1

Find the Fourier transform of $$\frac{\sin x}{x}$$

I know that question was asked before here, here on M.SE, but those solutions use contour integration. Can I solve it without that? One possible solution may be,

Take

$$ \mathscr F\{f(x)\}=F(\omega)=\int_{-\infty}^\infty \frac{\sin x}{x}e^{-i\omega x}dx $$

If we use Leibniz integral rule,

$$ \begin{align} \frac{d}{d\omega}F(\omega)&=\int_{-\infty}^\infty \frac{\sin x}{x}\frac{\partial}{\partial \omega}e^{-i\omega x}dx\\ &=-i\int_{-\infty}^\infty \sin xe^{-i\omega x}dx\\ &=-i\int_{-\infty}^\infty\frac{e^{ix}-e^{-ix}}{2i}e^{-i\omega x}dx\\ &=-\frac12\int_{-\infty}^\infty(e^{ix}e^{-i\omega x}-e^{-ix}e^{-i\omega x})dx\\ &=-\frac12\underbrace{\int_{-\infty}^\infty e^{-(i\omega-i)x}-e^{-(i+i\omega)x}dx}_I \end{align} $$

But it seems the integral is problematic for their limit. Any help will be appreciated.

  • Do you know $\int_{-\infty}^\infty e^{ikx}dx=2\pi\delta(k)$ for $k\in\Bbb R$? – J.G. Nov 19 '21 at 14:24
  • I am not familiar with that result @J.G. Is there any proof of that? – WhyMeasureTheory Nov 19 '21 at 14:52
  • @WhyMeasureTheory See this https://math.stackexchange.com/questions/2404980/dirac-delta-and-exponential-integral you essentially prove that the LHS behave like the dirac delta in the sense of distributions. – FeedbackLooper Nov 19 '21 at 15:06
  • To know that $\frac{\sin x}{x}$ is even function, $\int_{-\infty}^\infty\frac{\sin ax}{x}dx=sgn(a)\pi,$ and $,\sin x\cos y=\frac{1}{2}(\sin(x+y)+\sin(x-y))$ is enough to find the solution: $$\int_{-\infty}^\infty\frac{\sin x}{x}e^{-i\omega x}dx=\int_{-\infty}^\infty\frac{\sin x\cos(\omega x)}{x}dx$$ $$=\frac{1}{2}\int_{-\infty}^\infty\frac{\sin (x\omega +x)}{x}dx+\frac{1}{2}\int_{-\infty}^\infty\frac{\sin (x-x\omega)}{x}dx$$ $$=\frac{\pi}{2}(sgn(\omega+1)+sgn(1-\omega))=\pi ,(\omega\in [-1;1]),, \text{otherwise zero}$$ – Svyatoslav Nov 20 '21 at 17:53

2 Answers2

3

One "easy" way is to use the convolution theorem. I'll sketch out how I would do it:

First, depending on the definition of the Fourier transform, some will use a multiplicative factor depending if they use angular frequency $\omega$, or $2\pi f$ or even $2\pi k$. Lets stick with $\omega$ and "ignore" multiplicative factors for the moment. You can figure out those yourself at the end.

Now, note that the Fourier transform of $\sin(x)$ is something proportional to $\delta(\omega-1)-\delta(\omega+1)$ where $\delta(\bullet)$ is the Dirac delta. The reason is that $\sin(x)$ is composed by two complex exponentials as: $\sin(x) = \frac{\exp(ix)}{2i}-\frac{\exp(-ix)}{2i}$, and the Fourier transform of a complex exponential $\exp(ix)$ is:

$$ \int_{-\infty}^{\infty}\exp(ix)\exp(-i\omega x)\,\mathrm dx = -\int_{-\infty}^{\infty}\exp(i(\omega-1)x)\,\mathrm dx = 2\pi\delta(\omega-1) $$ similarly for $\exp(-i\omega x)$. More details on the last step can be found here or here for a different approach.

In addition, the Fourier transform of $1/x$ be found as well, resulting in something proportional to $\text{sgn}(\omega)$ where $\text{sgn}(x)$ is the sign function (returns 1 or -1 depending on the sign of the input). For more information on this part, see this.

Now, The Fourier transform of $\sin(x)\frac{1}{x}$ will be the convolution of the Fourier transforms of $\sin(x)$ and $1/x$ by separate. Now, since $\delta(\omega-a)\ast f(\omega) = f(\omega-a)$ for any $a$ and any function $f(\bullet)$, this results in $$ \begin{aligned} \mathcal{F}\left\{\sin(x)\frac{1}{x}\right\} = \mathcal{F}\left\{\sin(x)\right\}\ast\mathcal{F}\left\{\frac{1}{x}\right\} &\propto \left(\delta(\omega-1)-\delta(\omega+1)\right)\ast\left(\text{sgn}(\omega)\right)\\ & =\text{sgn}(\omega-1)-\text{sgn}(\omega+1) \end{aligned} $$

In the following you will find some additional Wolfram Alpha links that walks you through this process:

I hope this helps!

