Bourbaki

§ 8. Fourier transform

We retain the notations and conventions of the preceding paragraph.

5^5 The proof of loc. cit., which has been stated only for split semi-simple Lie algebras,

is valid without change in the case of split reductive Lie algebras.

1. Fourier transforms of integrable functions

In this number, we recall some definitions and results from Spectral Theories6^6.

Denote by G^\widehat{G} the set of classes of irreducible representations of G (on finite dimensional complex vector spaces). For all u∈G^u\in \widehat{G}, denote by EuEu the space of uu and d(u)d(u) its dimension. There exist separating positive hermitian forms on EuE_u invariant under uu, and any two such forms are proportional. Denote by A∗A^* (resp. ∥A∥∞\|A\|_{\infty}) the adjoint (resp. the norm) of an element A of End(Eu_u) relative to one of these forms; for all g∈Gg\in G, we have u(g)∗=u(g)−1=u(g−1)u(g)^*=u(g)^{-1}=u(g^{-1}) and ∥u(g)∥∞=1\|u(g)\|_{\infty}= 1; for all x∈gx\in \mathfrak{g}, we have u(x)∗=−u(x)=u(−x)u(x)^*=-u(x) =u(-x).

Give End(Eu_u) the Hilbert space structure for which the scalar product is

⟨A∣B⟩=d(u)Tr(A∗B)=d(u)Tr(BA∗)\langle A|B\rangle =d(u)\mathrm{Tr}(A^*B) =d(u)\mathrm{Tr}(BA^*), (1)

and put

∥A∥2=⟨A∣A⟩1/2=(d(u)Tr(A∗A))1/2\|A\|_2=\langle A|A\rangle^{1/2}= (d(u)\mathrm{Tr}(A^*A))^{1/2}. (2)

We have

√d(u)‾∥A∥∞≤∥A∥2≤d(u)∥A∥∞(3) \surd\overline{d(u)}\|A\|_{\infty}\leq \|A\|_2\leq d(u)\|A\|_{\infty} \tag{3}

so

∣⟨A∣B⟩∣≤d(u)2∥A∥∞∥B∥∞(4) |\langle A|B\rangle | \leq d(u)^2\|A\|_{\infty}\|B\|_{\infty} \tag{4}

For all g∈Gg\in G, we have ∥u(g)∥2=d(u)\|u(g)\|_2=d(u).

Denote by F(G^)F( \widehat{G}) the algebra ∏u∈G^\prod_{u\in\widehat{G}} End(Eu_u). Denote by L2(G^)L^2( \widehat{G}) the Hilbert sum of the Hilbert spaces End(Eu_u); this is the space of families A=(Au)∈F(G^)A= (A_u)\in F( \widehat{G}) such that ∑u∥Au∥22<∞\sum_u\|A_u\|^2_2<\infty, with the scalar product

⟨A∣B⟩=∑u∈G^⟨Au∣Bu⟩=∑u∈G^d(u\langle A|B\rangle =\sum_{u\in\widehat{G}}\langle A_u|B_u\rangle =\sum_{u\in\widehat{G}}d(u)Tr(Au∗Bu^*_uB_u). (5)

Denote the Hilbert norm on L2(G^)L^2( \widehat{G}) also by ∥∥2\| \|_2, so that ∥A∥22=∑u∈G^∥Au∥22\|A\|^2_2=\sum_{u\in\widehat{G}}\|A_u\|^2_2 for A∈L2(G^)A\in L^2( \widehat{G}).

If ff is an integrable complex function on G, put

u(f)=∫Gf(g)u(g)dg∈u(f) =\int_Gf(g)u(g)dg\in End(Eu_u) (6)

for all u∈G^u\in \widehat{G}. We have ∥u(f)∥∞≤∫G∣f(g)∣dg=∥f∥1\|u(f)\|_{\infty}\leq \int_G|f(g)|dg=\|f\|_1. The Fourier cotransform of ff, denoted by F(f)\mathscr{F}(f), is the family (u(f))u∈G^∈F(G^)(u(f))_{u\in\widehat{G}}\in F( \widehat{G}). If f∈L2(G^)f\in L^2( \widehat{G}),

6^6 See note 1^1, §7, p. 66.

∥f∥22=∑u∈G^⟨u(f)∣u(f)⟩=∥F‾(f)∥22 \|f\|^2_2=\sum_{u\in\widehat{G}}\langle u(f)|u(f)\rangle =\|\overline{\mathscr{F}}(f)\|^2_2

so F‾\overline{\mathscr{F}} induces an isometric linear map from the Hilbert space L2(G)L^2(G) to the Hilbert space L2(G^):L^2( \widehat{G}): in other words, for ff and f′f' in L2L^2(G), we have

∫Gf(g)‾f′(g)dg=⟨F‾(f)∣F‾(f′)⟩=∑u∈G^d(u\int_G\overline{f(g)}f'(g)dg=\langle \overline{\mathscr{F}}(f)|\overline{\mathscr{F}}(f')\rangle =\sum_{u\in\widehat{G}}d(u)Tr(u(f)∗u(f′)u(f)^*u(f')). (7)

For ff and f′f' in L1L^1(G), the convolution product f∗f′f*f' of ff and f′f' is defined by

(f∗f′)(h)=∫Gf(hg−1)f′(g)dg=∫Gf(g)f′(g−1h)dg (f*f')(h) =\int_Gf(hg^{-1})f'(g)dg=\int_Gf(g)f'(g^{-1}h)dg

(the integral makes sense for almost all h∈Gh\in G).

