Chapter IX · The logarithmic, exponential and circular functions of a real variable

The binomial series on the edge

§222 35 min

The binomial series ∑(mn)xn\sum \binom{m}{n} x^n now for every real mm, and its delicate behaviour at x=±1x = \pm 1: convergent at x=1x = 1 exactly when m>−1m > -1, at x=−1x = -1 exactly when m≥0m \ge 0.

Why this matters

In §168 the binomial theorem (1+x)m=∑(mn)xn(1+x)^m = \sum \binom{m}{n}x^n was proved for −1<x<1-1 < x < 1 and rational mm, because that was the only kind of power then available. Chapter IX has since defined xmx^m for every real mm (§214), so two questions remain, and §222 answers both. Does the theorem hold for irrational mm? (Yes, with no new work.) And what happens on the boundary x=±1x = \pm 1 of the interval, where the ratio test is silent? There the answer depends on mm in an interesting way, and it is a good example of a power series behaving differently at the two ends of its interval of convergence.

Throughout, (mn)=m(m−1)⋯(m−n+1)n!\binom{m}{n} = \dfrac{m(m-1)\cdots(m-n+1)}{n!} for any real mm, and mm is not a non-negative integer (otherwise the series stops and there is nothing to discuss).

The idea

Inside the interval the terms (mn)xn\binom{m}{n}x^n shrink geometrically, whatever mm is. On the boundary the factor xnx^n no longer helps, and everything depends on how the coefficients (mn)\binom{m}{n} themselves behave. Their size is governed by the ratio

∣(mn+1)(mn)∣=∣m−n∣n+1=1−m+1n+1(n>m),\left|\frac{\binom{m}{n+1}}{\binom{m}{n}}\right| = \frac{|m - n|}{n+1} = 1 - \frac{m+1}{n+1} \qquad (n > m),

which is just below 11 when m>−1m > -1 and at least 11 when m≤−1m \le -1. In fact ∣(mn)∣|\binom{m}{n}| behaves like a constant times n−(m+1)n^{-(m+1)}.

Explore On the edge of convergence

Try this. Start at x=1x = 1, m=−0.5m = -0.5. Slide mm down through −1-1 and watch the partial sums. Then set x=−1x = -1 and slide mm through 00. Finally pick x=0.95x = 0.95 and a negative mm such as −2-2.

∑n≥0(−0.5n)(1)n\sum_{n\ge 0} \binom{−0.5}{n} (1)^n
nsₙ
  • partial sums sₙ (red when the series diverges)
  • (1+x)m(1+x)^m
  • endpoint: converges
  • endpoint: diverges
  • your m
The series converges at x = 1, because m > −1; its sum is (1+x)m=0.7071068(1+x)^m = 0.7071068. After 40 terms s39=0.6623667s_{39} = 0.6623667, and the last term has size 0.0901.
What to notice

At x=1x = 1 the partial sums zig-zag towards 2m2^m for every m>−1m > -1, more and more slowly as mm approaches −1-1; at m=−1m = -1 they flip between 11 and 00 for ever, and below −1-1 the zig-zag widens. At x=−1x = -1 the sums tend to 00 when m>0m > 0, but climb steadily to +∞+\infty when m<0m < 0. Inside the interval everything converges, though slowly near the edge.

Precise statements

Lemma Size of the binomial coefficients Hardy: cf. Examples LXXXIV, 3

Let mm be real and not a non-negative integer. Then (mn)→0\binom{m}{n} \to 0 as n→∞n \to \infty if m>−1m > -1, while ∣(mn)∣\bigl|\binom{m}{n}\bigr| never decreases from some nn on (so does not tend to 00) if m≤−1m \le -1. For n>m+1n > m + 1 the coefficients alternate in sign.

Theorem The binomial series for any real exponent Hardy: §222

Let mm be real and not a non-negative integer.

  1. If −1<x<1-1 < x < 1 then (1+x)m=∑n=0∞(mn)xn(1+x)^m = \sum_{n=0}^\infty \binom{m}{n}x^n.
  2. At x=1x = 1 the series converges if and only if m>−1m > -1, and its sum is then 2m2^m.
  3. At x=−1x = -1 the series converges if and only if m>0m > 0, and its sum is then 00. For m<0m < 0 its partial sums tend to +∞+\infty.

Including the trivial case of a non-negative integer mm, the series converges at x=−1x = -1 exactly when m≥0m \ge 0 (with sum 11 for m=0m = 0), as Hardy states it.

Proofs

Proof The lemma
Strategy

Look at the ratio of consecutive coefficients. When it is at least 11 the coefficients cannot shrink; when it is 1−(m+1)/(n+1)1 - (m+1)/(n+1) with m+1>0m + 1 > 0, the product of these ratios tends to 00 because ∑1/n\sum 1/n diverges.

The ratio

(mn+1)=(mn)⋅m−nn+1\binom{m}{n+1} = \binom{m}{n}\cdot\frac{m-n}{n+1}. For n>mn > m the factor m−nn+1\frac{m-n}{n+1} is negative (so the signs alternate from then on) and its size is n−mn+1=1−m+1n+1\frac{n-m}{n+1} = 1 - \frac{m+1}{n+1}. The coefficients are never 00, because mm is not a non-negative integer.

m≤−1m \le -1

Then 1−m+1n+1≥11 - \frac{m+1}{n+1} \ge 1, so ∣(mn)∣\bigl|\binom{m}{n}\bigr| never decreases once n>mn > m, and stays at least as large as some fixed non-zero coefficient.

m>−1m > -1

Put c=m+1>0c = m + 1 > 0 and fix n1>mn_1 > m. For n>n1n > n_1, using 1−u≤e−u1 - u \le e^{-u},

∣(mn)∣=∣(mn1)∣∏j=n1n−1(1−cj+1)≤∣(mn1)∣exp⁡(−c∑j=n1n−11j+1),\left|\binom{m}{n}\right| = \left|\binom{m}{n_1}\right|\prod_{j=n_1}^{n-1}\Bigl(1 - \frac{c}{j+1}\Bigr) \le \left|\binom{m}{n_1}\right|\exp\Bigl(-c\sum_{j=n_1}^{n-1}\frac{1}{j+1}\Bigr),

and the sum tends to ∞\infty with nn (the harmonic series diverges), so the right side tends to 00.

Proof The theorem
Strategy

Part 1 is the old proof once the derivative of (1+x)m(1+x)^m is known for irrational mm. Part 2 uses the integral form of the remainder at the single point x=1x = 1. Part 3 uses an exact formula for the partial sums.

Part 1: irrational mm

Now (1+x)m=emlog⁡(1+x)(1+x)^m = e^{m\log(1+x)} (§214), so by the chain rule its derivative is emlog⁡(1+x)⋅m1+x=m(1+x)m−1e^{m\log(1+x)}\cdot\frac{m}{1+x} = m(1+x)^{m-1}, the same rule as for rational mm. The proof of §168 used nothing else about the power, so it applies unchanged.

Part 2, m≤−1m \le -1: divergence at x=1x = 1

The terms (mn)\binom{m}{n} do not tend to 00 (lemma), so the series cannot converge.

Part 2, m>−1m > -1: the sum is 2m2^m

Taylor’s theorem with the integral remainder (§167), applied to f(t)=(1+t)mf(t) = (1+t)^m on [0,1][0, 1], gives 2m=∑k<n(mk)+Rn2^m = \sum_{k<n}\binom{m}{k} + R_n with

Rn=m(m−1)⋯(m−n+1)(n−1)!∫01(1−t)n−1(1+t)m−n dt.R_n = \frac{m(m-1)\cdots(m-n+1)}{(n-1)!}\int_0^1 (1-t)^{n-1}(1+t)^{m-n}\,dt.

When n>mn > m we have (1+t)m−n≤1(1+t)^{m-n} \le 1 on [0,1][0, 1], so the integral is at most ∫01(1−t)n−1dt=1/n\int_0^1(1-t)^{n-1}dt = 1/n. Hence ∣Rn∣≤∣(mn)∣|R_n| \le \bigl|\binom{m}{n}\bigr|, which tends to 00 by the lemma.

Part 3: a closed form for the partial sums at x=−1x = -1

By Pascal’s rule (mk)=(m−1k)+(m−1k−1)\binom{m}{k} = \binom{m-1}{k} + \binom{m-1}{k-1}, the sum ∑k=0n(−1)k(mk)\sum_{k=0}^{n}(-1)^k\binom{m}{k} telescopes:

sn=∑k=0n(−1)k(mk)=(−1)n(m−1n)=(1−m)(2−m)⋯(n−m)n!=∏j=1n(1−mj).s_n = \sum_{k=0}^{n}(-1)^k\binom{m}{k} = (-1)^n\binom{m-1}{n} = \frac{(1-m)(2-m)\cdots(n-m)}{n!} = \prod_{j=1}^{n}\Bigl(1 - \frac{m}{j}\Bigr).

Part 3: conclude

If m>0m > 0 then m−1>−1m - 1 > -1, so sn=±(m−1n)→0s_n = \pm\binom{m-1}{n} \to 0 by the lemma, and 0=(1−1)m0 = (1 - 1)^m as it should be. If m<0m < 0, every factor 1+∣m∣/j1 + |m|/j exceeds 11, so sns_n increases, and sn≥1+∣m∣∑j≤n1/j→∞s_n \ge 1 + |m|\sum_{j \le n} 1/j \to \infty.

The picture at x=−1x = -1 matches the function: (1+x)m→0(1 + x)^m \to 0 as x→−1x \to -1 when m>0m > 0, but →+∞\to +\infty when m<0m < 0. At x=1x = 1 the function is perfectly finite (2m2^m) for every mm, and yet the series diverges for m≤−1m \le -1: a power series can fail to represent a function at a point where the function is harmless.

Worked examples and exercises

Exercise Two endpoint sums ●○○ Hardy: §222

Find the sums of 1−12+1⋅32⋅4−1⋅3⋅52⋅4⋅6+…1 - \frac12 + \frac{1\cdot 3}{2\cdot 4} - \frac{1\cdot3\cdot5}{2\cdot4\cdot6} + \dots and of ∑n≥0(−1)n(1/2n)\sum_{n \ge 0} (-1)^n\binom{1/2}{n}.

Hint
Identify each as a binomial series at x=±1x = \pm1: what are (−1/2n)\binom{-1/2}{n} and mm?

Solution

(−1/2n)=(−1)n1⋅3⋯(2n−1)2⋅4⋯(2n)\binom{-1/2}{n} = (-1)^n\frac{1\cdot3\cdots(2n-1)}{2\cdot4\cdots(2n)}, so the first series is the binomial series with m=−12m = -\tfrac12 at x=1x = 1. Since m>−1m > -1 it converges, to 2−1/2=1/2≈0.70712^{-1/2} = 1/\sqrt2 \approx 0.7071. The second is the series with m=12m = \tfrac12 at x=−1x = -1; since m>0m > 0 it converges, to 00. (By contrast m=−12m = -\tfrac12 at x=−1x = -1 gives ∑1⋅3⋯(2n−1)2⋅4⋯(2n)\sum \frac{1\cdot3\cdots(2n-1)}{2\cdot4\cdots(2n)}, which diverges.)

Exercise ●○○ Hardy: Examples XCIII, 1

Show that for −1<x<1-1 < x < 1,

11+x2=1−12x2+1⋅32⋅4x4−…,11−x2=1+12x2+1⋅32⋅4x4+… .\frac{1}{\sqrt{1+x^2}} = 1 - \frac12x^2 + \frac{1\cdot3}{2\cdot4}x^4 - \dots, \qquad \frac{1}{\sqrt{1-x^2}} = 1 + \frac12x^2 + \frac{1\cdot3}{2\cdot4}x^4 + \dots .

