Quick question. Could somebody please explain to me why it is that $$\int_{-\infty}^\infty \frac{ \sin x \sin nx}{x^2} \ dx = \pi$$ for every positive integer $n$? This integral showed up when I was computing a certain normalization constant. I was planning on just labeling it $I_n$ and moving on with my life but then Wolfram Alpha informed me it always equals $\pi$. Thanks!
-
2Here's a fun "not a proof": Replace $n$ by a real parameter $t$, call the integral $f(t)$. For $t>0$ We have $f'(t) "=" \int t \frac{\sin(x) \cos(tx)}{x^2}$, which is $0$ since the integrand is odd (or would be $0$, if it converged and we could legally differentiate under the integral sign). The same argument "proves" the integral is constant if you replace $\infty$ by $1$, which is false. – Kevin P. Costello Feb 07 '12 at 07:35
-
You can use Residue theorem – Feb 08 '12 at 13:43
6 Answers
It depends how rigorous you want to be. For $n=1$ this is a classic integral, that I'll assume you have seen before/can easily find. For $n>1$ we have the following generalization if we let $a>b\geqslant 0$
$$\begin{aligned}2\int_0^{\infty}\frac{\sin(ax)\sin(bx)}{x^2}\;dx &=\int_0^{\infty}\frac{\cos((a-b)x)-\cos((a+b)x)}{x^2}\;dx \\ &= \int_0^{\infty}\int_{a-b}^{a+b}\frac{\sin(xy)}{x}\;dy \;dx\\ &=\int_{a-b}^{a+b}\int_0^{\infty}\frac{\sin(xy)}{x}\;dx \;dy \\ &=\int_{a-b}^{a+b}\frac{\pi}{2}\;dy \\ &=\pi b\end{aligned}$$

- 22,196

- 54,059
-
Thanks for the reply. Unfortunately, I can't recall seeing the $n=1$ case at any point and can't seem to find it anywhere. I like your derivation. It's not quite a proof, but still fairly satisfying. One of the more troublesome things to deal with might be that $\int_0^\infty \left| \frac{\sin xy}{x} \right| \ dx = \infty$ - no? – Mike F Feb 07 '12 at 07:44
-
Could someone edit this by putting $\mathrm dx$ and $\mathrm dy$ in the appropriate places, please? – John Bentin Feb 07 '12 at 07:45
-
I agree with Mike. The third line has to be detailled. It is not clear for me that you have to assume $a>b \geq 0$. – user10676 Feb 07 '12 at 09:55
-
@Mike: You are correct, so that Fubini cannot be directly applied. That said if you write $x^{-2}$ as $\displaystyle \int_0^{\infty}te^{-tx}$ you can make everything rigorous with Tonelli's. – Alex Youcis Feb 07 '12 at 22:27
Another way to see why it should be so is to go to the frequency domain. Let $f_a(x)=\frac{\sin ax}x$ and $a\ge b>0\,$. The Fourier transform of $f_a(x)$ is a step: $$ F[f_a](\xi)=\sqrt{\frac\pi2}\theta(a-|x|), $$ there $\theta$ is the Heaviside step function. By the properties of Fourier transform we have $$ \int_{-\infty}^{\infty}f_a(x)f_b(x)\,dx=F[f_af_b](0)= F[f_a]*F[f_b](0)= $$ $$\int_{-\infty}^{\infty}F[f_a](\xi)F[f_b](-\xi)\,d\xi= \frac\pi2 \int_{-b}^b d\xi=\pi b. $$

- 11,912
Another approach is integration by parts:
$$ \begin{align} \int_{-\infty}^\infty \frac{ \sin x \sin nx}{x^2} \mathrm dx &= \int_{-\infty}^\infty\frac{\cos x\sin nx+n\sin x\cos nx}x\mathrm dx \\ &= \frac12\int_{-\infty}^\infty\frac{\sin(n+1)x+\sin(n-1)x+n(\sin(n+1)x-\sin(n-1)x)}x\mathrm dx \\ &= \frac12(1+1+n-n)\int_{-\infty}^\infty\frac{\sin x}x\mathrm dx \\ &= \pi\;. \end{align} $$
For $n=1$, the terms with $\sin(n-1)x$ don't occur, but the result is the same.

