15

I was working on an integral which I found on Quora. I simplified it a lot and ended up with this intgeral

$$\int_0^{\infty}\dfrac{\operatorname{sech}(\pi x)}{1+4x^2}\, \mathrm dx $$

I tried converting this into exponential form and using geometric series which ended up in this.

$$2\sum_{k=0}^\infty (-1)^k\int_0^\infty \dfrac{e^{-(2k+1)\pi x}}{4x^2+1}\, \mathrm dx $$

I didn't try to solve this using exponential integral, as I am not that much familiar with it.

Using Wolfram|Alpha, I figured out that this integral is equal to $\frac{1}{2}\ln{2}$. The simple answer makes me suspect if the integral is just a tricky one.

How can I evaluate this integral, using this method or any other method, except contour integration?

Zack
  • 2,850
  • Maybe try writing $\frac{1}{1+4x^2} $ as a geonetric series. That way you get integrals like $\int x^n e^{-x} dx $ which are related to the gamma function. – Vishu Jun 20 '21 at 15:29
  • @Tavish the integration bounds are from $0$ to $\infty $ so won't work directly , however you could just split the integral $\sum_{k=0}^\infty \int_{k}^{k+1} f(x)\mathrm{d}x$ and then use a substituiton to $x-k=u$ to get the bounds to $0$ to $1$, then maybe – Sarthak Sahoo Jun 20 '21 at 15:32
  • @Sarthak I had thought of don't this. But it won't work, as the denominator will have $u+k$. – Laxmi Narayan Bhandari Jun 20 '21 at 15:38
  • @Laxmi Narayan Bhandari Yeah that might be possible since it would ultimately depend if $|au^2+bu+c|\le 1$ – Sarthak Sahoo Jun 20 '21 at 15:45
  • 1
    @SarthakSahoo Splitting the integral as $\int_0^{1/2}+ \int_{1/2}^{\infty} $ would work. – Vishu Jun 20 '21 at 15:50
  • 3
    A shame they are closing the question for “lack of context”. Look at all these creative answers. Fourier, residues, expansions, clever changes of variables... this is the type of post that we all benefit from. This site’s rules are way too rigid sometimes. – Stefan Lafon Jun 26 '21 at 05:27
  • 1
    @Stefan thanks for understanding. Don't know why would someone report it. The question doesn't even require more context. More context will be useless. – Laxmi Narayan Bhandari Jun 26 '21 at 11:42

5 Answers5

16

Integrate as follows

\begin{align} &\int_0^{\infty}\frac{\operatorname{sech}(\pi x)}{1+4x^2}\>dx\\ \overset{t=2\pi x }=& \frac\pi2 \int_{-\infty}^{\infty}\frac{e^{\frac t2}}{(e^t+1)(\pi^2+t^2)}dt\\ =& \frac\pi2 \int_{-\infty}^{\infty}\frac{e^{\frac t2}}{e^t+1}Re\left(\frac1\pi \int_0^\infty e^{-(\pi-i t)y }dy\right)dt\\ =& \frac12 Re\int_{0}^{\infty}e^{-\pi y} \left(\int_{-\infty}^\infty \frac{e^{a t}}{e^t+1}dt \right)dy \>\>\>\>\>\>\>a= iy+\frac12\\ \overset{x=e^t}=& \frac12 Re\int_{0}^{\infty}e^{-\pi y} \left(\int_{0}^\infty \frac{x^{a-1}}{x+1}dx \right)dy \\ =& \frac12Re \int_{0}^{\infty}e^{-\pi y}\pi\csc(\pi a)\,dy \\ =& \int_{0}^{\infty} \frac\pi{e^{2\pi y}+1}dy \overset{t=e^{-2\pi y}}=\frac12\int_0^1 \frac1{1+t}dt\\ =&\frac12\ln2 \end{align}

Quanto
  • 120,125
12

All proposed different ways of solution to the problem are really nice. In my opinion, contour integration "catches" the symmetry of this problem deeper and allows to get the closed answer to more general problem.

Let's denote $$I(a)=\int_{-\infty}^\infty\frac{dx}{(e^{\pi x}+e^{-\pi x})(a^2+x^2)}$$ Then $$I_0=\int_0^{\infty}\dfrac{\operatorname{sech}(\pi x)}{1+4x^2}\, \mathrm dx=\frac{1}{4}\int_{-\infty}^\infty\frac{dx}{(e^{\pi x}+e^{-\pi x})(\bigl(\frac{1}{2}\bigr)^2+x^2)}=\frac{1}{4}I\Bigl(\frac{1}{2}\Bigr)$$ Let's also consider $$J(a)=\int_{-\infty}^\infty\frac{\log\bigl(a^2+x^2\bigr)}{e^{\pi x}+e^{-\pi x}}dx$$

Then $I(a)=\frac{1}{2a}\frac{\partial}{\partial a}J(a)$ and

$$J(a)=2\,\Re\int_{-\infty}^\infty\frac{\log\bigl(a-ix\bigr)}{e^{\pi x}+e^{-\pi x}}dx=\frac{2}{\pi}\Re\int_{-\infty}^\infty\frac{\log\bigl(a-\frac{it}{\pi}\bigr)}{e^t+e^{-t}}dt$$ $$=\frac{2}{\pi}\Re\int_{-\infty}^\infty\frac{\log\bigl(\frac{a}{2}-\frac{it}{2\pi}\bigr)+\log2}{e^t+e^{-t}}dt=J_1(a)+J_2$$ where $$J_2=\frac{2\log2}{\pi}\int_{-\infty}^\infty\frac{dt}{e^t+e^{-t}}=\frac{2\log2}{\pi}\int_{-\infty}^\infty\frac{e^tdt}{e^{2t}+1}=\frac{2\log2}{\pi}\int_{0}^\infty\frac{dx}{x^2+1}=\log2$$ To evaluate $J_1(a)$ we follow the approach developed by Iaroslav V. Blagouchine (here)

Noting that $\log z=\log\Gamma (z+1)-\log\Gamma(z)$ $$J_1(a)=\frac{2}{\pi}\Re\int_{-\infty}^\infty\frac{\log\bigl(\frac{a}{2}-\frac{it}{2\pi}\bigr)}{e^t+e^{-t}}dt=\frac{2}{\pi}\Re\int_{-\infty}^\infty\frac{\log\Gamma\bigl(\frac{a}{2}-\frac{it}{2\pi}+1\bigr)-\log\Gamma\bigl(\frac{a}{2}-\frac{it}{2\pi}\bigr)}{e^t+e^{-t}}dt$$

Next, we choose a rectangular contour C in the complex plane enter image description here

Noting that integral along the path $(1)$ and $(2)$ wanish as $R\to\infty$ and that $e^{2\pi i+t}=e^t$, we can write: $$J_1(a)=-\frac{2}{\pi}\Re\oint_C\frac{\log\Gamma\bigl(\frac{a}{2}-\frac{iz}{2\pi}\bigr)}{e^z+e^{-z}}dz=\Re\biggl(-2\pi i \frac{2}{\pi}\operatorname*{Res}_{\binom{z=\pi i/2}{z=3\pi i/2}}\frac{\log\Gamma\bigl(\frac{a}{2}-\frac{iz}{2\pi}\bigr)}{e^z+e^{-z}}\biggr)=2\Bigl(\log\Gamma\bigl(\frac{a}{2}+\frac{3}{4}\bigr)-\log\Gamma\bigl(\frac{a}{2}+\frac{1}{4}\bigr)\Bigr)$$ $$J(a)=\int_{-\infty}^\infty\frac{\log\bigl(a^2+x^2\bigr)}{e^{\pi x}+e^{-\pi x}}dx=J_1(a)+J_2=2\Bigl(\log\Gamma\bigl(\frac{a}{2}+\frac{3}{4}\bigr)-\log\Gamma\bigl(\frac{a}{2}+\frac{1}{4}\bigr)\Bigr)+\log2$$ Taking derivative with respect to $a$ we can easily evaluate expressions like $\int_{-\infty}^\infty\frac{dx}{(e^{\pi x}+e^{-\pi x})(a^2+x^2)^n}$.

For example, $$I_0=\int_0^{\infty}\dfrac{\operatorname{sech}(\pi x)}{1+4x^2}\, \mathrm dx=\frac{1}{4}I\bigl(a=\frac{1}{2}\bigr)=\frac{1}{4}\frac{1}{2a}\frac{\partial}{\partial a}J(a)|_{a=1/2}=\frac{1}{8a}\Bigl(\Psi\bigl(\frac{a}{2}+\frac{3}{4}\bigr)-\Psi\bigl(\frac{a}{2}+\frac{1}{4}\bigr)\Bigr)|_{a=1/2}$$ $$I_0=\frac{1}{4}\Bigl(\Psi(1)-\Psi(\frac{1}{2})\Bigr)=\frac{\log2}{2}$$ where $\Psi(x)=\frac{\partial}{\partial x}\log\Gamma(x)$ - digamma function.

Svyatoslav
  • 20,502
  • 2
    Nice solution and good reference. – Zack Jun 23 '21 at 15:48
  • 2
    You can also integrate the digamma function directly and use the recurrence relation of the digamma function. I intended to leave a comment a while ago, but I guess I forgot. – Random Variable Dec 04 '21 at 23:47
9

The question asks for anything "except contour integration", but actually the latter is the very first thing I would apply here. So this answer follows the contour integration approach for reference.

