Chapter X · The general theory of the logarithmic, exponential and circular functions

The general binomial theorem

§247 35 min

For every complex m and |z| < 1, the binomial series 1 + mz + m(m−1)z²/2! + … converges to the principal value of (1 + z)ᵐ.

Why this matters

The binomial theorem began as an identity about (1+z)m(1 + z)^m for positive integers mm; Newton used it for fractional mm; Chapter IX (§222) proved it for real mm and real zz between −1-1 and 11. The last section of the book proves it in full generality: complex exponent, complex zz in the unit disc. Since (1+z)m(1 + z)^m is now in general infinitely many-valued, the theorem must also say which value the series picks out, and the answer is the principal one.

The idea

The coefficients (mn)=m(m−1)⋯(m−n+1)n!\binom mn = \dfrac{m(m - 1)\cdots(m - n + 1)}{n!} make sense for any complex mm, and the ratio of successive coefficients, m−nn+1\dfrac{m - n}{n + 1}, tends to −1-1. So the series converges for ∣z∣<1|z| < 1 (§198), just like a geometric series. The series is one-valued and continuous in zz, and equals 11 at z=0z = 0; the values of (1+z)m(1 + z)^m are exp⁡{mlog⁡(1+z)+2kπim}\exp\{m\log(1 + z) + 2k\pi i m\}, and the only choice of kk that gives a function equal to 11 at z=0z = 0 is one with exp⁡(2kπim)=1\exp(2k\pi i m) = 1, which gives the same function as k=0k = 0. So the principal value is the only candidate. The proof confirms it with Taylor’s theorem.

Explore The binomial series and the values of (1 + z)ᵐ

Try this. With m=12m = \tfrac12, add terms and see which hollow dot the polygon approaches. Then set Im⁡m\operatorname{Im}m to 11, so that mm is complex. Drag zz around inside the disc, and then switch on “keep zz on the circle” and move zz towards −1-1.

1+(m1)z+(m2)z2+⋯  →  exp⁡{mlog⁡(1+z)}1 + \binom{m}{1}z + \binom{m}{2}z^2 + \cdots \;\to\; \exp\{m\log(1+z)\}
ReIm
  • z (drag it)
  • partial sums sₖ, one side per term
  • principal value of (1 + z)^m
  • two other values of (1 + z)^m
z=0.35+0.55iz = 0.35 + 0.55i, ∣z∣=0.652|z| = 0.652. Partial sum: 1.1975+0.22688i1.1975 + 0.22688i. Sum: 1.1848+0.2321i1.1848 + 0.2321i; distance 0.01370.0137. Next term has modulus 0.01730.0173.
What to notice

The polygon always converges to the black hollow dot, the principal value exp⁡{mlog⁡(1+z)}\exp\{m\log(1 + z)\}, never to the ochre ones (the values with k=±1k = \pm1). For m=12m = \tfrac12 the other value is just −1+z-\sqrt{1 + z}; for complex mm the other values have quite different moduli. Near the boundary the convergence slows down, like a geometric series with ratio close to 11; on the circle itself the behaviour depends on mm (Hardy defers this harder case to Bromwich).

Precise statements

Theorem The general binomial theorem Hardy: §247, Theorem

For all complex mm and all complex zz with ∣z∣<1|z| < 1,

1+(m1)z+(m2)z2+⋯=exp⁡{mlog⁡(1+z)},1 + \binom m1 z + \binom m2 z^2 + \cdots = \exp\{m\log(1 + z)\},

the logarithm having its principal value. That is, the binomial series converges to the principal value of (1+z)m(1 + z)^m.

For mm a positive integer the series stops and this is the elementary binomial theorem; for real mm and real zz in (−1,1)(-1, 1) it is §222.

Proofs

Proof The general binomial theorem
Strategy

Fix zz with ∣z∣=r<1|z| = r < 1 and put ϕ(t)=(1+tz)m\phi(t) = (1 + tz)^m (principal value) for 0≤t≤10 \le t \le 1, so that ϕ(1)\phi(1) is the number we want and ϕ(0)=1\phi(0) = 1. Apply Taylor’s theorem with the integral form of the remainder to the real and imaginary parts of ϕ\phi, and show that the remainder tends to 00. The work is all in bounding ∣(1+tz)m−1∣|(1 + tz)^{m-1}|, which for complex mm involves the amplitude of 1+tz1 + tz.

The derivatives of ϕ\phi

By §246, ϕ′(t)=mz(1+tz)m−1\phi'(t) = mz(1 + tz)^{m-1}, and repeating, ϕ(n)(t)=m(m−1)⋯(m−n+1) zn(1+tz)m−n\phi^{(n)}(t) = m(m-1)\cdots(m - n + 1)\,z^n(1 + tz)^{m-n}, all powers principal (valid since ∣tz∣<1|tz| < 1). At t=0t = 0: ϕ(n)(0)/n!=(mn)zn\phi^{(n)}(0)/n! = \binom mn z^n.

Taylor's theorem with integral remainder

Applied to the real and imaginary parts of ϕ\phi (§167, §170): ϕ(1)=∑k=0n−1(mk)zk+Rn\phi(1) = \sum_{k=0}^{n-1}\binom mk z^k + R_n with Rn=1(n−1)!∫01(1−t)n−1ϕ(n)(t) dt=m(m−1)⋯(m−n+1)(n−1)!zn∫01(1+tz)m−1(1−t1+tz)n−1dt.R_n = \dfrac{1}{(n-1)!}\displaystyle\int_0^1 (1 - t)^{n-1}\phi^{(n)}(t)\,dt = \dfrac{m(m-1)\cdots(m-n+1)}{(n-1)!}z^n\int_0^1(1 + tz)^{m-1}\Bigl(\frac{1 - t}{1 + tz}\Bigr)^{n-1}dt.

Bound (1−t)/(1+tz)(1 - t)/(1 + tz)

∣1+tz∣≥1−tr|1 + tz| \ge 1 - tr, and 1−tr≥1−t1 - tr \ge 1 - t because r≤1r \le 1. So ∣1−t1+tz∣≤1−t1−tr≤1\Bigl|\dfrac{1 - t}{1 + tz}\Bigr| \le \dfrac{1 - t}{1 - tr} \le 1.

Bound (1+tz)m−1(1 + tz)^{m-1}

Write m=μ+iνm = \mu + i\nu. The principal value is exp⁡{(m−1)log⁡(1+tz)}\exp\{(m - 1)\log(1 + tz)\}, whose modulus is ∣1+tz∣μ−1e−νam⁡(1+tz)|1 + tz|^{\mu - 1}e^{-\nu\operatorname{am}(1 + tz)} with −π≤am⁡(1+tz)≤π-\pi \le \operatorname{am}(1 + tz) \le \pi. Since 1−r≤∣1+tz∣≤21 - r \le |1 + tz| \le 2, the first factor is at most 2μ−12^{\mu - 1} if μ≥1\mu \ge 1 and at most (1−r)μ−1(1 - r)^{\mu - 1} if μ<1\mu < 1; the second is at most eπ∣ν∣e^{\pi|\nu|}. So ∣(1+tz)m−1∣≤K|(1 + tz)^{m-1}| \le K with KK independent of tt and nn.

Bound the remainder

Hence ∣Rn∣≤K ∣m(m−1)⋯(m−n+1)∣(n−1)! rn=ρn|R_n| \le K\,\dfrac{|m(m-1)\cdots(m - n + 1)|}{(n - 1)!}\,r^n = \rho_n, say.

ρn→0\rho_n \to 0

ρn+1ρn=∣m−n∣n r→r<1\dfrac{\rho_{n+1}}{\rho_n} = \dfrac{|m - n|}{n}\,r \to r < 1, so ρn→0\rho_n \to 0 (as for nxn→0n x^n \to 0, Examples XXVII, 6). Therefore Rn→0R_n \to 0 and the series sums to ϕ(1)=exp⁡{mlog⁡(1+z)}\phi(1) = \exp\{m\log(1 + z)\}.

Worked examples and exercises

Exercise Checking a case by hand ●○○ Hardy: §247

Show that for m=−1m = -1 the theorem reduces to the geometric series, and that for m=2m = 2 it reduces to (1+z)2=1+2z+z2(1 + z)^2 = 1 + 2z + z^2. Are principal values an issue in these cases?

Hint
Compute (−1n)\binom{-1}{n} and (2n)\binom 2n.

Solution

(−1n)=(−1)(−2)⋯(−n)n!=(−1)n\binom{-1}{n} = \dfrac{(-1)(-2)\cdots(-n)}{n!} = (-1)^n, so the series is 1−z+z2−⋯=1/(1+z)1 - z + z^2 - \cdots = 1/(1 + z) for ∣z∣<1|z| < 1. (2n)\binom 2n is 1,2,11, 2, 1 for n=0,1,2n = 0, 1, 2 and 00 afterwards. For integer mm, exp⁡{mLog⁡(1+z)}\exp\{m\operatorname{Log}(1 + z)\} has the single value (1+z)m(1 + z)^m (the 2kπim2k\pi i m terms are multiples of 2πi2\pi i), so there is no choice of value to make.

Exercise Real and imaginary parts ●●○ Hardy: Examples XCIX, 1

Let mm be real and 0≤r<10 \le r < 1. By putting z=irz = ir in the theorem, show that

1−(m2)r2+(m4)r4−⋯=(1+r2)m/2cos⁡(marctan⁡r),(m1)r−(m3)r3+⋯=(1+r2)m/2sin⁡(marctan⁡r),1 - \binom m2 r^2 + \binom m4 r^4 - \cdots = (1 + r^2)^{m/2}\cos(m\arctan r), \qquad \binom m1 r - \binom m3 r^3 + \cdots = (1 + r^2)^{m/2}\sin(m\arctan r),

with arctan⁡r\arctan r between 00 and 14π\tfrac14\pi.

Hint
log⁡(1+ir)=12log⁡(1+r2)+iarctan⁡r\log(1 + ir) = \tfrac12\log(1 + r^2) + i\arctan r.

Solution

The powers (ir)n(ir)^n are 1,ir,−r2,−ir3,r4,…1, ir, -r^2, -ir^3, r^4, \dots, so the real part of the series is 1−(m2)r2+(m4)r4−⋯1 - \binom m2 r^2 + \binom m4 r^4 - \cdots and the imaginary part (m1)r−(m3)r3+⋯\binom m1 r - \binom m3 r^3 + \cdots (the series converges absolutely, so it may be split). On the other side, 1+ir1 + ir has modulus 1+r2\sqrt{1 + r^2} and principal amplitude arctan⁡r∈[0,14π)\arctan r \in [0, \tfrac14\pi), so exp⁡{mlog⁡(1+ir)}=(1+r2)m/2{cos⁡(marctan⁡r)+isin⁡(marctan⁡r)}\exp\{m\log(1 + ir)\} = (1 + r^2)^{m/2}\{\cos(m\arctan r) + i\sin(m\arctan r)\} for real mm. Equate real and imaginary parts.

Exercise cos mθ by the binomial series ●●○ Hardy: Examples XCIX, 3

Let mm be real and −14π<θ<14π-\tfrac14\pi < \theta < \tfrac14\pi. Show that cos⁡mθ=cos⁡mθ{1−(m2)tan⁡2θ+(m4)tan⁡4θ−⋯ }\cos m\theta = \cos^m\theta\bigl\{1 - \binom m2\tan^2\theta + \binom m4\tan^4\theta - \cdots\bigr\}.

Hint
cos⁡mθ+isin⁡mθ=cos⁡mθ (1+itan⁡θ)m\cos m\theta + i\sin m\theta = \cos^m\theta\,(1 + i\tan\theta)^m; why does ∣tan⁡θ∣<1|\tan\theta| < 1 matter?

Solution

Since ∣θ∣<14π|\theta| < \tfrac14\pi, cos⁡θ>0\cos\theta > 0 and ∣tan⁡θ∣<1|\tan\theta| < 1. The number 1+itan⁡θ=(cos⁡θ+isin⁡θ)/cos⁡θ1 + i\tan\theta = (\cos\theta + i\sin\theta)/\cos\theta has modulus 1/cos⁡θ1/\cos\theta and principal amplitude θ\theta, so its principal mm-th power is cos⁡−mθ (cos⁡mθ+isin⁡mθ)\cos^{-m}\theta\,(\cos m\theta + i\sin m\theta). By the theorem with z=itan⁡θz = i\tan\theta (∣z∣<1|z| < 1), this equals ∑(mn)(itan⁡θ)n\sum\binom mn(i\tan\theta)^n. Multiply by cos⁡mθ\cos^m\theta and take real parts. (The imaginary parts give the companion formula for sin⁡mθ\sin m\theta.) For integer mm the series terminates and this is the classical formula from De Moivre’s theorem, valid for all θ\theta.

