Lecture 18

Paley-Wiener theorem for distribution with compact support

Growth of Fourier-Laplace transform of compactly supported distributions

Let TT be a distribution with compact support, let KK be a compact set containing support of KK. Then

FT(z)=FT(x+iy)=(T(t),e2πt(x+iy))Ce2πRy0lksupxKdldxle2πitxCe2πRy(1+z)k. \begin{split} |\mathcal FT(z)| &= |\mathcal FT(x+iy)|\\ & = |(T(t),e^{-2\pi t(x+iy)})|\\ & \leq C e^{2\pi R|y|}\sum_{0\leq l\leq k}\sup_{x\in K}|\frac{d^{l}}{dx^l}e^{-2\pi i tx}|\\ &\leq C'e^{2\pi R|y|}(1+|z|)^k. \end{split}

where the second last row follows from previous lecture.

The Paley-Wiener theorem for distributions

Theorem. Let UU be an entire function satisfying U(z)C(1+z)ke2πRIm(z)|U(z)|\leq C(1+|z|)^k e^{2\pi R|Im(z)|} for some integer k>0k>0. Then UU is the Fourier-Laplace transform of a distribution TT with compact support on [R,R][-R,R].

Proof. Let Vϵ(z)=U(z)Fjϵ(z)V_\epsilon(z) = U(z)\mathcal Fj_\epsilon(z). Since jϵj_\epsilon is supported on [ϵ,ϵ][-\epsilon,\epsilon], the Paley-Wiener theorem implies that Fjϵ(z)\mathcal F j_\epsilon(z) is an entire function satisfying the growth condition that

Fjϵ(z)Ck(1+z)ke2πRIm(z)|\mathcal Fj_\epsilon(z)|\leq C_k (1+|z|)^{-k}e^{2\pi R|Im(z)|}

for every positive integer kk. Since UU is “moderate increasing”, U(z)C(1+z)me2πRIm(z)|U(z)|\leq C(1+|z|)^m e^{2\pi R|Im(z)|} for some integer m>0m>0. Overall, these growth estimates implies that Vϵ(z)V_\epsilon(z) satisfies the growth condition of Paley-Wiener theorem for R+ϵR+\epsilon, i.e. Vϵ(z)Ck(1+z)ke2π(R+ϵ)Im(z)|V_\epsilon(z)|\leq C_{k'}(1+|z|)^{k'}e^{2\pi (R+\epsilon)|Im(z)|} for every positive integer kk'. The Paley-Wiener theorem implies that Vϵ(z)V_\epsilon(z) is the Fourier-Laplace transform of some smooth function vϵ(x)v_\epsilon(x) supported on [(R+ϵ),R+ϵ][-(R+\epsilon),R+\epsilon] and vϵ(x)=F1Vϵ(x)=F1U(x)F1Fjϵ(x)=F1Ujϵ(x)v_\epsilon(x) = \mathcal F^{-1}V_\epsilon(x) = \mathcal F^{-1}U(x) * \mathcal F^{-1}\mathcal Fj_\epsilon(x) = \mathcal F^{-1}U * j_\epsilon(x). Note that the identity holds pointwise because they are both smooth functions and equal in sense of distribution (identity principle).

We have showed that supp(F1Ujϵ)[(R+ϵ),R+ϵ]\mathrm{supp}(\mathcal F^{-1}U*j_\epsilon)\subset [-(R+\epsilon),R+\epsilon]. The support of convolution is equal to the Minkovski sum of supports the convolution elements, so this should imply suppF1U[R,R]\mathrm{supp}\mathcal F^{-1}U \subset [-R,R]. To be precise, let T=F1UT = \mathcal F^{-1}U, for every φD\varphi\in \mathcal D which is supported outside [R,R][-R,R], then eventually for some ϵ>0\epsilon > 0, suppφ\mathrm {supp}\varphi will be outside [(R+ϵ),R+ϵ][-(R+\epsilon),R+\epsilon] because suppφ\mathrm{supp}\varphi is compact, and this implies (Tjϵ,φ)=0(T*j_\epsilon,\varphi) =0 for small enough ε. In problem set 7 we have showed that TjϵTT*j_\epsilon \to T weakly, which means limϵ0(Tjϵ,φ)(T,φ)\lim_{\epsilon\to 0} (T*j_\epsilon,\varphi)\to (T,\varphi). This implies (T,φ)=0(T,\varphi) =0 and that suppT[R,R]\mathrm{supp} T\subset [-R,R].