While answering to the following question

Problem. Find the range of $x$ for which $\displaystyle \sum_{n=0}^{\infty} (-1)^{n} \sin(x^{-n})$ converges.

Unlike some careless answerers who merely claimed that this converges only for $|x| > 1$, I faced a hard question:

Question. Is there any $|A| > 1$ such that $\sin (A^{n})$ converges? (Of course, I put $A = x^{-1}$ here.)

I literally have no idea on it since the behavior of the sequence depends on $n \mapsto A^{n} \text{ mod } 2\pi$ and I have no idea how it behaves for values of $A$.

Actually I suspect that there is a number $A > 1$ for which $\sin (A^{n})$ converges. But this is just a groundless guess, anyway.

Do you have any idea regarding this one?

  • $\begingroup$ Do you have to answer your question to solve the problem? or you already have solution to the problem, but just curious about the question? $\endgroup$ – Sungjin Kim Feb 18 '14 at 7:08
  • $\begingroup$ @i707107 As you see, my question is logically just a necessary condition for the problem. I failed to manage both of them. In fact, this is a problem I saw in some QnA site, and none gave a satisfactory answer. I'm sure that even the one who made this question was not aware of this delicacy. $\endgroup$ – Sangchul Lee Feb 18 '14 at 7:47

It can be shown that $$ \mathcal{D}=\{|A|>1 : \sin(A^n) \textrm{ diverges.}\} $$ is dense. (I conjecture that $\mathcal{C}=\{|A|>1: \sin(A^n) \textrm{ converges.}\} $ has measure zero, do not know how to deal with complex exponentials ---> now I do know.)

This follows from a result on equidistribution:

For real numbers $\theta>1$, is is shown in Koksma's paper that $\theta^n$ is equidistributed modulo $1$ for almost all $\theta>1$. Equidistribution modulo $2\pi$ is just a minor modification of Satz 2 in the paper.

Then the density of $\mathcal{D}$ follows from considering $A=re^{i\pi\frac{k}{m}}$ for $k,m$ fixed integers. In fact $A^{2mn}=(r^{2m})^n$, and for almost all $r^{2m}>1$, $(r^{2m})^n$ is equidistributed modulo $2\pi$.

A stronger result of the above(in terms of error term) is shown in Leveque's paper.

Edit: Proof of my conjecture

Let $A=re^{i \theta}$, so $A^n=r^ne^{in\theta}$. Suppose that $\theta\neq k\pi$ for any integer $k$. Then there is a subsequence $n_j$ for $j=1,2,\cdots$ such that $$\sin n_j\theta >\frac 12 \ \ \textrm{ for all }j. $$

Now, $$\sin A^n = \frac{e^{-r^n\sin n\theta}e^{i r^n\cos n\theta}-e^{r^n\sin n\theta}e^{-i r^n\cos n\theta}}{2i}$$

Since $|v-w|\geq ||v|-|w||$ for any complex $v, w$,

$$|\sin A^{n_j}|\geq \frac{1}{2} \left|e^{r^{n_j}\sin n_j\theta}-e^{-r^{n_j}\sin n_j\theta}\right| $$

Since $r>1$, the above $\rightarrow\infty$ as $j\rightarrow\infty$.

Hence, if $A=re^{i\theta}$, with $\theta \neq\pi \textrm{ mod }2\pi$, then the sequence $\sin A^n$ diverges. This proves my conjecture. $$\mathcal{C}\textrm{ has measure zero.}$$

Further, we see that $\mathcal{C}\subset \mathbb{R}$, and has measure zero for the Lebesgue measure on $\mathbb{R}$.

Edit2 I finally found references that partially settles your problem.

Pisot and Kawapisz

We have (1) from Pisot, (2) from Kawapisz.

(1) If a real number $\lambda$ is transcendental, then the series $$ \sum_{n=0}^{\infty} \sin^2 (\pi\lambda x^n)$$ is divergent for all $x\geq 1$.

(2) If $x$ is algebraic, and there exists $\lambda\neq 0$ such that $$\sin (\pi\lambda x^n)$$ converges to $0$, then $\lambda \in \mathbb{Q}(x)$.

With $\lambda=1/\pi$, there does not exist any real number $x\geq 1$ such that $$ \sum_{n=0}^{\infty} \sin^2 x^n$$ converges.

Therefore, the a partial conclusion to your problem is:

The series $$ \sum_{n=0}^{\infty} (-1)^{n} \sin(x^{-n})$$ absolutely converges if and only if $|x|>1$. If $|x|<1$ and $x$ is algebraic, then the series does not converge.

Still, whether or not there exists a real number $x>1$ such that $\sin x^n$ converges, is open. What we know so far is: if $x>1$ algebraic then $\sin x^n$ does not converge to $0$.


One observation.

Let, w.l.o.g. $1<A<2\pi$. Then, let, $n=\min_{k}\{k\in \mathbb{N}:A^{k}<2\pi,\ A^{k+1}\ge2\pi\}$. So the $n+1$ st term of the sequence $A^n\mod{2\pi}$ is $A^{n+1}-2\pi m$ for some $m\ge 1$. Let, $A_1:=A^{n+1}-2\pi m$. Now, $A_1\le 2\pi A-2\pi$. So, if $2\pi A-2\pi \le A\Rightarrow A\le \frac{2\pi}{2\pi -1}$, then, $A_1\le A$. now, again define $n_1:=\min_{k}\{k\in \mathbb{N},k\ge n+1: A_1^{k}<2\pi, A_1^{k+1}\ge 2\pi\}$ and again set $A_2:=A_1^{n_1+1}-2\pi m_1$ and now, we can see that $A_2\le A_1$. In this way one can find a decreasing sub-sequence ${A_n}$ in the orbit of $A^n \mod{2\pi} $.

  • $\begingroup$ $\sin((A+2\pi)^2) \ne \sin(A^2)$. $\endgroup$ – aschepler Feb 22 '14 at 1:30
  • $\begingroup$ I am not using this anywhere, all I'm saying is that once you have a term $A^n$ which is the first term to cross $2\pi$, then call it $A_1$ and similarly if you define $A_2,A_3,\cdots $ etc. then they should form a non-increasing sequence. $\endgroup$ – Samrat Mukhopadhyay Feb 24 '14 at 9:32

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.