Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I ran across another challenging and interesting series, and I am wondering if someone could shed some light on how to evaluate it.

$$ \sum_{n=1}^{\infty}\sum_{k=1}^{\infty}\frac{1}{(n^{2}+k^{2})^{2}}=\zeta(2)\sum_{n=1}^{\infty}\frac{(-1)^{n-1}}{(2n-1)^{2}}-\zeta(4)$$

This has turned out to be rather challenging. At first glance, I thought it may be somewhere along the lines of the famous $$ \sum_{n=1}^{\infty}\frac{1}{n^{2}+k^{2}}=\frac{\pi}{2k}\coth(\pi k)-\frac{1}{2k^{2}}$$ that is often seen in Complex Analysis.

So, I ran the first sum through Maple and it gave me:

$$ \sum_{n=1}^{\infty} \sum_{k=1}^{\infty}\frac{1}{(n^{2}+k^{2})^{2}}=\frac{{\pi}^{2}}{4}\sum_{n=1}^{\infty}\frac{\coth^2(\pi n)}{n^{2}}+\frac{\pi}{4}\sum_{n=1}^{\infty}\frac{\coth(\pi n)}{n^{3}}-\frac{{\pi}^{2}}{4}\sum_{n=1}^{\infty}\frac{1}{n^{2}}-\frac{1}{2}\sum_{n=1}^{\infty}\frac{1}{n^{4}}$$

Of course, two of these are the very familiar $\zeta(2), \;\ \zeta(4)$. I managed to evaluate $\displaystyle \sum_{n=1}^{\infty}\frac{\coth(\pi n)}{n^{3}}=\frac{7{\pi}^{3}}{180}$ using Complex Analysis.

The one that has given me the fit is $\displaystyle\sum_{n=1}^{\infty}\frac{\coth^{2}(\pi n)}{n^{2}}$.

This evaluates to $\displaystyle\frac{2}{3}K+\frac{19{\pi}^{2}}{180}$. But, how?.

I tried a Complex Analysis method, but had trouble finding the residues at $ni$, which are the zeroes of $\sinh^{2}(\pi n)$ .

Anyone have any good ideas on how to evaluate the original sum at the top or even just this 'coth-squared' one?. Complex analysis or other wise. I thought maybe a clever use of Fourier would work, but maybe not.

share|improve this question
1  
It wasn't until I got to the end of the question that I realised "CA" meant "complex analysis"... also I personally think it's unhelpful to put "interesting" in your title – I already assume that your question is interesting, else why would you post it? –  Ben Millwood Sep 16 '12 at 11:48
    
Ha. I thought CA meant Computer Algebra. –  Gerry Myerson Sep 16 '12 at 11:51
    
    
Sorry, I did not realize that was such an issue. –  Cody Sep 16 '12 at 15:55

2 Answers 2

A more general form is known : $$\tag{1}f(c)=\sum_{n,m}'\frac 1{(n^2+m^2)^c}=4\,\zeta(c)\,L_{-4}(c)$$ with $\ L_{\sigma}(c):=\sum_{n\ge 1} \,\left(\frac {\sigma}n\right)\,n^{-c}\quad$ (where $\ \displaystyle\left(\frac {\sigma}n\right)$ is the Kronecker symbol and the sum is on all signed integers $n,m$ excluding $n=m=0$ as indicated by ' )

For $\;\sigma=-4\ $ only the odd terms remain with alternating signs : $$\tag{2}L_{-4}(c)=\sum_{n\ge 1} \,\left(\frac {-4}n\right)\,n^{-c}=\sum_{n\ge 1} \frac {(-1)^{n-1}}{(2n-1)^c}=\beta(c)$$ with $\beta$ the Dirichlet beta function so that (distinguishing the values $n=0$ and $m=0$) : $$4\,\zeta(c)\,\beta(c)=\sum_{n,m}'\frac 1{(n^2+m^2)^c}=2\sum_{m\ge 1}\frac 1{(0^2+m^2)^c}+2\sum_{n\ge 1}\frac 1{(n^2+0^2)^c}+4\sum_{n,m\ge 1}\frac 1{(n^2+m^2)^c}$$ and (dividing by $4$) a generalization of your nice result : $$\tag{3}\sum_{n,m\ge 1}\frac 1{(n^2+m^2)^c}=\,\zeta(c)\,\beta(c)-\zeta(2c)$$


The (rather non-trivial) derivation of a further generalization of $(1)$ is provided in Borwein, Bailey and Girgensohn's "Experimentation in Mathematics" (page $167-169$ of my $2004$ edition).

They define :

$$\tag{4}\zeta(a,b,c) = \sum_{n,m}' \frac{n^{2a}\,m^{2b}}{(n^2+m^2)^c}$$

($a,\;b,\;c\;$ are nonnegative integers)
(the sum was for $n,m\ge 0$ but I think that "$\ge 0$" is a typo that doesn't appear page $169$)

The page $168$ of the book contains (I hope that a one page re-transcription will not be a problem, extracts of the book may be viewed at Amazon) :

"The following identity expresses $\zeta(a,0,c)$ by using a Bessel function expansion of the normalized Mellin transform $$M_c(f) = \frac 1{\Gamma(c)} \int_0^\infty f(t)\,t^{c-1}\,dt$$ of the function $$t \to \sum_{n,m} n^{2a}q^{n^2+m^2} =\sum_n n^{2a}q^{n^2} \theta_3(q)$$ with $q=e^{-t}$ after using the theta transform $(2.15)$ $$\theta_3\bigl(e^{\pi t}\bigr)= \sqrt{\frac{\pi}t}\,\theta_3\left(e^{\frac{\pi}t}\right)$$ This leads to the identity $$\tag{5}\zeta(a,0,c)= 2\delta_{0a}\zeta(2c)+ 2\beta\left(c-\frac12,\frac12\right) \zeta(2c-2a-1)+4\sum_{p\ge 1} \sigma_{[2a+1-2c]}(p)\,{\cal E}_c(p)$$ This is valid for real $ac$ with $d=c-a>1$ and presumably provides an analytic continuation of $\zeta(a,0,c)$ for the sum over positive integers. Here $\sigma$ is a divisor function $$\sigma_{[d]}(p) =\sum_{n|p} n^d,$$ and $${\cal E}_c(p)= \frac{\sqrt{\pi}}{\Gamma(c)} 2(\pi p)^{c-1/2}K_{\bigl(c-\frac 12\bigr)}(2\pi p)$$ is derived from the modified Bessel function of half integer order, $K_{\bigl(c-\frac 12\bigr)}$, of the second kind.

When $c=N$ is integer, ${\cal E}_N(p)$ is of the form $$\pi\,e^{-2\pi p} P_N(\pi p)$$ where $P_N$ is a rational polynomial of degree $N-1$ with positive coefficients : $P_1(x)=1,\ P_2(x)=x+1/2,\ P_3(x)=x^2/2+3/4x+3/8,\ P_4(x)=1/6x^3+1/2x^2+5/8x+5/16$.

In general $$P_N(x)=\sum _{k=0}^{N-1}\frac{{{N+k-1}\choose{N-1}}x^{N-1-k}}{(N-k-1)!4^k}$$ This allows one to very efficiently compute $\zeta(a,0,c)$ via $(5)$, using roughly $D/4$ terms for $D$ digits. Note also that for fixed $c$ and variable $a$ only the powers in $\sigma_{[2a+1-2c]}$ vary so most of the computation can be saved.

Then the general integer case follows from $$\tag{6}\zeta(a,b,c)=\sum_{k=0}^e (-1)^{e-k}{{e}\choose{k}}\zeta(a+b-k,0,c-k)$$ where $\ e=\min(a,b)$."

The formula at the beginning was $\zeta(0,0,c)$ and you asked merely for $\zeta(0,0,2)$. I think there is a much simpler answer for the second one (so that the problem remains...) but generalizations are of value too!


Let's add some keywords and references I just found (MathWorld and others) :

Hoping this helped too,

share|improve this answer
    
Wow, thanks Raymond. Yes, help is always appreciated. I must admit, I am not familiar with most of this. I will have to look into it all further. I managed to evaluate the series by summing each of the four series I mentioned in my first post. The tricky one was the $\frac{\coth^{2}(n\pi)}{n^{2}}$. Thanks much. You are always helpful. –  Cody Sep 18 '12 at 18:34
    
Glad it helped @Cody ! For other generalizations see page $19$ of this. Page $25$ has even a sum with three squares as well as page $378$ of this paper ! For theta and Bessel functions DLMF may help. –  Raymond Manzoni Sep 18 '12 at 22:53
    
Thanks for the edit @Lucian! (concerning ^' I think it worked until around one and half year earlier... possibly a Mathjax evolution since I still use the same old safari browser). Cheers, –  Raymond Manzoni yesterday

I managed to evaluate the series by beginning with:

$\displaystyle \sum_{k=1}^{\infty}\frac{1}{n^{2}+k^{2}}=\frac{\pi n\coth(\pi n)-1}{2n^{2}}$, and using some known sums/identities.

Then, differentiate, do some manipulating and it works out to what Maple gave.

Then, sum these four series separately.

It works out to $ \frac{-{\pi}^{2}}{4}\zeta(2)+\frac{{\pi}^{2}}{4}\left(\frac{2}{3}K+\frac{19{\pi}^{2}}{180}\right)-\frac{1}{2}\zeta(4)+\frac{\pi}{4}\cdot \frac{7{\pi}^{3}}{180}$

This then simplifies down to $\zeta(2)K-\zeta(4)$. K being the Catalan constant represented by the series in the problem.

The trickiest sum was $\displaystyle \sum_{n=1}^{\infty}\frac{\coth^{2}(\pi n)}{n^{2}}$.

$=\displaystyle\sum_{n=1}^{\infty}\frac{csch^{2}(\pi n)}{n^{2}}+\sum_{n=1}^{\infty}\frac{1}{n^{2}}$

$=\displaystyle\frac{2}{3}K-\frac{11{\pi}^{2}}{180}+\frac{{\pi}^{2}}{6}=\frac{2}{3}K+\frac{19{\pi}^{2}}{180}$

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.