Let φ∈S(Rn). The function F(φ) is indefinitely differentiable on Rn and one has
F(Xαφ)=(−2iπ)−∣α∣∂α(F(φ))
F(∂αφ)=(2iπ)∣α∣XαF(φ)
for every α in Nn.
One may suppose that n⩾1. Let φ∈S(Rn). The function defined by (x,y)↦φ(x) exp(2iπx⋅y) from Rn×Rn into C satisfies the hypotheses of Corollary 1 of IV, p. 198 for every integer k. The Fourier transform of φ is therefore indefinitely differentiable and satisfies
∂α(Fφ)(y)=(−2iπ)∣α∣∫Rnxαφ(x) exp(−2iπx⋅y)dμ(x)
for every y∈Rn, which implies the first formula.
Let us prove the second formula. By induction on ∣α∣, it is enough to do so when ∣α∣=1, and one is easily reduced to the case α=(1,0,...,0). Write every x∈Rn in the form x=(x1,x′) with x′∈Rn−1, and denote by μ1 (resp. μ′) the Lebesgue measure on Rn−1 (resp. R). By the Lebesgue-Fubini theorem (INT, V, p. 96, § 8, no4, th. 1), for every y=(y1,y′)∈R×Rn−1, we get
F(∂1φ)(y)=∫ exp(−2iπx′⋅y′)
×Rn(∫R−1(∂1φ)(x1,x′) exp(−2iπx1y1)dμ1(x1))dμ′(x′).
For every compact interval [a,b] in R and every x′∈Rn−1, one has
∫ab(∂1φ)(x1,x′) exp(−2iπx1y1)dμ1(x1)=
[φ(x1,x′) exp(−2iπx1y1)]ab
+2iπy1∫abφ(x1,x′) exp(−2iπx1y1)dμ1(x1) by integration by parts (FVR, II, p. 10, formula (10)). When a tends to −∞ and b tends to +∞, the first term of the second member converges to 0 since φ∈S(Rn). The second term converges, by the Lebesgue theorem (INT, IV, p. 137, § 3, no7, th. 6), to
2iπy1∫Rφ(x1,x′) exp(−2iπx1y1)dμ1(x1),
since the mapping x1↦φ(x1,x′) is integrable on R. As x1↦∂1φ(x1,x′) is also integrable on R, one deduces that
∫R∂1φ(x1,x′) exp(−2iπx1y1)dμ1(x1)
=2iπy1∫Rφ(x1,x′) exp(−2iπx1y1)dμ1(x1) and finally, applying the Lebesgue-Fubini theorem again, one concludes that
F(∂1φ)(y)=2iπy1∫Rnφ(x) exp(−2iπx⋅y)dμ(x),
as desired.
Bourbaki sets the proof directly after the statement, and the corpus does not separate them, so what is above is the statement and its proof together.
Nothing.
Nothing.