Bourbaki

§ 3. Convolution of measures on groups

1. Algebras of measures

Let G be a locally compact group. It will be understood, once and for all, that the measures μ1,…,μn\mu_1, \ldots, \mu_n on G are said to be convolvable if they are so for the mapping

(x1,x2,…,xn)↦x1x2⋯xn; (x_1, x_2, \ldots, x_n) \mapsto x_1 x_2 \cdots x_n;

and it is by means of this mapping that the convolution product ∗iμi*_{i} \mu_i will always be taken. If s∈G,t∈Gs \in G, t \in G, then

(1)

εs∗εt=εst. \varepsilon_s * \varepsilon_t = \varepsilon_{st}.

If s∈Gs \in G and μ∈M(G)\mu \in \mathscr{M}(G), then

(2)

εs∗μ=γ(s)μ \varepsilon_s * \mu = \gamma(s) \mu
(3)
μ∗εs=δ(s−1)μ \mu * \varepsilon_s = \delta(s^{-1}) \mu

by §1, No. 1, Example 3. If G is abelian, to say that μ1\mu_1 and μ2\mu_2 are convolvable is equivalent to saying that μ2\mu_2 and μ1\mu_1 are convolvable, and one then has μ1∗μ2=μ2∗μ1\mu_1 * \mu_2 = \mu_2 * \mu_1. When G is not abelian, it can happen that μ1\mu_1 and μ2\mu_2 are convolvable, without μ2\mu_2 and μ1\mu_1 being so (Exer. 12).

Proposition 1

— Let G be a locally compact group, λ,μ,ν\lambda, \mu, \nu measures ≠0\neq 0 on G.

(i) If λ,μ,ν\lambda, \mu, \nu are convolvable, then so are λ\lambda and μ\mu, ∣λ∣∗∣μ∣|\lambda| * |\mu| and ν\nu, μ\mu and ν\nu, λ\lambda and ∣μ∣∗∣ν∣|\mu| * |\nu|, and one has

λ∗μ∗ν=(λ∗μ)∗ν=λ∗(μ∗ν). \lambda * \mu * \nu = (\lambda * \mu) * \nu = \lambda * (\mu * \nu).

(ii) If λ\lambda and μ\mu are convolvable, as well as ∣λ∣∗∣μ∣|\lambda| * |\mu| and ν\nu, then λ,μ,ν\lambda, \mu, \nu are convolvable. Similarly if μ\mu and ν\nu are convolvable, as well as λ\lambda and ∣μ∣∗∣ν∣|\mu| * |\nu|.

This follows from Prop. 1 of §1, No. 2.

There can exist measures λ,μ,ν\lambda, \mu, \nu on G such that the convolution products λ∗μ\lambda * \mu, (λ∗μ)∗ν(\lambda * \mu) * \nu, μ∗ν\mu * \nu, λ∗(μ∗ν)\lambda * (\mu * \nu) are all defined, and yet (λ∗μ)∗ν≠λ∗(μ∗ν)(\lambda * \mu) * \nu \neq \lambda * (\mu * \nu) (cf. Exer. 4).

Let ρ\rho be a lower semi-continuous finite function >0> 0 on GG such that ρ(st)≤ρ(s)ρ(t)\rho(st) \leq \rho(s)\rho(t) for all s,ts, t in GG. We denote by Mρ(G)\mathscr{M}^\rho(G) the vector space of measures λ\lambda on GG such that ρ\rho is λ\lambda-integrable, and by ∥λ∥ρ\| \lambda \|_\rho (or simply ∥λ∥\| \lambda \|) the norm ∫Gρ(s)d∣λ∣(s)\int_G \rho(s) d|\lambda|(s) on this space. When ρ=1\rho = 1, one recovers the set M1(G)\mathscr{M}^1(G) of bounded measures on GG.

