Bourbaki

Proposition 5

Tag 03C9, permanent. Cite it as [Bourbaki, Tag 03C9].

Théories spectrales, chapter V, § 3. Fonctions de type positif

Let A be an involutive Banach algebra admitting an approximate unit. Every continuous positive linear form on A admits a cyclic Hilbertian realization (E,x,φ)(E, x, \varphi ). If A is unital, then there exists such a realization in which the morphism φ\varphi is unital.

One may suppose that A is a unital algebra (Lemma 3, b)).

Let λ\lambda be a continuous positive linear form on A. Let (E,g)(E, g) be a cyclic Hilbertian realization of the universally positive kernel ff on A defined by f(a,b)=λ(a∗b)f(a, b) =\lambda (a^*b) (Prop. 4 and Theorem 1). The mapping gg is continuous, its image is total in E, and one has λ(a∗b)=⟨g(a)∣g(b)⟩\lambda (a^*b) =\langle g(a)|g(b)\rangle for every (a,b)∈A2(a, b)\in A^2. Put x=g(1)∈Ex=g(1)\in E.

The mapping gg of A into E is linear: in fact, for (a,b,c)∈A3(a, b, c)\in A^3 and (s,t)∈C2(s, t)\in \mathbf{C}^2, we have

⟨g(c)∣g(sa+tb)⟩=λ(c∗(sa+tb)) \langle g(c)|g(sa+tb)\rangle =\lambda (c^*(sa+tb))
=sλ(c∗a)+tλ(c∗b)=⟨g(c)∣sg(a)+tg(b)⟩ =s\lambda (c^*a) +t\lambda (c^*b) =\langle g(c)|sg(a) +tg(b)\rangle

whence the assertion, since the image of gg is total in E.

In particular, the image F of gg is a dense vector subspace of E. The kernel of gg is a left ideal of A: if g(b)=0g(b) = 0, then for all aa and cc in A, we have

⟨g(ab)∣g(c)⟩=λ((ab)∗c)=λ(b∗(a∗c))=⟨g(b)∣g(a∗c)⟩=0 \langle g(ab)|g(c)\rangle =\lambda ((ab)^*c) =\lambda (b^*(a^*c)) =\langle g(b)|g(a^*c)\rangle = 0

therefore g(ab)=0g(ab) = 0 since F is dense in E.

Let a∈Aa\in A. Since the kernel of gg is a left ideal of A, there exists a linear mapping φ~(a):F→F\widetilde{\varphi}(a) : F\rightarrow F such that φ~(a)(g(b))=g(ab)\widetilde{\varphi}(a)(g(b)) =g(ab) for every b∈Ab\in A. Moreover, for every b∈Ab\in A, we have

∥φ~(a)(g(b))∥2=∥g(ab)∥2=λ(b∗a∗ab)⩽∥a∗a∥λ(b∗b)⩽∥a∥2∥g(b)∥2 \|\widetilde{\varphi}(a)(g(b))\|^2=\|g(ab)\|^2=\lambda (b^*a^*ab)\leqslant \|a^*a\|\lambda (b^*b)\leqslant \|a\|^2\|g(b)\|^2

(lemma 3, c)), therefore φ~(a)\widetilde{\varphi}(a) is continuous and of norm ⩽∥a∥\leqslant \|a\|. Consequently, there exists a unique endomorphism φ(a)∈L(E)\varphi (a)\in \mathscr{L}(E) which induces φ~(a)\widetilde{\varphi}(a) by passing to the subspaces.

Let aa and bb be in A. We have

(φ(a)∘φ(b))(g(c))=g(abc)=φ(ab)(g(c)) (\varphi (a)\circ \varphi (b))(g(c)) =g(abc) =\varphi (ab)(g(c))

for every cc in A, therefore φ(ab)=φ(a)∘φ(b)\varphi (ab) =\varphi (a)\circ \varphi (b) since F is dense in E.

The mapping φ\varphi of A into L(E)\mathscr{L}(E) is linear: in fact, for every (a,b)∈A2(a, b)\in A^2 and every (s,t)∈C2(s, t)\in \mathbf{C}^2, we have

φ(sa+tb)(g(c))=g((sa+tb)c)=(sφ(a)+tφ(b))g(c) \varphi (sa+tb)(g(c)) =g((sa+tb)c) = (s\varphi (a) +t\varphi (b))g(c)

for every c∈Ac\in A, whence φ(sa+tb)=sφ(a)+tφ(b)\varphi (sa+tb) =s\varphi (a) +t\varphi (b) since F is dense in E. Since ∥φ(b)∥⩽∥b∥\|\varphi (b)\|\leqslant \|b\| for every b∈Ab\in A, the mapping φ\varphi is continuous. We also have φ(1)=1E\varphi (1) = 1_E since φ(1)(g(a))=g(a)\varphi (1)(g(a)) =g(a) for every a∈Aa\in A and F is dense in E.

Lastly, let a,b,ca,b,c be in A. We have

⟨g(c)∣φ(a∗)(g(b))⟩=⟨g(c)∣g(a∗b)⟩=λ(c∗a∗b) \langle g(c)|\varphi (a^*)(g(b))\rangle =\langle g(c)|g(a^*b)\rangle =\lambda (c^*a^*b)
=⟨g(ac)∣g(b)⟩=⟨φ(a)(g(c))∣g(b)⟩ =\langle g(ac)|g(b)\rangle =\langle \varphi (a)(g(c))|g(b)\rangle

whence φ(a∗)=φ(a)∗\varphi (a^*) =\varphi (a)^* since F is dense in E.

In conclusion, the mapping φ\varphi is a continuous morphism of involutive algebras of A into L(E)\mathscr{L}(E) and g(a)=φ(a)xg(a) =\varphi (a)x for every a∈Aa\in A; consequently, the triplet (E,x,φ)(E, x, \varphi ) is a cyclic Hilbertian realization of λ\lambda.

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.