Chapter IV · Limits of functions of a positive integral variable

The limit of xnx^n

§72 35 min

How xⁿ behaves for every real x, proved with monotone sequences, and the ratio arguments that grow out of it.

Why this matters

The geometric sequence xnx^n is the yardstick of analysis. Almost every convergence test in Chapter VIII works by comparing something with xnx^n, the geometric series of §78 is built from it, and its behaviour decides that of nkxnn^k x^n, xn/n!x^n/n! and many others. §72 settles xnx^n for every real xx, and shows the monotone theorem of §69 earning its keep: it gives the limit without any estimate of n0n_0.

The idea

Multiplying by xx again and again does one of a few things. If x>1x > 1, each step makes the term larger by a fixed factor, and the terms run off to infinity. If 0<x<10 < x < 1, each step shrinks the term by a fixed factor, and the terms die away to 00. A negative xx does the same in size but flips the sign every time. The borderline cases x=1x = 1 and x=−1x = -1 are exactly where “larger” or “smaller by a fixed factor” fails.

Explore Five regimes of xnx^n

Try this. Drag xx slowly from 1.61.6 down to −1.6-1.6. Stop at 1.051.05, 11, 0.950.95, −0.95-0.95, −1-1 and −1.05-1.05. Watch which part of the strip under the plot lights up.

ϕ(n)=xn,x=0.9\phi(n) = x^n,\quad x = 0.9
nxⁿ
  • xⁿ (points beyond ±4.5 are off the plot)
osc. infinitely→ 0→ +∞−11
xⁿ → 0. Here ∣x∣n<0.01|x|^n < 0.01 for every n≥44n \ge 44; the closer ∣x∣|x| is to 1, the later this happens, but it always does.
What to notice

The behaviour changes only at x=1x = 1 and x=−1x = -1. Just inside, ∣x∣=0.95|x| = 0.95, the terms still reach 00, only later (the read-out says when ∣x∣n<0.01|x|^n < 0.01); just outside, ∣x∣=1.05|x| = 1.05, they still escape, only slowly. There is no gradual transition in the limit, only in the speed.

The surprise in Hardy’s proof is that we never have to compute when xnx^n becomes small or large. For x>1x > 1 the sequence increases, so (§69) it tends to a limit or to +∞+\infty. A limit ll would have to satisfy l=xll = xl, because xn+1x^{n+1} is both x⋅xnx \cdot x^n and a shifted copy of xnx^n. With x>1x > 1 and l≥1l \ge 1 that is impossible, so xn→+∞x^n \to +\infty.

Precise statements

Theorem The limit of xnx^n Hardy: §72

As n→∞n \to \infty,

xn→+∞ (x>1);xn→1 (x=1);xn→0 (−1<x<1);x^n \to +\infty \ (x > 1); \qquad x^n \to 1 \ (x = 1); \qquad x^n \to 0 \ (-1 < x < 1);

xnx^n oscillates finitely when x=−1x = -1, and oscillates infinitely when x<−1x < -1.

Two small facts are used in the proof.

Lemma Shifting does not change a limit Hardy: Examples XXV, 7

If ϕ(n)→l\phi(n) \to l, then ϕ(n+1)→l\phi(n+1) \to l (and more generally ϕ(n+p)→l\phi(n + p) \to l for any fixed integer pp).

Lemma Bernoulli’s inequality Hardy: §72, Example

If δ>0\delta > 0 and nn is a positive integer, then (1+δ)n≥1+nδ(1 + \delta)^n \ge 1 + n\delta.

Proofs

Proof Shifting does not change a limit
Strategy

The same n0n_0 works, because n≥n0n \ge n_0 implies n+1≥n0n + 1 \ge n_0.

Use the threshold for ϕ\phi

Given δ>0\delta > 0, there is n0n_0 with ∣ϕ(m)−l∣<δ|\phi(m) - l| < \delta for all m≥n0m \ge n_0. If n≥n0n \ge n_0, then m=n+1≥n0m = n + 1 \ge n_0, so ∣ϕ(n+1)−l∣<δ|\phi(n+1) - l| < \delta. (For ϕ(n+p)\phi(n + p) with pp negative, use n0−pn_0 - p instead.)

Proof The limit of xnx^n (Hardy's proof)
Strategy

For x>0x > 0 the sequence is monotone, so it has a limit or tends to +∞+\infty; any limit ll must satisfy l=xll = xl. Negative xx reduces to positive xx through ∣xn∣=∣x∣n|x^n| = |x|^n.

x=0x = 0 and x=1x = 1

The sequences are constant (00 and 11), so they tend to 00 and 11.

x>1x > 1

xn+1=x⋅xn>xnx^{n+1} = x \cdot x^n > x^n, so the sequence increases. By §69 it tends to a limit ll or to +∞+\infty. If xn→lx^n \to l, then xn+1→lx^{n+1} \to l (shifting), but also xn+1=x⋅xn→xlx^{n+1} = x \cdot x^n \to xl (Theorem II). By uniqueness of limits, l=xll = xl, so (x−1)l=0(x - 1)l = 0 and l=0l = 0. But every term is greater than 11, and by Cor. 3 of §69 the limit is at least each term: l≥x>1l \ge x > 1. This contradiction leaves xn→+∞x^n \to +\infty.

