math.wiki
Analysis › Sequences and series › The monotone convergence theorem
Sign in

The monotone convergence theorem

A sequence that only climbs and never passes a ceiling must settle somewhere. Why is that fact exactly as strong as the completeness of the real numbers, and what can it build, from intervals that close in on a single point to the number e?

Where you are 5 in Sequences and series

Why this exists

Convergence of sequences proved the monotone convergence theorem: a nondecreasing sequence that is bounded above converges, to the least upper bound of its terms. It used the least-upper-bound property of Least upper bounds and completeness once. This lesson shows the theorem is worth exactly as much as completeness: an ordered field has the least-upper-bound property if and only if every bounded monotone sequence in it converges, so either statement can serve as the axiom that separates \mathbb{R} from \mathbb{Q}.

It also builds numbers: nested closed intervals close in on a point, and e is trapped this way between \displaystyle \left(1 + \frac{1}{n}\right)^n and \displaystyle \left(1 + \frac{1}{n}\right)^{n+1}.

Definition

A sequence \{a_n\} is nondecreasing if a_n \le a_{n+1} for every n \ge 1, increasing if a_n < a_{n+1} for every n, and nonincreasing or decreasing with the inequalities reversed. It is monotone if it is nondecreasing or nonincreasing. It is bounded above if some M has a_n \le M for every n, bounded below if some m has a_n \ge m for every n, and bounded if both hold.

Nested intervals. A sequence of intervals \{I_n\} is nested if I_{n+1} \subseteq I_n for every n \ge 1. Its intersection \bigcap_{n=1}^{\infty} I_n is the set of real numbers lying in every I_n. A closed bounded interval [a_n, b_n] has length b_n - a_n.

Convergence in an ordered field. Theorem 2 works in an ordered field F, as in The real numbers as an ordered field, that need not be \mathbb{R}. There a positive integer n stands for 1 + 1 + \cdots + 1 (n terms), and a_n \to L means: for every \varepsilon > 0 in F there is a positive integer N with \lvert a_n - L \rvert < \varepsilon for every n > N. In \mathbb{R} this is the definition of Limits of sequences, epsilon and N.

Intuition

A nondecreasing sequence bounded above has a set of terms with a least upper bound s. Every number below s is passed by some term, and once passed it stays passed, because the terms never come down. So the terms crowd in just below s.

Turned around: to find \sup E using only monotone sequences, halve an interval that straddles the top of E again and again. The left ends climb, the right ends fall, and their common limit is \sup E. Nested intervals are this picture in general (Figure 1).

2
2.5
3
3.5
4
e
n = 1
n = 2
n = 3
n = 4
n = 5
n = 6
Figure 1The intervals \displaystyle \left[\left(1 + \frac{1}{n}\right)^n, \left(1 + \frac{1}{n}\right)^{n+1}\right] for n = 1 to 6. Each lies inside the one before, their lengths shrink to 0, and the one point they all share is e.

Results

Results of Least upper bounds and completeness are cited as C1 to C7 (C2 is the approximation property, C3 gives \inf from \sup). The sum and product laws, uniqueness, tails theorem and order limit theorem are those of Limits of sequences, epsilon and N; the squeeze theorem is in Convergence of sequences.

Theorem 1 (monotone convergence). A monotone sequence of real numbers converges if and only if it is bounded. If \{a_n\} is nondecreasing and bounded above, then a_n \to \sup\{a_n : n \ge 1\}; if it is nonincreasing and bounded below, then a_n \to \inf\{a_n : n \ge 1\}. By the tails theorem, it is enough that the sequence be monotone from some index on.

Theorem 2 (monotone convergence gives completeness). If every nondecreasing sequence that is bounded above converges in the ordered field F, then F has the least-upper-bound property. Since the proof of Theorem 1 uses only the ordered field rules and the least-upper-bound property, the two properties are equivalent in every ordered field.

Theorem 3 (nested interval property). Let I_n = [a_n, b_n], n \ge 1, be nested closed bounded intervals of \mathbb{R}. Then a_n \to \alpha = \sup\{a_n : n \ge 1\} and b_n \to \beta = \inf\{b_n : n \ge 1\}, \alpha \le \beta, and

\bigcap_{n=1}^{\infty} I_n = [\alpha, \beta]

So the intersection is not empty, and if b_n - a_n \to 0 it is a single point.

Theorem 4 (Bernoulli’s inequality). If x > -1, x \ne 0 and m \ge 2 is an integer, then (1 + x)^m > 1 + mx.

Theorem 5 (the number e). Let \displaystyle a_n = \left(1 + \frac{1}{n}\right)^n and \displaystyle b_n = \left(1 + \frac{1}{n}\right)^{n+1}. Then \{a_n\} increases, \{b_n\} decreases, a_n < b_n, and b_n - a_n \to 0. The intervals [a_n, b_n] are nested and share exactly one point, called e; so a_n \to e, b_n \to e, and a_n < e < b_n for every n \ge 1.

From a_1 = 2 and \displaystyle b_5 = \frac{46656}{15625} \approx 2.986, 2 < e < 3. This is the e of The number e and the natural exponential function.

Why it is true

Theorem 1. Proof. A convergent sequence is bounded (Convergence of sequences). Conversely, a nondecreasing sequence has a_n \ge a_1, so it is bounded once it is bounded above. If it is, Convergence of sequences proves a_n \to s = \sup\{a_n : n \ge 1\}: for \varepsilon > 0, s - \varepsilon is not an upper bound (C2), so some a_N > s - \varepsilon, and for n > N, s - \varepsilon < a_N \le a_n \le s. If \{a_n\} is nonincreasing and bounded below, \{-a_n\} is nondecreasing and bounded above, so -a_n \to \sup\{-a_n : n \ge 1\} = -\inf\{a_n : n \ge 1\} by C3, and the product law with the constant -1 finishes. \square

Theorem 2. Proof. The sum and product laws, uniqueness, tails, squeeze and order limit theorems have the same proofs in F as in \mathbb{R}, since those proofs use only the ordered field rules.

F is Archimedean. If the increasing sequence a_n = n were bounded above, then n \to L for some L \in F. By the tails theorem n + 1 \to L, and by the sum law n + 1 \to L + 1, so uniqueness gives L = L + 1, which is false. So every element of F is below some positive integer, and \displaystyle \frac{1}{n} \to 0: given \varepsilon > 0, take an integer \displaystyle N > \frac{1}{\varepsilon}; for n > N, \displaystyle 0 < \frac{1}{n} < \varepsilon.

Bisection. Let E \subseteq F be nonempty and bounded above, with x_0 \in E and an upper bound u_0. Set a_1 = x_0, b_1 = u_0. Given a_n \le b_n, let \displaystyle m = \frac{a_n + b_n}{2}; if m is an upper bound of E, set a_{n+1} = a_n and b_{n+1} = m, and otherwise a_{n+1} = m and b_{n+1} = b_n. By induction: every b_n is an upper bound of E; every a_n has some x \in E with x \ge a_n (for a_1 take x_0; if m is not an upper bound, some x \in E exceeds m); a_n \le a_{n+1} \le b_{n+1} \le b_n; and \displaystyle b_n - a_n = \frac{u_0 - x_0}{2^{n-1}}.

The limits. \{a_n\} is nondecreasing and bounded above by b_1, so a_n \to s for some s. \{-b_n\} is nondecreasing and bounded above by -a_1, so it converges, and b_n \to t by the product law. Since 2^{n-1} \ge n (by induction, 2^n \ge 2n \ge n + 1), \displaystyle 0 \le b_n - a_n \le \frac{u_0 - x_0}{n}, so b_n - a_n \to 0 by the squeeze theorem, and t = s by the sum law.

s = \sup E. If x \in E, then x \le b_n for every n, so x \le t = s by the order limit theorem. If v is an upper bound of E, each a_n has some x \in E with a_n \le x \le v, so s \le v. \square

Theorem 3. Proof. Since a_{n+1}, b_{n+1} \in I_{n+1} \subseteq I_n, a_n \le a_{n+1} \le b_{n+1} \le b_n. Every a_n is at most every b_m: if n \le m, a_n \le a_m \le b_m; if n > m, a_n \le b_n \le b_m. So \{a_n\} is nondecreasing and bounded above by b_1, \{b_n\} is nonincreasing and bounded below by a_1, and Theorem 1 gives a_n \to \alpha, b_n \to \beta. Each b_m is an upper bound of the a_n, so \alpha \le b_m; then \alpha is a lower bound of the b_m, so \alpha \le \beta.

