I want to evaluate the double sum $\displaystyle \sum_{m=1}^{\infty}\sum_{n=0}^{m-1}\frac{(-1)^{m-n}}{(m^2-n^2)^2} $ where I know that the value is $-\frac{17\pi^4}{1440}$ , I have failed to evaluate to the inner finite sum cleverly so as to transform it into a reasonable infinite series.

  • $\begingroup$ Maybe show what you did, we may help you better :) $\endgroup$ – ParaH2 Aug 14 '16 at 19:00
  • $\begingroup$ Hint: sort by the value of $k=m^2-n^2$. Given such a $k$, the number of times $k$ appears in the sum is the number of factorizations of the type $k=(m+n)(m-n)$, that is, the number of factorizations of $k$ into two factors of the same parity. That should transform the problem into one involving values of (cousins of) the zeta function. $\endgroup$ – Greg Martin Aug 14 '16 at 19:09
  • $\begingroup$ I understand what you are aiming at, had it been $ m^2+n^2$ , the sum of squares function would have been useful then transforming it into a analytical NT problem. $\endgroup$ – Aditya Narayan Sharma Aug 14 '16 at 19:13
  • $\begingroup$ My first thought would be to factorize the denominator into $(m-n)^2(m+n)^2$, but then it's not entirely clear (to me) how to proceed from there. $\endgroup$ – Clement C. Aug 14 '16 at 19:20
  • 2
    $\begingroup$ Possibly related: math.stackexchange.com/questions/1294185/… $\endgroup$ – nospoon Aug 14 '16 at 19:24

Hint. One may observe that, for $m^2\neq n^2$, we have $$ \frac1{(m^2-n^2)^2}=\int_0^1\int_0^1x^{m+n-1}y^{m-n-1}\log x \log y\:dxdy \tag1 $$ giving, by using absolute convergence, $$ \sum_{m=1}^{\infty}\sum_{n=0}^{m-1}\frac{(-1)^{m-n}}{(m^2-n^2)^2}=\int_0^1\!\!\int_0^1\frac{-\log x \log y}{(1-x^2)(1+xy)}\:dxdy. \tag2 $$ Then, by partial fraction decomposition, one may write

$$ \begin{align} &\sum_{m=1}^{\infty}\sum_{n=0}^{m-1}\frac{(-1)^{m-n}}{(m^2-n^2)^2} \\\\&=\int_0^1\!\!\int_0^1\frac{-\log x \log y}{(1-x^2)(1+xy)}\:dxdy \\\\&=\int_0^1\!\!\int_0^1\left(\frac{-\log x \log y}{2 (1+x) (1-y)}+\frac{-\log x \log y}{2 (1-x)(1+y)}+\frac{y^2\log x \log y}{(1-y^2) (1+x y)}\right)\!dxdy \tag3 \\\\&=-2\cdot \frac{\pi^4}{144}+\frac{\pi^4}{480} \\\\&=-\frac{17\pi^4}{1440} \end{align} $$ as announced.

  • 1
    $\begingroup$ What a beautiful answer! $\endgroup$ – Omar Nagib Aug 14 '16 at 22:30
  • $\begingroup$ @Omar Nagib Thank you very much. $\endgroup$ – Olivier Oloa Aug 15 '16 at 0:33
  • 1
    $\begingroup$ Amazing stuff !!!! , Thanks ! $\endgroup$ – Aditya Narayan Sharma Aug 15 '16 at 5:40
  • 1
    $\begingroup$ @OlivierOloa: Nice and concise! (+1) $\endgroup$ – Markus Scheuer Aug 15 '16 at 6:36

Let $k\in\mathbb{N}$. The function $\tau:\mathbb{N}\to\mathbb{N}$ is the divisor-counting function. For an odd value of $k$, there are $\left\lceil\frac{1}{2}\tau(k)\right\rceil$ pairs $(m,n)$ with $m\in\mathbb{N}$ and $n\in\mathbb{N}_0$ such that $m>n$ and $k=m^2-n^2$, whereas there are $\left\lceil\frac{1}{2}\tau\left(\frac{k}{4}\right)\right\rceil$ such pairs if $k$ is divisible by $4$. If $k$ is even but not divisible by $4$, then there are no such pairs $(m,n)$.

