The p-series convergence test is a classic and well-known result for sums of the form $\sum_{n=1}^{\infty}\frac{1}{n^p}$ for a real number $p$. It is known that $\sum_{n=1}^{\infty}\frac{1}{n}$ diverges, but for every $\epsilon>0$, $\sum_{n=1}^{\infty}\frac{1}{n^{1+\epsilon}}$ converges.

It can be shown that series with terms asymptotically smaller than this will also converge, such as $$\sum_{n=2}^{\infty}\frac{1}{n\log^2n}\text{ and even }\sum_{n=2}^{\infty}\frac{1}{n\log^{1+\epsilon}n}\text{ for }\epsilon>0$$

I was introduced to a related series by a coworker of mine, which is the following: $$\sum_{n=1}^{\infty}\frac{1}{n^{1+\sin n}}$$ Supposedly, he was able to prove that this diverged. A natural generalization is to look at series of the form $$\sum_{n=1}^{\infty}\frac{1}{n^{c+\sin n}}$$ for some $c>0$. It is not hard to show that the series diverges when $c\leq0$ and converges when $c\geq2$. What I want is to find the smallest value of $c$ such that the series converges, or a tight lower bound. Formally, I want to find $$\inf\left\{c\,:\,\sum_{n=1}^{\infty}\frac{1}{n^{c+\sin n}}<\infty\right\}$$ Any progress on finding this number is appreciated. I would assume that it is greater than 1, but I haven't been able to prove much else.

  • 2
    $\begingroup$ In your scond paragraph do you mean "will also converge" or "will also diverge" ? $\endgroup$ – Delta-u Oct 12 '18 at 14:11
  • $\begingroup$ I mean converge, also in addition to $\sum_{n=1}^{\infty}\frac{1}{n^{1+\epsilon}}$. The two I referenced can be shown from the integral test $\endgroup$ – HackerBoss Oct 12 '18 at 14:29
  • $\begingroup$ @Delta-u Good catch though. Just realized I was dividing by zero. I have changed the lower bounds accordingly $\endgroup$ – HackerBoss Oct 12 '18 at 14:44
  • $\begingroup$ Anyway on different note I think that when $c<2$ the series diverges, because $\sin n \in[c-1,1]$ infinitely often and the probability that $\sin n \in [c-1,1]$ is non-zero (though I can't prove this statement). $\endgroup$ – kingW3 Oct 12 '18 at 14:57
  • $\begingroup$ @kingW3 I have just corrected that part of the question again. They actually do converge now $\endgroup$ – HackerBoss Oct 12 '18 at 14:58

The series clearly diverges for any $c<2$. To see this at a glance suppose $c = 2-\epsilon$. The idea is to look at the unit circle and note the proportion of angles for which the number of terms with $c+\sin(n)\leq1$ is finite. $$\begin{equation} \sum_{n=0}^N \frac{1}{n^{c + \sin n}} \geq \sum_{n \in S_N} \frac{1}{n^{2 - \epsilon + \sin n}} \end{equation}$$ where $S_N = \Big\{n: \sin n \leq -1 + \epsilon \ \ \text{and} \ \ 0 \leq n \leq N\Big\}$ and $0<\delta<\epsilon$. Then $|S_N|/N$ goes to $\cos^{-1}(1-\epsilon)$ as $N$ goes to infinity by equidistribution of $n\mod 2\pi$ and the RHS becomes of the same order as $\sum 1/n$.

Edit: $\delta$ was unnecessary

| cite | improve this answer | |
  • 1
    $\begingroup$ Over the real numbers, that would be true. The trick here is that we are looking only at integers. Can you prove that $|S_N|/N\rightarrow(\epsilon - \delta)/2\pi$ as $N\rightarrow\infty$ looking at only integral values of $n$? $\endgroup$ – HackerBoss Oct 12 '18 at 15:10
  • $\begingroup$ Last paragraph is unnecessary. Just refer to equidistribution of $n$ mod $2\pi$. en.wikipedia.org/wiki/Equidistributed_sequence $\endgroup$ – Sungjin Kim Oct 12 '18 at 15:35
  • $\begingroup$ The limit of $|S_N|/N$ is $\frac 1{\pi} \cos^{-1} (1-\epsilon+\delta)$. But this doesn't affect the validity of this argument. $\endgroup$ – Sungjin Kim Oct 12 '18 at 15:53

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.