We have f∗f′∈L1(G)f*f'\in L^1(G) and, for all u∈G^,u(f∗f′)=u(f)u(f′)u\in \widehat{G},u(f*f') =u(f)u(f'), so

F‾(f∗f′)=F‾(f).F‾(f′)(8) \overline{\mathscr{F}}(f*f') =\overline{\mathscr{F}}(f).\overline{\mathscr{F}}(f') \tag{8}

Conversely, let A=(Au)u∈G^A = (A_u)_{u\in\widehat{G}} be an element of F(G^)F( \widehat{G}); for all u∈G^u\in \widehat{G}, let FuA\mathscr{F}_uA be the (analytic) function on G defined by

(FuA)(g)=⟨u(g)∣Au⟩=d(u(\mathscr{F}_uA)(g) =\langle u(g)|A_u\rangle =d(u)Tr(Auu(g)−1_uu(g)^{-1}). (9)

If A∈L2(G^)A\in L^2( \widehat{G}), the family (FuA)u∈G^(\mathscr{F}_uA)_{u\in\widehat{G}} is summable in L2(G)L^2(G); the Fourier transform of A, denoted by F\mathscr{F}(A), is the sum of this family. The maps F\mathscr{F} and F\mathscr{F} are inverse isomorphisms between the Hilbert spaces L2(G)L^2(G) and L2(G^)L^2( \widehat{G}).

In other words:

Proposition 1 01GZ

Every square integrable complex function ff on G is the sum in the Hilbert space L2(G)L^2(G) of the family (fu)u∈G^(f_u)_{u\in\widehat{G}} where, for all h∈Gh\in G and all u∈G^u\in \widehat{G},

fu(h)=⟨u(h)∣u(f)⟩ f_u(h) =\langle u(h)|u(f)\rangle

=d(u)∫Gf(g=d(u)\int_Gf(g)Tr(u(gh−1)u(gh^{-1}))dg=d(u)∫Gf(ghdg=d(u)\int_Gf(gh)Tr(u(g)u(g))dgdg. (10)

For all u∈G^u\in \widehat{G} choose an orthonormal basis BuBu of EuE_u, and denote by (uij(g))(u_{ij}(g)) the matrix of u(g)u(g) in this basis. Prop. 1 also means that the family of functions √d(u)‾uij\surd\overline{d(u)}u_{ij}, for uu in G^\widehat{G} and i,ji, j in BuB_u, is an orthonormal basis of the space L2(G)L^2(G).

If ff is an integrable function on G such that the family (fu)(f_u) is uniformly summable, then the sum of this family is a continuous function which coincides almost everywhere with ff; in other words, if we assume in addition that ff is continuous, then for all h∈Gh\in G, f(h)=∑u∈G^d(u)∫Gf(ghf(h) =\sum_{u\in\widehat{G}}d(u)\int_Gf(gh)Tr(u(g)u(g))dgdg. (11)

Conversely, let A∈F(G^)A\in F( \widehat{G}); if the family (FuA)u∈G^(\mathscr{F}_uA)_{u\in\widehat{G}} is uniformly summable, the function

g→∑u∈G^(FuA)(g)=∑u∈G^d(ug \rightarrow \sum_{u\in\widehat{G}}(\mathscr{F}_uA)(g) =\sum_{u\in\widehat{G}}d(u)Tr(Auu(g)−1_uu(g)^{-1})

is a continuous function on G whose Fourier cotransform is A.

Let ff be an integrable function on G, and let s∈Gs\in G. Denote by γ(s)f\gamma (s)f and δ(s)f\delta (s)f the functions on G defined by γ(s)f=εs∗f,δ(s)f=f∗εs−1\gamma (s)f=\varepsilon_s*f,\delta (s)f=f*\varepsilon_{s^{-1}}, that is,

(γ(s)f)(g)=f(s−1g),(δ(s)f)(g)=f(gs)(\gamma (s)f)(g) =f(s^{-1}g),(\delta (s)f)(g) =f(gs) for g∈Gg\in G,

(Chap. III, §3, no. 4 and Integration, Chap. VII, §1, no. 1). We have

u(γ(s)f)=∫Gf(s−1g)u(g)dg=∫Gf(g)u(sg)dg u(\gamma (s)f) =\int_Gf(s^{-1}g)u(g)dg=\int_Gf(g)u(sg)dg

so

u(γ(s)f)=u(s)u(f)(12) u(\gamma (s)f) =u(s)u(f) \tag{12}

and similarly

u(δ(s−1)f)=u(f)u(s)(13) u(\delta (s^{-1})f) =u(f)u(s) \tag{13}

When G is commutative, G^\widehat{G} is the underlying set of the dual group of G (Spectral Theories, Chap. II, §1, no. 11),d(u)=1,d(u) = 1 for all u∈G^u\in \widehat{G}, and we recover the definitions of the Fourier transform given in Spectral Theories, Chap. II.

