8

While studying Fourier analysis last semester, I saw an interesting identity:

$$\sum_{n=1}^{\infty}\frac{1}{n^2-\alpha^2}=\frac{1}{2\alpha^2}-\frac{\pi}{2\alpha\tan\pi\alpha}$$ whenever $\alpha \in \mathbb{C}\setminus \mathbb{Z}$, which I learned two proofs using Fourier series and residue calculus.

More explicitly, we can deduce the theorem using Fourier series of $f(\theta)=e^{i(\pi - \theta)\alpha}$ on $[0,2\pi]$ or contour integral of the function $g(z)=\frac{\pi}{(z^2-\alpha^2)\tan\pi z}$ along large circles.

But these techniques, as long as I know, wasn't fully developed at Euler's time.

So what was Euler's method to prove this identity? Is there any proof at elementary level?

Mark Viola
  • 179,405
HyJu
  • 701

2 Answers2

12

According to Wikipedia (https://en.wikipedia.org/wiki/Basel_problem), Euler was the first to give a representation of the sine function as an infinite product: $$(*) \hspace{2cm}\sin (\pi \alpha)=\pi \alpha \prod\limits_{n=1}^{\infty}(\frac{n^2-\alpha^2}{n^2}),$$ which was formally proved by Weierstrass about 100 years later.

Now taking "$\ln$" on by sides of (*) gives $$\ln(\sin (\pi \alpha))=\ln(\pi \alpha)+ \sum \limits_{n=1}^{\infty} \ln (\frac{n^2-\alpha^2}{n^2}),$$ and after taking derivatives on both sides we arrive at

$$\sum_{n=1}^{\infty}\frac{1}{n^2-\alpha^2}=\frac{1}{2\alpha^2}-\frac{\pi}{2\alpha\tan\pi\alpha}.$$

math
  • 1,664
  • This sounds really reasonable, even though the completion of this idea was a later development. Thanks! – HyJu Jan 22 '17 at 05:30
  • 1
    @HyJu: Note that Euler's reasoning was probably seriously flawed, but he just was 'lucky'. Even under nice conditions, namely that the zeros are isolated and simple, the zeros only define an analytic function up to a factor of $\exp(g(z))$ for some analytic function $g$. It just happens that $\sin(πz)$ is the product of the linear factors needed to reproduce the zeros, with the $\exp(g(z))$ factor being constant. – user21820 Jan 22 '17 at 11:49
  • @HyJu: See https://en.wikipedia.org/wiki/Weierstrass_factorization_theorem for more. – user21820 Jan 22 '17 at 11:52
  • @user21820 Actually I'm studying that part now. Thanks. – HyJu Jan 22 '17 at 12:00
4

For the partial fraction decomposition of the cotangent $$\pi \cot \pi z = \frac{1}{z} + \sum_{n\ge 1}\left( \frac{1}{z-n} + \frac{1}{z+n}\right) = \lim_{k\to\infty} \sum_{n=-k}^k \frac{1}{z-n}$$ hence \begin{align*} \sum_{n=1}^\infty \frac{1}{n^2-\alpha ^2} &= \lim_{k\to\infty}\sum_{n=1}^k \frac{1}{n^2-\alpha^2}\\ &= -\frac{1}{2\alpha}\lim_{k\to\infty} \sum_{n=1}^k \frac{1}{\alpha-n} + \frac{1}{\alpha+n}\\ &= -\frac{1}{2\alpha}\lim_{k\to\infty} \left(-\frac{1}{\alpha} +\sum_{n=-k}^k \frac{1}{\alpha-n}\right)\\ &= -\frac{1}{2\alpha}\left(\pi \cot \pi \alpha - \frac{1}{\alpha}\right)\\ &= \frac{1-\pi \alpha\cot \pi \alpha}{2\alpha^2}\\ &=\frac{1}{2\alpha^2}-\frac{\pi}{2\alpha\tan\pi\alpha} \end{align*}

Mark Viola
  • 179,405
Renascence_5.
  • 5,281
  • 1
  • 27
  • 69
  • For me the partial fraction decomposition seems already using some aspect of complex analysis. Thanks. – HyJu Jan 22 '17 at 05:31
  • @HyJu The partial fraction representation can be derived directly from the infinite product representation of the sinc function. – Mark Viola Jan 22 '17 at 06:50
  • This development seems completely circular. Starting from the partial fraction representation, one can immediately deduce the coveted identity by simply writing $\cot(\pi \alpha)=\frac{1}{\tan(\pi \alpha)}$ and writing $\frac{1}{z-n}+\frac{1}{z+n}=\frac{2z}{z^2-n^2}$. – Mark Viola Jan 22 '17 at 06:54
  • 1
    I have a strong suspicion Euler didn't know of the partial fraction decomposition of cotangent, at least not before he has already solved Basel problem. – Wojowu Jan 22 '17 at 09:27