Proposition 2

(i) Any two elements of Mρ(G)\mathscr{M}^\rho(G) are convolvable. (ii) For convolution, and for the norm ∥λ∥\| \lambda \|, Mρ(G)\mathscr{M}^\rho(G) is a complete normed algebra, admitting εe\varepsilon_e as unity element. (iii) C′(G)\mathscr{C}'(G) is a subalgebra of Mρ(G)\mathscr{M}^\rho(G).

Let λ,μ\lambda, \mu be in Mρ(G)\mathscr{M}^\rho(G), and let us show that λ\lambda and μ\mu are convolvable. Let f∈K+(G)f \in \mathscr{K}_+(G). Since ρ\rho is >0> 0 and lower semi-continuous, there exists a constant k>0k > 0 such that f≤kρf \leq k\rho. Then

∫∗f(st)d∣λ∣(s)d∣μ∣(t)≤k∫∗ρ(st)d∣λ∣(s)d∣μ∣(t) \int^* f(st) d|\lambda|(s) d|\mu|(t) \leq k \int^* \rho(st) d|\lambda|(s) d|\mu|(t)
≤k∫∗ρ(s)ρ(t)d∣λ∣(s)d∣μ∣(t) \leq k \int^* \rho(s)\rho(t) d|\lambda|(s) d|\mu|(t)
=k(∫∗ρ(s)d∣λ∣(s))(∫∗ρ(t)d∣μ∣(t)) = k \left( \int^* \rho(s) d|\lambda|(s) \right) \left( \int^* \rho(t) d|\mu|(t) \right)

(Ch. V, §8, No. 3, Cor. 1 of Prop. 8). Therefore (s,t)↦f(st)(s,t) \mapsto f(st) is (λ⊗μ)(\lambda \otimes \mu)-integrable, so that λ\lambda and μ\mu are convolvable. On the other hand, using Ch. V (§1, Prop. 4, §6, Prop. 2, §8, Cor. 1 of Prop. 8) and the fact that (s,t)↦ρ(s)ρ(t)(s,t) \mapsto \rho(s)\rho(t) is lower semi-continuous in G×GG \times G, one has

∫G∗ρ(s)d∣λ∗μ∣(s)=∫G∗ρ(s)d∣λ∗μ∣(s) \int_G^* \rho(s) d|\lambda * \mu|(s) = \int_G^* \rho(s) d|\lambda * \mu|(s)
≤∫G×G∗ρ(st)d∣λ∣(s)d∣μ∣(t)≤∫G×G∗ρ(s)ρ(t)d∣λ∣(s)d∣μ∣(t) \leq \int_{G \times G}^* \rho(st) d|\lambda|(s) d|\mu|(t) \leq \int_{G \times G}^* \rho(s)\rho(t) d|\lambda|(s) d|\mu|(t)
=∫G×G∗ρ(s)ρ(t)d∣λ∣(s)d∣μ∣(t)=∥λ∥⋅∥μ∥. = \int_{G \times G}^* \rho(s)\rho(t) d|\lambda|(s) d|\mu|(t) = \| \lambda \| \cdot \| \mu \|.

One sees that λ∗μ∈Mρ(G)\lambda * \mu \in \mathscr{M}^\rho(G) and that ∥λ∗μ∥≤∥λ∥⋅∥μ∥\| \lambda * \mu \| \leq \| \lambda \| \cdot \| \mu \|. In view of Prop. 1, Mρ(G)\mathscr{M}^\rho(G) is an algebra. The mapping λ↦ρ⋅λ\lambda \mapsto \rho \cdot \lambda is an isometric linear mapping θ\theta of Mρ(G)\mathscr{M}^\rho(G) into M1(G)\mathscr{M}^1(G); if μ∈M1(G)\mu \in \mathscr{M}^1(G) then 1/ρ1/\rho, which is locally bounded and upper semi-continuous, is locally μ\mu-integrable, and ρ\rho is (1/ρ)⋅μ(1/\rho) \cdot \mu-integrable, thus (1/ρ)⋅μ∈Mρ(G)(1/\rho) \cdot \mu \in \mathscr{M}^\rho(G); this proves that θ\theta is surjective; therefore Mρ(G)\mathscr{M}^\rho(G) is a complete normed algebra. Finally, it is clear that εe\varepsilon_e is a unity element for Mρ(G)\mathscr{M}^\rho(G) and that C′(G)\mathscr{C}'(G) is a subalgebra of Mρ(G)\mathscr{M}^\rho(G) (§1, No. 4, Cor. of Prop. 5).

