4

I understand that when we have a polynomial a semicircle is used because the poles are included in the contour when doing so but in the following example a rectangle has been used and I am struggling to understand how to come up with this i.e. what the reasoning behind it is.

10.3.2 Computing $\widehat{f}(\xi)$, and hence $\phi(\xi, x)$, by using the CRT:
Next we are going to show that the function $f: \mathbb{R}^{1} \rightarrow \mathbb{R}^{1}, f(x)=\operatorname{sech}(\pi x)$, has Fourier transform $$ \hat{f}(\xi)=\operatorname{sech}(\xi / 2), \quad \xi \in \mathbb{R}^{1} $$ To do that, consider the contour integral $$ \int_{\gamma_{n}} e^{i \xi z} \operatorname{sech}(\pi z) d z \quad \text { with } \xi>0 $$ around the finite rectangle $\gamma_{n}$ in $\mathbb{C}=\{x+i y\}$ bounded by the lines $$ y=0, \quad y=n+1, \quad x=-n-1, \quad x=n+1 $$ with $n$ a positive integer.

K.defaoite
  • 12,536
  • The reasonning behind what ? Do you understand the reasonning when applying the residue theorem to $\displaystyle\int_{-\infty}^\infty \frac{e^{2i \pi \xi x}}{2 i \pi \xi +|a|}d\xi = e^{-|a| x} 1_{x > 0}$ ? – reuns Aug 15 '17 at 17:41

2 Answers2

2

Fourier transform integrals are commonly treated using a rectangular contour, possibly with semicircular bumps to avoid poles on the real axis. The reason for a rectangular contour is that the factor $e^{i\xi z}$ is easily understood on such a contour. If we write $z = x+iy$, then on vertical lines $x = M$, $e^{i\xi z} = e^{i M\xi}e^{-i\xi y}$. Similarly on horizontal lines $y=N>0$ we have $e^{i\xi z} = e^{i\xi x}e^{-\xi N}$.

For reasonable choices of $f$ in the contour integral $\int e^{i\xi z}f(z)~dz$, you can control three of the pieces of the contour easily. On the vertical lines you will want to use decay of $f(x+iy)$ in the horizontal direction, as you let $M \to \pm\infty$. On the horizontal line $y=N>0$, you can use the exponential decay from $e^{-\xi N}$ to help you bound the integral.

Gyu Eun Lee
  • 18,793
  • 2
    Addendum: The reason you often use semicircular contours when integrating rational functions is mostly about the fact that rational functions often decay quickly on the semicircle part of the contour, so you can ignore that contribution as the radius tends to infinity. In general, the way we choose contours has a lot to do with how we expect to show that certain pieces of the contour give vanishing contributions in the limit; this depends on the specific functions we're trying to work with. – Gyu Eun Lee Apr 29 '21 at 01:13
1

It appears this question is nearly four years old, but I'll answer it anyway for future reference. As you know, if we want to evaluate the Fourier transform of the function $f:x\mapsto \operatorname{sech}(\pi x)$ we need to compute the integral $$\mathcal{F}(f)(\xi)=\int_{\mathbb{R}}\operatorname{sech}(\pi x)e^{-2\pi i\xi x}\mathrm{d}x$$ In the complex plane this looks like this: enter image description here To appeal to the Residue Theorem, we think of another rectangular contour $$R(L)=R_1(L)\cup R_2(L)\cup R_3(L)\cup R_4(L)$$ Using $$R_1(L)=\{z\in\mathbb{C}:\operatorname{Im}z=0\text{ and }|z|<L\}$$ $$R_2(L)=\{z\in\mathbb{C}:\operatorname{Re}(z)=L\text{ and }0\leq\operatorname{Im}z\leq L\}$$ $$R_3(L)=\{z\in\mathbb{C}:\operatorname{Im}z=L\text{ and }-L\leq\operatorname{Re}z\leq L\}$$ $$R_4(L)=\{z\in\mathbb{C}:\operatorname{Re}(z)=-L\text{ and }0\leq\operatorname{Im}z\leq L\}$$ Oriented counterclockwise. In the complex plane this looks as such: enter image description here The trick is, if we can show that as we take $L\to\infty$, that the contribution to the integral from the parts of the contour off the real line vanishes, then we can say $$\lim_{L\to\infty}\oint\limits_{R(L)}\operatorname{sech}(\pi z)e^{-2\pi i\xi z}\mathrm{d}z=\int_{\mathbb{R}}\operatorname{sech}(\pi x)e^{-2\pi i\xi x}\mathrm{d}x$$ Once we've done this, we recall that the singularities of $f$ in the upper half plane occur at $$i\left(n+\frac{1}{2}\right)~~~~n\in\mathbb{N}_0$$ So $$\lim_{L\to\infty}\oint\limits_{R(L)}\operatorname{sech}(\pi z)e^{-2\pi i\xi z}\mathrm{d}z=2\pi i\sum_{n=0}^\infty \operatorname{Res}\left(f,i\left(n+\frac{1}{2}\right)\right)$$ I won't detail all of the computations here, but hopefully this is a good outline of the overall method.

K.defaoite
  • 12,536