0<x<10 < x < 1

Now the sequence decreases and is positive, so it is bounded below by 00 and tends to a limit l≥0l \ge 0. As before l=xll = xl, and since x≠1x \ne 1, l=0l = 0.

−1<x<0-1 < x < 0

∣xn−0∣=∣x∣n|x^n - 0| = |x|^n and 0<∣x∣<10 < |x| < 1, so ∣xn∣<δ|x^n| < \delta whenever ∣x∣n<δ|x|^n < \delta, which holds for large nn by the previous step. Hence xn→0x^n \to 0.

x=−1x = -1 and x<−1x < -1

(−1)n(-1)^n oscillates finitely (Oscillating sequences). If x<−1x < -1, then ∣xn∣=∣x∣n→+∞|x^n| = |x|^n \to +\infty, so xnx^n is unbounded; its terms alternate in sign, so it is positive for infinitely many nn and negative for infinitely many nn, and so it tends neither to +∞+\infty nor to −∞-\infty. It oscillates infinitely.

Hardy’s alternative for x>1x > 1 is more direct and gives an explicit n0n_0.

Proof Bernoulli’s inequality, and xn→+∞x^n \to +\infty again
Strategy

Expand (1+δ)n(1 + \delta)^n by the binomial theorem and throw away all terms after the second; they are positive.

The inequality

(1+δ)n=1+nδ+(n2)δ2+⋯+δn≥1+nδ(1 + \delta)^n = 1 + n\delta + \binom{n}{2}\delta^2 + \dots + \delta^n \ge 1 + n\delta, since every omitted term is positive (and for n=1n = 1 there is equality).

Apply it

Write x=1+δx = 1 + \delta with δ=x−1>0\delta = x - 1 > 0. Given Δ\Delta, if n>Δ/δn > \Delta/\delta then xn≥1+nδ>Δx^n \ge 1 + n\delta > \Delta. So n0=[Δ/δ]+1n_0 = [\Delta/\delta] + 1 works, and xn→+∞x^n \to +\infty. (For 0<x<10 < x < 1, apply this to 1/x>11/x > 1: (1/x)n→+∞(1/x)^n \to +\infty, so xn→0x^n \to 0.)

The explorer’s “Bernoulli” toggle draws the line 1+n(x−1)1 + n(x - 1) under the points. The bound is very generous (it grows linearly, the true terms exponentially), but it is enough.

Ratios: the idea that grows out of xnx^n

If each term is at least KK times the previous one, with K>1K > 1, the sequence grows at least as fast as KnK^n. If each term is at most KK times the previous one in size, with K<1K < 1, it shrinks at least as fast as KnK^n. This is the germ of the ratio test for series. Try it on nxnn x^n with x=0.9x = 0.9.

Explore nxnn x^n: a rise before the fall

Try this. With x=0.9x = 0.9, find an n0n_0 for δ=0.5\delta = 0.5 (show 250250 terms). Then move xx to 0.980.98.

ϕ(n)=nxn\phi(n) = n x^n , candidate limit l=0l = 0
nφ(n)
  • band l ± δ
  • l
  • n₀ (drag)
  • n ≥ n₀, good
  • n ≥ n₀, bad
42 terms with n ≥ n₀ fall outside the band: n = 1, 2, 3, 4, 5, ….
What to notice

The terms first grow (the factor nn wins early), then fall (the factor xnx^n wins). The ratio of consecutive terms is n+1nx\tfrac{n+1}{n}x, which tends to x<1x < 1; once it is below some K<1K < 1 the terms shrink geometrically. The closer xx is to 11, the later and higher the peak, but the limit is still 00.

Worked examples and exercises

Exercise Ratio at least K>1K > 1 ●○○ Hardy: Examples XXVII, 1–2

Suppose ϕ(n)>0\phi(n) > 0 and ϕ(n+1)≥Kϕ(n)\phi(n+1) \ge K\phi(n) for all n≥n0n \ge n_0, where K>1K > 1. Show that ϕ(n)→+∞\phi(n) \to +\infty.

