I am wondering if anyone recognises the sum

$\sum_{n=-\infty}^\infty \frac{1}{\cosh\left(x \; (n+p-\frac{1}{2}) \right) \cosh \left(x\; (n-\frac{1}{2}) \right) }$ ?

I am trying to evaluate sums of this (and related) forms, but I appear to be stuck...

(The full form I eventually want to find is $ \sum_{n , \vec{p} \; \in \;\mathbb{Z} \;} \; \frac{ \prod_{i} p_i ^{i-1} \; \prod_{i< j } \; \; [p_j - p_i] \; \; }{ \prod_{i} 2 \cosh \left[ x \; ( p_i - \frac{1}{2} )\right] \; 2\cosh \left[ x \; ( p_i+ n - \frac{1}{2} )\right] }\; $, but any help would be highly appreciated! )

I have been looking through the papers by Zucker -78 (http://epubs.siam.org/doi/abs/10.1137/0510019) and related works, (such as for example this paper by Bruckman, http://www.fq.math.ca/Scanned/15-4/bruckman.pdf), and know how to evaluate sums of the form $\sum_{n=-\infty}^\infty \frac{1}{ \cosh^2 \left(x\; (n-\frac{1}{2}) \right) }$, but I run into problems as soon as one of the arguments is shifted.

The answer to the sum above is given in terms of complete elliptic integrals, and I suspect that that will be the case for the shifted one as well, assuming there is a closed form.

And, on another note, this is my first post here so hope I haven't made any massive mistakes :)

Thanks a lot!


It is essentially a discrete convolution between $\frac{1}{\cosh}$ and $\frac{1}{\cosh}$. It is interesting to point out that $\frac{1}{\cosh}$ is more or less a fixed point of the Fourier transform: $$ \mathscr{F}\left(\frac{1}{\cosh x}\right)=\frac{\pi}{\cosh\frac{\pi s}{2}},$$ and its inverse Laplace transform is given by $$ \mathscr{L}^{-1}\left(\frac{1}{\cosh x}\right) = 2\sum_{n\geq 0}(-1)^n \delta(s-2n-1)=f(s),$$ so: $$ \frac{1}{\cosh\left(x\left(\frac{1}{2}-n\right)\right)}=\int_{0}^{+\infty}f(s) e^{-sx(n-1/2)}\,ds$$ and: $$ \frac{1}{\cosh\left(x\left(n-\frac{1}{2}\right)\right)\cosh\left(x\left(n+p-\frac{1}{2}\right)\right)}=\int_{0}^{+\infty}\int_{0}^{+\infty} f(s)\,f(t)\,e^{-ptx}\,e^{-(s+t)x(n+1/2)}\,ds\,dt$$ so: $$ \sum_{n\geq 0}\frac{1}{\cosh\left(x\left(n-\frac{1}{2}\right)\right)\cosh\left(x\left(n+p-\frac{1}{2}\right)\right)}=\int_{0}^{+\infty}\int_{0}^{+\infty}\frac{ 2\, f(s)\,f(t)\,e^{-ptx}}{\sinh\left((s+t)\frac{x}{2}\right)}\,ds\,dt$$ equals:

$$ \sum_{a\geq 0}\sum_{b\geq 0}\frac{8(-1)^{a+b} e^{-px(2b+1)}}{\sinh((a+b+1)x)}=e^{-px}\sum_{h=0}^{+\infty}\frac{8(-1)^h}{\sinh((h+1)x)}\cdot\frac{1-e^{-2(h+1)px}}{1-e^{-2px}} \tag{1}$$

that is a pretty fast convergent series.

Addendum. We may notice that the degenerate case $p=0$ is way easier to deal with.

