Bourbaki

§ 1. Convolution

1. Definition and examples

Recall (Ch. V, §6, Nos. 1 and 4; Ch. VI, §2, No. 10) that, if X and Y are locally compact spaces, μ\mu a measure on X, and φ\varphi a mapping of X into Y, φ\varphi is said to be μ\mu-proper if: a) φ\varphi is μ\mu-measurable; b) for every compact subset K of Y, φ−1(K)\varphi^{-1}(K) is essentially μ\mu-integrable. Then the image measure ν=φ(μ)\nu = \varphi(\mu) on Y exists and has the following property: for a function f on Y, with values in a Banach space or in R‾\overline{\mathbf{R}}, to be essentially integrable for ν\nu, it is necessary and sufficient that f∘φf \circ \varphi be so for μ\mu, in which case,

∫Yf(y) dν(y)=∫Xf(φ(x))dμ(x). \int_Y f(y)\, d\nu(y) = \int_X f(\varphi(x)) d\mu(x).

Definition 1

Let X1,…,XnX_1, \ldots, X_n be locally compact spaces, μi\mu_i a measure on XiX_i (1≤i≤n1 \leq i \leq n); let X be the product of the XiX_i, μ\mu that of the μi\mu_i. Let φ\varphi be a mapping of X into a locally compact space Y. One says that the sequence (μi)(\mu_i) is φ\varphi-convolvable, or that μ1,…,μn\mu_1, \ldots, \mu_n are φ\varphi-convolvable, if φ\varphi is μ\mu-proper; in this case, the image ν=φ(μ)\nu = \varphi(\mu) of μ\mu under φ\varphi is called the convolution product of the μi\mu_i for φ\varphi, and is denoted ∗φ(μi)1≤i≤n*_{\varphi}(\mu_i)_{1 \leq i \leq n}, or ∗i=1nμi*_{i=1}^n \mu_i, or μ1∗μ2∗⋯∗μn\mu_1 * \mu_2 * \cdots * \mu_n.

The last two notations are of course used only when there can be no doubt as to φ\varphi.

Let f be a function on Y, with values in a Banach space or in R‾\overline{\mathbf{R}}. In order that f be essentially integrable for μ1∗⋯∗μn\mu_1 * \cdots * \mu_n, it is necessary and sufficient that the function

(x1,…,xn)↦f(φ(x1,…,xn)) (x_1, \ldots, x_n) \mapsto f(\varphi(x_1, \ldots, x_n))

be essentially integrable for μ1⊗μ2⊗⋯⊗μn\mu_1 \otimes \mu_2 \otimes \cdots \otimes \mu_n, in which case

(1)∫f d(μ1∗⋯∗μn)=∫f(φ(x1,…,xn))dμ1(x1)…dμn(xn), (1) \quad \int f\ d(\mu_1 * \cdots * \mu_n) = \int f(\varphi(x_1, \ldots, x_n)) d\mu_1(x_1) \ldots d\mu_n(x_n),

a formula that may be regarded as defining μ1∗⋯∗μn\mu_1 * \cdots * \mu_n when one takes f∈K(Y)f \in \mathscr{K}(Y).

The definitions imply at once that the μi\mu_i are convolvable if and only if the ∣μi∣|\mu_i| are. When this is the case,

∣φ(μ1⊗⋯⊗μn)∣≤φ(∣μ1⊗⋯⊗μn∣)=φ(∣μ1∣⊗⋯⊗∣μn∣) |\varphi(\mu_1 \otimes \cdots \otimes \mu_n)| \leq \varphi(|\mu_1 \otimes \cdots \otimes \mu_n|) = \varphi(|\mu_1| \otimes \cdots \otimes |\mu_n|)
(Ch. VI, §2, No. 10), that is,

(2)∣∗iμi∣≤∗i∣μi∣. (2) \quad |*_i \mu_i| \leq *_i |\mu_i|.

If the μi\mu_i are convolvable and positive, and if νi\nu_i is a measure on XiX_i such that 0≤νi≤μi0 \leq \nu_i \leq \mu_i, then the νi\nu_i are convolvable and

∗iνi≤∗iμi. *_i \nu_i \leq *_i \mu_i.

Suppose μ1,μ2,…,μn\mu_1, \mu_2, \ldots, \mu_n are convolvable, and that μ1′,μ2,…,μn\mu'_1, \mu_2, \ldots, \mu_n are convolvable (μ1′\mu'_1 being a measure on X1X_1). By Ch. V, §6, No. 3, Cor. 1 of Prop. 6, μ1+μ1′,μ2,…,μn\mu_1 + \mu'_1, \mu_2, \ldots, \mu_n are convolvable and

(μ1+μ1′)∗μ2∗⋯∗μn=μ1∗μ2∗⋯∗μn+μ1′∗μ2∗⋯∗μn. (\mu_1 + \mu'_1) * \mu_2 * \cdots * \mu_n = \mu_1 * \mu_2 * \cdots * \mu_n + \mu'_1 * \mu_2 * \cdots * \mu_n.

Example

— 1) For any φ\varphi, the measures εxi\varepsilon_{x_i}, where xi∈Xix_i \in X_i for 1≤i≤n1 \leq i \leq n, are always convolvable and have convolution product εy\varepsilon_y, with y=φ(x1,x2,…,xn)y = \varphi(x_1, x_2, \ldots, x_n). Consequently, if each of the μi\mu_i has finite support, then the μi\mu_i are convolvable and μ1∗⋯∗μn\mu_1 * \cdots * \mu_n has finite support. In particular, let MM be a monoid1^1 equipped with a locally compact topology; if one takes φ\varphi to be the law of composition in MM then the measures on MM with finite support form, for convolution, an algebra that is none other than the algebra of the monoid MM (over R\mathbf{R} or over C\mathbf{C}, according as one considers real or complex measures) (A, III, §2, No. 6).

2) Let MM be a monoid equipped with the discrete topology; assume that for each m∈Mm \in M, there are only finitely many pairs (m′,m′′)∈M×M(m', m'') \in M \times M such that m′m′′=mm'm'' = m; this amounts to saying that the law of composition in MM is a proper mapping of M×MM \times M into MM; the measures on MM then form an algebra for convolution, an algebra that is none other than the *total

(1) Monoïde, in the sense of Exer. 17 of Ch. VII, §1.

algebra of the monoid MM (A, III, §2, No. 10); we note the following two special cases:

a) M=NM = N, the law of composition being addition. To every measure μ\mu on NN, let us associate the formal series

S(μ)=∑n=0∞μ({n})tn S(\mu) = \sum_{n=0}^{\infty} \mu(\{n\}) t^n

in an indeterminate tt. Then S(μ∗μ′)=S(μ)S(μ′)S(\mu * \mu') = S(\mu)S(\mu'). An analogous remark holds for formal series in any number of indeterminates.

∗b)*b) M=N∗M = N^*, the law of composition being multiplication. To every measure μ\mu on N∗N^*, let us associate the formal Dirichlet series

D(μ)=∑n=1∞μ({n})n−s. D(\mu) = \sum_{n=1}^{\infty} \mu(\{n\}) n^{-s}.

Then D(μ∗μ′)=D(μ)D(μ′)D(\mu * \mu') = D(\mu)D(\mu').

3) Let X,Y,ZX, Y, Z be locally compact spaces, φ\varphi a continuous mapping of X×YX \times Y into ZZ. If x∈Xx \in X and μ\mu is a measure on YY, to say that εx\varepsilon_x and μ\mu are φ\varphi-convolvable comes to saying that the mapping φ(x,⋅)\varphi(x, \cdot) of YY into ZZ is μ\mu-proper. One then has εx∗μ=φ(x,⋅)(μ)\varepsilon_x * \mu = \varphi(x, \cdot)(\mu).

2. Associativity

The following lemma completes Prop. 11 of Ch. V, §8, No. 5:²

Lemma 1

For 1≤i≤n1 \leq i \leq n, let Xi,YiX_i, Y_i be two locally compact spaces, μi\mu_i a measure on XiX_i, and φi\varphi_i a continuous mapping of XiX_i into YiY_i. Let X=∏iXiX = \prod_i X_i, Y=∏iYiY = \prod_i Y_i, μ=⨂iμi\mu = \bigotimes_i \mu_i, and φ\varphi the mapping of XX into YY that is the product of the φi\varphi_i. If φ\varphi is μ\mu-proper and μi≠0\mu_i \neq 0 for each ii, then the φi\varphi_i are μi\mu_i-proper and φ(μ)=⨂iφi(μi)\varphi(\mu) = \bigotimes_i \varphi_i(\mu_i).

