A function f(z) that is differentiable in the complex sense turns out to be far more rigid than any differentiable function of two real variables — rigid enough that its values on a boundary determine all its values inside.
UndergraduateFrom complex numbers to complex functions
"Complex numbers" treats C as a set of numbers to compute with. Here we ask a different question: what happens when a function f takes a complex number as input and produces a complex number as output, and we require its derivative to exist? The surprising answer is that this single requirement — complex differentiability — is far stronger than being differentiable as a map R2→R2, and it forces remarkable rigidity, some of which we prove in this topic.
Definition: Holomorphic function
Let U⊆C be open and f:U→C a function. At a point z0∈U, define f′(z0)=limh→0hf(z0+h)−f(z0), where h∈C may approach 0 from any direction. When this limit exists, f is holomorphic at z0; it is holomorphic on U if this holds at every point of U, and entire if U=C.
f′(z0)=h→0limhf(z0+h)−f(z0),h∈C
Definition: Cauchy–Riemann equations
Write f(x+iy)=u(x,y)+iv(x,y) for real functions u,v. Letting h approach 0 along the real axis gives f′(z0)=ux+ivx; letting h approach 0 along the imaginary axis gives f′(z0)=vy−iuy. If f is holomorphic at z0, these two expressions must be equal, so ux=vy and uy=−vx: the Cauchy–Riemann equations. Together with continuity of the partial derivatives, they are equivalent to f being holomorphic.
A grid of the complex plane and its image under z squared; the images of two perpendicular grid lines through a point still meet at a right angle, illustrating that a holomorphic map is conformal where its derivative is nonzero.
Grid lines under z↦z2: angles are preserved wherever f′(z)=0.
UndergraduateIntegrating along a path in C
For a path (contour) γ in C and a function f defined on it, the contour integral ∮γf(z)dz is defined the same way as a real integral: as a limit of sums ∑kf(zk)(zk+1−zk) over finer and finer partitions of γ. When γ is a closed curve and f is holomorphic, this integral turns out to be extraordinarily well-behaved.
If f is holomorphic on a simply connected open set U and γ is a closed contour in U, then ∮γf(z)dz=0.
Why is it true?
Writing f=u+iv and dz=dx+idy, the contour integral splits into two real line integrals, each of which can be rewritten as a double integral over the region enclosed by γ using Green's theorem. The two double integrals that appear are exactly ∬(vx−uy)dA and ∬(ux−vy)dA up to sign and a factor of i, and both vanish because of the Cauchy–Riemann equations ux=vy, uy=−vx.
Proof
Write f(z)=u(x,y)+iv(x,y) and dz=dx+idy along the closed contour γ=∂D enclosing the region D⊆U. Expanding the product separates the contour integral into real and imaginary line integrals: ∮γf(z)dz=∮γ(udx−vdy)+i∮γ(vdx+udy).
Since u and v have continuous partial derivatives on U, apply Green's theorem ∮∂D(Pdx+Qdy)=∬D(Qx−Py)dxdy to the real and imaginary parts separately to obtain ∮γf(z)dz=∬D(−vx−uy)dxdy+i∬D(ux−vy)dxdy.
Because f is holomorphic, the Cauchy–Riemann equations ux=vy and uy=−vx hold at every point of D. Thus −vx−uy=0 and ux−vy=0 identically on the region, which gives ∮γf(z)dz=0+i⋅0=0.
Example: A closed-contour integral of a holomorphic function
Compute ∮∣z∣=1ezdz.
Solution
ez is entire (holomorphic on all of C), so in particular it is holomorphic on the simply connected set C, and the circle ∣z∣=1 is a closed contour inside it. By Cauchy's integral theorem, ∮∣z∣=1ezdz=0 — no computation of an antiderivative is even needed.
If f is holomorphic on and inside a simple closed contour γ and a is a point inside γ, then f(a)=2πi1∮γz−af(z)dz.
Why is it true?
Apply Cauchy's integral theorem to g(z)=z−af(z) on the region between γ and a tiny circle of radius ε around a (where g is holomorphic, since z=a is excluded): the two contour integrals of g must agree. As ε→0, the integral over the tiny circle tends to f(a)⋅2πi, since f is continuous and nearly constant, equal to f(a), on a shrinking circle around a. This single formula shows that the values of a holomorphic function inside a region are completely determined by its values on the boundary.
Proof
Fix a strictly inside γ, and choose ε>0 small enough that the circle Cε={z:∣z−a∣=ε} lies entirely inside the contour. Since g(z)=z−af(z) is holomorphic on the region between the outer contour and the small circle, Cauchy's integral theorem gives ∮γz−af(z)dz=∮Cεz−af(z)dz.
Parameterize the small circle by z=a+εeiθ for θ∈[0,2π], so dz=iεeiθdθ and ∮Cεz−a1dz=∫02πidθ=2πi. Splitting the numerator as f(z)=f(a)+(f(z)−f(a)) yields ∮Cεz−af(z)dz=2πif(a)+∮Cεz−af(z)−f(a)dz.
Bounding the remainder integral by arc length gives ∮Cεz−af(z)−f(a)dz≤εmaxCε∣f(z)−f(a)∣⋅2πε=2πmaxCε∣f(z)−f(a)∣. Because the function is continuous at the center, as ε→0 we have maxCε∣f(z)−f(a)∣→0, so the remainder vanishes and f(a)=2πi1∮γz−af(z)dz.
The complex plane mapped by the exponential function; horizontal strips of height 2*pi map to the whole plane minus the origin, repeating periodically as the input moves vertically, unlike the strictly increasing real exponential.
The complex exponential z↦ez: entire, and periodic with imaginary period 2πi.
