Patrick Lei

All notes

Complex analysis

Lectures by Paul Hacking, UMass, Spring 2018.

Complex Analysis
Math 621
Notes by Patrick Lei, June 2020
Lectures by Paul Hacking, Spring 2018

University of Massachusetts Amherst

Disclaimer

These notes are a transcription of handwritten notes that were taken during lecture. Any errors are mine and not the instructor’s. In addition, my notes are picture-free (but will include commutative diagrams) and are a mix of my mathematical style (omit lengthy computations, use category theory) and that of the instructor. If you find any errors, please contact me at plei@umass.edu.

1 Basic Notions

A complex number is a sum z=x+i⁢y, where x,y∈ℝ and i is a symbol satisfying the identity i2=−1. Addition and multiplication work as one would expect. The set ℂ of complex numbers is a field. This means that (ℂ,+) is an abelian group, (ℂ∖{0},⋅) is an abelian group, and that multiplication distributes over addition.

Remark 1.0.1.

If F is a field and f∈F⁡[x] is irreducible, then we can construct a field extension K/F such that K has a root of f by setting K=F⁡[x]/(f). In this way, we have ℂ=ℝ⁡[x]/(x2+1). The Galois group Gal⁡ℂ/ℝ is generated by complex conjugation.

1.1 Holomorphic Functions

Let Ω⊂ℂ be an open set. Here, the topology is the Euclidean topology on ℂ=ℝ2. Then f:Ω→ℂ is holomorphic at a point z0∈Ω if the limit

exists. If it does, we write f′⁢(z0) for the derivative at z0.

Example 1.1.1.

The function f⁡(z)=z¯ is not holomorphic. To see this, the difference quotient has different limits on the real and imaginary axes.

Example 1.1.2.

The function f⁡(z)=zn is holomorphic for n∈ℕ, and f′⁢(z)=n⁢zn−1.

Remark 1.1.3.

The usual formulas for differentiation (chain rule, product rule, linearity) hold in this case.

We will now compare holomorphic and real differentiability. Rewrite F=u+i⁢v. Recall that F is real differentiable at 𝐚=(x0,y0) if

for some linear map A. We say that A is the derivative of F at 𝐚. Moreover, A is given by the Jacobian matrix

Note that F is differentiable if and only if the partials exist and are continuous. Then recalling that holomorphic means that

we see that f is holomorphic if and only if f is differentiable and JZ0⁢F corresponds to multiplication by a complex number.

Remark 1.1.4.

The standard definition of holomorphic is that the differentiability condition is satisfied in an open neighborhood around z0.

Now recall that if λ=a+i⁢b∈ℂ, multiplication by λ is given by the real matrix (a−bba). Therefore, we must have ux=vy and uy=−vx. These are the Cauchy-Riemann equations.

Theorem 1.1.5 (Cauchy-Riemann).

The function f:U→ℂ is holomorphic if and only if it is real differentiable and its Jacobian satisfies the Cauchy-Riemann equations.

We can also write λ=r⁡(cos⁡θ+i⁢sin⁡θ)=r⁢ei⁢θ∈ℂ in polar form. We know that multiplication by λ is rotation by θ followed by dilation by r, so in particular it preserves angles. Suppose f:U→ℂ is holomorphic at z0 and γ1,γ2:(−1,1)→U are parameterized curve with γi⁢(0)=z0, then we define the angle between γ1,γ2 to be the angle between γ1′⁢(0) and γ2′⁢(0). We also obtain the curves f⁢γ1,f⁢γ2. If f′⁢(z0≠0), then the angle between f⁢γ1,f⁢γ2 equals the angle between γ1,γ2. To see this, the chain rule gives us

Remark 1.1.6.

Note that the condition that f′⁢(z0)≠0 is necessary. In fact, f is conformal at z0 if and only if f′⁢(z0)≠0. For example, consider f⁡(z)=z2, which satisfies f′⁢(0)=0.

1.2 Power Series

Theorem 1.2.1.

Let ∑n=0∞an⁢zn be a power series with an∈ℂ. Then there exists R∈R≥0∪∞ such that

  1. 1.

    The series converges absolutely for |z|<R;

  2. 2.

    The series diverges for |z|>r.

Moreover, we have 1R=lim supn→∞|an|1/n.

Recall that if bn is a sequence of real numbers, we define

This exists if bn is bounded below because the sequence of supremums is decreasing.

Proof.

Define R by the formula. Given ε>0, there exists N∈ℕ such that |an|1/n≤1R+ε for n>N. Then

so if |z|<11R+ε, then the series converges absolutely by the comparison test. Then ε is arbitrary, so the series converges absolutely for |z|<R.

If |z|>R, then fix ρ such that |z|>ρ>R. For all N∈|N|, there exists n>N such that |an|1/n>1ρ. Now

as n→∞, because |z|ρ>1. □

Theorem 1.2.2.

Let f⁡(z)=∑n=0∞an⁢zn∈ℂ⁡[[z]] have radius of convergence R and Ω={z∣|z|<R}. Then f:Ω→C is holomorphic and f′⁢(z)=∑n=1∞n⁢an⁢zn−1 has the same radius of convergence.

Here is the most important example of a power series.

Example 1.2.3.

Define ez−∑n=0∞znn!. This has radius of convergence R=∞ because

Alternatively, we may use the ratio test instead of the root test.

It is easy to see that ddz⁡ez=ez from the power series. Then we have

because if ∑an,∑bn are absolutely convergent and cn=∑k=0nak⁢bn−k, then ∑cn is absolutely convergent and ∑cn=∑an⋅∑bn (see baby Rudin for a reference).

Now we may define

It is easy to see from the power series that e±i⁢z=cos⁡z±i⁢sin⁡z. The formulas for the derivatives from basic calculus hold.

Example 1.2.4.

Consider the series ∑n=0∞zn. This has radius of convergence R=1 and diverges for |z|≥1. In fact, f⁡(z)=11−z for |z|<1. In this case, the power series f⁡(z) extends to a holomorphic g⁡(z)=11−z defined on ℂ∖{1}. More generally, if g:Ω→C is holomorphic and z0∈Ω is a point, then it has a power series expansion

valid in the largest open disc contained in Ω. This means that g is an analytic continuation of f.

Note that convergence for |z|=R is very delicate.

  1. 1.

    The series ∑zn has R=1 and diverges when |z|=1;

  2. 2.

    The series ∑1n2⁢zn has R=1 and converges absolutely for |z|=1;

  3. 3.

    The series ∑1n⁢zn has R=1 and diverges for z=1 but converges for |z|=1,z≠1.

1.3 Integration along curves

A paramaterized curve is a continuous function z:[a,b]→ℂ. We will call it smooth if z is continuously differentiable. We also assume that z′⁢(t)≠0 for all t∈[a,b].

Example 1.3.1.

The circle centered at z0 with radius r is given by

Two parameterized curves z1:[a,b]→ℂ,z2:[c,d]→ℂ are equivalent if there exists a C1 homeomorphism t:[c,d]→[a,b] with t′⁢(s)=0 for all s∈[c,d] and z2=z1∘t. Therefore, we can define a smooth curve to be an equivalence class of parameterized curves. It is closed if z⁡(a)=z⁡(b) and simple if z⁡(s)≠z⁡(t) for all s≠t unless s,t=a,b. Given a curve γ with parameterization z:[a,b]→ℂ, we write −γ for the curve

We may finally define integration along a curve. If f is continuous and γ is a smooth curve, we define

for some parameterization z:[a,b]→ℂ of γ. We need to check that this is well-defined, and we have

Example 1.3.2.

Let γ be a circle of radius r centered at the origin with parameterization z:[0,2⁢π]→ℂ given by z⁡(t)=r⁢ei⁢t. Then we have

It is easy to see that ∫−γf⁡(z)⁢dz=−∫γf⁡(z)⁢dz. We may also define the length

This has an easy bound given by

Theorem 1.3.3 (Fundamental Theorem of Calculus).

Let f:Ω→ℂ be continuous ans assume there exists a holomorphic F:Ω→ℂ such that F′=f (in other words, a primitive for f). Then

In particular, if γ is closed, then ∫γf⁡(z)⁢dz=0.

Proof.

By definition, we have

by the fundamental theorem of calculus over ℝ. □

2 Local Theory

2.1 Integrals of Holomorphic Functions

Theorem 2.1.1 (Cauchy).

Let f:Ω→ℂ be holomorphic and γ be a simple closed curve in Ω such that the interior of γ is contained in Ω. Then

Warning 2.1.2.

It is very tricky to precisely describe what is meant by “interior” and we need the Jordan curve theorem from algebraic topology to do so.

Sketch of Proof.

We write

by Green’s theorem and the Cauchy-Riemann equations. □

Of course, this assumes that the partial derivatives of u,v are C∞, which we do not know yet. Instead, we will give a more careful proof of a weaker theorem.

Theorem 2.1.3 (Goursat).

Let f:Ω→ℂ be holomorphic and T⊂Ω be a triangle. Then

Proof.

Bisect each side of T and create four smaller triangles Ti1. Then we have

and thus

for some i. Now write T1=Ti1. We may repeat this process to obtain

and thus we have

and length⁡(∂Tn)=2−n⁢length⁡(∂T). Now set z0=⋂n≥1Tn. We now use the holomorphicity of f to estimate the integral, and we have f⁡(z0+h)=f⁡(z0)+h⁢f′⁢(z0)+h⁢ψ⁢(h), where limh→0ψ⁡(h)=0. Therefore

Therefore, we have

as n→∞. □

Corollary 2.1.4.

The same result holds for a rectangle.

Now we want to discuss the existence of local primitives.

Proposition 2.1.5.

Let f:D→ℂ be holomorphic on an open disc D=Dr⁢(z0). Then there exists F:D→ℂ holomorphic such that F′=f. In particular, ∫γf⁡(z)⁢dz=0 for any closed curve γ⊂D.

Proof.

Define F as an integral

Here, we define γz as

z0γzz
Figure 2.1: Path of integration

Now we need to prove that F′⁢(z)=f⁡(z). Note that

Then note that the path of integration is given by

zz+h≡zz+h≡zz+h
Figure 2.2: Equivalent paths

Call the final path [z,z+h]. Therefore,

so

but then

as h→0. □

Now we will relate the function f itself to some integral of points neearby f.