2. Fourier transforms of infinitely-differentiable functions

Recall (Chap. III, §3, no. 1, Def. 2) that U(G) denotes the algebra of distributions on G with support contained in {e}\{e\}. The canonical injection of g\mathfrak{g} into U(G) extends to an isomorphism from the enveloping algebra of the Lie algebra g\mathfrak{g} to U(G) (loc. cit., no. 7, Prop. 25); from now on we identify these two algebras by this isomorphism. If ff is an infinitely-differentiable complex function on G and if t∈U(G)t\in U(G), we denote by LtfL_tf and RtfR_tf the functions on G defined by

Ltf(g)=⟨εg∗t,f⟩,Rtf(g)=⟨t∗εg,f⟩ L_tf(g) =\langle \varepsilon_g*t, f\rangle ,R_tf(g) =\langle t*\varepsilon_g, f\rangle

(cf. loc. cit., no. 6). For all g∈Gg\in G,

Lt∘γ(g)=γ(g)∘Lt,Rt∘δ(g)=δ(g)∘Rt L_t\circ \gamma (g) =\gamma (g)\circ L_t,R_t\circ \delta (g) =\delta (g)\circ R_t

Let u∈G^u\in \widehat{G}; denote by EuEu the space of uu. The morphism of Lie groups u:G→GL(Eu)u: G\rightarrow \mathbf{G}\mathbf{L}(E_u) gives by differentiation a homomorphism of (real) Lie algebras g→\mathfrak{g}\rightarrow End(Eu_u), hence a homomorphism of algebras, also denoted by uu, from U(G) to End(Eu_u). If t∈U(G)t\in U(G) and if ff is an infinitely-differentiable function on G, then

u(Ltf)=u(f)u(t∨),u(Rtf)=u(t∨)u(f)(14) u(L_tf) =u(f)u(t^{\vee}), u(R_tf) =u(t^{\vee})u(f) \tag{14}

where t∨t^{\vee} denotes the image of tt under the principal anti-automorphism of U(G) (Chap. I, §2, no. 4); indeed, it suffices to verify this for t∈gt\in \mathfrak{g}, in which case it follows by differentiation from formulas (12) and (13) (cf. Chap. III, §3, no. 7, Prop. 27).

For all u∈G^u\in \widehat{G}, denote by λ(u)\lambda (u) the highest weight of u(u(§7, no. 2, Th. 1), so u→λ(u)u \rightarrow \lambda (u) is a bijective map from G^\widehat{G} to the set X++X_{++} of dominant elements of X(T).

Let Γ∈U(G)\Gamma \in U(G) be a Casimir element of G (§7, no. 6); for all u∈G^u\in \widehat{G}, the endomorphism u(Γ)u(\Gamma ) of EuE_u is a homothety, whose ratio we denote by Γ~(u)\widetilde{\Gamma}(u), so we have a map u→Γ~(u)u \rightarrow \widetilde{\Gamma}(u) from G^\widehat{G} to C\mathbf{C}.

If φ\varphi and ψ\psi are two functions on G^\widehat{G} with positive real values, denote by “φ≼ψ\varphi \preccurlyeq \psi” or “φ(u)≼ψ(u\varphi (u)\preccurlyeq \psi (u)” the relation “there exists M>0M>0 such that φ(u)≤Mψ(u)\varphi (u)\leq M\psi (u) for all u∈G^u\in \widehat{G}”; this is a pre-order relation on the set of functions on G^\widehat{G} with positive real values.

Proposition 2 01H0

Let m→∥m∥m \rightarrow \|m\| be a norm on the R\mathbf{R}-vector space R⊗X(T)\mathbf{R}\otimes X(T) and Γ\Gamma a Casimir element of G. Let φ\varphi be a function on G^\widehat{G} with positive real values.

a) The following conditions are equivalent:

(i) There exists an integer n>0n >0 such that φ(u)≼(∥λ(u)∥+1)n\varphi (u)\preccurlyeq (\|\lambda (u)\|+ 1)^n (resp. for every integer n>0n >0, we have φ(u)≼(∥λ(u)∥+1)−n\varphi (u)\preccurlyeq (\|\lambda (u)\|+ 1)^{-n}).

