Understanding Analysis · Chapter 2 · §2.5

Subsequences and the Bolzano-Weierstrass Theorem

Exercise 2.5.5

Assume (an)(a_{n}) is a bounded sequence with the property that every convergent subsequence of (an)(a_{n}) converges to the same limit aRa \in \mathbb{R}. Show that (an)(a_{n}) must converge to aa.

Proof.
Let’s prove by contradiction. Suppose (an)(a_{n}) does not converge to aa. Since (an)(a_{n}) does not converge to aa, then there exists ϵ0>0\epsilon_{0} > 0 such that for each NNN \in \mathbb{N}, we can find an nNn \geq N where anaϵ0|a_{n} - a| \geq \epsilon_{0}. We can use this sequence to construct a subsequence that does not converge to aa using induction.

Base case: Let N=1N = 1. There exists an index n11n_{1} \geq 1 such that an1aϵ0|a_{n_{1}} - a| \geq \epsilon_{0}.

Inductive step: Assume we have found n1<n2<n3<<nkn_{1} < n_{2} < n_{3} < \dots < n_{k} that satisfy ankaϵ0|a_{n_{k}} - a | \geq \epsilon_{0}. To find the next term, let N=nk+1N = n_{k} + 1. Then there exists an index nk+1nk+1n_{k+1} \geq n_{k} + 1 such that ank+1aϵ0|a_{n_{k+1}} - a |\geq \epsilon_{0}.

Because nk+1nkn_{k+1} \geq n_{k}, we have increasing indices and have constructed a subsequence (ank)(a_{n_{k}}) where ankaϵ0|a_{n_{k}} - a| \geq \epsilon_{0} for each kNk \in \mathbb{N}.

Since the original sequence (an)(a_{n}) is bounded, the subsequence (ank)(a_{n_{k}}) is also bounded. By the Bolzano-Weierstrass Theorem, the bounded sequence (ank)(a_{n_{k}}) contains a convergent subsequence. Let’s call this convergent subsequence (ym)(y_{m}). Since (ym)(y_{m}) is a subsequence of (ank)(a_{n_{k}}), and (ank)(a_{n_{k}}) is a subsequence of (an)(a_{n}), then (ym)(y_{m}) is a subsequence of (an)(a_{n}). From the problem statement, every convergent subsequence of (an)(a_{n}) must converge to aa. Therefore, since (ym)(y_{m}) is a convergent subsequence of (an)(a_{n}), then (ym)(y_{m}) must converge to aa.

Since every term in (ym)(y_{m}) comes from (ank)(a_{n_{k}}), then every term satisfies ymaϵ0|y_{m} - a| \geq \epsilon_{0}. If every term is at least ϵ0\epsilon_{0} distance away from aa, the limit of (ym)(y_{m}) can not be aa. We know this because by the Order Limit Theorem, the limit LL must satisfy Laϵ0>0|L - a| \geq \epsilon_{0} > 0. This is a contradiction: (ym)(y_{m}) converges to aa but is also bounded away from aa. Therefore our initial assumption that ana_{n} does not converge to aa is false, and we conclude (an)(a_{n}) converges to aa. \blacksquare

Stephen Abbott, Understanding Analysis, 2nd edition.