Exercise An imaginary exponent ●●○ Hardy: Examples XCIX, 5

Let μ\mu be real and −1<z<1-1 < z < 1 real. Show that ∑n(iμn)zn=cos⁡{μlog⁡(1+z)}+isin⁡{μlog⁡(1+z)}\displaystyle\sum_n\binom{i\mu}{n}z^n = \cos\{\mu\log(1 + z)\} + i\sin\{\mu\log(1 + z)\}.

Hint
Apply the theorem with m=iμm = i\mu; log⁡(1+z)\log(1 + z) is real here.

Solution

By the theorem, the sum is exp⁡{iμlog⁡(1+z)}\exp\{i\mu\log(1 + z)\}. For real z∈(−1,1)z \in (-1, 1), log⁡(1+z)\log(1 + z) is the real logarithm, so the exponent is purely imaginary and exp⁡{iμlog⁡(1+z)}=cos⁡{μlog⁡(1+z)}+isin⁡{μlog⁡(1+z)}\exp\{i\mu\log(1 + z)\} = \cos\{\mu\log(1 + z)\} + i\sin\{\mu\log(1 + z)\}. In particular the sum has modulus 11 for every such zz: the real number 1+z1 + z raised to an imaginary power lies on the unit circle (principal value).

Check yourself

Check yourself Choose one

For m=12m = \tfrac12 and z=−34+0.1iz = -\tfrac34 + 0.1i, the binomial series converges to

Why

The theorem names the value: exp⁡{mlog⁡(1+z)}\exp\{m\log(1 + z)\} with the principal logarithm. Here 1+z=14+0.1i1 + z = \tfrac14 + 0.1i lies in the right half-plane, its principal amplitude is small, and the principal square root has positive real part. The series converges because ∣z∣<1|z| < 1, whatever the sign of Re⁡z\operatorname{Re}z.

Check yourself True or false?

For complex mm, the binomial series converges for every ∣z∣<1|z| < 1.

Why

True. The ratio of consecutive terms has modulus ∣m−nn+1∣ ∣z∣→∣z∣<1\bigl|\tfrac{m - n}{n + 1}\bigr|\,|z| \to |z| < 1. (If mm is a non-negative integer the series terminates and converges everywhere.)

Check yourself Choose one

In the proof, why does the bound on ∣(1+tz)m−1∣|(1 + tz)^{m-1}| contain the factor eπ∣ν∣e^{\pi|\nu|}, where ν=Im⁡m\nu = \operatorname{Im}m?

Why

For a complex exponent, the amplitude of the base affects the modulus of the power: (m−1)log⁡w=(μ−1+iν)(log⁡∣w∣+iam⁡w)(m - 1)\log w = (\mu - 1 + i\nu)(\log|w| + i\operatorname{am}w) has real part (μ−1)log⁡∣w∣−νam⁡w(\mu - 1)\log|w| - \nu\operatorname{am}w. Using principal values keeps ∣am⁡w∣≤π|\operatorname{am}w| \le \pi, so the factor is at most eπ∣ν∣e^{\pi|\nu|}. With other values of the power no such uniform bound would follow.

Hardy’s language

A more complete discussion of the binomial series, including the more difficult case in which | z | = 1, will be found on pp. 287 et seq. of Bromwich’s Infinite series.

— G. H. Hardy, §247

Summary

Summary
  • For all complex mm and ∣z∣<1|z| < 1: ∑(mn)zn=exp⁡{mlog⁡(1+z)}\sum\binom mn z^n = \exp\{m\log(1 + z)\}, the principal value of (1+z)m(1 + z)^m.
  • Proof: Taylor’s theorem with integral remainder for ϕ(t)=(1+tz)m\phi(t) = (1 + tz)^m, using ϕ(n)(t)=m⋯(m−n+1)zn(1+tz)m−n\phi^{(n)}(t) = m\cdots(m - n + 1)z^n(1 + tz)^{m - n}.
  • The key bound is ∣(1+tz)m−1∣≤K|(1 + tz)^{m-1}| \le K, where principal amplitudes keep the factor e−νam⁡(1+tz)e^{-\nu\operatorname{am}(1 + tz)} under control.
  • Real and imaginary parts give trigonometric expansions such as cos⁡mθ=cos⁡mθ{1−(m2)tan⁡2θ+⋯ }\cos m\theta = \cos^m\theta\{1 - \binom m2\tan^2\theta + \cdots\}.