(ii) There exists an integer n>0n >0 such that φ(u)≼(Γ~(u)+1)n\varphi (u)\preccurlyeq ( \widetilde{\Gamma}(u) + 1)^n (resp. for every integer n>0n >0, we have φ(u)≼(Γ~(u)+1)−n\varphi (u)\preccurlyeq ( \widetilde{\Gamma}(u) + 1)^{-n}).

b) If G is semi-simple, conditions (i) and (ii) are also equivalent to:

(iii) There exists an integer n>0n >0 such that φ(u)≼d(u)n\varphi (u)\preccurlyeq d(u)^n (resp. for every integer n>0n >0, we have φ(u)≼d(u)−n\varphi (u)\preccurlyeq d(u)^{-n}).

Note first of all that condition (i) is clearly independent of the choice of norm. Thus we can use the norm defined by the quadratic form QΓQ_{\Gamma} associated to Γ(\Gamma (§7, no. 6, Prop. 4). Then

0≤Γ~(u)=∥λ(u)+ρ∥2−∥ρ∥2 0\leq \widetilde{\Gamma}(u) =\|\lambda (u) +\rho \|^2- \|\rho \|^2

so Γ~(u)+1≼(∥λ(u)∥+1)2≼Γ~(u)+1\widetilde{\Gamma}(u) + 1\preccurlyeq (\|\lambda (u)\|+ 1)^2\preccurlyeq \widetilde{\Gamma}(u) + 1, hence a).

Further, if G is semi-simple,

∥λ(u)+ρ∥≼d(u)≼∥λ(u)+ρ∥N\|\lambda (u) +\rho \|\preccurlyeq d(u)\preccurlyeq \|\lambda (u) +\rho \|^N, where N=1/N = 1/2(dim G−G- dim T)

(§7, no. 5, Cor. 1 of Th. 3), so ∥λ(u)∥+1≼d(u)≼(∥λ(u)∥+1)N\|\lambda (u)\|+ 1\preccurlyeq d(u)\preccurlyeq (\|\lambda (u)\|+ 1)^N, hence b).

It follows from Prop. 2 that condition (i) is independent of the choice of maximal torus, chamber, and norm, and that condition (ii) is independent of the choice of Casimir element. A function φ\varphi satisfying conditions (i) and (ii) is said to be moderately increasing (resp. rapidly decreasing). The product of two moderately increasing functions is moderately increasing; the product of a moderately increasing function and a rapidly decreasing function is rapidly decreasing. If φ\varphi is rapidly decreasing, the family (φ(u))u∈G^(\varphi (u))_{u\in\widehat{G}} is summable.

Example 01H1

The function u→d(u)u \rightarrow d(u) is moderately increasing (§7, no. 5, Cor. 1 of Th. 3); for any norm ∥∥\| \| on R⊗X(T)\mathbf{R}\otimes X(T), the function u→∥λ(u)∥u \rightarrow \|\lambda (u)\| is moderately increasing. For any Casimir element Γ\Gamma, the function u→Γ~(u)u \rightarrow \widetilde{\Gamma}(u) is moderately increasing; more generally:

Proposition 3 01H2

For all t∈t\in U(G), the functions u→∥u(t)∥∞u \rightarrow \|u(t)\|_{\infty} and u→∥u(t)∥2u \rightarrow \|u(t)\|_2 on G^\widehat{G} are moderately increasing.

Since the product of two moderately increasing functions is moderately increasing, it suffices to prove this when t∈g:t\in \mathfrak{g}: in that case the assertion follows from the Remark in §7, no. 6 and the inequality

∥u(t)∥2≤d(u)∥u(t)∥∞ \|u(t)\|_2\leq d(u)\|u(t)\|_{\infty}

Theorem 1 01H3

a) Let ff be an infinitely-differentiable complex function on G. Then the family (fu)u∈G^(f_u)_{u\in\widehat{G}}, where fu(g)=⟨u(g)∣u(f)⟩f_u(g) =\langle u(g)|u(f)\rangle, is uniformly summable on G and, for all h∈Gh\in G,

f(h)=∑u∈G^⟨u(h)∣u(f)⟩=∑u∈G^d(u)∫Gf(gf(h) =\sum_{u\in\widehat{G}}\langle u(h)|u(f)\rangle =\sum_{u\in\widehat{G}}d(u)\int_Gf(g)Tr(u(gh−1)u(gh^{-1}))dgdg.

b) Let ff be an integrable function on G; then ff is equal almost everywhere to an infinitely-differentiable function if and only if the function u→∥u(f)∥∞u \rightarrow \|u(f)\|_{\infty} is rapidly decreasing on G^\widehat{G}.