Solution

By induction, ϕ(n)≥Kn−n0ϕ(n0)\phi(n) \ge K^{n - n_0}\phi(n_0) for n≥n0n \ge n_0. Here ϕ(n0)>0\phi(n_0) > 0 is a fixed number and Kn−n0=Kn/Kn0→+∞K^{n - n_0} = K^n / K^{n_0} \to +\infty by the theorem. So given Δ\Delta, Kn−n0>Δ/ϕ(n0)K^{n - n_0} > \Delta/\phi(n_0) for large nn, and then ϕ(n)>Δ\phi(n) > \Delta.

Exercise Ratio tending to ll with ∣l∣<1|l| < 1 ●●○ Hardy: Examples XXVII, 4, 6

Show that if ϕ(n+1)/ϕ(n)→l\phi(n+1)/\phi(n) \to l with −1<l<1-1 < l < 1, then ϕ(n)→0\phi(n) \to 0.

Hint
Pick KK with ∣l∣<K<1|l| < K < 1, say K=12(1+∣l∣)K = \tfrac12(1 + |l|). For large nn, ∣ϕ(n+1)∣<K∣ϕ(n)∣|\phi(n+1)| < K|\phi(n)|.

Solution

Let K=12(1+∣l∣)K = \tfrac12(1 + |l|), so ∣l∣<K<1|l| < K < 1. Applying the definition with δ=K−∣l∣>0\delta = K - |l| > 0, there is n0n_0 with ∣ϕ(n+1)/ϕ(n)∣<∣l∣+δ=K|\phi(n+1)/\phi(n)| < |l| + \delta = K for n≥n0n \ge n_0, that is, ∣ϕ(n+1)∣<K∣ϕ(n)∣|\phi(n+1)| < K|\phi(n)|. By induction ∣ϕ(n)∣≤Kn−n0∣ϕ(n0)∣|\phi(n)| \le K^{n - n_0}|\phi(n_0)| for n≥n0n \ge n_0. Since Kn→0K^{n} \to 0, the right side tends to 00; so ∣ϕ(n)∣<δ′|\phi(n)| < \delta' for large nn, for any δ′>0\delta' > 0. Hence ϕ(n)→0\phi(n) \to 0.

Exercise nrxnn^r x^n ●●○ Hardy: Examples XXVII, 7

Let rr be a positive integer. Find the behaviour of nrxnn^r x^n for every real xx.

Hint
For x≠0x \ne 0 the ratio of consecutive terms is (n+1n)rx→x\left(\frac{n+1}{n}\right)^r x \to x.

Solution

If x=0x = 0 every term is 00. Otherwise the ratio is (1+1/n)rx→x(1 + 1/n)^r x \to x (Theorem II, rr times). For 0<x<10 < x < 1 or −1<x<0-1 < x < 0, the previous exercise gives nrxn→0n^r x^n \to 0. For x>1x > 1, the ratio is eventually above K=12(1+x)>1K = \tfrac12(1 + x) > 1, so the terms tend to +∞+\infty (Examples XXVII, 1). For x=1x = 1 the sequence is nr→+∞n^r \to +\infty. For x≤−1x \le -1, ∣nrxn∣=nr∣x∣n≥nr→+∞|n^r x^n| = n^r|x|^n \ge n^r \to +\infty and the signs alternate, so it oscillates infinitely. In short: the factor xnx^n decides everything, and nrn^r matters only when ∣x∣=1|x| = 1.

Exercise xn→1\sqrt[n]{x} \to 1 ●●○ Hardy: Examples XXVII, 10

Show that x1/n→1x^{1/n} \to 1 for every x>0x > 0.

Hint
For x>1x > 1, the sequence x1/nx^{1/n} decreases and stays above 11. If its limit ll were greater than 11, what would x≥lnx \ge l^n say?

Solution

If x=1x = 1 there is nothing to prove. Let x>1x > 1. Then x1/n>1x^{1/n} > 1, and x1/(n+1)<x1/nx^{1/(n+1)} < x^{1/n} (raise both sides to the power n(n+1)n(n+1): xn<xn+1x^n < x^{n+1}). So x1/nx^{1/n} decreases and is bounded below by 11; it tends to a limit l≥1l \ge 1, and by the decreasing form of Cor. 3, x1/n≥lx^{1/n} \ge l for every nn, that is, x≥lnx \ge l^n for every nn. If l>1l > 1 then ln→+∞l^n \to +\infty, which contradicts x≥lnx \ge l^n. So l=1l = 1. If 0<x<10 < x < 1, then x1/n=1/(1/x)1/n→1/1=1x^{1/n} = 1/(1/x)^{1/n} \to 1/1 = 1 by Theorem III.

Exercise nn→1\sqrt[n]{n} \to 1 ●●○ Hardy: Examples XXVII, 11

Show that n1/n→1n^{1/n} \to 1.

