In one of his letters to G. H. Hardy, Ramanujan gave the following sum $$\dfrac{1}{1^{7}\cosh\left(\dfrac{\pi\sqrt{3}}{2}\right)} - \dfrac{1}{3^{7}\cosh\left(\dfrac{3\pi\sqrt{3}}{2}\right)} + \dfrac{1}{5^{7}\cosh\left(\dfrac{5\pi\sqrt{3}}{2}\right)} - \cdots = \frac{\pi^{7}}{23040}\tag{1}$$ or using $\sum $ notation $$\sum_{n = 0}^{\infty} \dfrac{(-1)^{n}}{(2n + 1)^{7}\cosh\left(\dfrac{(2n + 1)\pi\sqrt{3}}{2}\right)} = \frac{\pi^{7}}{23040}$$ Since $\cosh y = (e^{y} + e^{-y})/2$ we can see that the sum is equal to $$S = 2\sum_{n = 0}^{\infty}\dfrac{(-1)^{n}\exp\left(-\dfrac{(2n + 1)\pi\sqrt{3}}{2}\right)}{(2n + 1)^{7}\left\{1 + \exp\left(-(2n + 1)\pi\sqrt{3}\right)\right\}}$$ Putting $$q = \exp\left(-\pi\sqrt{3}\right)$$ we get the sum as $$S = 2\sum_{n = 0}^{\infty}\frac{(-1)^{n}q^{n + 1/2}}{(2n + 1)^{7}(1 + q^{2n + 1})}$$ We can then use $$\dfrac{q^{n + 1/2}}{1 + q^{2n + 1}} = \frac{q^{n + 1/2} - q^{3(n + 1/2)}}{1 - q^{4n + 2}}$$ and hope to use the Ramanujan functions $P, Q, R$ given by $$P(q) = 1 - 24\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}\\ Q(q) = 1 + 240\sum_{n = 1}^{\infty}\frac{n^{3}q^{2n}}{1 - q^{2n}}\\ R(q) = 1 - 504\sum_{n = 1}^{\infty}\frac{n^{5}q^{2n}}{1 - q^{2n}}$$ to calculate the sum $S$ in terms of $P, Q, R$. However the stumbling block is the term $(2n + 1)^{7}$ appearing in the denominator. Therefore I am not sure if the functions $P, Q, R$ can be used in the evaluation of sum $S$. Please let me know if there are any other approaches to calculate the sum $S$.

Update: While going through Ramanujan's Notebooks Vol 3 (by Bruce C. Berndt) I found the proof for the above sum. But he uses a highly non-obvious formula (also discovered by Ramanujan) $$\frac{u^{6n}}{\cos u\cos(\omega u)\cos(\omega^{2}u)} = 12\sum_{k = 0}^{\infty}\dfrac{(-1)^{k}\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{6n + 5}}{\left[\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{6} - u^{6}\right]\cosh\left\{\left(k + \dfrac{1}{2}\right)\pi\sqrt{3}\right\}}\tag{2}$$ where $\omega$ is a primitive cube root of unity. Berndt goes on to say that Ramanujan probably obtained this formula $(2)$ via partial fractions and says that it can be obtained by a routine procedure with somewhat lengthy calculation. The method of partial fractions is used to express a rational function (for the purpose of integrating them) in the form of a sum of finite terms which can be integrated via standard formulas. I am nor sure how that could be extended to any general function (which is not rational) and thereby produce a series. Berndt says "the sum $(1)$ can be obtained by putting $n = 0$ in $(2)$ and then equating coefficients of $u^{6}$ on both sides". This part requires some reasonable amount of calculation, but it is not so difficult.

I want to understand the technique of partial fractions as applied to general functions (may be with some requirement of continuity and differentiability) and its proper justification so that I can provide a proof of formula $(2)$ for myself and thereby have a complete proof of the Ramanujan's sum $(1)$.

  • 2
    $\begingroup$ Just for diving in there... respect. $\endgroup$ – Patrick Da Silva Aug 16 '14 at 7:58
  • $\begingroup$ Updated with a solution by Bruce C Berndt, but need more details and justification of this approach. $\endgroup$ – Paramanand Singh Aug 23 '14 at 7:03
  • $\begingroup$ are you still interested in a proof of (2)? $\endgroup$ – tired May 12 '15 at 15:13
  • $\begingroup$ @tired: yes I am interested in a proof. $\endgroup$ – Paramanand Singh May 13 '15 at 2:35

We want to prove that $$ \frac{1}{\cos z\cos(\omega z)\cos(\omega^{2}z)}=12\sum_{k = 0}^{\infty}\dfrac{(-1)^{k}\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{5}}{\left[\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{6} - z^{6}\right]\cosh\left\{\left(k + \dfrac{1}{2}\right)\pi\sqrt{3}\right\}}. $$ ["Expand both sides in powers of z and equate coefficients of $z^6$ on both sides to achieve the proposed formula."(Ramanujan's notebooks, part 3, p. 162) This means that the case $n=0$ of eq.$(2)$ is enough to calculate the sum in the title.]

$\it{Proof}.$ Let $$ f(z)=\frac{1}{\cos z\cos(\omega z)\cos(\omega^{2}z)},\qquad \omega=e^{2\pi i/3}, $$ $$ g(z)=12\sum_{k = 0}^{\infty}\dfrac{(-1)^{k}\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{5}}{\left[\left\{\left(k + \dfrac{1}{2}\right)\pi\right\}^{6} - z^{6}\right]\cosh\left\{\left(k + \dfrac{1}{2}\right)\pi\sqrt{3}\right\}}. $$

It is plain that $g(e^{2\pi i /6}z)=g(z)$ and $f(e^{2\pi i /6}z)=f(-\omega^2 z)=f(z)$. Also $f(z)$ ad $g(z)$ have simple poles at the same set of points: $z_{k,m}=\left(k + \dfrac{1}{2}\right)\pi e^{2\pi im /6},\ k=0,1,2,...; 0\le m\le 5$ ($m-$ integer). Simple calculation shows that $$ \operatorname{res}f(z)\Bigr\rvert_{z=\left(k + \frac{1}{2}\right)\pi}=\frac{1}{(-1)^{k+1}\cos\left(\frac{-1+i\sqrt{3}}{2}\left(k + \frac{1}{2}\right)\pi\right)\cos\left(\frac{1+i\sqrt{3}}{2}\left(k + \frac{1}{2}\right)\pi\right)}=\\ \frac{(-1)^{k+1}}{\cos^2\left(k + \frac{1}{2}\right)\frac{\pi}{2}\cdot \cosh^2\frac{\sqrt{3}}{2}\left(k + \frac{1}{2}\right)\pi+\sin^2\left(k + \frac{1}{2}\right)\frac{\pi}{2}\cdot \sinh^2\frac{\sqrt{3}}{2}\left(k + \frac{1}{2}\right)\pi}=\\ \frac{(-1)^{k+1}}{\cosh^2\frac{\sqrt{3}}{2}\left(k + \frac{1}{2}\right)\pi-\sin^2\left(k + \frac{1}{2}\right)\frac{\pi}{2}}=\frac{2(-1)^{k+1}}{\cosh\sqrt{3}\left(k + \frac{1}{2}\right)\pi}, $$ $$ \operatorname{res}g(z)\Bigr\rvert_{z=\left(k + \frac{1}{2}\right)\pi}=\frac{2(-1)^{k+1}}{\cosh\sqrt{3}\left(k + \frac{1}{2}\right)\pi}. $$ One can see that $f(z)$ and $g(z)$ have equal residues at $z_{k,0}$, but by the symmetry mentioned above $g(e^{2\pi i /6}z)=g(z)$, $f(e^{2\pi i /6}z)=f(z)$ one can also see that they have equal residues at all other poles. So $f(z)-g(z)$ doesn't have any poles and since it is a meromorphic function, according to Liouville's theorem it is a constant: $f(z)-g(z)=C$. $C$ can be found considering the point $z=+i\infty$ and we get $C=0$, which completes the proof.

  • $\begingroup$ I forgot to mention that the function $f(z)-g(z)$ is bounded. $\endgroup$ – user82588 Nov 12 '15 at 9:11

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.