0
$\begingroup$

Suppose $f(x)$ is a probability density on $\mathbb{R}$. Let $\varphi(t)=\int e^{itx}f(x)dx$ denote the Fourier transform (characteristic function). It is well-known that if $\int |x|^p f(x)dx<\infty$, then $\varphi^{(p)}(t)= i^p \int x^p e^{itx} f(x) dx $. If in addition $\varphi^{(p)}\in L^1$, then through inverse Fourier transform $$ (ix)^pf(x)=\frac{1}{2\pi} \int e^{-ixt} \varphi^{(p)}(t)dt $$ one has the bound $$|f(x)|\le \frac{1}{2\pi|x|^p} \int |\varphi^{(p)}(t)|dt,$$ and hence one obtains the tail decay rate $|f(x)|=O(|x|^{-p})$ as $|x|\rightarrow\infty$.

I wonder if there is a generic way to obtain an estimate better than $O(|x|^{-p})$ by working with suitable conditions on $\varphi(t)$. I have this question because for "nice" $f$ (e.g., $f$ decays like a power function) to satisfy $\int |x|^p f(x)dx<\infty$, one typically needs $|f(x)|=o(|x|^{-p-1})$ in order to have integrability for large $|x|$. Hence there is a gap between the expected order of decay $o(|x|^{-p-1})$ and the order $O(|x|^{-p})$ obtained from $\varphi$.

Another perspective is that for the inverse Fourier transform, one has by Riemann Lebesgue Lemma that $\int e^{-ixt} \varphi^{(p)}(t)dt$ tends to zero when $|x|$ tends to infinity. So it seems possible to extract extra decay rate under suitable conditions.

$\endgroup$
4
  • $\begingroup$ Isn't the loss coming just from chaining together non-sharp inequalities? You certainly do not need $f = o(|x|^{-p-1})$ for it to have finite moment $\int |x|^p f(x) ~dx < \infty$. And there are also $f\in O(|x|^{-p})$ which do not have Fourier transform with integrable $p$th derivatives. This is to say nothing of changing between $L^\infty$ and $L^1$ scales. It is unclear to me what you are hoping for as an answer here. $\endgroup$ Commented Nov 19, 2021 at 2:12
  • $\begingroup$ Thanks for the comment. Yes I am aware that in general the estimate may not be improved. However like I illustrated for function of power decay, there is a gap. So maybe an alternative question is, how is the power decay precisely reflected by Fourier transform. $\endgroup$
    – Uchiha
    Commented Nov 19, 2021 at 12:13
  • $\begingroup$ One way to relate the decay of $f$ at infinity with regularity of $\varphi$ is provided by the theory of regular variation — if your $f$ is regularly varying at infinity, you may find this interesting. This definitely applies to symmetric densities $f$; I can give some references if you are interested. I never really needed that for non-symmetric $f$. $\endgroup$ Commented Nov 19, 2021 at 17:30
  • 1
    $\begingroup$ The general principle is that decay of the FT corresponds to smoothness of the function and vice versa. $\endgroup$ Commented Nov 19, 2021 at 18:06

0

You must log in to answer this question.

Browse other questions tagged .