Taylor series of $\frac{\sin(x)}{x}$ is given by $1 - \frac{x^2}{6} + \frac{x^4}{120} + ... $ We know that the Fourier Transform of this sinc function is a scaled Rect function. Why is it incorrect to simply take a Fourier Transform of each term of the Taylor Series shown above and sum them up? Taking the definition of FT as $\int_{-\infty}^{\infty} f(t) e^{-i 2 \pi\omega t} dt$,

$$FT(1) = \delta(\omega)$$ $$FT(x^2) = - \frac{\delta^{"}(\omega)}{4\pi^2}$$

And so on.

The sum doesn't resemble a Rect function. I do realize that $\delta(x)$ is not a function in traditional sense etc. So at least I was hoping to use it as an operator under integral sign for doing something like $FT(\frac{\sin(x)}{x} \frac{\sin(x)}{x}) = FT(\frac{\sin(x)}{x}) \star FT(\frac{\sin(x)}{x})$, where $\star$ is convolution.I wanted to use Rect function for the first term on the right hand side of this equation, but for the second term, instead of using another Rect function, I wanted to use the term wise FT of the Taylor series and then sum the infinite set of convolutions. Note, I already know that the FT of this particular squared sinc function is a scaled unit triangle function in the frequency domain. So my question is: why doesn't the term-wise FT give correct result (either directly for sinc or as an operator under integral sign for this squared sinc)?


1 Answer 1


Let $f(x)=\frac{\sin(x)}{x}$. Then, we have for $\phi\in\mathbb{S}$ (i.e., $\phi$ is a Schwartz function)

$$\begin{align} \pi \int_{-1/2\pi}^{1/2\pi}\phi(k)\,dk&=\langle \mathscr{F}\{f\},\phi\rangle\\\\ &=\langle f,\mathscr{F}\{\phi\}\rangle\\\\ &=\sum_{n=0}^\infty \frac{(-1)^n}{(2n+1)!}\int_{-\infty}^\infty x^{2n}\int_{-\infty}^\infty \phi(k) e^{i2\pi kx}\,dk\,dx \\\\ &=\sum_{n=0}^\infty \frac{(-1)^n}{(2n+1)!} \int_{-\infty}^\infty \int_{-\infty}^\infty \phi(k) \frac1{(i2\pi )^{2n}}\frac{d^{2n}}{dk^{2n}}e^{i2\pi k x}\,dk\,dx\\\\ &=\sum_{n=0}^\infty \frac{1}{(2\pi)^{2n}(2n+1)!}\int_{-\infty}^\infty \int_{-\infty}^\infty \phi^{(2n)}(k)e^{i2\pi kx}\,dk\,dx\\\\ &=\sum_{n=0}^\infty \frac{\phi^{(2n)}(0)}{(2\pi)^{2n}(2n+1)!} \end{align}$$

Alternatively, we can expand $\phi$ in a Taylor series to arrive at the same result. Proceeding, we see that

$$\begin{align} \pi \int_{-1/2\pi}^{1/2\pi}\phi(k)\,dk&=\pi \int_{-1/2\pi}^{1/2\pi} \sum_{n=0}^\infty \frac{\phi^{(n)}(0)}{n!} k^n\,dk\\\\ &=2\pi \sum_{n=0}^\infty \frac{\phi^{(2n)}(0)}{(2n)!}\int_0^{1/2\pi} k^{2n}\,dk\\\\ &=2\pi \sum_{n=0}^\infty \frac{\phi^{(2n)}(0)}{(2\pi)^{2n+1}(2n+1)!}\\\\ &=\sum_{n=0}^\infty \frac{\phi^{(2n)}(0)}{(2\pi)^{2n}(2n+1)!} \end{align}$$

Therefore, in distribution we see that

$$\begin{align} \pi \text{rect}(\pi k)&=\mathscr{F}\{f\}(k)\\\\ &= \sum_{n=0}^\infty \frac{\delta^{2n}(k)}{(2\pi)^{2n}(2n+1)!} \end{align}$$

as was to be shown!

  • $\begingroup$ My comment vanished somehow. So trying again. Thanks Mark for the answer. I am happy that this received a response and happier still that you showed the two representations to be equivalent (even though I still can't visualize the RHS). Do you think it will be useful to start with a nasent delta function like $\frac{1}{|b|} e^(\frac{-x^2}{b^2})$ and sum up a few terms including their derivatives and then take the limit b->0? Will it start to look a bit like rect function or that is a hopeless direction as those dirac delta on the RHS will never lend to any visualization? $\endgroup$
    – Srini
    Mar 29, 2023 at 19:01
  • $\begingroup$ My reply vanished too. Yes, you could certainly proceed as you suggest. That would require your selecting a value for $b$, taking derivatives of even order, and evaluating the series at selected points within and outside the interval $(-1/2\pi,1/2\pi)$. That sounds like a lot of work. $\endgroup$
    – Mark Viola
    Mar 29, 2023 at 20:17
  • $\begingroup$ I am running into a wall with this approach of using Taylor expansion. Please help. If I want to find the value of $\int_{-\infty}^{\infty} cos^2(2 \pi x) dx$, I know it doesn't converge and is reflected by evaluating its fourier transform ($(\delta(y-2)+2\delta(y)+\delta(y+2))/4$ at $y=0$ resulting in a $\delta(0)$. But if I treat the time domain as $cos(2 \pi x) cos(2 \pi x)$ and try to expand one of the $cos(2 \pi x)$ as Taylor series and then take their respective FT and convolve them in frequency domain and then substitute $y=0$, all the terms result in $0$. $\endgroup$
    – Srini
    Apr 17, 2023 at 16:10

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .