Let t∈T(s)(G) with s≤r. Let f:X→F be a function of class C^r with values in a Hausdorff polynormed space (for example F=K). The convolution of t and f, denoted by t∗f, is the function on X with values in F defined by
(t∗f)(x)=⟨t∨∗εx,f⟩. Then
(t∗f)(x)=⟨ϱ(x)∗(t∨),f⟩(no. 3, Proposition 14 (ii))=⟨t∨,f∘ϱ(x)⟩(Diff. & Anal. Man., R, 13.2.3)=⟨t,(f∘ϱ(x))∨⟩(Diff. & Anal. Man., R, 13.2.3). Note that Definition 3 can also be written in the more symmetric form
⟨εx,t∗f⟩=⟨t∨∗εx,f⟩. The function
(g,x)↦f(gx)=(f∘ϱ(x))(g) on
G×X is of class C^r. By Differentiable and Analytic Manifolds, R, 13.4.4, the function
x↦⟨t∨,f∘ϱ(x)⟩ is therefore of class
Cr−s if
s<∞. In other words, if
s<∞,
t∗f is of class
Cr−s.
Clearly t∗f depends linearly on t and f.
Formula (4) implies in particular, for g∈G,
(εg∗f)(x)=f(g−1x) that is
εg∗f=γ(g)f.
Suppose that K=R or C, that G and X are finite-dimensional and that X has a positive measure invariant under G. The definition of εg∗f agrees with that of Integration, Chapter VIII, § 4, no. 1 (cf. formula (2), loc. cit.).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.