Attempt: I computed the complex Fourier Series of $f(x) = e^{x + 2πxi}$

$f(x) = ∑_n(e^{2in}−e^2)e^{inx−in−1}/(2π)(in−2iπ−1)$ where $n$ is any integer.

How do I use this to calculate $∑_{n>0}1/(1+π^2n^2)$? I tried equating f(1), f(-1), f(0) but they didn't work.

  • 1
    $\begingroup$ Is it $\sum_{k=1}^{\color{red}\infty}\frac{(-1)^\color{red}{n}}{1+k^2 \pi^2}$ ? $\endgroup$ – Claude Leibovici Jul 23 '17 at 4:12
  • $\begingroup$ A bounty for a question already solved multiple times on the site (and in every textbook on the subject)? Hmmm... $\endgroup$ – Did Jul 25 '17 at 16:53
  • 3
    $\begingroup$ @Did Don't worry, he accepted his own answer. $\endgroup$ – adfriedman Jul 25 '17 at 19:38
  • $\begingroup$ @adfriedman ...Which does not even solve the question (as you noted). Strange character, I must admit. $\endgroup$ – Did Jul 25 '17 at 19:44

Using $f(x)=\cosh(x/\pi)$, the Fourier coefficients are given by

\begin{align} a_n &= \frac{1}{\pi}\int_{-\pi}^{\pi} \cosh(x/\pi) \cos(n x)\,dx = 2 \sinh(1) \frac{(-1)^n}{1+\pi^2n^2}\\ b_n &= \frac{1}{\pi}\int_{-\pi}^{\pi} \cosh(x/\pi) \sin(n x)\,dx = 0 \end{align}

So on $[-\pi,\pi]$, \begin{align} f(x) = \cosh(x/\pi)&= \frac{1}{2}a_0 + \sum_{n=1}^{\infty} a_n \cos(n x) + \sum_{n=1}^{\infty} b_n \sin(n x) \\ &= \sinh(1) + 2\sinh(1)\sum_{n=1}^{\infty} \frac{(-1)^n}{1+\pi^2n^2}\cos(nx) \end{align} at $x=0$ and with some rearranging we have

\begin{align} \sum_{n=1}^{\infty} \frac{(-1)^n}{1+\pi^2n^2} &= \frac{\cosh(0) - \sinh(1)}{2\sinh(1)} = \frac{1 - \sinh(1)}{2\sinh(1)}\\ &= \frac{1}{2}\big( \mathrm{csch}(1) - 1\big) \end{align} giving $$\boxed{\sum_{n=1}^{\infty} \frac{(-1)^n}{1+\pi^2n^2} = \frac{1}{2}\left( \frac{1+2e-e^2}{e^2-1}\right)}$$

Motivation As to why I chose $\cosh(x/\pi)$, I recalled that the Fourier coefficients for $e^x$ were something like ''$\frac{\cos(n\pi)}{1+n^2}$'', so it made sense to try and compute coefficients after some kind of scaling. $e^{x/\pi}$ turned out to work quite well for the cosine terms, but not for the sine terms. The same turned out to be true of $e^{-x/\pi}$, so the hope was that some linear combination of the two would cancel out all of the sine terms. I took the two most obvious linear combinations, $\sinh(x/\pi)$ and $\cosh(x/\pi)$, where the latter turned out to work.

Update To address the change of question description, we will solve the more general form first. We just established that for $x\in [-\pi,\pi]$,

$$\cosh(x/\pi) = \sinh(1) + 2\sinh(1) \sum_{n=1}^{\infty} \frac{(-1)^n}{1+\pi^2n^2}\cos(nx)$$ which may be arranged to \begin{align} \sum_{n=1}^{\infty} \frac{(-1)^n}{1+\pi^2n^2}\cos(nx) &= \frac{1}{2}\left(\frac{\cosh(x/\pi)}{\sinh(1)} - 1\right) \end{align}

Of course $\cos(n\pi)=(-1)^n$, so evaluating at $x=\pi$ gives $$\sum_{n=1}^{\infty} \frac{1}{1+\pi^2n^2} = \sum_{n=1}^{\infty} \frac{(-1)^n\cos(n\pi)}{1+\pi^2n^2} = \frac{1}{2}\left(\frac{\cosh(\pi/\pi)}{\sinh(1)} - 1\right) = \frac{1}{2}\big(\coth(1) - 1\big)$$ after expanding the definition of $\coth(1)$ and a bit of algebra you get $$\boxed{\sum_{n=1}^{\infty} \frac{1}{1+\pi^2n^2} = \frac{1}{e^2-1}}$$

  • $\begingroup$ The OP has changed the question a bit: the $(-1)^n$ has been replaced by $1$. Why not modify your answer accordingly and get the bounty? $\endgroup$ – Alex M. Jul 29 '17 at 9:46
  • $\begingroup$ @AlexM. I guess he did. I have updated accordingly $\endgroup$ – adfriedman Jul 30 '17 at 10:28
  • 2
    $\begingroup$ @Jeffrey, I strongly suggest that you unaccept your own answer (and delete it, to recover some reputation points), and accept this one instead. $\endgroup$ – Alex M. Jul 30 '17 at 11:41
  • $\begingroup$ Good answer although it was chaotic question. (+1) $\endgroup$ – user90369 Aug 1 '17 at 9:39

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.