If ρ=1\rho = 1, Prop. 2, (i) and (ii) also follow from §1, Prop. 2.

Proposition 3

Let μ1,…,μn\mu_1, \ldots, \mu_n be measures on G. If all of the μi\mu_i, except at most one, have compact support, then the μi\mu_i are convolvable.

For, let SiS_i be the support of μi\mu_i, and suppose that SiS_i is compact for i≠i0i \neq i_0. Let K be a compact subset of G. The set of (x1,…,xn)∈∏iSi(x_1, \ldots, x_n) \in \prod_i S_i such that x1x2⋯xn∈Kx_1 x_2 \cdots x_n \in K is compact, because the conditions xi∈Six_i \in S_i for all ii, x1x2⋯xn∈Kx_1 x_2 \cdots x_n \in K imply

xi0∈Si0−1−1⋯S1−1KSn−1⋯Si0+1−1. x_{i_0} \in S_{i_0-1}^{-1} \cdots S_1^{-1} K S_n^{-1} \cdots S_{i_0+1}^{-1} .

Therefore the μi\mu_i are convolvable (§1, No. 4, Prop. 4).

Proposition 4

The mapping (λ,μ)↦λ∗μ(\lambda, \mu) \mapsto \lambda * \mu (resp. (λ,μ)↦μ∗λ(\lambda, \mu) \mapsto \mu * \lambda), where λ∈C′(G)\lambda \in \mathscr{C}'(G), μ∈M(G)\mu \in \mathscr{M}(G), defines on M(G)\mathscr{M}(G) the structure of a left (resp. right) module over the algebra C′(G)\mathscr{C}'(G).

This follows from Props. 1 and 3.

Proposition 5

Let λ\lambda be a left (resp. right) Haar measure on G, and μ∈M1(G)\mu \in \mathscr{M}^1(G). Then μ\mu and λ\lambda (resp. λ\lambda and μ\mu) are convolvable, and μ∗λ=μ(1)λ\mu * \lambda = \mu(1)\lambda (resp. λ∗μ=μ(1)λ\lambda * \mu = \mu(1)\lambda).

We can suppose that μ⩾0\mu \geqslant 0. Let f∈K+(G)f \in \mathscr{K}_+(G). When λ\lambda is a left Haar measure,

∫∗dμ(x)∫∗f(xy) dλ(y)=∫∗dμ(x)∫f(y) dλ(y)=λ(f)∥μ∥, \int^* d\mu(x) \int^* f(xy)\, d\lambda(y) = \int^* d\mu(x) \int f(y)\, d\lambda(y) = \lambda(f)\|\mu\| ,

therefore the function (x,y)↦f(xy)(x, y) \mapsto f(xy) is (μ⊗λ)(\mu \otimes \lambda)-integrable, and its integral for μ⊗λ\mu \otimes \lambda is λ(f)∥μ∥\lambda(f)\|\mu\|. One argues similarly when λ\lambda is a right Haar measure.

Proposition 6

Let μ\mu and ν\nu be two convolvable measures on G. Let χ\chi be a continuous representation of G in C∗\mathbf{C}^*. Then χ⋅μ\chi \cdot \mu and χ⋅ν\chi \cdot \nu are convolvable and (χ⋅μ)∗(χ⋅ν)=χ⋅(μ∗ν)(\chi \cdot \mu) * (\chi \cdot \nu) = \chi \cdot (\mu * \nu).

