I want to find the value to which this series converges $$\sum_{n=0}^\infty \frac{(-1)^n}{n^2+1}$$

I tried looking at the sequence of partial sums $$S_k = \sum_{n=0}^k \frac{(-1)^n}{n^2+1}$$ and I noticed that $$\frac{-1}{n^2+1} \leq \frac{(-1)^n}{n^2+1} \leq \frac{1}{n^2 +1}$$ and so I think that by the squeeze rule I can see ( I could have just noticed it by logic, but okay) that the terms converge to zero. How do I find the value of the original series though? I could only show that it converged

  • $\begingroup$ wolframalpha.com/input/…, has a complicated converged value! $\endgroup$ – rookie Feb 20 '17 at 18:16
  • $\begingroup$ @stud_iisc that's trivial, but to derive a real representation is much more interesting. $\endgroup$ – Simply Beautiful Art Feb 20 '17 at 18:20
  • $\begingroup$ @stud_iisc however I am asked to find the value of it! I tried on wolfram alpha as well, in my computer gave me the answer in terms of cosecants or such things.. it shouldn't be that hard, it's a second year university exercise $\endgroup$ – Euler_Salter Feb 20 '17 at 18:20
  • 5
    $\begingroup$ For example,$$\sum_{n=0}^\infty\frac{(-1)^n}{n^2+1}=\int_0^1\sum_{n=0}^\infty(-x)^{n^2}\ dx$$ $\endgroup$ – Simply Beautiful Art Feb 20 '17 at 18:22
  • 4
    $\begingroup$ If you were asked to give the value then perhaps you've already dealed with series from $\;-\infty\;$ to $\;\infty\;$ , Dirichlet series or stuff like that. Then (obsderve the index!) $$\sum_{n=-\infty}^\infty\frac{(-1)^n}{n^2+1}=\pi\,\text{csch}\,\pi $$ $\endgroup$ – DonAntonio Feb 20 '17 at 18:22

You may notice that: $$ \frac{1}{n^2+1} = \int_{0}^{+\infty}\sin(x)e^{-nx}\,dx\tag{1}$$ from which$^{(*)}$: $$ S=\sum_{n\geq 0}\frac{(-1)^n}{n^2+1}=1+\int_{0}^{+\infty}\sin(x)\sum_{n\geq 1}(-1)^n e^{-nx}\,dx \tag{2}$$ and: $$ S = 1-\int_{0}^{+\infty}\frac{\sin(x)}{e^x+1}\,dx =\color{red}{\frac{1}{2}\left(1+\frac{\pi}{\sinh\pi}\right)}\tag{3}$$ where the last equality follows from integration by parts and the residue theorem. The same can be proved by considering the Fourier cosine series of $\cosh(x)$ over the interval $(-\pi,\pi)$.

Yet another (Eulerian) approach. It is clearly enough to compute $\sum_{n\geq 1}\frac{1}{n^2+1}$ and $\sum_{n\geq 1}\frac{1}{4n^2+1}$.
From the Weierstrass product for the $\sinh$ function we have $$ \frac{\sinh(\pi z)}{\pi z}=\prod_{n\geq 1}\left(1+\frac{z^2}{n^2}\right)\tag{4}$$ and by applying $\frac{d}{dz}\log(\cdot)$ to both sides: $$ -\frac{1}{z}+\pi\coth(\pi z) = \sum_{n\geq 1}\frac{2z}{z^2+n^2}\tag{5}$$ At last we just need to evaluate the LHS of $(5)$ at $z=1$ and $z=\frac{1}{2}$.

$(*)$ The exchange of $\sum$ and $\int$ is allowed by the absolute convergence of the series $\sum_{n\geq 0}\frac{(-1)^n}{n^2+1}$, the trivial inequality $\left|\sin(x)\right|\leq x$ and the dominated convergence theorem. For any $x>0$ we have $$ \sum_{n=1}^{N} e^{-nx}\leq \frac{1}{e^x-1} $$ and $\frac{x}{e^x-1}$ is a function belonging to $\mathcal{L}^1(\mathbb{R}^+)$, whose integral over $\mathbb{R}^+$ equals $\zeta(2)=\frac{\pi^2}{6}$.

  • $\begingroup$ Many ways to write the same thing. Was just in the middle of trying this, though it is the first time I've tackled a series like this. +1 $\endgroup$ – Simply Beautiful Art Feb 20 '17 at 18:46
  • $\begingroup$ The Fourier cosine series comes out to my last comment on the main post, right? $\endgroup$ – Simply Beautiful Art Feb 20 '17 at 18:49
  • $\begingroup$ @SimplyBeautifulArt: I am noticing it just now, but sure, it is a pretty standard technique in this case. $\endgroup$ – Jack D'Aurizio Feb 20 '17 at 19:20
  • 2
    $\begingroup$ (+1) The Eisenstein series for $\cot$ (and friends) deserves to be better-known... it seems to have fallen off the map in recent years! $\endgroup$ – Chappers Feb 21 '17 at 1:18
  • 1
    $\begingroup$ @MartinArgerami I didn't know about the theorem indeed, I do know you always need to be careful from our course in Analysis though. $\endgroup$ – Euler_Salter Feb 21 '17 at 16:09

Extend the sum $S$ to $$ 2S-1 = \sum_{n=-\infty}^{\infty} \frac{(-1)^n}{n^2+1}, $$ which we can get away with because the summand is even in $n$. We want to find something that gives these as residues, then use Cauchy's theorem, hoping the integral goes to zero. Therefore we take $$ \frac{\pi\csc{\pi z}}{z^2+1}: $$ this has poles at $z=n$ with residue $ (-1)^n/(n^2+1) $, and two others, at $\pm i$, with residues $ \frac{\pi \operatorname{csch}{(\pm \pi)}}{\mp 2i} $. Also, if we take the integral to be around a large square that between the poles, one can show using $ \lvert\sin{(x+iy)} \rvert^2 = \sin^2{x}+\sinh^2{y} $ and the corresponding one for cosine that the integrand is bounded on the border of the square by a multiple of $1/(x^2+y^2)$, and hence the integral tends to zero as we make the square larger and larger. Hence $$ \frac{1}{2\pi i}\left( 2S_k-1 - \frac{2\pi\operatorname{csch}{\pi}}{2} \right) = \int_{\square_k} \frac{\pi\csc{\pi z}}{z^2+1} \, dz \to 0 $$ as $k \to \infty$, and then you can rearrange to get the answer.

The general idea is that $\pi\cot{\pi z}$ has residue $1$ at every integer $n$, while $\csc{\pi z}$ has residue $(-1)^n$ at every integer $n$. The same technique works on any sum of the form $1/p(n^2)$ (or $(-1)^n/p(n^2)$) where $p$ is a nonconstant polynomial, although you do have to find the roots.

  • 2
    $\begingroup$ I was just in the middle of preparing this way forward as a solution. (+1) for the faster typing! $\endgroup$ – Mark Viola Feb 20 '17 at 18:44

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.