86
$\begingroup$

Is the following true?
Let $x_n$ be a sequence with the following property: Every subsequence of $x_n$ has a further subsequence which converges to $x$. Then the sequence $x_n$ converges to $x$.

I guess that it is true but I am not sure how to prove this.

$\endgroup$
4
  • $\begingroup$ What's the setting here? Metric spaces? $\endgroup$ Aug 18, 2014 at 9:53
  • $\begingroup$ Possible duplicate of Sufficient condition for convergence of a real sequence $\endgroup$
    – Guy Fsone
    Dec 5, 2017 at 15:39
  • $\begingroup$ math.stackexchange.com/questions/173590/… $\endgroup$
    – Guy Fsone
    Jan 10, 2018 at 22:16
  • $\begingroup$ I would like to add that this result easily generalizes to any topological space (as already mentioned). Furthermore, this result implies that in a limit compact space, if every convergent subsequence of the sequence converges to some point, then the original sequence will converge to that point. $\endgroup$
    – Eigenfield
    Nov 6, 2023 at 3:58

4 Answers 4

77
$\begingroup$

True. If not, there exists an $\epsilon > 0$, such that for all $k$, there exists an $n_k > k$ satisfying $|x_{n_k}−x| \ge \epsilon$ since if there is some $k$ which doesn't have such $n_k$, then we can take it as $N$, so $x_n$ converges to $x$. The subsequence $x_{n_{k}}$ does not have any subsequence converging to $x$.

$\endgroup$
2
  • 22
    $\begingroup$ This works in an arbitrary topological space: "If not, there exists a neighborhood $U$ of $x$ such that for all $k$, there is an $n_k > k$ such that $x_{n_k}\notin U$. Otherwise, $\{x_n\}$ is eventually in all neighborhoods of $x$, which contradicts $x_n\not\to x$. Then, $\{x_{n_k}\}$ has no subsequences converging to $x$, since it is never in $U$." $\endgroup$
    – Michael L.
    Aug 22, 2017 at 11:08
  • 9
    $\begingroup$ Even better, the result can be extended to nets. $\endgroup$
    – Michael L.
    Aug 22, 2017 at 11:13
13
$\begingroup$

This is much easier to prove than the other answers here might suggest, if you remember the handy little fact (given, for instance, in Baby Rudin but which aught to be more well-known) that a sequence $x_n$ does not converge to $x$ if, and only if, there exists $\epsilon>0$ such that $|{x_n-x}| \geq \epsilon$ for infinitely many $n$.

On the one hand, if $x_n \to x$ then this fact gives that any subsequence $(x_{n_i}) \subset (x_i)$ can also have finitely many terms a distance $\epsilon$ from $x$, and therefore also converges to $x$. In particular, if $x_n$ converges to $x$ then every subsequence has a subsequence converging to $x$, namely itself

On the other hand, if $x_n \not\to x$ then there exist infinitely many $n$ such that $|x_n-x| \geq \epsilon$ and the subsequence of these terms has no subsequence converging to $x$ as any subsequence has all terms a distance of at least $\epsilon$ away from $x$

Note, however, that it is necessary that the sub-subsequnces converge to the same limit: for instance, any subsequence of the alternating sequence $0,1,0,1,0,1,\ldots$ has a convergent subsequence by the pigeonhole principle, but such sub-subsequences can converge to either 0 or 1

$\endgroup$
7
$\begingroup$

proof by Contradiction

Suppose that $\{X_n\}$ does not converge to $\ell$. Then, there is $\varepsilon_0>0$ such that $$\forall N\in\mathbb N,\exists \hspace{.2cm}n=n(N) : n>N~~~and ~~~ |X_n -\ell|>\varepsilon_0 $$

