11

While trying to answer this question I stumbled on a paper by Bruce C. Berndt which contains the following formula by Ramanujan $$\frac{\pi}{2}\cot\sqrt{w\alpha}\coth\sqrt{w\beta}=\frac{1}{2w}+\sum_{m=1}^{\infty} \left(\frac{m\alpha\coth m\alpha} {w+m^2\alpha} +\frac{m\beta\coth m\beta} {w-m^2\beta} \right) \tag{1}$$ which is supposed to hold for all positive numbers $\alpha, \beta$ with $\alpha\beta=\pi^2$.

Berndt mentions that this formula is wrong and missing a term. The corrected version stands as $$\frac{\pi}{2}\cot\sqrt{w\alpha}\coth\sqrt{w\beta}=\frac{1}{2w}+\frac{1}{2}\log\frac{\beta}{\alpha}+\sum_{m=1}^{\infty} \left(\frac{m\alpha\coth m\alpha} {w+m^2\alpha} +\frac{m\beta\coth m\beta} {w-m^2\beta} \right) \tag{2}$$ for $\alpha>0<\beta,\alpha\beta=\pi^2$. Bruce gives some references which contain a proof of the above formula or its equivalents.

Bruce himself derives the above identity by using a change of variables in the following identity established by R. Sitaramchandrarao $$\pi^2xy\cot (\pi x) \coth (\pi y) =1+\frac{\pi^2}{3}(y^2-x^2)-2\pi xy\sum_{n=1}^{\infty} \left(\frac{y^2\coth (\pi n x/y)} {n(n^2+y^2)}-\frac{x^2\coth(\pi n y/x)}{n(n^2-x^2)}\right) \tag{3}$$ Ramanujan gave a similar (but wrong) formula and Sitaramachandrarao fixed it to arrive at $(3)$.

The derivation of $(2)$ from $(3)$ is not that difficult. The RHS of $(3)$ is modified using the identities $$\frac{y^2} {n(n^2+y^2)}=\frac{1}{n}-\frac{n}{n^2+y^2},\frac{x^2}{n(n^2-x^2)}=\frac{n}{n^2-x^2}-\frac{1}{n}$$ and $$\coth z =1+\frac{2}{e^{2z}-1}$$ The derivation also involves a transformation formula for logarithm of Dedekind eta function. However the proof of $(3)$ is omitted in Berndt's paper.

Unfortunately I have not been able to find those references online which contain a proof for $(2)$ or $(3)$. It is also mentioned that the formula could be proved using Mittag-Leffler expansion but I am barely a novice in complex analysis.

It is desirable to find a direct proof of the above result $(2)$ (or $(3)$) which avoids complex analytic methods. I tried to multiply the partial fractions of $\cot a$ and $\coth b$ but I could not manage to get the desired result.

  • Is the LHS minus the RHS meromorphic in $\alpha \in \Bbb{C}^*$ ? (it is locally meromorphic thus it suffices to check that it stays the same when continued analytically along the unit circle) Where are its poles, of what order and residue ? Is it bounded ? If so then it is constant. – reuns Feb 29 '20 at 08:44
  • @reuns: well the poles of both sides match. I need to evaluate the residues. But from your approach I don't see how we get the constant term $(1/2)\log(\beta/\alpha)$. Ramanujan also missed it somehow. – Paramanand Singh Feb 29 '20 at 08:56
  • There is a proof without complex functions in the google book version of Ramanujan's Lost Notebook, Part 4 by George E. Andrews, Bruce C. Berndt page 273. – Jean Marie Feb 29 '20 at 10:06
  • @JeanMarie: It would be great if you can give the proof here as an answer. Even a brief outline or rough sketch would do. I was not able to find this reference online. – Paramanand Singh Feb 29 '20 at 11:56

2 Answers2

8

Here are the exceirpts from "Ramanujan's Lost Notebook, Part 4" by George E. Andrews, Bruce C. Berndt pages 273 and 274 :

enter image description here

enter image description here

Jean Marie
  • 81,803
  • 1
    Thanks and +1. Can you share the Google book link so that I can browse a few more pages? The current pages mention about equation 12.3.9 which is another equivalent formulation. And that also needs to be proved. – Paramanand Singh Mar 01 '20 at 04:37
  • 1
    I was later able to find the desired reference on Google books but page 272 is not available for preview. Anyway by looking at available pages it appears that the content is almost same as that mentioned in paper of Berndt referenced in my question. – Paramanand Singh Mar 01 '20 at 05:00
6

I thought it would be worthwhile to at least mention how to use the Mittag-Leffler pole expansion theorem to show that $$ \begin{align}\frac{\pi}{2}\cot (\sqrt{w\alpha})\coth(\sqrt{w\beta}) &=\frac{1}{2w}+ \frac{\beta-\alpha}{6} +\sum_{m=1}^{\infty} \left(\frac{m\alpha\coth (m\alpha)} {w+m^2\alpha} +\frac{m\beta\coth (m\beta)} {w-m^2\beta} \right) \\ &-2 \sum_{m=1}^{\infty} \frac{1}{m} \left(\frac{1}{e^{2m \alpha}-1}- \frac{1}{e^{2m \beta}-1}\right). \end{align} $$


Using the princicpal branch of the square root, and under the assumption that $\alpha$ and $\beta$ are positive parameters such that $\alpha \beta = \pi^{2}$, let $$f(w) = \frac{\pi}{2} \cot (\sqrt{w \alpha}) \coth (\sqrt{w \beta}) - \frac{1}{2w}.$$

The function $f(w)$ is meromorphic with simple poles at $w = \frac{m^{2}\pi^{2}}{\alpha} = m^{2} \beta$ and $w = -\frac{m^{2}\pi^{2}}{\beta} = -m^{2} \alpha$, where $m$ is a positive integer.

