Cauchy sequences

Proving that a sequence converges from the definition needs the limit in hand. When the limit is the very number we cannot name, can the terms alone show that a limit exists?

Where you are 5 in Sequences and series

Why this exists

To prove a_n \to L from the definition in Limits of sequences, epsilon and N, one names L first, yet often the limit is exactly what cannot be named: the root of an equation no formula solves, or a real number being built from rationals. The monotone convergence theorem avoids naming the limit, but only for monotone sequences. Cauchy’s test looks at the terms alone: if they eventually lie as close as we like to each other, they should pile up at some point. In \mathbb{R} they do, through the least-upper-bound property and the Subsequences and Bolzano-Weierstrass theorem. In \mathbb{Q} they need not, which makes the Cauchy condition a second way to say what completeness means.

Definition

Cauchy sequence. In plain words: past some index, any two terms are as close as we like. Formally, a sequence \{a_n\} of real numbers is a Cauchy sequence if for every \varepsilon > 0 there is N such that \lvert a_m - a_n \rvert < \varepsilon for every m > N and every n > N.

The two indices vary independently: m may be far beyond n. The definition makes sense in any ordered field F, with \varepsilon taken in F.

Negation. Negating the quantifiers in turn (Negating a quantified statement), \{a_n\} is not Cauchy if and only if there is \varepsilon > 0 such that for every N there are m, n > N with \lvert a_m - a_n \rvert \ge \varepsilon.

Cauchy-complete. An ordered field is Cauchy-complete if every Cauchy sequence in it converges to an element of it.

Contractive. A sequence \{a_n\} is contractive if there is a constant C with 0 < C < 1 such that \lvert a_{n+2} - a_{n+1} \rvert \le C \lvert a_{n+1} - a_n \rvert for every n \ge 1: each step is at most a fixed fraction of the step before.

Intuition

Terms crowding toward a limit L crowd toward each other: two terms within \displaystyle \frac{\varepsilon}{2} of L are within \varepsilon of each other. The converse is the substance. A Cauchy sequence is eventually trapped in a window of width \varepsilon, for every \varepsilon, but no window is centered at a named point. Bolzano-Weierstrass supplies the point: the sequence is bounded, so a subsequence converges to some L, and every late term is close to a late term of that subsequence, hence close to L.

The condition is not “consecutive terms get close”. In Figure 1 the steps of \displaystyle a_n = \cos(\pi\sqrt{n}) shrink to 0, yet the terms keep returning to 1 and to -1 (Example 2). The Cauchy condition compares every pair of late terms, however far apart.

25
50
75
100
1
0
-1
n
a_n
Figure 1The terms \displaystyle a_n = \cos(\pi\sqrt{n}) for n = 1 to 100. Neighbors draw closer, yet the terms return to 1 at n = (2k)^2 and to -1 at n = (2k+1)^2 (larger dots), so the sequence is not Cauchy.

Results

The negation, tails and order limit theorems are Theorems 1, 3 and 8 of Limits of sequences, epsilon and N; monotone convergence and its converse are Theorems 1 and 2 of The monotone convergence theorem. Theorems 1, 2 and 3 below hold in any ordered field, with the same proofs.

Theorem 1 (convergent implies Cauchy). If a_n \to L, then \{a_n\} is a Cauchy sequence.

Theorem 2 (Cauchy implies bounded). Every Cauchy sequence is bounded.

Theorem 3 (one convergent subsequence is enough). If \{a_n\} is Cauchy and some subsequence a_{n_k} \to L, then a_n \to L.

Theorem 4 (Cauchy criterion). A sequence of real numbers converges if and only if it is a Cauchy sequence. So \mathbb{R} is Cauchy-complete.

Theorem 5 (contractive sequences). Let \{a_n\} be a sequence of real numbers, contractive with constant C, and let d = \lvert a_2 - a_1 \rvert. Then for all m > n \ge 1,

\lvert a_m - a_n \rvert \le \frac{d\,C^{n-1}}{1 - C}

so \{a_n\} is Cauchy and converges to some L, and \displaystyle \lvert L - a_n \rvert \le \frac{d\,C^{n-1}}{1 - C} for every n \ge 1.

Theorem 6 (Cauchy completeness and the least-upper-bound property). Let F be an Archimedean ordered field: for every x \in F there is a positive integer greater than x. If F is Cauchy-complete, then F has the least-upper-bound property. With Theorem 4, an Archimedean ordered field is Cauchy-complete if and only if it has the least-upper-bound property.

Why it is true

Theorem 1. Proof. Let \varepsilon > 0 and take N with \displaystyle \lvert a_n - L \rvert < \frac{\varepsilon}{2} for every n > N. For m, n > N the Triangle inequality gives \lvert a_m - a_n \rvert \le \lvert a_m - L \rvert + \lvert L - a_n \rvert < \varepsilon. \square

Theorem 2. Proof. Take \varepsilon = 1: there is N \ge 1 (enlarge N if needed) with \lvert a_m - a_n \rvert < 1 for all m, n > N. For n > N, \lvert a_n \rvert \le \lvert a_{N+1} \rvert + \lvert a_n - a_{N+1} \rvert < \lvert a_{N+1} \rvert + 1. So every term satisfies \lvert a_n \rvert \le B, where B = \max\{\lvert a_1 \rvert, \ldots, \lvert a_N \rvert, \lvert a_{N+1} \rvert + 1\} is the largest of finitely many numbers. \square

Theorem 3. Proof. Let \varepsilon > 0. Take N with \displaystyle \lvert a_m - a_n \rvert < \frac{\varepsilon}{2} for all m, n > N, and K with \displaystyle \lvert a_{n_k} - L \rvert < \frac{\varepsilon}{2} for every k > K. Fix one k > \max\{K, N\}; then n_k \ge k > N (Theorem 1 of Subsequences and Bolzano-Weierstrass). For every n > N, \lvert a_n - L \rvert \le \lvert a_n - a_{n_k} \rvert + \lvert a_{n_k} - L \rvert < \varepsilon. \square

Theorem 4. Proof. A convergent sequence is Cauchy by Theorem 1. Conversely, a Cauchy sequence is bounded by Theorem 2, so the Bolzano-Weierstrass theorem (Theorem 5 of Subsequences and Bolzano-Weierstrass) gives a subsequence converging to some L, and Theorem 3 gives a_n \to L. \square

This proof uses the least-upper-bound property only through Bolzano-Weierstrass and monotone convergence, so it holds in every ordered field with that property.

Theorem 5. Proof. If d = 0, every step is 0 and the sequence is constant, so let d > 0. First, C^n \to 0: \{C^n\} is nonincreasing and bounded below by 0, so it has a limit \ell by monotone convergence, and C^{n+1} = C \cdot C^n gives \ell = C\ell (tails and the product law), so \ell = 0.

By induction, \lvert a_{k+1} - a_k \rvert \le C^{k-1} d for every k \ge 1. For m > n, write a_m - a_n as the sum of the steps from n to m, then use the triangle inequality and the finite geometric sum:

\begin{aligned} \lvert a_m - a_n \rvert &\le \sum_{k=n}^{m-1} C^{k-1} d = \frac{d\,C^{n-1}(1 - C^{m-n})}{1 - C} \\ &\le \frac{d\,C^{n-1}}{1 - C} \end{aligned}

Given \varepsilon > 0, choose N with \displaystyle C^N < \frac{\varepsilon(1 - C)}{d}. For m > n > N, C^{n-1} \le C^N, so \lvert a_m - a_n \rvert < \varepsilon. So \{a_n\} is Cauchy, and Theorem 4 gives a limit L. Fix n and let \displaystyle b = \frac{d\,C^{n-1}}{1 - C}. Every m > n has -b \le a_m - a_n \le b, and the order limit theorem, as m \to \infty, gives -b \le L - a_n \le b. \square

Theorem 6. Proof. By Theorem 2 of The monotone convergence theorem, it is enough that every nondecreasing sequence \{a_n\} in F bounded above, say by b, converges; as F is Cauchy-complete, it is enough that \{a_n\} be Cauchy. Suppose it is not. Then there is \varepsilon > 0 such that every N has indices m > n > N with \lvert a_m - a_n \rvert \ge \varepsilon, that is a_m - a_n \ge \varepsilon, as the sequence is nondecreasing.

Let n_1 = 1. Given n_k, take such m > n > n_k; then a_m - a_{n_k} \ge a_m - a_n \ge \varepsilon, and let n_{k+1} = m. By induction a_{n_k} \ge a_1 + (k - 1)\varepsilon for every k. As F is Archimedean, some positive integer k exceeds \displaystyle \frac{b - a_1}{\varepsilon}, and then a_{n_{k+1}} \ge a_1 + k\varepsilon > b, against b being an upper bound.