The integral is $\frac12\int_{-\infty}^\infty$, so that if $I_{R,N}=\int_{C_{R,N}}\frac{\operatorname{sech}\pi z}{1+4z^2}\,dz$ where, for $R>0$ and $N$ a positive integer, we take $C_{R,N}$ to be the boundary of $[-R,R]+i[0,N]$, then $\lim\limits_{R\to\infty}\lim\limits_{N\to\infty}I_{R,N}$ is twice the given integral, and it equals $2\pi i$ times the sum of the residues of the integrand at $z=i(n+1/2)$ over $n\in\mathbb{Z}_{\geqslant 0}$ ($n=0$ is a double pole, $n>0$ are simple poles): $$\operatorname*{Res}_{z=i/2}\frac{\operatorname{sech}\pi z}{1+4z^2}=\frac1{4\pi i},\quad\operatorname*{Res}_{z=i(n+1/2)}\frac{\operatorname{sech}\pi z}{1+4z^2}=\frac1{4\pi i}\frac{(-1)^{n-1}}{n(n+1)}.\quad(n>0)$$ Hence the value of the given integral is, as conjectured, $$\frac14\left[1+\sum_{n=1}^\infty(-1)^{n-1}\left(\frac1n-\frac1{n+1}\right)\right]=\frac14\big(1+\ln2-(1-\ln2)\big)=\frac{\ln2}{2}.$$

metamorphy
  • 43,591
  • Well, I had not requested for contour integration approach, because I haven't learned it yet. I knew that it would be much straightforward. – Laxmi Narayan Bhandari Jun 20 '21 at 18:26
  • 1
    Both solutions are really nice. I also think that contour integration provides a shortcut to more general solution (for an arbitrary parameter, for instance, or more general type of integral). I tried to use the method developed by I. Blagouchine – Svyatoslav Jun 21 '21 at 22:53
7

A short solution by using the series of $\operatorname{sech}(x)$.

Since $$\operatorname{sech}(x)= 4\sum_{k=0}^{\infty}\frac{(-1)^k(2k+1)\pi}{(2k+1)^2\pi^2+4x^2}$$ and then we have $$\int_0^{\infty}\frac{\operatorname{sech}(\pi x)}{1+4x^2}dx=\frac{4}{\pi}\sum_{k=0}^{\infty}{(-1)^k(2k+1)}\int_0^{\infty}\frac{dx}{(m^2+4x^2)(1+4x^2)}=$$ where $m=2k+1$ and by partial fraction decomposition we easily can deduce the integral $$\int_0^{\infty}\frac{dx}{(m^2+4x^2)(1+4x^2)}=\frac{\pi}{4m(m+1)}=\frac{\pi}{4(2k+1)(2k+2)}$$ and hence we have $$\sum_{k=0}^{\infty}\frac{(-1)^k}{2(k+1)}=\frac{1}{2}\sum_{k=1}^{\infty}\frac{(-1)^{k-1}}{k}=\frac{\log(2)}{2}$$

Naren
  • 3,562
4

Note that your integrand is even in $x$, so that you are really concerned with

$$ \frac{1}{2}\int_{\mathbb{R}} \operatorname{sech}(\pi x) \frac{1}{1+4x^2}\,dx.$$

If you're willing to accept Fourier transform techniques, we have

$$ \frac{1}{\sqrt{2\pi}}\int_{\mathbb{R}} e^{-ixy}\operatorname{sech}(\pi x)\,dx = \frac{1}{\sqrt{2\pi}}\operatorname{sech}\bigg(\frac{y}{2}\bigg) $$

and also

$$ \frac{1}{\sqrt{2\pi}} \int_{\mathbb{R}} e^{-ixy} \frac{1}{1+4x^2}\,dx = \frac{1}{2}\sqrt{\frac{\pi}{2}} e^{-|y|/2}. $$

Then from Parseval's theorem,

\begin{align} \frac{1}{2}\int_{\mathbb{R}} \operatorname{sech}(\pi x) \frac{1}{1+4x^2}\,dx &= \frac{1}{2} \int_{\mathbb{R}} \frac{1}{\sqrt{2\pi}} \operatorname{sech}\bigg(\frac{y}{2}\bigg) \frac{1}{2}\sqrt{\frac{\pi}{2}} e^{-|y|/2}\,dy \\ &= \frac{1}{4} \int_0^{\infty} \operatorname{sech}\bigg(\frac{y}{2}\bigg) e^{-y/2}\,dy. \end{align}

As for this integral, note that $\operatorname{sech}(x) = \frac{2}{e^x + e^{-x}}$ so that we are left to evaluate

$$ \frac{1}{2}\int_0^{\infty} \frac{e^{-y/2}}{e^{y/2} + e^{-y/2}}\,dy. $$

Let $z = e^{-y/2}$, then $dz = -\frac{1}{2} e^{-y/2}\,dy$, to give

$$ \int_0^1 \frac{1}{z^{-1}+z}\,dz = \int_0^1 \frac{z}{1+z^2}\,dz = \frac{\log(2)}{2}.$$