Given the Weierstrass product for the $\cosh$ function, we have: $$ \frac{1}{\cosh^2(x)}=\frac{d^2}{dx^2}\log\cosh(x) = -\sum_{m\geq 0}\left(\frac{1}{\left(x+\pi i\frac{2m+1}{2}\right)^2}+\frac{1}{\left(x-\pi i\frac{2m+1}{2}\right)^2}\right)$$ so we may recognize in the RHS a meromorphic function with equally-spaced and equally-behaving double poles along the imaginary axis. It follows that the meromorphic function defined through $$ f(x)=\sum_{n\in\mathbb{Z}}\frac{1}{\cosh^2(x+n)} $$ is a constant multiple of a Weierstrass $\wp$ function, whose connection with elliptic integrals is well-known. It is interesting to point out that the RHS of $(1)$, as $p\to 0$, gives a fast-converging expansion for a Weierstrass $\wp$ function. On the other hand, the meromorphic function $$ g(x)=\sum_{n\in\mathbb{Z}}\frac{1}{\cosh(x+n)\cosh(x+n+t)} $$ has only simple poles, due to: $$\frac{1}{\cosh x}=\sum_{n\geq 0}(-1)^n\left(\frac{i}{x+(2m+1)\frac{\pi i}{2}}-\frac{i}{x-(2m+1)\frac{\pi i}{2}}\right)$$ so $g(x)$ is related with a Jacobi elliptic function; a useful identity (already exploited in a slightly disguised form through the inverse Laplace transform) is:

$$ \frac{1}{\cosh x} = \frac{2}{\pi}\int_{0}^{+\infty}\frac{\cos\left(\frac{2x}{\pi}u\right)}{\cosh u}\,du\tag{2} $$

that makes us aware of the following fact: a discrete convolution of $\frac{1}{\cosh}$ is the integral over $\mathbb{R}^+$ of $\frac{1}{\cosh u}$ times a discrete convolution of $\cos$, way easier to deal with. It turns out that the RHS of $(1)$ is related with $\frac{1}{\cosh(px)}$, too.

  • $\begingroup$ Thank you so much! I've got two questions though. The first one is a pretty stupid one, but it's bugging me. In eq (1), how did you do the last step? I'm assuming its just by defining h=a+b, but then I've got issues with the limits of h... Secondly, you don't happen to know if there's a closed expression for the series in (1)? I want to continue manipulating for quite a while and that would be very useful! As you pointed out, we know these series are related to the jacobi theta-functions, so my guess would be it can be expressed in terms of elliptic integrals or something.. $\endgroup$ – Lou May 9 '16 at 10:49
  • $\begingroup$ @Louise: yes, I put $h=a+b$. Given $a,b\geq 0$, $h$ ranges from $0$ to $+\infty$. The RHS of $(1)$ surely has a closed form in terms of elliptic integrals if $p=0$, as you already know. In the general case, you have to exploit the identity above $(2)$ to understand the distribution of poles, then to get a value of $\text{cn}$ (Jacobi elliptic function). $\endgroup$ – Jack D'Aurizio May 9 '16 at 11:02

