10
$\begingroup$

Working with some integrals I stumbled upon the following slowly converging series:

$$ S = \sum_{n = 0}^{\infty}\left(-1\right)^{n} \left[n + \frac{3}{2} + \left(n + 1\right)\left(n + 2\right) \log\left(1 - \frac{1}{n + 2}\right)\right] $$

I have reasons to suspect that the series has a closed form:

$$S=\frac{1}{2}-\frac{7 \zeta(3)}{2 \pi^2}=0.073721601182494209 \ldots$$

The actual proof eludes me so far.

Can you prove or disprove this conjecture?

Writing the logarithm as a series we have:

\begin{align} &\left(n + 2\right) \log\left(1 - \frac{1}{n+2}\right) = -\sum_{k = 1}^{\infty} \frac{1}{k\left(n + 2\right)^{k - 1}} \\ = &\ -1-\sum_{k=1}^\infty \frac{1}{(k+1) (n+2)^k} \end{align} Which turns the series into:

$$S=\sum_{n=0}^\infty (-1)^n \left(\frac{1}{2}-(n+1) \sum_{k=1}^\infty \frac{1}{(k+1) (n+2)^k} \right)$$


I can provide the way I came to this expression, but it's very long and complicated, as usual. I'd like some clear proof, if possible.


To get the sense of how slowly the series converges, for $20000$ terms the result agrees with the stated closed form in $4$ first significant digits.


The integral from which this series was obtained is (again, conjectured):

$$\int_0^1 {_2 F_1} (1,-t;2-t;-1) dt = \frac{7 \zeta(3)}{\pi^2}+\frac{1}{2}$$

I don't think it's very useful, except for numerical confirmation.

$\endgroup$
5
  • $\begingroup$ Since you were working with some integrals, do you have an equivalent integral representation for that sum? $\endgroup$ – Zacky Aug 10 '19 at 19:00
  • 1
    $\begingroup$ By Maple it agrees for at least $200$ digits, although it does not compute the closed form. If there was an integral form though, I think it could.(not saying it would be a proof, but still verification) $\endgroup$ – Sil Aug 10 '19 at 19:02
  • 1
    $\begingroup$ @ㄴㄱ, sure, please see the edit $\endgroup$ – Yuriy S Aug 10 '19 at 19:03
  • $\begingroup$ As it is already pointed out in one of the answers, the best way to attack such a problem is by functional equations. From seeing the conjectured result this was my first bet too, since the combination of Apéry's constant divided by $\pi^2$ is a rather strange one apart from derivatives of zeta-like functions via the functional equations (at least from what I have experienced). $\endgroup$ – mrtaurho Aug 11 '19 at 12:20
  • $\begingroup$ $\texttt{Relative Error}$ is $\displaystyle\large 10^{-13} \%$. $\endgroup$ – Felix Marin Aug 11 '19 at 20:09
8
$\begingroup$

For $\Re(s)> 0$, let $F(s)$ be defined as$$ \small F(s) = \sum_{n=0}^\infty (-1)^n\left[(n+1)^{1-s} + \frac 1 2 (n+1)^{-s} + \color{red}{(n+1)(n+2)}\left[(n+1)^{-s}\ln(n+1) - \color{red}{(n+2)^{-s}\ln(n+2)}\right]\right]. $$ Then $F(s)$ is analytic and $\displaystyle S= \lim_{\substack{s\to 0\\\Re(s)>0}}F(s)$ holds. Assume $\Re(s)>3$ for a moment so that each term in the summand is absolutely summable. Then by making $n+1\mapsto n$ to the red-colored term, we get \begin{align*} F(s) =& \sum_{n=0}^\infty (-1)^n\left[(n+1)^{1-s} + \frac 1 2 (n+1)^{-s} + 2(n+1)^{2-s} \ln(n+1)\right]\\ =&\eta(s-1) +\frac 1 2 \eta(s) -2\eta'(s-2) ,\qquad \Re(s)>3 \end{align*} where $\eta(s)$ is the Dirichlet's eta function. By analytic continuation, this should also hold for all $\Re(s)>0$ and it follows $$ S = \lim_{\substack{s\to 0\\\Re(s)>0}}F(s) = \eta(-1) + \frac 1 2 \eta(0)-2\eta'(-2)=\frac 1 2 - \frac{7\zeta(3)}{2\pi^2} $$ where the result can be derived from the functional equations satisfied by $\eta(s)$ and $\zeta(s)$, i.e. $$ \eta(s) = (1-2^{1-s})\zeta(s),\qquad \zeta(s) = 2^s\pi^{s-1}\sin\left(\frac{\pi s}{2}\right)\Gamma(1-s)\zeta(1-s). $$

$\endgroup$
1
  • $\begingroup$ Great answer - compact and elegant $\endgroup$ – Klangen Jan 7 '20 at 7:06
3
$\begingroup$

Splitting up the positive and negative terms gives

$$S=\sum_{n=1}^\infty-1+2n(2n-1)\ln\left(1-\frac1{2n}\right)-2n(2n+1)\ln\left(1-\frac1{2n+1}\right)$$

Since $\displaystyle\ln\left(1-\frac1n\right)=\ln(n-1)-\ln(n)=-\int_0^1\frac1{x+n-1}~\mathrm dx$ we provide the more suitable form

$$S=\int_0^1\sum_{n=1}^\infty-1+\frac{2n(2n+1)}{x+2n}-\frac{2n(2n-1)}{x+2n-1}~\mathrm dx$$

The summand can be managed with some long division and the digamma function, reducing this down to

$$S=\int_0^1x(x-1)\sum_{n=1}^\infty\frac{(-1)^n}{x+n}~\mathrm dx\\=\frac12\int_0^1x(1-x)\left[\psi^{(0)}\left(\frac{x+2}2\right)-\psi^{(0)}\left(\frac{x+1}2\right)\right]~\mathrm dx$$

Substituting $x\mapsto1-x$ in the second digamma function and using the reflection formula gives us

$$S=\frac12\int_0^1x(1-x)\left[\frac2x-\pi\cot\left(\frac{\pi x}2\right)\right]~\mathrm dx$$

which reduces to a few polylogarithms and the rest is plugging and chugging.

$\endgroup$
1
  • $\begingroup$ Nice! By the way last integral Maple handles by convert((1/2)*(int(x*(1-x)*(2/x-Pi*cot((1/2)*Pi*x)), x = 0 .. 1)), Elliptic_related) and gives $\frac{1}{2}-\frac{7 \zeta(3)}{2 \pi^2}$ . $\endgroup$ – Sil Aug 10 '19 at 21:04

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.