7
$\begingroup$

I was able to solve the integral $$\int_0^\infty\frac x{e^x-e^\frac x2}dx=4\left(\frac{\pi^2}6-1\right)$$ I want to see other approaches to solving it. Here is my solution: $$\int_0^\infty\frac x{e^x-e^\frac x2}dx=\int_0^\infty\frac{xe^{-x}}{1-e^{-\frac x2}}dx=\int_0^\infty xe^{-x}\sum_{k=0}^\infty e^{-\frac{kx}2}dx=\sum_{k=0}^\infty\int_0^\infty xe^{-(1+\frac{kx}2)x}dx=\sum_{k=0}^\infty\frac1{(1+\frac k2)^2}\int_0^\infty xe^{-x}dx=\sum_{k=0}^\infty\frac1{(1+\frac k2)^2}=\sum_{k=0}^\infty\frac{4}{(k+2)^2}=4\left(\frac{\pi^2}6-1\right)$$I came up with this integral when I was trying to prove that $\int_0^\infty\frac x{e^x-1}dx$ converges using the comparison test. I replaced the $1$ with $e^x/2$ while checking on Desmos whether this integral exists or not, but I put $e^{x/2}$ instead. I decided to have a go at this integral.

$\endgroup$
4
  • 3
    $\begingroup$ Contour integration might be a good candidate here. My first attempt would probably be integrating over the boundary of the rectangle $[0, R] \times [0, 4 \pi]$, which quarter-circular notches at $z = 0, 4 \pi i$. $\endgroup$ Commented Jul 24, 2023 at 2:00
  • 5
    $\begingroup$ $$\int_0^\infty\frac x{e^x-e^\frac x2}dx=4\int_0^\infty\frac{e^{-2t}}{1-e^{-t}}dt\overset{e^{-t}=x}{=}-4\int_0^1\frac {x\ln x}{1-x}dx$$ $$=4\int_0^1\ln xdx-4\int_0^1\frac{\ln x}{1-x}dx=-4-4\int_0^1\frac{\ln (1-x)}xdx=4(\zeta(2)-1)$$ $\endgroup$
    – Svyatoslav
    Commented Jul 24, 2023 at 2:40
  • 2
    $\begingroup$ Nice solution and (+1) $\endgroup$ Commented Jul 24, 2023 at 4:34
  • $\begingroup$ plugging it into Wolframalpha I can see that $\sum_0^{\infty} e^{-kx/2} = \frac{\sqrt{ e^x}}{\sqrt{e^x}-1}$ which we can multiply by 1 to get $\frac{e^{x/2}*\frac{1}{e^{x/2}}}{(e^{x/2}-1)*\frac{1}{e^{x/2}}} = \frac{1}{1-e^{-x/2}}$. And so working back from that you can make that substitution and then you can interchange integrals and sums as well as terms respective to each since the integrand is treated as a constant since it's not dependent on k, and boom. Yep that works. I wouldn't have come anywhere close to solving this on my own $\endgroup$ Commented Jul 26, 2023 at 3:48

3 Answers 3

7
$\begingroup$

Less simple that @Svyatoslav's solution.

Working the antiderivative $$I=\int\frac x{e^x-e^{\frac x2}}\,dx$$ $$x=2\log(t)\quad \implies \quad I=4\int \frac{ \log (t)}{(t-1) t^2} \,dt$$ that is to say $$I=4\int \left(\frac{1}{t-1}-\frac{1}{t^2}-\frac{1}{t} \right)\log(t)\,dt$$ Using integration by parts $$I=4 \left(\frac{1}{t}-\frac{1}{2} \log ^2(t)+\frac{\log (t)}{t}-\text{Li}_2(1-t)\right)$$ Back to $x$ $$I=-4 \text{Li}_2\left(1-e^{x/2}\right)-\frac{x^2}{2}+2 e^{-x/2} (x+2)$$

$$J=\int_0^\infty\frac x{e^x-e^{\frac x2}}\,dx=\frac{2 \pi ^2}{3}-4=4 \left(\frac{\pi ^2}{6}-1\right)$$

Working asymptotics $$K=\int_0^t\frac x{e^x-e^{\frac x2}}\,dx$$ $$K=4 \left(\frac{\pi ^2}{6}-1\right)+2 t\,e^{-t/2}+e^{-t}-\frac{4}{9} e^{-3 t/2}+\frac 14 e^{-2t}+\cdots $$ whose absolute error is less than $0.01$ if $t \geq 16$ and less than $0.001$ if $t \geq 22$.

