7
$\begingroup$

Compute the following sum $$\sum_{n=1}^{+\infty}\frac{1}{n^3 \sin(n \pi \sqrt{2})}$$

Source :

Very difficult sum of series

Like jmerry on AoPS I have no idea how to compute the sum.

Any ideas ? Perhaps someone knows already the result..

Thank you in advance for your time.

$\endgroup$
1
  • 3
    $\begingroup$ I would say the first question is whether it converges. That sine could be very small for various $n$, making occasional big terms in the sum. $\endgroup$ May 19, 2014 at 12:47

2 Answers 2

2
$\begingroup$

This is not a solution, but it doesn't quite fit as a comment and with some hope it can be finished. Let $a=\pi\sqrt{2}$ and $$g_k(x)=\sum_{n\in\mathbb{Z},n\neq0}\frac{1}{n^k}\frac{\exp(inx)}{\exp(ina)-\exp(-ina)},$$ you want to know $g_3(0)$. For convergence, $1/\sin na=O(n)$, so $g_k$ converges absolutely and uniformly for $k\geq3$, and we can use $g_k'=ig_{k-1}$ to get lower $k$ (the series converges unconditionally in the sense of distributions for every $k$).

Notice now $$g_k(x+a)-g_k(x-a)=\sum_{n\neq0}\frac{\exp(inx)}{n^k}=c_kB_k((x\text{ mod }2\pi)/2\pi)\quad(*)$$ where $c_k$ is an easy constant and $B_k$ is Bernoulli polynomial. Also $g_k(-x)=(-1)^{k+1}g_k(x)$.

The difference equation $(*)$ would have a polynomial solution - if the RHS were a polynomial (not just a piecewise polynomial) and if $g_k$ didn't have to be $2\pi$-periodic. Nonetheless, I hope that it can be solved also in the periodic case, and thus finally get $g_3(0)$.

Sorry for this irresponsible suggestion.

$\endgroup$
3
  • $\begingroup$ How do you get $1/\sin na=O(n^2)$? $\endgroup$ May 20, 2014 at 13:18
  • 1
    $\begingroup$ @GerryMyerson I actually made a mistake and should have written $O(n)$ (will fix it). The thing is that $\sqrt{2}$, as any quadratic irrational, can't be well approximated by a rational number, i.e. $\exists c>0$ s.t. for every $m,n\in\mathbb Z$, $n\neq0$, $|\sqrt{2}-m/n|>1/n^2$ - and thus $|n\sqrt{2}-m|>1/n$. (for proof - the continued fraction of $\sqrt{2}$ is eventually periodic). $\endgroup$
    – user8268
    May 20, 2014 at 15:00
  • $\begingroup$ @GerryMyerson errata: "$>1/n^2$" and "$>1/n$" should have been "$>c/n^2$" and "$>c/n$" $\endgroup$
    – user8268
    May 20, 2014 at 15:26
0
$\begingroup$

Partial answer:

$$\sum_{n\ge 1}\frac{1}{n^3\sin(n\pi \sqrt{2})}=\sum_{n\ge 1}\sum_{k=0}^{\infty}\frac{e^{-jn\pi\sqrt{2}}}{2j n^3}e^{-2jkn\pi\sqrt{2}}=\frac{1}{2j}\sum_{k\ge 0}\mbox{Li}_{3}\left(e^{-j(2k+1)\pi \sqrt{2}}\right)$$ where $\mbox{Li}_s(\cdot)$ is the polylogarithm function. We need to see if there is a closed form for this kind of summation.

$\endgroup$
2
  • 2
    $\begingroup$ So you actually replaced the summand with another more complex summand :-) $\endgroup$ May 19, 2014 at 11:47
  • $\begingroup$ I was hoping that there may be a sum formula for polylogarithms, which would be more manageable then the sum of $1/\sin(\cdot)$ :-). $\endgroup$ May 19, 2014 at 11:50

You must log in to answer this question.