We can suppose that the μi\mu_i are positive and n=2n = 2. Let f1∈K+(Y1)f_1 \in \mathscr{K}_+(Y_1). Since μ2≠0\mu_2 \neq 0, there exists an f2∈K+(Y2)f_2 \in \mathscr{K}_+(Y_2) such that f2∘φ2f_2 \circ \varphi_2 is not μ2\mu_2-negligible. The function (x1,x2)↦f1(φ1(x1))f2(φ2(x2))(x_1, x_2) \mapsto f_1(\varphi_1(x_1)) f_2(\varphi_2(x_2)) is essentially μ\mu-integrable and continuous, hence μ\mu-integrable. Therefore there exists an x2∈X2x_2 \in X_2 such that f2(φ2(x2))≠0f_2(\varphi_2(x_2)) \neq 0 and such that the function x1↦f1(φ1(x1))f2(φ2(x2))x_1 \mapsto f_1(\varphi_1(x_1)) f_2(\varphi_2(x_2)) is μ1\mu_1-integrable. Therefore f1∘φ1f_1 \circ \varphi_1 is μ1\mu_1-integrable,

²The lemma follows by induction on part b) of the cited Prop. 11, which is the case n=2n = 2; the corresponding result in the first edition of Ch. V (§8, No. 3, Prop. 7) did not include the result of part b).

which proves that φ1\varphi_1 is μ1\mu_1-proper. One argues similarly for φ2\varphi_2. Then φ(μ)=⨂iφi(μi)\varphi(\mu) = \bigotimes_i \varphi_i(\mu_i) by Prop. 11 of Ch. V, §8, No. 5.

The following lemma completes Prop. 4 of Ch. V, §6, No. 3.\footnote{The assertion of this lemma is in fact part b) of the cited Prop. 4, a part that was not included in the first edition of Ch. V.}

Lemma 2

Let T, T', T'' be three locally compact spaces, μ\mu a measure on T, π\pi a μ\mu-measurable mapping of T into T', π′\pi' a continuous mapping of T' into T'', and π′′=π′∘π\pi'' = \pi' \circ \pi. If π′′\pi'' is μ\mu-proper, then π\pi is μ\mu-proper, π′\pi' is π(μ)\pi(\mu)-proper, and π′′(μ)=π′(π(μ))\pi''(\mu) = \pi'(\pi(\mu)).

Let K' be a compact subset of T'. Then K′′=π′(K′)K'' = \pi'(K') is compact, therefore π′′(K′′)\pi''(K'') is essentially μ\mu-integrable, therefore π−1(K′)⊂π′′(K′′)\pi^{-1}(K') \subset \pi''(K'') is essentially μ\mu-integrable, thus π\pi is μ\mu-proper. Then π′\pi' is π(μ)\pi(\mu)-proper and π′′(μ)=π′(π(μ))\pi''(\mu) = \pi'(\pi(\mu)) by Ch. V, §6, No. 3, Prop. 4.

Proposition 1

Let XijX_{ij} (1≤i≤m,1≤j≤ni1 \leq i \leq m, 1 \leq j \leq n_i), YiY_i (1≤i≤m1 \leq i \leq m), and Z be locally compact spaces; for each i, let φi\varphi_i be a mapping of Xi=∏jXijX_i = \prod_j X_{ij} into YiY_i; let φ\varphi be the product of the φi\varphi_i, mapping X=∏iXiX = \prod_i X_i into Y=∏iYiY = \prod_i Y_i; let ψ\psi be a mapping of Y into Z.

(i) *Let μij\mu_{ij} be measures given, respectively, on the XijX_{ij}, such that, for each i, the μij\mu_{ij} (1≤j≤ni1 \leq j \leq n_i) are φi\varphi_i-convolvable, and such that the measures ∗∣μij∣*| \mu_{ij} | are ψ\psi-convolvable; then the μij\mu_{ij}, for 1≤i≤m,1≤j≤ni1 \leq i \leq m, 1 \leq j \leq n_i, are (ψ∘φ)(\psi \circ \varphi)-convolvable and

∗i,jμij=∗i(∗jμij). *_{i,j} \mu_{ij} = *_i \left( *_j \mu_{ij} \right).

