Let f∈L1(G) be such that the restriction to H⊥ of the continuous function FG(f) is integrable with respect to γ. Then, for almost every x∈G, the function y↦f(xy) on H is α-integrable, and one has:
∫Hf(xy)dα(y)=∫H⊥⟨χ,x⟩FG(f)(χ)dγ(χ)
From the above, the function f♭ defined almost everywhere on G/H by
f♭(x˙)=∫Hf(xy)dα(y)
belongs to L1(G/H). The Fourier transform of f♭ is identified with the function on H⊥=G/H given for χ∈H⊥ by
FG/H(f♭)(χ)=∫G/H⟨χ,x˙⟩f♭(x˙)dγ(x˙)
=∫G⟨χ,x⟩f(x)dβ(x)=FG(f)(χ)
by formula (33), applied to the integrable function x↦⟨χ,x⟩f(x). By hypothesis, the function F(f)∣H⊥=FG/H(f♭) belong to L1(H⊥), and therefore the function f♭ belong to the space B(G/H). It follows (Theorem 3 of II, p. 222) that f♭ coincides almost everywhere with FG/H(FG/H(f♭)). For almost all x˙∈G/H, one therefore has