Conversely, if F has the least-upper-bound property, the proof of Theorem 4 applies in F, so F is Cauchy-complete. \square

Worked examples

Example 1

A cutoff for every pair

Let \displaystyle a_n = \frac{2n - 1}{n + 3}. Prove from the definition that \{a_n\} is Cauchy, and find the least N that works for \displaystyle \varepsilon = \frac{1}{100}.

Solution

Since 2(n + 3) - 7 = 2n - 1, \displaystyle a_n = 2 - \frac{7}{n + 3}, so \displaystyle a_m - a_n = \frac{7}{n + 3} - \frac{7}{m + 3}. A difference of two positive numbers is smaller in absolute value than the larger of them, and for m, n > N both denominators are at least N + 4. Hence, for m, n > N,

\begin{aligned} \lvert a_m - a_n \rvert &< \max\left\{\frac{7}{n + 3}, \frac{7}{m + 3}\right\} \\ &\le \frac{7}{N + 4} \end{aligned}

Given \varepsilon > 0, any integer \displaystyle N \ge \frac{7}{\varepsilon} - 4 makes this at most \varepsilon, so the sequence is Cauchy; the proof never used the limit.

For \displaystyle \varepsilon = \frac{1}{100}, \displaystyle \frac{7}{N + 4} \le \frac{1}{100} when N \ge 696. A smaller N fails: the pair n = 696 and m = 489{,}297 lies beyond every N \le 695, and

\lvert a_m - a_n \rvert = \frac{7}{699} - \frac{7}{489{,}300} = \frac{1}{100}

which is not less than \varepsilon. So the least N is 696.

Example 2

Steps that shrink, terms that never settle

Let \displaystyle a_n = \cos(\pi\sqrt{n}). Show that a_{n+j} - a_n \to 0 for every fixed j \ge 1, and that consecutive terms differ by less than \displaystyle \frac{1}{100} from n = 24{,}675 on, yet \{a_n\} is not Cauchy.

Solution

Since \displaystyle \cos u - \cos v = -2\sin\frac{u + v}{2}\sin\frac{u - v}{2} and \lvert \sin t \rvert \le \lvert t \rvert (The squeeze theorem and the limit of sine x over x for \displaystyle \lvert t \rvert < \frac{\pi}{2}; otherwise \lvert \sin t \rvert \le 1 < \lvert t \rvert), \lvert \cos u - \cos v \rvert \le \lvert u - v \rvert. Hence

\begin{aligned} \lvert a_{n+j} - a_n \rvert &\le \pi\left(\sqrt{n + j} - \sqrt{n}\right) = \frac{\pi j}{\sqrt{n + j} + \sqrt{n}} \\ &< \frac{\pi j}{2\sqrt{n}} \end{aligned}

which tends to 0 with j fixed, so a_{n+j} - a_n \to 0 by the squeeze theorem. For j = 1 the bound is at most \displaystyle \frac{1}{100} when \displaystyle \sqrt{n} \ge 50\pi, that is n \ge 2500\pi^2 \approx 24{,}674.01.

Not Cauchy: take \varepsilon = 1 in the negation. Given N, choose a positive integer k with (2k)^2 > N, and let n = (2k)^2, m = (2k + 1)^2. Then a_n = \cos(2k\pi) = 1 and a_m = \cos((2k + 1)\pi) = -1, so \lvert a_m - a_n \rvert = 2 with m, n > N. By Theorem 1 the sequence diverges. The gap 4k + 1 grows, so no fixed j catches these pairs (Figure 1).

Example 3

A Cauchy sequence of rationals with no rational limit

Let q_1 = 2 and \displaystyle q_{n+1} = \frac{1}{2}\left(q_n + \frac{3}{q_n}\right). Working only in \mathbb{Q}, show that \{q_n\} is Cauchy and has no rational limit. What is its limit in \mathbb{R}?

Solution

By induction every q_n is a positive rational. Let e_n = q_n^2 - 3. Expanding,

q_{n+1}^2 - 3 = \frac{(q_n^2 + 3)^2 - 12q_n^2}{4q_n^2} = \frac{e_n^2}{4q_n^2}

The first terms are \displaystyle q_2 = \frac{7}{4}, \displaystyle q_3 = \frac{97}{56}, \displaystyle q_4 = \frac{18817}{10864}, with \displaystyle e_2 = \frac{1}{16}, \displaystyle e_3 = \frac{1}{3136} and \displaystyle e_4 = \frac{1}{118{,}026{,}496}.

