I am trying to find a continuous compactly supported function $f$ such that the Fourier transform $f^{ft}$ and derivative $(f^{ft})'$ of the $f^{ft}$ decay faster than exponential rates, that is $$|f^{ft}(t)|=O(e^{-k\left|t\right|^{\gamma}}),\, \left|(f^{ft})'(t)\right|=O(e^{-k\left|t\right|^{\gamma}}).$$ So far I have not found such a function. Can someone help me? Thank so much.

  • $\begingroup$ Doesn't this run into problems in view of the following form of the uncertainty principle? en.wikipedia.org/wiki/Uncertainty_principle#Harmonic_analysis $\endgroup$ – Yemon Choi Aug 23 '14 at 12:42
  • $\begingroup$ Note than by the Schwartz's Paley–Wiener theorem such a function should be an entire function of finite order, i.e. growing not faster than $e^{a |z|}$ for some $a>0$. Roughly speaking it is a sum of exponents. So may be a related question would be: can such $f$ decay faster than exponent on a line? $\endgroup$ – Andrew Aug 23 '14 at 13:04

There are no such functions if $\gamma>1$. Even more is true:

Suppose $f\in L^2(\mathbb R)$ is supported in a set $S$ of finite measure and assume its Fourier transform $g=\hat f$ satisfies $|g(x)|\leq Ae^{-k|x|^\gamma}$ for $k>0$ and $\gamma>1$. Denote $I_m=[-m,m]$.

By Benedick's theorem there is a constant $C>0$ so that $$ \|f\|_{L^2(\mathbb R)} \leq Ce^{2Cm|S|}\|g\|_{L^2(\mathbb R\setminus I_m)}. $$ If $m\geq\gamma^{-1/(\gamma-1)}$, then $x\geq m$ implies $x^\gamma-x\geq m^\gamma-m$. Therefore $$ \|g\|_{L^2(\mathbb R\setminus I_m)}^2 = \sum_\pm\int_m^\infty|g(\pm x)|^2dx \leq \sum_\pm\int_m^\infty A^2e^{-2kx^\gamma} dx \leq 2A^2e^{-2km^\gamma}\int_m^\infty e^{-2k(x-m)} dx = 2A^2e^{-2km^\gamma}(2k)^{-1}. $$ Inserting this to the previous estimate gives $$ \|f\|_{L^2(\mathbb R)} \leq Ck^{-1/2}Ae^{2Cm|S|-km^\gamma}. $$ Since $\gamma>1$, taking the limit $m\to\infty$ gives $\|f\|_{L^2(\mathbb R)}=0$ so $f=g=0$.

  • 1
    $\begingroup$ I think there is no such function for $\gamma\geq 1$, not only for $\gamma>1$ as it is proven above. Indeed the inverse Fourier transform of $g$ defines an entire function for $\gamma>1$, and complex analytic function in the strip $|Im(x)|<k$ for $\gamma=1$. In either case $f$ must be real analytic on $\mathbb{R}$, and hence cannot have compact support. However my argument does not work for $\gamma<1$. $\endgroup$ – makt Aug 23 '14 at 13:49
  • $\begingroup$ Your proof is very nice, and I understand well. Thank you so much. In the proof, you used the assumption $\gamma>1$. What happen if $0<\gamma<1$ ? $\endgroup$ – Cao Aug 23 '14 at 13:56
  • $\begingroup$ @Cao, I don't know what happens if $0<\gamma<1$. My approach only really works when $\gamma>1$, but on the other hand I only need to assume that $f$ has a support of finite measure and I need no bound on $g'$. It would be very convenient if there were bounds on $g$ and $g'$ in the whole complex plane instead of only on the real axis. (Also: Welcome to MO! Have you taken the intro tour that you can find under the help button yet?) $\endgroup$ – Joonas Ilmavirta Aug 23 '14 at 14:04
  • 3
    $\begingroup$ Thank you. In case of $0<\gamma<1$, I have just found a example on a continous compactly supported function that its Fourier transform decays faster than exponential rate. This function is the standard bump function, $\Phi \left( x \right) = {e^{ - \frac{1}{{1 - {x^2}}}}}{\chi _{\left( { - 1,1} \right)}}\left( x \right)$ having $\left| {{\Phi ^{\operatorname{ft} }}\left( t \right)} \right| = O\left( {{{\left| t \right|}^{ - \frac{3}{4}}}{e^{ - \sqrt {\left| t \right|} }}} \right)$ as $\left|t\right|\to\infty$ (\gamma=\frac{1}{2}). See in en.wikipedia.org/wiki/Bump_function $\endgroup$ – Cao Aug 23 '14 at 14:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.