Theorem 2.1.6 (Cauchy integral formula).

Let f:Ω→ℂ be holomorphic and D be an open disc such that D¯⊂Ω. Then

for all z∈D.

Proof.

First, if Ω′ is a disk with a segment of a radius removed, then any holomorphic f has a primitive on Ω′ by the same argument as before. Now define the path γε,δ by

z0z
Figure 2.3: Contour γε,δ

where the radius of the circle around z is δ and the width of the corridor is 2⁢ε. Now f⁡(z)z−2 is holomorphic on the interior of γε,δ, so there eixists a primitive. Therefore, ∫γε,δf⁡(w)w−z⁢dz=0. Now we let ε→0. Because f⁡(w)w−z is continuous on the rectangular corridor, we have

as ε→0. This is because if f is continuous on some rectangle, then

as s→s0 by uniform continuity on compact sets. Now we see that

where αδ is the circle of radius δ centered at z. Then we can rewrite

and the last term approaches f′⁢(z) as w→z. Therefore it is bounded, so

as δ→0 because length⁡(αδ)→0 as δ→0. Therefore, we have

Corollary 2.1.7 (Higher derivatives).

Let f:Ω→ℂ be holomorphic and D¯⊂Ω be a disk. Then

This formula is obtained by differentiating under the integral sign with respect to z, and this is Ok because in general

as long as φ,∂φ∂x are continuous. Again, a reference for this is baby Rudin. Alternatively, we can use the following theorem.

Theorem 2.1.8.

Let f:Ω→ℂ be holomorphic and D¯⊂Ω be a disk. Then

for all z∈D.

Proof.

Fix z∈D. By the Cauchy integral formula, we have f⁡(z)=∫∂Df⁡(z)w−z⁢dw. Expanding out

we obtain

Note that f⁡(w) is bounded on ∂D and the sum ∑n=0∞(w−z)n+1(z−z0)n converges uniformly on ∂D, so we can interchange the sum and the integral. □

Corollary 2.1.9 (Cauchy inequality).

Let z0 be the center of our disk with radius R. Then we see that

Theorem 2.1.10 (Liouville).

Let f:ℂ→ℂ be holomorphic. If f is bounded, then f is constant.

Proof.

Write f⁡(z)=∑n=0∞an⁢zn, and this is valid for all z∈ℂ. By the Cauchy inequality, we have

as R→∞. Therefore, an=0 for n≠0. □

Theorem 2.1.11 (Fundamental Theorem of Algebra).

Let f∈ℂ⁡[z] be a nonconstant polynomial. Then there exists α∈ℂ such that f⁡(α)=0.

Proof.

Suppose no root exists, so we may consider the entire function g⁡(z)=1f⁡(z). Then there exists R>0 such that |f⁡(z)|>12⁢|an⁢zn| for |z|≥R. Then

On the other hand, g is bounded for |z|≤R, so by Liouville’s theorem g is constant. Therefore f is constant. □

In particular, if f has degree n, then it has n roots counting multiplicity.

2.2 Analytic Continuation

Theorem 2.2.1.

Let f:Ω→ℂ be holomorphic with Ω open and connected. Suppose there exists a sequence zn of distinct points of Ω such that zn→α∈Ω as n→∞ and f⁡(zn)=0. Then f=0.

Proof.

Expand f⁡(z)=∑n=0∞an⁢(z−α)n=∑n=m∞an⁢(z−α)n for some m. In particular, we have f⁡(z)=(z−α)m⁢g⁢(z), where g⁡(z) is holomorphic on D⊂Ω centered at α. Thus f⁡(z)≠0 for z close to α. But then g⁡(z)≠0 for |z−α|<ε and g⁡(α)=am≠0. But this gives a contradiction, so we must have an=0 for all n. Now we show that f=0 on all of Ω. If Ω1 is the interior of the set {z∈Ω∣f⁡(z)=0}, then we know Ω1 is closed by the argument above, so by connectedness of Ω, we see that Ω=Ω1. □

Given f:Ω→ℂ holomorphic and Ω⊂Ω~⊂ℂ, an analytic continuation of f to Ω~ is a holomorphic function f~:Ω~→ℂ such that f~|Ω=f. By the theorem, the analytic continuation is unique.

Example 2.2.2.

The Riemann zeta function ζ⁡(s)=∑n=1∞1ns admits an analytic continuation to ℂ∖{1} from {s∣Re(s)>1}.

Theorem 2.2.3.

Let f:Ω→ℂ be holomorphic with Ω connected. Assume that f⁢n⁢e⁢q⁢0. Given z0∈Ω, then there exists an open neighborhood U of z0, g:U→ℂ holomorphic, and n∈ℤ≥0 such that f⁡(z)=(z−z0)n⁢g⁢(z), where g⁡(z)≠0 for all z∈U.

Proof.

Expand f⁡(z)=∑k=0∞ak⁢(z−z0)k=an⁢(z−z0)n+O⁡(zn+1)=(z−z0)n⁢(an+O⁡(z)). □

We say that f has a zero of order n at z=z0.

2.3 Poles

Let f:Ω→ℂ be holomorphic and D×={z∣0<|z−z0|<r}⊂Ω. Then we say that f has an isolated singularity at z=z0. We say that f has a pole at z0 if

is holomorphic in a neighborhood of z0. In particular, if z=z0 is a pole, then 1f⁡(z)→0 as z→z0.

Better, write g⁡(z)=(z−z0)n⁢h⁢(z) for some holomorphic h with h⁡(z)≠0 for z near z0. Then we can write f⁡(z)=1g⁡(z)=(z−z0)−n⁢k⁢(z) for some holomorphic k. Then we say that f has a pole of order n.

Example 2.3.1.

Consider f⁡(z)=g⁡(z)h⁡(z)=(z−z0)n⁢k⁢(z)(z−z0)m⁢ℓ⁢(z)=(z−z0)n+m⁢k⁡(z)ℓ⁡(z). Then if m>n we have a pole of order m−n and if m≤n we have a removable singularity and can extend to a holomorphic function with a zero of order n−m.

Now we say that f is meromorphic on Ω if there exists a discrete set S⊂Ω such that f:Ω∖S→ℂ is holomorphic and f has a pole at each s∈S. If f has a pole of order n at z0, then f⁡(z)=(z−z0)−n⁢g⁢(z) near z0. Expanding g as a power series, we have

where h⁡(z) is holomorphic. We call a−1≕Resz0⁡f the residue of f at z0.

Theorem 2.3.2.

Let f:Ω→ℂ be holomorphic with a pole at z0. Let C be a small circle centered at z0. Then

Proof.

We have

By Cauchy, the integral of h⁡(z) vanishes, and each a−k⁢(z−z0)−k has a primitive for k≠1, so by the computation of the integral of 1z from before, we obtain the desired result. □

Theorem 2.3.3 (Residue Theorem).

Let f:Ω→ℂ be holomorphic and assume γ is a simple closed curve in Ω and the interior of γ is contained in Ω∪{z1,…,zN}, where the zj are the poles of f. then

Proof.

Recall the contour in Figure 2.3, except now with more than one keyhole. Then by Cauchy’s theorem, we know

so letting ε→0, we have

Now if cj is a small circle around zj, we have

Remark 2.3.4.

We gave a sketch of a proof of Cauchy relying on Green’s theorem. This required the partial derivatives to be continuous, but this is now fine because we proved that f is infinitely differentiable. However, we still have the issue of the interior of a simple closed curve.

Example 2.3.5.

Consider the integral

If we set z=ei⁢θ, then dz=i⁢ei⁢θ, so we obtain

Now we would actually like to be able to compute residues. This is important if we want to actually be able to compute integrals.

Example 2.3.6.

Consider f⁡(z)=sin⁡zz6. Then there is a pole at z=0, and the residue is 15!=1120 by inspeection of the power series.

Example 2.3.7.

Consider f⁡(z)=tan⁡z=sin⁡zcos⁡z. We would like to compute the residue at z0=π2. If f⁡(z) has a simple pole (order 1) at z=z0, then

In particular, if f=gh and h has a simple zero at z0, then f has a simple pole at z0. This gives us

so Resπ/2⁡tan⁡z=sin⁡π2−sin⁡π2=−1.

Similarly, if f has a pole of order k at z0, then

Example 2.3.8.

Consider f⁡(z)=1ez−(1+z), which has a pole of order 2 at z=0. In particular, we have

so

Alternatively, we could formally invert the power series 12+z6+⋯. Returning to our integral over a real variable θ, we see the poles of z2+2⁢a⁢z+1=0 are at z=−a±a2−1, so our integral is

Of course, we may consider a general form ∫02⁢πF⁡(cos⁡θ,sin⁡θ)⁢dθ.

Consider the integral ∫−∞∞F⁡(x)⁢dx, where F∈ℝ⁡(x) has the form F=P/Q with deg⁡P+2≤deg⁡Q and no poles in ℝ. Now if we consider the contour

0−RR
Figure 2.4: The contour γR

then we see that

and the second term approaches 0 as R→∞, so

To see that the integral of the arc disappears, observe that

But then if deg⁡P=n,deg⁡Q=m, zm−n⁢F⁢(z)→anbm as |z|→∞, so we can bound this quantity by some constant C, and thus

as R→∞ because n+1−m<0.

Example 2.3.9.

Consider the integral

Then we have simple poles at z=±i,±3⁢i, so we have

Now consider F∈ℝ⁡(x) with deg⁡P+1≤deg⁡Q and Q⁡(x)≠0 for x∈ℝ. We want to compute the integral

We will consider the contour

0−x2x1y
Figure 2.5: The contour of integration

Now we note that

Then we know |F⁡(z)|≤C⁢|z|deg⁡P−deg⁡Q, so

as x1→∞. By a similar argument, ∫down vanishes in the limit. Finally, we have

and this clearly vanishes in the limit. This gives us

This is because

As x1,x2,y→∞, this vanishes.

Example 2.3.10.