Let f∈K(G)f \in \mathscr{K}(G). Then fχ∈K(G)f\chi \in \mathscr{K}(G), therefore the function

(x,y)↦f(xy)χ(xy)=f(xy)χ(x)χ(y) (x, y) \mapsto f(xy)\chi(xy) = f(xy)\chi(x)\chi(y)

on G×GG \times G is integrable for μ⊗ν\mu \otimes \nu; therefore the function (x,y)↦f(xy)(x, y) \mapsto f(xy) is integrable for (χ⋅μ)⊗(χ⋅ν)(\chi \cdot \mu) \otimes (\chi \cdot \nu); therefore χ⋅μ\chi \cdot \mu and χ⋅ν\chi \cdot \nu are convolvable. Moreover,

⟨χ⋅μ∗χ⋅ν,f⟩=∫f(xy)χ(x)χ(y) dμ(x) dν(y)=∫(fχ)(xy) dμ(x) dν(y)=⟨μ∗ν,χf⟩, \langle \chi \cdot \mu * \chi \cdot \nu, f \rangle = \int f(xy)\chi(x)\chi(y)\, d\mu(x)\, d\nu(y) = \int (f\chi)(xy)\, d\mu(x)\, d\nu(y) = \langle \mu * \nu, \chi f \rangle ,

whence (χ⋅μ)∗(χ⋅ν)=χ⋅(μ∗ν)(\chi \cdot \mu) * (\chi \cdot \nu) = \chi \cdot (\mu * \nu).

Proposition 7

Let G and G' be two locally compact groups, u a continuous representation of G in G', and μ1,…,μn\mu_1, \ldots, \mu_n measures on G, all ≠0\neq 0. Then the following assertions are equivalent: (i) uu is μi\mu_i-proper for all i, and the measures u(∣μi∣)u(|\mu_i|) are convolvable; (ii) the μi\mu_i are convolvable and uu is proper for ∗i(∣μi∣)*_{i}(|\mu_i|). When these conditions are satisfied,

∗iu(μi)=u(∗iμi). *_{i} u(\mu_i) = u(*_{i} \mu_i).

This follows from §1, No. 2, Cor. of Prop. 1.

Corollary

Let G be a locally compact group, μ1,…,μn\mu_1, \ldots, \mu_n measures on G. For the sequence (μi)1≤i≤n(\mu_i)_{1 \leq i \leq n} to be convolvable, it is necessary and sufficient that the sequence (μˇn−i)0≤i≤n−1(\check{\mu}_{n-i})_{0 \leq i \leq n-1} be so, in which case

(μ1∗⋯∗μn)∨=μˇn∗⋯∗μˇ1. (\mu_1 * \cdots * \mu_n)^{\vee} = \check{\mu}_n * \cdots * \check{\mu}_1.

This follows from Prop. 7 on considering the isomorphism x↦x−1x \mapsto x^{-1} of G onto the opposite group.

2. The case of a group operating on a space

Let X be a locally compact space on which a locally compact group G operates on the left continuously by

(s,x)↦s⋅x. (s, x) \mapsto s \cdot x.

If μ1,…,μn\mu_1, \ldots, \mu_n are measures on G and ν\nu is a measure on X, these will be said to be convolvable if they are so for the mapping (s1,…,sn,x)↦s1⋯snx(s_1, \ldots, s_n, x) \mapsto s_1 \cdots s_n x of Gn×XG^n \times X into X, and their convolution product is to be understood in the sense of this mapping.

If s∈Gs \in G and x∈Xx \in X, then

(4)

εs∗εx=εsx. \varepsilon_s * \varepsilon_x = \varepsilon_{sx}.

If s∈Gs \in G and μ∈M(X)\mu \in \mathscr{M}(X), then

(5)

εs∗μ=γ(s)μ \varepsilon_s * \mu = \gamma(s)\mu
by §1, No. 1, Example 3.

