16

Wolfram Alpha numerical calculation shows that the quantity $$ \sum_{n=1}^\infty\frac{12\pi^2n^2}{(-1)^n \cosh(\pi n\sqrt{3})-1} $$ is 1 with high accuracy. Can anybody prove the resulting conjecture:

$$ \sum_{n=1}^\infty\frac{n^2}{(-1)^n \cosh(\pi n\sqrt{3})-1}\overset{?}{=}\frac{1}{12\pi^2} $$

Thomas Andrews
  • 177,126
Tyrell
  • 1,685
  • 1
    Do you mean $\cosh (\pi n \sqrt{3}) - 1$ or $\cosh (\pi n\sqrt{3} - 1)$? – anomaly Sep 22 '16 at 15:38
  • 1
    I'm too lazy to work out the details myself, but this looks like a standard sort of sum-by-contour-integration problem. – anomaly Sep 22 '16 at 15:38
  • I cannot discern the denominator. Is it $(-1)^n \cosh(\pi)n\sqrt{3}-1$, or $(-1)^n \cosh(\pi n)\sqrt{3}-1$, or $(-1)^n \cosh(\pi n\sqrt{3})-1$ or $(-1)^n \cosh(\pi n\sqrt{3}-1)$? – Mark Viola Sep 22 '16 at 15:47
  • @anomaly looking at the WA input i think it is the first option – tired Sep 22 '16 at 15:48
  • @Dr.MV the WA input answers this question – tired Sep 22 '16 at 15:50
  • I get to a sum like this $$\sum_{n=1}^\infty\frac{6n^2\pi^2}{\sinh^2\left(\frac{n\pi}{2}(1+i\sqrt{3})\right)}$$ - if it helps. – Chinny84 Sep 22 '16 at 16:01
  • Based on the WA input, and also trying the sum numerically in Mathematica, it's clear that it should be $$ \sum_{n=1}^{\infty}\frac{n^2}{1 - {(-1)}^n \cosh(\pi n \sqrt{3})} $$ – John Barber Sep 22 '16 at 16:31
  • 4
    Where did you come across this sum? My bet is that there is a simple derivation using the residue theorem here. – Winther Sep 22 '16 at 16:38
  • 2
    One idea is to consider a function like $f(z) = \pi\cot(\pi z)\cdot \frac{z^2}{\cos (\pi z) \cosh \left(\sqrt{3} \pi z\right)-1}$. This is constructed such that the residue at $z = n\in\mathbb{N}$ is equal to your $n$'th summand. If we can show that the integral over some contour (e.g. circle of radius $R \to \infty$) encompassing all integers vanishes then the sum will be equal to minus the sum of the other residues of $f$. One probably have to tune the function $f$ a bit to ensure that the integral vanishes. – Winther Sep 22 '16 at 16:58

2 Answers2

14

Take the function $$ f(z) = \frac{z^2}{\sin (\pi z) \left(\cosh \left(\sqrt{3} \pi z\right)-\cos (\pi z) \right)}. $$ Observe that $\sin (\pi z) \left(\cosh \left(\sqrt{3} \pi z\right)-\cos (\pi z) \right)=2\sin (\pi z) \sin (\pi z \omega) \sin (\pi z\omega^2) $, where $\omega=e^{2\pi i/3}$. This means that $f(z)dz=\frac{z^3}{\sin (\pi z) \left(\cosh \left(\sqrt{3} \pi z\right)-\cos (\pi z) \right)}\frac{dz}{z}$ is symmetric under $z\to z\omega$.

The rest is a routine procedure in application of residue theorem. Consider a closed contour $C$ composed of 2 rays, $z=x$, and $z=x\omega$ $(x>0)$ closed by a circle of large radius $R$. The poles inside the contour are $z=ne^{\pi i/3}$. Note that the integral along the circle vanishes in the limit $R\to\infty$, and the integrals along straight lines cancel each other out. Now one carefully calculates the contribution of residues at $z=n$ and $z=n\omega$ and the contribution of the residue at $z=0$: $$ \lim_{R\to\infty}\int_Cf(z)dz=-\pi i~\sum_{n=1}^\infty\underset{z=n}{\text{res}}f(z)\cdot 2-\frac{2\pi i}{3}\underset{z=0}{\text{res}}f(z)=2\pi i~\sum_{n=1}^\infty\underset{z=ne^{\pi i/3}}{\text{res}}f(z). $$ As a result $$ 2\sum_{n=1}^\infty\frac{n^2}{\pi(-1)^n \left(\cosh(\pi n\sqrt{3})-(-1)^n\right)}=-\frac{1}{3}\frac{1}{2\pi^3} $$ Finally $$ \sum_{n=1}^\infty\frac{n^2}{1-(-1)^n \cosh(\pi n\sqrt{3})}{=}\frac{1}{12\pi^2} $$

9

If we put $q=e^{-\pi\sqrt{3}}$ then the desired sum in question is given by $F(-q) $ where $$F(q) =2\sum_{n=1}^{\infty}\frac{n^2q^n}{(1-q^n)^2}$$ The above function is connected to well known functions $P, Q, R$ of Ramanujan : \begin{align} P(q) &=1-24\sum_{n\geq 1}\frac{nq^n}{1-q^n}\tag{1}\\ Q(q) &=1+240\sum_{n\geq 1}\frac{n^3q^n}{1-q^n}\tag{2}\\ R(q) &=1-504\sum_{n\geq 1}\frac{n^5q^n}{1-q^n}\tag{3} \end{align} We have the following relation $$q\frac{dP(q)} {dq} =-12F(q)=\frac{P^2(q)-Q(q)}{12}\tag{4}$$ The first equality above is a matter of differentiating the series for $P(q) $ but the second identity is deep and was proved by Ramanujan.

Thus the desired sum $F(-q)$ is given by $$\frac{Q(-q)-P^2(-q)}{144}$$ The calculation of $P(-q), Q(-q) $ can be done via their link with elliptic integrals $K, E$ with modulus $k$ corresponding to nome $q$. We have \begin{align} P(-q) &=\left(\frac{2K}{\pi}\right) ^2\left(\frac{6E}{K}+4k^2-5\right)\tag{5}\\ Q(-q) &=\left(\frac{2K}{\pi}\right) ^4(1-16k^2k'^2)\tag{6} \end{align} Let us also note that the modulus $k$ corresponding to $q$ is given by $$k=\frac{\sqrt{3}-1}{2\sqrt{2}}$$ This value of $k$ is obtained from the modular equation of degree $3$ $$\sqrt{kl} +\sqrt{k'l'} =1$$ by putting $l=k', l'=k$. Using the value of $k$ in $(6)$ we get that $Q(-q) =0$ and it remains to evaluate $P(-q) $. It can be proved with some effort that $P(-q) =2\sqrt{3}/\pi$ which gives the sum of series in question as $-1/12\pi^2$.


The value of $P(-q) $ is obtained via the value of $P(q^2)$. The formula for $P(q^2)$ is $$P(q^2)=\left(\frac{2K}{\pi}\right)^2\left(\frac{3E}{K}+k^2-2\right)\tag{7}$$ Using $(5),(7)$ we get $$2P(q^2)-P(-q)=\left(\frac{2K}{\pi}\right)^2(1-2k^2)\tag{8}$$ Ramanujan gave a technique to evaluate the value of $P(q^2)$ using the identity $$3P(q^6)-P(q^2)=\frac{4KL}{\pi^2}(1+kl+k'l')$$ where $l, l', L$ are related to nome $q^3$. Putting $q=e^{-\pi/\sqrt{3}},k=(\sqrt{3}+1)/2\sqrt{2},l=(\sqrt {3}-1)/2\sqrt{2}$ and noting that $l=k'$, $K/L=L'/L=\sqrt{3}$ we get $$3P(e^{-2\pi\sqrt {3}})-P(e^{-2\pi/\sqrt {3}})=\frac{3\sqrt {3}}{2}\cdot\frac{4L^2}{\pi^2}\tag{9}$$ Further using logarithmic differentiation of transformation formula for Dedekind eta function one can prove that $$nP(e^{-2 \pi\sqrt {n}}) +P(e^{-2\pi/\sqrt {n}}) =\frac{6\sqrt{n}}{\pi}\tag{10}$$ Putting $n=3$ in above equation and adding it to $(9)$ we get $$P(e^{-2\pi\sqrt{3}})=\frac{\sqrt{3}}{4}\cdot\frac{4L^2}{\pi^2}+\frac{\sqrt{3}} {\pi}$$ Changing notation a bit so that $q=e^{-\pi\sqrt{3}}$ and then $L$ can be replaced by $K$ and we have $$P(q^2)=\frac{\sqrt{3}}{4}\left(\frac{2K}{\pi}\right) ^2+\frac{\sqrt{3}}{\pi}$$ Using this value in $(8) $ and $k=(\sqrt{3}-1)/2\sqrt{2}$ we get $P(-q) =2\sqrt {3}/\pi$.


Alternatively we can use the notation $q=\exp(2\pi i\tau) $ where $\tau$ has positive imaginary part and define $$E_2(\tau)=P(q), E_4(\tau)=Q(q) $$ and use the complex version of the identity $(10)$ as $$E_2(-1/\tau)=\tau^2E_2(\tau)-\frac {6i\tau}{\pi}\tag{11}$$ (the identity $(10)$ is arrived at by putting $\tau=i\sqrt{n} $). The identity for $E_4$ is $$E_4(-1/\tau)=\tau^4E_4(\tau)$$ If we put $\tau=(-1+i\sqrt{3})/2$ so that $\tau$ is the cube root of unity with positive imaginary part then we have $-1/\tau=\tau+1$ and hence $$\tau^4E_4(\tau)=E_4(-1/\tau)=E_4(\tau+1)=E_4(\tau)$$ so that $E_4(\tau)=0$ or $Q(-e^{-\pi\sqrt{3}})=0$.

Further we have $$E_2(-1/\tau)=E_2(\tau+1)=E_2(\tau)$$ and from equation $(11)$ we get $$E_2(\tau)=\frac{6i\tau}{(\tau^2-1)\pi}=\frac{2\sqrt{3}}{\pi}$$ as the value of $P(-e^{-\pi\sqrt{3}})$. This avoids the difficult but more general technique of Ramanujan described earlier.

  • Wow. Thanks a lot for your answer. This answer is more general than the earlier one, albeit this generality is achieved at the expense of requiring a lot more knowledge to understand it than basic complex integration. Out of curiosity, do you know all these formulas by heart? What do you do in order not to forget them over time? I always forget stuff that I do not use regularly... – Tyrell Oct 19 '23 at 12:17
  • @Tyrell: its common to forget things which one doesn't use regularly and I am prone to it. – Paramanand Singh Oct 19 '23 at 16:28