A real x lies in every I_n exactly when it is an upper bound of the a_n and a lower bound of the b_n, that is, when \alpha \le x \le \beta. If b_n - a_n \to 0, the sum law gives \beta - \alpha = 0. \square

Theorem 4. Proof. By induction on m. For m = 2, (1 + x)^2 = 1 + 2x + x^2 > 1 + 2x since x \ne 0. If (1 + x)^m > 1 + mx, multiply by 1 + x > 0:

\begin{aligned} (1 + x)^{m+1} &> (1 + mx)(1 + x) \\ &= 1 + (m+1)x + mx^2 \\ &\ge 1 + (m+1)x \end{aligned}

\square

Theorem 5. Proof. \{a_n\} increases. With a common exponent,

\begin{aligned} \frac{a_{n+1}}{a_n} &= \left(\frac{n(n+2)}{(n+1)^2}\right)^{n+1} \frac{n+1}{n} \\ &= \left(1 - \frac{1}{(n+1)^2}\right)^{n+1} \frac{n+1}{n} \end{aligned}

Theorem 4 with \displaystyle x = -\frac{1}{(n+1)^2} and m = n + 1 makes the power greater than \displaystyle 1 - \frac{1}{n+1} = \frac{n}{n+1}, so \displaystyle \frac{a_{n+1}}{a_n} > 1.

\{b_n\} decreases. For n \ge 2,

\begin{aligned} \frac{b_{n-1}}{b_n} &= \left(\frac{n^2}{(n-1)(n+1)}\right)^{n} \frac{n}{n+1} \\ &= \left(1 + \frac{1}{n^2 - 1}\right)^{n} \frac{n}{n+1} \end{aligned}

Theorem 4 with \displaystyle x = \frac{1}{n^2 - 1} and m = n makes the power greater than \displaystyle 1 + \frac{n}{n^2 - 1} > \frac{n+1}{n}, so \displaystyle \frac{b_{n-1}}{b_n} > 1.

Nested and shrinking. Since \displaystyle b_n = a_n\left(1 + \frac{1}{n}\right) > a_n, a_n < a_{n+1} < b_{n+1} < b_n. The lengths satisfy \displaystyle 0 < b_n - a_n = \frac{a_n}{n} < \frac{b_1}{n} = \frac{4}{n}, so they tend to 0. Theorem 3 gives the single point e, and strict monotonicity gives a_n < a_{n+1} \le e \le b_{n+1} < b_n. \square

Worked examples

Example 1

Partial sums with no formula

Show that \displaystyle a_n = \sum_{k=1}^{n} \frac{1}{k\,2^k} converges, and find an interval of length \displaystyle \frac{1}{80} that contains its limit.

Solution

Each step adds a positive term, so the sequence increases. Each term is at most \displaystyle \frac{1}{2^k}, and the geometric sum gives \displaystyle a_n \le 1 - \frac{1}{2^n} < 1. By Theorem 1, a_n \to L, with a_n < a_{n+1} \le L.

To bracket L, compare the tail with a geometric one: for m > n, every k > n has k \ge n + 1, so

\begin{aligned} a_m - a_n &= \sum_{k=n+1}^{m} \frac{1}{k\,2^k} \\ &\le \frac{1}{n+1} \sum_{k=n+1}^{m} \frac{1}{2^k} < \frac{1}{(n+1)\,2^n} \end{aligned}

Letting m \to \infty, the order limit theorem gives \displaystyle a_n < L \le a_n + \frac{1}{(n+1)2^n}. With n = 4, \displaystyle a_4 = \frac{1}{2} + \frac{1}{8} + \frac{1}{24} + \frac{1}{64} = \frac{131}{192} and the bound is \displaystyle \frac{1}{80}, so \displaystyle \frac{131}{192} < L \le \frac{667}{960}, that is, 0.6822 < L < 0.6949.

The sum is the Taylor and Maclaurin series of -\ln(1 - x) at \displaystyle x = \frac{1}{2}, so L = \ln 2 \approx 0.6931, inside the bracket.

Example 2

Personal bests at the speedcubing club

At a speedcubing club, Hiro’s nth solve takes \displaystyle t_n = 15 + 2(-1)^n + \frac{6}{n} seconds. Jasmine, the record keeper, tracks his personal best p_n = \min\{t_1, \ldots, t_n\} and his slowest time w_n = \max\{t_1, \ldots, t_n\}. Do the times converge? Do the records?

Solution

Odd solves take \displaystyle 13 + \frac{6}{n} and even solves \displaystyle 17 + \frac{6}{n}; the first five are 19, 20, 15, 18.5, 14.2. Every L is at least 2 from 13 or from 17. Odd n \ge 7 have t_n within 1 of 13, even n \ge 6 within 1 of 17, so either way \lvert t_n - L \rvert \ge 1 infinitely often; by the negation theorem of Limits of sequences, epsilon and N the times diverge.

A minimum over more times can only be lower, so p_{n+1} \le p_n, and every p_n > 13. By Theorem 1, p_n \to \inf\{p_n : n \ge 1\}, which equals \inf\{t_n : n \ge 1\} because each p_n is a time and each t_n \ge p_n. Now 13 is a lower bound of the times, and every odd \displaystyle n > \frac{6}{\varepsilon} has t_n < 13 + \varepsilon, so the infimum is 13 by C2. The personal best tends to 13 seconds, a time Hiro never records (Figure 2).

5
10
15
13
17
20
n
t_n
p_n
Figure 2Hiro’s times \displaystyle t_n = 15 + 2(-1)^n + \frac{6}{n} do not converge, but his personal best p_n steps down and converges to \inf\{t_n : n \ge 1\} = 13, a time he never records.

The slowest time is nondecreasing and bounded above by 20 (odd times are at most 19, even times at most 20), so it converges to \sup\{t_n : n \ge 1\} = 20 = t_2; indeed w_n = 20 for n \ge 2.

Example 3

One rule, three starting values

Let \displaystyle a_{n+1} = \frac{a_n^2 + 2}{3}. Decide whether \{a_n\} converges, and find the limit, for a_1 = 0, \displaystyle a_1 = \frac{3}{2} and a_1 = 3.

Solution

Everything rests on one factorization,

\begin{aligned} a_{n+1} - a_n &= \frac{a_n^2 - 3a_n + 2}{3} \\ &= \frac{(a_n - 1)(a_n - 2)}{3} \end{aligned}

and on one fact: if a_n \to L, then a_{n+1} \to L (tails) and \displaystyle \frac{a_n^2 + 2}{3} \to \frac{L^2 + 2}{3} (product and sum laws), so uniqueness gives (L - 1)(L - 2) = 0.

Start at 0. If 0 \le a_n < 1, then \displaystyle \frac{2}{3} \le a_{n+1} < 1. By induction every term lies in [0, 1), both factors are negative, and the sequence increases. It converges, and the order limit theorem gives L \le 1, so L = 1. The terms run \displaystyle 0, \frac{2}{3}, \frac{22}{27}, \ldots

Start at \displaystyle \frac{3}{2}. If 1 < a_n < 2, then \displaystyle 1 < a_{n+1} < \frac{4 + 2}{3} = 2. Every term lies in (1, 2), the factors have opposite signs, and the sequence decreases, so it converges, and \displaystyle L \le \frac{3}{2} rules out 2. Again L = 1, through \displaystyle \frac{3}{2}, \frac{17}{12}, \frac{577}{432}, \ldots

Start at 3. If a_n > 2, then a_{n+1} > 2, so every term exceeds 2 and the sequence increases. If it were bounded, it would converge to some L \ge 3, but L is 1 or 2. So it is unbounded, and diverges: \displaystyle 3, \frac{11}{3}, \frac{139}{27}, \ldots

Example 4

Pinning down e

Theorem 5 traps e in [a_n, b_n]. Find the first n for which this interval has length less than 0.01, and the interval.

