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

Functions of a complex variable

§§227–228 25 min

Why “any complex function of x and y” is too broad to be called a function of z, and which functions of z this chapter will study.

Why this matters

Chapter X extends the logarithm, the exponential and the sine and cosine from real to complex numbers. Before doing so, Hardy pauses on a question that sounds pedantic but is not: what should we mean by a function of zz? The naive answer turns out to be useless, and the reason why it is useless points straight at what makes complex analysis special.

The idea

For a real variable, a function is any rule that gives a value yy for each xx (§20). Copy that for complex numbers: ZZ is a function of zz if some rule gives a value ZZ for each zz. But giving z=x+iyz = x + iy is exactly the same as giving the pair of real numbers (x,y)(x, y). So under this definition a “function of zz” is nothing but a complex-valued function f(x,y)+ig(x,y)f(x, y) + i g(x, y) of two real variables. Hardy’s examples are x−iyx - iy, xyxy, ∣z∣=x2+y2|z| = \sqrt{x^2 + y^2} and am⁡z\operatorname{am} z. Each is a perfectly good rule, but calling it a function “of zz” adds nothing: the definition, as Hardy says, does not define a new idea.

What is new about polynomials such as z2z^2, or rational functions such as 1/z1/z? They are built from zz by the four operations of complex arithmetic, and that gives them a geometric property the others lack. Near any point, such a function behaves like multiplication by a fixed complex number, which is a rotation followed by a stretch (§44). A rotation and a stretch carry a tiny square to a tiny square. Watch this happen.

Explore Which maps keep small squares square?

Try this. Start with z2z^2 and look at the images of the little grid squares away from 00. Then switch to 1/z1/z, to Hardy's x−iyx - iy and xyxy, and to the shear x+i(x+y)x + i(x + y).

w=z2=(x2−y2)+2ixyw = z^2 = (x^2 - y^2) + 2ixy
z-plane
xy
w-plane, w = f(z)
uv
  • horizontal lines and their images
  • vertical lines and their images
  • circle (drag centre and rim) and its image
  • w = 0
Grid
Circle: centre 0.4+0.2i0.4 + 0.2i, radius 0.50.5 (it encloses z = 0). Its image winds 2 times round w=0w = 0 (counter-clockwise counts +).
What to notice

Under z2z^2 and 1/z1/z the images of the grid lines still cross at right angles, and small squares come out as (curved) small squares, turned and resized. x−iyx - iy keeps squares square but reverses their orientation: it is a reflection, not a rotation. xyxy squashes the whole plane onto the real axis. The shear turns squares into parallelograms. Only the genuine functions of zz act locally like multiplication by a complex number.

So the useful notion is a restricted one: out of all complex functions of xx and yy, pick a special class that deserves the name “function of zz”. Describing that class in general is the business of complex analysis proper, and Hardy deliberately stops short of it. He works instead with particular functions, each defined directly: polynomials, rational functions and roots (all from Chapter III), algebraical functions, and now the logarithm, exponential and circular functions.

Precise statements

Definition Complex function of two real variables Hardy: §227

A complex function of the real variables xx and yy is a rule assigning to each pair (x,y)(x, y) in some region a complex number

Z=f(x,y)+i g(x,y),Z = f(x, y) + i\, g(x, y),

with ff and gg real. Since z=x+iyz = x + iy determines (x,y)(x, y) and conversely, every such rule could be called a function of zz in the wide sense; Hardy declines to do so.

Convention The functions of zz studied in this chapter Hardy: §228

In Chapter X a “function of zz” is always one of a list of functions defined directly: polynomials P(z)P(z) (§39), rational functions P(z)/Q(z)P(z)/Q(z) (§46), roots (§47), algebraical functions defined by polynomial equations in zz and ww, and the logarithmic, exponential and circular functions defined in §230 onwards. The number zz at which the function is evaluated is called the argument of the function.

The property seen in the explorer can be stated and proved for every polynomial.