$\endgroup$
3
  • $\begingroup$ The part from "Working asymptotics" and below was completely optional ;) I noticed that your solution is the one that Wolfram Alpha used, as the answer it gave was $\frac23(\pi^2-6)$, which looks more similar to your form of the answer than mine. $\endgroup$ Commented Jul 24, 2023 at 6:14
  • $\begingroup$ @KamalSaleh. I enjoy to see asymptotics. $\endgroup$ Commented Jul 24, 2023 at 6:38
  • $\begingroup$ I have a question. for the sum of exponential functions in your asymptotics, is it possible to re-write it as $O(e^{-\frac t2}$? $\endgroup$ Commented Sep 15, 2023 at 17:39
6
$\begingroup$

$$\int_0^\infty\frac x{e^x-{e^{x/2}}}dx=4\int_0^\infty \frac{xe^{-x}}{e^x-1}dx\\ =4\int_0^\infty\left(\frac x{e^x-1}-xe^{-x}\right)dx=4\left(\frac{\pi^2}{6}-1\right)$$

$\endgroup$
4
  • $\begingroup$ I see why the second and third expressions are equal, but how did you come up with this transformation of the integrand? $\endgroup$ Commented Jul 24, 2023 at 6:17
  • $\begingroup$ I think steps are very clear. First step is the subs $x\to 2x$. $\endgroup$
    – Bob Dobbs
    Commented Jul 24, 2023 at 6:50
  • $\begingroup$ I understand the first step. I meant how you thought of $\frac x{e^x-1}-xe^{-x}=\frac{xe^{-x}}{e^x-1}$ (I know it is true, I just don't know where it came from). $\endgroup$ Commented Jul 24, 2023 at 17:27
  • 4
    $\begingroup$ @KamalSaleh Partial fraction decomposition with respect to $e^x$:$$\frac1{(e^x-1)e^x}=\frac1{e^x-1}-\frac1{e^x}$$ $\endgroup$
    – user170231
    Commented Jul 24, 2023 at 18:43
4
$\begingroup$

Map $x \mapsto 2x$ so that

$$I := \int_{0}^{\infty}\frac{x}{e^{x}-e^{\frac{x}{2}}}dx = 4\int_{0}^{\infty}\frac{x}{e^{2x}-e^{x}}dx\,.$$

Let $\displaystyle f(z) = \frac{z^2}{e^{2z}-e^z}$. Its set of poles is $\left\{z \in \mathbb{C} \text{ } \wedge \text{ } n \in \mathbb{Z} \backslash \left\{0\right\}: z = 2\pi ni\right\}$. There is a removable pole at the origin because the principal part of the Laurent expansion of $f(z)$ equals $0$. Let $r \ll \pi \ll R$. We integrate along the following contour in the positive direction.

Puzzle Piece Contour

Cauchy's Integral Theorem allows us to write $\displaystyle \oint_C f(z)dz$ as

$$ 0 = \left(\int_{0}^{R}+\int_{R}^{R+2\pi i}+\int_{R+2\pi i}^{r+2\pi i}+\int_{r+2\pi i}^{2\pi i-ir}+\int_{2\pi i-ir}^{0}\right)f(z)dz \,. $$

We equate the imaginary part to avoid divergence issues and then apply $r \to 0^+$ and $R \to \infty$ on both sides.

