9
$\begingroup$

While trying to prove $\int_0^{\infty } \frac{\log (x)}{e^x+1} \, dx = -\frac{1}{2} \log ^2(2)$ How to show? in an alternative way, I came to this solution:

$$\sum_{k=0}^{+\infty}(-1)^{k+1}\frac{\log (k+1)+\gamma }{(k+1)}.$$

As both solutions have to be the same, the following equality should be valid:

$$\sum_{k=0}^{+\infty}(-1)^{k+1}\frac{\log (k+1)+\gamma }{(k+1)}=- \frac{1}{2}{{\log }^2(2)}. $$

Can anyone give me some advice on how to prove this equality.

p.s. You can be sure that the equality is correct, as I checked it numerically.

$\endgroup$
  • 1
    $\begingroup$ So, obviously the hard part is $\displaystyle \sum_k \frac{(-1)^k\log(k)}{k!}$. I think a possible route would be to see if you can find $\displaystyle \sum_k \frac{(-1)^kk^x}{k!}$ for a general $x$ (it's easy for integral $x$) and then you can differentiate both sides and let $x=0$. I have to run, but that should work, assuming that sum isn't actually too crazy for non-integral $x$. $\endgroup$ – Alex Youcis Jan 20 '12 at 23:26
  • 2
    $\begingroup$ Are you sure you wrote this right? I get $\sum_{k=0}^\infty (-1)^{k+1} \frac{\log(k+1)+\gamma}{(k+1)!} = -.1548995048$ and $-\frac{1}{2} \log^2(2) = -.2402265070$ approximately $\endgroup$ – Robert Israel Jan 20 '12 at 23:39
  • 2
    $\begingroup$ The ! should be removed from the denominator: $\sum_{k=0}^{\infty} (-1)^{k+1} \frac{\log(k+1)+\gamma}{k+1}=-\frac{1}{2}(\log 2)^2.$ This is obtained from the integral by expanding $(1+e^x)^{-1}=e^{-x}-e^{-2x}+{\dots}$ and integrating term-by-term. $\endgroup$ – David Moews Jan 20 '12 at 23:52
  • $\begingroup$ This is really annoying. I removed the !. I am very ver sorry for this error. $\endgroup$ – wnvl Jan 21 '12 at 0:18
8
$\begingroup$

Note that $$ \begin{align} \sum_{k=1}^\infty(-1)^k\frac{\log(k)+\gamma}{k} &=\lim_{n\to\infty}\left(2\sum_{k=1}^{n}\frac{\log(2k)+\gamma}{2k}-\sum_{k=1}^{2n}\frac{\log(k)+\gamma}{k}\right)\\ &=\lim_{n\to\infty}\left(\sum_{k=1}^{n}\frac{\log(2)+\log(k)+\gamma}{k}-\sum_{k=1}^{2n}\frac{\log(k)+\gamma}{k}\right)\tag{1} \end{align} $$ Using the Euler-Maclaurin Sum Formula, we get that $$ \sum_{k=1}^n\frac{1}{k}=\log(n)+\gamma+\frac{1}{2n}-\frac{1}{12n^2}+O\left(\frac{1}{n^4}\right)\tag{2} $$ and that $$ \sum_{k=1}^n\frac{\log(k)}{k}=\frac12\log(n)^2+C+\frac{\log(n)}{2n}-\frac{\log(n)-1}{12n^2}+O\left(\frac{\log(n)}{n^4}\right)\tag{3} $$ Applying $(2)$ and $(3)$ to $(1)$, leaving out the terms which vanish, we get $$ \begin{align} &\sum_{k=1}^\infty(-1)^k\frac{\log(k)+\gamma}{k}\\ &=\small{\lim_{n\to\infty}\left(\log(2)(\log(n){+}\gamma)+\left(\frac12\log(n)^2+C+\gamma(\log(n){+}\gamma)\right)-\left(\frac12\log(2n)^2+C+\gamma(\log(2n){+}\gamma)\right)\right)}\\ &=\lim_{n\to\infty}\left(\log(2)(\log(n)+\gamma)-\log(2)\log(n)-\frac12\log(2)^2-\gamma\log(2)\right)\\ &=-\frac12\log(2)^2\tag{4} \end{align} $$

$\endgroup$
  • $\begingroup$ What is C in (3)? $\endgroup$ – wnvl Jan 21 '12 at 14:09
  • 1
    $\begingroup$ @wnvl: I don't know if it has a closed form or simple name, but it doesn't matter since it gets canceled. $\endgroup$ – robjohn Jan 21 '12 at 15:45
  • 1
    $\begingroup$ @wnvl: The error in $(3)$ turns out to be less than $\dfrac{log(n)}{120n^4}$, so we can compute $C = -0.07281584548367672486$ to $20$ places by setting $n=100000$. $\endgroup$ – robjohn Jan 21 '12 at 17:22
  • 1
    $\begingroup$ @wnvl: I ran that constant through the Inverse Symbolic Calculator, and it could not find a match. That leads me to think that the constant may not have a closed form or simple name. $\endgroup$ – robjohn Jan 21 '12 at 18:24
6
$\begingroup$

This can be solved similarly to the original problem. The Dirichlet eta function is defined by $$ \eta(s):=\sum_{n\ge 1} \frac{(-1)^{n-1}}{n^s}. $$ The given sum can be rewritten as $$ \sum_{n\ge 1} (-1)^n \frac{\log n}{n}+\sum_{n\ge 1} (-1)^n \frac{\gamma}{n}= \eta'(1)-\gamma \log 2.\qquad (*)$$ We have $$ \eta(s)=\sum_{n\ge 1} \frac{1}{n^s}-2\sum_{n\ge 1} \frac{1}{(2n)^s}=(1-2^{1-s})\zeta(s) $$ so, using the expansions $$\zeta(s)=\frac{1}{s-1}+\gamma+O(s-1),$$ $$ 2^{1-s}=e^{(1-s)\log 2}=1-(s-1)\log 2+\frac{1}{2}(\log 2)^2 (s-1)^2+O((s-1)^3), $$ we get $\eta'(1)=\gamma \log 2 -\frac{1}{2}(\log 2)^2$, so (*) equals $-\frac{1}{2}(\log 2)^2$.

$\endgroup$
  • $\begingroup$ In fact, I use the result from my answer to derive the expansion for $\zeta(s)$ you cite above. See Riemann zeta near 1 $\endgroup$ – robjohn Jan 28 '12 at 14:16
  • $\begingroup$ Great answer but something's off by a sign and I can't figure out what. The term $-\frac{1}{2}(\log 2)^2$ comes from the term $\left ( \zeta'=-\frac{1}{(s-1)^2}\right )* \color{red}{-}(2^{1-s}_{\text{third term}})=\color{red}{+}\frac{1}{(s-1)^2}*\frac{1}{2}(\log 2)^2 (s-1)^2=\frac{1}{2}(\log 2)^2$ $\endgroup$ – user5389726598465 Jun 5 '17 at 5:23

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.