Cauchy's integral formula gives an estimate for the derivatives of a holomorphic function in terms of its size on a large circle: ∣f′(z0)∣≤RM, where M bounds ∣f∣ on a circle of radius R around z0. If f is entire and bounded by M everywhere, this estimate holds for every R, however large; letting R→∞ forces f′(z0)=0 at every point z0, so f is constant.
Proof
Suppose f:C→C is entire and bounded, so there exists a constant M≥0 such that ∣f(z)∣≤M for all z∈C. Fix an arbitrary point z0∈C and any radius R>0, and consider the circle CR={z:∣z−z0∣=R}.
By Cauchy's integral formula for the first derivative, f′(z0)=2πi1∮CR(z−z0)2f(z)dz. Since the circle has arc length 2πR and satisfies ∣z−z0∣2=R2 along the contour, the standard integral estimate yields ∣f′(z0)∣≤2π1⋅R2M⋅2πR=RM.
Because the function is entire, this bound holds for every radius; letting R→∞ forces f′(z0)=0 at every point, so the function is constant.
Liouville's theorem gives a strikingly short proof of the fundamental theorem of algebra from "Complex numbers": if a non-constant polynomial p had no root, then 1/p would be a bounded entire function, hence constant by Liouville — contradicting that p is non-constant. The worked solution below carries out this argument in full.
AdvancedIsolated singularities and residues
Near a point z0 where f fails to be holomorphic but is holomorphic on a punctured neighbourhood (an isolated singularity), f has a Laurent series f(z)=∑n=−∞∞cn(z−z0)n, allowing finitely or infinitely many negative powers. The singularity is removable if no negative powers appear, a pole of order k if the lowest power is −k, and essential if infinitely many negative powers appear. The coefficient c−1 has a name of its own: the residue of f at z0.
Classifying an isolated singularity of f at z0
Type
Laurent series
Example
Removable
no negative powers
zsinz at z0=0
Pole of order k
lowest power is (z−z0)−k
1/z2 at z0=0 (k=2)
Essential
infinitely many negative powers
e1/z at z0=0
The complex plane mapped by 1/z; points near the origin, where the function has a pole, map to points far from the origin, and the map turns circles around the origin inside-out.
If f is holomorphic on and inside a simple closed contour γ except for finitely many isolated singularities z1,…,zk inside γ, then ∮γf(z)dz=2πi∑j=1kResz=zjf(z).
Why is it true?
Deform γ into k tiny circles, one around each singularity zj, connected by thin corridors that cancel in pairs; by Cauchy's integral theorem applied to the (holomorphic) region between γ and these circles, the integral over γ equals the sum of the integrals over the small circles. Each small circle integral picks out exactly 2πi times the coefficient c−1 of the Laurent series at zj — the residue — because ∮(z−zj)ndz=0 for every power except n=−1, where it equals 2πi.
Proof
Surround the isolated singularities with small disjoint positively oriented circles C1,…,Ck centered at z1,…,zk of radii εj>0 inside γ. Because the function is holomorphic on the multiply connected region between the outer contour and the small circles, Cauchy's integral theorem gives ∮γf(z)dz=∑j=1k∮Cjf(z)dz.
On the punctured neighborhood of zj, expand the function into its Laurent series f(z)=∑n=−∞∞cj,n(z−zj)n, which converges uniformly on Cj. Integrating term by term around the small circle yields ∮Cjf(z)dz=∑n=−∞∞cj,n∮Cj(z−zj)ndz.
Using the parameterization z=zj+εjeiθ, direct calculation gives ∮Cj(z−zj)ndz=0 for every integer n=−1 and ∮Cj(z−zj)−1dz=2πi. Hence only the residue term survives on each circle, giving ∮Cjf(z)dz=2πicj,−1=2πiResz=zjf(z), and summing over all circles yields ∮γf(z)dz=2πi∑j=1kResz=zjf(z).
Example: Evaluating a contour integral with two poles via residues
Compute ∮∣z∣=2z2−1zdz, where the circle ∣z∣=2 is oriented counterclockwise.
Solution
Factor the denominator as z2−1=(z−1)(z+1). The integrand f(z)=(z−1)(z+1)z has two simple poles at z1=1 and z2=−1; since ∣1∣=1<2 and ∣−1∣=1<2, both poles lie strictly inside ∣z∣=2.
Compute the residue at each simple pole: Resz=1f(z)=limz→1(z−1)f(z)=limz→1z+1z=21, and Resz=−1f(z)=limz→−1(z+1)f(z)=limz→−1z−1z=−2−1=21.
Applying the residue theorem gives ∮∣z∣=2z2−1zdz=2πi(21+21)=2πi.
The complex plane mapped by sine; along the real axis the image stays within [-1, 1] as for the familiar real sine, but moving away from the real axis in the imaginary direction the image grows rapidly without bound.
The complex sine z↦sinz: entire, but unbounded — consistent with Liouville's theorem, since sinz is not constant.
ResearchRiemann surfaces and an open problem
Functions such as z or logz are naturally multi-valued: going once around the origin changes their value. Riemann's 1851 doctoral dissertation resolved this by gluing together several copies of the plane into a single geometric object, a Riemann surface, on which the function becomes single-valued and holomorphic. This geometric viewpoint on complex functions became one of the seeds of modern topology and algebraic geometry.
For f(x+iy)=u(x,y)+iv(x,y) to be holomorphic, u and v must satisfy the Cauchy–Riemann equations:
What is ∮∣z∣=1z2dz?
Liouville's theorem says that if an entire function f is bounded on all of C, then f must be
By the residue theorem, if f has finitely many isolated singularities z1,…,zk inside a simple closed contour γ, then ∮γf(z)dz equals