(ii) Assume ψ\psi and the φi\varphi_i continuous, and let μij\mu_{ij} be measures ≠0\neq 0 given, respectively, on the XijX_{ij} and (ψ∘φ)(\psi \circ \varphi)-convolvable; then, for each i, the μij\mu_{ij} (1≤j≤ni1 \leq j \leq n_i) are φi\varphi_i-convolvable, the measures ∗j∣μij∣*_j |\mu_{ij}| are ψ\psi-convolvable, and the formula (3) holds.

It suffices to consider the case that all of the measures in question are ≥0\geq 0.

Let us place ourselves under the hypotheses of (i). The mapping φ\varphi is proper for ⨂i,jμij\bigotimes_{i,j} \mu_{ij}, and

φ(⨂i,jμij)=⨂iφi(⨂jμij)=⨂i(∗jμij) \varphi \left( \bigotimes_{i,j} \mu_{ij} \right) = \bigotimes_i \varphi_i \left( \bigotimes_j \mu_{ij} \right) = \bigotimes_i \left( *_j \mu_{ij} \right)

(Ch. V, §8, No. 5, Prop. 11). The mapping ψ∘φ\psi \circ \varphi is proper for ⨂i,jμij\bigotimes_{i,j} \mu_{ij}, and

(ψ∘φ)(⨂i,jμij)=ψ(⨂i(∗jμij))=∗i(∗jμij) (\psi \circ \varphi)\left( \bigotimes_{i,j} \mu_{ij} \right) = \psi\left( \bigotimes_i (*_j \mu_{ij}) \right) = *_i (*_j \mu_{ij})
(Ch. V, §6, Prop. 4). Therefore the μij\mu_{ij} (1≤i≤m, 1≤j≤ni1 \leq i \leq m,\ 1 \leq j \leq n_i) are (ψ∘φ)(\psi \circ \varphi)-convolvable and formula (3) holds.

Let us place ourselves under the hypotheses of (ii). First of all, Lemma 2 proves that φ\varphi is proper for ⨂i,jμij\bigotimes_{i,j} \mu_{ij}. Lemma 1 then proves that for every ii, φi\varphi_i is proper for ⨂jμij\bigotimes_j \mu_{ij}, and that

φ(⨂i,jμij)=⨂i(∗jμij). \varphi\left( \bigotimes_{i,j} \mu_{ij} \right) = \bigotimes_i (*_j \mu_{ij}).
By Lemma 2, ψ\psi is proper for ⨂i(∗jμij)\bigotimes_i (*_j \mu_{ij}). Whence the proposition.

Corollary

— Let Xi,Xi′X_i, X'_i (1≤i≤n1 \leq i \leq n), Y,Y′Y, Y' be locally compact spaces; let φ,φ′\varphi, \varphi' be continuous mappings of X=∏iXiX = \prod_i X_i into YY and of X′=∏iXi′X' = \prod_i X'_i into Y′Y', respectively; let fif_i be continuous mappings of XiX_i into Xi′X'_i (1≤i≤n1 \leq i \leq n) and gg a continuous mapping of YY into Y′Y', such that φ′∘f=g∘φ\varphi' \circ f = g \circ \varphi, ff being the mapping of XX into X′X' that is the product of the fif_i. Let μi\mu_i be measures given respectively on the XiX_i, all ≠0\neq 0. Then the following two assertions are equivalent: (i) fif_i is μi\mu_i-proper for all ii, and the measures fi(∣μi∣)f_i(|\mu_i|) are φ′\varphi'-convolvable; (ii) the μi\mu_i are φ\varphi-convolvable, and gg is proper for ∗φ(∣μi∣)*_{\varphi}(|\mu_i|).

Moreover, when these assertions are verified,

∗φ′(fi(μi))=g(∗φ(μi))=∗g∘φ(μi). *_{\varphi'}(f_i(\mu_i)) = g(*_{\varphi}(\mu_i)) = *_{g \circ \varphi}(\mu_i).

