Skip to content
Topos Uranos

Subsequences, Bolzano-Weierstrass, limit superior and inferior

Summary

This lesson studies subsequences, which are formed by choosing infinitely many terms of a sequence in the order in which they appear, and with them a new face of the completeness axiom, after the supremum and monotone convergence: the Bolzano-Weierstrass theorem, according to which every bounded sequence has a convergent subsequence. It proves it on the real line, with a lemma that extracts a monotonic subsequence from any sequence, and in the plane of complex numbers, componentwise. It defines the limit superior and the limit inferior as limits of the suprema and of the infima of the tails of the sequence, characterizes them with ε\varepsilon and NN, proves that they are the largest and the smallest of the limits of the subsequences, and that a bounded sequence converges if and only if they coincide. Finally, it proves the fundamental theorem of algebra, which Algebra and Geometry had admitted: the modulus of a polynomial attains a minimum in the plane, and d’Alembert’s lemma prevents that minimum from being positive.

Learning objectives

  1. Recognize subsequences, prove that those of a convergent sequence have the same limit as the sequence, and use subsequences to prove that a sequence diverges.
  2. Prove the Bolzano-Weierstrass theorem on the real line and in the plane, and explain why boundedness cannot be omitted.
  3. Compute limits superior and inferior, characterize them with ε\varepsilon and NN, and relate them to the limits of subsequences and to convergence.
  4. Prove the fundamental theorem of algebra with the Bolzano-Weierstrass theorem in the plane and d’Alembert’s lemma.

Subsequences

To show that (−1)n(-1)^n diverges, the lesson on sequences and their limit looked separately at its terms of even index, all equal to 11, and at those of odd index, equal to −1-1; and the lesson on monotonic sequences and the number e studied the recurrence an+1=11+ana_{n + 1} = \frac{1}{1 + a_n} by separating the sequences bm=a2m−1b_m = a_{2m - 1} and cm=a2mc_m = a_{2m}. In both cases infinitely many terms of the sequence are chosen, without altering their order, and a new sequence is formed with them.

DefinitionSubsequence

Let (an)(a_n) be a sequence. If (nk)(n_k) is a strictly increasing sequence of natural numbers, that is, n1<n2<n3<⋯n_1 < n_2 < n_3 < \cdots, the sequence whose kkth term is anka_{n_k} is called a subsequence of (an)(a_n) and is written (ank)\left(a_{n_k}\right).

With the indices nm=2m−1n_m = 2m - 1 and nm=2mn_m = 2m, which are strictly increasing, the sequences (bm)(b_m) and (cm)(c_m) of the lesson on monotonic sequences are subsequences of (an)(a_n): the one of the terms of odd index and the one of the terms of even index. So is, for a fixed natural number pp, the sequence (an+p)(a_{n + p}), with the indices n+pn + p; and every sequence is a subsequence of itself. On the other hand, a1,a1,a2,a2,…a_1, a_1, a_2, a_2, \ldots is not a subsequence, because it repeats indices, nor is a2,a1,a4,a3,…a_2, a_1, a_4, a_3, \ldots, which puts them out of order. The letter of the index is irrelevant; what matters is that the chosen indices increase strictly.

LemmaThe indices of a subsequence

If (nk)(n_k) is a strictly increasing sequence of natural numbers, then nk≥kn_k \geq k for every natural number kk. If, moreover, (kj)(k_j) is another strictly increasing sequence of natural numbers, the indices mj=nkjm_j = n_{k_j} are strictly increasing as well, so that a subsequence of a subsequence of (an)(a_n) is a subsequence of (an)(a_n). The first claim is an induction: each index exceeds the one before it by at least one.

Proof

  1. nk+1≥nk+1≥k+1n_{k + 1} \geq n_k + 1 \geq \resaltar{k + 1}

    For k=1k = 1, the natural number n1n_1 is at least 11. If nk≥kn_k \geq k, since nk+1n_{k + 1} is a natural number greater than nkn_k, it is at least nk+1n_k + 1, because there are no others between two consecutive natural numbers.

  2. j<l→kj<kl→mj<mlj < l \rightarrow k_j < k_l \rightarrow \resaltar{m_j < m_l}

    Since (nk)(n_k) is strictly increasing, nk<nln_k < n_l whenever k<lk < l, by induction on ll: if nk<nln_k < n_l, then nk<nl<nl+1n_k < n_l < n_{l + 1}. If j<lj < l, then kj<klk_j < k_l, and therefore nkj<nkln_{k_j} < n_{k_l}. The jjth term of the subsequence of (ank)\left(a_{n_k}\right) with the indices kjk_j is amja_{m_j}.

TheoremThe subsequences of a sequence with a limit

If an→La_n \to L, every subsequence (ank)\left(a_{n_k}\right) of (an)(a_n) converges to LL. If an→∞a_n \to \infty, every subsequence tends to ∞\infty, and if an→−∞a_n \to -\infty, to −∞-\infty. The natural number that responds for the sequence also responds for the subsequence, because nk≥kn_k \geq k.

Proof

  1. k≥N→nk≥k≥N→∣ank−L∣<εk \geq N \rightarrow n_k \geq k \geq N \rightarrow \resaltar{\abs{a_{n_k} - L} < \varepsilon}

    Let ε>0\varepsilon > 0, and let NN be the response of the definition of the limit for (an)(a_n), which depends on ε\varepsilon and on the sequence. If k≥Nk \geq N, by the lemma on the indices, nk≥k≥Nn_k \geq k \geq N, and the term anka_{n_k} lies in the band.

  2. k≥N→nk≥N→ank>Kk \geq N \rightarrow n_k \geq N \rightarrow \resaltar{a_{n_k} > K}

    If an→∞a_n \to \infty, the same argument, with the definition of sequences that tend to infinity, from the lesson on sequences and their limit, and a bound KK instead of the challenge ε\varepsilon, gives ank>Ka_{n_k} > K for k≥Nk \geq N; the case of −∞-\infty reduces to this one with the sequence (−an)(-a_n).

CorollaryDivergence criterion by subsequences

If a sequence has two subsequences with different limits, or a divergent subsequence, then it diverges. If it converged to LL, all its subsequences would converge to LL.

Proof

  1. ank→L1,amk→L2,an→L→L1=L=L2a_{n_k} \to L_1, \quad a_{m_k} \to L_2, \quad a_n \to L \rightarrow \resaltar{L_1 = L = L_2}

    If (an)(a_n) converged to a number LL, by the preceding theorem every subsequence would converge to LL. Two subsequences with limits L1L_1 and L2L_2 would give L1=L=L2L_1 = L = L_2, by the uniqueness of the limit, and no subsequence would be divergent.

For (−1)n(-1)^n, the subsequences of even index and of odd index tend to 11 and to −1-1, and the corollary sums up the argument of the lesson on sequences and their limit. The sequence sin⁡nπ2\sin \frac{n\pi}{2}, which runs through 1,0,−1,0,…1, 0, -1, 0, \ldots, has constant subsequences with limits 11, 00 and −1-1, with the indices 4m−34m - 3, 2m2m and 4m−14m - 1, and diverges. In the opposite direction, if the subsequences of even index and of odd index tend to the same number LL, the sequence tends to LL, by the argument of the third move of the solved problem “A recurrence with a decreasing function” of the lesson on monotonic sequences, which merges the responses of the two subsequences into a single one; the proposed problem “The evens, the odds and the multiples of three” shows how to merge three subsequences.