Theorem A polynomial acts locally like multiplication

Let PP be a polynomial and zz a fixed complex number. Then

P(z+h)−P(z)=h {P′(z)+ϵ(h)},ϵ(h)→0 as h→0,P(z + h) - P(z) = h\,\{P'(z) + \epsilon(h)\}, \qquad \epsilon(h) \to 0 \text{ as } h \to 0,

where P′P' is the polynomial obtained by differentiating PP formally. So for small hh the map z+h↦P(z+h)z + h \mapsto P(z + h) moves points by approximately h⋅P′(z)h \cdot P'(z): a rotation through am⁡P′(z)\operatorname{am} P'(z) and a stretch by ∣P′(z)∣|P'(z)|, when P′(z)≠0P'(z) \ne 0.

Contrast the map z↦zˉ=x−iyz \mapsto \bar z = x - iy, which is not of this kind at all.

Theorem x−iyx - iy is not locally a multiplication

There is no complex number cc such that hˉ=c h\bar h = c\,h for all small hh. Equivalently, the quotient (z+h‾−zˉ)/h=hˉ/h(\overline{z + h} - \bar z)/h = \bar h / h has no limit as h→0h \to 0.

Proofs

Proof A polynomial acts locally like multiplication
Strategy

It is enough to prove the statement for a single power znz^n, because a polynomial is a sum of constant multiples of powers and both sides of the claim add up. For znz^n expand (z+h)n(z + h)^n by the binomial theorem, which holds for complex numbers because it uses only the rules of algebra.

Expand (z+h)n−zn(z+h)^n - z^n

By the binomial theorem, (z+h)n−zn=nzn−1h+(n2)zn−2h2+⋯+hn(z + h)^n - z^n = n z^{n-1} h + \binom n2 z^{n-2} h^2 + \dots + h^n. Take out the factor hh: the bracket is nzn−1+ϵn(h)n z^{n-1} + \epsilon_n(h), where ϵn(h)=(n2)zn−2h+⋯+hn−1\epsilon_n(h) = \binom n2 z^{n-2} h + \dots + h^{n-1}.

ϵn(h)→0\epsilon_n(h) \to 0

Each term of ϵn(h)\epsilon_n(h) contains at least one factor hh. If ∣h∣≤1|h| \le 1 then ∣ϵn(h)∣≤∣h∣ {(n2)∣z∣n−2+⋯+1}=K∣h∣|\epsilon_n(h)| \le |h|\,\bigl\{\binom n2 |z|^{n-2} + \dots + 1\bigr\} = K|h| with KK independent of hh, so ϵn(h)→0\epsilon_n(h) \to 0.

Add up the powers

If P(z)=a0zm+a1zm−1+⋯+amP(z) = a_0 z^m + a_1 z^{m-1} + \dots + a_m, apply the previous steps to every power and add: P(z+h)−P(z)=h{P′(z)+ϵ(h)}P(z + h) - P(z) = h\{P'(z) + \epsilon(h)\} where ϵ(h)\epsilon(h) is a fixed combination of the ϵn(h)\epsilon_n(h) and so tends to 00.

Read it geometrically

Multiplying by c=P′(z)≠0c = P'(z) \neq 0 rotates every displacement hh through am⁡c\operatorname{am} c and multiplies its length by ∣c∣|c| (§44). Two short displacements at right angles therefore go to two displacements at right angles, and a tiny square goes, up to an error that is small compared with its size, to a tiny square.

Proof x−iyx - iy is not locally a multiplication
Strategy

Test the supposed identity hˉ=ch\bar h = c h on two different directions of hh and get a contradiction.

Real hh

If h=th = t is real and non-zero, then hˉ=t\bar h = t, so c t=tc\,t = t forces c=1c = 1.

Imaginary hh

If h=ith = it with tt real and non-zero, then hˉ=−it\bar h = -it, so c (it)=−itc\,(it) = -it forces c=−1c = -1.

