0% found this document useful (0 votes)
12 views21 pages

Complexanalysis 5 Local Behavior of Holomorphic Functions

Uploaded by

iantyler329
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
12 views21 pages

Complexanalysis 5 Local Behavior of Holomorphic Functions

Uploaded by

iantyler329
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 21

Complex Analysis (part 5): Local Behavior of Holomorphic Functions (by Evan Dummit, 2022, v. 1.

00)

Contents
5 Local Behavior of Holomorphic Functions 1
5.1 Locations of Zeroes and Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

5.1.1 Counting Zeroes and Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

5.1.2 Rouché's Theorem and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

5.1.3 The Open Mapping Theorem, Local Invertibility and Local Preimages . . . . . . . . . . . . . 6

5.2 Conformal Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

5.2.1 The Point at Innity and the Extended Complex Plane . . . . . . . . . . . . . . . . . . . . . 9

5.2.2 Fractional Linear Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5.2.3 Conformal Maps and Analytic Isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

5.2.4 Functions on the Unit Disc, Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . 19

5 Local Behavior of Holomorphic Functions


In this chapter, our goal is to study the local behavior of holomorphic functions from various perspectives. We
begin by examining ways to count the number of zeroes and poles of a meromorphic function in a given region
and then describe ways to estimate the number of zeroes. Next, we establish the open mapping theorem: every
nonconstant holomorphic function maps open sets to open sets. We then study holomorphic functions on the unit
disc in some detail, providing some simple but powerful growth estimates and describing the action of fractional
linear transformations on such functions. Finally, we discuss conformal mapping, which allows us to use holomorphic
functions to transform complex regions into one another.

5.1 Locations of Zeroes and Poles

• In this section we will study the locations of zeroes and poles of holomorphic functions.

5.1.1 Counting Zeroes and Poles


• Our rst goal is to establish a procedure for counting the number of zeroes and poles a meromorphic function
f possesses in a region R.

◦ Suppose that f is meromorphic on R and not identically zero. Then by direct calculation we can see that
f 0 /f has singularities only where f has a zero or a pole, since otherwise both f and f 0 are holomorphic
0
and f is nonzero (so that the quotient f /f is also holomorphic).

◦ If f has a zero or pole at z0 k , then by taking out a factor of (z −z0 )k from the Laurent expansion
of order

k f 0 (z) k(z − z0 )k−1 g(z) + (z − z0 )k g 0 (z)


we see that f (z) = (z − z0 ) g(z) where g(z0 ) 6= 0. Then = =
f (z) (z − z0 )k g(z)
0 0
k g (z) g 0
+ , and since is holomorphic at z0 because g(z0 ) 6= 0, we see that f /f has a simple pole
z − z0 g(z) g
at z0 with residue k (the residue being positive when f has a zero and negative when f has a pole).

◦ By applying the residue theorem to f 0 /f we can count zeroes and poles of f:

1
• Theorem (Counting Zeroes and Poles): Suppose that f is meromorphic on R, not identically zero, and
γ is a simple closed counterclockwise contour in R such that f has no zeroes and no poles on γ . Then
´ f 0 (z)
γ f (z)
dz = 2πi · [N0 (f ) − N∞ (f )] where N0 (f ) is the number of zeroes of f inside γ and N∞ (f ) is the
number of poles of f inside γ , both counted with multiplicity.

f 0 (z)
◦ Proof: As noted above, the poles of are all simple and the residue at z0 is equal to the order of
f (z)
the zero of f at z0 (if f has a zero) or the negative of the order of the pole at z0 (if f has a pole).

◦ Therefore, by the residue theorem, if we sum over all zeroes and poles, we see immediately that
´ f 0 (z) P
γ f (z)
dz = 2πi zi ∈poles of f 0 /f Resf 0 /f (zi ) = 2πi[N0 (f ) − N∞ (f )] as claimed.

• As an immediate obvious corollary, we can count the number of zeroes of a holomorphic function on a region:

• Corollary: Suppose that f is holomorphic on R, not identically zero, and γ is a simple closed counterclockwise
´ f 0 (z)
contour in R such that f has no zeroes on γ . Then γ dz = 2πi · N0 (f ) where N0 (f ) is the number of
f (z)
zeroes of f inside γ counted with multiplicity.

◦ Proof: Apply the theorem above and observe that f has no poles because it is holomorphic.

• As examples we can verify the calculation for some simple functions.

• Example: Verify the zero-counting result for f (z) = z 3 inside the circle |z| = 1.

◦ Obviously f has a triple zero at z = 0. We have f 0 /f = 3z 2 /z 3 = 3/z , and with γ the boundary of the
´ 3
unit circle we indeed get
γ z
dz = 3 · 2πi, as expected.

8z
• Example: Verify the zero-and-pole-counting result for f (z) = inside the circle |z| = 2.
z−1
8
◦ Obviously f has a simple zero at z = 0 and a simple pole at z = 1. We have f 0 (z) = − so
(z − 1)2
1 1 1
f 0 /f = − = − .
z(z − 1) z z−1
´ f 0 (z) ´ 1 1
◦ With γ the boundary of |z| = 2 we get
γ f (z)
dz = γ [ − ] dz = 2πi − 2πi = 0 as expected.
z z−1

f 0 (z)
• It is not obvious why precisely one would think of considering the integral of at all, but there are various
f (z)
ways to explain why this is in fact a natural idea.

f 0 (z) d
◦ The most direct is to observe that is the logarithmic derivative [log[f (z)]], and so by the
f (z) dz
´ f (z)0
fundamental theorem of calculus, the integral
γ f (z)
dz should be equal to the dierence in the

values of log[f (z)] at the two endpoints of the contour.

◦ Of course, the complex logarithm is multivalued, so this evaluation does not really make sense as written,
but if we allow some abuse of notation, the integral should be the dierence of two values of the complex
logarithm at the endpoints of the contour, which (since the contour is closed) is an integer multiple of
2πi.
◦ We have encountered a similar phenomenon previously when we formalized the notion of winding num-
bers: in fact, we can rephrase this argument entirely in terms of winding numbers.
´ f 0 (z) ´ b f 0 (γ(t)) 0
◦ Explicitly, suppose we have a parametrization of γ as γ(t) for a ≤ t ≤ b: then
γ f (z)
dz = a γ (t) dt.
f (γ(t))

2
◦ But, motivated by the substitution s = f (γ(t)) with ds = f 0 (γ(t))γ 0 (t) dt, we can see that this integral
´ 1
also equals the line integral
Γ z
dz on the contour Γ=f ◦γ dened with Γ(t) = f (γ(t)) for a ≤ t ≤ b.
´ 1
◦ From our discussion of winding numbers,
Γ z
dz equals 2πi times the winding number of the contour

f ◦γ around z = 0.
◦ Thus, in other words, the zero-and-pole-counting calculation above can be equivalently formulated in
terms of the winding behavior of the contour Γ=f ◦γ around the origin.

◦ This interpretation is often called the argument principle since it is equivalent to tabulating the net change
in the argument (i.e., the polar angle) of f (γ(t)) as t varies from a to b, which in turn is equivalent to
counting the number of net crossings that the contour f ◦γ makes across the positive real axis.

• Here is an illustration of the argument principle with the function f (z) = z 2 and the contour given by the
counterclockwise boundary of the triangle with vertices −1 − i, 1 − i, and i:

◦ We can see that the triangle has winding number 1 around 0, whereas its image under the squaring map
has winding number 2, reecting the fact that f (z) = z 2 has 2 total zeroes (and no poles) counted with
multiplicity inside the contour.

´ f 0 (z)
• Of course, as a practical matter, it is not generally so easy to compute the integral
γ f (z)
dz if f is suciently

complicated in a way that makes it dicult to identify its zeroes and poles.

◦ The zero-and-pole-counting theorem is of far more interest as a theoretical device.

5.1.2 Rouché's Theorem and Applications


• Suppose that f is holomorphic on a simply connected region R with boundary γ, and f is nonzero on γ.

◦ Then, per the argument principle, the number of zeroes of f inside γ is equal to the winding number of
f ◦γ around the origin.

◦ If we perturb f by a small amount to obtain a function g , then since γ is continuous the composition
f ◦γ will also be perturbed by a small amount to yield g ◦ γ.
◦ As long as the perturbation is small enough that it does not cause f ◦ γ to pass through the origin, then
by our results on deformation of contours, f ◦ γ and g ◦ γ will have the same winding number around 0
and thus by the argument principle, f and g will have the same number of zeroes inside γ .
◦ More precisely, we require that the homotopy function ht (s) = tf (s) + (1 − t)g(s) not pass through the
origin for any s∈γ and any 0 ≤ t ≤ 1.