Proposition 8

— Let μ\mu be a measure on GG, ν\nu a measure on XX. (i) If μ\mu has compact support, then μ\mu and ν\nu are convolvable. (ii) If ν\nu has compact support, and if GG operates properly in XX, then μ\mu and ν\nu are convolvable. This follows from Prop. 4 of §1, No. 4.

Proposition 9

— For convolution, M1(X)\mathscr{M}^1(X) is a left module over M1(G)\mathscr{M}^1(G), while M(X)\mathscr{M}(X) and C′(X)\mathscr{C}'(X) are left modules over C′(G)\mathscr{C}'(G). This follows from Prop. 8, and from §1, Props. 1, 3 and the Cor. of Prop. 5.

Proposition 10

— Let μ\mu be a measure on GG, ν\nu a measure on XX, μ\mu and ν\nu being convolvable. Suppose in addition that there exists a positive measure β\beta on XX such that γ(s)ν\gamma(s)\nu has base β\beta for every s∈Gs \in G. Then μ∗ν\mu * \nu has base β\beta. Let KK be a β\beta-negligible compact subset of XX. Then KK is γ(s)∣ν∣\gamma(s)|\nu|-negligible for every s∈Gs \in G. Now,

∣μ∣∗∣ν∣=∫G(εs∗∣ν∣)d∣μ∣(s) |\mu| * |\nu| = \int_G (\varepsilon_s * |\nu|) d|\mu|(s)
(§1, No. 5, Prop. 7), and the mapping s↦εs∗∣ν∣s \mapsto \varepsilon_s * |\nu| is vaguely continuous (§2, Prop. 6). Therefore KK is ∣μ∣∗∣ν∣|\mu| * |\nu|-negligible by Ch. V, §3, No. 3, Th. 1. Therefore ∣μ∣∗∣ν∣|\mu| * |\nu| has base β\beta (Ch. V, §5, No. 5, Th. 2).

3. Convolution and linear representations

Proposition 11

— Let GG be a locally compact group, EE a quasi-complete locally convex space, UU a continuous representation of GG in EE. (i) If λ∈C′(G)\lambda \in \mathscr{C}'(G), μ∈C′(G)\mu \in \mathscr{C}'(G), then U(λ∗μ)=U(λ)U(μ)U(\lambda * \mu) = U(\lambda)U(\mu). (ii) Suppose that EE is a Banach space, and let ρ(s)=∥U(s)∥\rho(s) = \|U(s)\| for s∈Gs \in G. If λ∈Mρ(G)\lambda \in \mathscr{M}^\rho(G), μ∈Mρ(G)\mu \in \mathscr{M}^\rho(G), then U(λ∗μ)=U(λ)U(μ)U(\lambda * \mu) = U(\lambda)U(\mu). Let λ,μ\lambda, \mu be in C′(G)\mathscr{C}'(G). For any x∈Ex \in E one has, by applying notably Props. 1 and 4 of Ch. VI, §1, No. 1,

U(λ∗μ)x=∫GU(s)x d(λ∗μ)(s) U(\lambda * \mu)x = \int_G U(s)x\, d(\lambda * \mu)(s)
=∫G×GU(st)x dλ(s) dμ(t)=∫G×GU(s)U(t)x dλ(s) dμ(t) = \int_{G \times G} U(st)x\, d\lambda(s)\, d\mu(t) = \int_{G \times G} U(s)U(t)x\, d\lambda(s)\, d\mu(t)
=∫GU(λ)U(t)x dμ(t)=U(λ)∫GU(t)x dμ(t)=U(λ)U(μ)x, = \int_G U(\lambda)U(t)x\, d\mu(t) = U(\lambda) \int_G U(t)x\, d\mu(t) = U(\lambda)U(\mu)x ,
whence (i). An analogous argument may be applied in case (ii).