Consider the integral I=∫−∞∞x⁢sin⁡xx2+a2⁢dx for a∈ℝ>0. We see that sin⁡x=Im(ei⁢x), so we must have

Finally, let α∈[0,1) and F=P/Q with deg⁡P+2≤deg⁡Q and consider the integral

Suppose that Q⁡(x)≠0 for x∈ℝ>0 and at worst has a simple zero at x=0. Then we will consider the contour γr,R,ε, where r is the inner radius, R the outer radius, and ε the width of the corridor.

Figure 2.6: Contour γr,R,ε

Now as ε→0, we have

so we obtain the equation

But now we have

as R→∞ because |zα⁢F|≤C⋅Rα⁢Rdeg⁡P−deg⁡Q≤C⋅Rα−2. We also have

as r→0 because |zα⁢F|≤C⁢rα−1. Therefore, we have

Example 2.3.11.

Consider the integral I=∫0∞x1/21+x2⁢dx. Then we have

Therefore, I=π2.

2.4 Singularities

We will now consider the next type of singularity. These are the removable singularities.

Theorem 2.4.1 (Riemann, removable singularity theorem).

Let f:Ω∖{z0}→ℂ be holomorphic and assume f is bounded on D∖{z0} for some disk D centered at z0. Then f extends to a holomorphic function on Ω and z0 is a removable singularity.

Proof.

Consider the same keyhole as in Figure 2.3 except now with keyholes at z0 and z.

z0z
Figure 2.7: Contour γ

Then we have

by Cauchy’s theorem. Therefore, ∫∂D−∫C1−∫C2=0, so we see that

where C1 is a circle centered at z0 and C2 is a circle centered at z. Then as r→0, we see that ∫C1→0. But now we can set f⁡(z)≔12⁢π⁢i⁢∫∂Df⁡(w)w−z⁢dw, so f is holomorphic at z. □

Corollary 2.4.2.

Let f:Ω∖{z0}→ℂ be holomorphic. Then f has a pole at z0 if and only if f⁡(z)→∞ as z→z0.

Proof.

We know that f has a pole if and only if

is holomorphic near z0. Therefore f has a pole if and only if 1f→0 as z→z0, which is equivalent to f⁡(z)→∞. □

Now we will consider the final type of singularity: essential singularities. This is every singularity that is not a pole or a removable singularity. The canonical example is f⁡(z)=e1/z:ℂ∖{0}→ℂ.

Theorem 2.4.3 (Casorati-Weierstrass).

Let f:Ω∖{z0}→ℂ be holomorphic and z0 be an essential singularity. Then f⁡(Ω∖{z0}) is dense in ℂ. Equivalently, for all α∈ℂ, there exists zn∈Ω∖{z0} such that zn→z0 and f⁡(zn)→α.

Proof.

Assume there exists α∈ℂ such that |f⁡(z)−α|>δ for all z∈Ω and some δ>0. Then if we write g⁡(z)=1f⁡(z)−α, we see that |g⁡(z)|<δ−1, so g is holomorphic on Ω. Therefore, we have f⁡(z)=1g⁡(z)+α. If g⁡(z0)≠0, then f has a removable singularity, and if g⁡(z0)=0, then f has a pole. □

Finally, we will consider Laurent series expansions.

Proposition 2.4.4.

Let f:Ω→ℂ be holomorphic and A={z∣r1<|z−z0|<r2} and assume |A|⊂Ω. Then f⁡(z)=∑n=−∞∞an⁢(z−z0)n for all z∈A.

Proof.

Consider the keyhole

z
Figure 2.8: Contour γε

Now note that f⁡(z)=12⁢π⁢i⁢∫γεf⁡(w)w−z⁢dw, so as ε→0, we see that ∫γε→∫Cr2−∫Cr1. Now we obtain

Similarly, we have

Combining, we obtain the desired result. □

In the special case where r1=0, then if f:Ω∖{z0}→ℂ is holomorphic with D⊂Ω, then we can write

on D∖{z0}. Therefore the types of singularities correspond to

Removable:

For all n<0, an=0.

Pole:

There exists k>0 such that an=0 for all n<−k.

Essential:

For all N>0, there exists n>N such that a−n≠0.

Now we see that e1/z=∑n=0∞z−nn! has an essential singularity at 0.

Remark 2.4.5.

The Laurent series expansion is unique.

Warning 2.4.6.

We cannot write an=f(n)⁢(z0)n!.

2.5 Meromorphic Functions

First, we will define the Riemann sphere ℙ1=ℂ⁢ℙ1=ℂ∪{∞}. Here, we consider stereographic projection S2∖{(0,0,1)}→ℂ. This is given by the formula φ⁡(x,y,z)=11−z⁢(x+i⁢y). This gives us a homeomorphism S2→ℂ⁢ℙ1, and the topology on ℂ⁢ℙ1 is induced by this identification. This is a special case of the one-point compactification. In fact, we will see that ℂ⁢ℙ1 has more structure: that of a Riemann surface, or a 1-dimensional complex manifold.

Definition 2.5.1.

A Riemann surface X is a topological space X with an atlas of charts X=⋃αUα with homeomorphisms φα:Uα→Vα⊆ℂ such that the transition maps

are holomorphic. We also require that X is connected, Hausdorff, and second-countable.

Example 2.5.2.

On ℂ⁢ℙ1 we may consider the two charts S2=U1∪U2 with the two charts given by φ1 being stereographic projection and φ2 being stereographic projection from the south pole followed by conjugation. We can check that the transition map

so φ2∘φ−1⁢(w)=1w. Therefore ℙ1 can be given the structure of a compact Riemann surface with two charts U1=ℂ→idℂ and U2=ℂ∪∞∖0→z↦1/zℂ.

Example 2.5.3.

The second example of a Riemann surface is an elliptic curve. Choose λ1,λ2∈ℂ linearly independent over ℝ. Then if we write Λ=ℤ⁢λ1+ℤ⁢λ2, we can set X=ℂ/Λ. This has the quotient topology from ℂ. To define charts, let Uz0 be a small disk in ℂ centered at z0. Then under the map q:ℂ→X, we have a homeomorphism D→q⁡(D) and transition functions are given by z↦z+λ for some λ∈Λ.

Remark 2.5.4.

It is also possible to define Riemann surfaces without using topology. We simply define the Uα as sets with bijections to open subsets of ℂ and then specify holomorphic transition maps. Then the topology is given by the topology from ℂ, where each Uα is open. This allows for non-Hausdorff examples, like two copies of ℂ glued by the identity map on ℂ∖{0}, called the affine line with two origins.1

Remark 2.5.5.

If X,Y are Riemann surfaces, we can define holomorphic maps F:X→Y. If we choose charts p∈U→𝜑ℂ and F⁡(p)∈V→𝜓ℂ, then we require that ψ⁢F⁢φ−1 is holomophic and that F is continuous.

Now the definition of a pole is equivalent to the map f~:Ω→ℙ1 given by f~⁢(z)=f⁡(z) when z≠z0 and f⁡(z0)=∞ being holomorphic. Similarly, being meromorphic is equivalent to being a holomorphic map to ℂ⁢ℙ∞.

Definition 2.5.6.

Let f:Ω→ℂ be holomorphic and suppose there exists R>0 such that {z∣|z|>R}⊂Ω. Then we can define g:{x∣0<|z|<1R}→ℂ by g⁡(z)=f⁡(1/z). Then we say that f has a (removable singularity, pole of order k, essential singularity) at ∞ if g has the same thing at z=0.

Example 2.5.7.

If f=PQ is a rational function and deg⁡P=n,deg⁡Q=m, then f has a removable singularity at ∞ if m≥n and a pole of order n−m if m<n. In particular, f is meromorphic on ℂ∪∞.

Example 2.5.8.

We know that f⁡(z)=ez has an essential singularity at ∞ because g⁡(z)=ez−1 has an essential singularity at 0.

Theorem 2.5.9.

The meromorphic functions on ℙ1 are the rational functions.

Proof.

Let f:ℙ1→ℙ1 be meromorphic. Note that the number of poles is finite because ℙ1=S2 is compact. Then let the poles be z1,…,zn and possibly ∞. Then at zi, we expand f as a Laurent series and write

where gi is holomorphic near zi. By construction, g=f−∑i=1nfi is holomorphic on ℂ. Now consider h⁡(z)=g⁡(z−1) near z=0. We have

so

Therefore If we write the first term as a polynomial p⁡(z), g∞⁢(z−1) is bounded as z→∞, so by Liouville it is constant. Finally, we have f=c+p⁡(z)+∑fi⁢(z), which is rational. □

Remark 2.5.10.

Let f=pq be rational and suppose p,q are coprime and nonconstant. Then the number of zeros of f is equial to the number of poles of f, and both are given by max⁡(deg⁡p,deg⁡q). This follows from the fundamental theorem of algebra and the calculuation of the behavior of f at ∞.

More generally, if f:X→Y is a nonconstant holomorphic map between compact Riemann surfaces, we can define a well-defined degree of f by |f−1⁢(q)|=deg⁡f counting multiplicity. We can now consider the automorphisms of ℙ1. Clearly these are rational maps, and the maps given by

have inverse w↦d⁢z−b−c⁢w+a. In fact, these are actually all automorphisms because any automorphism f=pq must satisfy deg⁡f=max⁡(deg⁡p,deg⁡q)=1. Also, for all but finitely many c∈ℂ, all solutions of f⁡(z)=c have multiplicity 1. Therefore, we have Aut⁡(ℙ1)=P⁢G⁢L2⁢(ℂ) because we can identify ℙ1=ℂ2∖{0}/ℂ× and then the point (z1:z2) corresponds to z1z2. Then it is easy to see that these fractional linear transformations correspond to matrix multiplication.

With the prior discussion, we may now use the residue theorem at ∞.

Example 2.5.11 (Fall 2016 Qualifying Exam, Q9).

Consider C={z∈ℂ∣|z|=2} oriented counterclockwise and let n∈ℤ satisfy n≥2. We want to compute

Note that the residue theorem applies to essential singularities, so we may set w=1z,dz=−1w2⁢dw. Therefore we have

2.6 Estimates for Holomorphic Functions

Theorem 2.6.1 (Argument principle).