3
◦ Since ht (s) = 0 precisely when tf (s) = (1 − t)(−g(s)), we must avoid the situation where one of f (s)
and −g(s) is a nonnegative real multiple of the other. Conveniently, this is precisely the situation where
the triangle inequality holds: |f (s) − g(s)| = |f (s)| + |g(s)| if and only if one of f (s) and −g(s) is a
nonnegative real multiple of the other.

◦ |f (s) − g(s)| < |f (s)| +


Therefore, if we insist on strict inequality in the triangle inequality, namely, that
|f (s)|, then the homotopy function ht (s) cannot take the value zero, and so we may deform the contour
f ◦ γ into g ◦ γ without passing through the origin. By the argument principle, we can then conclude
that f and g have the same number of zeroes inside R.

◦ We can formalize this as follows:

• Theorem (Rouché's Theorem): Suppose that f and g are holomorphic on a simply connected bounded region
R with counterclockwise boundary γ. If |f (z) − g(z)| < |f (z)| + |g(z)| for all z ∈ γ , then f and g have the
same number of zeroes inside γ, counted with multiplicity.

◦ The argument we gave above using deformation of contours is already essentially rigorous, but we will
give another approach for illustration.

◦ Proof: Suppose that |f (z)| < |f (z) − g(z)| + |g(z)| requires both f (z) and g(z) not to vanish on γ, so (in
particular) they also cannot be identically zero.

´ f 0 (z) ´ g 0 (z)
◦ It therefore suces to show that
γ f (z)
dz = γ dz since these integrals calculate 2πi times the
g(z)
number of zeroes that f and g respectively have inside γ .

f (z)
◦ Additionally, we note that cannot be a nonpositive real number −α on γ : if f (z) = −αg(z)
g(z)
f (z) f (z)
then − 1 = |−α − 1| = −α − 1 = − 1 so multiplying by g(z) would give |f (z) − g(z)| =
g(z) g(z)
|f (z)| − |g(z)|, contradicting the given hypothesis.
´ f 0 (z) g 0 (z) ´ d
 
f (z)
◦ Now observe that γ − dz = γ [log ] dz where the logarithm's branch cut is taken
f (z) g(z) dz g(z)
f (z)
along the negative real axis. Since does not take any nonpositive real values on γ , the integrand
g(z)
f (z)
log is holomorphic along γ .
g(z)
´ d f (z)
◦ Thus, by the fundamental theorem of calculus, the integral γ [log ] dz is zero, and so we imme-
dz g(z)
´ f (z)
0 ´ g (z)
0
diately conclude
γ f (z)
dz = γ dz as desired.
g(z)
◦ Remark: The connection between this argument and the one above in terms of winding numbers is that
for F (z) = f (z)/g(z), the contour F ◦ γ never crosses the nonpositive real axis, and therefore it has
winding number 0 around the origin.

• In practice, one often uses the weaker hypothesis |f (z) − g(z)| < |f (z)| when applying Rouché's theorem.

◦ Indeed, Rouché's original formulation was in terms of the perturbation h(z) = f (z) − g(z), in which case
the condition is |h(z)| < |f (z)|: the perturbation is smaller in absolute value than the original function.

◦ By running the condition the other way one can also take the hypothesis |h(z)| < |g(z)|: the perturbation
is smaller in absolute value than the new function.

◦ The version using the weaker hypothesis |f (z) − g(z)| < |f (z)| + |g(z)|, which is symmetric in f and g,
was given by Estermann in 1962. This version is also equivalent to the more obviously symmetric version
of the theorem with hypothesis |f (z) + g(z)| < |f (z)| + |g(z)| upon replacing g with −g (since g and −g
obviously have the same zeroes with the same orders).

• Although the hypothesis in Rouché's theorem may seem rather unintuitive, there is a quite nice nice geometric
interpretation.

4
◦ Imagine a person walking their leashed dog around a agpole. The dog (being a dog) will run around its
owner, sometimes staying close and other times moving far away, with the owner letting the leash out
and reeling it back in as appropriate.

◦ The principle is that as long as the dog's leash is not long enough to reach all the way to the agpole,
from wherever the owner is currently located, then once the owner nishes circulating around the agpole
and returns to their starting position, the dog has circled the agpole exactly the same number of times
as the owner (or, in terms of a more likely situation in real life: the dog has not managed to get the leash
tangled around the agpole!).

◦ If the owner's position at time t is f (z) while the dog's position is g(z) for all z ∈ γ , then the leash
represents the dierence f (z) − g(z), and so the condition simply says that |f (z) − g(z)| < |f (z)|: this
is Rouché's original hypothesis. (Our formulation in terms of the owner and dog walking around the
agpole are rephrasing the zero-counting in terms of winding numbers using the argument principle.)

◦ In this formulation, the more general hypothesis |f (z) − g(z)| < |f (z)| + |g(z)| states that the leash can
never intersect the agpole: although the owner and dog can be on opposite sides of the agpole, the
line joining them can never pass through the agpole itself. (Intuitively, any tangling of the leash
necessarily must start with the leash touching the agpole.)

• Example: Show that f (z) = z 4 + 2z + 5 has no zeroes inside |z| = 1 but has four zeroes inside |z| = 2.

◦ For |z| = 1 the largest contribution to f comes from the constant term 5, so we try comparing f (z) =
z 4 + 2z + 5 to g(z) = 5 with associated perturbation h(z) = f (z) − g(z) = z 4 + 2z .
4
◦ When |z| = 1 we have |h(z)| = z 4 + 2z ≤ |z| + 2 |z| ≤ 3 < 5 = |g(z)|, so the condition |h(z)| < |g(z)|
is satised.

◦ Thus by Rouché's theorem, f and g have the same number of zeroes inside |z| = 1: namely, none, since
g is nonzero everywhere.

◦ For |z| = 2 the largest contribution to f comes from the term z 4 , so we try comparing f (z) = z 4 + 2z + 5
to g(z) = z 4 with associated perturbation h(z) = 2z + 5.
◦ When |z| = 2 we have |h(z)| = |2z + 5| ≤ 2 |z| + 5 ≤ 9 < 16 = |g(z)|, so the condition |h(z)| < |g(z)| is
satised.

◦ Thus by Rouché's theorem, f and g have the same number of zeroes inside |z| = 1: namely, 4, since g
clearly has a single zero of multiplicity 4 at the origin.

• Example: Show that f (z) = z 9 − 5iz 4 + 2 has exactly ve zeroes in the annulus 1 < |z| < 2.

◦ First observe that for |z| = 1 the middle term −5iz 4 is largest, while for |z| = 2 the leading term z9 is
largest.

◦ For |z| = 1 we compare f (z) = z 9 − 5iz 4 + 2 to g(z) = −5iz 4 with associated perturbation h(z) = z 9 + 2.
9
◦ When |z| = 1 we have |h(z)| = z 9 + 2 ≤ |z| + 2 = 3 < 5 = −5iz 4 = |g(z)| so by Rouché's theorem
we see that f has the same number of zeroes as g inside |z| = 1, which is 4.

◦ Likewise, for |z| = 2 we compare f (z) to g(z) = z 9 with associated perturbation h(z) = −5iz 4 + 2.
4
◦ When |z| = 2 we have |h(z)| = −5iz 4 + 2 ≤ 5 |z| + 2 ≤ 82 < 29 = |g(z)| so by Rouché's theorem we
see that f has the same number of zeroes as g inside |z| = 2, which is 9.

◦ So since there are 9 zeroes inside |z| = 2 but only 4 inside |z| = 1, and there are no zeroes on either
circle, the remaining 5 are in the annulus 1 < |z| < 2.

• Example: Show that f (z) = 3 + z + ez has exactly one zero in the left half-plane Re(z) ≤ 0, and that it is
real.

◦ Rouché's theorem only applies on bounded regions, so in order to count the zeroes of f (z) on the
unbounded left half-plane, we instead count the number of zeroes of f in the semicircular region |z| ≤ R
with Re(z) ≤ 0 and then take R → ∞.
z Re(z)
◦ Since |e | = e ≤e is relatively small for z in the left half-plane, we can compare f (z) = 3 + z + ez
to g(z) = 3 + z with associated perturbation h(z) = f (z) − g(z) = ez .

5
◦ Then for Re(z) ≤ 0 (hence in particular on the semicircle of interest) we have |h(z)| ≤ eRe(z) ≤ e, while
|g(z)| ≥ 3 on the imaginary axis and |f (z)| ≥ |z| − 3 ≥ R − 3 on the semicircle |z| = R, so for R > 3 + e
we see that |h(z)| < |g(z)|.

◦ Thus for R > 3 + e we see that f and g have the same number of zeroes inside the semicircle, which is
clearly 1 since g has the single zero z = −3. Taking R → ∞ shows that f has exactly one zero in the
left half-plane, as desired.

◦ To see that the zero is real we simply invoke the intermediate value theorem: since f (−4) = e−4 − 1 < 0
−3 x
and f (−3) = e >0 andf (z) = 3 + x + e is continuous, by the intermediate value theorem f has a
real root in the interval (−4, −3), which by the above is the only root of f in the left half-plane.

• We can also use Rouché's theorem to give another quick proof of the fundamental theorem of algebra:

• Theorem (Fundamental Theorem of Algebra): If p(z) is a polynomial of degree d, then p(z) has d complex
roots, counted with multiplicity.

Pd−1
◦ Proof: Supposep(z) = ad z d + k=0 ak z k with ad 6= 0. For large |z| the largest contribution to p
d d
comes from the leading term ad z so we apply Rouché's theorem to f (z) = p(z) and g(z) = ad z , with
Pd−1 k
perturbation h(z) = k=0 ak z , on a suciently large circle |z| = R.
Pd−1 Pd−1
◦ Explicitly, take R > max(1, k=0 |ak | / |ad |), so that 1 < R and k=0 |ak | < R |ad |.
Pd−1 k
Pd−1 k
Pd−1 k
Pd−1 k
◦ Then for |z| = R we have |h(z)| = k=0 ak z ≤ k=0 ak z = k=0 |ak | R ≤ k=0 |ak | R ≤
Pd−1 d−1 d−1
= Rd−1 k=0 |ak | < Rd−1 |ad | R = ad z d = |f (z)|.
P
k=0 |ak | R
◦ Therefore, the condition |h(z)| < |f (z)| for Rouché's theorem is satised, and so f and g have the same
number of roots inside |z| = R, which is clearly d since g has a zero of multiplicity d at the origin.
◦ Taking R→∞ yields that f has d roots in C, as desired.

5.1.3 The Open Mapping Theorem, Local Invertibility and Local Preimages
• Using the argument principle and Rouché's theorem, we can establish another very fundamental local property
of nonconstant holomorphic functions: namely, that they are open mappings, in the sense that they map open
sets to open sets.

◦ Before proving the main result, we make some preliminary remarks about the non-obviousness of this
result: namely, that it fails completely for real-valued functions.

◦ As a simple example, observe that the innitely dierentiable real-valued function f (x) = sin x maps the
open interval (0, 6π) to the non-open set [−1, 1] (which is in fact closed).

◦ Even non-holomorphic functions from C to C need not map open sets to open sets: for example, f (z) =
2
|z| maps the open disc |z| < 1 to the interval [0, 1), which is not even open as a subset of R (much less
C), even though as a function on real and imaginary parts f (x + iy) = x2 + y 2 has partial derivatives of
all orders.

◦ In general topology, the open mapping property is used as a denition of continuity: if X Y are
and
topological spaces, then f :X →Y is continuous precisely when the inverse-image relation f −1 maps
open sets to open sets: more precisely, f is continuous precisely when for any open subset V of Y , the
set f −1 (V ) = {x ∈ X : f (x) ∈ V } is an open subset of X.
◦ Thus, the fact that nonconstant holomorphic functions map open sets to open sets is yet another reection
that their local behavior is far more pleasant than can be expected for more general functions.

◦ More explicitly, nonconstant holomorphic functions are locally onto: if f (z0 ) = w0 , then near z0 , f
takes all values suciently near w0 , in the sense that for any >0 there exists an r > 0 such that for
any w with |w − w0 | < r, there exists some z with |z − z0 | <  with f (z) = w.
◦ Indeed, this property is equivalent to saying that f maps open sets, since the image under f of the disc
|z − z0 | <  necessarily contains an open disc |w − w0 | < r around any arbitrary point w0 in the image
of f .

• We now establish the open mapping property of holomorphic functions:

6
• Theorem (Open Mapping Theorem): Suppose that U is a connected open set and f : U → C is a nonconstant
holomorphic function on U. Then f (U ) is open.

◦ Proof: Let z0 ∈ U : then since U is open it contains some open disc |z − z0 | < R. Let f (z0 ) = w0 and
set g(z) = f (z) − w0 , so that g is zero at z0 .
◦ Since f hence g is not constant on U , by our results on analytic functions since g is zero at z0 , it is
nonzero on some punctured disc 0 < |z − z0 | < r, for an appropriate positive 0 < r < R. Take γ to be
the circle |z − z0 | = r/2.
◦ Then since g is continuous and γ is closed, by the extreme value theorem |g(z)| attains its minimum M
on γ, so |g(z)| ≥ M , and since g is nonzero on γ we have M > 0.
◦ Now for any w ∈ C with |w − w0 | < M , we have |w − w0 | < M ≤ |g(z)|, so by Rouché's theorem, we see
that g(z) − (w − w0 ) has the same number of zeroes inside γ as g does.
◦ Since g(z0 ) = 0 that means g(z) − (w − w0 ) = 0 for some |z − z0 | < r/2, which is to say, for any w with
|w − w0 | < M there exists z with |z − z0 | < r/2 such that f (z) = w.
◦ This shows that the image of f is open, and so f is an open mapping as claimed.

• As noted above, the open mapping theorem states that nonconstant holomorphic functions are locally onto.
Another natural question is: are nonconstant holomorphic functions locally one-to-one?

◦ Equivalently, we are asking whether nonconstant holomorphic functions are locally invertible, in the
sense that for any z0 there exists an open set U containing z0 and an open set V containing f (z0 ) such
that f :U →V has an inverse function f −1 : V → U .
1
◦ We have previously invoked the inverse function dierentiation formula (f −1 )0 (w) = , which
f 0 (f −1 (w))
1
with w = f (z0 ) equivalently states that (f −1 )0 (f (z0 )) = , and so under the reasonable suspicion
f 0 (z0 )
that f −1 should be holomorphic, we must necessarily
0
have f (z0 ) 6= 0 in order for (f −1 )0 to be dened.
−1
◦ In fact, this condition is also sucient, and in such a case the local inverse f will also be holomorphic:

• Theorem (Local Invertibility): Suppose that f is holomorphic on an open region R. Then for any z0 ∈ R with
f 0 (z0 ) 6= 0, f is locally one-to-one near z0 , in the sense that there exists an open set U containing z0 and an
open set V containing f (z0 ) such that there exists a holomorphic function g : V → U such that f (g(v)) = v
for all v ∈ V and g(f (u)) = u for all u ∈ U .

◦ The analogous result for real-dierentiable functions is known as the inverse function theorem.

◦ Proof: By translating in the domain and translating and rescaling f , we may assume that z0 = 0,
P∞
f (z0 ) = 0 and f 0 (z0 ) = 1, so that f (z) has a series expansion f (z) = z + z 2 q(z) where q(z) = n=2 an z n .
◦ Then by our results on local series expansions, q is holomorphic near 0 hence is bounded on some disc
|z| < R: say with |q(z)| ≤ M on this disc.

◦ Now take r < 1/(2M ), select any ζ with |ζ| < r/2, and consider fα (z) = f (z) − α. We perturb fα by
h(z) = z 2 q(z) − α to compare fα to g(z) = z .
◦ For |z| = r, we have |h(z)| ≤ z 2 q(z) + |α| ≤ r2 M + r/2 < r = |z| = |g(z)|, so the hypothesis of Rouché's
theorem is satised.

◦ Therefore since g has a unique zero of order 1 inside |z| < r, so does fα .
◦ This means f (z) = α has a unique solution inside |z| < r for any |α| < r/2.
◦ So now dene U = {z : |z| < r and |f (z)| < r/2}. This set is open because it is the intersection of the
open disc |z| < r with the open set |f (z)| < r/2 (which is open because f is continuous).
◦ We have just shown that f maps U bijectively onto the open set V = {z : |z| < r/2}. Since f is an open
mapping by the open mapping theorem, the inverse bijection g : V → U is continuous.
◦ Next, since g is continuous, we may apply the inverse function dierentiation formula to see that g is
0 1 1
dierentiable at z0 and that g (0) = 0 = 0 .
f (g(0)) f (0)

7
◦ Finally, since f 0 (z0 ) 6= 0, since f 0 is continuous all of this discussion also applies to all points in some open
disc containing z0 , so in fact g is dierentiable on an open disc around z0 , meaning it is holomorphic.

• Let us now examine more closely the situation where f (z) is not locally invertible at z = z0 .

◦ For f (z) = z 2 , we see that f is not locally invertible at z = 0, since f 0 (0) = 0. Indeed, we can see directly
that f is not one-to-one on any disc |z| < r centered at the origin: it is in fact two-to-one, since each
2
point other than z = 0 in the image disc |z| < r has exactly 2 preimages in the disc |z| < r .

◦ In the same way we can see that f (z) = z d is d-to-one inside any disc |z| < r, since f (z) = f (ze2πik/d )
for each 0 ≤ k ≤ d, so each point other than z = 0 in the image disc |z| < rd has d preimages in the disc
|z| < r.
◦ It may seem that this local behavior is determined by the order of the zero of f (z) at z = z0 , but the
function f (z) = z 2 + 1 is also two-to-one on any disc |z| < r, for the same reason: shifting the image of
the function does not aect the number of preimages.

◦ Indeed, if we replace f (z) by f (z) − f (z0 ), then the local behavior appears to be controlled by the order
of the zero of f (z) − f (z0 ) at z0 .
◦ This is in fact quite sensible if we look at the series expansion for
P∞ f : if f (z0 ) = w0 , and f (z0 ) − w0 has a
zero of order d at z0 , then f (z0 ) = w0 + ad (z − z0 )d + n=d+1 an (z − z0 )n , and so for small |z − z0 | we
d d+1
have f (z0 ) − w0 = ad (z − z0 ) + O(|z − z0 | ), and the O-term will be a suciently small perturbation
that f (z) will still be locally d-to-one near z0 .

◦ As with the local invertibility theorem, we can use Rouché's theorem to make the argument precise.

• Denition: If f is holomorphic and nonconstant on R and z0 ∈ R, the multiplicity of z0 is the order of the
zero of f (z) − f (z0 ) at z0 . A point of multiplicity 1 is called simple.

◦ We observe that points of multiplicity greater than 1 are the same as the zeroes of f 0 (z), and are thus
isolated under the assumption that f is nonconstant.

• Theorem (Local Preimages): Suppose that f is holomorphic and nonconstant on a connected open region R.
For any z0 ∈ R, if z0 is a point of multiplicity d for f, then there exist positive  and r such that each w
with 0 < |α − f (z0 )| <  has exactly d distinct preimages inside the disc |z − z0 | < r, and each preimage is a
simple point.

◦ The proof is essentially the same as the local invertibility theorem, aside from making a minor adjustment
to avoid points of higher multiplicity.

◦ Proof: If d = 1 the result follows immediately from the local invertibility theorem, so otherwise assume
d > 1. As in the local invertibility theorem we may translate and rescale to assume z0 = 0, f (z0 ) = 0,
(d)
and f (z0 ) = 1. Then f (z) = z d + z d+1 q(z) for some holomorphic q(z) with |q(z)| ≤ M for |z| < R1 .
◦ As noted above, since the points of f of multiplicity > 1 are isolated, we may select R such that the only
point of multiplicity >1 for f inside |z| < R2 is z = 0 itself.
◦ Then take r < min(R1 , 1/(2M ), R2 ), select any ζ with |ζ| < rd /2, and consider fα (z) = f (z) − α. We
perturb fα by h(z) = z d+1 q(z) − α to compare fα to g(z) = z d .
◦ For |z| = r, we have |h(z)| ≤ z d+1 q(z) + |α| ≤ rd+1 M + r/2 < rd = z d = |g(z)|, so the hypothesis of
Rouché's theorem is satised.

◦ Therefore since g has d total zeroes inside |z| < r, so does fα .


◦ This means f (z) = α has a total of d solutions (with multiplicity) |z| < r for any |α| < r/2. But since
the only points with 0 < |z| < r < R2 are simple points, there are exactly d solutions to f (z) = α and
they are all simple.

• We will remark that the branch points (or ramied points), which as we have seen are simply the zeroes
of f 0 (z), carry substantial geometric information about the behavior of the function f.
• We nish by noting the following useful property about images of boundaries under holomorphic maps.

◦ Recall that if R is a region, then ∂R denotes the boundary of R.

8
• Theorem (Images of Boundaries): Suppose R is a bounded region and f : R → C is a nonconstant holomorphic
function. If U is an open subset of R, then ∂f (U ) ⊆ f (∂U ). Furthermore, if f is one-to-one (equivalently, if
f0 is nonzero on U ), then in fact ∂f (U ) = f (∂U ).

◦ We also remark here that if U is connected, then f (U ) is connected since f is continuous.

◦ Proof: Since U is open, by the open mapping theorem, we see that f (U ) is open.

◦ Also let U = U ∪ ∂U be the closure of U : then U is closed and bounded, so since f is continuous, the
closure f (U ) is equal to the set f (U ). (Explicitly, the elements in f (U ) are limits of the form limn→∞ f (zi )
for zi ∈ U , which by passing f through the limit are the same as limits of the form f (limn→∞ zi ), and
these are elements of f (U ).)

◦ Then the boundary ∂f (U ) = f (U )\f (U ) = f (U )\f (U ) ⊆ f (U \U ) = f (∂U ), as claimed.

◦ If f is one-to-one then necessarily f (U ) and f (∂U ) are disjoint, and so since f (∂U ) ⊆ f (U ) = f (U ), we
see f (∂U ) ⊆ f (U )\f (U ) = ∂f (U ). This is the reverse containment so ∂f (U ) = f (∂U ).

• We remark that the condition that f be one-to-one is necessary to have ∂f (U ) = f (∂U ) in the theorem above.

◦ Specically, take U to be the open semidisc |z| < 1 with Im(z) > 0 and f (z) = z 4 . Then f (U ) is the
punctured disc 0 < |z| < 1 while ∂U is the half-circle |z| = 1 with Im(z) ≥ 0 along with the interval
[−1, 1] on the real line.

◦ Then f (∂U ) is the circle |z| = 1 along with the interval [0, 1] on the real line, but ∂f (U ) is only the circle
|z| = 1 along with the point 0.

5.2 Conformal Mapping

• Now that we have established various local properties of holomorphic functions, we broaden our discussion to
encompass more global properties.

◦ Our main goal is to study holomorphic functions f :U →V that are both one-to-one and onto.

◦ Such functions are variously called analytic isomorphisms (reecting the fact that they are structure-
preserving analytic maps), biholomorphic functions (reecting the fact that they are holomorphic func-
tions having a holomorphic inverse), or conformal mappings (reecting the fact that they have the
conformal angle-preserving property we discussed long ago).

◦ The main utility of having such a map from U to V is that they allow us to transfer holomorphic functions
from U to V (and vice versa) via composition: thus, U and V behave essentially equivalently from the
perspective of complex analysis, and so understanding behavior on U allows us to understand behavior
on V.
◦ We will begin by studying the class of conformal mappings on the entire complex plane: these are the
fractional linear transformations. So as to present a more unied discussion, we rst introduce the notion
of the point at innity and the extended complex plane in the context of projective geometry.

◦ We then discuss various ways to construct conformal mappings among various sets, with a particular
emphasis on conformal maps involving the unit disc, whose behavior is rather natural to study.

5.2.1 The Point at Innity and the Extended Complex Plane


• For real-valued functions, we have the useful notion of a function diverging to +∞ or −∞ at a point, which
we can formalize using variations on the -δ denition of limit.

◦ Rather than bothering with the details, we will just observe that, for example, 1/x2 → +∞ and −1/x2 →
−∞ as x → 0.
◦ We also have the related notion of a function having a limit (or diverging) as x → +∞ or x → −∞: for
example, we have ex → +∞ as x → +∞ and ex → 0 as x → −∞.

• For complex-valued functions, there are many more ways for a function to diverge to ∞.

9
◦ For example, the function f (z) = 1/ |z| tends uniformly to +∞ (positive real innity) as z → 0, whereas
g(z) = i/ |z| tends uniformly to +i∞ (imaginary innity), in the sense that 1/ |z| grows large along the
positive real axis and i/ |z| grows large along the positive imaginary axis.
◦ Like with real-valued functions we could make precise all of the dierent directions in which a function
can diverge to ∞, but in fact for holomorphic functions, all of these directions can only occur together.

◦ As an illustration, consider f (z) = 1/z as z → 0. If we select dierent directions of approach (e.g., along
the path γ(t) = t · e−iθ as t → 0 for xed θ) then f (z) will approach ∞ along the ray corresponding to
the polar angle θ .

◦ So in this sense, we can simply write 1/z → ∞ as z → 0, with the understanding that 1/z approaches
∞ in every possible direction as z → 0.
◦ Likewise, even though there are many possible limits we could evaluate for a meromorphic function f (z)
as  z → ∞ (along the positive real axis, along the negative imaginary axis, along the polar ray θ = π/4,
along the curve γ(t) = t + t2 i as t → ∞, etc.), all of these limits will in fact display the same behavior:
either the function f is bounded and has a limit as |z| → ∞, the function is unbounded and tends to ∞
in every direction, or the function is bounded along some paths and unbounded along others.

◦ This observation follows by considering the nature of the singularity of f (1/z) at z = 0, since as |z| → ∞
we have 1/z → 0, the behavior of f (z) as |z| → ∞ is the same as that of f (1/z) as z → 0.
◦ Thus, if f (1/z) has a removable singularity at z = 0 then f (z) will have a limit as z → ∞, if f (1/z) has
a pole at z = 0 then f (z) will tend to ∞ (uniformly) as z → 0, and if f (1/z) has an essential singularity
at z = 0 then f (z) will be bounded along some paths and unbounded along others.

• Let us record these observations:

• Denition: Suppose f is holomorphic on the region |z| > R for some R. Then there are three types of behavior
as z→∞ based on the singularity of f (1/z) at z = 0.

1. If f (1/z) has a removable singularity at z = 0, we say f is holomorphic at ∞. In this case f (z) tends to
a limit L as |z| → ∞.
2. If f (1/z) has a pole at z = 0, we say f has a pole at ∞. In this case |f (z)| → ∞ uniformly as |z| → ∞.
3. If f (1/z) has an essential singularity at z = 0, we say f has an essential singularity at ∞. In this case
|f (z)| is bounded along some paths but unbounded along others as |z| → ∞. In fact, by Picard's little
theorem, for any R > 0, on the region |z| > R the function f (z) will take all values in C with at most
one exception.

◦ Example: For f (z) = z/(z − 1), we have f (1/z) = 1/(1 − z) which is holomorphic at z = 0. Thus, f (z)
is holomorphic at ∞, and f (z) → 1 (the value of f (1/z) at z = 0) as z → ∞.
◦ Example: For f (z) = z 2 , we have f (1/z) = 1/z 2 which has a pole at z = 0. Thus, f (z) has a pole at ∞,
2
and |f (z)| → ∞ as z → ∞ (which here is obvious, since |f (z)| = |z| ).
◦ Example: For f (z) = ez ,f (1/z) = e1/z which has an essential singularity at z = 0. Thus, f (z)
we have
has an essential singularity at ∞, and f (z) displays erratic behavior as z → ∞ along dierent paths.
Indeed, along the positive real axis we have f (z) → +∞ while along the negative real axis we have
f (z) → 0 and along the imaginary axis the values of f simply circulate around the unit circle (so f is
bounded, indeed periodic, but does not approach a limit).

◦ As with other singularities, we can classify the nature of the singularity at ∞ using the Laurent expansion
of f.
P∞ P∞ P∞
◦ Explicitly, if f (z) = n=−∞ an z n on |z| > R, then f (1/z) = n=−∞ an z −n = n=−∞ a−n z n .
◦ Thus, if there are no terms a−n with n < 0 (in other words, no terms an with n > 0) then f has a
removable singularity at ∞, if there are only nitely many a−n with n < 0 (i.e., nitely many an with
n > 0) then f has a pole at ∞, and if there are innitely many a−n with n < 0 (i.e., innitely many an
with n > 0) then f has an essential singularity at ∞.
◦ As an immediate consequence, we see that if f is entire, then f is holomorphic at ∞ if and only if f is
constant, f has a pole at ∞ if and only if f is a polynomial, and otherwise (if f is not a polynomial)
then f has an essential singularity at ∞.

10
• We can make our discussion a bit more precise by including the point ∞ explicitly in our discussion. Intuitively,
we glue the innite extent of C into a single point at innity, and extend the topology of C to include the
point ∞ using the map z 7→ 1/z :
• Denition: The extended complex plane is the set C ∪ {∞}. A subset U of C ∪ {∞} is dened to be open if
(i) U ∩C is open in C and (ii) if ∞∈U then U ∩ C must also contain the set |z| > R for some R.

◦ The idea is that we dene the open discs around ∞ to be ∞ along with the sets |z| > R, which are the
images of the open discs |z| < 1/R under the map z 7→ 1/z . Then a subset U of C ∪ {∞} is open if and
only if it contains an open disc around each of its points.

◦ By construction, the map z 7→ 1/z on C ∪ {∞} is continuous, and so since this map is its own inverse,
it is in fact a homeomorphism of the extended complex plane.

◦ The extended complex plane is a special case of the topological construction known as one-point com-
pactication (which takes a topological space X and makes it compact by adding one extra point, whose
open neighborhoods are selected in a way that closes up all of the non-compact closed sets).

• Viewing ∞ as an actual point in the extended complex plane also claries why meromorphic functions are in
fact quite natural to consider, while essential singularities are less natural.

◦ Explicitly, if f is holomorphic on a region R\{z0 } and has a pole at z0 , we can now dene f (z0 ) = ∞.
◦ Since f (z) → ∞ uniformly as z approaches the pole z0 , the function f : R → C ∪ {0} is now actually
continuous at z0 .
◦ So we see that a pole is just a point where the (now everywhere continuous) function f takes the value
∞.
◦ Hence, the meromorphic functions on R are simply the functions that are continuous everywhere (as
functions from R to C ∪ {∞} and are holomorphic except at an isolated set of points).

◦ We can also dene what it means for a function to be meromorphic at ∞: we simply require that f be
holomorphic or have a pole at ∞, and also that the singularity at ∞ be isolated (meaning that there is
some R for which there is no pole of f in the open disc |z| > R around ∞).

• There is an appealing geometric description of the extended complex plane using stereographic projection on
the Riemann sphere.

◦ Explicitly, consider the unit sphere x2 + y 2 + z 2 = 1 in 3-dimensional real space. For any point P = z =
x + iy = (x, y, 0) in the xy -plane, let l be the line joining (x, y, 0) to the north pole (0, 0, 1) of the sphere,
and consider the intersection of l with the sphere.

◦ Explicitly, we can see that l l(t) = t hx, y, 0i + (1 − t) h0, 0, 1i = htx, ty, 1 − ti, which
is parametrized as
2 2 2 2 2 2
intersects the sphere when t x + t y + (1 − t) = 1, yielding t =
2 . The point of intersection
1 + |z|
2
2Re(z) 2Im(z) 1 − |z|
itself is then π(z) = ( , ,
2 2 2 ).
1 + |z| 1 + |z| 1 + |z|
◦ We can see that this map is continuous, and as |z| → ∞ approaches (0, 0, 1) uniformly, so it is also
one-to-one and onto. Additionally, the inverse function is also continuous (it is given explicitly by
−1 x0 + iy0
π (x0 , y0 , z0 ) = for z0 6= 1 with π −1 (0, 0, 1) = ∞), so π gives a homeomorphism of the
1 − z0
extended complex plane with the unit sphere.

◦ The point at ∞ corresponds to the north pole, and all of the nite points in C form the rest of the sphere:
the equator (the great circle lying in the xy -plane) corresponds to the unit circle, while the northern
hemisphere corresponds to the exterior of the unit disc and the southern hemisphere corresponds to the
interior of the unit disc.

• Another geometric description of the extended complex plane arises naturally from projective geometry.

◦ Explicitly, we dene the complex projective line P1 (C) to be the set of lines through the origin in C2 .
1
More explicitly, dene P (C) to be the set of points [z0 : z1 ] with z0 , z1 ∈ C not both zero, where P ∼ Q
if P = λQ for some nonzero λ ∈ C.

