7
$\begingroup$

With reference to my previous question : Show that $\int_0^\infty \frac{x\log(1+x^2)}{e^{2\pi x}+1}dx=\frac{19}{24} - \frac{23}{24}\log 2 - \frac12\log A$

The following integral should be manageable :

$$\int_0^\infty \frac{\log(1+x^2)}{e^{2\pi x}+1}dx = ?$$

But it seams that wolframalpha struggles to find a closed form solution.

Maybe the answer should be in terms Barnes G function or the derivative of the Hurwtiz zeta function.


Update

My attempt so far, consider

$$F(a,b) = \int^\infty_0 \frac{\log(a+t^2)}{e^{2\pi (t+b)}-1}\,dt$$

We consider the derivative

$$\frac{\partial }{\partial a}F(a,b) = \int^\infty_0 \frac{dt}{(t^2+a)(e^{2\pi (t+b)}-1)}\,dt$$

Use the following expansion

$$\frac{2t}{e^{2\pi t}-1} =\frac{1}{\pi}-t+\frac{2t^2}{\pi}\sum_{k=1}^\infty\frac{1}{k^2+t^2} $$

Hence have

$$\frac{1}{e^{2\pi (t+b)}-1} =\frac{1}{2(t+b)\pi}-\frac{1}{2}+\frac{t+b}{\pi}\sum_{k=1}^\infty\frac{1}{k^2+(t+b)^2} $$

$$\frac{\partial }{\partial a}F(a,b) = \int^\infty_0 \frac{1}{t^2+a}\left\{ \frac{1}{2(t+b)\pi}-\frac{1}{2}+\frac{t+b}{\pi}\sum_{k=1}^\infty\frac{1}{k^2+(t+b)^2}\right\}dt$$

This reduces to $$\frac{\partial }{\partial a}F(a,b)= \frac{1}{2\pi}\int^\infty_0 \frac{1}{(t^2+a)(t+b)} dt-\frac{1}{2}\int^\infty_0 \frac{dt}{t^2+a}+\\\frac{1}{\pi}\sum_{k=1}^\infty\int^\infty_0\frac{t+b}{(t^2+a)(k^2+(t+b)^2)}dt$$

The first and second are simple it remains to evaluate

$$\sum_{k=1}^\infty\int^\infty_0\frac{t+b}{(t^2+a)(k^2+(t+b)^2)}dt$$

Then integrate with respect to $a$.

Or evaluate

$$\sum_{k=1}^\infty\int^\infty_0\frac{\log(t^2+1)(t+b)}{k^2+(t+b)^2}dt$$

Where $b = i+1/2$.

$\endgroup$
  • $\begingroup$ If you make a few substitutions and take the Taylor expansion of the log, you get a series in terms of the Riemann zeta function. $\endgroup$ – Simply Beautiful Art Jan 1 '17 at 13:59
  • $\begingroup$ i have only found a numerical value $\endgroup$ – Dr. Sonnhard Graubner Jan 1 '17 at 14:33
  • $\begingroup$ @SimpleArt, that series is not that simple. $\endgroup$ – Zaid Alyafeai Jan 2 '17 at 3:45
  • 4
    $\begingroup$ Have a look at the answer by pisco125 and you will be surprised: math.stackexchange.com/questions/1311929/… The best you can get here is something in terms of @Olivier Oloa's Poly Stieltjes constants...:/ $\endgroup$ – tired Jan 2 '17 at 9:44
  • $\begingroup$ @ZaidAlyafeai Obviously, if it were, I'd already have my answer checked in :-( $\endgroup$ – Simply Beautiful Art Jan 2 '17 at 12:35
7
$\begingroup$