Let f be meromorphic on Ω⊂ℂ and γ be a simple closed curve such that γ and its interior are contained in Ω. Suppose f is holomorphic and nonzero on γ, oriented positively. Then

Proof.

Use the residue theorem. We have

But then if z0 is a zero or pole of order n, we have

□

Remarks 2.6.2.

The left hand-side of the formula is the winding number of f∘γ around z=0. In addition, we can think of f′f⁢dz as d(log⁡f), but log⁡z is not well-defined.

Theorem 2.6.3 (Rouché).

Let f,g:Ω→ℂ be holomorphic and γ be a simple closed curve with γ+int⁡(γ)⊂Ω. Assume that |f⁡(z)|>|g⁡(z)| for all z∈γ. Then the number of zeroes of f inside γ is equal to the number of zeroes of f+g inside γ, counting multiplicity.

Proof.

Define for t∈[0,1] the function ft=f+t⁢g:Ω→ℂ. This is holomorphic on Ω and nonzero on γ. Then the function

is continuous because the integrand depends continuously on t, but ℤ is discrete, so it must be constant. □

Example 2.6.4.

Consider f=z5, g=3⁢z+1. Then if |z|=2, |f⁡(z)|=32 while |g⁡(z)|≤3⋅2+1=7, so f+g has 5 zeroes with |z|<2. On the other hand, if |z|=1, then |g|≥2 while |f|=1, so f+g has a single zero with |z|<1.

Theorem 2.6.5 (Open mapping theorem).

Let f:Ω→ℂ be holomorphic and nonconstant on a connected Ω. If U⊂Ω is open, then f⁡(U) is open.

Proof.

Let z0∈Ω and write w0=f⁡(z0). We need to show that for some ε>0, the set

Equivalently, we want to show that f⁡(z)−w=0 has a solution z∈Ω for |w−w0|<ε. Choose δ>0 such that D¯={z∣|z−z0|≤δ}⊂Ω. For δ small enough, f⁡(z)≠w0 for |z−z0|=δ. Then we know |f⁡(z)−w0|>ε on ∂D for some ε>0, so for |w−w0|<ε, Rouche’s theorem tells us that the number of zeroes of (f⁡(z)−w0)+(w0−w) is equal to the number of zeroes of f⁡(z)−w0 in D, and this is positive. □

Theorem 2.6.6 (Maximum Principle).

Let f:Ω→ℂ be holomorphic on some open set Ω⊂ℂ. Then |f| does not attan a maximum on Ω.

Proof.

Given z0∈Ω, we know f⁡(z0)∈{w∣|w−f⁡(z0)|<ε}⊂f⁡(Ω)⊂ℂ. Clearly there exists w∈D such that |w|>|f⁡(z0)|. □

Here is another formulation. Suppose f:Ω→ℂ is holomorphic and f extends continuously to Ω¯, which is bounded. Then the maximum of f on Ω¯ is obtained on the boundary, so we cannot obtain the maximum on Ω.

Remark 2.6.7.

Suppose f:Ω→ℂ is holomorphic and nonconstant on a connected Ω. Then the real part of f does not attain a maximum on Ω. Here, we simply apply the maximum principle to g=ef.

Now recall that if f=u+i⁢v and f is holomorphic, then u,v are harmonic. This means that ∂2u∂x2+∂2u∂y2=0. Conversely, if u is a harmonic funciton on a disk D⊂ℂ=ℝ2, there exists a harmonic v such that f=u+i⁢v is holomorphic on D. Therefore we obtain a maximum principle for harmonic functions. The physical intuition for this is that if u:Ω¯→ℝ is harmonic and we fix u|∂Ω, then this corresponds to the steady state temperature of points in the region Ω. Then heat would flow away from a maximum, so the maximum cannot be attained on the interior.

Now recall from multivariable calculus that if f:ℝ2→ℝ, we have critical points of f when

Then we may consider the matrix

This is called the Hessian of f. If we assume that p=(0,0), then we can write

If det⁡H≠0, then we have a nondegenerate quadratic form. After a linear change of coordinates, theere are three cases:

  1. 1.

    The form looks like x2+y2 and det⁡H>0. This is a minimum.

  2. 2.

    The form looks like −(x2+y2) and det⁡G>0. This is a maximum.

  3. 3.

    The form looks like x2−y2 and det⁡H<0. This is a saddle point.

For any harmonic function, any nondegenerate critical point is a saddle point because

Example 2.6.8.

Consider f⁡(z)=z2. Then u=x2−y2=(x+y)⁢(x−y) and v=2⁢x⁢y. Clearly the origin is a saddle point for both. The graph of u looks like

−4−2024−505−20020
Figure 2.9: Graph of z=x2−y2
Example 2.6.9.

Consider f⁡(z)=z3. Then u=x3−3⁢x⁢y2=x⁢(x−3⁢y)⁢(x+3⁢y) is not nondegenerate at the origin. The graph looks like

−4−2024−505−2000200
Figure 2.10: Graph of z=x3−3⁢x⁢y2

Now we can give another proof of the maximum principle using the mean-value property. Recall that

Now we parameterize γ by w=z0+r⁢ei⁢θ, so we obtain

and therefore f⁡(z0) is the mean value of f on the circle γ. Therefore, if f is nonconstant, it obtains some larger value on the circle γ.

2.7 The Logarithm

Let Ω⊂ℂ and γ0,γ1 be two curves in Ω with the same endpoints α,β and parameterizations z0,z1:[a,b]→Ω. Then γ0,γ1 are homotopic in Ω if there exists a continuous

Theorem 2.7.1.

Let f be holomorphic on Ω and γ0,γ1 curves in Ω. If γ0,γ1 are homotopic, then

Proof.

Choose a homotopy F. Then the image K of F is compact and K⊂Ω⊂ℂ, so there exists ε>0 such that for all w∈K the disk D⁡(z,ε)⊂Ω. We need to show that the distance d⁡(p,q),p∈K,q∈ℂ∖Ω is bounded below by ε>0. Otherwise, there exists pn∈K and qn∈ℂ such that d⁡(pn,qn)→0. But then K is compact, so pn→p∈K, so qn→p as well. But then ℂ∖Ω is closed, and thus p∈ℂ∖Ω, which is a contradiction.

F is uniformly continuous, so given ε>0, there exists δ>0 such that |F⁡(s1,t1)−F⁡(s2,t2)|<ε whenever |(s1,t1)−(s2,t2)|<δ. If we fix t1,t2 such that |t1−t2|<12⁢δ, then we can choose a subdivision a=s0<s1<⋯<sn=b, where si+1−si<δ2. Then we see that |zt⁢(s)−zt′⁢(s′)|<ε for all (s,t),(s′,t′)∈[si,si+1]×[t1,t2]. But now we choose D to be a disk of radius ε such that F⁡([si,si+1]×[t1,t2])⊂D. By the Cauchy formula for the disk, we see that

Summing over i, we see that ∫γt1−∫γt2=0. Subdividing in the t direction, we get ∫γ0=∫γ1. □

Now let Ω⊂ℂ be open. We sau that Ω is simply connected if it is path connected and any two paths with the same endpoints are homotopic. Equivalently, Ω is path-connected and any loop γ is homotopic to the constant loop.

Example 2.7.2.

Any convex set is homotopic by the straight line homotopy

Example 2.7.3.

The set ℂ∖(−∞,0] is simply connected. Note that this set is homeomorphic to ℝ>0×(−π,π) because any z∈Ω can be uniquely written as z=r⁢ei⁢θ for any r>0 and θ∈(−π,π). Therefore, Ω is homeomorphic to a convex set and is thus simply connected.

Theorem 2.7.4.

Let Ω⊂ℂ be open and connected. Then Ω is simply connected if and only if ℙ1∖Ω is connected.

Example 2.7.5.

Consider ℝ×(0,1). Then the complement in ℙ1 is connected because ∞ joins the two components of ℂ∖(ℝ×(0,1)).

Theorem 2.7.6.

Let Ω⊂ℂ be simply connected and f:Ω→ℂ be holomorphic. Then f has a primitive.

Proof.

Fix a basepoint z0∈Ω. Then set F⁡(z)≔∫γzf⁡(w)⁢dw, where γz is a path from z0 to z in Ω. Because Ω is simply-connected, this is well-defined. Now as in the case of the disk, we see that F′⁢(z)=f⁡(z). □

Example 2.7.7.

Consider Ω=ℂ∖{0} and f⁡(z)=1z. Then we see that

and therefore 1z does not have a primitive, so Ω is not simply-connected.

Now we may define the complex logarithm.

Theorem 2.7.8.

Let Ω∈ℂ be simply connected with 1∈Ω and 0∉Ω. Then there exists a unique holomorphic function F=logΩ:Ω→ℂ such that eF⁡(z)=z for all z∈Ω and F⁡(1)=0.

Proof.

Let F be a primitive of 1z on Ω normalized such that F⁡(1)=0. Then we show that eF⁡(z)=z. Here, we have

so z⁢e−F⁡(z)=c for some constant c∈ℂ. Setting z=1, we obtain c=1, so eF⁡(z)=z.

Now we consider uniqueness of the logarithm. Let G be another such function. Then G−F:Ω→2⁢π⁢i⁢ℤ, but Ω is connected, so G−F is constant and thus evaluating at z=1 gives us G=F. □

Example 2.7.9.

Consider Ω=ℂ∖(−∞,0]. Then we defined earlier that for z=r⁢ei⁢θ, log⁡z=log⁡r+i⁢θ.

More generally, we have the following result.

Theorem 2.7.10.

Let Ω⊂ℂ be simply connected and f:Ω→ℂ be holomorphic and nowhere vanishing. Then there exists g:Ω→ℂ such that eg=f.

Proof.

Let g be a primitive of f′f. Then we have

so we can simply adjust g by a constant until f⁢e−g=1. □

Now we may define fα=eα⁢log⁡f for any α∈ℂ.

3 Global Theory

3.1 Conformal Mappings

Question 3.1.1.

Let Ω1,Ω2⊂ℂ be open. Does there exist a holomorphic bijection Ω1→Ω2?