11
◦ We use the notation [z0 : z1 ] to evoke the idea of considering only the ratios between the coordinates,
since for example we consider the points [1 : 1] and [2 : 2] to be the same since [1 : 1] = 21 [2 : 2].
◦ When z1 6= 0 we see that [z0 : z1 ] = [z0 /z1 : 1] whereas when z1 = 0 (in which case z0 6= 0 since both
coordinates cannot be zero) we have [z0 : z1 ] = [z0 : 0] = [1 : 0].
◦ Therefore, if we identify the point [α : 1] with α ∈ C and [1 : 0] with ∞, we see that the extended
complex plane corresponds with the points on the complex projective line P1 (C).
◦ Intuitively, the points on the projective line simply label the possible slopes of lines in C2 : all elements
from C along with ∞ (for vertical lines).

◦ Furthermore, P1 (C) inherits a natural quotient topology from C2 \{0}, and one may check directly that
this topology agrees with the topology we dened above on the extended complex plane.

◦ The projective line (or more generally, projective n-space over an arbitrary eld F, which is dened to
be the set of lines through the origin in F n+1 ) arises naturally in classical geometry.

◦ Geometrically speaking, parallel lines in C, when extended to the projective line P1 (C), will intersect at
the point at ∞.
◦ Because of the entirely uniform treatment of the point at ∞ as being entirely equivalent to any other
point (rather than our approach above that requires special treatment of ∞, e.g., in classifying the
behavior of the singularity there), many results have more natural statements when viewed projectively.

5.2.2 Fractional Linear Transformations


• We now study holomorphic bijections on various regions. To begin, we classify the holomorphic bijections
from the extended complex plane to itself (one may reasonably view these as meromorphic automorphisms
of the extended complex plane).

• We begin by showing that the only meromorphic functions on the extended complex plane are rational
functions:

• Proposition (Meromorphic Functions on C ∪ {∞}): A function f : C ∪ {∞} → C ∪ {∞} is meromorphic


everywhere if and only if it is a rational function of z.

◦ Proof: Clearly each rational function is meromorphic on the extended complex plane, since any rational
function has only nitely many poles (including potentially at ∞).
◦ Conversely, suppose f is meromorphic on C ∪ {∞}. Then f has no poles in |z| > R for some R, and also
f can only have nitely many poles in |z| ≤ R since otherwise the poles could not be isolated (as any
innite bounded sequence in C has a convergent subsequence by the Heine-Borel theorem).

◦ Hence f has only nitely many poles in C, so we may remove all of them by multiplying by an appropriate
polynomial q(z).
P∞ n
◦ Then q(z) · f (z) = n=0 an z can have only nitely many nonzero coecients an since q(z)f (z) cannot
have an essential singularity at ∞. Therefore q(z)f (z) = p(z) is a polynomial so that f (z) = p(z)/q(z)
is a rational function, as claimed.

• Now we can classify the meromorphic bijections of the extended complex plane using our characterization of
one-to-one maps:

• Proposition (Meromorphic Bijections on C ∪ {∞}): A function f : C ∪ {∞} → C ∪ {∞} is meromorphic and


az + b
one-to-one if and only if it is a nonconstant function of the form f (z) = for some a, b, c, d ∈ C.
cz + d
az + b
◦ Proof: Each nonconstant function f (z) = is meromorphic everywhere and has an inverse function
cz + d
dz − b
f −1 (z) = by direct calculation, so each of these functions are meromorphic bijections.
−cz + a
◦ Conversely, by the proposition above, if f is meromorphic everywhere then f must be a rational function
p(z)
of the form for some polynomials p and q.
q(z)

12
◦ Now observe that this function can have at most one pole, since any pole z0 is a location where f (z0 ) = ∞
and f is a bijection. If the pole is at ∞ then replacing f (z) with f (1/z) yields a function with a pole at
0, so we may assume the pole is at a nite point z0 .
p(z)
◦ Then we must have f (z) = for some p(z) with p(z0 ) 6= 0. Now if f (z) is one-to-one then
(z − z0 )k
1 (z − z0 )k
= is also one-to-one, but since this function is locally k -to-one near z0 by our results, we
f (z) p(z)
must have k = 1.
1
◦ By applying the same argument to , we see that p(z) must also have degree 1, and so f (z) is a
f (z)
quotient of linear polynomials as claimed (and obviously it must be nonconstant in order to be one-to-
one).

• These quotients of linear polynomials are quite fundamental and have many interesting properties:

az + b
• Denition: The fractional linear transformations are the nonconstant functions of the form f (z) = for
cz + d
a, b, c, d ∈ C. They are all bijections from the extended complex plane to itself.

◦ We can see easily from the description that the unique zero of f is at −b/a while the unique pole is at
−d/c.
◦ It is also easy to see that any fractional linear transformation can be obtained as a composition of scalings
z 7→ az (which are in turn obtained as a composition of a dilation z 7→ rz and a rotation z 7→ eiθ z ),
translations z 7→ z + b, and inversions z 7→ 1/z .
◦ By basic linear algebra the function f (z) is nonconstant whenever the coecient vectors ha, bi and hc, di
are linearly independent in C2 , which is equivalent to saying that the determinant ad − bc is nonzero.
◦ The fact that there is an obvious linear algebra condition arising here should raise some suspicion that
linear algebra is more involved than may immediately be apparent.

• In fact, the fractional linear transformations arise very naturally, as actual linear transformations, when we
work with the complex projective line P1 (C).

◦ Explicitly, any linear transformation T : C2 → C 2 is of the form T (z0 , z1 ) = (az0 + bz1 , cz0 + dz1 ) for
some a, b, c, d ∈ C.
◦ These transformations naturally descend to linear transformations on P1 (C) via setting T [z0 : z1 ] =
[az0 + bz1 : cz0 + dz1 ], and they are well-dened on projective points since they are homogeneous:
T [λz0 : λz1 ] = [λ(az0 + bz1 ) : λ(cz0 + dz1 )] = [az0 + bz1 : cz0 + dz1 ] = T [z0 : z1 ] for any λ 6= 0.
◦ The only concern is that we may attempt to map a point [z0 : z1 ] to an element not in P1 (C), which
would require the coordinates az0 + bz1 and cz0 + dz1 to be zero simultaneously for some (z0 , z1 ) 6= 0.
    
a b z0 0
◦ Thus we need to avoid the situation where the linear system = has a nonzero
c d z1  0
a b
solution, which is equivalent to requiring the determinant ad − bc of be nonzero.
c d
◦ Therefore, the linear transformations T (z0 , z1 ) = (az0 + bz1 , cz0 + dz1 ) of C2 that yield well-dened maps
1
on P (C) are precisely the ones with ad − bc 6= 0.
◦ Finally, when we make the identication of the projective point [z0 : z1 ] = [z0 /z1 : 1] with the complex
az0 + bz1 az + b
point z = z0 /z1 , since T [z0 : z1 ] = [az0 +bz1 : cz0 +dz1 ] = [ : 1] = [ : 1], the corresponding
cz0 + dz1 cz + d
az + b
complex map is simply T (z) = , our fractional linear transformation.
cz + d
az + b
◦ Indeed, the fact that T (z) = arises as an actual linear transformation on the complex projective
cz + d
line is precisely the reason that T is called a fractional linear transformation.

◦ Using the projective approach, however, gives us much more.

13
◦ For example, we can see immediately that the composition of fractional linear transformations is precisely
the same as composition of invertible linear transformations (namely, via matrix multiplication).
 
az + b ãz + b̃ Az + B A B
◦ Explicitly, if T (z) = and T̃ (z) = , then (T ◦ T̃ )(z) = where =
 0  cz + d ˜ + d˜
cz Cz + D C D
a b0

a b
.
c d c0 d0
 
az + b a b −1
◦ Likewise, for T (z) = corresponding to , the inverse transformation T corresponds to
cz + d c d
 −1  
a b 1 d −b −1 (dz − b)/(ad − bc) dz − b
the inverse matrix = hence T = = .
c d ad − bc −c a (−cz + a)/(ad − bc) −cz + a
◦ Furthermore, we also see that the composition of fractional linear transformations has a group structure,
since composition of linear transformations (equivalently, multiplication of invertible matrices) also has
a group structure.

◦ Specically, the group of fractional linear transformations is isomorphic to the group GL2 (C) of invertible
2×2 matrices, modulo the subgroup that reduces to the identity transformation on C ∪ {∞}, which one
may check directly is simply the nonzero diagonal matrices.