LL 3.14
  • 12,457
  • Thanks for your response @FeedbackLooper. I really like the convolution part. Let me read your answer several times to understand it correctly. I will update you if I have any query. Thanks again. – WhyMeasureTheory Nov 19 '21 at 15:17
  • Nice! I'll be around in case you have some issue. Good luck! – FeedbackLooper Nov 19 '21 at 15:18
  • whoops! there were some typos. I confused $x$ and $t$ since I usually use $t$. I have corrected the answer. Sorry! – FeedbackLooper Nov 19 '21 at 15:26
  • Okay. I think I understand your answer. But when I try to get the Fourier transform of $\sin x$ and $\frac{1}{x}$ by my own, it's again problematic for me. Another thing I notice that the Wolfram Alpha use different version of definition for Fourier transform which it OK. Could you add some help for those two transforms? I think then the answer will be self-contained. – WhyMeasureTheory Nov 19 '21 at 15:27
  • Ok! I added some more details. I didn't wanted to add all explicit steps, since those has been extensively covered in other posts, so I just linked to them. Hope this helps. Regarding the definition WA uses for the FT, you will find that many text books use different definitions, differing in multiplicative factors, or even the sign of the exponent in the $\exp(-i\omega x)$ term. I myself use a different convention than yours, so be careful. – FeedbackLooper Nov 19 '21 at 15:47
  • Thank you very much @FeedbackLooper. I have one last query. In here $$ \mathcal{F}{\operatorname{sgn}(t)}=\lim_{\alpha\to 0}\mathcal{F}{f_{\alpha}(t)}=\frac{2}{i\omega} \underbrace{\implies}{\text{Duality}} \mathcal{F}\left{\frac{1}{\pi t}\right}=-i \operatorname{sgn}(\omega) $$But from what I knew $(f(t)\rightarrow F(\omega)\underbrace{\implies}{\text{Duality}} F(t)\rightarrow 2\pi f(-\omega) )$, hence $$\mathcal{F}{\frac{2}{it}}=2\pi\operatorname{sgn}(-\omega)$$. Another is how to get $\mathcal{F}{\frac{1}{t}}$ from the above? – WhyMeasureTheory Nov 19 '21 at 16:18
  • Please help me with the above part. Then, I will accept the answer and close this thread. Thanks for your effort @FeedbackLooper. – WhyMeasureTheory Nov 19 '21 at 16:21
  • 1
    If you do not know distribution theory, it is better to avoid the above method. Indeed, the Fourier transform of $1/x$ seen as a function is not well defined and one more precisely should use the renormalized version of $1/x$ called the principal value. – LL 3.14 Nov 19 '21 at 16:33
  • You mean that you get $sgn(-\omega)$ instead of $sgn(\omega)$? Perhaps the relation $sgn(-\omega)=-sgn(\omega)$ may be useful since the sgn function is an odd function. – FeedbackLooper Nov 19 '21 at 17:22
  • Regarding the suggestion to avoid this method if the OP is not familiar with distribution theory, note that there is a constraint on not to use contour integration. This is the approach that came into my mind, with the disadvantage that I have to be somewhat handwavy here, but the details are perfectly formalized using distributions and principal values. However, if someone has another approach not using either distributions nor contour integration, please post it, maybe I could learn something new myself. Thanks for the feedback! – FeedbackLooper Nov 19 '21 at 17:26
3

There is a much simpler way by using the Fourier inversion theorem!

Then the result just follows from the fact that $$ {\cal F}^{-1}(\mathbf 1_{[-1,1]})(x) = \frac{1}{2\pi}\int_{-1}^1 e^{i\,\omega\,x}\,\mathrm d \omega = \frac{e^{ix}-e^{-ix}}{2i\pi\,x} = \frac{\sin x}{\pi\,x}. $$ and so $$ {\cal F}\left(\frac{\sin x}{x}\right) = \pi\,{\cal F}\left(\frac{\sin x}{\pi\,x}\right) = \pi\,\mathbf 1_{[-1,1]}(\omega) $$


Let me add details if you are not familiar with the Fourier transform: the Fourier inversion theorem tells that you can find the function from its Fourier transform by writing $$ f(x) = \frac{1}{2\pi} \int_{\Bbb R} \widehat f(\omega) \, e^{i\omega\,x}\,\mathrm d \omega = \frac{1}{2\pi} \cal F\big(\hat f\big)(-x) $$ where $\widehat f = {\cal F}(f)$. Equivalently, it means that the inverse operation to the Fourier transform is given by $$ {\cal F}^{-1}(g)(x) = \frac{1}{2\pi} \int_{\Bbb R} g(\omega) \, e^{i\omega\,x}\,\mathrm d \omega = \frac{1}{2\pi} \cal F(g)(-x) $$

LL 3.14
  • 12,457
  • Wow, that seems a way more less work. From this result, how can I say the Fourier transform of ${\cal F} {\frac{\sin x}{x}}$? @LL3.14 Actually, I am introduced to this topic recently. And thanks for your response $(+1)$ – WhyMeasureTheory Nov 19 '21 at 16:39
  • I added details – LL 3.14 Nov 19 '21 at 16:58
  • I am little bit confused. Like from $$f(x)=\frac{1}{2\pi} \cal F\big(\hat f\big)(-x)$$Isn't $$\cal F\left(\frac{\sin x}{x\pi}\right)(-x)=2\pi\mathbf 1_{[-1,1]}(x)$$But then $$ {\cal F}\left(\frac{\sin x}{x}\right) = \pi,{\cal F}\left(\frac{\sin x}{\pi,x}\right) = 2\pi^2,\mathbf 1_{[-1,1]}(x) $$Maybe it was a simple thing, but I don't know why I can't see that. If possible, then please add some clarity here. Thanks a lot @LL3.14 – WhyMeasureTheory Nov 19 '21 at 17:24
  • Nice approach! I think my answer was too complicated. Hope it helped anyway. – FeedbackLooper Nov 19 '21 at 17:31
  • In the first equation, I wrote ${\cal F}^{-1}$, not ${\cal F}$. So then since ${\cal F}^{-1} g = f \iff g = {\cal F}f$ you get $\mathbf 1_{[-1,1]} = {\cal F}(\frac{\sin x}{πx})$ (i.e. the $1/\pi$ comes from the inversion here: the $1/(2\pi)$ was multiplied by $2$ because $(e^{ix}-e^{-ix})/i = 2\sin(x)$ ...) – LL 3.14 Nov 19 '21 at 18:11