Let ff be an infinitely-differentiable function on G, and let Γ\Gamma be a Casimir element for G; by formula (14),

Γ~(u)nu(f)=u(f)u(Γ)n=u((LΓ)nf) \widetilde{\Gamma}(u)^nu(f) =u(f)u(\Gamma )^n=u((L_{\Gamma})^nf)

for all n≥0n\geq 0, and consequently

Γ~(u)n∥u(f)∥∞≤∥(LΓ)nf∥1≤\widetilde{\Gamma}(u)^n\|u(f)\|_{\infty}\leq \|(L_{\Gamma})^nf\|_1\leq supg∈G∣((LΓ)nf)(g)∣_{g\in G}|((L_{\Gamma})^nf)(g)|; (15)

thus, the function u→∥u(f)∥∞u \rightarrow \|u(f)\|_{\infty} is indeed rapidly decreasing.

Conversely, let A=(Au)u∈G^A = (A_u)_{u\in\widehat{G}} be an element of F(G^)F( \widehat{G}) such that the function u→∥Au∥∞u \rightarrow \|A_u\|_{\infty} is rapidly decreasing. Put fu(g)=⟨u(g)∣Au⟩f_u(g) =\langle u(g)|A_u\rangle; the function g→fu(g)g \rightarrow f_u(g) is analytic, hence infinitely-differentiable. By Chap. III, §3, no. 7, Prop. 27,

(Lxfu)(g)=⟨u(g)u(x)∣Au⟩ (L_xf_u)(g) =\langle u(g)u(x)|A_u\rangle

for all x∈gx\in \mathfrak{g}. Let t∈U(G)t\in U(G); by the preceding formula,

(Ltfu)(g)=⟨u(g)u(t)∣Au⟩ (L_tf_u)(g) =\langle u(g)u(t)|A_u\rangle

and consequently

∣(Ltfu)(g)∣=∣⟨u(g)u(t)∣Au⟩≤d(u)2∥u(t)∥∞∥u(g)∥∞∥Au∥∞ |(L_tf_u)(g)|=|\langle u(g)u(t)|A_u\rangle \leq d(u)^2\|u(t)\|_{\infty}\|u(g)\|_{\infty}\|A_u\|_{\infty}
=d(u)2∥u(t)∥∞∥Au∥∞ =d(u)^2\|u(t)\|_{\infty}\|A_u\|_{\infty}

Since d(u)d(u) and ∥u(t)∥∞\|u(t)\|_{\infty} are moderately increasing (Prop. 3) and ∥Au∥∞\|A_u\|_{\infty} is rapidly decreasing, the function u→u \rightarrow supg∣(Ltfu)(g)∣_g|(L_tf_u)(g)| is rapidly decreasing; thus, the family (Ltfu)u∈G^(L_tf_u)_{u\in\widehat{G}} is uniformly summable. It follows7^7 that the sum of the family (fu)(f_u) is an infinitely-differentiable function on G, whose Fourier cotransform is (Au)(A_u), hence the theorem.

Denote by S(G^)\mathscr{S}( \widehat{G}) the vector subspace of L2(G^)L^2( \widehat{G}) consisting of the families A=(Au)u∈G^A = (A_u)_{u\in\widehat{G}} such that the function u→∥Au∥∞u \rightarrow \|A_u\|_{\infty} is rapidly decreasing on G^\widehat{G}. It follows from the theorem that the maps F:f→(u(f))u∈G^\mathscr{F}:f \rightarrow (u(f))_{u\in\widehat{G}} and F:A→∑u∈G^⟨u(g)∣Au⟩\mathscr{F}: A \rightarrow \sum_{u\in\widehat{G}}\langle u(g)|A_u\rangle induce inverse isomorphisms between the complex vector spaces C∞(G;C)\mathscr{C}^{\infty}(G;\mathbf{C}) and S(G^)\mathscr{S}( \widehat{G}). Give the space C∞(G;C)\mathscr{C}^{\infty}(G;\mathbf{C}) the topology of uniform C∞C^{\infty}-convergence (§6, no. 4) which can be defined by the family of semi-norms f→f \rightarrow supg∈G∣Ltf(g)∣_{g\in G}|L_tf(g)| for t∈U(G)t\in U(G), and the space S(G^)\mathscr{S}( \widehat{G}) the topology defined by the sequence of semi-norms pn:A→p_n: A \rightarrow supu∈G^(Γ~(u)+1)n∥Au∥∞_{u\in\widehat{G}}( \widetilde{\Gamma}(u) + 1)^n\|A_u\|_{\infty}. Formula (15) of the preceding proof shows that F‾\overline{\mathscr{F}} is continuous. Let t∈U(G)t\in U(G), and let A=(Au)u∈G^A = (A_u)_{u\in\widehat{G}} be an element of S(G^)\mathscr{S}( \widehat{G}); put fn(g)=⟨u(g)∣Au⟩f_n(g) =\langle u(g)|A_u\rangle. Let pp be an integer such that ∑u∈G^Γ~(u)−p=M<∞\sum_{u\in\widehat{G}}\widetilde{\Gamma}(u)^{-p}= M<\infty. By the preceding proof, there exists a positive integer mm such that, for all g∈Gg\in G,

∣(Ltfu)(g)∣≤d(u)2∥u(t)∥∞∥Au∥∞≤m.(1+Γ~(u))mΓ~(u)−p∥Au∥∞ |(L_tf_u)(g)| \leq d(u)^2\|u(t)\|_{\infty}\|A_u\|_{\infty}\leq m.(1 + \widetilde{\Gamma}(u))^m\widetilde{\Gamma}(u)^{-p}\|A_u\|_{\infty}

so ∣(LtF(A))(g)∣≤mMpm(A)|(L_t\mathscr{F}(A))(g)| \leq mMp_m(A); this proves that F\mathscr{F} is continuous. Consequently:

Corollary 01H4

The maps F‾:f→(u(f))u∈G^\overline{\mathscr{F}}:f \rightarrow (u(f))_{u\in\widehat{G}} and F:A→∑u∈G^⟨u(g)∣Au⟩\mathscr{F}: A \rightarrow \sum_{u\in\widehat{G}}\langle u(g)|A_u\rangle induce inverse isomorphisms between the topological vector spaces C∞(G;C)\mathscr{C}^{\infty}(G;\mathbf{C}) and S(G^)\mathscr{S}( \widehat{G}).

7^7 This follows from the fact that the space C∞(G;C)\mathscr{C}^{\infty}(G;\mathbf{C}), with the topology of uniform

C∞C^{\infty}-convergence (§6, no. 4), is complete.

3. Fourier transforms of central functions

For all u∈G^u\in \widehat{G}, denote by χu\chi_u the character of uu; thus,

χu(g)=\chi_u(g) = Tr(u(g)u(g)),(g∈G),(g\in G). (16) Recall from Spectral Theory the formulas

χu∗χv=0(u,v∈G^,u≠v)(17) \chi_u*\chi_v= 0(u, v\in \widehat{G}, u\not=v) \tag{17}
χu∗χu=1d(u)χu(u∈G^)(18) \chi_u*\chi_u=\frac{1}{d(u)}\chi_u(u\in \widehat{G}) \tag{18}

For all u∈G^u\in \widehat{G}, denote by εu\varepsilon_u the identity map of EuE_u. Recall (§7, no. 4) that ZL2(G)^2(G) denotes the subspace of L2(G)L^2(G) consisting of the classes of the functions ff that are central, that is, such that f∘f\circ Int s=fs=f for all s∈Gs\in G, or equivalently that γ(s)f=δ(s−1)f\gamma (s)f=\delta (s^{-1})f for all s∈Gs\in G.

Proposition 4 01H5

Let f∈L2(G)f\in L^2(G). Then ff is central if and only if u(f)u(f) is a homothety for all u∈G^u\in \widehat{G}. In that case

u(f)=εud(u)∫Gf(g)χu(g)dg(19) u(f) =\frac{\varepsilon_u}{d(u)}\int_Gf(g)\chi_u(g)dg \tag{19}

By Prop. 1 (no. 1), to say that ff is central means that u(γ(s)f)=u(\gamma (s)f) = u(δ(s−1)f)u(\delta (s^{-1})f) for all s∈Gs\in G and all u∈G^u\in \widehat{G}; but this can also be written as u(s)u(f)=u(f)u(s)u(s)u(f) =u(f)u(s) for all s∈Gs\in G and all u∈G^u\in \widehat{G} (formulas (12) and (13)), hence the first assertion of Prop. 4 (Schur’s lemma). If u(f)u(f) is a homothety, then u(f)=λuεuu(f) =\lambda_u\varepsilon_u with

λu=1d(u)\lambda_u=\frac{1}{d(u)} Tr(u(f)u(f)) =1d(u)∫Gf(g=\frac{1}{d(u)}\int_Gf(g)Tr(u(g)u(g))dg=1d(u)∫Gf(g)χu(g)dgdg=\frac{1}{d(u)}\int_Gf(g)\chi_u(g)dg.

Consequently, for f∈f\in ZL2(G)^2(G) we have

u(f)=⟨χ‾u∣f⟩εud(u)⟩(20) u(f) =\langle \overline{\chi}_u|f\rangle \frac{\varepsilon_u}{d(u)}\rangle \tag{20}

so

F‾(f)=(⟨χ‾u∣f⟩εud(u))u∈G^(21) \overline{\mathscr{F}}(f) =(\langle \overline{\chi}_u|f\rangle \frac{\varepsilon_u}{d(u)})_{u\in\widehat{G}} \tag{21}

with

∥F‾(f)∥22=∑∣∣∣∣⟨χ‾u∣f⟩εud(u)∣∣∣∣22=∑∣⟨χ‾u∣f⟩∣2 \|\overline{\mathscr{F}}(f)\|^2_2=\sum||||\langle \overline{\chi}_u|f\rangle \frac{\varepsilon_u}{d(u)}||||^{22}=\sum|\langle \overline{\chi}_u|f\rangle |^2
uuuu