Since e_1 = 1 and e_{n+1} \ge 0, every q_n^2 \ge 3, so \displaystyle q_n > \frac{3}{2} and 4q_n^2 \ge 12. Then \displaystyle e_{n+1} \le \frac{e_n^2}{12}, and by induction 0 \le e_n \le 12^{-(n-1)}. Also \displaystyle q_n - q_{n+1} = \frac{e_n}{2q_n} \ge 0, and for m > n

\begin{aligned} 0 \le q_n - q_m &= \frac{q_n^2 - q_m^2}{q_n + q_m} = \frac{e_n - e_m}{q_n + q_m} \\ &\le \frac{e_n}{3} \le \frac{12^{-(n-1)}}{3} \end{aligned}

Given a rational \varepsilon > 0, take an integer \displaystyle N > \frac{1}{3\varepsilon}; as 12^N > N, \displaystyle \lvert q_m - q_n \rvert \le \frac{12^{-N}}{3} < \varepsilon for m > n > N. So \{q_n\} is Cauchy in \mathbb{Q}.

If q_n \to r with r rational, then q_n^2 \to r^2 (product law) and e_n \to 0 (squeeze), so r^2 = 3, impossible because \displaystyle \sqrt{3} is irrational (Irrational numbers). So \mathbb{Q} is not Cauchy-complete. In \mathbb{R}, Theorem 4 gives a limit r with r^2 = 3 and \displaystyle r \ge \frac{3}{2}, so \displaystyle q_n \to \sqrt{3}; already \displaystyle 0 < q_4 - \sqrt{3} \le \frac{e_4}{3} < 3 \times 10^{-9}.

Example 4

Kepler’s equation for a CubeSat orbit

Chloe and Daniel, on a university CubeSat team, test their orbit software on an ellipse of eccentricity \displaystyle \frac{1}{4}. At mean anomaly 1 radian, the angle they need (the eccentric anomaly x) solves Kepler’s equation

x - \frac{1}{4}\sin x = 1

which no formula solves. Daniel iterates x_1 = 1, \displaystyle x_{n+1} = 1 + \frac{1}{4}\sin x_n. Prove that the iterates converge to the only solution, and find the least n for which Theorem 5 guarantees \lvert L - x_n \rvert < 10^{-6}.

Solution

As in Example 2, now from \displaystyle \sin u - \sin v = 2\cos\frac{u + v}{2}\sin\frac{u - v}{2}, \lvert \sin u - \sin v \rvert \le \lvert u - v \rvert. So

\lvert x_{n+2} - x_{n+1} \rvert = \frac{1}{4}\lvert \sin x_{n+1} - \sin x_n \rvert \le \frac{1}{4}\lvert x_{n+1} - x_n \rvert

and the sequence is contractive with \displaystyle C = \frac{1}{4}. Theorem 5 gives a limit L. Since \lvert \sin x_n - \sin L \rvert \le \lvert x_n - L \rvert, \sin x_n \to \sin L, and letting n \to \infty in the recursion (tails, sum and product laws) gives \displaystyle L = 1 + \frac{1}{4}\sin L. A second solution L' would have \displaystyle \lvert L - L' \rvert = \frac{1}{4}\lvert \sin L - \sin L' \rvert \le \frac{1}{4}\lvert L - L' \rvert, forcing L' = L.

Here \displaystyle d = \frac{1}{4}\sin 1 and \displaystyle 1 - C = \frac{3}{4}, so Theorem 5 reads

\lvert L - x_n \rvert \le \frac{\sin 1}{3} \cdot 4^{-(n-1)}

with \displaystyle \frac{\sin 1}{3} \approx 0.28049. This is below 10^{-6} exactly when 4^{n-1} > 280{,}490.3; 4^9 = 262{,}144 falls short and 4^{10} = 1{,}048{,}576 suffices, so n = 11. The iterates x_2 \approx 1.210368, x_3 \approx 1.233936, x_4 \approx 1.235949, x_5 \approx 1.236115 settle on L \approx 1.236130. The bound is cautious: in fact \lvert L - x_7 \rvert \approx 1.0 \times 10^{-7}, but only Theorem 5 certifies a term in advance.

