We were given a challenge by our calculus professor and I've been stuck on it for a while now.

Show that the set of subsequence limits of $A_n=\sin(n)$ is $[-1, 1]$ (another way to phrase this would be: $\forall r\in [-1,1]$ there exists a subsequence of $A_n=\sin(n)$ that converges to $r$).

What would be a good way to start?

  • 8
    $\begingroup$ Pick r. What are the values of x such that sin x = r? How can you get n to be close to one of these values? These are the obvious questions you should be asking yourself. $\endgroup$ Nov 7 '10 at 20:32
  • 2
    $\begingroup$ @daniel.jackson maybe you want to delete the 0 from your image. That way your cardinality grows with your reputation. $\endgroup$ Nov 7 '10 at 22:43
  • $\begingroup$ @Qiaochu: not sure this is what you meant, but for x = arcsin(r) + 2πk? $\endgroup$ Nov 8 '10 at 9:15
  • $\begingroup$ @daniel.jackson: yes. How can you get n to be close to one of these numbers? $\endgroup$ Nov 8 '10 at 10:22
  • $\begingroup$ @Qiaochu: not really sure how. $\endgroup$ Nov 8 '10 at 15:45

Following Qiaochu hint, I'll try to elaborate a bit.

Note: This is a complete rewrite of the proof to fix a flaw pointed out by Qiaochu and make it overall clearer.

First some notation. Let $S^1 = [0, 2\pi)$ with 0 and $2\pi$ identified. For two points $a, b \in S^1$ we'll denote by $a \oplus b = a + b \mod 2\pi$ (and likewise for $\ominus$) and let $A_n = \{n \mod 2\pi\}_{n=1}^{\infty}$. Also denote by $U_{\varepsilon}(p)$ a punctured $\varepsilon$-neighborhood of $p \in S^1$.

Our strategy will be to show that $\{A_n\}$ has at least one limit point in $S^1$. From this we will conclude that 0 is also a limit point and finally we will use this fact to show that every point in $S_1$ is a limit point of $A_n$ (this means that $\{A_n\}$ is dense in S^1). Having established this, we will use continuity of $\sin$ to finally resolve the problem.


As a preparation we will note the relation $A_{n+m} = A_n \oplus A_m \quad (1)$. As a corollary of this we have that the sequence $\{A_n\}$ is injective (for if not, there would exist $n, m \in \mathbb{N}, k \in \mathbb{Z} \quad n>m$ such that $A_n = A_m$ and so $n - m = 2 k \pi$, a contradiction with irrationality of $\pi$). This implies simple but crucial fact that the image of the sequence contains infinitely many points.

Now, to establish the density of ${A_n}$ in $S^1$ we will first show that there exists at least one limit point $p \in S^1$. This is established by a standard argument: consider intervals $I = [0, \pi]$ and $J = [\pi, 2\pi]$ and take the one which contains infinitely many points of $\{A_n\}$ (if both do, take the "bottom" one, i.e. $I$). Call this interval $I_0$. Now again divide it in two intervals of same length and let $I_1$ be the one with infinitely many points. Continue in this way to obtain an infinite sequance of intervals $I_0 \supset I_1 \supset \cdots$. Then the set $K = \cap_{n=0}^{\infty} I_n$ is non-empty (this should be covered in standard calculus course, I hope) and any point $p \in K$ is surely a limit point (by construction of $\{I_n\}$).

Thanks to the above we now know that for each $\varepsilon > 0$ there exist infinitely many points in $U_\varepsilon(p)$, i.e. infinitely many $n, m \quad n > m$ such that $|A_n - A_m - 2k\pi| < 2\varepsilon$ for some $k \in \mathbb{Z}$. But from (1) we get that $|A_{n-m} - 2k\pi| < 2\varepsilon$. Because we identify 0 and $2\pi$ in $S^1$ we can see that these differences are concentrated around 0. Therefore we have shown that 0 is also a limit point.

Now we will show that every $p \in S^1$ is a limit point. Suppose we are given $\varepsilon > 0$. We just take any $A_k$ from $U_{\varepsilon}(0)$ and note that for $l = \lfloor {p \over A_k} \rfloor$ we have $A_{lk} \in U_{\varepsilon}(p)$.

To conclude the answer we will show that any $r \in [-1, 1]$ is indeed a limit point of $\{\sin(n)\}$. Take $s = \arcsin (r) \mod 2\pi$. Then by the above, we know that $s$ is a limit point of $\{A_n\}$ so there is a subsequence $\{A_{n_k}\} \to s$ and observe by continuity of $\sin$ that $\{\sin(A_{n_k})\} = \{\sin(n_k)\} \to sin(s) = r$.

  • $\begingroup$ You don't need Bolzano-Weierstrass, and the last line of the argument is wrong. $\endgroup$ Nov 7 '10 at 22:32
  • $\begingroup$ Thank you for pointing that out. I now see that it suffices to use the injectivity of the sequence to obtain the desired approximation property. $\endgroup$
    – Marek
    Nov 7 '10 at 23:22
  • $\begingroup$ Thanks for the detailed answer, unfortunately I don't think I've learned the required material to understand your solution. $\endgroup$ Nov 8 '10 at 9:16
  • $\begingroup$ @Marek: The last line of the argument is still wrong. $\endgroup$ Nov 8 '10 at 10:23
  • $\begingroup$ I am aware of that Qiaochu. But I wonder what I should do? Rewrite the solution completely? Or post new answer with correct solution? I am not sure what the policy is, being new to the site. $\endgroup$
    – Marek
    Nov 8 '10 at 11:32

Suppose you want to find positive integers $n, k$ such that $|n - 2 \pi k - \arcsin r| < \frac{1}{m}$ for some large positive integer $m$ and $r \in [-1, 1]$. Then I claim that some value of $n$ less than $2\pi m$ works. This is the pattern I was trying to get you to see, although I wasn't doing a very good job of it (I should've asked you to plot $k$ instead of $n$).

Unfortunately I think this is a little hard to prove except when $r = 0$; fortunately, using the second half of Marek's argument you can prove the same result with a weaker bound on $n$ for general $r$ using the case of $r = 0$, so I would encourage you to try that case first. To do this, try plotting the fractional parts of the numbers $\frac{n}{2\pi}$ in $[0, 1)$.

  • 1
    $\begingroup$ This works because $1/2\pi$ is irrational. More generally, the fractional parts of the sequence $n\alpha$ is dense in the unit interval for any irrational $\alpha$, and is not dense for any rational $\alpha$. So, at the very least, any proof should mention irrationality. $\endgroup$ Nov 9 '10 at 1:21

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.