For, let h=φ′∘f=g∘φh = \varphi' \circ f = g \circ \varphi. By Prop. 1, the conditions (i) and (ii) are each equivalent to the following condition: (iii) the μi\mu_i are hh-convolvable. When this is so,

∗φ′(fi(μi))=∗h(μi)=g(∗φ(μi)). *_{\varphi'}(f_i(\mu_i)) = *_h(\mu_i) = g(*_{\varphi}(\mu_i)).

3. The case of bounded measures

Proposition 2

— Let X1,…,Xn,YX_1, \ldots, X_n, Y be locally compact spaces, μi\mu_i a bounded measure on XiX_i (1≤i≤n1 \leq i \leq n), μ\mu the product of the μi\mu_i, φ\varphi a μ\mu-measurable mapping of ∏i=1nXi\prod_{i=1}^n X_i into YY. Then the μi\mu_i are φ\varphi-convolvable and

∥∗i=1nμi∥≤∏i=1n∥μi∥. \left\| *_{i=1}^n \mu_i \right\| \leq \prod_{i=1}^n \| \mu_i \|.
If the μi\mu_i are moreover positive, then
∥∗i=1nμi∥=∏i=1n∥μi∥. \left\| *_{i=1}^n \mu_i \right\| = \prod_{i=1}^n \| \mu_i \|.
For, μi′=∣μi∣\mu'_i = |\mu_i| is bounded and ∥μi′∥=∥μi∥\| \mu'_i \| = \| \mu_i \| (Ch. III, §1, No. 8, Cor. 1 of Prop. 10). One has ∣μ1⊗⋯⊗μn∣=μ1′⊗⋯⊗μn′|\mu_1 \otimes \cdots \otimes \mu_n| = \mu'_1 \otimes \cdots \otimes \mu'_n (Ch. III, §4, Nos. 2, 4), therefore μ1⊗⋯⊗μn\mu_1 \otimes \cdots \otimes \mu_n is bounded and
∥μ1⊗⋯⊗μn∥=∥μ1∥⋯∥μn∥ \| \mu_1 \otimes \cdots \otimes \mu_n \| = \| \mu_1 \| \cdots \| \mu_n \|
(ibid., Prop. 4). Therefore φ\varphi is μ\mu-proper (Ch. V, §6, No. 1, Remark 1), that is, the μi\mu_i are φ\varphi-convolvable. One has ∥∗i=1nμi′∥=∥μ1′⊗⋯⊗μn′∥\| *_{i=1}^n \mu'_i \| = \| \mu'_1 \otimes \cdots \otimes \mu'_n \| (Ch. V, §6, No. 2, Th. 1), consequently ∥∗i=1nμi′∥=∥μ1′∥⋯∥μn′∥\| *_{i=1}^n \mu'_i \| = \| \mu'_1 \| \cdots \| \mu'_n \|. Finally, ∣∗iμi∣≤∗iμi′| *_{i} \mu_i | \leq *_{i} \mu'_i (No. 1, formula (2)), therefore
∥∗iμi∥≤∥∗iμi′∥=∏i=1n∥μi∥. \left\| *_{i} \mu_i \right\| \leq \left\| *_{i} \mu'_i \right\| = \prod_{i=1}^n \| \mu_i \|.

Proposition 3

— Let X1,…,Xn,YX_1, \ldots, X_n, Y be locally compact spaces, φ\varphi a continuous mapping of ∏i=1nXi\prod_{i=1}^n X_i into YY. Then the mapping

(μ,…,μn)↦∗φ(μi) (\mu, \ldots, \mu_n) \mapsto *_{\varphi} (\mu_i)
of ∏i=1nM1(Xi)\prod_{i=1}^n \mathscr{M}^1(X_i) into M1(Y)\mathscr{M}^1(Y) is a continuous multilinear mapping. This follows from Prop. 2 and what has been said in No. 1.

4. Properties concerning supports

Proposition 4

— Let X1,…,Xn,YX_1, \ldots, X_n, Y be locally compact spaces, μi\mu_i a measure on XiX_i (1≤i≤n1 \leq i \leq n), SiS_i its support, and φ\varphi a continuous mapping of ∏iXi\prod_i X_i into YY such that the restriction of φ\varphi to ∏iSi\prod_i S_i is proper. Then the μi\mu_i are φ\varphi-convolvable.

