0

Maybe it is now a boring question. Since it took my several minutes to write the answer of WA.

C: Catalan's constant

A: Glaisher–Kinkelin constant

Following the unrigorous method I used here:

Evaluating Accelerator's integral $\int_0^{\infty} \frac{e^{4x}-e^{2x}}{x(e^{2x}+1)^3}dx$

I showed that $$\int_0^{\infty} \frac{e^{2x}-e^{x}}{x(e^{2x}+1)^2}dx=3\ln A - \frac{1}{4}+\sum_{k=1}^{\infty}(-1)^{k-1}k\ln(2k+1).$$

My question is: How can I calculate the divergent sum $\sum_{k=1}^{\infty}(-1)^{k-1}k\ln(2k+1)$ from the derivative of and its zeta function?

EDIT: After J.G.'s harsh comments I tried to be a little more rigorous just for the computation of the value of that integral, without some famous letters like A, B. I found $$\frac{e^{2x}-e^{x}}{x(e^{2x}+1)^2}=\sum_{k=1}^{\infty}(-1)^kk\left(\frac{e^{-2kx}-e^{-(2k+1)x}}{x}\right).$$ I think there will be no objection here. But after I applied Frullani's integral theorem, $$\int_0^{\infty} \frac{e^{2x}-e^{x}}{x(e^{2x}+1)^2}dx=\sum_{k=1}^{\infty}(-1)^kk\ln\left(\frac{2k+1}{2k}\right)\approx -0.06230246...$$ which is wrong since the actual value is approximately $0.187691$. It would be pretty quick computation if it was true. Where is the mistake? Should I leave this way?

Bob Dobbs
  • 10,988
  • 3
    You're thinking about this the wrong way. One can indeed obtain a finite value for this integral by careful methods, but the lesson is not that your divergent series "equals" that values minus the other terms on your RHS. The lesson is you only got that series because of illegitimate $\infty-\infty$ manipulations. Having said that, there might be some summation method that turns your divergent series into whatever the value should be. – J.G. Oct 01 '22 at 10:31
  • 1
    Something's wrong. The integral you show is elementary since it only has powers of $e^x$; it evaluates to $\frac12-\frac\pi8$. Is there anything missing? – Parcly Taxel Oct 01 '22 at 10:41
  • @J.G. How is WA calculating in seconds? – Bob Dobbs Oct 01 '22 at 10:58
  • 3
    Since my and @ParclyTaxel's comments were written, the integrand in this question gained a $\frac1x$ factor, and that takes the integral from a very easy one to a much harder one. – J.G. Oct 01 '22 at 10:59

1 Answers1

3

I present a different solution utilising elementary real analytic techniques, namely Mellin transforms.

Consider the following classical Mellin transform $$\int_{0}^{\infty} \frac{x^{s-1}}{e^{n x}+1} \, dx \stackrel{x \mapsto u/n}{=} n^{-s} \int_{0}^{\infty} \frac{u^{s-1}}{e^u+1}\, du$$ By writing $$\frac{1}{e^u+1} = \frac{e^{-u}}{1+e^{-u}} = -\sum_{k=1}^{\infty} (-e^u)^{-k}=\sum_{k=1}^{\infty} (-1)^{k-1} e^{-k u}$$ we have \begin{align*}\int_{0}^{\infty} \frac{u^{s-1}}{e^u+1}\, du &= \sum_{k=1}^{\infty} (-1)^{k-1} \int_{0}^{\infty} u^{s-1}e^{-k u}\, du\\&\stackrel{u \mapsto t/k}=\sum_{k=1}^{\infty} \frac{(-1)^{k-1}}{k^s} \int_{0}^{\infty} t^{s-1} e^{-t} \, dt \\&= \eta (s)\Gamma (s)=(1-2^{1-s})\zeta(s) \Gamma(s)\end{align*}

Thus $$\int_{0}^{\infty} \frac{x^{s-1}}{e^{n x}+1} \, dx = n^{-s} (1-2^{1-s})\zeta(s) \Gamma(s)$$ for $\Re(n)>0 \land \Re (s)>0$.