By integration by parts, such integral equals $$ I=\frac{1}{\pi}\int_{0}^{+\infty}\frac{x\log\left(1+e^{-2\pi x}\right)}{1+x^2}\,dx $$ and since $\mathcal{L}^{-1}\left(\frac{x}{1+x^2}\right)=\cos(s)$ and $\mathcal{L}\left(\frac{(-1)^{k+1} e^{-2\pi k x}}{k}\right) = \frac{(-1)^{k+1}}{k(s+2\pi k)}$, we also have: $$ I = 2\int_{0}^{+\infty}\cos(s)\sum_{k\geq 1}\frac{(-1)^{k+1}}{2\pi k(s+2\pi k)}\,ds $$ so $I$ is a series of weighted sine and cosine integrals depending on a Lerch transcendent.
That is not surprising since $$ \int_{0}^{+\infty}\log(1+x^2)e^{-2k\pi x}\,dx = -\frac{\text{Ci}(2\pi k)}{\pi k}=\frac{1}{\pi k}\int_{2\pi k}^{+\infty}\frac{\cos x}{x}\,dx=\frac{1}{\pi k}\int_{0}^{+\infty}\frac{\cos x}{x+2\pi k}\,dx $$ So: $$\begin{eqnarray*} I &=& \frac{1}{\pi}\int_{0}^{+\infty}\cos(2\pi u)\sum_{k\geq 1}\frac{(-1)^{k+1}}{k (u+k)}\,du\\&=&\int_{0}^{+\infty}\frac{\cos(2\pi u)}{2\pi u}\left[\log(4)+\psi\left(\tfrac{1+u}{2}\right)-\psi\left(\tfrac{2+u}{2}\right)\right]\,du\tag{1}\end{eqnarray*}$$ Additionally, from $$ \zeta(s)=\frac{1}{\Gamma(s)}\left(1-\frac{2}{2^s}\right)^{-1}\int_{0}^{+\infty}\frac{x^{s-1}}{e^x+1}\,dx \tag{2}$$ holding for every $s$ such that $\text{Re}(s)>0$, it is not difficult to derive $$ \int_{0}^{+\infty}\frac{\log x}{e^{2\pi x}+1}\,dx = -\frac{\log(2)\log(8\pi^2)}{4\pi} \tag{3} $$ through a substitution and differentiation under the integral sign. We also have: $$\begin{eqnarray*} I &=& \int_{0}^{+\infty}\frac{\log(x+i)}{e^{2\pi x}+1}\,dx + \int_{0}^{+\infty}\frac{\log(x-i)}{e^{2\pi x}+1}\,dx\\ &=& \left(\int_{i}^{+\infty+i}+\int_{-i}^{+\infty-i}\right)\frac{\log x}{e^{2\pi x}+1}\,dx \tag{4}\end{eqnarray*}$$ hence our problem is equivalent to finding a closed form for the principal value of the integral $$ J = \text{PV}\int_{0}^{i}\text{Re}\left(\frac{\log x}{e^{2\pi x}+1}\right)\,dx \tag{5}$$ by choosing a suitable contour avoiding the singularity at $x=\frac{i}{2}$.
The singularity is associated with the term corresponding to $h=0$ in $$ \frac{1}{e^{2\pi x}+1}=\frac{1}{2}-\sum_{h\geq 0}\frac{4x}{(2h+1)^2 \pi+4\pi x^2}\tag{6}$$ and for every $h>0$ we have $$ \text{Re}\int_{0}^{i}\log(x)\frac{4x}{(2h+1)^2 \pi+4\pi x^2}\,dx = \frac{1}{4\pi}\,\text{Li}_2\left(\frac{4}{(2h+1)^2}\right)\tag{7}$$ while the singular term associated with $h=0$ gives a contribute of $\frac{1}{2}\text{arctanh}\frac{1}{2}$. It follows that the problem of finding a closed form for $I$ is equivalent to finding a closed form for: $$ S = \sum_{h\geq 1}\text{Li}_2\left(\frac{1}{(h+1/2)^2}\right) =\sum_{n\geq 1}\frac{\zeta\left(2n,\tfrac{3}{2}\right)}{n^2}.\tag{8}$$ Now we may exploit: $$ \sum_{n\geq 1}\zeta\left(2n,\tfrac{3}{2}\right)x^{2n} = \frac{4x^2}{4x^2-1}+\frac{\pi x\tan(\pi x)}{2}\tag{9} $$ and check that $S$ just depends on $$\text{PV}\int_{0}^{1}\frac{x\log x}{4x^2-1}\,dx,\qquad \color{blue}{\text{PV}\int_{0}^{1}\log(x)\tan(\pi x)\,dx}.\tag{10}$$ According to Mathematica, the first integral is a combination of $\pi^2,\log^2(2)$ and $\text{Li}_2\left(\tfrac{1}{4}\right)$.
That is not difficult to prove, and reduces the whole problem to finding a closed form for the blue integral. The computation of the blue integral boils down to the computation of $$ \int_{0}^{1}\text{arctanh}(x)\cot\left(\tfrac{\pi x}{2}\right)\,dx =-\frac{2}{\pi}\int_{0}^{1}\frac{\log\sin\tfrac{\pi x}{2}}{1-x^2}\,dx=-\frac{2}{\pi}\int_{0}^{1}\frac{\log\cos\tfrac{\pi x}{2}}{x(2-x)}\,dx\tag{11}$$ where $$ \log\cos\tfrac{\pi x}{2}=\sum_{k\geq 1}(-1)^k\frac{1-\cos(\pi k x)}{k} $$ has a nice Fourier series, but $\frac{1}{x(2-x)}$ does not seem to. However, a decent numerical approximation can be recovered from the fact that over the interval $\left(0,1\right)$ we have: $$ \text{arctanh}(x)\cot\left(\tfrac{\pi x}{2}\right)\approx\frac{2}{\pi}\sqrt{1-x^2}. \tag{12}$$

$\endgroup$
  • $\begingroup$ Nice work (+1). I'm curious if you are able to find a closed form (i have to confess i'm sceptical)...good luck! $\endgroup$ – tired Jan 3 '17 at 10:25
  • $\begingroup$ @tired: after my last update, I am starting to being skeptical too :D $\endgroup$ – Jack D'Aurizio Jan 3 '17 at 18:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.