Example 5

Reciprocals of a Cauchy sequence

Let F be an ordered field and \{a_n\} a Cauchy sequence in F that does not converge to 0. Using only the definitions, prove: (a) there are \delta > 0 and N_0 with \lvert a_n \rvert > \delta for every n > N_0; (b) \displaystyle \left\{\frac{1}{a_n}\right\}_{n > N_0} is Cauchy.

Solution

Proof. (a) Since a_n \not\to 0, the negation theorem gives \varepsilon > 0 with \lvert a_n \rvert \ge \varepsilon for infinitely many n. Take N_0 with \displaystyle \lvert a_m - a_n \rvert < \frac{\varepsilon}{2} for all m, n > N_0, and fix m > N_0 with \lvert a_m \rvert \ge \varepsilon. For every n > N_0, \displaystyle \lvert a_n \rvert \ge \lvert a_m \rvert - \lvert a_m - a_n \rvert > \frac{\varepsilon}{2}, so \displaystyle \delta = \frac{\varepsilon}{2} works.

(b) For m, n > N_0,

\left\lvert \frac{1}{a_m} - \frac{1}{a_n} \right\rvert = \frac{\lvert a_n - a_m \rvert}{\lvert a_m \rvert \lvert a_n \rvert} < \frac{\lvert a_n - a_m \rvert}{\delta^2}

Given \eta > 0, take N_1 with \lvert a_m - a_n \rvert < \eta\delta^2 for all m, n > N_1. For m, n > \max\{N_0, N_1\}, \displaystyle \left\lvert \frac{1}{a_m} - \frac{1}{a_n} \right\rvert < \eta. \square

No limit appears, so the proof holds in \mathbb{Q}.

Boundaries

  • Completeness is needed. Example 3 is Cauchy in \mathbb{Q} with no limit in \mathbb{Q}: Theorem 4 is a property of \mathbb{R}, not of every ordered field.
  • Shrinking steps are not enough. In Example 2 every fixed gap j has a_{n+j} - a_n \to 0, and the sequence diverges. Theorem 5 needs one constant C < 1 for all steps (Exercise 13).
  • The Archimedean hypothesis. There are non-Archimedean ordered fields in which every Cauchy sequence converges but the least-upper-bound property fails; this is stated without proof. In them the positive integers are bounded, which is where the proof of Theorem 6 breaks.
  • Other spaces. Being Cauchy depends only on distances; converging also depends on which points the space contains: \displaystyle \left\{\frac{1}{n + 1}\right\} is Cauchy in (0, 1), but its limit 0 is not in (0, 1). A metric space in which every Cauchy sequence converges is called complete (Metric spaces, a first look).

Where this goes next

Series convergence, proved applies Theorem 4 to partial sums, giving the Cauchy criterion for series and, from it, that absolute convergence implies convergence. Pointwise and uniform convergence carries the idea to sequences of functions, and Metric spaces, a first look makes completeness a property of a space. Theorem 5 is the one-variable case of the contraction mapping principle, proved in a later analysis course.

Exercises

From the definition

  1. Let \displaystyle b_n = \frac{5n + 2}{2n + 1}. Prove from the definition that \{b_n\} is Cauchy, and find the least N that works for \displaystyle \varepsilon = \frac{1}{50}.
  2. Let \displaystyle c_n = \frac{(-1)^n}{n^2}. Show that \displaystyle \lvert c_m - c_n \rvert < \frac{2}{(N + 1)^2} for all m, n > N, and find the least N this bound certifies for \displaystyle \varepsilon = \frac{1}{10{,}000}.
  3. Decide which of these sequences are Cauchy, with a reason: (a) \displaystyle (-1)^n \frac{n}{n + 1}; (b) \displaystyle \frac{(-1)^n}{\sqrt{n}}; (c) \displaystyle \sqrt{n + 1} - \sqrt{n}; (d) \displaystyle n \sin\frac{1}{n}.
  4. Let a_n = \ln n. Show that a_{n+j} - a_n \to 0 for every fixed j \ge 1. For \displaystyle \varepsilon = \frac{1}{2} and n = 1001, find the least m > n with \lvert a_m - a_n \rvert \ge \varepsilon, and show that \{a_n\} is not Cauchy.