The required sum is then $$S:=\sum_{m=1}^\infty\,\sum_{n=0}^{m-1}\,\frac{(-1)^{m-n}}{\left(m^2-n^2\right)^2}=\sum_{k\equiv0\!\pmod{4}}\,\frac{\left\lceil\frac{1}{2}\tau\left(\frac{k}{4}\right)\right\rceil}{k^2}-\sum_{k\text{ odd}}\,\frac{\left\lceil\frac{1}{2}\tau(k)\right\rceil}{k^2}\,.$$ Now, by handling perfect-square values of $k$, we see that $$S=\frac{1}{2}\,\sum_{k\equiv0\!\pmod{4}}\,\frac{\tau(k/4)}{k^2}-\frac{1}{2}\sum_{k\text{ odd}}\,\frac{\tau(k)}{k^2}+\frac{1}{2}\,\sum_{j=1}^\infty\,\frac{(-1)^j}{j^4}\,.$$ Hence, with $j:=\frac{k}{4}$ in the first sum, we have $$S=\frac{1}{32}\,\sum_{j=1}^\infty\,\frac{\tau(j)}{j^2}-\frac{1}{2}\,\sum_{k\text{ odd}}\,\frac{\tau(k)}{k^2}-\frac{1}{2}\,\eta(4)\,,$$ where $\eta$ is the Dirichlet eta function. Note that $$\eta(4)=\left(1-\frac{1}{2^3}\right)\,\zeta(4)=\frac{7}{8}\left(\frac{\pi^4}{90}\right)=\frac{7\pi^4}{720}\,,$$ where $\zeta$ is the Riemann zeta function.

Notice that $$\sum_{j=1}^\infty\,\frac{\tau(j)}{j^2}=\prod_{p\text{ prime}}\,\sum_{r=0}^\infty\,\frac{r+1}{p^{2r}}=\prod_{p\text{ prime}}\,\frac{p^4}{\left(p^2-1\right)^2}$$ and that $$\sum_{k\text{ odd}}\,\frac{\tau(k)}{k^2}=\prod_{p\text{ odd prime}}\,\sum_{r=0}^\infty\,\frac{r+1}{p^{2r}}=\prod_{p\text{ odd prime}}\,\frac{p^4}{\left(p^2-1\right)^2}\,,$$ whence $$\sum_{k\text{ odd}}\,\frac{\tau(k)}{k^2}=\frac{9}{16}\,\sum_{j=1}^\infty\,\frac{\tau(j)}{j^2}\,.$$ Now, observe that $$\sum_{j=1}^\infty\,\frac{\tau(j)}{j^2}=\left(\prod_{p\text{ prime}}\,\frac{1}{1-p^{-2}}\right)^2=\big(\zeta(2)\big)^2=\left(\frac{\pi^2}{6}\right)=\frac{\pi^4}{36}\,.$$ Combining all the pieces, we have $$S=\frac{1}{32}\left(\frac{\pi^4}{36}\right)-\frac{1}{2}\left(\frac{9}{16}\right)\left(\frac{\pi^4}{36}\right)-\frac{1}{2}\left(\frac{7\pi^4}{720}\right)=-\frac{17\pi^4}{1440}\,.$$

P.S. We can show that $$\sum_{j=1}^\infty\,\frac{\tau(j)}{j^s}=\big(\zeta(s)\big)^2$$ for all $s\in\mathbb{C}$ with $\text{Re}(s)>1$.

  • $\begingroup$ (+1) So this is the NT method for this sum, Nice observation and thanks $\endgroup$ – Aditya Narayan Sharma Aug 15 '16 at 5:41

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.