Conclude

No number is both 11 and −1-1. In the language of limits: hˉ/h\bar h/h equals 11 on the real axis and −1-1 on the imaginary axis, however small hh is, so it has no limit as h→0h \to 0.

Worked examples and exercises

Hardy’s §§227–228 contain no numbered examples; these exercises make the ideas concrete.

Exercise Real and imaginary parts ●○○ Hardy: §227

Write z3z^3 and 1/z1/z in the form u(x,y)+iv(x,y)u(x, y) + i v(x, y). Check that in both cases ∂u/∂x=∂v/∂y\partial u/\partial x = \partial v/\partial y and ∂u/∂y=−∂v/∂x\partial u/\partial y = -\partial v/\partial x (away from z=0z = 0 for 1/z1/z).

Hint
Expand (x+iy)3(x + iy)^3; for 1/z1/z multiply numerator and denominator by x−iyx - iy.

Solution

z3=(x3−3xy2)+i(3x2y−y3)z^3 = (x^3 - 3xy^2) + i(3x^2 y - y^3). Then ux=3x2−3y2=vyu_x = 3x^2 - 3y^2 = v_y and uy=−6xy=−vxu_y = -6xy = -v_x.

1/z=xx2+y2−iyx2+y21/z = \dfrac{x}{x^2 + y^2} - i \dfrac{y}{x^2 + y^2}. With u=x/(x2+y2)u = x/(x^2+y^2), v=−y/(x2+y2)v = -y/(x^2+y^2): ux=y2−x2(x2+y2)2u_x = \dfrac{y^2 - x^2}{(x^2+y^2)^2} and vy=−(x2+y2)+2y2(x2+y2)2=y2−x2(x2+y2)2v_y = \dfrac{-(x^2+y^2) + 2y^2}{(x^2+y^2)^2} = \dfrac{y^2 - x^2}{(x^2+y^2)^2}; also uy=−2xy(x2+y2)2u_y = \dfrac{-2xy}{(x^2+y^2)^2} and vx=2xy(x2+y2)2v_x = \dfrac{2xy}{(x^2+y^2)^2}, so uy=−vxu_y = -v_x. These are the Cauchy–Riemann equations; see the note at the end.

Exercise Hardy's examples fail the test ●○○ Hardy: §227

Show that u+iv=xyu + iv = xy and u+iv=∣z∣u + iv = |z| do not satisfy ux=vyu_x = v_y, uy=−vxu_y = -v_x on any open set.

Hint
In both examples v=0v = 0, so the equations would force ux=uy=0u_x = u_y = 0.

Solution

For xyxy: u=xyu = xy, v=0v = 0, and ux=yu_x = y, uy=xu_y = x; both vanish only at the origin. For ∣z∣|z|: u=x2+y2u = \sqrt{x^2 + y^2}, v=0v = 0, and ux=x/∣z∣u_x = x/|z|, uy=y/∣z∣u_y = y/|z|, which never both vanish (and uu is not differentiable at 00). In either case the equations hold at no point of an open set. A real-valued function satisfies them on an open disc only if it is constant there.

Exercise Everything is a function of zz and zˉ\bar z ●●○ Hardy: §227

Show that x=12(z+zˉ)x = \frac12(z + \bar z) and y=12i(z−zˉ)y = \frac{1}{2i}(z - \bar z), and use this to write xyxy and ∣z∣2|z|^2 in terms of zz and zˉ\bar z. Which of Hardy’s examples can be written using zz alone?

Hint
∣z∣2=zzˉ|z|^2 = z\bar z.

Solution

Adding and subtracting z=x+iyz = x + iy and zˉ=x−iy\bar z = x - iy gives the formulas. Then xy=14i(z+zˉ)(z−zˉ)=14i(z2−zˉ2)xy = \frac{1}{4i}(z + \bar z)(z - \bar z) = \frac{1}{4i}(z^2 - \bar z^2) and ∣z∣2=zzˉ|z|^2 = z\bar z. None of Hardy’s four examples (x−iyx - iy, xyxy, ∣z∣|z|, am⁡z\operatorname{am} z) can be written with zz alone by algebraic operations: if one could, it would act locally like a multiplication, and the explorer and the exercise above show that it does not. Roughly speaking, the functions that deserve the name “function of zz” are the ones in which zˉ\bar z does not occur.