Remark 3.1.2.

A holomorphic bijection has holomorphic inverse.

Our goal will be to prove the following remarkable theorem.

Theorem 3.1.3 (Riemann Mapping Theorem).

Let Ω⊂ℂ be open. Then there exists a holomorphic bijection F:Ω→D if and only if Ω is simply connected and Ω≠ℂ. Here, D is the unit disk.

Example 3.1.4.

Let Ω=ℋ={z=x+i⁢y∣y>0}⊂ℂ. Then there exists a holomorphic bijection F:ℋ→D. Note that if z∈ℋ, then |z−i|<|z+i|. Therefore, we can set F⁡(z)=i−zi+z. To see that this is a bijection, we note that ∞↦−1, −i↦∞. Therefore, F:ℂ∖{−i}→ℂ∖{−1} is a bijection, and thus it restricts to a bijection ℋ→D on the northern hemisphere.

Remark 3.1.5.

Consider the map φ:S2→ℙ1. Then

Then if r is the rotation by π2 about the x-axis, the diagram

commutes, where G~=F−1∘ψ, where ψ⁡(w)=i⁢w.

Here, rotation of S2 corresponds to a holomorphic automorphism of ℙ1 under φ. This is because stereographic projection preserves angles and rotations also preserve angles. Therefore G~=φ∘r∘φ−1 will preserve angles, so it is holomorphic. This gives us a homomorphism S⁢O⁢(3)↪Aut⁡(ℙ1)=P⁢G⁢L2⁢(ℂ). Recall that

The image of the homomorphism is P⁢S⁢U⁢(2), where we have

In fact, any B∈S⁢U⁢(2) has the form B=(ab−b¯a¯), so S⁢U⁢(2)≃S3. Then we have PSU(2)=SU(2)/±I. Now we have a chain of isomorphisms

and this is somehow related to the spin of an electron.

Now here are some examples of conformal maps:

  1. 1.

    If f:ℂ→ℂ is a holomorphic bijection, then there exists x,y such that f⁡(z)=a⁢z+b.

  2. 2.

    Let n∈ℕ and set S={z∈ℂ∣0<arg⁡(z)<πn}. Then z↦zα gives a holomorphic bijection S→ℋ. If α>12 is real, then we may consider z↦zα.

  3. 3.

    Consider log⁡z=log⁡r+i⁢θ. This gives a holomorphic bijection ℂ∖(−∞,0]→ℝ×i⁡(−π,π). Restricting to ℋ, we have a holomorphic bijection ℋ→ℝ×i⁡(0,π).

Example 3.1.6.

Consider the map ℂ∖{0}→𝐹ℂ given by z↦z+1z. Then if z+1z=w, we have z2−w⁢z+1=0 and thus z=w±w2−42. Therefore F is 2-to-1 onto ℂ and branched over w2−4=0, or w=±2. Then let z1,z2 be two roots of z2−w⁢z+1=0 for a fixed w. We know z1⁢z2=1, so exactly one of z1,z2 lies in ℋ unless z1,z2∈ℝ, which happens if and only if w∈ℝ and w2−4≥0. Therefore F restricts to a bijection

We want to compute the preimage of ℋ. Then we have

For y>0, we see that 1−1x2+y2>0 if x2+y2>1. Therefore we have Ω=ℋ∖{|z|≤1}. By the reasoning, the map

is a holomorphic bijection.

Example 3.1.7.

Set Ω={z∣ℜ⁡(z)>0}∖[0,1]. Then the sequence

gives a holomorphic bijection Ω→ℋ.

Example 3.1.8.

We have the following sequence taking a vertical corridor to ℋ.

→π⁢i⁢z→z↦ez→−z−z−1ℋ
Figure 3.1: A sequence of holomorphic bijections

Now we will study fractional linear transformaitons in more detail. Let f:ℙ1→ℙ1 be given by z↦a⁢z+bc⁢z+d with a⁢d−b⁢c≠0. Then we know that f is a holomorphic bijection and that Aut⁡(ℙ1)=P⁢G⁢L2⁢(ℂ). Another important property is the following: given z1,z2,z3∈ℙ1 and w1,w2,w3∈ℙ1 distinct points, there exists a unique f such that f⁡(zj)=wj for j=1,2,3.

To prove existence, we consider the case when w1,w2,w3=0,1,∞. Then we simply set

To prove uniqueness, we need to do this for the case where z1,z2,z3=w1,w2,w3=0,1,∞. First, fixing ∞ means that our transformation is given by a⁢x+b. Fixing 0 means that b=0, and fixing 1 gives us a=1.

Now let z1,z2,z3,z4∈ℙ1. Then we define the cross ratio

Then if z1,z2,z3,z4 are distinct, the cross ratio is a complex number. Note that if f is the unique Möbius transformation sending z2,z3,z4↦1,0,∞, then C⁢R⁢(z1,z2,z3,z4)=f⁡(z1). Therefore, fractional linear transformations preserve the cross ratio. Alternatively, we can compute using brute force.

Proposition 3.1.9.

Fractional linear transformations take circles and lines to circles and lines (for a nicer formulation, these are all circles on ℙ1=S2, where lines are circles passing through ∞).

Proof.

We will prove that C⁢R∈ℝ⁢ℙ1 if and only if z1,z2,z3,z4 lie on a circle or a line. To see this, write arg⁡(z1−z3z1−z4)=α or −(π−α). Then arg⁡(C⁢R)=0 if z1,z2 lie on the same side of the line connecting z3,z4 and π if they lie on opposite sides, so C⁢R∈ℝ.

Conversely, if we fix the circle passing through z2,z3,z4, then z1∈C if and only if arg⁡(C⁢R)=0 or arg⁡(C⁢R)=π if and only if C⁢R∈ℝ⁢ℙ1. □

For an alternative proof, we know the result is true for z↦a⁢z and z↦z+b. Then we can check that the result holds for z↦1z, and finally we see that these transformations generate P⁢G⁢L2⁢(ℂ).

Example 3.1.10.

We may apply this to conformal mapping problems. Set

Then f⁡(1)=0,f⁡(i)=1,f⁡(−i)=∞. We also know that f⁡(∂D)=∂ℋ=ℝ⁢ℙ1, and f⁡(D)=ℋ because f⁡(0)=i.

Lemma 3.1.11.

Let U,V⊂ℂ and f:U→V be holomorphic and injectivve. Then f′⁢(z)≠0 for all z∈U and f−1:f⁡(U)→U is holomorphic.

Proof.

Near z0, write f⁡(z)=w0+(z−z0)m⁢g⁢(z), where g⁡(z0)≠0. In fact, we can write f⁡(z)=w0+(h⁡(z))m because we can locally define an m-th root of g. Then near z0, we see that f is given as the composition

Then if m≥2, the map z↦zm is not injective near 0, so f is not injective near z0. Finally, f−1 is holomorphic by the inverse function theorem. □

Theorem 3.1.12 (Inverse function theorem).

Let F:U→ℝ2 and p∈U such that det⁡(D⁢F⁢(p))≠0. Then there exist open neighborhoods p∈U′⊂U and F⁡(p)∈V⊂ℝ2 such that F:U′→V is a bijection with F−1 differentiable and D⁢F−1⁢(F⁡(p))=D⁢F⁢(p)−1.

Lemma 3.1.13 (Schwarz).

Let D be the unit disk and f:D→D be holomorphic such that f⁡(0)=0. Then

  1. 1.

    For all z∈D, |f⁡(z)|≤|z|;

  2. 2.

    If there exists z0∈D such that |f⁡(z0)|=|z0|, then f is a rotation.

  3. 3.

    |f′⁢(0)|≤1 with equality if and only if f is a rotation.

Proof.

Consider the function g⁡(z)=f⁡(z)z. Then if f⁡(z)=a0+a1⁢z+a2⁢z2+⋯, we know a0=f⁡(0)=0, so g⁡(z)=a1+a2⁢z+a3⁢z2+⋯ is holomorphic on D. Then we apply the maximum principle to g, and we note that |g⁡(z)|<1r for |z|=r<1, so |g⁡(z)|<1r for |z|≤r. Allowing r→1 from below, we see that |g⁡(z)|≤1 for all z∈D and therefore |f|⁢(z)≤|z|.

Now if |f⁡(z)|=|z| for some z0∈D, then g⁡(z) must be constant and therefore f⁡(z)=c⁢z with |c|=1. Finally, we note that g⁡(0)=f′⁢(0) and then if g⁡(0)=1, we see that f is a rotation by the second part. □

Now we are ready to consider automorphisms of the disk and upper half plane. In the case of the disk, we have two kinds of automorphisms:

  1. 1.

    Rotations z↦ei⁢θ⁢z;

  2. 2.

    Blaschke factors ψα⁢(z)=α−z1−α¯⁢z, where α∈D. We can check that ψα⁢(∂D)=∂D.

Proposition 3.1.14.

Every automorphism of the disk is of the form

for some θ,α.

Proof.

Given an automorphism f:D→D, then the map f∘ψα fixes the origin and is a rotation by the Schwarz lemma because it also fixes α. Therefore, f=r∘ψα−1=r∘ψα, where r is some rotation. □

Note that if f:D→D is an automorphism with f⁡(0)=0, then the Schwarz lemma implies that |f⁡(z)|≤|z| and |f−1⁢(z)|≤|z| and therefore |z|=|f⁡(z)| and therefore f is a rotation.o

Now we want to study automorphisms of ℋ. We know that ℋ,D are isomorphic, so Aut⁡D=Aut⁡ℋ.

Theorem 3.1.15.

We have Aut⁡(ℋ)≅P⁢S⁢L2⁢(ℝ).

Proof.

Let f⁡(z)=a⁢z+bc⁢z+d. Then we see that

and thus we require a⁢d−b⁢c>0. Therefore we have P⁢S⁢L2⁢(ℝ)⊂Aut⁡ℋ.

Now we prove surjectivity. First, we note that P⁢S⁢L2⁢(ℝ) acts transitively on ℋ. First, if b∈ℝ, z↦z+b preserves ℋ and if λ>0, z↦λ⁢z preserves ℋ. Next, we consider the effect of rotation (cos⁡φ−sin⁡φsin⁡φcos⁡φ). But we can compute that

To finish, we show that given f:ℋ→ℋ, then f∈P⁢S⁢L2⁢(ℝ). Suppose f⁡(α)=i. Then there exists g∈G such that g⁡(i)=α. Then g∘f fixes i, so F∘g∘f∘F−1 fixes 0 and is thus a rotation. Therefore there exists h∈G such that F∘g∘f∘F−1=F∘h∘F−1, so g∘f=h and f=h∘g−1. □

3.2 Riemann Mapping Theorem

Our goal in this section is to prove the Riemann mapping theorem.

Theorem 3.2.1.

Let Ω⊂ℂ be open and {fn}n=1∞ be a sequence of holomorphic functions fn:Ω→ℂ which converges uniformly to f on every compact subset K⊂Ω. Then f is holomorphic.

Proof.

Let D={z∈ℂ∣|z−z0|<r}⊂Ω. Then fn being holomorphic implies that ∫γfn⁢(z)⁢dz=0 for all closed curves γ in D. Because fn→f uniformly, we have

Also we know that f is continuous. Now we define

Here, γz is a path from z0 to z for a fixed basepoint z0. Then F is well-defined and F′=f, so f is holomorphic. □

Lemma 3.2.2.

Let Ω⊂ℂ be open and fn:Ω→ℂ be holomorphic. Suppose that fn→f uniformly on compact sets. Then fn′→f′ uniformly on compact sets.

Proof.

Let K′={z∈ℂ∣d⁡(z,K)⁢l⁢e⁢q⁢r} for some r∈ℝ such that K′⊂Ω. Then we know fn→f uniformly on K′, and we will use this to show that fn′→f′ uniformly on K. Using the Cauchy integral formula to some holomorphic function h, we have

so then we can bound

Applying this to h=fn−f, we see that for all ε>0, there exists N∈ℕ such that |fn⁢(w)−f⁡(w)|<ε on K′ for all n>N, and this implies that |fn′⁢(z)=f′⁢(z)|<1r⁢ε on K. □

Now let Ω⊂ℂ be open and ℱ be a set of holomorphic functions on Ω. We say that ℱ is normal if every sequence in ℱ has a subsequence which converges uniformly on compact sets to some function f:Ω→ℂ. We say that ℱ is uniformly bounded on compact sets if for all compact K⊂Ω, there exists M∈ℝ such that |f⁡(z)|≤M for all f∈ℱ,z∈K. Finally, we say that ℱ is equicontinuous on a compact set K⊂Ω if for all ε>0 there exists δ>0 such that |f⁡(z)−f⁡(w)|<ε for all z,w∈K such that |z−w|<δ and f∈ℱ.

Theorem 3.2.3 (Montel).

Let ℱ be a family of holomorphic functions on Ω⊂ℂ that is uniformly bounded on comact sets. Then

  1. 1.

    ℱ is equicontinuous on compact sets;

  2. 2.

    ℱ is normal.

Remark 3.2.4.

The first part uses holomorphicity, and the second part uses a general fact, the Arzelà-Ascoli theorem.

Example 3.2.5.

Let fn⁢(z)=zn. Then ℱ={fn} is not equicontinuous on K=[0,1]. In fact, |fn⁢(z)−fn⁢(1)|=|zn−1|→1 as n→∞ for z∈(0,1).

Example 3.2.6.

Let ℱ={fn⁢(z)=sin⁡(n⁢z)}. Then ℱ is not equicontinuous on K=[0,1] because fn⁢(π2⁢n)=1, but fn⁢(0)=0.

Proof of Montel’s theorem.
  1. 1.

    We will use the Cauchy integral formula to bound |f⁡(z)−f⁡(w)|. Let K⊂Ω be compact and choose r>0 such that K′={z∈ℂ∣d⁡(z,K)≤2⁢r}⊂Ω. Given z,w∈K such that |z−w|<r, we will let γ be the circle with center z and radius 2⁢r. We know γ⊂K′. Now we have

    Then we can bound

    so

    where |f⁡(u)|≤M for all u∈K,f∈ℱ. Therefore ℱ is equicontinuous on K.

  2. 2.

    Let K⊂Ω be compact and fn be a sequence in ℱ. First let (wj)j=1∞ be a sequence in K which is dense. To construct such a sequence, recall that

    By compactness, there exists a finite Sn⊂K such that K=⋃x∈SnD⁡(x,1/n)∩K, and now we can take S=⋃n≥1Sn and obtain a countable dense subset.

    Now we will use a “diagonal argument.” If fn is uniformly bounded, then the sequence fn⁢(w1) has a convergent subsequence fn,1⁢(w1). Continuing, there exists a subsequence {fn,2} of {fn,1} such that fn,2⁢(w2) converges. For each m∈ℕ, there exists a subsequence {fn,m} of {fn} such that fn,m⁢(wk) converges for all k≤m. Now take gn=fn,n. This is a subsequence of {fn} such that gn⁢(wk) converges for all k∈ℕ. If z∈K, then

    by the triangle inequality.

    Now carefully, we know that given ε>0 there exists δ>0 such that |gn⁢(z)−gn⁢(w)|<ε whenever |z−w|<δ and z,w∈K. Because K is compact, K⊆⋃j=1JD⁡(wj,δ) for some J∈ℕ. Given z∈K, there exists wj such that |z−wj|<δ, so |gn⁢(z)−gm⁢(z)|<ε. Now for n,m large, we have |gn⁢(z)−gm⁢(z)|<3⁢ε because there exists N∈ℕ such that |gn⁢(wj)−gm⁢(wj)|<ε for n,m≥N. Therefore gn⁢(z) converges uniformly for all z∈K.

    For all compact sets K, we need another diagonal argument. There exists an exhaustion

    such that Kℓ is contained in the interior of Kℓ+1 and for all K⊂Ω compact, K⊂Kℓ for some ℓ. For example, set

    Finally, given a sequence fn∈ℱ, we have a subsequence gn,1 converging uniformly on K, a subsequence gn,2 of gn,1 converging uniformly on K2, and now the sequence hn=gn,n converges uniformly on all Kℓ and hence on all K. □

We are now ready to prove the Riemann mapping theorem, and in fact we have a stronger statement.

Theorem 3.2.7 (Riemann mapping theorem).

Let Ω⊂ℂ be a nonempty proper simply-connected open subset of ℂ. Fix z0∈Ω. Then there exists a unique holomorphic bijection F:Ω→B such that F⁡(z0)=0,F′⁢(z0)∈ℝ>0.

Remark 3.2.8.

Uniqueness follows from the Schwarz lemma. If we have F1,F2 two such functions, then G=F2∘F1−1:D→D has G⁡(0)=0 and G′⁢(0)∈ℝ>0, so G⁡(z)=ei⁢θ⁢z. But then G′⁢(0)=ei⁢θ∈ℝ>0, so G=id.

Proof.

Consider

By construction, ℱ is uniformly bounded and thus normal.

First, we show that ℱ is nonempty. By assumption Ω≠ℂ, so let a∈ℂ∖Ω. Then we can define f:Ω→ℂ defined by z↦log⁡(z−a). Then ef⁡(z)=z−a. Note that f is injective and recall that ew+2⁢π⁢i⁢k=ew for all k∈ℤ. Take w0∈f⁡(Ω). Then f⁡(Ω) is open, so D⁡(w0,δ)⊂f⁡(Ω) for some δ>0. Then we see that D⁡(w0+2⁢π⁢i,δ)∩f⁡(Ω)=∅, so we can take

and we have |g|<1δ. Because f is injective, g is injective. Composing with translation and rotation, we can assume that g⁡(z0)=0 and g′⁢(z0)∈ℝ>0. Composing with scaling, we may assume that |g|<1, so g∈ℱ.

Now let gn be a sequence in ℱ such that gn′⁢(z0)→supg∈ℱg′⁢(z0). By Montel’s theorem, gn→g uniformly on compact sets for some g:Ω→ℂ. then g is holomorphic and gn′→g′ uniformly on compact sets, so in particular g′⁢(z0)=suph∈ℱh′⁢(z0). We will prove that g is a bijection. To prove injectivity, fix z1,z2∈Ω. Define g~n⁢(z)=gn⁢(z)=gn⁢(z1) and g~⁢(z)=g⁡(z)−g⁡(z1). Then g~n⁢(z)≠0 for z≠z1. Given z2∈Ω∖{z1}, there exists δ>0 such that g~⁢(z)≠0 for 0<|z−z2|≤δ. Also assume |z1−z2|>δ. Let γ={z∣|z−z2|=δ}. Then we have

so g is injective.

We will prove surjectivity by contradiction. Suppose there exists α∈D∖g⁡(Ω). Define U=ψα⁢(g⁡(Ω)) and U⊂D is open, simply connected, and 0∉U. We can define k:U→D by k⁡(z)=z=e12⁢log⁡z. Now we can set G=rθ∘ψβ∘k∘ψα∘g, where β=k⁡(α). Equivalently, we have g=ψα∘ℓ∘ψβ∘r−θ∘G, and set ψα∘ℓ∘ψβ∘r−θ=L. Then L:D→D fixes 0 and is not a rotation, so |L′⁢(0)|<1. Therefore |g′⁢(z0)|=|L′⁢(0)⋅G′⁢(0)|<|G′⁢(z0)| by the chain rule, which is a contradiction. □

3.3 Elliptic Functions

Recall that meromorphic functions on ℙ1 are the rational functions. We would like to study meromorphic functions on the complex torus X=ℂ/Λ, where Λ=ℤ⁢ω1+ℤ⁢ω2 for 0≠ω1,ω2∈ℂ with ω2/ω!∉ℝ. Equivalently, we want to consider doubly periodic meromorphic functions on ℂ.

Lemma 3.3.1.

If f is holomorphic on X, then f is constant.

Proof.

Clearly, f is given by values on a parallelogram, so f is bounded and therefore constant. □

Remark 3.3.2.

If X is a compact Riemann surface, then all holomorphic functions f:X→ℂ are constant. This is because |f| is bounded and attains its maximum on some chart, so it must be constant.

Lemma 3.3.3.