• The fractional linear transformations also have an extremely useful geometric property: namely, they preserve
the class of generalized circles: circles together with lines (where we view a line as being a circle with radius
∞).

◦ One may verify this directly for each of the simpler maps that can be composed to obtain a general
fractional linear transformation: it is obvious that scalings and translation map circles to circles and
lines to lines, and for the inversion map one may just compute directly the image of the circle |z − α| = r
under the inversion map.

◦ Using the Riemann sphere model, one may see that lines in C correspond to circles on the Riemann
sphere passing through the north pole of the sphere while circles in C correspond to circles on the sphere
not passing through the north pole (these calculations follow from some tedious uses of the explicit
coordinate conversions between the Riemann sphere and the extended plane).

◦ However, we can also obtain this result in a more conceptually clean way using the projective approach.

◦ Explicitly, the circle |z − α| = r for α ∈ C and r > 0 has equation (z − α)(z − α) = r2 which yields
2
zz − αz − αz + (|α| − r2 ) = 0, which in projective coordinates z = z0 /z1 has the form Az0 z0 + Bz0 z1 +
2
Bz0 z1 + Cz1 z 1 = 0 for A = 1, B = −α, and C = |α| − r2 .
◦ The line ax + by = c for reala, b, c also has equation (a + bi)z + (a − bi)z = 2c which in projective
coordinates also has the form Az0 z0 + Bz0 z1 + Bz0 z1 + Cz1 z 1 = 0 with A = 0, B = a + bi, and C = 2c.
◦ Therefore we see that the generalized circles are precisely the curves with projective equation Az0 z0 +
Bz0 z1 + Bz0 z1 + Cz1 z 1 = 0 for some real A and C and some complex B (not all zero): lines have A=0
while circles have A 6= 0.
  
  A B z0
◦ The equation Az0 z0 +Bz0 z1 +Bz0 z1 +Cz1 z 1 = 0 can be written in matrix form as z0 z1 =
B C z1
 
z0
0, or more compactly as z† M z = 0 where z = , the dagger represents the adjoint (conjugate trans-
z1

pose), and the matrix M has M = M (i.e., M is Hermitian).

◦ If we then apply a fractional linear transformation T to z , which corresponds to applying a left matrix
multiplication to z, the new equation is (T z)† M (T z) = 0 which is equivalent to z† (T † M T )z = 0: this

has the same form, except the new coecient matrix is T M T (which is still Hermitian).

◦ Therefore, we see in particular that applying a fractional linear transformation to a generalized circle
yields another generalized circle.

• As an actual practical matter, to compute a fractional linear transformation taking one generalized circle to
another, it is usually easiest to compose various simpler maps such as scalings, translations, and inversions.

14
◦ In particular, since the image of any generalized circle is another generalized circle, we only need to
nd the images of a few points to identify the image of any generalized circle under a fractional linear
transformation.

◦ Of particular note is that a fractional linear transformation only maps one point to ∞ (namely, its pole),
and so only the generalized circles passing through the pole will be mapped to lines; the others will be
mapped to circles.

z
• Example: Find the images of |z − 1| = 1, |z| = 2, and |z| = 1 under the map f (z) = .
z−2
◦ The pole of f (z) is at z = 2, which lies on |z − 1| = 1. Thus this circle will be mapped to a line.

◦ To identify which line we just nd two points on it. For example, both z = 0 and z = 1+i lie on
|z − 1| = 1, so f (0) = 0 and f (1 + i) = −i will lie on the image of |z − 1| = 1. So the desired line is

simply the imaginary axis Re(z) = 0 .


◦ Likewise, the pole of f (z) also lies on |z| = 2 so it is also mapped to a line. Both z = 2i and z = −2i lie
1−i 1+i 1
on the circle, so f (2i) = and f (−2i) = both lie on the line. So the desired line is Re(z) = .
2 2 2
◦ The circle |z| = 1 does not contain the pole of f so it is mapped to another circle. This circle will
1 1 − 2i
contain the points f (1) = −1, f (−1) = , and f (i) = . Using some basic geometry to construct
3 5
1 2
the circumcenter of this triangle shows that the circumcenter is − and has radius , so the circle is
3 3
1 2
z+ = .
3 3

• Example: Find a fractional linear transformation mapping |z| = 1 to |z − 2i| = 3.

◦ The original circle has radius 1 and the new circle has radius 3, so rst we want to scale by a factor of
3. Then we just need to translate the original center 0 to the new center 2i.
◦ This yields the linear map T (z) = 3z + 2i .

• Example: Find a fractional linear transformation mapping |z| = 1 to the real axis.

◦ Here we want to map a circle to a line, so the pole of the fractional linear transformation needs to be
on the circle. Since we may choose arbitrarily, let us take the pole at z = 1, in which case we want
az + b
T (z) = for some a, b.
z−1
◦ Since the image is determined by the images of any two points on|z| = 1, let us try sending z = −1 to
a = b, and then try to map z = i to a real number.
0, which requires

a(i − 1)
◦ Since T (i) = = a, we can choose any real number a that makes T nonconstant (i.e., a 6= 0). So
i−1
z+1
for example T (z) = will map |z| = 1 to the real axis.
z−1

• In fact, since one-to-one holomorphic functions map boundaries to boundaries, we can even identify the images
of regions under fractional linear transformations, as long as the region's boundary consists of generalized
circles.

◦ To identify which of the various possible regions having a particular generalized circle as its boundary is
the actual image f (R), we can simply calculate f (z) for some z in the interior of R: then f (z) must lie
inside f (R).
◦ Furthermore, since holomorphic functions are continuous, they preserve connectedness of regions, so if
the original region is connected, so is the new region.

z
• Example: Find the images of |z − 1| ≤ 1, |z| > 2, and 1 < |z| < 2 under the map f (z) = .
z−2

15
◦ We calculated previously that |z − 1| = 1 maps to the line Re(z) = 0, so the closed disc |z − 1| ≤ 1 must
map either to the upper half-plane or the lower half-plane, since these are the only two connected regions
having the real line as their boundary.

◦ To identify which one it is we can observe that the center of the circle z =1 is in the interior of the
region, so f (1) = −1 must lie in the interior of the image. Thus the image of |z − 1| ≤ 1 is the lower

half-plane Re(z) ≤ 0 .

◦ Likewise, |z| = 2 maps to the line Re(z) = 1/2 so the open region |z| > 2 maps either to the region above
or below the line. Since z = 3 lies in the original region, f (3) = 3 lies in the image, so the image is the
upper half-plane Re(z) ≥ 1/2 .

1 2
◦ Finally, since f maps |z| = 1 to the circle z+ = and f maps |z| = 2 to the line Re(z) = 1/2, it
3 3
must map the region 1 < |z| < 2 to a region bordering both the circle and the line, and there is only
1 2
one possibility: the region outside the circle and to the left of the line: z+ > and Re(z) < 1/2.
3 3
Indeed, one can check that z = 3/2 with f (z) = −3 lies in this region, as expected.

• There are many other interesting properties of fractional linear transformations acting on the extended complex
plane. For example, their action is strictly 3-transitive, in the following sense:

• Proposition (3-Transitivity of FLTs): Suppose z0 , z1 , z∞ are three distinct elements of the extended complex
plane. Then there exists a unique fractional linear transformation T such that T (z0 ) = 0, T (z1 ) = 1, and
T (z∞ ) = ∞. More generally, if w0 , w1 , w∞ are any three distinct elements of the extended complex plane, then
there exists a unique fractional linear transformation T such that T (z0 ) = w0 , T (z1 ) = w1 , and T (z∞ ) = w∞ .

◦ Proof: Suppose that T is a fractional linear transformation such that T (z0 ) = 0, T (z1 ) = 1, and
T (z∞ ) = ∞.
◦ Then z0 T and z∞ is the unique pole of T , which determines T up to a scaling factor:
is the unique zero of
z − z0
T (z) = a
explicitly, we must have , where we interpret this equation as saying T (z) = a(z − z0 ) if
z − z∞
a
z∞ = ∞ and T (z) = if z0 = ∞.
z − z∞
◦ Then the constant a is uniquely characterized by the condition T (z1 ) = 1, and so T exists and is unique,
as claimed.

