3

Consider the next statement:

Let $(u_n)_{n \in \mathbb{N}^*}$ be defined implicitely by: $$u_n = \tan(u_n) \\ -\frac{\pi}{2}+n\pi\leq u_n\leq\frac{\pi}{2}+n\pi$$ Then: $$\sum_{n=1}^{\infty} \frac{1}{u_n^2}=\frac{1}{10}$$

I found this statement in an undergraduate exam, but I can't find it anymore (I only have the result in mind). I have no clue on how to tackle this problem, especially how to get such an intriguing result.

More generally, are there somewhat general methods/theorems to compute series whose terms are defined implicitely?

KRPO
  • 33
  • 1
    It requires a bit of knowledge on complex analysis, but the upshot is that we have $$\prod_{n=1}^{\infty}\left(1-\frac{z}{u_n^2}\right)=\frac{3}{z}\left(\frac{\sin\sqrt{z}}{\sqrt{z}}-\cos\sqrt{z}\right)=1-\frac{z}{10}+\cdots.$$ So by comparing the coefficient of $z$ in both sides, we get the desired sum. – Sangchul Lee Dec 21 '19 at 02:43
  • @aziiri It does indeed! – KRPO Dec 21 '19 at 03:28

1 Answers1

2

The problem reminds me of Euler's (very famous) heuristic proof of the Basel problem, as described here. Inspired by that, we can concoct a similar heuristic explanation for this problem. In fact, our heuristic will be detailed enough that it actually yields a rigorous proof using the Weierstrass factorization theorem.

First, rewrite the fixed point equation $x=\tan x$ as $$\frac{\sin x}{x}=\cos x,$$ which (a la Euler) can be written as a product: $$ \prod_{n=1}^{\infty}\left(1-\frac{x^2}{n^2\pi^2}\right)=\prod_{n=1}^{\infty}\left(1-\frac{x^2}{(n-\tfrac12)^2\pi^2}\right). $$ Now (just as if these were polynomials) we want to move everything over to the left side and factor. When we consider $$ \prod_{n=1}^{\infty}\left(1-\frac{x^2}{n^2\pi^2}\right)-\prod_{n=1}^{\infty}\left(1-\frac{x^2}{(n-\tfrac12)^2\pi^2}\right), $$ we see that the constant term $1$ cancels out, and there is no coefficient of $x$ (since all powers of $x$ appearing are even) so that leaves the $x^2$ coefficient: $$ \sum_{n=1}^{\infty}\frac{1}{(n-\tfrac12)^2\pi^2}-\sum_{n=1}^{\infty}\frac{1}{n^2\pi^2}=\frac{1}{2}-\frac{1}{6}=\frac{1}{3}, $$ where we calculated sum involving $n-\tfrac12$ from the sum involving $n$ by thinking of it as $4$ times the odd terms in the $n$ sum, and calculated the odd sum as the whole sum minus the even terms (which contribute $\tfrac14$ times the whole sum). (We will use this trick once more at the very end...)

Therefore we have the factorization $$\prod_{n=1}^{\infty}\left(1-\frac{x^2}{n^2\pi^2}\right)-\prod_{n=1}^{\infty}\left(1-\frac{x^2}{(n-\tfrac12)^2\pi^2}\right)=\frac{x^2}{3}\prod_{n=1}^{\infty}\left(1-\frac{x^2}{u_n^2}\right).$$ Thus we will be able to compute $\sum_{n=1}^{\infty}\frac{1}{u_n^2}$ by computing the $x^{4}$ coefficient of the left side.

Therefore we find that $$ \frac{1}{3}\sum_{n=1}^{\infty}\frac{1}{u_n^2}=\sum_{1\leq n<m}\left[\frac{1}{n^2m^2\pi^4}-\frac{1}{(n-\tfrac12)^2(m-\tfrac12)^2\pi^4}\right]. $$ Since in general $$ \sum_{n<m}a_na_m=\frac{\Bigl(\sum_{n}a_n\Bigr)^2-\sum_n a_n^2}{2}, $$ we can apply this trick separately to both terms to write them in terms of $$\sum_{n=1}^{\infty}\frac{1}{n^4\pi^4}=\frac{1}{90}\qquad\text{and}\qquad \sum_{n=1}^{\infty}\frac{1}{(n-\tfrac12)^4\pi^4}=\frac{1}{6},$$ (where again we get the $n-\tfrac12$ sum from the $n$ sum using the $\textrm{odd}=\textrm{all}-\textrm{even}$ trick) and after some straightforward manipulations we obtain that $$ \sum_{n=1}^{\infty}\frac{1}{u_n^2}=\frac{1}{10}. $$

pre-kidney
  • 30,884
  • It reminds me of what I read on Stein&Shakarchi's Complex Analysis. Just as Euler's approach, this one might be justified using functions on the complex plane, I guess. – Hyeongmuk LIM Dec 21 '19 at 03:13
  • Note, I seem to have a numerical mistake in my answer that I didn't catch yet, since the explicit integer sum at the end doesn't equal what it is supposed to equal. Rather than delete the answer, I will leave it up in the interim since it is still the correct idea. – pre-kidney Dec 21 '19 at 03:18
  • @Seawatcher the wikipedia page I linked to gives three different approaches for making the proof rigorous, if you scroll to the bottom: https://en.wikipedia.org/wiki/Basel_problem#A_rigorous_proof_using_Euler's_formula_and_L'H%C3%B4pital's_rule – pre-kidney Dec 21 '19 at 03:19
  • Looks like there is more than a numerical issue. My answer would compute $\sum_{n=-\infty}^{n=\infty}\frac{1}{u_n^2}$ but the question asks for $\sum_{n=1}^{\infty}$ instead, since I am not limiting which roots to sum over. – pre-kidney Dec 21 '19 at 03:39
  • @KRPO I have fixed my answer. – pre-kidney Dec 21 '19 at 04:15