Let f be meromorphic on ℂ/Λ. Then f has at least 2 poles.

Proof.

Integrating around the boundary of the parallelogram, we see that

because ∑Resp⁡f=0 by an analog of the residue theorem. □

Remark 3.3.4.

There is a suitably generalized version of the residue theorem for any compact Riemann surface. First, define a meromorphic 1-form ω. Choose charts φi:Ui≃Vi. On each Vi, we have a form fi⁢(z)⁢dz with fi⁢(z) meromorphic. Then compatibility with transitions gi⁢j is given by

and thus we can define the integral of ω along a curve γ⊂X. Now the residue theorem says that if X is a compact Riemann surface and ω is a meromorphic 1-form on X, then ∑p⁢poleResp⁡ω=0.

For example, on X=ℙ1, we can consider ω=1z⁢dz. Then on the second chart, we see that ω=w⋅d(1w)=−1w⁢dw. Then we see that ∑pResp⁡ω=1+(−1)=0.

To prove the residue theorem, triangulate X such that each triangle is contained in some Ui. Then ∑Δi∫∂Δiω=0=2⁢π⁢i⁢∑pResp⁡ω.

Corollary 3.3.5.

Let f be meromorphic on a compact Riemann surface X. Then the number of zeroes of f equals the number of poles of f.

Proof.

Given f, we may define the meromorphic form df. Then we can consider the meromorphic form ω=dff. Then by the residue theorem we see that

The geometric meaning is that if f:X→ℙ1 is holomorphic, then for all α, f−1⁢(α)=deg⁡f, counting multiplicity.

Now we will define an explicit meromorphic function ℘ on ℂ/Λ with a unique double pole at 0∈X. Equivalently, ℘:ℂ→ℙ1 has a double pole at each point in Λ. Define

because the first attempt ℘⁡(z)=∑ω∈Λ1(z+ω)2 doesn’t converge.

Lemma 3.3.6.

The function ℘⁡(z) converges absolutely and uniformly on compact sets K∈ℂ∖Λ and has a double pole at each ω∈Λ.

Proof.

First note that

for |ω| large and z bounded. Now we show that ∑0≠ω∈Λ1|ω|3 converges. First, we know that ω=x1⁢ω1+x2⁢ω2. Then for any x1,x2∈ℝ, we know |x1⁢ω1+x2⁢ω2|≥C⁢x12+x22. This is because the function f⁡(x1,x2)=|x1⁢ω1+x2⁢ω2| satisfies f⁡(λ⁢𝐱)=λ⁢f⁢(𝐱). Therefore if ‖x‖=1, we know f lands in the positive reals and has a positive minimum. Therefore, we have reduced to the case where ω1=1,ω2=i. Now we want to show that

converges. Using the integral test, we see that

and therefore it converges for α>2. □

Now we need to check that ℘ is doubly periodic. Here, we simply note that

is doubly periodic, so ℘⁡(z+ω)−℘⁡(z) is constant. Now checking ω=ω1 and z=−12⁢ω, we see that ℘⁢(12⁢ω)−℘⁢(−12⁢ω)=0 because ℘ is even.

Theorem 3.3.7.

We have the differential equation

where α1,α2,α3=℘⁢(ω12),℘⁢(ω22),℘⁢(ω1+ω22).

Proof.

We will compute the zeroes and poles of the two sides of the equations and show that they agree. Then their quotient is holomorphic on X=ℂ/Λ and thus constant, so we can compute the constant. Recall that ℘ has a pole of order 2 at 0 and ℘′ has a pole of order 3 at 0. Next, ℘ is even so ℘′ is odd. Then we know that

and thus ℘′⁢(ω2)=0. Then ℘−℘⁢(ω2) has a zero of order 2 at ω2.

Now to compute the constant, we use the Laurent series at the origin. We know that

and thus (℘′)2=4⁢∏i=13(℘−αi). □

Now we have the identity

where w=℘⁡(z) and (℘′)2=∏i=13(℘−αi).

Theorem 3.3.8.

Every meromorphic function on X=ℂ/Λ is a rational function of ℘,℘′. This means that the field of meromorphic funcions on X is given by

Therefore we can write f∈ℂ⁡(x) uniquely as a⁡(℘)+b⁡(℘)⁢℘′ for a,b∈ℂ⁡(x).

Proof.

First suppose that f is even. We will show that f is rational in ℘. We may assume that f has no poles on Λ⊂ℂ up to replacing f by ℘−m⁢f. Now if we write down a function with the same zeroes and poles and is a rational function of ℘. Recall that ℘⁡(z)−℘⁡(a) has zeroes {a,−aa≠−aa⁢(twice)a∈12⁢Λ. But then for our function f, because f is even, the zeroes and poles of f come in pairs a,−a or a∈12⁢Λ has even multiplicity. Then f has the same poles as

for some ai,bj∈ℂ. Note that ℘ has a pole of multiplicity 2 at 0∈X and no other poles and f has no zero or pole at 0∈X by assumption. But we are fine because #⁢{ai}=#⁢{bj}.

In the general case, suppose f is meromorphic on X. Then write

Then if g is odd, g℘′ is even and thus rational in ℘, so we obtain the desired description. □

Now we can give a geometric description of this. Note that X=ℂ/Λ→℘ℙ1 has degree 2. It has branch points at a1=℘⁢(ω12,a2=℘⁢(ω22)),a3=℘⁢(ω1+ω22),∞. Now we have a map

Now we can make two branch cuts, then expand the branch cuts into circles, and obtain half of a torus. Gluing two halves together, we obtain the torus. There is an involution (x,y)↦(x,−y) that exchanges the two sheets of the torus. As a picture, it looks like

⟵2:1
Figure 3.2: Branched cover of sphere by torus

3.4 Winding Numbers

Let γ⊂ℂ be a closed path. For a∈ℂ∖γ, define the winding number

This is reasonable because if we consider the covering space ℝ→S1=ℝ/2⁢π⁢ℤ. Then if we write z⁡(t)=a+r⁡(t)⁢ei⁢θ⁢(t), we have

Theorem 3.4.1.
  1. 1.

    For all a,γ, we have n⁡(γ,a)∈ℤ.

  2. 2.

    If a,b lie in the same component of ℂ∖γ, then n⁡(γ,a)=n⁡(γ⁡(b)).

  3. 3.

    If a lies in the unbounded component of ℂ∖γ, then n⁡(γ,a)=0.

  4. 4.

    If γ,δ are homotopic in ℂ∖{a}, then n⁡(γ,a)=n⁡(δ,a).

Proof.
  1. 1.

    Let z:[0,1]→γ⊂ℂ. Then define

    Then we see that ((z⁡(t)−a)⁢e−g⁡(t))′=z′⁢(t)⁢e−g⁡(t)−(z⁡(t)−a)⁢g′⁢(t)⁢e−g⁡(t)=0. But then eg⁡(1)=eg⁡(0)≡c⋅(z⁡(0)−a), so g⁡(1)−g⁡(0)∈2⁢π⁢i⁢ℤ. Therefore n⁡(γ,a)∈ℤ.

  2. 2.

    We know that the winding number ℂ∖γ→ℤ⊂ℂ is continuous, so it must be locally constant.

  3. 3.

    Note that

    so as a→∞, n⁡(γ,a)→0, but it must be constant, so n⁡(γ,a)=0.

  4. 4.

    Note that 1z−a is holomorphic on ℂ∖{a}. Then the desired result follows from Theorem 2.7.1. □

We can compute winding numbers using Alexander numbering.

120-100
Figure 3.3: Winding numbers

We know the winding number vanishes on the unbounded component, and given α↑β, we have n⁡(γ,α)=n⁡(γ,β)+1 because in the diagram

αγβ↝αγβδ⁢R≡αβγ+δ⁢R
Figure 3.4: Pictoral proof of Alexander numbering

we have n⁡(γ,α)=n⁡(γ+δ⁢R,α)=n⁡(γ+δ⁢R,β)=n⁡(γ,β)+1.

Theorem 3.4.2 (Cauchy).

Let f:Ω→ℂ and γ⊂Ω be a closed curve such that n⁡(γ,a)≠0 for all a∈ℂ∖Ω. Then

Corollary 3.4.3 (General Residue Theorem).

Let Ω⊂ℂ be open and S⊂Ω be a discrete set. Suppose f:Ω∖S→ℂ is holomorphic and γ⊂Ω∖S is a closed curve. Then

Remark 3.4.4.

The assumption that S is discrete implies that there are only finitely many a∈S with nonzero winding number.

Proof.

We will reduce to Cauchy. Set δ≔γ−∑a∈Sn⁡(γ,a)⋅γa, where γa is a small circle around a. By construction, n⁡(δ,a)=0 for all a∈ℂ∖(Ω∖S) and therefore the ordinary Cauchy theorem implies

as desired. □

Before we prove Cauchy, we need the following lemma:

Lemma 3.4.5.

Let Ω⊂ℂ and γ:[a,b]→Ω be a path. Then there exists a rectangular path η:[a,b]→Ω such that there exists a subdivision a=a0<a1<⋯<an=b such that γ⁡(ai)=η⁡(ai) for all i and there exist disks Di⊂Ω such that γ⁡([ai,ai+1]),η⁡([ai,ai+1])⊂Di.

Proof.

There exists ε>0 such that if z∈γ, then D⁡(z,ε)⊂Ω. Then there exists δ>0 such that |γ⁡(s)−γ⁡(t)|<ε for |s−t|<δ. Subdivide a=a0<a1<⋯<an=b such that |ai+1−ai|<δ and set Di=D⁡(γ⁡(ai),ε)⊂Ω. Now we can build the rectangular path inside each Di. □

Proof of Cauchy.

We may assume that γ is a rectangular path. Then there exist rectangles Ri such that γ=∑mi⁢∂Ri for some mi∈ℤ. To see this, just draw a fine enough rectangular grid until every segment of γ is aprt of the grid.

Now we need to show that if n⁡(γ,ai)≠0, then Ri⊂Ω, where ai is in the interior of Ri. But this is because Ri∘⊂A for some connected component A of C∖γ. By assumption, A⊂Ω, so Ri⊂A¯⊂Ω.