Contractive sequences

  1. Chloe runs the software of Example 4 on an ellipse of eccentricity \displaystyle \frac{1}{3} at mean anomaly 2 radians, solving \displaystyle x - \frac{1}{3}\sin x = 2 by x_1 = 2, \displaystyle x_{n+1} = 2 + \frac{1}{3}\sin x_n. Give x_2 and x_3 to six decimal places, and the least n for which Theorem 5 guarantees \lvert L - x_n \rvert < 10^{-8}.
  2. Let a_1 = 1, a_2 = 5 and \displaystyle a_{n+2} = \frac{2a_{n+1} + a_n}{3}. (a) Show that \displaystyle a_{n+2} - a_{n+1} = -\frac{1}{3}(a_{n+1} - a_n). (b) Show that \displaystyle a_{n+1} + \frac{1}{3}a_n does not depend on n, and find \lim_{n \to \infty} a_n. (c) Compare the bound of Theorem 5 with the exact value of \lvert L - a_n \rvert.
  3. Let q_1 = 2 and \displaystyle q_{n+1} = \frac{1}{2}\left(q_n + \frac{5}{q_n}\right). Find q_2, q_3 and q_n^2 - 5 for n = 2, 3, and adapt Example 3 to show that \{q_n\} is Cauchy in \mathbb{Q} with no rational limit.

Bounds and subsequences

  1. A Cauchy sequence \{a_n\} has \displaystyle a_{k^2} = \frac{3k - 1}{k + 2} for every k \ge 1. Find \lim_{n \to \infty} a_n and \lim_{n \to \infty} a_{2n}.
  2. Let \displaystyle c_n = \frac{n^2 - 10}{n^2 + 1}. Find the least N with \lvert c_m - c_n \rvert < 1 for all m, n > N, and the bound B that the proof of Theorem 2 then gives. Is it the least bound for \lvert c_n \rvert?