For, let KK be a compact subset of YY. The support of μ=μ1⊗⋯⊗μn\mu = \mu_1 \otimes \cdots \otimes \mu_n is S=∏iSiS = \prod_i S_i (Ch. III, §4, No. 2, Prop. 2). Therefore φ‾(K)∩(∏iXi−S)\overline{\varphi}(K) \cap \left( \prod_i X_i - S \right) is μ\mu-negligible. On the other hand, φ‾(K)∩S\overline{\varphi}(K) \cap S is compact. Therefore φ‾(K)\overline{\varphi}(K) is μ\mu-integrable.

Proposition 5

Let X1,…,Xn,YX_1, \ldots, X_n, Y be locally compact spaces, μi\mu_i a measure on XiX_i (1≤i≤n1 \leq i \leq n), μ\mu the product of the μi\mu_i, φ\varphi a μ\mu-proper mapping of ∏iXi\prod_i X_i into YY, and SiS_i the support of μi\mu_i.

a) *The support of ∗iμi*_{i} \mu_i is contained in the closure of φ(∏iSi)\varphi \left( \prod_i S_i \right).

b) If φ\varphi is continuous and the μi\mu_i are positive, then the support of ∗iμi*_{i} \mu_i is the closure of φ(∏iSi)\varphi \left( \prod_i S_i \right).

Let S=∏iSiS = \prod_i S_i be the support of μ\mu. The support of ∗iμi*_{i} \mu_i is contained in φ(S)‾\overline{\varphi(S)} by Ch. V, §6, No. 2, Cor. 3 of Prop. 2. If φ\varphi is continuous and the μi\mu_i are positive, then the support of ∗iμi*_{i} \mu_i is φ(S)‾\overline{\varphi(S)} (loc. cit., Cor. 4 of Prop. 2).

Corollary

If φ\varphi is continuous and the μi\mu_i have compact support, then the μi\mu_i are convolvable and ∗iμi*_{i} \mu_i has compact support.

5. Vectorial expression of the convolution product

Proposition 6

Let X,Y,ZX, Y, Z be locally compact spaces, φ\varphi a continuous mapping of X×YX \times Y into ZZ, and λ,μ\lambda, \mu measures on X,YX, Y. For λ\lambda and μ\mu to be φ\varphi-convolvable, it is necessary and sufficient that the mapping (x,y)↦εφ(x,y)=εx∗εy(x, y) \mapsto \varepsilon_{\varphi(x,y)} = \varepsilon_x * \varepsilon_y of X×YX \times Y into M(Z)\mathscr{M}(Z) be scalarly (λ⊗μ)(\lambda \otimes \mu)-integrable for the topology σ(M(Z),H(Z))\sigma(\mathscr{M}(Z), \mathscr{H}(Z)), in which case

λ∗μ=∫X×Y(εx∗εy) dλ(x) dμ(y). \lambda * \mu = \int_{X \times Y} (\varepsilon_x * \varepsilon_y) \, d\lambda(x) \, d\mu(y).

To say that λ\lambda and μ\mu are φ\varphi-convolvable signifies that, for every f∈H(Z)f \in \mathscr{H}(Z), f∘φf \circ \varphi is (λ⊗μ)(\lambda \otimes \mu)-integrable, that is, for every f∈H(Z)f \in \mathscr{H}(Z) the function (x,y)↦⟨f,εφ(x,y)⟩(x, y) \mapsto \langle f, \varepsilon_{\varphi(x,y)} \rangle is (λ⊗μ)(\lambda \otimes \mu)-integrable, that is, again, that the mapping (x,y)↦εφ(x,y)(x, y) \mapsto \varepsilon_{\varphi(x,y)} of X×YX \times Y into M(Z)\mathscr{M}(Z) is scalarly (λ⊗μ)(\lambda \otimes \mu)-integrable for σ(M(Z),K(Z))\sigma(\mathscr{M}(Z), \mathscr{K}(Z)). If this is the case, then

⟨λ∗μ,f⟩=∫f(φ(x,y))dλ(x)dμ(y)=∫X×Y⟨εφ(x,y),f⟩dλ(x)dμ(y), \langle \lambda * \mu, f \rangle = \int f(\varphi(x, y)) d\lambda(x) d\mu(y) = \int_{X \times Y} \langle \varepsilon_{\varphi(x,y)}, f \rangle d\lambda(x) d\mu(y),

whence λ∗μ=∫X×Yεφ(x,y)dλ(x)dμ(y)\lambda * \mu = \int_{X \times Y} \varepsilon_{\varphi(x,y)} d\lambda(x) d\mu(y).

