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

The power series for exp z

§242 35 min

The series 1 + z + z²/2! + … converges to exp z for every complex z; two proofs, one of which needs nothing about the series for sine and cosine.

Why this matters

We now have two candidates for “ee to the power zz” when zz is complex. One is exp⁡z=ex(cos⁡y+isin⁡y)\exp z = e^x(\cos y + i\sin y), obtained by inverting the logarithm. The other is the sum of the series 1+z+z2/2!+⋯1 + z + z^2/2! + \cdots, which converges for every complex zz (§198) and equals exe^x when zz is real (§219). §242 proves they are the same function. Many modern books start from the series; Hardy’s order shows that the series is a theorem about a function already known.

The idea

Draw the partial sums sn=1+z+⋯+zn/n!s_n = 1 + z + \dots + z^n/n! in the Argand diagram. Each new term zn/n!z^n/n! is a step from sn−1s_{n-1} to sns_n: its direction turns by am⁡z\operatorname{am} z from one step to the next (multiplying by zz turns by am⁡z\operatorname{am} z), and its length is multiplied by ∣z∣/n|z|/n, which eventually becomes small. So the partial sums form a polygonal spiral whose sides first grow, as long as n<∣z∣n < |z|, then shrink rapidly, and the spiral closes in on a point. The theorem says that point is ex(cos⁡y+isin⁡y)e^x(\cos y + i\sin y).

Explore Partial sums of the exponential series

Try this. Drag zz to πi\pi i (on the imaginary axis, at height about 3.143.14) and increase the number of terms from 00 to 1515. Then drag zz further out, to modulus about 44, and see how many terms are needed. Try a negative real zz as well.

1+z+z22!+⋯  →  exp⁡z1 + z + \frac{z^2}{2!} + \cdots \;\to\; \exp z
ReIm
  • z (drag it)
  • partial sums sₖ, one side per term
  • the sum
z=0.6+2.4iz = 0.6 + 2.4i, ∣z∣=2.47|z| = 2.47. Partial sum: −2.792+1.968i-2.792 + 1.968i. Sum: −1.3436+1.2308i-1.3436 + 1.2308i; distance 1.631.63. Next term has modulus 1.561.56.
What to notice

At z=πiz = \pi i the polygon turns a right angle at every vertex and spirals in to −1-1: Euler's exp⁡(πi)=−1\exp(\pi i) = -1. The sides grow while n<∣z∣n < |z| and shrink after that; once nn is a few times ∣z∣|z| the partial sums are indistinguishable from the sum. For negative real zz the polygon zigzags along the real axis, overshooting in both directions, because the terms alternate in sign.

Precise statements

Theorem The exponential series Hardy: §242 (1)

For every complex zz,

exp⁡z=1+z+z22!+z33!+⋯=∑n=0∞znn!,\exp z = 1 + z + \frac{z^2}{2!} + \frac{z^3}{3!} + \cdots = \sum_{n=0}^{\infty}\frac{z^n}{n!},

the series being absolutely convergent.

Write F(z)F(z) for the sum of the series. The proof rests on one property of FF that comes straight from the series.

Lemma The functional equation of the series Hardy: §242 (2); cf. Examples LXXXI, 7

For all complex z1z_1, z2z_2: F(z1)F(z2)=F(z1+z2)F(z_1)F(z_2) = F(z_1 + z_2).

Proofs

Proof The functional equation of the series
Strategy

Multiply the two absolutely convergent series term by term (§202), collect the terms of total degree nn, and recognise the binomial expansion of (z1+z2)n(z_1 + z_2)^n.

Absolute convergence

∑∣z∣n/n!\sum |z|^n/n! converges for every zz (ratio test: ∣z∣/(n+1)→0|z|/(n+1) \to 0), so both series are absolutely convergent and their product may be formed as a Cauchy product.

Collect terms of degree nn

The terms of degree nn in F(z1)F(z2)F(z_1)F(z_2) are ∑k=0nz1kk!z2n−k(n−k)!=1n!∑k=0n(nk)z1kz2n−k=(z1+z2)nn!\displaystyle\sum_{k=0}^{n}\frac{z_1^k}{k!}\frac{z_2^{n-k}}{(n-k)!} = \frac{1}{n!}\sum_{k=0}^n\binom nk z_1^k z_2^{n-k} = \frac{(z_1 + z_2)^n}{n!}.

Add up

So F(z1)F(z2)=∑n(z1+z2)n/n!=F(z1+z2)F(z_1)F(z_2) = \sum_n (z_1 + z_2)^n/n! = F(z_1 + z_2).

Proof The exponential series (first proof)
Strategy

Split z=x+iyz = x + iy and use the functional equation: F(z)=F(x)F(iy)F(z) = F(x)F(iy). The real factor is exe^x by Chapter IX. For the imaginary factor, separate the series into real and imaginary parts and recognise the series for cos⁡y\cos y and sin⁡y\sin y.

Split

By the lemma, F(x+iy)=F(x)F(iy)F(x + iy) = F(x)F(iy).

The real factor

For real xx, F(x)=exF(x) = e^x (§219).

The imaginary factor

The powers of iyiy cycle through 1,iy,−y2,−iy3,y4,…1, iy, -y^2, -iy^3, y^4, \dots. Grouping the real and imaginary terms (allowed, since the series converges absolutely), F(iy)=(1−y22!+y44!−⋯ )+i(y−y33!+⋯ )=cos⁡y+isin⁡yF(iy) = \bigl(1 - \frac{y^2}{2!} + \frac{y^4}{4!} - \cdots\bigr) + i\bigl(y - \frac{y^3}{3!} + \cdots\bigr) = \cos y + i\sin y, by the power series for the real cosine and sine (§152).

Conclude

F(z)=ex(cos⁡y+isin⁡y)=exp⁡zF(z) = e^x(\cos y + i\sin y) = \exp z by §233.

Hardy adds a second proof that avoids the sine and cosine series and uses only differentiation. It is worth knowing because the same pattern (show a quotient has zero derivative) recurs everywhere.

Proof The exponential series (second proof)
Strategy

Put f(y)=F(iy)f(y) = F(iy) for real yy. The functional equation shows f′(y)=if(y)f'(y) = if(y), the differential equation satisfied by cos⁡y+isin⁡y\cos y + i\sin y. Then show that f(y)(cos⁡y−isin⁡y)f(y)(\cos y - i\sin y) has derivative zero, so it is constant, equal to its value 11 at y=0y = 0.

A difference quotient

For real k≠0k \ne 0, f(y+k)−f(y)k=f(y)f(k)−1k\dfrac{f(y + k) - f(y)}{k} = f(y)\dfrac{f(k) - 1}{k}, and f(k)−1k=i{1+ik2!+(ik)23!+⋯ }=i(1+ρ)\dfrac{f(k) - 1}{k} = i\Bigl\{1 + \dfrac{ik}{2!} + \dfrac{(ik)^2}{3!} + \cdots\Bigr\} = i(1 + \rho).

The error term is small

∣ρ∣≤∣k∣2!+∣k∣23!+⋯≤(e−2)∣k∣|\rho| \le \dfrac{|k|}{2!} + \dfrac{|k|^2}{3!} + \cdots \le (e - 2)|k| for ∣k∣≤1|k| \le 1, so ρ→0\rho \to 0 as k→0k \to 0. Hence ff is differentiable and f′(y)=if(y)f'(y) = if(y).

A quotient with zero derivative

Let g(y)=f(y)(cos⁡y−isin⁡y)g(y) = f(y)(\cos y - i\sin y). By the product rule, g′(y)=if(y)(cos⁡y−isin⁡y)−f(y)(sin⁡y+icos⁡y)=f(y){icos⁡y+sin⁡y−sin⁡y−icos⁡y}=0g'(y) = if(y)(\cos y - i\sin y) - f(y)(\sin y + i\cos y) = f(y)\{i\cos y + \sin y - \sin y - i\cos y\} = 0.

Conclude

The real and imaginary parts of gg have zero derivative, so they are constant (mean value theorem). Since g(0)=f(0)=1g(0) = f(0) = 1, g(y)=1g(y) = 1, and f(y)=1/(cos⁡y−isin⁡y)=cos⁡y+isin⁡yf(y) = 1/(\cos y - i\sin y) = \cos y + i\sin y. As before, F(x+iy)=exf(y)=exp⁡zF(x + iy) = e^x f(y) = \exp z.

Worked examples and exercises

Exercise Euler's identity from the series ●○○ Hardy: §242

Use the series to show that exp⁡(πi)=−1\exp(\pi i) = -1 and exp⁡(2πi)=1\exp(2\pi i) = 1, and estimate how many terms of the series give exp⁡(πi)\exp(\pi i) to within 0.0010.001.

Hint
The theorem identifies the sum with cos⁡π+isin⁡π\cos\pi + i\sin\pi. For the estimate, compare the tail with a geometric series once n>∣z∣n > |z|.

Solution

By the theorem, exp⁡(πi)=cos⁡π+isin⁡π=−1\exp(\pi i) = \cos\pi + i\sin\pi = -1 and exp⁡(2πi)=1\exp(2\pi i) = 1. For the tail: if n+1>2∣z∣n + 1 > 2|z|, then each term after zn+1/(n+1)!z^{n+1}/(n+1)! is at most half the previous one in modulus, so the error after the term in znz^n is at most 2∣z∣n+1/(n+1)!2|z|^{n+1}/(n+1)!. With ∣z∣=π|z| = \pi: n=11n = 11 gives 2π12/12!≈2×9.24×105/4.79×108≈0.00392\pi^{12}/12! \approx 2 \times 9.24 \times 10^5 / 4.79 \times 10^8 \approx 0.0039; n=12n = 12 gives 2π13/13!≈0.00093<0.0012\pi^{13}/13! \approx 0.00093 < 0.001. So the terms up to z12z^{12} (thirteen terms) suffice; the explorer shows the polygon is indistinguishable from −1-1 by then.

Exercise exp of (1 + i)z ●●○ Hardy: Examples XCVII, 4

Show that exp⁡{(1+i)z}=∑n=0∞2n/2exp⁡(14nπi)znn!\exp\{(1 + i)z\} = \displaystyle\sum_{n=0}^\infty 2^{n/2}\exp(\tfrac14 n\pi i)\frac{z^n}{n!}.

Hint
Apply the theorem to (1+i)z(1 + i)z and write 1+i1 + i in polar form.

Solution

By the theorem with (1+i)z(1 + i)z in place of zz, exp⁡{(1+i)z}=∑(1+i)nzn/n!\exp\{(1 + i)z\} = \sum (1 + i)^n z^n/n!. Now 1+i=2(cos⁡14π+isin⁡14π)=21/2exp⁡(14πi)1 + i = \sqrt2(\cos\tfrac14\pi + i\sin\tfrac14\pi) = 2^{1/2}\exp(\tfrac14\pi i), so (1+i)n=2n/2exp⁡(14nπi)(1 + i)^n = 2^{n/2}\exp(\tfrac14 n\pi i) by the addition theorem. Taking real parts for real z=xz = x gives excos⁡x=∑2n/2cos⁡(14nπ) xn/n!e^x\cos x = \sum 2^{n/2}\cos(\tfrac14 n\pi)\,x^n/n!.

Exercise Two inequalities ●○○ Hardy: §242

Show that ∣exp⁡z∣≤exp⁡∣z∣|\exp z| \le \exp|z| and ∣exp⁡z−1∣≤exp⁡∣z∣−1≤∣z∣exp⁡∣z∣|\exp z - 1| \le \exp|z| - 1 \le |z|\exp|z| for every complex zz.

Hint
Compare the series term by term with the series for exp⁡∣z∣\exp|z|.

Solution

∣exp⁡z∣=∣∑zn/n!∣≤∑∣z∣n/n!=exp⁡∣z∣|\exp z| = |\sum z^n/n!| \le \sum |z|^n/n! = \exp|z|. Similarly ∣exp⁡z−1∣=∣∑n≥1zn/n!∣≤∑n≥1∣z∣n/n!=exp⁡∣z∣−1|\exp z - 1| = |\sum_{n\ge1} z^n/n!| \le \sum_{n \ge 1}|z|^n/n! = \exp|z| - 1. Finally exp⁡∣z∣−1=∣z∣∑n≥1∣z∣n−1/n!≤∣z∣∑m≥0∣z∣m/m!=∣z∣exp⁡∣z∣\exp|z| - 1 = |z|\sum_{n\ge1}|z|^{n-1}/n! \le |z|\sum_{m\ge0}|z|^m/m! = |z|\exp|z|. (The first inequality is also clear from ∣exp⁡z∣=ex≤e∣z∣|\exp z| = e^x \le e^{|z|}.)

Exercise The derivative of exp along a line ●●○ Hardy: §242

Let z0z_0 and cc be complex and tt real. Show that ddtexp⁡(z0+ct)=cexp⁡(z0+ct)\dfrac{d}{dt}\exp(z_0 + ct) = c\exp(z_0 + ct).

Hint
Copy the difference-quotient step of the second proof, using the functional equation.

Solution

exp⁡(z0+c(t+k))−exp⁡(z0+ct)k=exp⁡(z0+ct) exp⁡(ck)−1k\dfrac{\exp(z_0 + c(t + k)) - \exp(z_0 + ct)}{k} = \exp(z_0 + ct)\,\dfrac{\exp(ck) - 1}{k}, and exp⁡(ck)−1k=c{1+ck2!+(ck)23!+⋯ }\dfrac{\exp(ck) - 1}{k} = c\{1 + \tfrac{ck}{2!} + \tfrac{(ck)^2}{3!} + \cdots\}, which tends to cc as k→0k \to 0 because the bracket differs from 11 by at most ∣ck∣exp⁡∣ck∣|ck|\exp|ck| (previous exercise). This is the rule used in §246.

Check yourself

Check yourself True or false?

The series ∑zn/n!\sum z^n/n! converges for every complex zz, and its sum is never 00.

Why

True. It converges absolutely everywhere (ratio ∣z∣/(n+1)→0|z|/(n+1) \to 0), and by the theorem its sum is exp⁡z=ex(cos⁡y+isin⁡y)\exp z = e^x(\cos y + i\sin y), whose modulus exe^x is positive. (From the series alone: F(z)F(−z)=F(0)=1F(z)F(-z) = F(0) = 1.)

Check yourself Choose one

In the polygon of partial sums for exp⁡z\exp z, the side added at step nn is zn/n!z^n/n!. When do the sides start to get shorter?

Why

∣zn/n!∣÷∣zn−1/(n−1)!∣=∣z∣/n|z^n/n!| \div |z^{n-1}/(n-1)!| = |z|/n. While n<∣z∣n < |z| this ratio exceeds 11 and the sides grow; afterwards they shrink faster and faster. The series converges for every zz; large ∣z∣|z| only delays the shrinking. The amplitude controls the turning of the sides, not their length.

Check yourself Choose one

In the first proof, what is the functional equation F(z1)F(z2)=F(z1+z2)F(z_1)F(z_2) = F(z_1 + z_2) used for?

Why

Convergence comes from the ratio test. The functional equation splits the problem into two known cases: F(x)=exF(x) = e^x (Chapter IX) and F(iy)=cos⁡y+isin⁡yF(iy) = \cos y + i\sin y (the real series for sine and cosine).

Hardy’s language

Summary

Summary
  • exp⁡z=∑n≥0zn/n!\exp z = \sum_{n \ge 0} z^n/n! for every complex zz; the series converges absolutely.
  • The key step is F(z1)F(z2)=F(z1+z2)F(z_1)F(z_2) = F(z_1 + z_2), from multiplying series and the binomial theorem.
  • First proof: F(x+iy)=F(x)F(iy)=ex(cos⁡y+isin⁡y)F(x + iy) = F(x)F(iy) = e^x(\cos y + i\sin y). Second proof: f′=iff' = if, and f(y)(cos⁡y−isin⁡y)f(y)(\cos y - i\sin y) is constant.
  • The partial sums form a spiralling polygon; its sides grow until n≈∣z∣n \approx |z| and then shrink factorially.