For $N_1=1$ there exists $n_1$ such that $$n_1>N_1 ~~~and ~~~ |X_{n_1} -\ell|>\varepsilon_0 $$ Taking successively $N_{k+1}> \max\{N_k, n_k,k+1\}$ there exists $n_{k+1>N_{k+1}}$ such that,

$$ |X_{ n_{k+1}} -\ell|>\varepsilon_0 $$

It is easy to see that, $\{X_{ n_k}\}_k$ is a subsequence of $\{X_{ n}\}_n$ since $$ n_k< n_{k+1} \quad i.e ~~\text{the map }~~k\mapsto n_k~~~\text{Is one-to-one}$$

However, $$\forall k,~~ |X_{ n_{k}} -\ell|>\varepsilon_0 \qquad \text{and}~~~\{X_{ n_{k}} \}~~~\text{is bounded} $$

Therefore By Bolzano-Weierstrass Theorem's there exists $\{X_{ n_{k_p} }\}_p$ subsequence of $\{X_{ n_{k} }\}_k$ which converges to some limit $\ell_1 $ but $\{X_{ n_{k_p} }\}_p\to \ell_1$ is also a congering subsequence of $\{X_n\}_n$

By assumption, $\ell=\ell_1$ that is together with the fact $\{X_{ n_{k_p} }\}_p$ is a subsequence of $\{X_{ n_{k} }\}_k$ we have

$$0=\lim_{p\to\infty } |X_{ n_{k_p} }-\ell|>\varepsilon_0>0~~~\text{CONTRADICTION}$$

Note that $$\forall p,~~|X_{ n_{k_p} }-\ell|>\varepsilon_0$$ Since $$\forall k,~~|X_{ n_{k}} -\ell|>\varepsilon_0$$

Direct proof(Update) Since $(|x_n-\ell|)_n$ is bounded, $M=\limsup_{n\to\infty} |x_n-\ell|$ exists and thus by definition of the $\limsup$, there exists a subsequnce $(x_{n_k})_k$ such that $$\lim_{k\to \infty} |x_{n_k}-\ell|=M= \limsup_{n\to\infty} |x_n-\ell| $$

By assumption there exists a further subsequence $(x_{n_{k_p}})_p$ converging to $\ell$. Hence we have $$0= \lim_{p\to \infty} |x_{n_{k_p}}-\ell|= \lim_{k\to \infty} |x_{n_k}-\ell|=M =\limsup_{n\to\infty} |x_n-\ell|.$$ Consequently, $$0\leq \liminf_{n\to\infty} |x_n-\ell|\leq \limsup_{n\to\infty} |x_n-\ell|=0$$

that is $$ \liminf_{n\to\infty} |x_n-\ell|= \limsup_{n\to\infty} |x_n-\ell|=0$$

that is $\lim_{n\to\infty} |x_n-\ell|=0$ or $x_n\to x$ as $n\to \infty$.

$\endgroup$
3
  • 2
    $\begingroup$ BW seems overkill here... $\endgroup$ Feb 25, 2019 at 17:19
  • $\begingroup$ (+1), very comprehensive answer. :) By the way, do you know any direct proof for this? If, yes, please take a look at my question here. $\endgroup$ Feb 19, 2020 at 12:04
  • $\begingroup$ @HoseinRahnama Sorry for the late reply. I have updated the result $\endgroup$
    – Guy Fsone
    Aug 11, 2021 at 14:56
4
$\begingroup$

Assume $(x_n)$ doesn't converge to $x$. In case $1, (x_n)$ isn't bounded, thus has at least one sub-sequence going to either $\infty$ or $-\infty.$ This subsequence clearly doesn't have any subsequence converging to $x$ (thus a contradiction).

In case $2, (x_n)$ is bounded. If $(x_n)$ converges to a finite real limit different from $x$ then you get an immediate contradiction to the property. Otherwise $(x_n)$ is divergent and bounded thus has $\liminf$ and $\limsup.$ These two converge, and have a sub-sequence which converges to $x$, thus must converge to $x$ themselves (otherwise you get a similar contradiction). And this means that : $$\liminf=\limsup=0 \implies \lim(x_n)=x$$

$\endgroup$
0

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .