Chapter VIII · The convergence of infinite series and infinite integrals

Chapter VIII problems

Miscellaneous examples 120 min

Ten of Hardy’s Miscellaneous Examples on Chapter VIII, from comparison tests to power series and infinite integrals, with graded hints and full solutions.

How to use this page

These problems come from the Miscellaneous Examples at the end of Hardy’s Chapter VIII, chosen to cover the whole chapter: series of positive terms, telescoping sums, conditional convergence, power series, multiplication of series, recurrences and infinite integrals. Each is restated briefly; Hardy’s wording is in the book. Try each one before opening the first hint, and open the hints one at a time: the first only points in a direction, the last is nearly a solution. Dots show difficulty (one: routine; three: needs a real idea).

Series of positive terms

Exercise A second difference ●●○ Hardy: Misc. Examples VIII, 1

For which real kk does ∑n≥1nk{n+1−2n+n−1}\displaystyle\sum_{n \ge 1} n^k\big\{\sqrt{n+1} - 2\sqrt n + \sqrt{n-1}\big\} converge?

Hint
The bracket is a second difference of x\sqrt x, so expect it to be of order n−3/2n^{-3/2}. Find its exact size by rationalising twice.
Hint
n+1−n=1n+1+n\sqrt{n+1} - \sqrt n = \dfrac{1}{\sqrt{n+1} + \sqrt n}, and similarly for n−n−1\sqrt n - \sqrt{n-1}.

Solution

Write dn=n+1−2n+n−1=(n+1−n)−(n−n−1)d_n = \sqrt{n+1} - 2\sqrt n + \sqrt{n-1} = (\sqrt{n+1} - \sqrt n) - (\sqrt n - \sqrt{n-1}). Rationalising each bracket,

dn=1n+1+n−1n+n−1=n−1−n+1(n+1+n)(n+n−1)=−2(n+1+n−1)(n+1+n)(n+n−1).d_n = \frac{1}{\sqrt{n+1} + \sqrt n} - \frac{1}{\sqrt n + \sqrt{n-1}} = \frac{\sqrt{n-1} - \sqrt{n+1}}{(\sqrt{n+1} + \sqrt n)(\sqrt n + \sqrt{n-1})} = \frac{-2}{(\sqrt{n+1} + \sqrt{n-1})(\sqrt{n+1} + \sqrt n)(\sqrt n + \sqrt{n-1})} .

Each of the three factors in the denominator is ∼2n\sim 2\sqrt n, so dn<0d_n < 0 and dn∼−14n−3/2d_n \sim -\tfrac14 n^{-3/2}. The terms nkdnn^k d_n are all negative, and nkdn/(−14nk−3/2)→1n^k d_n \big/ \big(-\tfrac14 n^{k - 3/2}\big) \to 1, so by comparison (in the limit form of §173) the series converges if and only if ∑nk−3/2\sum n^{k-3/2} does, that is, if and only if k−32<−1k - \tfrac32 < -1:

convergent for k<12,divergent (to −∞) for k≥12.\text{convergent for } k < \tfrac12, \qquad \text{divergent (to } -\infty \text{) for } k \ge \tfrac12 .
Exercise Partial fractions and telescoping ●●○ Hardy: Misc. Examples VIII, 3

Show that ∑n=1∞n2+9n+5(n+1)(2n+3)(2n+5)(n+4)=536\displaystyle\sum_{n=1}^\infty \frac{n^2 + 9n + 5}{(n+1)(2n+3)(2n+5)(n+4)} = \frac{5}{36}.

Hint
Resolve the general term into partial fractions An+1+B2n+3+C2n+5+Dn+4\dfrac{A}{n+1} + \dfrac{B}{2n+3} + \dfrac{C}{2n+5} + \dfrac{D}{n+4}, using the “cover-up” rule.
Hint
You should find A=−DA = -D and B=−CB = -C. Group the terms in pairs that telescope.

Solution

Evaluating the numerator over the other factors at each root of the denominator: at n=−1n = -1, A=−31⋅3⋅3=−13A = \dfrac{-3}{1\cdot 3 \cdot 3} = -\tfrac13; at n=−32n = -\tfrac32, B=−25/4(−12)(2)(52)=52B = \dfrac{-25/4}{(-\frac12)(2)(\frac52)} = \tfrac52; at n=−52n = -\tfrac52, C=−45/4(−32)(−2)(32)=−52C = \dfrac{-45/4}{(-\frac32)(-2)(\frac32)} = -\tfrac52; at n=−4n = -4, D=−15(−3)(−5)(−3)=13D = \dfrac{-15}{(-3)(-5)(-3)} = \tfrac13. So

un=13(1n+4−1n+1)+52(12n+3−12n+5).u_n = \frac13\Big(\frac{1}{n+4} - \frac{1}{n+1}\Big) + \frac52\Big(\frac{1}{2n+3} - \frac{1}{2n+5}\Big).

Both brackets telescope. ∑n=1N(12n+3−12n+5)=15−12N+5→15\sum_{n=1}^{N}\big(\frac{1}{2n+3} - \frac{1}{2n+5}\big) = \frac15 - \frac{1}{2N+5} \to \frac15, and ∑n=1N(1n+4−1n+1)=−(12+13+14)+(1N+2+1N+3+1N+4)→−1312\sum_{n=1}^{N}\big(\frac{1}{n+4} - \frac{1}{n+1}\big) = -\big(\frac12 + \frac13 + \frac14\big) + \big(\frac{1}{N+2} + \frac{1}{N+3} + \frac{1}{N+4}\big) \to -\frac{13}{12}. Hence the sum is 52⋅15−13⋅1312=12−1336=536\tfrac52\cdot\tfrac15 - \tfrac13\cdot\tfrac{13}{12} = \tfrac12 - \tfrac{13}{36} = \tfrac{5}{36}.

Exercise Near the edge of convergence ●●○ Hardy: Misc. Examples VIII, 11

Prove that lim⁡α→0+α∑n=1∞n−1−α=1\displaystyle\lim_{\alpha \to 0+} \alpha\sum_{n=1}^\infty n^{-1-\alpha} = 1.

Hint
Use the inequalities behind the integral test (§180) for ϕ(x)=x−1−α\phi(x) = x^{-1-\alpha}.

Solution

Since x−1−αx^{-1-\alpha} decreases, ∫nn+1x−1−αdx≤n−1−α≤∫n−1nx−1−αdx\int_n^{n+1} x^{-1-\alpha}dx \le n^{-1-\alpha} \le \int_{n-1}^{n} x^{-1-\alpha}dx for n≥2n \ge 2. Summing (and keeping the term n=1n = 1 separately on the right),

1α=∫1∞x−1−α dx≤∑n=1∞n−1−α≤1+∫1∞x−1−α dx=1+1α.\frac1\alpha = \int_1^\infty x^{-1-\alpha}\,dx \le \sum_{n=1}^\infty n^{-1-\alpha} \le 1 + \int_1^\infty x^{-1-\alpha}\,dx = 1 + \frac1\alpha .

Multiplying by α>0\alpha > 0: 1≤α∑n−1−α≤1+α1 \le \alpha\sum n^{-1-\alpha} \le 1 + \alpha, and the right-hand side tends to 11. So ∑n−1−α\sum n^{-1-\alpha} is about 1/α1/\alpha: it blows up exactly as fast as the integral does when α→0\alpha \to 0.

Exercise A telescoping series in xx ●●○ Hardy: Misc. Examples VIII, 12

Find the sum of ∑n≥1un\sum_{n \ge 1} u_n, where

un=xn−x−n−1(xn+x−n)(xn+1+x−n−1),u_n = \frac{x^n - x^{-n-1}}{(x^n + x^{-n})(x^{n+1} + x^{-n-1})},

for every real x≠0x \neq 0 for which the series converges.

Hint
Show that un=1x−1(an−an+1)u_n = \dfrac{1}{x-1}\big(a_n - a_{n+1}\big) with an=1xn+x−na_n = \dfrac{1}{x^n + x^{-n}} (when x≠1x \ne 1).
Hint
What happens to ana_n when ∣x∣≠1|x| \ne 1? When x=−1x = -1?

Solution

For x≠1x \ne 1,

an−an+1=xn+1+x−n−1−xn−x−n(xn+x−n)(xn+1+x−n−1)=(x−1)(xn−x−n−1)(xn+x−n)(xn+1+x−n−1),a_n - a_{n+1} = \frac{x^{n+1} + x^{-n-1} - x^n - x^{-n}}{(x^n + x^{-n})(x^{n+1} + x^{-n-1})} = \frac{(x - 1)(x^n - x^{-n-1})}{(x^n + x^{-n})(x^{n+1} + x^{-n-1})} ,

so un=(an−an+1)/(x−1)u_n = (a_n - a_{n+1})/(x - 1) and ∑n=1Nun=(a1−aN+1)/(x−1)\sum_{n=1}^{N} u_n = (a_1 - a_{N+1})/(x - 1). (For real x≠0x \ne 0 the denominators never vanish: xnx^n and x−nx^{-n} have the same sign.)

  • If ∣x∣≠1|x| \ne 1, then ∣xN+x−N∣→∞|x^{N} + x^{-N}| \to \infty, so aN+1→0a_{N+1} \to 0 and the sum is a1x−1=x(x−1)(x2+1)\dfrac{a_1}{x - 1} = \dfrac{x}{(x - 1)(x^2 + 1)}.
  • If x=1x = 1, every un=0u_n = 0 and the sum is 00.
  • If x=−1x = -1, an=12(−1)na_n = \tfrac12(-1)^n, and the partial sums (a1−aN+1)/(−2)=14(1+(−1)N+1)(a_1 - a_{N+1})/(-2) = \tfrac14\big(1 + (-1)^{N+1}\big) alternate between 12\tfrac12 and 00: the series oscillates finitely.

Signs, cancellation and rearrangement

Exercise Five series in a/na/n ●●○ Hardy: Misc. Examples VIII, 6

Let aa be real. Discuss the convergence of

∑sin⁡an,∑1nsin⁡an,∑(−1)nsin⁡an,∑(1−cos⁡an),∑(−1)nn(1−cos⁡an).\sum \sin\frac an, \qquad \sum\frac1n\sin\frac an, \qquad \sum(-1)^n\sin\frac an, \qquad \sum\Big(1 - \cos\frac an\Big), \qquad \sum (-1)^n n\Big(1 - \cos\frac an\Big).
Hint
For large nn, sin⁡(a/n)≈a/n\sin(a/n) \approx a/n and 1−cos⁡(a/n)=2sin⁡2(a/2n)≈a2/(2n2)1 - \cos(a/n) = 2\sin^2(a/2n) \approx a^2/(2n^2).
Hint
For the two alternating series you need the terms to decrease steadily from some point on. Look at t↦(sin⁡t)t \mapsto (\sin t) and t↦(1−cos⁡t)/tt \mapsto (1 - \cos t)/t for small t>0t > 0.
Solution

If a=0a = 0 every term is 00; take a≠0a \ne 0, and by symmetry (sin⁡\sin is odd, cos⁡\cos even) a>0a > 0.

  1. sin⁡(a/n)>0\sin(a/n) > 0 for n>a/πn > a/\pi and sin⁡(a/n)/(a/n)→1\sin(a/n)\big/(a/n) \to 1, so by comparison with ∑a/n\sum a/n the series diverges (to +∞+\infty; to −∞-\infty if a<0a < 0).
  2. ∣1nsin⁡an∣≤∣a∣/n2\big|\tfrac1n\sin\tfrac an\big| \le |a|/n^2: absolutely convergent.
  3. For n>2a/πn > 2a/\pi, a/n∈(0,π/2)a/n \in (0, \pi/2) where sin⁡\sin increases, so sin⁡(a/n)\sin(a/n) decreases steadily to 00: convergent by the alternating series test, and only conditionally, by case 1.
  4. 0≤1−cos⁡an=2sin⁡2a2n≤a22n20 \le 1 - \cos\tfrac an = 2\sin^2\tfrac{a}{2n} \le \tfrac{a^2}{2n^2}: convergent (absolutely).
  5. Here φn=n(1−cos⁡an)=a h(a/n)\varphi_n = n(1 - \cos\tfrac an) = a\,h(a/n) with h(t)=(1−cos⁡t)/th(t) = (1 - \cos t)/t. Since h′(t)=(tsin⁡t−1+cos⁡t)/t2h'(t) = (t\sin t - 1 + \cos t)/t^2 and tsin⁡t−(1−cos⁡t)=t2−12t2+O(t4)>0t\sin t - (1 - \cos t) = t^2 - \tfrac12 t^2 + O(t^4) > 0 for small t>0t > 0, hh increases near 00; so φn\varphi_n decreases steadily for large nn, and φn∼a2/(2n)→0\varphi_n \sim a^2/(2n) \to 0. The series is convergent, but only conditionally, since ∑φn\sum \varphi_n diverges like ∑1/n\sum 1/n.
Exercise Signs in growing blocks ●●● Hardy: Misc. Examples VIII, 8

Show that 1−12−13+14+15+16−17−18−19−110+⋯1 - \tfrac12 - \tfrac13 + \tfrac14 + \tfrac15 + \tfrac16 - \tfrac17 - \tfrac18 - \tfrac19 - \tfrac1{10} + \cdots, in which successive blocks of terms of the same sign have 1,2,3,4,…1, 2, 3, 4, \dots terms, converges; but that if the blocks have 1,2,4,8,…1, 2, 4, 8, \dots terms the series oscillates finitely.

Hint
First reduce the problem to the series of block sums G1−G2+G3−⋯G_1 - G_2 + G_3 - \cdots: within a block, the partial sums move monotonically from one block-end value to the next.
Hint
Estimate GkG_k by integrals of 1/x1/x. For blocks 1,2,3,…1, 2, 3, \dots show Gk=2k+O(1/k2)G_k = \frac2k + O(1/k^2). For blocks 1,2,4,…1, 2, 4, \dots show that GkG_k decreases to log⁡2\log 2.

Solution

Reduction. Let GkG_k be the sum of the moduli of the kk-th block and Sk=G1−G2+⋯±GkS_k = G_1 - G_2 + \cdots \pm G_k the partial sum at the end of block kk. Inside block k+1k+1 all terms have one sign, so every partial sum in it lies between SkS_k and Sk+1S_{k+1}. Hence the series converges iff SkS_k converges, and it is bounded iff SkS_k is bounded.

Blocks 1,2,3,…1, 2, 3, \dots. Block kk consists of 1/j1/j for m<j≤Mm < j \le M, where m=12k(k−1)m = \tfrac12 k(k-1) and M=12k(k+1)M = \tfrac12 k(k+1). Comparing with ∫dx/x\int dx/x,

log⁡M+1m+1≤Gk≤log⁡Mm(k≥2).\log\frac{M+1}{m+1} \le G_k \le \log\frac Mm \quad (k \ge 2).

The two bounds differ by log⁡(1+1m)−log⁡(1+1M)≤1m=O(1/k2)\log(1 + \frac1m) - \log(1 + \frac1M) \le \frac1m = O(1/k^2), and log⁡Mm=log⁡k+1k−1=2k+O(1/k3)\log\frac Mm = \log\frac{k+1}{k-1} = \frac2k + O(1/k^3). So Gk=2k+ekG_k = \frac{2}{k} + e_k with ∣ek∣≤C/k2|e_k| \le C/k^2. Then Sk=∑i≤k(−1)i−12i+∑i≤k(−1)i−1eiS_k = \sum_{i \le k} (-1)^{i-1}\frac2i + \sum_{i \le k}(-1)^{i-1}e_i: the first converges by the alternating series test, the second absolutely. So SkS_k converges, and so does the series.

Blocks 1,2,4,…1, 2, 4, \dots. Block kk consists of 1/j1/j for 2k−1≤j<2k2^{k-1} \le j < 2^k, so Gk=f(2k−1)G_k = f(2^{k-1}) with f(N)=∑j=N2N−11/jf(N) = \sum_{j=N}^{2N-1} 1/j. Now f(N)−f(N+1)=1N−12N−12N+1=12N−12N+1>0f(N) - f(N+1) = \frac1N - \frac1{2N} - \frac1{2N+1} = \frac{1}{2N} - \frac{1}{2N+1} > 0, so GkG_k decreases. Comparing with ∫dx/x\int dx/x again, log⁡2=∫N2Ndxx≤f(N)≤∫N−12N−1dxx=log⁡2N−1N−1\log 2 = \int_N^{2N}\frac{dx}{x} \le f(N) \le \int_{N-1}^{2N-1}\frac{dx}{x} = \log\frac{2N-1}{N-1} for N≥2N \ge 2, so Gk→log⁡2>0G_k \to \log 2 > 0. By Examples LXXVIII, 5, ∑(−1)k−1Gk\sum (-1)^{k-1}G_k oscillates finitely: its partial sums are bounded but the even and odd ones have limits differing by log⁡2\log 2. By the reduction, the original series oscillates finitely too.

Power series and multiplication

Exercise The central binomial coefficients ●●● Hardy: Misc. Examples VIII, 23

Discuss the convergence of ∑n≥0(2n)!(n!)2zn\displaystyle\sum_{n \ge 0} \frac{(2n)!}{(n!)^2} z^n for complex zz.

Hint
Find the radius by the ratio test. On the circle put z=14wz = \tfrac14 w with ∣w∣=1|w| = 1.
Hint
Let φn=(2nn)4−n=∏k=1n2k−12k\varphi_n = \binom{2n}{n}4^{-n} = \prod_{k=1}^{n}\frac{2k-1}{2k}. Show 12n≤φn≤12n+1\frac{1}{2\sqrt n} \le \varphi_n \le \frac{1}{\sqrt{2n+1}} by comparing each factor with 2k2k+1\frac{2k}{2k+1} and with 2k−22k−1\frac{2k-2}{2k-1}.

Solution

With cn=(2n)!/(n!)2c_n = (2n)!/(n!)^2, cn+1cn=(2n+2)(2n+1)(n+1)2=2(2n+1)n+1→4\dfrac{c_{n+1}}{c_n} = \dfrac{(2n+2)(2n+1)}{(n+1)^2} = \dfrac{2(2n+1)}{n+1} \to 4, so the radius of convergence is 14\tfrac14 (§200): absolute convergence for ∣z∣<14|z| < \tfrac14, no convergence for ∣z∣>14|z| > \tfrac14.

On the circle write z=14wz = \tfrac14 w, ∣w∣=1|w| = 1, so the series is ∑φnwn\sum \varphi_n w^n with φn=cn4−n=∏k=1n2k−12k\varphi_n = c_n 4^{-n} = \prod_{k=1}^n \frac{2k-1}{2k}. Each factor is less than 11, so φn\varphi_n decreases. Since 2k−12k<2k2k+1\frac{2k-1}{2k} < \frac{2k}{2k+1}, φn2<∏2k−12k⋅2k2k+1=12n+1\varphi_n^2 < \prod\frac{2k-1}{2k}\cdot\frac{2k}{2k+1} = \frac{1}{2n+1}, so φn→0\varphi_n \to 0. Since 2k−12k≥2k−22k−1\frac{2k-1}{2k} \ge \frac{2k-2}{2k-1} for k≥2k \ge 2 and 12≥12\frac12 \ge \frac12, φn2≥12⋅12∏k=2n2k−22k=14n\varphi_n^2 \ge \frac12\cdot\frac12\prod_{k=2}^{n}\frac{2k-2}{2k} = \frac{1}{4n}, so ∑φn\sum\varphi_n diverges.

  • At w=1w = 1 (z=14z = \tfrac14): ∑φn\sum\varphi_n diverges.
  • At every other point of the circle: the partial sums of ∑wn\sum w^n are bounded by 2/∣1−w∣2/|1 - w|, and φn\varphi_n decreases steadily to 00, so the series converges by Dirichlet’s test (§196); it is not absolutely convergent, since ∑∣φnwn∣=∑φn\sum |\varphi_n w^n| = \sum\varphi_n diverges. (At z=−14z = -\tfrac14 this is the alternating series test.)

So the series converges exactly on the closed disc ∣z∣≤14|z| \le \tfrac14 with the point 14\tfrac14 removed, absolutely only inside. (Its sum is (1−4z)−1/2(1 - 4z)^{-1/2}, a binomial series.)

Exercise A product series that oscillates ●●● Hardy: Misc. Examples VIII, 26

Let an=(−1)n/n+1a_n = (-1)^n/\sqrt{n+1}, An=a0+⋯+anA_n = a_0 + \cdots + a_n, bn=a0an+a1an−1+⋯+ana0b_n = a_0a_n + a_1a_{n-1} + \cdots + a_na_0 and Bn=b0+⋯+bnB_n = b_0 + \cdots + b_n. Show that (i) ∑an\sum a_n converges to a sum AA; (ii) An=A+O(n−1/2)A_n = A + O(n^{-1/2}); (iii) bnb_n oscillates finitely; (iv) Bn=a0An+a1An−1+⋯+anA0B_n = a_0A_n + a_1A_{n-1} + \cdots + a_nA_0; (v) BnB_n oscillates finitely.