Solution

The length is \displaystyle \frac{a_n}{n}, and these lengths decrease: \displaystyle \frac{a_{n+1}}{a_n} is a power of \displaystyle 1 - \frac{1}{(n+1)^2} < 1 times \displaystyle \frac{n+1}{n}, so \displaystyle \frac{a_{n+1}}{n+1} < \frac{a_n}{n}. So we want the first n with n > 100\,a_n. Computing, a_{271} \approx 2.71328, so 100\,a_{271} \approx 271.33 > 271 and n = 271 fails, while 100\,a_{272} \approx 271.33 < 272. The first n is 272, and

2.71330 < e < 2.72328
Example 5

A sequence that may step down a little

Let \{a_n\} be bounded above, with \displaystyle a_{n+1} \ge a_n - \frac{1}{2^n} for every n \ge 1. Prove that \{a_n\} converges.

Solution

Proof. The sequence may decrease, so Theorem 1 does not apply directly. Absorb the drops into a correction: let \displaystyle b_n = a_n - \frac{1}{2^{n-1}}. Then

b_{n+1} - b_n = a_{n+1} - a_n + \frac{1}{2^n} \ge 0

So \{b_n\} is nondecreasing, and b_n < a_n \le M for an upper bound M of \{a_n\}. By Theorem 1, b_n \to L. Since \displaystyle 0 < \frac{1}{2^{n-1}} \le \frac{1}{n}, the squeeze theorem gives \displaystyle \frac{1}{2^{n-1}} \to 0, and the sum law gives \displaystyle a_n = b_n + \frac{1}{2^{n-1}} \to L. \square

Exercise 14 allows drops of \displaystyle \frac{1}{n^2}.

Example 6

The arithmetic-geometric mean of 1 and 9

Let a_1 = 1, b_1 = 9, \displaystyle a_{n+1} = \sqrt{a_n b_n} and \displaystyle b_{n+1} = \frac{a_n + b_n}{2}. Prove that both sequences converge to the same limit, and find it to seven decimal places.

Solution

Proof. For positive x \ne y,

\frac{x + y}{2} - \sqrt{xy} = \frac{\left(\sqrt{x} - \sqrt{y}\right)^2}{2} > 0

So if 0 < a_n < b_n, then

\begin{aligned} a_n = \sqrt{a_n a_n} &< a_{n+1} < b_{n+1} \\ &< \frac{b_n + b_n}{2} = b_n \end{aligned}

By induction the intervals [a_n, b_n] are nested. Their lengths at least halve, since \displaystyle b_{n+1} - a_{n+1} < b_{n+1} - a_n = \frac{b_n - a_n}{2}, so \displaystyle b_n - a_n \le \frac{8}{2^{n-1}} \le \frac{8}{n} \to 0. Theorem 3 gives one number M with a_n \to M and b_n \to M. \square

The intervals are [1, 9], [3, 5], \displaystyle [\sqrt{15}, 4] \approx [3.872983, 4], [3.9359793, 3.9364917] and [3.93623550, 3.93623551]. Both ends of the last round to 3.9362355, so M \approx 3.9362355. The lengths 8, 2, 0.127, 0.00051, 8 \times 10^{-9} shrink far faster than halving. The nested intervals define M, the arithmetic-geometric mean of 1 and 9.

Boundaries

  • Monotone and bounded, both. Each alone fails, as Convergence of sequences shows with (-1)^n and n.
  • Closed and bounded intervals, both. The intervals \displaystyle (0, \frac{1}{n}] are nested and bounded but not closed, and share no point: x \le 0 is in none, and x > 0 misses those with \displaystyle \frac{1}{n} < x. The intervals [n, \infty) are closed and nested but unbounded, and share no point by the The Archimedean property and density of the rationals. Theorem 3 is a guarantee, not a description: the open intervals \displaystyle (-\frac{1}{n}, \frac{1}{n}) still share 0.
  • Completeness is the hypothesis. In \mathbb{Q}, which lacks the least-upper-bound property (C6), Theorem 2 says some nondecreasing bounded sequence has no limit; the decimal truncations of \displaystyle \sqrt{2} are one. The nested interval property fails in \mathbb{Q} too (Exercise 16).
  • Nested intervals return completeness only with the Archimedean property. Exercise 15 proves that an Archimedean ordered field with the nested interval property is complete. Some non-Archimedean ordered fields, the hyperreal numbers among them, have the nested interval property but are not complete; this is stated without proof, and belongs to mathematical logic.

Where this goes next

Subsequences and Bolzano-Weierstrass finds a convergent subsequence of every bounded sequence, by a monotone subsequence or by bisection. Cauchy sequences tests convergence without naming a limit. Series convergence, proved applies Theorem 1 to partial sums, as in Example 1, to prove the comparison test. Countable and uncountable sets uses nested intervals to show \mathbb{R} is uncountable, and Extreme and intermediate value theorems, proved bisects to prove the extreme and intermediate value theorems.

Exercises

Monotone and bounded

  1. Show that \displaystyle a_n = \sum_{k=1}^{n} \frac{1}{3^k + 1} converges, and that its limit L satisfies \displaystyle \frac{27}{70} < L \le \frac{382}{945}.
  2. Let a_n = \dfrac{1 \cdot 3 \cdot 5 \cdots (2n - 1)}{2 \cdot 4 \cdot 6 \cdots (2n)}. (a) Show that \{a_n\} converges. (b) Show that \displaystyle a_{n+1} = \frac{2n + 1}{2n + 2}\,a_n does not determine the limit. (c) Prove that \displaystyle a_n^2 < \frac{1}{2n + 1}, conclude that a_n \to 0, and find the least n for which this bound guarantees a_n < 0.01.

Recursive sequences

  1. Let a_1 = 1 and a_{n+1} = \dfrac{2a_n + 1}{a_n + 1}. Show that \{a_n\} is increasing and bounded above by \displaystyle \varphi = \frac{1 + \sqrt{5}}{2}, and find its limit.
  2. Let a_{n+1} = 4 - \dfrac{3}{a_n}. Find the limit when a_1 = 2 and when a_1 = 5, proving monotonicity and a bound in each case.
  3. Let a_1 = 1 and a_{n+1} = a_n + \dfrac{1}{a_n}. Prove that \{a_n\} is increasing and not bounded above. Show that a_n^2 \ge 2n - 1, and conclude that a_n > 10 for every n \ge 51.

Nested intervals

  1. Find \bigcap_{n=1}^{\infty} I_n, or show it is empty, for (a) \displaystyle I_n = \left[1 - \frac{1}{n}, 1 + \frac{1}{n^2}\right]; (b) \displaystyle I_n = \left[-\frac{1}{n}, 2 + \frac{1}{n}\right]; (c) \displaystyle I_n = \left(2, 2 + \frac{1}{n}\right); (d) I_n = (-\infty, -n]. For each empty intersection, name the hypothesis of Theorem 3 that fails.
  2. Let a_1 = 2, b_1 = 8, \displaystyle a_{n+1} = \sqrt{a_n b_n} and \displaystyle b_{n+1} = \frac{a_n + b_n}{2}. List [a_n, b_n] for n \le 4, and use [a_5, b_5] to give the common limit to six decimal places.

The number e

  1. Using that \displaystyle b_n = \left(1 + \frac{1}{n}\right)^{n+1} decreases, find the first n with b_n < 2.75, and state what it proves about e.
  2. Show that \displaystyle c_n = \left(1 - \frac{1}{n}\right)^n, for n \ge 2, is increasing and converges to \displaystyle \frac{1}{e}.
  3. Show that \displaystyle \left(1 + \frac{1}{2n}\right)^n \to \sqrt{e}, and compute the term at n = 10.