Conversely, if φ\varphi is a square-integrable complex function on G^\widehat{G}, the element (φ(u)d(u)εu)u∈G^(\frac{\varphi(u)}{d(u)}\varepsilon_u)_{u\in\widehat{G}} of F(G^)F( \widehat{G}) belongs to L2(G^)L^2( \widehat{G}), and we have (formula (9)) (Fu(φ(u)d(u)εu))(g)=d(u(\mathscr{F}_u(\frac{\varphi(u)}{d(u)}\varepsilon_u))(g) =d(u)Tr (φ(u)d(u)εuu(g)−1)=φ(u)χ‾u(g)(\frac{\varphi(u)}{d(u)}\varepsilon_uu(g)^{-1})=\varphi (u)\overline{\chi}_u(g),

so

F((φ(u)d(u)εu))=∑φ(u)χ‾u(22) \mathscr{F}((\frac{\varphi(u)}{d(u)}\varepsilon_u))=\sum\varphi (u)\overline{\chi}_u \tag{22}
u∈G^u\in \widehat{G}

Note, in particular, that formulas (20) and (21) give, for u,vu, v in G^,8\widehat{G},^8

u(χ‾v)=0u(\overline{\chi}_v) = 0 if u≠vu\not=v, (23)

u(χ‾u)=εud(u)∈u(\overline{\chi}_u) =\frac{\varepsilon_u}{d(u)}\in End(Eu_u), (24)

F‾(χu)=εud(u)∈\overline{\mathscr{F}}(\chi_u) =\frac{\varepsilon_u}{d(u)}\in End(Eu_u)⊂F(G^)\subset F( \widehat{G}). (25)

Proposition 5 01H6

Let ff be a continuous central function on G. Then ff is infinitely-differentiable if and only if the function u→∣⟨χu∣f⟩∣u \rightarrow |\langle \chi_u|f\rangle | is rapidly decreasing on G^\widehat{G}; in that case,

f(g)=∑u∈G^⟨χu∣f⟩χu(g) f(g) =\sum_{u\in\widehat{G}}\langle \chi_u|f\rangle \chi_u(g)

for all g∈Gg\in G.

By Th. 1b)1b), the function f‾\overline{f} is infinitely-differentiable if and only if the function u→∥u(f‾)∥∞u \rightarrow \|u(\overline{f})\|_{\infty} is rapidly decreasing; but, by (20),

∥u(f‾)∥∞=1d(u)∣⟨χu∣f⟩∣ \|u(\overline{f})\|_{\infty}=\frac{1}{d(u)}|\langle \chi_u|f\rangle |

hence the first assertion, since the functions d(u)d(u) and 1d(u)\frac{1}{d(u)} are moderately increasing.

Assume that ff is infinitely-differentiable; by Th. 1a),f(g)=∑u∈G^fu(g)1a),f(g) =\sum_{u\in\widehat{G}}f_u(g) for all g∈Gg\in G, so

fu(g)=⟨u(g)∣u(f)⟩=d(uf_u(g) =\langle u(g)|u(f)\rangle =d(u)Tr(u(g)−1.u(f)u(g)^{-1}.u(f)) =d(u=d(u)Tr(u(g)−1⟨χ‾u∣f⟩εud(u))(u(g)^{-1}\langle \overline{\chi}_u|f\rangle \frac{\varepsilon_u}{d(u)})

=⟨χ‾u∣f⟩=\langle \overline{\chi}_u|f\rangleTr(u(g)−1u(g)^{-1}) =⟨χ‾u∣f⟩χ‾u(g)=\langle \overline{\chi}_u|f\rangle \overline{\chi}_u(g).

Hence, f(g)=∑u∈G^⟨χ‾u∣f⟩χ‾u(g)f(g) =\sum_{u\in\widehat{G}}\langle \overline{\chi}_u|f\rangle \overline{\chi}_u(g); but, for all u∈G^u\in \widehat{G}, the contragredient representation u′u' of uu satisfies χ‾u=χu′\overline{\chi}_u=\chi_{u'} and the map u→u′u \rightarrow u' is a permutation of G^\widehat{G}; so we also have f(g)=∑u∈G^⟨χu∣f⟩χu(g)f(g) =\sum_{u\in\widehat{G}}\langle \chi_u|f\rangle \chi_u(g), hence the proposition.

8^8 We embed End(Eu_u) in the product F(G^)=∏v∈G^F( \widehat{G}) =\prod_{v\in\widehat{G}} End(Ev_v) by associating to any

A∈A\in End(Eu_u) the family (Av)v∈G^(A_v)_{v\in\widehat{G}} such that Au=AA_u= A and Av=0A_v= 0 for v≠uv\not=u.

