I am trying to compute the Fourier transform of $f(x) = \frac{1-\cos(tx)}{x^2}$, $(t > 0)$ directly. I tried contour integration, and could not seem to get it to work. So, I am wondering if it can be done in this way, or, even better, if there is a simpler way to do it.
2 Answers
There is another way, if you are OK working with a FT that you might know.
Use the half-angle formula to write
$$\hat{f}(\omega) = \int_{-\infty}^{\infty} dx \frac{1-\cos{t x}}{x^2} e^{i \omega x} = 2 \int_{-\infty}^{\infty} dx \frac{\sin^2{(t x/2)}}{x^2} e^{i \omega x}$$
Substitute $u = t x/2$ to get
$$\hat{f}(\omega) = t \int_{-\infty}^{\infty} du \frac{\sin^2{u}}{u^2} e^{i (2 \omega/t) u}$$
So we are looking at the FT of $\sin^2{u}/u^2$. Let's assume you do not know this, but that you do know that
$$\int_{-\infty}^{\infty} du \frac{\sin{u}}{u} e^{i y u} = \pi \, \text{rect}(y) = \begin{cases}\\ \pi & |y| \lt 1 \\ 0 & |y| > 1 \end{cases}$$
The FT of $\sin^2{u}/u^2$ is evaluated using the convolution theorem. Here, we convolve the function $\text{rect}(y)$ with itself:
$$\int_{-\infty}^{\infty} du \frac{\sin^2{u}}{u^2} e^{i y u} = \frac{\pi}{2} \int_{-\infty}^{\infty} dy' \, \text{rect}(y')\, \text{rect}(y-y')$$
The latter integral is evaluated directly: take the product of the area of two rectangles. One rectangle ($\text{rect}(y')$) is fixed about the origin, the other ($\text{rect}(y-y')$) is centered at $y'=y$. Draw a picture: it should be clear that this product, the convolution, is zero when $|y| \gt 2$. For the rest, just calculate the area of the overlap between the two rectangles. The result is, using the original expression for $\hat{f}(\omega)$ above:
$$\hat{f}(\omega) = \begin{cases}\ \pi (t-|\omega|) & |\omega| \lt t\\ 0 & |\omega| \gt t\end{cases}$$
- 138,521
-
The more I look at this answer (and compare it with my own failed attempt), the more convinced I get that this is the right way. And in case the OP is unhappy with the requirement for the “known” transform used here, let me point out that the inverse transform (i.e., that of the characteristic function of an interval) is trivial to compute, hence the forward transform is easily verified, if not directly computed. – Harald Hanche-Olsen Apr 26 '13 at 21:53
Here is what I suggested, with unsuccessfully as it turned out (see also the comments below): Consider $$\oint_C \frac{1-e^{itz}}{z^2}e^{i\omega z}\,dz=0$$ where the contour $C$ is composed of the real intervals $[-R,-r]$ and $[r,R]$ joined by two semicircles in the upper halfplane centered at $0$ and with radii $r$ and $R$. Provided that $\omega>0$ (which is all you need) you will find that the integral over the large semicircle vanishes in the limit as $R\to\infty$. You need to be more careful about the small semicircle as $r\to0$, but I expect you can fill that in for yourself. (It will not vanish, of course, or else $\hat f$ would vanish.)
Now the integral around the small semicircle works out to be $-\pi t$ in the limit as $r\to0$. Thus we end up with $$\int_{-\infty}^\infty\frac{1-\cos tx}{x^2}\cos\omega x\,dx+\int_{-\infty}^\infty\frac{\sin tx\sin\omega x}{x^2}\,dx=\pi t$$ for $\omega>0$, which is not as useful as I had thought. (The second integral must be considered a principal value.) Oops.
- 31,960
-
-
@RonGordon: Because of symmetry. Since $f$ is real-valued and symmetric, the Fourier transform is also real valued and symmetric. – Harald Hanche-Olsen Apr 26 '13 at 18:46
-
1I agree. But how are you going to exploit that in your contour integral? You can't just take the real part of that integral and be done. Perhaps the PV of the integral is useful, but you have to incorporate the symmetry in a non-trivial way, or just use $\omega < 0$. – Ron Gordon Apr 26 '13 at 18:53
-
@RonGordon: Hmm. I wasn't going to spend time on this one, but now you got me intrigued enough to actually try carrying out the computation. We shall see what snags I hit. – Harald Hanche-Olsen Apr 26 '13 at 19:19
-
BTW I think you will find some way to do it - there must be a way. I just don't see it right away and I am usually pretty good at sorting these things out. – Ron Gordon Apr 26 '13 at 19:32
-
5@RonGordon: I see the snag now. I edited the answer accordingly. As penance, I shall go read your rant on meta. – Harald Hanche-Olsen Apr 26 '13 at 19:56
-