Proofs

  1. Let \{a_n\} be nondecreasing with a_n \to L. Without Theorem 1, prove that a_n \le L for every n and that L = \sup\{a_n : n \ge 1\}.
  2. Let \{a_n\} be nondecreasing and not bounded above. Prove that for every M there is N with a_n > M for every n > N.
  3. Let \displaystyle h_n = 1 + \frac{1}{2} + \cdots + \frac{1}{n}. Prove that \displaystyle h_{2n} - h_n \ge \frac{1}{2}, that \{h_n\} diverges, and, by Theorem 1, that it is not bounded above, although h_{n+1} - h_n \to 0.
  4. Let \{a_n\} be bounded above with \displaystyle a_{n+1} \ge a_n - \frac{1}{n^2} for every n \ge 1. Prove that \{a_n\} converges.
  5. Let F be an Archimedean ordered field in which every sequence of nested closed bounded intervals has a common point. Prove that F has the least-upper-bound property.
  6. Find nested closed intervals with rational endpoints whose intersection in \mathbb{R} is a single irrational number, and conclude that the nested interval property fails in \mathbb{Q}.
Answers
  1. The terms are positive, so \{a_n\} increases, and \displaystyle a_n < \sum_{k=1}^{n} \frac{1}{3^k} < \frac{1}{2}; Theorem 1 applies. Then \displaystyle a_3 = \frac{1}{4} + \frac{1}{10} + \frac{1}{28} = \frac{27}{70} and L \ge a_4 > a_3. For m > 3, \displaystyle a_m - a_3 < \sum_{k=4}^{m} \frac{1}{3^k} < \frac{1}{54}, so \displaystyle L \le \frac{27}{70} + \frac{1}{54} = \frac{382}{945}. (Numerically L \approx 0.40406.)
  2. (a) The terms are positive and \displaystyle \frac{a_{n+1}}{a_n} = \frac{2n + 1}{2n + 2} < 1, so the sequence decreases and is bounded below by 0. (b) Passing to the limit gives L = 1 \cdot L, true for every L. (c) n = 5000. Since (2k - 1)(2k + 1) < 4k^2, \displaystyle \frac{2k - 1}{2k} < \frac{2k}{2k + 1}, so \displaystyle a_n^2 < \prod_{k=1}^{n} \frac{2k - 1}{2k + 1} = \frac{1}{2n + 1}, the product telescoping. The squeeze theorem gives a_n \to 0, and \displaystyle \frac{1}{\sqrt{2n + 1}} \le 0.01 exactly when 2n + 1 \ge 10{,}000.
  3. Limit \varphi; the terms are \displaystyle 1, \frac{3}{2}, \frac{8}{5}, \frac{21}{13}, \ldots The map \displaystyle f(x) = 2 - \frac{1}{x + 1} increases for x > -1, and f(\varphi) = \varphi because \varphi^2 = \varphi + 1. So 0 < a_n < \varphi gives 0 < a_{n+1} < \varphi, and \displaystyle a_{n+1} - a_n = \frac{1 + a_n - a_n^2}{a_n + 1} > 0, since x^2 - x - 1 < 0 between its roots \displaystyle \frac{1 - \sqrt{5}}{2} and \varphi. The limit solves L^2 - L - 1 = 0 (quotient law), and L \ge 1 gives L = \varphi.
  4. Limit 3 both times, with \displaystyle a_{n+1} - a_n = -\frac{(a_n - 1)(a_n - 3)}{a_n}. From 2: if 1 < a_n < 3, then \displaystyle 1 < \frac{3}{a_n} < 3, so 1 < a_{n+1} < 3 and the difference is positive; the terms \displaystyle 2, \frac{5}{2}, \frac{14}{5}, \frac{41}{14} increase. From 5: if a_n > 3, then \displaystyle 0 < \frac{3}{a_n} < 1, so a_{n+1} > 3 and the difference is negative; the terms \displaystyle 5, \frac{17}{5}, \frac{53}{17} decrease. The limit solves L^2 - 4L + 3 = 0, and L \ge 2 in both cases gives L = 3.
  5. By induction a_n \ge 1, and \displaystyle a_{n+1} - a_n = \frac{1}{a_n} > 0. If \{a_n\} were bounded, Theorem 1 would give a_n \to L \ge 1 and then \displaystyle L = L + \frac{1}{L}, impossible. Next \displaystyle a_{n+1}^2 = a_n^2 + 2 + \frac{1}{a_n^2} > a_n^2 + 2 and a_1^2 = 1, so a_n^2 \ge 2n - 1. For n \ge 51, a_n^2 \ge 101, so a_n > 10.
  6. (a) \{1\}: left ends increase to 1, right ends decrease to 1. (b) [0, 2]. (c) Empty: a common point needs x > 2 and \displaystyle x < 2 + \frac{1}{n} for every n, so x \le 2; not closed. (d) Empty: a common point would have x \le -n for every n, against the The Archimedean property and density of the rationals; not bounded.