Proofs

  1. Prove from the definition alone that if \{a_n\} and \{b_n\} are Cauchy, so are \{a_n + b_n\} and \{a_n b_n\}.
  2. Prove that \{a_n\} is Cauchy if and only if for every \varepsilon > 0 there is N with \lvert a_n - a_N \rvert < \varepsilon for every n \ge N.
  3. Suppose \displaystyle \lvert a_{n+1} - a_n \rvert \le \frac{1}{n(n + 1)} for every n \ge 1. Prove that \{a_n\} is Cauchy and that its limit L satisfies \displaystyle \lvert L - a_n \rvert \le \frac{1}{n}.
  4. Let \displaystyle a_n = \sqrt{n}. Prove that \lvert a_{n+2} - a_{n+1} \rvert < \lvert a_{n+1} - a_n \rvert for every n \ge 1 but that \{a_n\} is not Cauchy, and explain why Theorem 5 does not apply.
  5. Give a second proof that a Cauchy sequence of real numbers converges, by showing \liminf a_n = \limsup a_n and using Theorem 7 of Subsequences and Bolzano-Weierstrass.
  6. Prove that every Cauchy sequence has a subsequence with \lvert a_{n_{k+1}} - a_{n_k} \rvert < 2^{-k} for every k \ge 1.
Answers
  1. N = 11. As \displaystyle b_n = \frac{5}{2} - \frac{1}{2(2n + 1)}, the argument of Example 1 gives \displaystyle \lvert b_m - b_n \rvert < \frac{1}{2(2N + 3)} for m, n > N, at most \displaystyle \frac{1}{50} when N \ge 11. For N \le 10 the pair n = 11, m = 144 gives \displaystyle \lvert b_m - b_n \rvert = \frac{1}{46} - \frac{1}{578} > \frac{1}{50}.
  2. N = 141. For m, n > N, \displaystyle \lvert c_m - c_n \rvert \le \frac{1}{m^2} + \frac{1}{n^2} \le \frac{2}{(N + 1)^2}, with equality only if m = n, when the difference is 0. The bound is at most \displaystyle \frac{1}{10{,}000} when (N + 1)^2 \ge 20{,}000; 141^2 = 19{,}881 and 142^2 = 20{,}164.
  3. (a) Not Cauchy: \displaystyle \lvert a_{n+1} - a_n \rvert = \frac{n}{n + 1} + \frac{n + 1}{n + 2} \ge \frac{1}{2} + \frac{2}{3} > 1 for every n. (b) Cauchy: it converges to 0 (Theorem 1). (c) Cauchy: it equals \displaystyle \frac{1}{\sqrt{n + 1} + \sqrt{n}} \to 0. (d) Cauchy: it converges to 1, as \displaystyle \frac{\sin x}{x} \to 1 when x \to 0 (The squeeze theorem and the limit of sine x over x).
  4. m = 1651. \displaystyle a_{n+j} - a_n = \ln\left(1 + \frac{j}{n}\right) \to 0 by continuity of \ln. For m > 1001, \displaystyle \ln\frac{m}{1001} \ge \frac{1}{2} if and only if \displaystyle m \ge 1001\sqrt{e} \approx 1650.37. For any N, the pair n = N + 1, m = 2n gives \displaystyle a_m - a_n = \ln 2 > \frac{1}{2}, so the sequence is not Cauchy.
  5. x_2 \approx 2.303099, x_3 \approx 2.247879, and n = 18. Here \displaystyle C = \frac{1}{3} and \displaystyle d = \frac{1}{3}\sin 2, so \displaystyle \lvert L - x_n \rvert \le \frac{\sin 2}{2} \cdot 3^{-(n-1)} with \displaystyle \frac{\sin 2}{2} \approx 0.454649. We need 3^{n-1} > 45{,}464{,}871.3; 3^{16} = 43{,}046{,}721 falls short and 3^{17} = 129{,}140{,}163 suffices.
  6. (a) \displaystyle a_{n+2} - a_{n+1} = \frac{2a_{n+1} + a_n - 3a_{n+1}}{3}, which is \displaystyle -\frac{1}{3}(a_{n+1} - a_n), so \displaystyle C = \frac{1}{3}. (b) \displaystyle a_{n+2} + \frac{1}{3}a_{n+1} = \frac{2a_{n+1} + a_n}{3} + \frac{a_{n+1}}{3}, which is \displaystyle a_{n+1} + \frac{1}{3}a_n; so the value is always \displaystyle a_2 + \frac{1}{3}a_1 = \frac{16}{3}. In the limit \displaystyle \frac{4}{3}L = \frac{16}{3}, so L = 4. (c) With d = 4, Theorem 5 gives \lvert L - a_n \rvert \le 6 \cdot 3^{-(n-1)}. By induction \displaystyle a_n = 4 - 3\left(-\frac{1}{3}\right)^{n-1}, so exactly \lvert L - a_n \rvert = 3^{2-n}, half the bound.
  7. \displaystyle q_2 = \frac{9}{4}, \displaystyle q_3 = \frac{161}{72}, \displaystyle q_2^2 - 5 = \frac{1}{16}, \displaystyle q_3^2 - 5 = \frac{1}{5184}. With e_n = q_n^2 - 5, Example 3’s computation gives \displaystyle e_{n+1} = \frac{e_n^2}{4q_n^2} \ge 0, so for n \ge 2, q_n > 2 and \displaystyle e_{n+1} \le \frac{e_n}{20}, hence \displaystyle 0 \le e_n \le \frac{1}{16} \cdot 20^{-(n-2)}. For m > n \ge 2, \displaystyle 0 \le q_n - q_m = \frac{e_n - e_m}{q_n + q_m} \le \frac{e_n}{4}, below any rational \varepsilon > 0 for large n. A rational limit r would have r^2 = 5, but \displaystyle \sqrt{5} is irrational.
