3

Is it possible to compute $I=\int_0^\infty \frac{\sin{t}}{e^t-1}dt$?

I encountered this problem while calculating sum $\sum_{n=1}^\infty \frac{1}{n^2+1}$. This integral converges $I=\int_0^\varepsilon+\int_\varepsilon^\infty$, first integral is finite since $\sin{t}/(e^t-1)\tilde{} 1$ and the second is finite since $\sin{t}/(e^t-1)<e^{-t}$.

Any help is welcome. Thanks in advance.

alans
  • 6,643
  • 1
  • 26
  • 51
  • 1
    WolframAlpha gives $\frac{1}{2}(\pi\coth(\pi) - 1) \approx 1.07667$. I'm not sure how you would go about showing that, though. – DMcMor Oct 25 '21 at 16:18
  • 1
    @DMcMor you get that by taking a complex contour approach on $\sum(n^2+1)^{-1}$ where coth is due to a rectangular contour (plenty of duplicates on mse), so I get op wants a different integral method – TheSimpliFire Oct 25 '21 at 16:19
  • To op: it would be of benefit to you and others if you provide more context to your question such as the procedure to go $\sum_{n>0}1/(n^2+1)=\int_{t>0}\sin t/(e^t-1),dt$ (to prevent closure) – TheSimpliFire Oct 25 '21 at 16:21
  • $$im\left(\int_0^{\infty} \frac{e^{it}}{e^t-1}\right)$$ –  Oct 25 '21 at 16:22
  • A couple of solutions are here: https://math.stackexchange.com/questions/4235131/on-the-integral-int-0-infty-frac-sin-lambda-x-mathrmdxe2-pi-x-1/4235298#4235298 – Svyatoslav Oct 25 '21 at 16:28
  • You could try setting $x=2\pi t$ and use the Abel-Plana formula, but there are better methods. – Тyma Gaidash Oct 25 '21 at 16:29
  • @Svyatoslav I would like to see solution which is not from complex analysis, if possible. – alans Oct 25 '21 at 16:36
  • Do you allow the use of Fourier series or Poisson summation formula ? – Célestin Oct 25 '21 at 16:46
  • @Célestin Yes, I would like to see your solution by using Fourier series – alans Oct 25 '21 at 16:50

2 Answers2

5

First method. Using Poisson summation formula
If a continuous integrable function $\varphi$ and its Fourier transform are rapidly going to zero at infinity (check its Wikipedia page for more details) then $$ \sum_{n=-\infty}^\infty \varphi(n) = \sum_{n=-\infty}^\infty \hat \varphi(n) $$ Since the Fourier transform tends to transform rapidly decreasing (to zero) functions to slowly decreasing (to zero) functions, and vice versa, Poisson's formula is well suited to the calculations of certain series.

You can check that for $\varphi(x) = \pi e^{-2\pi |x|}$ we have $\hat \varphi(s) = \frac{1}{1+s^2}$. Poisson tells us that $$ \sum_{n=-\infty}^\infty \frac{1}{n^2 + 1} = \sum_{n=-\infty}^\infty \pi e^{-2\pi |n|} $$ Use parity and geometric series to complete the computation.

Second method. Using Parseval's identity for Fourier series
Let $e_n(t) = e^{2\pi i nt}$ be the Fourier orthonormal system of $L^2[0,1]$ and recall Parseval's identity : $$ \forall f,g \in L^2[0,1] ~~,~~ \langle f,g\rangle = \int_0^1 f(t)\overline{g(t)} \,\mathrm dt= \sum_{n=-\infty}^\infty \langle f,e_n\rangle \overline{\langle g,e_n\rangle} $$ Take $s \in \mathbf C\setminus \mathbf Z$ and let $z=\overline{s}$. Compute the Fourier coefficients of $f(t) = e^{2 \pi i st}$ and $g(t)=e^{2\pi i zt}$ $$ \langle f,e_n\rangle = \frac{1}{2\pi i} \frac{e^{2\pi i s}-1}{s-n} ~~\text{ and } ~~ \overline{\langle g,e_n\rangle} = -\frac{1}{2\pi i} \frac{e^{-2\pi i s}-1}{s-n} $$ Using Parseval's identity, it comes, after some simplifications : $$ \frac{\pi^2}{\sin^2(\pi s)} = \sum_{n=-\infty}^\infty \frac{1}{(s-n)^2} \tag{1} $$ Break the sum into three pieces, and write $$\frac{\pi^2}{\sin^2(\pi s)} = \frac{1}{s^2} + \sum_{n=1}^\infty \frac{1}{(s-n)^2} + \frac{1}{(s+n)^2} \tag{2} $$ Integrate $(2)$ both sides, then multiply by $-1$ to get the famous formula : $$ \pi \cot(\pi s) = \frac{1}{s} + \sum_{n=1}^\infty \frac{2s}{s^2-n^2} $$ Replacing $s$ by $is$ leads to : $$ \pi \coth (\pi s) = \frac{1}{s} + \sum_{n=1}^\infty \frac{2s}{s^2 +n^2} $$ If you really want to avoid the use of any complex variable, replace $s$ by $is$ in $(1)$ or in $(2)$ then integrate in the « real » sense. Again, it is easy to recover the result given by WolframAlpha.

Also, using Fourier series and Dirichlet's theorem (for pointwise convergence) you may check other topics on MSE, for example. :)

ADDENDUM. We can reach our goal a bit faster. Apply Parseval's identity to $f(t)= e^{2\pi ist}$ and $g(t) = \overline{f(t)}$ : $$ \int_0^1 f(t)^2 \mathrm dt = \sum_{n=-\infty}^\infty \langle f,e_n\rangle \langle f,e_{-n}\rangle $$ Substitute things... $$ \frac{1}{2\pi i} \frac{e^{4 \pi i s}-1}{2s} = \sum_{n=-\infty}^\infty \bigg(\frac{e^{2\pi is} - 1}{2\pi i} \bigg)^2 \frac{1}{s^2-n^2} $$ Simplify some terms... $$ 2\pi \cot(\pi s) = \sum_{n=-\infty}^\infty \frac{2s}{s^2-n^2} $$ Finally, by symmetry in $n\neq 0$ $$ \pi \cot(\pi s) = \frac{1}{s} + \sum_{n=1}^\infty \frac{2s}{s^2-n^2} $$

Célestin
  • 663
4

$$I=\int_{0}^{\infty} \frac{\sin t}{e^t-1}d t=\int_{0}^{\infty} \frac{e^{-t}\sin t}{1-e^{-t}}d t= \sum_{n=0}^{\infty}\int_{0}^{\infty}\sin t~ e^{-(n+1)x}dt$$ Use $\int_{0}^{\infty} e^{-at} \sin t~dt=\frac{1}{1+a^2}$ $$I=\sum_{n=1}^{\infty} \frac{1}{1+n^2}$$ See Approximate $\coth(x)$ around $x = 0$ for the proof of $$\coth z-1/z=\sum_{n=1}^{\infty} \frac{2z}{\pi^2 n^2+ z^2}$$ Taking $z=\pi$, we prove that $$I=\frac{\pi}{2} \coth \pi-\frac{1}{2}$$

Z Ahmed
  • 46,319