- 238,052
-
Thanks this is nice! Summary: integrate by parts (differentiating the top); make use of the identity $2 \sin A \cos B = \sin(A+B) + \sin (A-B)$; notice that $\int_{-\infty}^\infty \frac{g(ax)}{x} dx = \int_{-\infty}^\infty \frac{g(x)}{x} dx$ ($g$ suitable, $a>0$); know that $\int_{-\infty}^\infty \frac{\sin x}{x} dx = \pi$. – Mike F May 09 '12 at 04:05
-
1I do find it a tad strange that $\int \frac{sin x}{x}$ keeps showing up in these derivations since that's only a "conditionally convergent" integral and I should think the original integral converges in the absolute sense - since it is bounded at $x=0$ and decays like $1/x^2$... – Mike F May 09 '12 at 04:12
-
Let $n\ge 1$. You can integrate $$f(z) = \frac{\sin(z)e^{inz}}{z^2}$$ around a big half-disc $U_R$ in the upper half-plane. The integral over the circle-part will go to $0$ for $R\to\infty$ (that's where $n\ge 1$ is needed). Therefore
$$\int_{-\infty}^\infty \frac{\sin(x)\sin(nx)}{x^2}\, dx = \lim_{R\to \infty} \mathrm{Im}\left[\oint_{\partial U_R} f(z) \, dz\right] = \mathrm{Im}\left[\pi i \;\mathrm{Res}_{z=0}(f(z))\right] = \pi$$

- 15,908
-
Awesome! This one is my favorite. Now that we can't deal with $e^{i(n+1)z}$ in whole, still, we extract $\sin z$ alone and apply the common trick to it. Thank you. – Bach Apr 29 '19 at 07:25
$\newcommand{\+}{^{\dagger}} \newcommand{\angles}[1]{\left\langle\, #1 \,\right\rangle} \newcommand{\braces}[1]{\left\lbrace\, #1 \,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\, #1 \,\right\rbrack} \newcommand{\ceil}[1]{\,\left\lceil\, #1 \,\right\rceil\,} \newcommand{\dd}{{\rm d}} \newcommand{\down}{\downarrow} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\half}{{1 \over 2}} \newcommand{\ic}{{\rm i}} \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow} \newcommand{\isdiv}{\,\left.\right\vert\,} \newcommand{\ket}[1]{\left\vert #1\right\rangle} \newcommand{\ol}[1]{\overline{#1}} \newcommand{\pars}[1]{\left(\, #1 \,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\vphantom{\large A}\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}} \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert} \newcommand{\wt}[1]{\widetilde{#1}}$ $\ds{\int_{-\infty}^{\infty}{\sin\pars{x}\sin\pars{nx} \over x^{2}}\,\dd x = \pi :\ {\large ?}.\qquad n \in {\mathbb N}_{>0}}$
With the identity $\ds{{\sin{x} \over x} = \half\int_{-1}^{1}\expo{\pm\ic kx}\,\dd k}$: \begin{align}&\color{#c00000}{% \int_{-\infty}^{\infty}{\sin\pars{x}\sin\pars{nx} \over x^{2}}\,\dd x} =\int_{-\infty}^{\infty}\pars{\half\int_{-1}^{1}\expo{\ic kx}\,\dd k} \pars{n\,\half\int_{-1}^{1}\expo{-\ic qnx}\,\dd q}\,\dd x \\[3mm]&={1 \over 4}\,n\int_{-1}^{1}\int_{-1}^{1}\ \underbrace{\int_{-\infty}^{\infty}\expo{\ic\pars{k - nq}x}\,\dd x} _{\ds{=\ 2\pi\,\delta\pars{k - nq}}}\ \,\dd k\,\dd q \end{align} where $\ds{\delta\pars{x}}$ is the Dirac Delta Function.
\begin{align}&\color{#c00000}{% \int_{-\infty}^{\infty}{\sin\pars{x}\sin\pars{nx} \over x^{2}}\,\dd x} =\half\,n\pi\int_{-1}^{1}\Theta\pars{1 - n\verts{q}}\,\dd q \end{align} $\ds{\Theta\pars{x}}$ is the Heaviside Step Function.
\begin{align}&\color{#c00000}{% \int_{-\infty}^{\infty}{\sin\pars{x}\sin\pars{nx} \over x^{2}}\,\dd x} =\half\,n\pi\int_{-1}^{1}\Theta\pars{{1 \over n} - \verts{q}}\,\dd q =\half\,n\pi\int_{-1/n}^{1/n}\dd q=\half\,n\pi\pars{2 \over n} \end{align}
$$ \color{#00f}{\large% \int_{-\infty}^{\infty}{\sin\pars{x}\sin\pars{nx} \over x^{2}}\,\dd x = \pi}\,\qquad\qquad n = 1,2,3,\ldots $$