  8. Both limits are 3. The indices k^2 increase and \displaystyle \frac{3k - 1}{k + 2} \to 3, so Theorem 3 gives a_n \to 3, and the subsequence \{a_{2n}\} also tends to 3 (Theorem 2 of Subsequences and Bolzano-Weierstrass).
  9. N = 3 and \displaystyle B = \frac{9}{2}, the least bound. As \displaystyle c_n = 1 - \frac{11}{n^2 + 1}, \displaystyle \lvert c_m - c_n \rvert < \frac{11}{(N + 1)^2 + 1} for m, n > N, which is \displaystyle \frac{11}{17} < 1 for N = 3; for N = 2 the pair n = 3, m = 11 gives \displaystyle \frac{11}{10} - \frac{11}{122} > 1. Then \displaystyle B = \max\left\{\frac{9}{2}, \frac{6}{5}, \frac{1}{10}, \frac{6}{17} + 1\right\} = \frac{9}{2} = \lvert c_1 \rvert.
  10. Proof. Let \varepsilon > 0 and take N with \displaystyle \lvert a_m - a_n \rvert < \frac{\varepsilon}{2} and \displaystyle \lvert b_m - b_n \rvert < \frac{\varepsilon}{2} for all m, n > N. By the triangle inequality the sums differ by at most \lvert a_m - a_n \rvert + \lvert b_m - b_n \rvert < \varepsilon. For the product, Theorem 2 gives A, B > 0 with \lvert a_n \rvert \le A and \lvert b_n \rvert \le B. Since a_m b_m - a_n b_n = a_m(b_m - b_n) + b_n(a_m - a_n), the products differ by at most A \lvert b_m - b_n \rvert + B \lvert a_m - a_n \rvert, which is less than \varepsilon once \displaystyle \lvert b_m - b_n \rvert < \frac{\varepsilon}{2A} and \displaystyle \lvert a_m - a_n \rvert < \frac{\varepsilon}{2B}, true for all m, n beyond some N. \square
  11. Proof. If \{a_n\} is Cauchy and \varepsilon > 0, take N with \lvert a_m - a_n \rvert < \varepsilon for all m, n > N; the anchor N + 1 has \lvert a_n - a_{N+1} \rvert < \varepsilon for every n \ge N + 1. Conversely, given \varepsilon > 0, take N with \displaystyle \lvert a_n - a_N \rvert < \frac{\varepsilon}{2} for every n \ge N. For m, n > N, \lvert a_m - a_n \rvert \le \lvert a_m - a_N \rvert + \lvert a_N - a_n \rvert < \varepsilon. \square
  12. Proof. Since \displaystyle \frac{1}{k(k + 1)} = \frac{1}{k} - \frac{1}{k + 1}, for m > n the triangle inequality and telescoping give \displaystyle \lvert a_m - a_n \rvert \le \frac{1}{n} - \frac{1}{m} < \frac{1}{n}. Given \varepsilon > 0, take an integer \displaystyle N > \frac{1}{\varepsilon} (The Archimedean property and density of the rationals); for m > n > N, \displaystyle \lvert a_m - a_n \rvert < \frac{1}{n} < \varepsilon. So \{a_n\} is Cauchy, with a limit L by Theorem 4. For fixed n and every m > n, \displaystyle -\frac{1}{n} \le a_m - a_n \le \frac{1}{n}, and the order limit theorem gives \displaystyle \lvert L - a_n \rvert \le \frac{1}{n}. \square
  13. Proof. \displaystyle a_{n+1} - a_n = \frac{1}{\sqrt{n + 1} + \sqrt{n}}, whose denominator increases with n, so the steps strictly decrease. But \displaystyle a_{4n} - a_n = \sqrt{n} \ge 1; given any N, the pair n = N + 1, m = 4n breaks the condition with \varepsilon = 1. Theorem 5 needs one C < 1 for every step, but the ratio of consecutive steps, \displaystyle \frac{\sqrt{n + 1} + \sqrt{n}}{\sqrt{n + 2} + \sqrt{n + 1}}, tends to 1. \square
  14. Proof. A Cauchy sequence is bounded (Theorem 2), so S = \limsup a_n and I = \liminf a_n are subsequential limits by Theorem 7(b) of Subsequences and Bolzano-Weierstrass: a_{n_k} \to S and a_{m_k} \to I. Let \varepsilon > 0. Take N with \displaystyle \lvert a_p - a_q \rvert < \frac{\varepsilon}{3} for all p, q > N, and k > N so large that \displaystyle \lvert a_{n_k} - S \rvert < \frac{\varepsilon}{3} and \displaystyle \lvert a_{m_k} - I \rvert < \frac{\varepsilon}{3}. As n_k, m_k \ge k > N, also \displaystyle \lvert a_{n_k} - a_{m_k} \rvert < \frac{\varepsilon}{3}, and the triangle inequality gives \lvert S - I \rvert < \varepsilon. As \varepsilon is arbitrary, S = I, and Theorem 7(c) gives convergence. \square
  15. Proof. For each k \ge 1 choose N_k with \lvert a_m - a_n \rvert < 2^{-k} for all m, n > N_k. A larger cutoff also works, so replacing N_k by \max\{N_1, \ldots, N_k\} + k we may assume N_1 < N_2 < \cdots. Let n_k = N_k + 1. Then n_1 < n_2 < \cdots, and n_k and n_{k+1} both exceed N_k, so \lvert a_{n_{k+1}} - a_{n_k} \rvert < 2^{-k}. \square
math.wiki · CC-BY-SA-4.0