I can evaluate your first sum in terms of the log-derivative of the theta function$$\vartheta(p,x)=e^{-p^2x}\prod_{k\ge1}(1+e^{2px-(2k-1)x})(1+e^{-2px-(2k-1)x}).\tag1$$ First, split the sum between positive and negative $n$:\begin{align}\sum_{n\in\mathbb{Z}}\frac1{\cosh[(n-\tfrac12\!+p)x]\cosh[(n-\tfrac12)x]}&=\Bigg[\sum_{n\le0}+\sum_{n\ge1}\Bigg]\frac1{\cosh[(n-\tfrac12\!+p)x]\cosh[(n-\tfrac12)x]}\\[2ex] &=\sum_{n\ge1}\frac1{\cosh[(n-\tfrac12\!+p)x]\cosh[(n-\tfrac12)x]}\\[1ex] &\quad+\sum_{n\ge1}\frac1{\cosh[(-n+\tfrac12\!+p)x]\cosh[(-n+\tfrac12)x]}\\[2ex] &=\sum_{n\ge1}\frac1{\cosh[(n-\tfrac12\!+p)x]\cosh[(n-\tfrac12)x]}\\ &\quad+\sum_{n\ge1}\frac1{\cosh[(n-\tfrac12\!-p)x]\cosh[(n-\tfrac12)x]} \end{align} Then, rewrite each $\cosh$ as its exponential form.... $$\sum_{n\ge1}\frac1{\cosh[(n-\tfrac12\!+p)x]\cosh[(n-\tfrac12)x]}+\sum_{n\ge1}\frac1{\cosh[(n-\tfrac12\!-p)x]\cosh[(n-\tfrac12)x]} $$ \begin{align}&=4\sum_{n\ge1}\frac1{(e^{(n-1/2+p)x}+e^{-(n-1/2+p)x})(e^{(n-1/2)x}+e^{-(n-1/2)x})}\\[1ex] &\quad\;+4\sum_{n\ge1}\frac1{(e^{(n-1/2-p)x}+e^{-(n-1/2-p)x})(e^{(n-1/2)x}+e^{-(n-1/2)x})}\\[2ex] &=4\sum_{n\ge1}\frac{e^{(2n-1)x+px}}{(e^{(2n-1+2p)x}+1)(e^{(2n-1)x}+1)}+4\sum_{n\ge1}\frac{e^{(2n-1)x-px}}{(e^{(2n-1-2p)x}+1)(e^{(2n-1)x}+1)}\\[3ex] &=4\sum_{n\ge1}\frac{e^{(2n-1)x}}{e^{(2n-1)x}+1}\bigg(\frac{e^{px}}{e^{(2n-1)x+2px}+1}+\frac{e^{-px}}{e^{(2n-1)x-2px}+1}\bigg)\\[4ex] &=4\sum_{n\ge1}\frac{e^{px}}{e^{2px}-1}\bigg(\frac1{e^{(2n-1)x}+1}-\frac1{e^{(2n-1)x+2px}+1}\bigg)\\[1ex] &\quad+4\sum_{n\ge1}\frac{e^{-px}}{e^{-2px}-1}\bigg(\frac1{e^{(2n-1)x}+1}-\frac1{e^{(2n-1)x-2px}+1}\bigg)\\[4ex] &=2\operatorname{csch}px\cdot-\sum_{n\ge1}\frac1{e^{(2n-1)x+2px}+1}+2\operatorname{csch}px\sum_{n\ge1}\frac1{e^{(2n-1)x-2px}+1}\tag{*} \end{align} Now take the $p$-log-derivative of $(1)$: \begin{align}\frac{\vartheta'}\vartheta(p,x)=D_p\ln\vartheta(p,x)&=D_p\Big(-p^2x+\sum_{k\ge1}\ln\left(1+e^{2px-(2k-1)x}\right)\Big)+\sum_{k\ge1}\ln\left(1+e^{-2px-(2k-1)x}\right)\Big)\\[2ex] &=-2px+2x\sum_{k\ge1}\left(\frac{e^{2px-(2k-1)x}}{1+e^{2px-(2k-1)x}}-\frac{e^{-2px-(2k-1)x}}{1+e^{-2px-(2k-1)x}}\right)\\[2ex] &=-2px+2x\sum_{k\ge1}\left(\frac1{e^{(2k-1)x-2px}+1}-\frac1{e^{2px+(2k-1)x}+1}\right), \end{align} and finally plug the above result into $(\,^*)$ to get that your sum equals$$\frac2x\operatorname{csch}px\bigg[\frac{\vartheta'}\vartheta(p,x)+2px\bigg].$$ As for your general sum, you could rearrange the summations, then try to evaluate $\sum_{n\in\mathbb{Z}}\operatorname{sech}[x(n-1/2+p_i)]$ (which should have a simple relation to a Jacobi elliptic function); the $p_i, p_j$ terms may force some inevitable complexities later...


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.