Now differentiate this Mellin transform with respect to $n$ to obtain $$\int_{0}^{\infty} \frac{x^{s}e^{n x}}{(1+e^{n x})^2}\,dx = s (1-2^{1-s})n^{-1-s} \zeta(s) \Gamma (s)$$ Take $n \mapsto 2$ to obtain $$\int_{0}^{\infty} \frac{x^s e^{2 x}}{(1+e^{2x})^2} \,dx=s\, 2^{-1-s} (1-2^{1-s}) \zeta (s) \Gamma (s)\tag{1}$$

Now similarly consider the following Mellin transform $$\int_{0}^{\infty} \frac{x^{s-1}e^{-x}}{e^{n x}+1} \, dx \stackrel{x \mapsto u/n}{=} n^{-s} \int_{0}^{\infty} \frac{u^{s-1}e^{-u/n}}{e^u+1}\, du$$ By once again writing $$\frac{1}{e^u+1} = \frac{e^{-u}}{1+e^{-u}} = -\sum_{k=1}^{\infty} (-e^u)^{-k}=\sum_{k=1}^{\infty} (-1)^{k-1} e^{-k u}$$

we have \begin{align*}\int_{0}^{\infty} \frac{u^{s-1}e^{-u/n}}{e^u+1}\, du &= \sum_{k=1}^{\infty} (-1)^{k-1} \int_{0}^{\infty} u^{s-1} e^{-u(k+1/n)}\, du\\&=\Gamma (s) \sum_{k=1}^{\infty} (-1)^{k-1} \left(k+\frac{1}{n}\right)^{-s}\end{align*} Separating this alternating sum into odd and even terms, we can conclude $$\int_{0}^{\infty} \frac{x^{s-1}e^{-x}}{e^{n x}+1} \, dx = 2^{-s} n^{-s} \left(\zeta\left(s, \frac{1+n}{2n}\right)-\zeta \left(s, \frac{1+2n}{2n} \right)\right)\Gamma (s)$$ for $\Re (n)>0 \land \Re(s)>0$ where $\zeta(s,a)$ is the Hurwitz zeta function. Once again differentiate with respect to $n$ and take $n \mapsto 2$ to obtain $$-\int_{0}^{\infty} \frac{x^s e^x}{(1+e^{2 x})^2}\,dx=s\, 2^{-3-2s}\left(4\zeta \left(s,\frac{5}{4}\right)-4\zeta\left(s,\frac{3}{4}\right)+\zeta \left(1+s,\frac{3}{4}\right)-\zeta \left(1+s,\frac{5}{4}\right)\right)\Gamma(s)\tag{2}$$

Proceed to add equations $(1)$ and $(2)$ to obtain $$\int_{0}^{\infty} \frac{x^s (e^{2x}-e^x)}{(1+e^{2x})^2}\,dx=2^{-2 s-3} s \Gamma (s) \left(4 \left(2^s-2\right) \zeta (s)+4 \zeta \left(s,\frac{1}{4}\right)-4 \zeta \left(s,\frac{3}{4}\right)-\zeta \left(s+1,\frac{1}{4}\right)+\zeta\left(s+1,\frac{3}{4}\right)\right)$$

or upon shifting $s \mapsto s-1$ $$\int_{0}^{\infty} \frac{x^{s-1} (e^{2x}-e^x)}{(1+e^{2x})^2}\,dx=\frac{1}{2}\Gamma (s) \left(2^{1-s}\eta(s-1)+\beta(s-1)-\beta(s)\right)\tag{3}$$

where $\eta$ and $\beta$ are the Dirichlet eta and beta functions respectively.

(This next limit swapping bit needs to be rigorously shown to be allowed)

Now taking the limit of $(3)$ as $s \to 0$, we obtain $$\int_{0}^{\infty} \frac{e^{2x}-e^x}{x(1+e^{2x})^2} = -\frac{1}{4}+\frac{G}{\pi}-\frac{1}{3} \ln(2)-\frac{1}{2}\ln (\pi)+3\ln (A)+\ln \Gamma \left(\frac{3}{4}\right)$$ as desired based on known values of the derivative of the Dirichlet eta and beta functions. $\square$

KStarGamer
  • 5,089
  • 1
  • 18
  • 50