Do the series converge at x=±1x = \pm 1?

Hint
Use m=−12m = -\tfrac12 with x2x^2 or −x2-x^2 in place of xx.

Solution

With m=−12m = -\tfrac12, (−1/2n)=(−1)n1⋅3⋯(2n−1)2⋅4⋯(2n)\binom{-1/2}{n} = (-1)^n\frac{1\cdot3\cdots(2n-1)}{2\cdot4\cdots(2n)}. Substituting x2x^2 (which lies in [0,1)[0,1)) gives the first series and −x2-x^2 gives the second. At x=±1x = \pm1 the first becomes the series of the previous exercise at x=1x = 1, convergent to 1/21/\sqrt2 since m>−1m > -1. The second becomes the series at x=−1x = -1 with m<0m < 0, which diverges to +∞+\infty, matching 1/1−x2→∞1/\sqrt{1-x^2} \to \infty.

Exercise Square roots by hand ●●○ Hardy: Examples XCIII, 2

Write 67=81+3/64\sqrt{67} = 8\sqrt{1 + 3/64} and expand. Show that the first two terms give 8316=8.18758\frac{3}{16} = 8.1875, that this is too large, and that the error is less than (3/64)2<0.003(3/64)^2 < 0.003.

Hint
With u=3/64u = 3/64, 1+u=1+12u−18u2+116u3−…\sqrt{1+u} = 1 + \frac12u - \frac18u^2 + \frac1{16}u^3 - \dots. Is this an alternating series with decreasing terms?

Solution

81+u=8(1+12u−18u2+116u3−… )8\sqrt{1 + u} = 8\bigl(1 + \frac12u - \frac18u^2 + \frac1{16}u^3 - \dots\bigr) with u=364u = \frac3{64}. After the first term the signs alternate (the coefficients of (1+u)1/2(1+u)^{1/2} alternate from n=1n = 1 on), and the terms decrease in size, since the ratio of successive terms is n−1/2n+1u<u<1\frac{n - 1/2}{n+1}u < u < 1. So the sum lies between consecutive partial sums. Stopping after 8(1+12u)=8+316=8.18758(1 + \frac12u) = 8 + \frac{3}{16} = 8.1875, the next term −8⋅18u2=−u2-8\cdot\frac18u^2 = -u^2 is negative, so the approximation is too large, by less than u2=94096≈0.0022u^2 = \frac{9}{4096} \approx 0.0022. Indeed 67=8.18535…\sqrt{67} = 8.18535\ldots, an error of 0.002150.00215.

Exercise ●●○ Hardy: §222 (2); Examples LXXXI, 5

Prove the formula ∑k=0n(−1)k(mk)=(−1)n(m−1n)\sum_{k=0}^{n}(-1)^k\binom{m}{k} = (-1)^n\binom{m-1}{n} used in part 3, and check it for m=−1m = -1.

Hint
Induction on nn, using (m−1n)+(m−1n+1)=(mn+1)\binom{m-1}{n} + \binom{m-1}{n+1} = \binom{m}{n+1}.

Solution

For n=0n = 0 both sides are 11. If the formula holds for nn, then

∑k=0n+1(−1)k(mk)=(−1)n(m−1n)+(−1)n+1(mn+1)=(−1)n+1[(mn+1)−(m−1n)]=(−1)n+1(m−1n+1),\sum_{k=0}^{n+1}(-1)^k\binom{m}{k} = (-1)^n\binom{m-1}{n} + (-1)^{n+1}\binom{m}{n+1} = (-1)^{n+1}\Bigl[\binom{m}{n+1} - \binom{m-1}{n}\Bigr] = (-1)^{n+1}\binom{m-1}{n+1},

by Pascal’s rule (which holds for real mm: (m−1n)+(m−1n+1)=(m−1n)(n+1)+(m−1−n)n+1=(mn+1)\binom{m-1}{n} + \binom{m-1}{n+1} = \binom{m-1}{n}\frac{(n+1) + (m-1-n)}{n+1} = \binom{m}{n+1}). For m=−1m = -1: (−1k)=(−1)k\binom{-1}{k} = (-1)^k, so the left side is n+1n + 1; and (−1)n(−2n)=(−1)n(−1)n(n+1)=n+1(-1)^n\binom{-2}{n} = (-1)^n(-1)^n(n+1) = n+1.

Check yourself

Check yourself Choose one

For which mm (not a non-negative integer) does 1+m+m(m−1)2!+m(m−1)(m−2)3!+…1 + m + \frac{m(m-1)}{2!} + \frac{m(m-1)(m-2)}{3!} + \dots converge?

Why

This is the series at x=1x = 1. It converges exactly when the terms (mn)\binom{m}{n} tend to 00, which happens exactly when m>−1m > -1. At m=−1m = -1 the terms are ±1\pm1 and the partial sums alternate between 11 and 00. That 2m2^m exists for all mm is irrelevant: the series need not represent the function at the edge.

Check yourself Enter a number

What is the sum of 1−12+38−516+35128−…1 - \frac12 + \frac38 - \frac5{16} + \frac{35}{128} - \dots, the binomial series with m=−12m = -\tfrac12 at x=1x = 1? (Decimals are fine.)

Why

m=−12>−1m = -\tfrac12 > -1, so the series converges to 2−1/2=1/2≈0.70712^{-1/2} = 1/\sqrt2 \approx 0.7071. It converges very slowly: the terms decrease like n−1/2n^{-1/2}, the case m+1=12m + 1 = \tfrac12 of the rule ∣(mn)∣≈Cn−(m+1)|\binom{m}{n}| \approx C n^{-(m+1)}.

Check yourself True or false?

The binomial series for (1+x)−1/2(1+x)^{-1/2} converges at x=−1x = -1.

Why

False. At x=−1x = -1 the series converges only for m≥0m \ge 0. For m=−12m = -\tfrac12 all its terms are positive and its partial sums ∏j≤n(1+12j)\prod_{j\le n}(1 + \frac{1}{2j}) grow without bound, as does (1+x)−1/2(1 + x)^{-1/2} when x→−1x \to -1.

Check yourself Choose one

What new fact was needed to extend the binomial theorem of §168 to irrational mm?

Why

The old proof used only the derivatives of (1+x)m(1+x)^m. Once the power is defined for irrational mm via the exponential and logarithm, the chain rule gives the same derivatives, and the proof goes through unchanged. A limit argument in mm would also be possible but is not needed.

Hardy’s language

If m+1>0m + 1 > 0 then unu_n ultimately alternates in sign and decreases steadily to zero, so that the series is convergent.

— G. H. Hardy, §222

Summary

Summary
  • For every real mm and −1<x<1-1 < x < 1, (1+x)m=∑(mn)xn(1+x)^m = \sum\binom{m}{n}x^n: the proof of §168 needs only ddx(1+x)m=m(1+x)m−1\frac{d}{dx}(1+x)^m = m(1+x)^{m-1}.
  • (mn)→0\binom{m}{n} \to 0 exactly when m>−1m > -1 (for mm not a non-negative integer); roughly ∣(mn)∣≈Cn−(m+1)|\binom{m}{n}| \approx C n^{-(m+1)}.
  • At x=1x = 1: convergent exactly when m>−1m > -1, with sum 2m2^m; the integral remainder is at most ∣(mn)∣|\binom{m}{n}|.
  • At x=−1x = -1: the partial sums are ∏j≤n(1−m/j)\prod_{j\le n}(1 - m/j); convergent (to 00) for m>0m > 0, divergent to +∞+\infty for m<0m < 0.