The first integral vanishes:

$$ I_1 := \lim_{R\to\infty} \Im \int_{0}^{R}\frac{z^{2}}{e^{2z}-e^{z}}dz=0\,. $$

The second integral also vanishes. Note that

$$ \left|\Im \int_{R}^{R+2\pi i}\frac{z^{2}}{e^{2z}-e^{z}}dz\right| \leq \left|\int_{R}^{R+2\pi i}\frac{z^{2}}{e^{2z}-e^{z}}dz\right|\le MR $$

where $R$ is the length of the contour and $M$ is a sufficiently tight upper bound of $|f(z)|$. Parameterize $z=R+iy$ where $y \in [0,2\pi]$. We get that $M$ by

$$ \begin{align} |f(z)| &= \frac{\left|z\right|^{2}}{\left|e^{2z}-e^{z}\right|} \\ &\leq \frac{\left|z\right|^{2}}{\left|\left|e^{2z}\right|-\left|e^{z}\right|\right|} \\ &= \frac{\left|R+iy\right|^{2}}{\left|\left|e^{2\left(R+iy\right)}\right|-\left|e^{R+iy}\right|\right|} \\ &\leq \frac{R^{2}+2Ry+y^{2}}{e^{2R}-e^{R}}\,. \\ \end{align} $$

We can use the Estimation Lemma and the Squeeze Theorem to prove that

$$ I_2 := \lim_{R\to\infty} \Im \int_{R}^{R+2\pi i}\frac{z^{2}}{e^{2z}-e^{z}}dz=0\,. $$

We can retrieve the integral we want from evaluating the third integral. Let $z = x + 2\pi i$ so that

$$ \begin{align} I_3 &= \lim_{R\to\infty}\lim_{r\to 0^+}\Im\int_{R+2\pi i}^{r+2\pi i}\frac{z^{2}}{e^{2z}-e^{z}}dz \\ &= -\lim_{R\to\infty}\lim_{r\to 0^+}\Im\int_{r}^{R}\frac{\left(x+2\pi i\right)^{2}}{e^{2\left(x+2\pi i\right)}-e^{x+2\pi i}}dx \\ &= 4\pi^2 \lim_{R\to\infty}\lim_{r\to 0^+}\Im\int_{r}^{R}\frac{dx}{e^{2x}-e^{x}} - 4\pi\lim_{R\to\infty}\lim_{r\to 0^+}\Im i \int_{r}^{R}\frac{xdx}{e^{2x}-e^{x}} - \lim_{R\to\infty}\lim_{r\to 0^+}\Im\int_{r}^{R}\frac{x^{2}dx}{e^{2x}-e^{x}} \\ &= 0 - \pi I - 0 \\ &= -\pi I \,. \\ \end{align} $$

We evaluate the fourth integral through a residue: $$ \begin{align} I_4 &:= \lim_{r\to0^+}\Im \int_{r+2\pi i}^{2\pi i-ir}\frac{z^{2}}{e^{2z}-e^{z}}dz \\ &= -\Im \frac{i\pi}{2} \mathop{\mathrm{Res}}_{z = 2\pi i} \frac{z^2}{e^{2z}-e^z} \\ &= -\frac{\pi}{2} \Re \lim_{z \to 2\pi i} \frac{z^2}{\frac{d}{dz} \left(e^{2z}-e^z\right)} \\ &= 2\pi^3 \,. \end{align} $$ For the last integral, parameterize $z=iy$ so that $$ \begin{align} I_5 &:= \lim_{r \to 0^+}\Im \int_{2\pi i-ir}^{0}\frac{z^{2}}{e^{2z}-e^{z}}dz \\ &= -\lim_{r \to 0^+}\Im i \int_{0}^{2\pi-r}\frac{\left(iy\right)^{2}}{e^{2iy}-e^{iy}}dy \\ &= \lim_{r \to 0^+} \Re \int_{0}^{2\pi-r}\frac{y^{2}}{e^{2iy}-e^{iy}}dy \\ &= \lim_{r \to 0^+} \Re\int_{0}^{2\pi-r}\left(-\frac{1}{2}y^{2}\left(2\cos y+1\right)-\frac{i}{2}y^{2}\cos\left(\frac{3y}{2}\right)\csc\left(\frac{y}{2}\right)\right)dy \\ &= -\frac{1}{2}\int_{0}^{2\pi}y^{2}\left(2\cos y+1\right)dy-0 \\ &= -\frac{4}{3}\pi^{3}-4\pi \,. \\ \end{align} $$

We gather the contributions together and conclude with

$$ 0 = 0+0-\pi I + 2\pi^3 - \frac{4}{3}\pi^3-4\pi $$ $$ \implies I = \frac{2}{3} \pi^2 - 4 $$

and we're done!

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .