The one in question is as followed:

$$ \sum_{n=1}^{\infty} e^{-n}\tan n $$

This comes up in one of those coffee-shop discussion...

  • $\begingroup$ I don't know how to make it rigorous, but I'm inclined to guess it diverges from a probabilistic argument: each congruence class of subintervals of length $\pi/k$ of $(-\pi/2, \pi/2)$ occurs equally often, and so the magnitude of contribution from each is $c \int_a^{a+\pi/k} \tan(t) dt$ for some constant $c$, which goes to $\infty$ as $a \to 0$. $\endgroup$ – David Schneider-Joseph Aug 2 '17 at 3:49
  • $\begingroup$ My guess is that it would converge since $exp(-n)$ gets nicely small while $tan(n)$ depends on how close $n$ is to $(k+1/2)\pi$ for some $k$, and this would not happen too often. $\endgroup$ – marty cohen Aug 2 '17 at 4:10
  • 6
    $\begingroup$ Using the fact that $\pi$ has finite irrationality measure, you can check that $\tan n$ grows at most polynomially fast. So the series converges absolutely. $\endgroup$ – Sangchul Lee Aug 2 '17 at 5:07
  • $\begingroup$ @SangchulLee: Can you please add an answer elaborating it? I am unable to see how. $\endgroup$ – Aryabhata Aug 2 '17 at 5:44

For each $x \in \mathbb{R}$, its irrationality measure $\mu(x)$ is defined as

$$ \mu(x) = \inf \left\{ c \in \mathbb{R} : \left| x - \frac{a}{b} \right| \leq \frac{1}{|b|^c} \text{ for at most finitely many } (a, b) \in \mathbb{Z}\times\mathbb{Z}^{*}\right\}. $$

Let us collect some basic properties of $\mu$:

  1. It is clear that $\mu(x) = 1$ if $x$ is rational. On the other hand, if $x$ is irrational, then by Dirichlet's approximation theorem we have $\mu(x) \geq 2$.

  2. We have $\mu(x) = \mu(1/x)$. Indeed, it suffices to prove this for irrational $x$. To this end, notice that if $0 < c < \mu(x)$, then there exists $(a_j, b_j) \in \mathbb{Z}^* \times \mathbb{Z}^*$ such that $|b_j| \to \infty$ and that $|x - (a_j/b_j)| \leq |b_j|^{-c}$. In particular, $a_j/b_j \to x$. Then it follows that

    $$ \left| \frac{1}{x} - \frac{b_j}{a_j}\right| = \frac{|b_j/a_j|}{|x|} \left| x - \frac{a_j}{b_j} \right| \leq \frac{\text{const}}{|b_j|^c} \leq \frac{\text{const}}{|a_j|^c} \leq \frac{1}{|a_j|^{c-\epsilon}} $$

    if $\epsilon > 0$ and $j$ is sufficiently large. So we have $c-\epsilon \leq \mu(1/x)$ and this is enough to conclude the claim.

  3. It is well-known that $\mu(\pi) < \infty$.

Using these properties, we find that $\mu(1/\pi) < \infty$. So we can pick $c > \mu(1/\pi)$. Then

$$ |\cos n| = \left|\sin\pi\left(\frac{n}{\pi} - \frac{1}{2} - a \right) \right| $$

for any $a \in \mathbb{Z}$. Pick $a$ such that $\left| \frac{n}{\pi} - \frac{1}{2} - a \right| \leq \frac{1}{2}$. Then using the inequality $|\sin(\pi x)| \geq 2|x|$ for $|x| \leq \frac{1}{2}$, for large $n$ we have

$$ |\cos n| \geq |2n|\left|\frac{1}{\pi} - \frac{2a+1}{2n} \right| \geq |2n|\cdot\frac{1}{|2n|^{c}} = \frac{1}{|2n|^{c-1}}. $$

This shows that $|\tan n| \leq |2n|^{c-1}$ for large $n$ and hence the series converges absolutely by comparison test.


This does not prove anything.

Sharing the same opinion as Marty Cohen, I just performed numerical evaluations of $$S_k=\sum_{n=1}^{10^k}e^{-n}\, \tan(n)$$ and obtained the following results $$\left( \begin{array}{cc} k & S_k \\ 1 & 0.2663227174666176381276531 \\ 2 & 0.2625520215463125833311472 \\ 3 & 0.2625520215463125833311472 \\ 4 & 0.2625520215463125833311472 \\ 5 & 0.2625520215463125833311472 \\ 6 & 0.2625520215463125833311472 \end{array} \right)$$

  • $\begingroup$ Using the known bounds for the irrationality measure of $\pi,$ it's possible to prove $|\tan n|\ge n^C$ for sufficiently large $n$ and some $0<C<8$, that's sufficient for convergence. Ah, I see that @Sangchul Lee wrote that, already. $\endgroup$ – Professor Vector Aug 2 '17 at 6:08

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.