Hint
Write n1/n=1+hnn^{1/n} = 1 + h_n with hn≥0h_n \ge 0, and keep only the third term of the binomial expansion of (1+hn)n(1 + h_n)^n.

Solution

For n≥2n \ge 2, n1/n>1n^{1/n} > 1, so hn>0h_n > 0, and n=(1+hn)n≥(n2)hn2=12n(n−1)hn2n = (1 + h_n)^n \ge \binom{n}{2}h_n^2 = \tfrac12 n(n-1)h_n^2. Hence hn2≤2/(n−1)h_n^2 \le 2/(n-1) and 0<hn≤2/(n−1)0 < h_n \le \sqrt{2/(n-1)}. Given δ>0\delta > 0, this is less than δ\delta as soon as n−1>2/δ2n - 1 > 2/\delta^2. So hn→0h_n \to 0 and n1/n→1n^{1/n} \to 1. (Hardy’s proof is different: he shows that n1/nn^{1/n} decreases from n=3n = 3 on, using (1+1/n)n<3(1 + 1/n)^n < 3 from the next lesson, and then argues as in the previous exercise.)

Exercise xn/n!→0x^n/n! \to 0 ●○○ Hardy: Examples XXVII, 12–13

Show that xn/n!→0x^n/n! \to 0 for every real xx, and deduce that n!n→+∞\sqrt[n]{n!} \to +\infty.

Solution

With un=xn/n!u_n = x^n/n! (and x≠0x \ne 0), un+1/un=x/(n+1)→0u_{n+1}/u_n = x/(n+1) \to 0, so by the ratio exercise un→0u_n \to 0. For the second claim, let Δ>0\Delta > 0. Taking x=Δx = \Delta, we get Δn/n!<1\Delta^n/n! < 1 for large nn, that is, n!>Δnn! > \Delta^n, that is, n!n>Δ\sqrt[n]{n!} > \Delta. So n!n→+∞\sqrt[n]{n!} \to +\infty.

Check yourself

Check yourself Choose one

How does (−1)n(-1)^n, that is xnx^n with x=−1x = -1, behave?

Why

It takes the values −1,1,−1,1,…-1, 1, -1, 1, \dots: bounded, with no limit. Only for x<−1x < -1 do the sizes grow, giving infinite oscillation; only for ∣x∣<1|x| < 1 do the terms tend to 00.

Check yourself True or false?

(0.9999)n(0.9999)^n does not tend to 00, because it is still greater than 0.90.9 after a thousand terms.

Why

False. (0.9999)1000≈0.905(0.9999)^{1000} \approx 0.905, but (0.9999)n<0.01(0.9999)^n < 0.01 once n>log⁡0.01/log⁡0.9999≈46 050n > \log 0.01/\log 0.9999 \approx 46\,050. A slow start says nothing about the limit; the theorem covers every xx with ∣x∣<1|x| < 1.

Check yourself Enter a number

What is the smallest nn with (0.9)n<0.01(0.9)^n < 0.01?

Why

(0.9)n<0.01(0.9)^n < 0.01 means n>log⁡0.01/log⁡0.9=43.7…n > \log 0.01 / \log 0.9 = 43.7\ldots, so n=44n = 44. Since (0.9)n(0.9)^n decreases, every later nn also works.

Check yourself Choose one

In Hardy’s proof for x>1x > 1, the equation l=xll = xl comes from:

Why

xn+1→lx^{n+1} \to l because shifting does not change a limit, and xn+1=x⋅xn→xlx^{n+1} = x \cdot x^n \to xl by Theorem II. A sequence has only one limit, so l=xll = xl. Monotonicity is used separately, to know that a limit or +∞+\infty is the only possibility.

Hardy’s language

Summary

Summary
  • xn→+∞x^n \to +\infty for x>1x > 1, →1\to 1 for x=1x = 1, →0\to 0 for ∣x∣<1|x| < 1; it oscillates finitely for x=−1x = -1 and infinitely for x<−1x < -1.
  • Hardy’s proof: monotone, so a limit or +∞+\infty; a limit would satisfy l=xll = xl. Bernoulli’s (1+δ)n≥1+nδ(1 + \delta)^n \ge 1 + n\delta gives an explicit alternative.
  • Shifting a sequence does not change its limit; this, with uniqueness, is how "l=xll = xl" arises.
  • Ratio principle: if ∣ϕ(n+1)/ϕ(n)∣→l<1|\phi(n+1)/\phi(n)| \to l < 1 then ϕ(n)→0\phi(n) \to 0; if the ratio tends to l>1l > 1 (positive terms) then ϕ(n)→+∞\phi(n) \to +\infty. So nrxn→0n^r x^n \to 0 and xn/n!→0x^n/n! \to 0 for ∣x∣<1|x| < 1 and all xx respectively.