  7. [2, 8], [4, 5], \displaystyle [2\sqrt{5}, 4.5] \approx [4.472136, 4.5], [4.4860463, 4.4860680]. Then a_5 \approx 4.48605716057 and b_5 \approx 4.48605716058, so the limit is 4.486057.
  8. n = 43: b_{42} \approx 2.75058 and b_{43} \approx 2.74983. Since e < b_{43}, e < 2.75.
  9. \displaystyle c_n = \left(\frac{n - 1}{n}\right)^n = \frac{1}{b_{n-1}}. The b_n are positive and decrease, so the c_n increase, and b_{n-1} \to e \ne 0 (tails), so the quotient law gives \displaystyle c_n \to \frac{1}{e}.
  10. About 1.6289. The term is \displaystyle \sqrt{a_{2n}}. If \lvert a_m - e \rvert < \varepsilon for m > N, the same holds for m = 2n when n > N, so a_{2n} \to e, and square roots pass to the limit (Example 3 of Limits of sequences, epsilon and N): the terms tend to \displaystyle \sqrt{e} \approx 1.6487. At n = 10 the term is \displaystyle \left(\frac{21}{20}\right)^{10} \approx 1.6289.
  11. Fix m. For n \ge m, a_n \ge a_m, so the order limit theorem on the tail from m gives L \ge a_m: L is an upper bound. If v is any upper bound, a_n \le v for every n, so L \le v. Hence L = \sup\{a_n : n \ge 1\}.
  12. Since M is not an upper bound, some a_N > M, and for n > N, a_n \ge a_N > M.
  13. \displaystyle h_{2n} - h_n = \sum_{k=n+1}^{2n} \frac{1}{k} \ge n \cdot \frac{1}{2n} = \frac{1}{2}. If h_n \to L, take N for \displaystyle \varepsilon = \frac{1}{4}; for n > N also 2n > N, so \displaystyle h_{2n} - h_n < \frac{1}{2}, a contradiction. So \{h_n\} diverges; it increases, and by Theorem 1 a bounded one would converge, so it is not bounded above. Yet \displaystyle h_{n+1} - h_n = \frac{1}{n+1} \to 0.
  14. Let \displaystyle S_n = \sum_{k=1}^{n} \frac{1}{k^2}, S_0 = 0. For k \ge 2, \displaystyle \frac{1}{k^2} < \frac{1}{k - 1} - \frac{1}{k}, so S_n < 2, and S_n \to S by Theorem 1. Let b_n = a_n + S_{n-1}. Then \displaystyle b_{n+1} - b_n = a_{n+1} - a_n + \frac{1}{n^2} \ge 0 and b_n < M + 2 for an upper bound M of \{a_n\}, so b_n \to B. Since S_{n-1} \to S (tails), a_n = b_n - S_{n-1} \to B - S.
  15. Let E be nonempty and bounded above, and bisect as in the proof of Theorem 2: each b_n is an upper bound of E, each a_n is at most some point of E, the intervals are nested, and \displaystyle b_n - a_n \le \frac{u_0 - x_0}{n}. The hypothesis gives s in every [a_n, b_n]. If some x \in E had x > s, the Archimedean property gives n with \displaystyle \frac{u_0 - x_0}{n} < x - s; since a_n \le s, b_n - s \le b_n - a_n < x - s, so b_n < x, a contradiction. If an upper bound v had v < s, choose n with b_n - a_n < s - v; then a_n > b_n - (s - v) \ge v, and the point x \in E with x \ge a_n exceeds v, a contradiction. So s = \sup E.
  16. Let \displaystyle r_n = 10^{-n}\lfloor 10^n \sqrt{2} \rfloor, the truncation of \displaystyle \sqrt{2} to n decimal places (the integer part exists by the The Archimedean property and density of the rationals), and I_n = [r_n, r_n + 10^{-n}]. Then \displaystyle r_n \le \sqrt{2} < r_n + 10^{-n} and r_n \le r_{n+1}; also r_{n+1} < r_n + 10^{-n} with both multiples of 10^{-(n+1)}, so r_{n+1} + 10^{-(n+1)} \le r_n + 10^{-n}, and the intervals are nested. Their lengths tend to 0 and each contains \displaystyle \sqrt{2}, so by Theorem 3 their intersection in \mathbb{R} is \displaystyle \{\sqrt{2}\}, which is irrational (Irrational numbers). So the sets I_n \cap \mathbb{Q}, nested closed bounded intervals of \mathbb{Q}, share no point.
math.wiki · CC-BY-SA-4.0