Bourbaki

Proposition 1

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

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

We have the formulas

∣p1∗∣=W(u)1/2(1) |p^*_1|= W(u)^{1/2} \tag{1}
b(u)=u∘W(u)1/2=u∘∣p1∗∣(2) b(u) =u\circ W(u)^{1/2}=u\circ |p^*_1| \tag{2}
1E−b(u)∗b(u)=W(u)(3) 1_E-b(u)^*b(u) = W(u) \tag{3}
b(u)W(u)=W(u∗)b(u),b(u)W(u)1/2=W(u∗)1/2b(u)(4) b(u)W(u) = W(u^*)b(u),b(u)W(u)^{1/2}= W(u^*)^{1/2}b(u) \tag{4}

Let x∈Ex\in E. Put y=W(u)(x)∈y= W(u)(x)\in dom(u∗uu^*u). We have y∈y\in dom(uu) and u(y)∈u(y)\in dom(u∗u^*). For every element (y1,u(y1))(y_1, u(y_1)) of Γu\Gamma_u, where y1∈y_1\in dom(uu), we compute

⟨(y,u(y))∣(y1,u(y1))⟩=⟨y∣y1⟩+⟨u(y)∣u(y1)⟩ \langle (y, u(y))|(y_1, u(y_1))\rangle =\langle y|y_1\rangle +\langle u(y)|u(y_1)\rangle
=⟨y+u∗u(y)∣y1⟩=⟨x∣p1(y1,u(y1))⟩ =\langle y+u^*u(y)|y_1\rangle =\langle x|p_1(y_1, u(y_1))\rangle

This means that p1∗(x)=(y,u(y))p^*_1(x) = (y, u(y)), whence p1∘p1∗(x)=y=W(u)(x)p_1\circ p^*_1(x) =y= W(u)(x). Consequently, we have p1∘p1∗=W(u)p_1\circ p^*_1= W(u), hence ∣p1∗∣=W(u)1/2|p^*_1|= W(u)^{1/2}, that is, formula (1). In particular, since Im(p1p_1) == Im(∣p1∗∣|p^*_1|) by the Cor. to Prop. 11 of I, p. 140, it follows that dom(uu) == Im(p1p_1) == Im(W(uu)1/2^{1/2}). The partial operator u∘W(u)1/2u\circ W(u)^{1/2} therefore has the space E as domain. The relation p1∘j=(p1∗)∗∘j=∣p1∗∣=W(u)1/2p_1\circ j= (p^*_1)^*\circ j=|p^*_1|= W(u)^{1/2} (Prop. 11, a) of I, p. 140) then implies that

b(u)=p2∘j=u∘p1∘j=u∘W(u)1/2 b(u) =p_2\circ j=u\circ p_1\circ j=u\circ W(u)^{1/2}

which is formula (2).

On has Ker(jj) == Ker(∣p1∗∣|p^*_1|) (prop. 10, b) of I, p. 139) and formula (1) therefore implies that the partial isometry j:E→Γuj: E\rightarrow \Gamma_u is injective, hence isometric, whence j∗∘j=1Ej^*\circ j= 1_E. Writing j=(p1∘j,b(u))j= (p_1\circ j, b(u)), we find

1E=j∗∘j=(p1∘j)∗(p1∘j)+b(u)∗b(u)=W(u)+b(u)∗b(u) 1_E=j^*\circ j= (p_1\circ j)^*(p_1\circ j) +b(u)^*b(u) = W(u) +b(u)^*b(u)

whence formula (3).

Let x∈x\in dom(uu) and put y=W(u)(x)y= W(u)(x). We have y∈y\in dom(uu) and formula y+u∗u(y)=xy+u^*u(y) =x implies that u∗u(y)∈u^*u(y)\in dom(uu). We then have

W(u∗)−1(u(y))=(1E+uu∗)(u(y)) W(u^*)^{-1}(u(y)) = (1_E+uu^*)(u(y))
=u(y)+uu∗u(y)=u(y+u∗u(y))=u(x) =u(y) +uu^*u(y) =u(y+u^*u(y)) =u(x)

which means that the reduction of u∘W(u)u\circ W(u) to the domain of uu is equal to W(u∗)∘uW(u^*)\circ u. Since the image of ∣p1∗∣|p^*_1| is equal to that of p1p_1 (cor. of prop. 11 of I, p. 140) and hence to the domain of uu, it follows that

u∘W(u)∘∣p1∗∣=W(u∗)∘u∘∣p1∗∣=W(u∗)∘b(u) u\circ W(u)\circ |p^*_1|= W(u^*)\circ u\circ |p^*_1|= W(u^*)\circ b(u)

(formula (2)). Moreover, ∣p1∗∣=W(u)1/2|p^*_1|= W(u)^{1/2} commutes with W(u)W(u), hence we obtain

b(u)∘W(u)=u∘∣p1∗∣∘W(u)=u∘W(u)∘∣p1∗∣=W(u∗)∘b(u) b(u)\circ W(u) =u\circ |p^*_1| \circ W(u) =u\circ W(u)\circ |p^*_1|= W(u^*)\circ b(u)

This relation implies that b(u)∘f(W(u))=f(W(u∗))∘b(u)b(u)\circ f(W(u)) =f(W(u^*))\circ b(u) for every function f∈C(R+)f\in \mathscr{C}(\mathbf{R}_+) (prop. 11 of I, p. 113), hence in particular that b(u)∘W(u)1/2=W(u∗)1/2∘b(u)b(u)\circ W(u)^{1/2}= W(u^*)^{1/2}\circ b(u). This proves formula (4) and concludes the proof.

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.