4
$\begingroup$

Let the Fourier transform of $f(t)$ be defined as $F(\omega) = \int_{-\infty}^\infty dt f(t) e^{i\omega t}$ for values of $\omega$ where the integral exists. What are the precise conditions on $F(\omega)$ so that $f(t)$ has the large time asymptotics, $t\to\infty$, as

$$f(t) = C \frac{ e^{-at} }{ t^\nu} \left( 1 + O\left( \frac{1}{t^\mu} \right) \right)$$

where $\mu > 0$? Once this type of asymptotics is present, how does one recover $a, C$ and $\nu$ from $F(\omega)$?

Examples:

1.

$F(\omega) = \frac{2\alpha}{\omega^2+\alpha^2}$, with $\alpha>0$, in this case we know that $f(t) = e^{-\alpha|t|}$ So this corresponds to $a = \alpha$, $\nu=0$, $C=1$. In this case the pole of $F(\omega)$ results in the exponential fall-off.

2.

$F(\omega) = \frac{1}{\sqrt{1-\omega^2}} \;\; {\rm if} \;\; -1 < \omega < 1,\;\; {\rm otherwise} \;\; 0$, which gives $f(t) = \pi J_0(t)$ a Bessel function. Asymptotically we have $J_0(t) = \sin(t+\pi/4) \sqrt{\frac{2}{\pi t}} + \ldots$. Now $\nu \neq 0$ and indeed we have a branch cut in $F(\omega)$ at $\omega = \pm 1$.

3.

$F(\omega) = \frac{1}{\sqrt{1+\omega^2}}$, which gives $f(t) = 2K_0(|t|)$, another Bessel function. Asymptotically $f(t) = 2 e^{-t} \sqrt{\frac{2}{\pi t}} + \ldots$, so again we have $\nu = 1/2$ from a branch cut at $\omega = \pm i$.

Basically: what is the precise correspondence between singularities of $F(\omega)$ and the asymptotic form of $f(t)$? Is it true that poles lead to $\nu=0$ and branch cuts are needed for $\nu \neq 0$?

$\endgroup$
3
  • $\begingroup$ I presume this is motivated by a question you deleted --- perhaps you can make contact to provide some motivation. $\endgroup$ Commented Feb 6, 2022 at 19:02
  • $\begingroup$ Correct. The particular form of $F(\omega)$ in that question was not relevant, I'm after a general statement, that's why I deleted that post. The context is basically that I have a very compicated $F(\omega)$ and no way I can do the inverse Fourier transform analytically. But that's okay, I'm only after the large $t$ behavior. So, presumably, one can get $C,a,\nu$ from $F(\omega)$ somehow, without doing the full inverse Fourier transform. I guess if the branch cut of $F(\omega)$ is at $\omega_0$ then $a = - i \omega_0$. I'm pretty sure the general theory is in some (elementary) text book. $\endgroup$ Commented Feb 6, 2022 at 19:22
  • $\begingroup$ The good old book "Introduction to Fourier analysis and generalized functions" by Lighthill has a section 4.3 titled "The asymptotic expression for the Fourier transform of a function with a finite number of singularities" which addressed this question. $\endgroup$
    – Mahdiyar
    Commented Apr 18 at 19:15

0

You must log in to answer this question.