Exercise An algebraical function of zz ●●○ Hardy: §228; cf. §47

The equation w2=zw^2 = z defines ww as a two-valued algebraical function of zz. Find both values when z=−4z = -4 and when z=2iz = 2i. Follow the value w=1w = 1 at z=1z = 1 continuously as zz goes once round the unit circle z=cos⁡θ+isin⁡θz = \cos\theta + i\sin\theta, 0≤θ≤2π0 \le \theta \le 2\pi: where does it end?

Hint
Try w=cos⁡12θ+isin⁡12θw = \cos\frac12\theta + i\sin\frac12\theta.

Solution

For z=−4z = -4: w=±2iw = \pm 2i. For z=2iz = 2i: w=±(1+i)w = \pm(1 + i), since (1+i)2=2i(1 + i)^2 = 2i.

w(θ)=cos⁡12θ+isin⁡12θw(\theta) = \cos\frac12\theta + i\sin\frac12\theta satisfies w2=zw^2 = z by De Moivre’s theorem, is continuous, and equals 11 at θ=0\theta = 0. At θ=2π\theta = 2\pi it equals cos⁡π+isin⁡π=−1\cos\pi + i\sin\pi = -1. Going once round the origin carries one value of z\sqrt z into the other. The logarithm, as we shall see, does something similar but never returns: each turn adds 2πi2\pi i.

Check yourself

Check yourself True or false?

In the wide sense of the word, xyxy is a function of z=x+iyz = x + iy.

Why

True: to each zz there corresponds one value of xyxy. That is exactly Hardy’s complaint. The wide definition admits everything, so it carries no information; this is why he restricts the phrase to functions defined directly from zz.

Check yourself Choose one

In §228 Hardy calls zz “the argument” of f(z)f(z). What does “argument” mean here?

Why

Hardy uses “argument” in its old sense of the input of a function. For the angle he uses amplitude, written am⁡z\operatorname{am} z. Modern books use “argument”, arg⁡z\arg z, for the angle, so read Hardy carefully.

Check yourself Choose one

Near z=1+iz = 1 + i, the map w=z2w = z^2 moves a short displacement hh to approximately

Why

(z+h)2−z2=2zh+h2(z + h)^2 - z^2 = 2zh + h^2, and for small hh the term h2h^2 is negligible compared with 2zh2zh. At z=1+iz = 1 + i the multiplier is 2z=2+2i2z = 2 + 2i, whose amplitude is 14π\frac14\pi and modulus 222\sqrt 2. The option (1+i)2+h(1+i)^2 + h forgets that the displacement is also rotated and stretched.

Check yourself True or false?

The map z↦x−iyz \mapsto x - iy rotates the plane through an angle π\pi.

Why

False. It is the reflection in the real axis: it fixes every real number, while a rotation through π\pi sends 11 to −1-1. A reflection preserves the shape of a square but reverses its orientation, and no multiplication by a complex number does that.

Hardy’s language

The definition, although quite legitimate, is futile because it does not really define a new idea at all.

— G. H. Hardy, §227

Summary

Summary
  • Any rule (x,y)↦f(x,y)+ig(x,y)(x, y) \mapsto f(x, y) + ig(x, y) is a “function of zz” in a wide sense that is too wide to be useful.
  • Hardy restricts the term to functions defined directly from zz: polynomials, rational and algebraical functions, and the log, exp and circular functions to come.
  • Such functions act near each point like multiplication by a complex number: a rotation and a stretch, so small squares stay square. x−iyx - iy, xyxy and ∣z∣|z| do not.
  • “Argument” means the input zz; the angle of zz is its amplitude am⁡z\operatorname{am} z, which is many-valued.