Is the following proof correct? Can it be improved?

Let $x\in \mathbb{R}$ and suppose that $\{x_n\}$ is a sequence that does not converge to $x$. Does there exist a subsequence $\{x_{n_i}\}$ which has no subsequence that converges to $x$?

Call a sequence $\{y_m\}$ perfect if

  1. every subsequence $\{y_{m_i}\}\subseteq \{y_m\}$ converges to the same limit, or

  2. every subsequence $\{y_{m_i}\}\subseteq \{y_m\}$ is divergent.

Note that the assertion is clearly true for perfect subsequences. Therefore, it suffices to prove that $\{x_n\}$ contains a perfect subsequence that does not converge to $x$.

If $\{x_n\}$ is convergent, let $E$ denote the set of all subsequential limits of $\{x_n\}$, pick $L\in E\setminus\{x\}$, and define $\{x_{n_i}\}$ by $$n_i=\operatorname{max}\left\{m:\left| x_{m}-L \right|=\operatorname{min}\{\left| x_j-L \right|:1\leq j < i\}\right\}.$$
If $\{x_n\}$ is divergent, define $\{x_{n_i}\}$ by $$n_i=\operatorname{max}\left\{m:\left| x_{m} \right|=\operatorname{max}\{\left| x_j \right|:1\leq j < i\}\right\}.$$

In both cases, it is clear by construction that $n_i\leq n_j \Leftrightarrow i<j$, so $\{x_{n_i}\}$ is a subsequence of $\{x_n\}$. In the convergent case, $x_{n_i}$ is perfect with every subsequence converging to $L\ne x$. In the divergent case case, every subsequence of $\{x_{n_i}\}$ is divergent. This exhausts all possibilities, so the assertion is true.

  • $\begingroup$ Note, $\mathbb{R}$ is Hausdorff (i.e. limits are unique, so the first case is meaningless). In the second case, its somewhat unclear how you're defining $n_i$, note that divergent doesn't mean going off to infinity ($1,-1,1,-1,\cdots$ diverges). I also think its better not to worry about the perfectness of the subsequence you choose, and simply choose a sequence which stays away from $x$. $\endgroup$ – Deven Ware Oct 7 '13 at 20:00
  • $\begingroup$ If $(x_n)$ is divergent and bounded, you can't find a subsequence such that every further subsequence diverges. Instead, if $(x_n)$ does not converge to $x$, choose $\epsilon>0$ such that for every $n$, there is an $m>n$ with $|x_m-x|>\epsilon$. Use this to construct a subsequence of $(x_n)$ that is "bounded away" from $x$. $\endgroup$ – David Mitra Oct 7 '13 at 20:01
  • $\begingroup$ @DavenWare You are right- what I meant to say in case $1$ was that the set of all subsequential limits of $\{x_n\}$ has cardinality $1$, and in case $2$ was that $\{x_n\}$ is unbounded. $\endgroup$ – Samuel Handwich Oct 7 '13 at 20:04
  • $\begingroup$ @Samuel: Even those two corrected cases don't cover everything. Daven's example gives a sequence that is bounded, having $2$ subsequential limits. Neither of the two cases covers it. Take a look at the answers given below, and see if they help you. $\endgroup$ – Cameron Buie Oct 7 '13 at 20:21
  • $\begingroup$ @CameronBuie Read the rest of my answer. $\endgroup$ – Samuel Handwich Oct 8 '13 at 16:46

It is simpler to note that if $x_n$ does not converge to $x$, then for all $\epsilon >0$, and for all $N$, there exists some $n \ge N$ such $x_n \notin B(x, \epsilon)$. Let $\nu(N)$ denote the first index greater than or equal to $N$ for which this is true.

Let $n_1 = \nu(1)$. Then let $n_{k+1} = \nu(n_k+1)$. Then $x_{n_k} \notin B(x, \epsilon)$ for all $k$.

In particular, $|x_{n_k} -x| \ge \frac{1}{2} \epsilon$ for all $k$, hence $x_{n_k}$ cannot contain as subsequence that converges to $x$.


In the divergent case, your argument does not work. Consider $x_n=(-1)^n,$ for an example. Since $|x_n|=1$ for all $n,$ then your definition of $n_i$ fails, as the natural numbers have no greatest element.

A better approach is as follows: Since the sequence $\{x_n\}$ does not converge to $x,$ then by definition, there is some $\epsilon>0$ such that for all $N\in\Bbb N,$ there is some $n\ge N$ such that $|x_n-x|\ge\epsilon.$ Let $A=\{n\in\Bbb N:|x_n-x|\ge\epsilon\},$ so that $A$ is an infinite subset of $\Bbb N$ by assumption. Enumerate the elements of $A$ in increasing order by $i\mapsto n_i.$ Show that no subsequence of $\{x_{n_i}\}$ converges to $x$.


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.