Skip to content
Topos Uranos

Summary

This lesson studies Cauchy sequences, whose terms come as close to one another as desired from some index on, without the condition mentioning any limit. It proves that in the reals they are exactly the convergent sequences (the Cauchy criterion), with the Bolzano-Weierstrass theorem, and with this it presents the last face of the completeness axiom in this unit: in an Archimedean ordered field, the supremum axiom is equivalent to the convergence of every Cauchy sequence. The rationals, which are Archimedean and not complete, have Cauchy sequences with no rational limit. Finally, it proves that contractive sequences are Cauchy and, with this, the fixed-point theorem for contractive functions, with an error bound that is known before iterating; a bound on the derivative, through the mean value theorem, makes it possible to recognize a contraction and to approximate the solutions of equations that cannot be solved with algebraic operations.

Learning objectives

  1. Define Cauchy sequences, prove that in the reals they are exactly the convergent ones, and use the criterion to prove a convergence without knowing the limit.
  2. Prove that, in an Archimedean ordered field, the completeness axiom is equivalent to the convergence of Cauchy sequences, distinguishing what is assumed in each direction, and explain why the rationals do not satisfy it.
  3. Prove that contractive sequences are Cauchy, and the fixed-point theorem for contractive functions, with its two error bounds.
  4. Recognize a contraction through a bound on the derivative, with the mean value inequality, and choose the form of an equation whose iteration converges.

The Cauchy condition

To prove with the definition that an→La_n \to L, one must know LL in advance. The monotone convergence theorem, from the lesson on monotonic sequences and the number e, and the Bolzano-Weierstrass theorem, from the lesson on subsequences and the Bolzano-Weierstrass theorem, guarantee that a limit exists without computing it, but the first holds only for monotonic sequences, and the second gives the limit of some subsequence, not that of the sequence. If a sequence converges, its terms approach the limit and, therefore, approach one another; the following condition expresses the latter, which speaks only of the terms.

DefinitionCauchy sequence

A sequence (an)(a_n) of real numbers is said to be Cauchy if for every ε>0\varepsilon > 0 there exists a natural number NN such that ∣am−an∣<ε\abs{a_m - a_n} < \varepsilon for all natural numbers mm and nn greater than or equal to NN:

(∀ε>0)(∃N∈N)(∀m,n∈N)(m≥N∧n≥N→∣am−an∣<ε)(\forall \varepsilon > 0)(\exists N \in \mathbb{N})(\forall m, n \in \mathbb{N})\left(m \geq N \wedge n \geq N \rightarrow \abs{a_m - a_n} < \varepsilon\right)

As in the definition of the limit, NN is chosen after ε\varepsilon, and depends on it and on the sequence, but not on mm or on nn. Writing m=n+pm = n + p, the condition says that, for n≥Nn \geq N, the difference ∣an+p−an∣\abs{a_{n + p} - a_n} is less than ε\varepsilon for every natural number pp at once. It is not enough that each difference an+p−ana_{n + p} - a_n, with pp fixed, tend to 00: the solved problem “Differences that tend to zero without being Cauchy” shows a divergent sequence with that property.

PropositionEvery convergent sequence is Cauchy

If an→La_n \to L, the sequence (an)(a_n) is Cauchy. Two terms whose distance from LL is less than ε2\frac{\varepsilon}{2} are at a distance less than ε\varepsilon from each other.

Proof

  1. ∣am−an∣≤∣am−L∣+∣L−an∣<ε2+ε2=ε\abs{a_m - a_n} \leq \abs{a_m - L} + \abs{L - a_n} < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \resaltar{\varepsilon}

    Let ε>0\varepsilon > 0, and let NN be the response of the definition of the limit to the challenge ε2\frac{\varepsilon}{2}. If mm and nn are greater than or equal to NN, the triangle inequality, from the lesson on the real line, with LL as an intermediate point, gives the Cauchy condition with the same NN.

PropositionEvery Cauchy sequence is bounded

If (an)(a_n) is Cauchy, there exists some M>0M > 0 such that ∣an∣≤M\abs{a_n} \leq M for every natural number nn. From some index on, all the terms are close to one of them; before it, there are finitely many.

Proof

  1. n≥N→∣an∣≤∣an−aN∣+∣aN∣<1+∣aN∣n \geq N \rightarrow \abs{a_n} \leq \abs{a_n - a_N} + \abs{a_N} < \resaltar{1 + \abs{a_N}}

    With the challenge ε=1\varepsilon = 1, there exists a natural number NN such that ∣an−aN∣<1\abs{a_n - a_N} < 1 for every n≥Nn \geq N; for those nn, the triangle inequality bounds the term.

  2. M=max⁡(∣a1∣,…,∣aN−1∣,1+∣aN∣),∣an∣≤MM = \max\left(\abs{a_1}, \ldots, \abs{a_{N - 1}}, 1 + \abs{a_N}\right), \qquad \resaltar{\abs{a_n} \leq M}

    The terms of index less than NN are finitely many. The largest of the following numbers, which exists because there are finitely many of them, is greater than or equal to the absolute value of each term of index less than NN, and to 1+∣aN∣1 + \abs{a_N}, which bounds the others.

LemmaA Cauchy sequence with a convergent subsequence converges

If (an)(a_n) is Cauchy and one of its subsequences (ank)\left(a_{n_k}\right) converges to LL, then an→La_n \to L. A far-off term of the subsequence is close to LL, and all the terms beyond the Cauchy index are close to it.

Proof

  1. k≥max⁡(N,K)→nk≥N,∣ank−L∣<ε2\dato{k}{k \geq \max(N, K) \rightarrow n_k \geq N, \qquad \abs{a_{n_k} - L} < \frac{\varepsilon}{2}}

    Let ε>0\varepsilon > 0. Let NN be the response of the Cauchy condition to the challenge ε2\frac{\varepsilon}{2}, and KK that of the definition of the limit of the subsequence to the same challenge. Choose some kk greater than or equal to NN and to KK; by the lemma on the indices of a subsequence, from the lesson on subsequences, nk≥k≥Nn_k \geq k \geq N.

  2. n≥N→∣an−L∣≤∣an−ank∣+∣ank−L∣<ε2+ε2=εn \geq N \rightarrow \abs{a_n - L} \leq \abs{a_n - a_{n_k}} + \abs{a_{n_k} - L} < \frac{\varepsilon}{2} + \frac{\varepsilon}{2} = \resaltar{\varepsilon}

    If n≥Nn \geq N, the Cauchy condition applies to the indices nn and nkn_k, and the triangle inequality, with anka_{n_k} as an intermediate point, gives the limit; the natural number NN depends only on ε\varepsilon and on the sequence.

TheoremCauchy criterion

A sequence of real numbers converges if and only if it is Cauchy. One direction is the proposition on convergent sequences; the other combines boundedness, the Bolzano-Weierstrass theorem and the preceding lemma.

Proof

  1. m≥N∧n≥N→∣am−an∣<εm \geq N \wedge n \geq N \rightarrow \resaltar{\abs{a_m - a_n} < \varepsilon}

    If (an)(a_n) converges, it is Cauchy, by the proposition according to which every convergent sequence is Cauchy: for each ε>0\varepsilon > 0 there is some NN with the following condition.

  2. ∣an∣≤M,ank→L,an→L\abs{a_n} \leq M, \qquad a_{n_k} \to L, \qquad \resaltar{a_n \to L}

    If (an)(a_n) is Cauchy, it is bounded, by the preceding proposition. By the Bolzano-Weierstrass theorem, from the lesson on subsequences, it has a subsequence (ank)\left(a_{n_k}\right) that converges to a number LL, and by the lemma, the whole sequence converges to LL.

The criterion proves that a sequence converges without naming its limit, like the monotone convergence theorem, but without requiring monotonicity. Its condition requires comparing all pairs of terms from NN on, and in practice it is verified with a bound ∣am−an∣≤cn\abs{a_m - a_n} \leq c_n, for every m>nm > n, with cn→0c_n \to 0; the last section obtains such a bound for contractive sequences. The lesson on numerical series and their convergence will apply the criterion to the sequence of partial sums of a series, whose limit is almost never known. The proposed problem “The Cauchy criterion in the plane” extends it to sequences of complex numbers.