Now we show that γ=∑n⁡(γ,ai)⁢∂Ri, where ai∈Ri∘. We will show that if δ≔γ−∑n⁡(γ,ai)⁢∂Ri, then δ=0. First, we know that n⁡(δ,a)=0 for all a∈ℂ∖δ (simply check in each rectangle). Now if δ≠0, then δ contains some component m⁢σ, so δ=m⁢σ+δ′ for some m≠0 and σ⊄δ′. But then if σ splits α,α′, we have n⁡(δ,α)=n⁡(δ,α′)+m, a contradiction. □

Theorem 3.4.6.

Let Ω⊂ℂ be open and connected. Then the following are equivalent:

  1. 1.

    Ω is simply connected.

  2. 2.

    ℙ1∖Ω is connected.

  3. 3.

    For all closed curves γ in Ω, then n⁡(γ,a)=0 for all a∈ℂ∖Ω.

Proof.
2 implies 3:

If ℙ1∖Ω is connected, then ℂ∖Ω is contained in the unbounded component of ℂ∖γ, and thus n⁡(γ,a)=0 for all a∈ℂ∖Ω.

3 implies 2:

Suppose ℙ1∖Ω is not connected. Then we can write ℙ1∖Ω=A∪B for closed and disjoint A,B where B∈∞. Now we need to produce γ⊂Ω such that n⁡(γ,a)≠0 for some a∈ℂ∖Ω. There exists δ>0 such that d⁡(A,B)>δ. Then take the square grid with side length less than δ2 and now take

Observe that γ∩A=∅ and γ∩A=∅, so γ⊂Ω. But then we see that n⁡(γ,a)=1 for all a∈A∩R∘, where R is a rectangle in the grid.

1 implies 3:

Recall that n⁡(γ,a)=12⁢π⁢i⁢∫γ1z−a⁢dz. If Ω is simply connected, then ∫γf⁡(z)⁢dz=0 for all γ⊂Ω and f:Ω→ℂ. Now apply this to f=1z−a for a∉Ω.

3 implies 1:

Recall that there exists a holomorphic bijection F:Ω→∼D if Ω is simply conencted. However, the simply connected assumption was only used to define log⁡(z−a),z−a for a∈Ω. Then log⁡(z−a) is defined as the primitive of 1z−a. Now we know that if f is holomorphic on Ω, then ∫γf⁡(z)⁢dz=0 for all γ⊂Ω closed curves, so f has a primitive g. But then we obtain a holomorphic bijection F:Ω→D, and thus Ω is simply connected. □

Remark 3.4.7.

In general, n⁡(γ,a)=0 for all a∈ℂ∖Ω does not imply that γ is homotopic to a constant path. For example, if Ω=ℂ∖{a,b} and γ is not homotopic and u,v generate π1⁢(Ω)=ℤ∗ℤ, then the loop γ=u⁢v⁢u−1⁢v−1 satisfies n⁡(γ,a)=n⁡(γ,b)=0.

Now we would like to apply the general Cauchy’s theorem in practice. Let Ω⊂ℂ be open and connected and suppose ℙ1∖Ω=A1∪⋯∪AN∪A∞∋∞ has finitely many components. Then there exist curves γj in Ω such that if we choose ai∈Ai, then n⁡(γi,aj)=δi⁢j. Now if γ⊂Ω, it has the same winding numbers around a∈Ω as ∑n⁡(γ,ai)⁢γi, so by Cauchy, we have

for f:Ω→ℂ holomorphic. Therefore

In particular, f has a primitive if and only if ∫γif⁡(z)⁢dz=0 for i=1,…,N.

Example 3.4.8.

Consider Ω={z∈ℂ∣|z|>4} and f⁡(z)=z(z−1)⁢(z−2)⁢(z−3). Now we see that if γ is a closed curve in Ω, then

so f does have a primitive.

Example 3.4.9.

We want to compute the integral

If we consider Ω=ℂ∖[−1,1], then we may consider γε,δ given by

−11
Figure 3.5: Contour γε,δ

with radius ε and width δ. Then ∫γ=∫γε,δ. Then we have

Alternatively, we may use the transformation z=12 to obtain

Example 3.4.10.

Consider the contour γ given by

01
Figure 3.6: Contour γ

Then we have

3.5 Harmonic Functions

We say u:Ω→ℝ is harmonic if it has continuous second partial derivatives and satisfies Laplace’s equation

Harmonic functions have applications in electricity and magnetism, fluid dynamics, gravity, and heat conduction.

Lemma 3.5.1.

Let f:Ω→ℂ be holomorphic and write f⁡(x+i⁢y)=u⁡(x,y)+i⁢v⁢(x,y). Then u,v are harmonic.

Proof.

By the Cauchy-Riemann equations, we have

and similar for v. □

There is a converse to this.

Theorem 3.5.2.

Let Ω⊂ℂ be simply connected and u:Ω→ℝ be harmonic. Then there exists a holomorphic f:Ω→ℂ such that u=Re(f). Moreover, f is unique up to f↝f+i⁢c for c∈ℝ.

Proof.

Define g≔∂u∂u−i⁢∂u∂y and integrate. Then we see that g is holomorphic because the components have continuous first derivatives and the Cauchy-Riemann equations are satisfied because u is harmonic and by symmetry of mixed partials.

Then because Ω is simply connected, there exists a primitive f:Ω→ℂ such that f′=g. Now write f=u~+i⁢v~ and note that

Therefore u~=u+a for some a∈ℝ, so up to f↝f−a, we have Re(f)=u.

To show uniqueness, note that if f1,f2 are two such functions, then Re(f1−f2)=0 and thus f1−f2∈i⁢ℝ. But this implies that Ω is sent to something that is not open so f1−f2 is a constant. □

Theorem 3.5.3.

Let u:Ω→ℝ be harmonic. Then for all z∈Ω,r∈ℝ>0 such that D¯⁢(z,r)⊂Ω, we have

Proof.

There exists D=D⁡(z,r′)⊂Ω for r′>r and f holomorphic on D such that u|D=Re(f). Applying the Cauchy integral formula, we have

Now the desired result follows by taking real parts. □

Theorem 3.5.4 (Maximum principle for harmonic functions).

Let u be harmonic on a connected Ω⊂ℂ. Then if u has a maximum in Ω, then u is constant.

Proof.

If Ω is simply connected, then we know u is the real part of some holomorphic f, so using the maximum principle for g=ef, we obtain the desired result. In general, suppose u has a maximum c at a point z0∈Ω. Then for z0∈D⊂Ω we know u is constant on D. Now define U={z∈Ω∣u≡c near z}. By definition, U is open, but U is also closed in Ω, so U=Ω. □

Example 3.5.5.

The function u⁡(x,y)=log⁡(x2+y2):ℝ2∖{(0,0)}→ℝ is harmonic.

Corollary 3.5.6.

Let Ω⊂ℂ=ℝ2 be bounded. Suppose u1,u2:Ω¯→ℝ are continuous on Ω¯ and harmonic on Ω. Further suppose that u1=u2 on ∂Ω. Then u1=u2.

Proof.

Write u=u1−u2. We know u=0 on ∂Ω and is harmonic on Ω and continuous on Ω¯. By the maximum principle, if u has a max in Ω then it is constant, but we know u attains a maximum on Ω¯ Therefore u≤0. Applying the same reasoning to −u, we see that u≥0. □

Now we would like to study the Dirichlet problem. Given Ω⊂ℂ=ℝ2 and g:∂Ω→ℝ continuous, we want to find u:Ω¯→ℝ continuous on Ω¯ and harmonic on Ω such that u|∂Ω=g.

Example 3.5.7.

Let Ω=A={z∈ℂ∣R1<|z|<R2} and set u⁡(z)=0 if |z|=R1 and u⁡(z)=1 when |z|=R2. Then the Laplace equation is a linear PDE and we expect u=u⁡(r)=u⁡(x2+y2). In fact, we can set

Now the generalized Dirichlet problem is as follows: Suppose Ω is simply connected with piecewise smooth boundary. Then the Riemann mapping theorem gives us a holomrophic bijection F:Ω→D. Then this extends to a homeomorphism F¯:Ω¯→D¯. Now fix a finite set S⊂∂Ω and continuous g:∂Ω∖S→ℝ.

Problem 3.5.8 (Generalized Dirichlet problem).

Find a continuous function u:Ω¯∖S→ℝ that is harmonic on Ω such that u|∂Ω∖S=g. Note that if such a u exists, then it is unique.

Example 3.5.9.

Let Ω=ℋ, S=0, and set g⁡(x)={1x<00x>0.

g=0g=1
Figure 3.7: Generalized Dirichlet problem on upper half plane.

Then we take u=arg⁡(z)π.

Example 3.5.10.

Consider the generalized Dirichlet problem given by

g=0g=1
Figure 3.8: Generalized Dirichlet problem on first quadrant.

Then we may take u=2π⁢arg⁡(z).

Lemma 3.5.11.

Let f:Ω2→Ω1 be holomorphic and u1:Ω1→ℝ be harmonic. Then u2=u1∘f is harmonic.

Proof.

Locally, we can write u1=Re(g) for some holomorphic g. Then u2=Re(g∘f) must be harmonic. □

Example 3.5.12.

Consider Ω={z∣|z|<1} and S={1,i}.

1ig=0g=1
Figure 3.9: Generalized Dirichlet problem on disk

Then a Möbius transformation f:ℙ1→ℙ1 taking D→ℋ and 1,i,−1↦∞,0,1 is given by

so we may take

Now we will give a general solution of the Dirichlet problem for the disk. Recall the mean value property

for 0<r<1. Assume that u extends continuously to D¯∖S and is bounded. Then as r→0, we see that u⁡(0)=12⁢π⁢∫02⁢πu⁡(ei⁢θ)⁢dθ. Now we use the Blaschke factor

Then if u~=u∘ψz, we have

Because we have

Now if Ω is simply connected, we have F¯:Ω¯→∼D¯ that restricts to a holomorphic bijection F:Ω→∼D. Then if u~:D→ℝ is a solution on D, u~∘F¯ is a solution on Ω.