Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Anybody know of a clever way of finding the Fourier transform of $$ F(x) = P(x)^n e^{-\pi x^2}$$ very $n$ is fairly large integer. It is to be used as part of a computation, so I would rather not multiply out the polynomial and then find the Fourier transform of each of the terms individually, hence I thought perhaps there is a way to use that my polynomial has few roots with high multiplicity to get a closed expression where $n$ features in an manageble way - perhaps something akin to the way factorials occur when $P(x)=x^n$. Any thoughts will be welcome.

EDIT: The polynomial $P$ can be assumed to be either linear or quadratic - in case that simplifies.

share|cite|improve this question
What do we know about $P$? – Davide Giraudo Feb 12 '12 at 21:16
Here's a thought: by the convolution theorem, you have $$\mathcal{F}\left\{P(x)^{2n}e^{-\pi x^2}\right\}=\mathcal{F}\left\{P(x)^{n}e^{-\frac{1}{2}\pi x^2}\right\}*\mathcal{F}\left\{P(x)^{n}e^{-\frac{1}{2}\pi x^2}\right\}$$ where * means convolution. If convoluting is more efficient (I'm not sure this is the case) then you can calculate once the expression $$\mathcal{F}\left\{P(x)e^{-\frac{1}{k}\pi x^2}\right\}$$ and then convolving it $\sim\log_2(n)$ times – yohBS Feb 12 '12 at 21:53
Interesting idea about the convolutions - though it will probably still be too slow. I might try it out none the less though. – testcase Feb 13 '12 at 11:04
up vote 0 down vote accepted

Warning: I am probably going to mess up factors of $\pm i$, $\pi$, etc. So take what I say modulo these constants.

We know that if $f(x)$ is sufficiently nice, then integration by parts yields $\mathcal{F}\{x^n f(x)\} = i^n \frac{d^n}{dk^n}\hat{f}(k)$. Since the Fourier transform is linear, if $p(x)$ is a polynomial, then we can express $\mathcal{F}(p(x)f(x))$ as a polynomial in the derivatives of $\hat{f}(k)$. In your case, the function $f(x)$ is a Gaussian, hence has Gaussian Fourier transform, so the question is reduced to:

If $p$ is a polynomial in $i d/dk$, is there a closed form expression for $p^n e^{-k^2}$? Let $h_n(k) = e^{k^2} p^n e^{-k^2}$ Note that $h_n(k)$ is a polynomial in $k$, and we can find an explicit recurrence relation via \begin{align} h_{n+1}(k) &= e^{k^2} p p^n e^{-k^2} \\ &= e^{k^2} p ( e^{-k^2} h_n(k) ) \end{align}

For example, in the linear case we have $p = id/dk + a$, and we obtain \begin{align} h_{n+1}(k) &= e^{k^2} (id/dk + a)(e^{-k^2} h_n(k)) \\ &= e^{k^2}(-2ki e^{-k^2} h_n(k) + i e^{-k^2} h_n'(k) + a e^{-k^2} h_n(k) \\ &= -2ki h_n(k) + i h_n'(k) + ah_n(k) \\ &= (-2ki + a) h_n(k) + i h_n'(k). \end{align} Note that since the $h_n(k)$ are polynomials, all of this can be computed symbolically. Also note that setting $a=0$ recovers (up to constants and conventions) the usual Hermite polynomials. Actually I expect that in this case it should be easy to work out an explicit relation to the Hermite polynomials (by shifting $k$), but I'll leave that to you.

The quadratic case will certainly be much harder as you'll get a more complicated recurrence relation. It might be possible to find a generating function (as in the case of Hermite polynomials) and to use it to find a closed-form expression.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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