- 89,464
Consider,
$$f(z) = \Im \frac{e^{iz[n+1]}}{z^2}$$
Consider a semi circle, with a small empty semi circle.
The residue is $0$
$$\oint_{C} f(z) dz = 0 = \int_{BDEFG} f(z) dz + \int_{-R}^{-\epsilon} f - (i)\int_{0}^{\pi} \frac{e^{i\epsilon e^{i\theta}}}{\epsilon e^{i\theta}} + \int_{\epsilon}^{R} f$$
$$\lim_{R \to \infty, \epsilon \to 0} \oint_{C} f(z) dz= 0 + \int_{-\infty}^{0} f(x) dx - (i)\int_{0}^{\pi} \lim_{\epsilon \to 0} \frac{e^{i\epsilon e^{i\theta}}}{\epsilon e^{i\theta}} + \int_{0}^{\infty} f(x) dx \space \space \space \space \space \space \space \space \space \space \space \space \space (1) $$
$$\int_{-\infty}^{\infty} f(x) dx - (i)\int_{0}^{\pi} \lim_{\epsilon \to 0} \frac{e^{i\epsilon e^{i\theta}}}{\epsilon e^{i\theta}} = 0 \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space \space (2)$$
$$\int_{-\infty}^{\infty} f(x) dx = \Im \{i\pi\} = \pi$$
We took the imaginary part because we took the imaginary part of $e^{iz}$ to get $\sin(z)$ in the first place.
In $(1)$ $\displaystyle \int_{BDEFG} f(z) dz$ vanishes. Below is how: Read Jordan's Lemma first.
We see that: $$f(z) = e^{iz[n+1]}g(z) \space \text{where} \space g(z) = \frac{1}{z^2}$$
$$M_R = \max|g(Re^{i\theta})| = \frac{1}{\left| Re^{i\theta} \right |^2} = \frac{1}{R^2}$$
Jordan's lemma states:
$$\left| \int_{BDEFG} f(z) dz \right| \le \frac{\pi}{n+1}M_R = \frac{\pi}{R^2(n+1)}$$
$$\lim_{R \to \infty} \frac{\pi}{R^2(n+1)} = 0$$
Therefore, the integral on the top contour vanished (becomes 0).
In $(2)$ we take the limit inside the integral due to the dominated convergence theorem. Find $g(z)$ such that, $|f(z)| \le g(z)$

- 10,465
-
1
-
@M.N.C.E., how? $$\sin(z) = \Im e^{iz}$$ then
$$\sin(nz) = \Im e^{niz}$$
$$\sin(nz)\sin(z) = \Im e^{iz(n+1)}$$
– Amad27 Jan 04 '15 at 16:09 -
M.N.C.E is right. For two complex numbers $w, u$ it is $\Im(uw) \neq \Im(u)\Im(w)$. Example $w=1+i$ and $u=1-i$. $uw=2$. Thus $\Im(u)=1, \Im(w)=-1, \Im(uw)=0$ and thus you'd end up with $1\cdot(-1) = 0$ which is obviously wrong. – Snake707 Feb 19 '19 at 22:01
-
@Amad27 $\sin(z)=\frac{e^{iz}-e^{-iz}}{2i}$ is not equal to $\Im e^{iz}$. – Bach Apr 29 '19 at 07:13