Bourbaki

Proposition 5

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

Théories spectrales, chapter V, § 2. Représentations des groupes localement compacts

Let p∈[1,+∞[p\in [1,+\infty [. The space Lπp(G)\mathscr{L}_{\pi}^p(G) is the space of functions f∈Fπ(G)f\in \mathscr{F}_{\pi}(G) such that ff is μ\mu-measurable and the function ∥f∥\|f\| belong to Lp(G/H)\mathscr{L}^p(G/H).

Let f∈Lπp(G)f\in \mathscr{L}_{\pi}^p(G). It is the limit in Lπp(G)\mathscr{L}_{\pi}^p(G) of a sequence of elements of Kπ(G)\mathscr{K}_{\pi}(G). By prop. 3, b) and Egoroff's theorem (INT, IV, p. 175, § 5, No.o4^o4, th. 2), the function ff is therefore μ\mu-measurable; consequently, the function ∥f∥\|f\| on G/HG/H is ν\nu-measurable (INT, VII, p. 47, § 2, No.o3^o3, prop. 6, b)). Since Np(f)N_p(f) is finite, the function ∥f∥\|f\| belong to Lp(G/H)\mathscr{L}^p(G/H) (INT, IV, p. 184, § 5, No.o6^o6, th. 5).

Let us prove the converse assertion. Let ff be a μ\mu-measurable function belonging to Fπ(G)\mathscr{F}_{\pi}(G) such that ∥f∥∈Lp(G/H)\|f\| \in \mathscr{L}^p(G/H). Let ε>0\varepsilon >0. Let us prove that there exists f~∈Kπ(G)\widetilde{f}\in \mathscr{K}_{\pi}(G) such that

Np(f−f~)p=∫G/H∗∥f−f~∥pdν⩽ε N_p(f-\widetilde{f})^p=\int_{G/H}^*\|f-\widetilde{f}\|^pd\nu \leqslant \varepsilon

which will conclude the proof.

Suppose first that there exists a compact subset C of G such that ff is zero outside C⋅HC\cdot H. Let u∈K+(G)u\in \mathscr{K}_+(G) be a function such that u(x)>0u(x)>0 for all x∈Cx\in C and define v=u1v=u^1 as above. Let us denote by φ\varphi the characteristic function of the support of uu.

Let qq be the conjugate exponent of pp. If p=1p= 1, let ww be the constant function on G equal to supx∈Gu(x)_{x\in G}u(x). If p>1p >1, define

w(x)=(∫Hu(xh)q dβ(h))1/q w(x)=\left(\int_Hu(xh)^q\,d\beta(h)\right)^{1/q}

for every x∈Gx\in G. In both cases, the function ww is continuous and positive; it belongs to K1(G)\mathscr{K}_1(G) (Prop. 4 applied to the trivial representation of dimension 1 and to the function uqu^q), hence it is bounded on G. Put W = supx∈Gw(x)_{x\in G}w(x).

Let ss be the function on G with values in E defined by Lemma 5 applied to ff. Put g=φsg=\varphi s. The function gg is μ\mu-measurable, and satisfies ∥g∥⩽∥f∥/\|g\|\leqslant \|f\|/ infx∈C⋅Hv(x)_{x\in C\cdot H}v(x). Since gg vanishes outside the support of uu and κ\kappa is continuous, we have g∈LEp(G,κ⋅μ)g\in \mathscr{L}_E^p(G, \kappa \cdot \mu). Let g~\widetilde{g} be a function in KE(G)\mathscr{K}_E(G) such that

∫∗∥g−g~∥pκdμ⩽εp \int^*\|g-\widetilde{g}\|^p\kappa d\mu\leqslant \varepsilon_p

G_G W

We have us=ugus=ug, hence f=(us)π=(ug)πf= (us)^{\pi}= (ug)^{\pi} almost everywhere modulo H (Lemma 5, b)). Let f~=(ug~)π\widetilde{f}= (u\widetilde{g})^{\pi}; we have f~∈Kπ(G)\widetilde{f}\in \mathscr{K}_{\pi}(G) (Prop. 4, b)). For every x∈Gx\in G - SugS_{ug}, we obtain