Proposition 7

Let X, Y, Z be locally compact spaces, φ\varphi a continuous mapping of X×YX \times Y into Z, and λ,μ\lambda, \mu measures on X, Y. Assume that for every x∈Xx \in X, εx\varepsilon_x and μ\mu are φ\varphi-convolvable. For λ\lambda and μ\mu to be φ\varphi-convolvable, it is necessary and sufficient that the mapping x↦εx∗∣μ∣x \mapsto \varepsilon_x * |\mu| of X into M(Z)\mathscr{M}(Z) be scalarly λ\lambda-integrable for the topology σ(M(Z),K(Z))\sigma(\mathscr{M}(Z), \mathscr{K}(Z)), in which case λ∗μ=∫X(εx∗μ)dλ(x)\lambda * \mu = \int_X (\varepsilon_x * \mu) d\lambda(x).

Suppose that λ\lambda and μ\mu are φ\varphi-convolvable. For every f∈K(Z)f \in \mathscr{K}(Z), f∘φf \circ \varphi is (∣λ∣⊗∣μ∣)(|\lambda| \otimes |\mu|)-integrable, therefore the function x↦∫Yf(φ(x,y))d∣μ∣(y)=⟨f,εx∗∣μ∣⟩x \mapsto \int_Y f(\varphi(x, y)) d|\mu|(y) = \langle f, \varepsilon_x * |\mu| \rangle (which by hypothesis is defined for all x∈Xx \in X) is λ\lambda-integrable; thus x↦εx∗∣μ∣x \mapsto \varepsilon_x * |\mu| is scalarly λ\lambda-integrable for σ(M(Z),K(Z))\sigma(\mathscr{M}(Z), \mathscr{K}(Z)), and

⟨f,λ∗μ⟩=∫Xdλ(x)∫Yf(φ(x,y))dμ(y)=∫X⟨f,εx∗μ⟩dλ(x), \langle f, \lambda * \mu \rangle = \int_X d\lambda(x) \int_Y f(\varphi(x, y)) d\mu(y) = \int_X \langle f, \varepsilon_x * \mu \rangle d\lambda(x),

whence λ∗μ=∫X(εx∗μ)dλ(x)\lambda * \mu = \int_X (\varepsilon_x * \mu) d\lambda(x). Conversely, suppose that the mapping x↦εx∗∣μ∣x \mapsto \varepsilon_x * |\mu| of X into M(Z)\mathscr{M}(Z) is scalarly λ\lambda-integrable for σ(M(Z),K(Z))\sigma(\mathscr{M}(Z), \mathscr{K}(Z)). Let f∈K+(Z)f \in \mathscr{K}_+(Z). Then the function (x,y)↦f(φ(x,y))(x, y) \mapsto f(\varphi(x, y)) is continuous and (Ch. V, §8, No. 3, Prop. 5)

∬∗f(φ(x,y))d∣λ∣(x)d∣μ∣(y)=∫∗d∣λ∣(x)∫∗f(φ(x,y))d∣μ∣(y)=∫∗⟨f,εx∗∣μ∣⟩d∣λ∣(x)<+∞. \iint^* f(\varphi(x, y)) d|\lambda|(x) d|\mu|(y) = \int^* d|\lambda|(x) \int^* f(\varphi(x, y)) d|\mu|(y) = \int^* \langle f, \varepsilon_x * |\mu| \rangle d|\lambda|(x) < +\infty.

Therefore f∘φf \circ \varphi is (λ⊗μ)(\lambda \otimes \mu)-integrable, so that λ\lambda and μ\mu are φ\varphi-convolvable.

Exercises

See the exercises for § 1.

Transcribed from the printed volume. Not translated.