Hint
(i)–(ii): alternating series test and its error bound. (iii): this is §202, Examples LXXXI, 9, plus an upper bound for ∑r1/(r+1)(n+1−r)\sum_r 1/\sqrt{(r+1)(n+1-r)}.
Hint
(v): write Am=A+εmA_m = A + \varepsilon_m in (iv), with ∣εm∣≤1/m+2|\varepsilon_m| \le 1/\sqrt{m+2}.

Solution

(i), (ii). 1/n+11/\sqrt{n+1} decreases steadily to 00, so ∑an\sum a_n converges (§195), and ∣A−An∣≤1/n+2=O(n−1/2)|A - A_n| \le 1/\sqrt{n+2} = O(n^{-1/2}).

(iii). As in Examples LXXXI, 9, bn=(−1)nσnb_n = (-1)^n\sigma_n with σn=∑r=0n1(r+1)(n+1−r)≥2(n+1)n+2≥1\sigma_n = \sum_{r=0}^{n}\frac{1}{\sqrt{(r+1)(n+1-r)}} \ge \frac{2(n+1)}{n+2} \ge 1. For an upper bound, put N=n+2N = n + 2 and j=r+1j = r + 1, so σn=∑j=1N−11j(N−j)\sigma_n = \sum_{j=1}^{N-1}\frac{1}{\sqrt{j(N - j)}}. For j≤N/2j \le N/2, N−j≥N/2N - j \ge N/2, so these terms total at most 2/N∑j≤N/2j−1/2≤2/N⋅2N/2=2\sqrt{2/N}\sum_{j \le N/2} j^{-1/2} \le \sqrt{2/N}\cdot 2\sqrt{N/2} = 2; the terms with j>N/2j > N/2 are the same by symmetry. So 1≤∣bn∣≤41 \le |b_n| \le 4 with alternating signs: bounded, and not convergent. (In fact ∣bn∣→π|b_n| \to \pi.)

(iv). BnB_n is the sum of aiaja_ia_j over all i+j≤ni + j \le n. Grouping by ii: for each i≤ni \le n the jj run from 00 to n−in - i, giving aiAn−ia_iA_{n-i}.

(v). Write Am=A+εmA_m = A + \varepsilon_m with ∣εm∣≤1/m+2|\varepsilon_m| \le 1/\sqrt{m+2}. Then

Bn=A An+∑i=0naiεn−i,∣∑i=0naiεn−i∣≤∑i=0n1(i+1)(n−i+2)≤4B_n = A\,A_n + \sum_{i=0}^{n} a_i\varepsilon_{n-i}, \qquad \Big|\sum_{i=0}^{n} a_i\varepsilon_{n-i}\Big| \le \sum_{i=0}^{n}\frac{1}{\sqrt{(i+1)(n-i+2)}} \le 4

by the estimate of (iii) (with N=n+3N = n + 3). So BnB_n is bounded. It does not converge, because its terms bnb_n do not tend to 00. Hence BnB_n oscillates finitely: the Cauchy product of ∑an\sum a_n with itself oscillates, although An2→A2A_n^2 \to A^2.

Exercise Heads and tails ●●○ Hardy: Misc. Examples VIII, 20

A player scores 11 for each head and 22 for each tail of a fair coin, and plays until the score reaches or passes nn. Show that the probability pnp_n of hitting nn exactly is 13{2+(−12)n}\tfrac13\{2 + (-\tfrac12)^n\}.

Hint
Condition on the last toss: pn=12pn−1+12pn−2p_n = \tfrac12p_{n-1} + \tfrac12p_{n-2} for n≥2n \ge 2, with p0=1p_0 = 1, p1=12p_1 = \tfrac12.
Hint
Let P(z)=∑pnznP(z) = \sum p_n z^n (a recurring series, Misc. Examples VIII, 15). Multiply by 1−12z−12z21 - \tfrac12 z - \tfrac12 z^2.

Solution

The total nn is hit exactly iff either n−1n - 1 is hit and the next toss is a head, or n−2n - 2 is hit and the next toss is a tail. So pn=12pn−1+12pn−2p_n = \tfrac12 p_{n-1} + \tfrac12 p_{n-2} (n≥2n \ge 2), p0=1p_0 = 1, p1=12p_1 = \tfrac12.

Since 0≤pn≤10 \le p_n \le 1, P(z)=∑pnznP(z) = \sum p_n z^n converges absolutely for ∣z∣<1|z| < 1. Multiplying by 1−12z−12z21 - \tfrac12 z - \tfrac12 z^2 (a product of power series, §202), every coefficient from z2z^2 on vanishes by the recurrence, and the first two are p0=1p_0 = 1 and p1−12p0=0p_1 - \tfrac12 p_0 = 0. So

P(z)=11−12z−12z2=2(1−z)(2+z)=2/31−z+1/31+z/2=∑n≥0{23+13(−12)n}znP(z) = \frac{1}{1 - \frac12 z - \frac12 z^2} = \frac{2}{(1 - z)(2 + z)} = \frac{2/3}{1 - z} + \frac{1/3}{1 + z/2} = \sum_{n \ge 0}\Big\{\frac23 + \frac13\Big(-\frac12\Big)^n\Big\}z^n

for ∣z∣<1|z| < 1. By the uniqueness of power series (§201), pn=13{2+(−12)n}p_n = \tfrac13\{2 + (-\tfrac12)^n\}. As n→∞n \to \infty, pn→23p_n \to \tfrac23: the average step is 32\tfrac32, and a random walk with that step hits a given total about 1/321/\tfrac32 of the time.

Infinite integrals

Exercise An integral by substitution ●●○ Hardy: Misc. Examples VIII, 30

Show that for n>1n > 1

∫0∞dx{x2+1+x}n=∫0∞{x2+1−x}n dx=nn2−1.\int_0^\infty \frac{dx}{\{\sqrt{x^2+1} + x\}^n} = \int_0^\infty \{\sqrt{x^2 + 1} - x\}^n\,dx = \frac{n}{n^2 - 1} .
Hint
(x2+1+x)(x2+1−x)=1(\sqrt{x^2+1} + x)(\sqrt{x^2+1} - x) = 1. Then substitute y=x2+1+xy = \sqrt{x^2+1} + x.
Hint
From y−1/y=2xy - 1/y = 2x, find xx and dxdx in terms of yy.
Solution

The two integrands are equal because (x2+1+x)(x2+1−x)=1(\sqrt{x^2+1} + x)(\sqrt{x^2+1} - x) = 1. The integral converges: the integrand is continuous on [0,∞)[0, \infty) and x2+1+x>2x\sqrt{x^2+1} + x > 2x, so it is less than (2x)−n(2x)^{-n}, integrable at ∞\infty for n>1n > 1.

Put y=x2+1+xy = \sqrt{x^2+1} + x, which increases from 11 to ∞\infty as xx goes from 00 to ∞\infty. Then 1/y=x2+1−x1/y = \sqrt{x^2+1} - x, so x=12(y−1/y)x = \tfrac12(y - 1/y) and dx=12(1+y−2) dydx = \tfrac12(1 + y^{-2})\,dy. By the substitution rule for infinite integrals (§186),

∫0∞dx(x2+1+x)n=12∫1∞(y−n+y−n−2) dy=12(1n−1+1n+1)=nn2−1.\int_0^\infty \frac{dx}{(\sqrt{x^2+1} + x)^n} = \frac12\int_1^\infty \big(y^{-n} + y^{-n-2}\big)\,dy = \frac12\Big(\frac{1}{n-1} + \frac{1}{n+1}\Big) = \frac{n}{n^2 - 1} .

Where to go next

Chapter IX defines the logarithm, the exponential and the circular functions and expands them in power series. Almost every tool of this chapter is used there: the integral test for ∑1/(nlog⁡n)\sum 1/(n\log n), the ratio test for ∑xn/n!\sum x^n/n!, the multiplication theorem for exey=ex+ye^x e^y = e^{x+y}, and the alternating series test and Abel’s theorem for the logarithmic and arctangent series on their circles of convergence.