Bourbaki

Proposition 4

Tag 09SD, permanent. Cite it as [Bourbaki, Tag 09SD].

Integration, IX, p. 78, § 6. Promeasures and measures on a locally convex space

Let E be a locally convex space. For every positive quadratic form Q on E', there exists one and only one promeasure ΓQ\Gamma_Q on E such that FΓQ=e−Q/2\mathscr{F} \Gamma_Q = e^{-Q/2}. The total mass of ΓQ\Gamma_Q is equal to 1.

The uniqueness of ΓQ\Gamma_Q follows from Prop. 3 of No. 3. The total mass of ΓQ\Gamma_Q is equal to (FΓQ)(0)=e−Q(0)/2=1(\mathscr{F}\Gamma_Q)(0) = e^{-Q(0)/2} = 1. We will prove existence in stages.

A) E of finite dimension n, and Q nondegenerate.

By Lemma 2 of No. 4, the measure γ1\gamma_1 on R\mathbf{R} having density t↦(2π)−1/2e−t2/2t \mapsto (2\pi)^{-1/2}e^{-t^2/2} is bounded, of total mass 1. Set γ=γ1⊗⋯⊗γ1\gamma = \gamma_1 \otimes \cdots \otimes \gamma_1 (n factors). From Lemma 2 of No. 4, one deduces

∫Rnei(a1t1+⋯+antn)dγ(t1,…,tn)=∏j=1n∫Reiajtdγ1(t) \int_{\mathbf{R}^n} e^{i(a_1 t_1 + \cdots + a_n t_n)} d\gamma(t_1, \ldots, t_n) = \prod_{j=1}^n \int_{\mathbf{R}} e^{ia_j t} d\gamma_1(t)
=∏j=1n(2π)−1/2∫Reiajte−t2/2dt = \prod_{j=1}^n (2\pi)^{-1/2} \int_{\mathbf{R}} e^{ia_j t} e^{-t^2/2} dt
=∏j=1ne−aj2/2 = \prod_{j=1}^n e^{-a_j^2/2}
=exp⁡(−12(a12+⋯+an2)). = \exp \left( -\frac{1}{2}(a_1^2 + \cdots + a_n^2) \right).

Since Q is positive and nondegenerate, there exists a basis (e1′,…,en′)(e'_1, \ldots, e'_n) of E' orthonormal for Q (Alg., Ch. IX, §7, No. 1). Let us denote by f the isomorphism x↦(e1′(x),…,en′(x))x \mapsto (e'_1(x), \ldots, e'_n(x)) of E onto Rn\mathbf{R}^n, and by ΓQ\Gamma_Q the measure f−1(γ)f^{-1}(\gamma) on E. Let x′=a1e1′+⋯+anen′x' = a_1 e'_1 + \cdots + a_n e'_n be in E'; then x′(f−1(t1,…,tn))=∑j=1ntjajx'(f^{-1}(t_1, \ldots, t_n)) = \sum_{j=1}^n t_j a_j for t1,…,tnt_1, \ldots, t_n real, whence

∫Eei⟨x,x′⟩dΓQ(x)=∫Rnei(a1t1+⋯+antn)dγ(t1,…,tn) \int_E e^{i\langle x, x' \rangle} d\Gamma_Q(x) = \int_{\mathbf{R}^n} e^{i(a_1 t_1 + \cdots + a_n t_n)} d\gamma(t_1, \ldots, t_n)
=exp⁡(−12(a12+⋯+an2))=exp⁡(−12Q(x′)). = \exp \left( -\frac{1}{2}(a_1^2 + \cdots + a_n^2) \right) = \exp \left( -\frac{1}{2}Q(x') \right).

Consequently, FΓQ=e−Q/2\mathscr{F}\Gamma_Q = e^{-Q/2}.

B) E finite-dimensional.

Let N be the linear subspace of E' formed by the x′x' such that Q(x′)=0Q(x') = 0. Denote by M the orthogonal of N in E, and by j the canonical injection of M into E. The linear mapping tj:E′→M′{}^t j : E' \to M' is surjective, with kernel N, therefore there exists on M′M' a nondegenerate positive quadratic form q such that Q=q∘tjQ = q \circ {}^t j. By the foregoing, there exists a bounded measure Γ\Gamma on M such that FΓ=e−q/2\mathscr{F}\Gamma = e^{-q/2}. Setting ΓQ=j(Γ)\Gamma_Q = j(\Gamma), we have

FΓQ=(FΓ)∘tj=exp⁡(−q∘tj/2)=e−Q/2 \mathscr{F}\Gamma_Q = (\mathscr{F}\Gamma) \circ {}^t j = \exp(-q \circ {}^t j/2) = e^{-Q/2}

by formula (4) of No. 3.

C) The general case. Let V∈F(E)V \in \mathscr{F}(E). Denote by pVp_V the canonical mapping of EE onto E/VE/V, and by QVQ_V the positive quadratic form Q∘tpVQ \circ {}^t p_V on (E/V)′(E/V)'; finally, let μV\mu_V be the measure on E/VE/V with Fourier transform e−QV/2e^{-Q_V/2} (cf. B)). If W∈F(E)W \in \mathscr{F}(E) is contained in VV, then pV=pVW∘pWp_V = p_{VW} \circ p_W, whence QV=QW∘tpVWQ_V = Q_W \circ {}^t p_{VW}; by formula (4) of No. 3, the measure pVW(μW)p_{VW}(\mu_W) has as Fourier transform the function (e−QW/2)∘tpVW=e−QV/2(e^{-Q_W/2}) \circ {}^t p_{VW} = e^{-Q_V/2}, hence is equal to μV\mu_V. The family (μV)V∈F(E)(\mu_V)_{V \in \mathscr{F}(E)} is therefore a promeasure μ\mu on EE. Formula (5) of No. 3 shows that Fμ\mathscr{F}\mu is equal to e−Q/2e^{-Q/2}.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Cites

Transcribed from the printed volume. Not translated.