6
$\begingroup$

Suppose that we have a $2\pi$-periodic, integrable function $f: \mathbb{R} \rightarrow \mathbb{R}$, whose Fourier coefficients are known. Parseval's identity tells us that:

$$\displaystyle \frac{1}{2\pi}\int_{-\pi}^{\pi}|f(x)|^{2}dx = \sum_{n = -\infty}^{\infty}|\widehat{f(n)}|^2,$$

where $\widehat{f(n)}$ are the Fourier coefficients of $f$.

Suppose we instead want to replace $f(x)$ with $f(x)^{q}$, say: then it would suffice to determine the Fourier coefficients of the $q$-th power of $f$. Is repeated application of the convolution theorem the usual (or, most efficient) way of finding powers of the Fourier coefficients of functions, where the Fourier coefficients of the original function are already known? Moreover, can Parseval's identity be extended in this way, by replacing $f$ with a power of $f$ instead?

For example, suppose that we are interested in the following integral:

$$\displaystyle \int_{-\pi}^{\pi}|f(x)|^{4} dx.$$

I would like to know if it is valid to say the following:

$$\displaystyle \frac{1}{2\pi}\int_{-\pi}^{\pi}|f(x)|^{4}dx = \frac{1}{2\pi}\int_{-\pi}^{\pi}|(f(x))^{2}|^{2}dx = \sum_{n = -\infty}^{\infty} |\widehat{f(n)^{2}}|^{2} = \sum_{n = -\infty}^{\infty} | (\hat{f} \ast \hat{f})(n)|^{2},$$

where $f \ast g$ denotes the convolution of $f$ and $g$, given by $(f \ast g)(t) := \int_{-\infty}^{\infty} f(\tau)g(t - \tau)d\tau,$ and $\widehat{f \ast g} = \hat{f} \cdot \hat{g}$ is the convolution theorem for the Fourier transforms of $f$ and $g$.

Is this manipulation valid?

$\endgroup$
5
  • 2
    $\begingroup$ Note that technically, the $\hat{f}*\hat{f}$ convolution is the convolution on $\mathbb{Z}$, not the convolution on $\mathbb{R}$. Other than that, sure, the manipulation is valid whenever everything converges: an integrable function need not be square integrable, so that the Fourier transform of $f^q$ need not be well-defined. $\endgroup$ Aug 22, 2016 at 20:48
  • 1
    $\begingroup$ I once asked Fourier transform of exponential of a function and was lead to Fourier transform of function composition. Maybe this answer helps. $\endgroup$
    – bers
    Aug 23, 2016 at 0:14
  • $\begingroup$ @WillieWong Thanks for clearing that up for me. Since the convolution is on $\mathbb{Z}$, does that mean one should use the discrete convolution, rather than the continuous version I defined in the original post? If so, does that not mean that we cannot use the convolution theorem here, since we are taking the discrete convolution of continuous Fourier coefficients? $\endgroup$
    – user363087
    Aug 27, 2016 at 18:03
  • $\begingroup$ @bers Thanks for those links. I particularly liked the answer to your original question (regarding the Fourier transform of the exponential of a function), especially since it provokes some thoughts about perhaps constructing a more general convolution operator to define what could be a "fractional convolution" of $f$, much in the same way the fractional Fourier transform is defined. $\endgroup$
    – user363087
    Aug 27, 2016 at 18:06
  • $\begingroup$ Your comment makes no sense to me. You are looking a a periodic function, so your Fourier coefficients have domain $\mathbb{Z}$; in other words, you are looking at the Fourier series. What do you mean by "continuous Fourier coefficients?" $\endgroup$ Aug 28, 2016 at 15:34

0

You must log in to answer this question.

Browse other questions tagged .