =v^*(v(x)). Hence there follows x=y+v∗v(x)=y+u∗u(y)x=y+v^*v(x) =y+u^*u(y), whence y=W(u)(x)y= W(u)(x), that is to say

(1E−v∗v)(x)=W(u)(x) (1_E-v^*v)(x) = W(u)(x)

Thus it has been proved that 1E−v∗v=W(u)1_E-v^*v= W(u), as desired.

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.

Cited by

Nothing.

Cites

Nothing.

Machine translation of the English, by gpt-5.4, gpt-5-6-mini, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.