∥(ug)π(x)−(ug~)π(x)∥⩽∫H∗u(xh)∥g(xh)−g~(xh)∥dβ(h) \|(ug)^{\pi}(x)-(u\widetilde{g})^{\pi}(x)\|\leqslant \int_H^*u(xh)\|g(xh)-\widetilde{g}(xh)\|d\beta (h)

whence

∥(ug)π(x)−(ug~)π(x)∥p⩽w(x)p∫H∗∥g(xh)−g~(xh)∥pdβ(h) \|(ug)^{\pi}(x)-(u\widetilde{g})^{\pi}(x)\|^p\leqslant w(x)^p\int_H^*\|g(xh)-\widetilde{g}(xh)\|^pd\beta (h)

by Hölder's inequality in the case p>1p >1. Since SugS_{ug} is negligible modulo H, it follows that

∫G/H∗∥f−f~∥pdν⩽Wp∫G/H∗(∫H∗∥g(xh)−g~(xh)∥dβ(h))pdν(xH) \int_{G/H}^*\|f-\widetilde{f}\|^pd\nu \leqslant W^p\int_{G/H}^*(\int_H^*\|g(xh)-\widetilde{g}(xh)\|d\beta (h))^pd\nu (xH)
=Wp∫G∗∥g−g~∥pκdμ⩽ε = W^p\int_G^*\|g-\widetilde{g}\|^p\kappa d\mu\leqslant \varepsilon

by INT, VII, p. 46, § 2, no3^o3, Prop. 5, b), which is applicable since g−g~g-\widetilde{g} vanishes outside a compact subset of G. This implies the required property when ff vanishes outside a compact subset modulo H.

Consider now the general case. Since ∥f∥∈Lp(G/H)\|f\| \in \mathscr{L}^p(G/H) by hypothesis, there exists a compact subset L of G/HG/H such that

∫∗pε \int^*p\varepsilon
∥f∥dν⩽ \|f\|d\nu \leqslant

(G/H)−L(G/H)-L 2

(cf. INT, IV, p. 152, § 4, no6^o6, th. 4). Let φL\varphi_L be the characteristic function of L and put fL=(φL∘ϖ)ff_L= (\varphi_L\circ \varpi )f. We have

∫∗p∫∗pε \int^*p\int^*p\varepsilon
∥f−fL∥dν=∥f∥dν⩽ \|f-f_L\|d\nu =\|f\|d\nu \leqslant

G/H(G/H)−LG/H(G/H)-L 2

The function fLf_L is μ\mu-measurable and vanishes outside a compact set modulo H. It belongs to Lπp\mathscr{L}_{\pi}^p(G), hence by the preceding case, there exists f~∈Kπ(G)\widetilde{f}\in \mathscr{K}_{\pi}(G) such that Np(fL−f~)⩽(ε2)1/pN_p(f_L-\widetilde{f})\leqslant (\frac{\varepsilon}{2})^{1/p}, whence

∫G/H∗∥f−f~∥pdν⩽ε \int_{G/H}^*\|f-\widetilde{f}\|^pd\nu \leqslant \varepsilon

as desired.

Let us consider the case p=2p= 2. For f1f_1 and f2f_2 in Fπ\mathscr{F}_{\pi}(G), the function x↦⟨f1(x)∣f2(x)⟩x\mapsto \langle f_1(x)|f_2(x)\rangle belongs to F1(G)\mathscr{F}_1(G) since the representation π\pi is unitary, and therefore defines by passing to the quotient a function on G/HG/H, which is identified as above with ⟨f1∣f2⟩\langle f_1|f_2\rangle. One has the majorization ∣⟨f1∣f2⟩∣⩽∥f1∥∥f2∥|\langle f_1|f_2\rangle |\leqslant \|f_1\|\|f_2\| in F(G/H)\mathscr{F}(G/H).

If f1f_1 and f2f_2 belong to Lπ2\mathscr{L}_{\pi}^2(G), then the function ⟨f1∣f2⟩\langle f_1|f_2\rangle belongs to L11(G)\mathscr{L}_1^1(G). In particular, it is integrable on G/HG/H. The mapping

(f1,f2)↦∫G/H⟨f1∣f2⟩dν (f_1, f_2)\mapsto \int_{G/H}\langle f_1|f_2\rangle d\nu

is a positive Hermitian form on Lπ2(G)\mathscr{L}_{\pi}^2(G); the associated seminorm is the seminorm N2N_2. In particular, the space Lπ2(G)\mathscr{L}_{\pi}^2(G) is a prehilbertian space and Lπ2(G)L^2_{\pi}(G) is the Hilbert space associated with Lπ2(G)\mathscr{L}_{\pi}^2(G).

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.