Let $\mathcal F$ be the Fourier transform. I would like to understand whether being in a Sobolev space implies that the Fourier transform of a function is necessarily supported on a compact ball up to some controlled error:

The Sobolev space $W^{k,2}(\mathbb R)$ with $k>0$ is defined by all the $L^2$ functions $f$ such that $ \int_{\mathbb R} (1+\left\lvert x \right\rvert^2)^{k} \left\lvert \mathcal F(f)(x) \right\rvert^2 \ dx< \infty.$

Let $C_1\le C_2$ be given. Consider all functions with $\left\lVert f \right\rVert_{L^2}=C_1$ and $ \int_{\mathbb R} (1+\left\lvert x \right\rvert^2)^{k} \left\lvert \mathcal F(f)(x) \right\rvert^2 \ dx\le C_2.$

Let $\varepsilon>0. $For which $k$ does there exist an $R(C_2,C_1,\varepsilon)>0$ such that

$\left\lVert \mathcal F(f) 1_{B(0,R)^C} \right\rVert< \varepsilon?$


If $k>0.5$ then one can use the Cauchy Schwartz inequality to construct such an $R$. If $C_2=C_1$ then $\left\lvert \mathcal F(f)(x) \right\rvert=0$ almost surely and the statement is true for all $k>0.$

So the only question is: Does there exist for $k \in (0,0.5]$ and $C_2>C_1$ such a radius $R(C_1,C_2,\varepsilon)$?

  • 2
    $\begingroup$ This just follows from straightforward book-keeping: if $\|\chi_{|x|>R}g\|^2\ge \epsilon^2$ (writing $g=\widehat{f}$) and you need to keep $\|x^k g\|$ below a certain bound, then the best you can do is concentrate the whole mass near $x=R$, so you'd get $C_2\ge \epsilon^2 R^{2k}$, and your inequality follows if $R$ is so large that this fails. $\endgroup$ – Christian Remling Jul 9 '18 at 18:49
  • 1
    $\begingroup$ To be more precise, the sharp function $R(C_2, C_1, \epsilon)$ can be found by Christian Remling's argument to be $$ R = \sqrt{\left[1 + \frac{C_2 - C_1}{\epsilon}\right]^{1/k} - 1} $$ $\endgroup$ – Willie Wong Jul 10 '18 at 3:54

No. You may ask the same equivalent question for a function in $L^2$. A function $u$ belongs to $L^2$ means measurability and $$ \int \vert u(x) \vert^2 dx<+\infty, $$ which implies $ \lim_{R\rightarrow +\infty}\int_{\vert x\vert\ge R} \vert u(x) \vert^2 dx=0, $ but the rate of convergence can be arbitrarily slow.

  • $\begingroup$ The OP also assumes that $\int x^2 |u|^2 \le C$, though, and this does give a bound (see my comment above for this). $\endgroup$ – Christian Remling Jul 9 '18 at 19:41

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.