With G still a locally compact group, let us assume that G operates continuously on the left in a locally compact space X. This defines (§2, No. 4) a continuous linear representation γ\gamma of G in M(X)\mathscr{M}(X) (equipped with the topology of compact convergence in H(X)\mathscr{H}(X)).

Proposition 12

If λ∈C′(G)\lambda \in \mathscr{C}'(G) and μ∈M(X)\mu \in \mathscr{M}(X), then

γ(λ)μ=λ∗μ. \gamma(\lambda)\mu = \lambda * \mu.

By Prop. 7 of §1, No. 5,

λ∗μ=∫G(εs∗μ)dλ(s). \lambda * \mu = \int_G (\varepsilon_s * \mu) d\lambda(s).

Now, εs∗μ=γ(s)μ\varepsilon_s * \mu = \gamma(s)\mu (No. 2, formula (5)) and

∫G(γ(s)μ)dλ(s)=γ(λ)μ \int_G (\gamma(s)\mu) d\lambda(s) = \gamma(\lambda)\mu

by the definition of γ(λ)\gamma(\lambda).

Corollary

The mapping (λ,μ)↦λ∗μ(\lambda, \mu) \mapsto \lambda * \mu of C′(G)×M(X)\mathscr{C}'(G) \times \mathscr{M}(X) into M(X)\mathscr{M}(X) is hypocontinuous relative to the equicontinuous subsets of C′(G)\mathscr{C}'(G) and the compact subsets of M(X)\mathscr{M}(X) (C′(G)\mathscr{C}'(G) and M(X)\mathscr{M}(X) being equipped with the topology of compact convergence in C(G)\mathscr{C}(G) and H(X)\mathscr{H}(X), respectively).

For, M(X)\mathscr{M}(X), equipped with the topology of compact convergence in H(X)\mathscr{H}(X), is quasi-complete. Therefore the mapping (λ,μ)↦γ(λ)μ(\lambda, \mu) \mapsto \gamma(\lambda)\mu of C′(G)×M(X)\mathscr{C}'(G) \times \mathscr{M}(X) into M(X)\mathscr{M}(X) is hypocontinuous relative to the equicontinuous subsets of C′(G)\mathscr{C}'(G) and the compact subsets of M(X)\mathscr{M}(X) (§2, No. 6). It then suffices to apply Prop. 12.

Remark

— 1) Let λ0∈C′(G)\lambda_0 \in \mathscr{C}'(G). The mapping μ↦λ0∗μ\mu \mapsto \lambda_0 * \mu of M(X)\mathscr{M}(X) into M(X)\mathscr{M}(X) is vaguely continuous. For, let f∈H(X)f \in \mathscr{H}(X). One has

⟨λ0∗μ,f⟩=∫f(sx)dλ0(s)dμ(x)=⟨μ,g⟩, \langle \lambda_0 * \mu, f \rangle = \int f(sx) d\lambda_0(s) d\mu(x) = \langle \mu, g \rangle,

where g(x)=∫f(sx)dλ0(s)g(x) = \int f(sx) d\lambda_0(s). Now, gg is continuous (Ch. VII, §1, No. 1, Lemma 1). On the other hand, let S be the support of λ0\lambda_0 and K that of ff. The conditions sx∈Ksx \in K and s∈Ss \in S imply x∈S−1Kx \in S^{-1}K; therefore the support of gg is contained in S−1KS^{-1}K, so that g∈H(X)g \in \mathscr{H}(X). Then ⟨λ0∗μ,f⟩=⟨μ,g⟩\langle \lambda_0 * \mu, f \rangle = \langle \mu, g \rangle is a vaguely continuous function of μ\mu, which proves our assertion.

2) Let μ0∈M(X)\mu_0 \in \mathscr{M}(X). The mapping λ↦λ∗μ0\lambda \mapsto \lambda * \mu_0 of C′(G)\mathscr{C}'(G) into M(X)\mathscr{M}(X) is continuous for the topologies σ(C′(G),C(G))\sigma(\mathscr{C}'(G), \mathscr{C}(G)) and σ(M(X),H(X))\sigma(\mathscr{M}(X), \mathscr{H}(X)). For, let f∈K(X)f \in \mathscr{K}(X). Setting h(s)=∫f(sx) dμ0(x)h(s) = \int f(sx)\, d\mu_0(x), we have ⟨f,λ∗μ0⟩=⟨h,λ⟩\langle f, \lambda * \mu_0 \rangle = \langle h, \lambda \rangle, and h∈C(G)h \in \mathscr{C}(G) (Ch. VII, §1, No. 1, Lemma 1).

Proposition 13

The mapping* (s,μ)↦γ(s)μ(s, \mu) \mapsto \gamma(s)\mu of G×M+(X)G \times \mathscr{M}_+(X) into M+(X)\mathscr{M}_+(X) is continuous when the set M+(X)\mathscr{M}_+(X) of positive measures on XX is equipped with the vague topology.

Since γ(s)μ=γ(ss0−1)γ(s0)μ\gamma(s)\mu = \gamma(ss_0^{-1})\gamma(s_0)\mu, it follows from Remark 1 that it suffices to prove the continuity of the mapping under consideration at a point of the form (e,μ0)(e, \mu_0) with μ0∈M+(X)\mu_0 \in \mathscr{M}_+(X). Given a function f∈K(X)f \in \mathscr{K}(X) and a number ε>0\varepsilon > 0, it is thus a matter of showing that there exist a neighborhood U of e in G and a neighborhood W of μ0\mu_0 in M+(X)\mathscr{M}_+(X) such that the relations s∈U,μ∈Ws \in U, \mu \in W imply

∣∫f(sx) dμ(x)−∫f(x) dμ0(x)∣≤ε. \left| \int f(sx)\, d\mu(x) - \int f(x)\, d\mu_0(x) \right| \leq \varepsilon .

Let V be a compact neighborhood of the support K of ff in X, and let φ∈K+(X)\varphi \in \mathscr{K}_+(X) be such that φ(x)=1\varphi(x) = 1 on V; there exists a neighborhood W0W_0 of μ0\mu_0 in M+(X)\mathscr{M}_+(X) such that a=sup⁡μ∈W0μ(V)a = \sup_{\mu \in W_0} \mu(V) is finite: it suffices to take for W0W_0 the set of μ∈M+(X)\mu \in \mathscr{M}_+(X) such that ∣⟨φ,μ−μ0⟩∣≤1|\langle \varphi, \mu - \mu_0 \rangle| \leq 1. Since the mapping (s,x)↦sx(s, x) \mapsto sx is continuous, there is, on the other hand, a compact neighborhood U0U_0 of e in G such that sK⊂VsK \subset V for all s∈U0s \in U_0; the function (s,x)↦f(sx)(s, x) \mapsto f(sx) is then uniformly continuous in U0×VU_0 \times V and so there is a neighborhood U⊂U0U \subset U_0 of e such that ∣f(sx)−f(x)∣≤ε/2a|f(sx) - f(x)| \leq \varepsilon / 2a for all s∈Us \in U and x∈Vx \in V. For s∈Us \in U and μ∈W0\mu \in W_0, we therefore have

∣∫f(sx) dμ(x)−∫f(x) dμ(x)∣≤ε/2; \left| \int f(sx)\, d\mu(x) - \int f(x)\, d\mu(x) \right| \leq \varepsilon / 2 ;

if W⊂W0W \subset W_0 is the neighborhood of μ0\mu_0 in M+(X)\mathscr{M}_+(X) formed by the measures μ∈W0\mu \in W_0 such that ∣∫f(x) dμ(x)−∫f(x) dμ0(x)∣≤ε/2\left| \int f(x)\, d\mu(x) - \int f(x)\, d\mu_0(x) \right| \leq \varepsilon / 2, U and W meet the requirements.

Exercises

See the exercises for § 3.

Transcribed from the printed volume. Not translated.