Corollary 01H7

Let ff be a continuous central function on G. Then ff is infinitely-differentiable if and only if the restriction of ff to T is infinitely-differentiable.

Indeed, by Cor. 4 of §7, no. 4,

⟨χu∣f⟩=∫Gλ(u)(t)‾φ(t)dt\langle \chi_u|f\rangle =\int_G\overline{\lambda(u)(t)}\varphi (t)dt, where φ(t)=∏α>0(1−α(t)−1)f(t)\varphi (t) =\prod_{\alpha >0}(1-\alpha (t)^{-1})f(t).

If f∣Tf|T is infinitely-differentiable, so is φ\varphi; by Prop. 5, applied to the group T, the function μ→∫Tμ(t)‾φ(t)dt\mu \rightarrow \int_T\overline{\mu(t)}\varphi (t)dt on T^=X(T)\widehat{T} = X(T) is then rapidly decreasing, and so is the function u→⟨χu∣f⟩u \rightarrow \langle \chi_u|f\rangle; hence the function ff is infinitely-differentiable (Prop. 5). The converse is clear.

4. Central functions on g and functions on t

Denote by C(G)\mathscr{C}(G) the space of continuous complex-valued functions on G and by C∞(G)\mathscr{C}^{\infty}(G) the subspace of infinitely-differentiable functions. Then we have a sequence of inclusions

Θ(G)⊂C∞(G)⊂C(G)⊂L2(G) \Theta (G)\subset \mathscr{C}^{\infty}(G)\subset \mathscr{C}(G)\subset L^2(G)

Denote by ZΘ(G),ZC∞(G),ZC(G)Z\Theta (G),Z\mathscr{C}^{\infty}(G),Z\mathscr{C}(G), ZL2^2(G), respectively, the subspaces consisting of the central functions in these various spaces. Introduce similarly the spaces Θ(T),C∞(T),C(T),L2(T)\Theta (T),\mathscr{C}^{\infty}(T),\mathscr{C}(T),L^2(T); for any space E in this list, denote by EWE^W (resp. E−WE^{-W}) the subspace consisting of the invariant (resp. anti-invariant) elements for the operation of the Weyl group W. We have a commutative diagram

ZC(G)⟶acC(T)W Z\mathscr{C}(G)\longrightarrow^{a_c}\mathscr{C}(T)^W
ZC∞(G)⟶a∞C∞(T)W Z\mathscr{C}^{\infty}(G)\longrightarrow^{a_{\infty}}\mathscr{C}^{\infty}(T)^W
ZΘ(G)⟶aΘΘ(T)W Z\Theta (G)\longrightarrow^{a_{\Theta}}\Theta (T)^W

where the vertical arrows represent the canonical injections, and the maps ac,a∞,aΘa_c, a_{\infty}, a_{\Theta} are induced by the restriction map from C(G)\mathscr{C}(G) to C(T)\mathscr{C}(T).

The maps ac,a∞,aΘa_c, a_{\infty}, a_{\Theta} are bijective (§2, no. 5, Cor. 1 of Prop. 5, §8, no. 3, Cor. of Prop. 5, and §7, no. 3, Cor. of Prop. 2).

Assume now that the semi-sum ρ\rho of the positive roots belongs to X(T) and consider the map bb which to every continuous function φ\varphi on T associates φ.J(ρ)\varphi .J(\rho ). We have a commutative diagram

ZL2(G)−−−−−−−−−−−−−−−−−−−−⟶uL2(T)−W^2(G)--------------------\longrightarrow^uL^2(T)^{-W}

ZC(G)⟶acC(T)W⟶bcC(T)−W Z\mathscr{C}(G)\longrightarrow^{a_c}\mathscr{C}(T)^W\longrightarrow^{b_c}\mathscr{C}(T)^{-W}
ZC∞(G)⟶a∞C∞(T)W⟶b∞C∞(T)−W Z\mathscr{C}^{\infty}(G)\longrightarrow^{a_{\infty}}\mathscr{C}^{\infty}(T)^W\longrightarrow^{b_{\infty}}\mathscr{C}^{\infty}(T)^{-W}
ZΘ(G)⟶aΘΘ(T)W⟶bΘΘ(T)−W Z\Theta (G)\longrightarrow^{a_{\Theta}}\Theta (T)^W\longrightarrow^{b_{\Theta}}\Theta (T)^{-W}

where the vertical arrows are the canonical inclusions, the maps bc,b∞,bΘb_c, b_{\infty}, b_{\Theta} are induced by bb, and uu extends bc∘acb_c\circ a_c by continuity (§7, no. 4, Cor. 3 of Th. 2). The maps uu and bΘb_{\Theta} are bijective (loc. cit.); so is b∞b_{\infty} (Exerc. 5); on the other hand, bcb_c is not surjective in general (Exerc. 6).

Exercises

See the exercises for § 8.

Transcribed from the printed volume. Not translated.