(Separately, $\cot(\sqrt{w \alpha})$ and $\coth(\sqrt{w \beta})$ have branch points at the origin, but their product has a simple pole at the origin with residue $\frac{1}{2w}$.)

At $w= m^{2} \beta$, the residue of $f(w)$ is $$\begin{align} \lim_{w \to m^{2} \beta} \frac{\pi}{2}\frac{\coth (\sqrt{w \beta})}{\left( \tan (\sqrt{w \alpha})\right)'} &= \lim_{w \to m^{2} \beta} \, \frac{\pi}{2}\frac{2\coth (\sqrt{w \beta}) \, \sqrt{w \alpha}}{\alpha \sec^{2} (\sqrt{w \alpha)}} \\ &= \frac{\pi \coth(m \beta) m \pi }{\alpha} \\ &= m \beta \coth(m \beta). \end{align}$$

Similarly, at $w = - m^{2} \alpha $, the residue of $f(w)$ is $$\begin{align} \lim_{w \to -m^{2} \alpha} \frac{\pi}{2}\frac{\cot (\sqrt{w \alpha})}{\left( \tanh(\sqrt{w \beta})\right)'} &= \lim_{w \to -m^{2} \alpha} \, \frac{\pi}{2}\frac{2\cot (\sqrt{w \alpha}) \, \sqrt{w \beta}}{\beta \operatorname{sech}^{2} (\sqrt{w \beta)}} \\ &= \frac{-\pi i \coth(m \alpha)i m \pi }{\beta } \\ &= m \alpha \coth(m \alpha). \end{align} $$

And the Taylor expansion of $\frac{\pi}{2} \cot (\sqrt{w \alpha}) \coth (\sqrt{w \beta}) - \frac{1}{2w}$ at the origin is $$\frac{\pi(\beta-\alpha)}{6 \sqrt{\alpha \beta}} + \mathcal{O}(w) = \frac{\beta- \alpha}{6} + \mathcal{O}(w). $$


The most basic version of the Mittag-Leffler pole expansion theorem states that if $f(w)$ is a meromorphic function with simple poles at $w= a_{1}, a_{2}, \ldots$ (where $0 < |a_{1}| < |a_{2}| < \ldots$ ) with associated residues $b_{1}, b_{2}, \ldots$, then

$$f(w) = \lim_{w \to 0} f(w) + \sum_{m=1}^{\infty} \left(\frac{b_{m}}{w-a_{m}} + \frac{b_{m}}{a_{m}} \right) $$ provided that

$$\lim_{M \to \infty} \oint_{C_{M}} \frac{f(s)}{s(s-w)} \mathrm ds = 0, $$ where $C_{M}$ is a sequence of contours that encloses $M$ poles without passing through any poles.

In our case the integral vanishes on a sequence of rectangular contours since $\cot(\sqrt{w \alpha}) \coth({\sqrt{w \beta}}) $ is continuous in the strips between the poles on the real axis and tends to $\mp i$ as $\Im(w) \to \pm \infty$.


Applying the Mittag-Leffler pole expansion theorem to $$\frac{\pi}{2}\cot(\sqrt{w\alpha})\coth(\sqrt{w\beta})-\frac{1}{2w}, $$ we get

$$ \begin{align} \frac{\pi}{2}\cot(\sqrt{w\alpha})\coth(\sqrt{w\beta})-\frac{1}{2w} &= \frac{\beta-\alpha}{6} + \sum_{m=1}^{\infty} \left(\frac{m\alpha\coth (m\alpha)} {w+m^2\alpha} +\frac{m\beta\coth (m\beta)} {w-m^2\beta} \right) \\ &+ \sum_{m=1}^{\infty} \frac{1}{m} \left(\coth(m \beta) - \coth(m \alpha) \right), \end{align} $$

where $$ \begin{align} \sum_{m=1}^{\infty} \frac{1}{m} \left(\coth(m \beta) - \coth(m \alpha) \right) &= \sum_{m=1}^{\infty} \frac{1}{m} \left(\coth(m \beta) -1 -\left(\coth(m \alpha)-1 \right) \right) \\ &= - 2 \sum_{m=1}^{\infty} \frac{1}{m} \left(\frac{1}{e^{2 m \alpha}-1}-\frac{1}{e^{2m \beta}-1} \right). \end{align}$$

  • 1
    Thanks (+1)! The last sum involving $e^{2m\alpha}-1$ in denominator is handled via logarithm of Dedekind eta function. If $\eta(q) =q^{1/12}\prod_{n=1}^{\infty} (1-q^{2n})$ then $$\log \eta(q) =\frac{\log q} {12}-\sum_{n=1}^{\infty}\frac{q^{2n}}{n(1-q^{2n})}$$ and then we get desired sum if we put $q=e^{-\alpha} $. So the last part in your answer is simplified by use of transformation formula for Dedekind eta function. – Paramanand Singh Mar 02 '20 at 01:25
  • Also the reason I prefer to avoid complex analysis is that I find the powerful tools of analytic function theory to be a bit mysterious. It appears that I think I understand the proofs of basic theorems but in reality I don't get them. – Paramanand Singh Mar 02 '20 at 01:30
  • @ParamanandSingh I'm not overly familiar with the Dedekind eta function other than knowing some basic properties needed to evaluate some definite integrals. – Random Variable Mar 02 '20 at 01:54
  • @ParamanandSingh I have had the same feeling (mysterious, unintuitive) for a long time. I gained confidence and some intuition into the "behavior" of functions $Z=f(z)$ in particular by plotting different representations of them. – Jean Marie Mar 02 '20 at 05:22