◦ For the second part, suppose S(z0 ) = w0 , S(z1 ) = w1 , and S(z∞ ) = w∞ . Then by the above there exists
a unique Tw with Tw (w0 ) = 0, Tw (w1 ) = 1, Tw (w∞ ) = ∞. Then Tw ◦ S maps z0 to 0, z1 to 1, and z∞
to ∞.
◦ Since (again by the above) there is a unique Tz with Tz (z0 ) = 0, Tz (z1 ) = 1, Tz (z∞ ) = ∞, we have
Tw ◦ S = Tz and so S must equal Tw−1 ◦ Tz . But since Tw−1 ◦ Tz has the desired property, we see S exists
and is unique.

◦ Remark: One may check more explicitly that the unique fractional linear transformation with T (z0 ) = 0,
z − z0 z1 − z∞
T (z1 ) = 1, and T (z∞ ) = ∞ is given by the cross ratio T (z) = .
z − z∞ z1 − z0

5.2.3 Conformal Maps and Analytic Isomorphisms


• Now that we have studied fractional linear transformations, which allow us to construct maps among regions
bounded by lines and circles, we investigate how to use other holomorphic functions to solve conformal mapping
problems.

◦ If U is an open region, the classical terminology is to say that f :U →C is a conformal map when f
preserves angles.

◦ As we showed during our original discussion of holomorphic functions, f preserves angles if and only if
f is holomorphic on U and f0 is everywhere nonzero on U.

16
◦ Per our results on local invertibility, the condition that f0 is nonzero on U is equivalent to saying that f
is locally one-to-one on U.
◦ If f is in fact globally one-to-one on U , then if we dene V = f (U ), which is also an open set by the open
mapping theorem, then the inverse function f −1 : V → U is also holomorphic, and f is a homeomorphism
from U to V.
◦ In this situation we say that f is biholomorphic (holomorphic with holomorphic inverse), or equivalently
that f is an analytic isomorphism between U and V.
0
◦ We remark that local invertibility (namely, that f is nonvanishing) is not equivalent to global invertibility:
for example, the exponential function f (z) = ez is locally invertible but also periodic hence not globally
invertible on C (or indeed on any set with two points that dier by an integer multiple of 2πi).

◦ Thus, analytic isomorphisms always preserve angles, but angle-preserving maps need not be analytic
isomorphisms.

• As a practical matter, we are interested specically in analytic isomorphisms (which we will refer to as
conformal maps, with the tacit insistence that these actually be bijections) since they have better properties.

◦ For example, the composition f ◦g of any two analytic isomorphisms f :V →W and g:U →V is also
an analytic isomorphism (its inverse is g −1 ◦ f −1 ), as is the inverse f −1 : V → U .
◦ Thus in particular, for any open set U, the set of analytic isomorphisms f :U →U forms a group under
composition (such maps are analytic automorphisms of U ).
◦ We say that two open sets U and V are conformally equivalent if there exists an analytic isomorphism
f : U → V. The observations above (along with the trivial observation that the identity function is
an isomorphism of any open set with itself ) show that conformal equivalence is indeed an equivalence
relation.

◦ The main utility of conformal equivalence is that it allows us to transfer questions involving holomorphic
functions between U and V by composing appropriately with f or f −1 .
◦ For example, we can see that if f : U → V and g : U → V are both analytic isomorphisms, then there
exists a unique analytic automorphism h : V → V such that g = h ◦ f : namely, h = g ◦ f −1 .
◦ Likewise, if f : U → V is an analytic isomorphism, then there is a bijection between analytic auto-
morphisms of U and analytic automorphisms of V given by ϕ 7→ f ◦ ϕ ◦ f −1 (i.e., via conjugation by
f ).
◦ Thus, if we want to understand the behavior of analytic isomorphisms involving a given set U, it suf-
ces to construct an analytic isomorphism of U with a xed set V, and then characterize the analytic
automorphisms of V.

• A particular general case is the situation where U is simply connected. There turn out to be two distinct
cases:

• Proposition (Analytic Isomorphisms of C): Suppose f :C→V is an analytic isomorphism. Then V =C and
f (z) = az + b is a nonconstant linear polynomial; conversely all such polynomials are analytic isomorphisms.

◦ Proof: Suppose f :C→V is a holomorphic bijection. Then f and f −1 map boundaries to boundaries,
so since C has no boundary points, neither does V.
◦ But the only nonempty region in C with no boundary points is C itself, so V = C.
◦ Since f is nonconstant, by Liouville's theorem f must be unbounded as z→∞ so f cannot be holomor-
phic at ∞, nor can it have an essential singularity by Picard's theorem (or Casorati-Weierstrass) since
it would then fail to be one-to-one.

◦ Thus f has a pole at ∞ so f (∞) = ∞. This means f extends to be an analytic automorphism of the
extended complex plane, hence by our previous characterization of such maps, f is a fractional linear
transformation. Since the pole of f is at ∞ that means f (z) = az + b for some nonzero a, as desired.

◦ Conversely, it is obvious that f (z) = az + b is an analytic automorphism of C, since f has an inverse


f −1 (z) = (z − b)/a.

17
• In the other situation, U is simply connected and a proper subset of C. In this case, it turns out that there
always exists an analytic isomorphism of U with the unit disc:

• Theorem (Riemann Mapping Theorem): Suppose U is a simply connected open region that is not the entire
complex plane C. Then there exists an analytic isomorphism of U with the open unit disc D = {z : |z| < 1}.
More precisely, given z0 ∈ U there exists a unique analytic isomorphism f :U →D such thatf (z0 ) = 0 and
f 0 (z0 ) > 0.

◦ As an immediate consequence, the Riemann mapping theorem implies that any two simply connected
open regions U, V (excluding C) are analytically isomorphic to one another. Furthermore, if we can nd
analytic isomorphisms f : U → D and g : V → D, then g −1 ◦f will give the desired analytic isomorphism.
◦ Therefore, it suces to construct analytic isomorphisms mapping regions onto the open unit disc.

◦ We will not really use the result of this theorem except as motivation (namely, why we now just focus on
writing down maps from a region to the open unit disc), so for brevity we will omit some of the technical
details of the proof of this theorem.

◦ Proof (Outline): For a given simply connected U 6= C, consider the family F of all injective holomorphic
functions f :U →D such that f (z0 ) = 0.
◦ This family is nonempty: because U 6= C, U omits some point α, and then because U is simply connected,
there exists a branch g(z) of the complex logarithm log(z − α) that is holomorphic and (thus) one-to-one
on U.
◦ Then for any z0 ∈ U , g(z) must be bounded away from g(z0 ) + 2πi (otherwise by taking an appropriate
limit, there would exist points with g(z) = g(z̃) + 2πi), and so [g(z) − g(z0 ) − 2πi]−1 is bounded. By
−1
translation and rescaling one obtains a function f (z) = a[g(z) − g(z0 ) − 2πi] + b such that f (U ) ⊆ D
and f (z0 ) = 0.
◦ Now observe that for any f ∈ F, the values |f 0 (z0 )| γ to be the coun-
are uniformly bounded: take

0 1 ´ f (z)
terclockwise boundary of U and apply Cauchy's integral formula f (z0 ) = dz . Since
2πi γ (z − z0 )2
0
|f (z)| ≤ 1 for all z ∈ γ since the image of f lies in the unit disc, we see |f (z0 )| is uniformly bounded
1
above (namely by times the arclength of γ divided by the square of the distance from z0 to γ ).

◦ The next claim is that there exists a function f ∈ F maximizing |f 0 (z0 )|. If M is the least upper bound of
0 0
the values |f (z0 )| for f ∈ F , then either there is a function with |f (z0 )| = M or there is a sequence of fi
0
with |fi (z0 )| → M . In the latter case one can show that this sequence has a subsequence that converges
0
uniformly on compact subsets of U , so by uniform convergence the limit function f has |f (z0 )| = M .
This function is also necessarily injective by a Rouché's theorem argument, and since |f (z)| ≤ 1 for all
z ∈ U , by the maximum modulus principle in fact |f (z)| < 1 for all z ∈ U , so this f ∈ F .
◦ Finally, we show that any injective holomorphic f :U →D |f 0 (z0 )| is necessarily onto: if
maximizing
the image of f does not contain some point z1 , then by moving z0 to 0 and z1 to 0 one obtains an
0
injective holomorphic g : U → D with |g (0)| = M and 0 6∈ im(g). Then since the image V of g is simply
connected, there exists a branch of the complex logarithm holomorphic on im(g), and hence by extension
2 [log g(z)]/2
there exists an h : U → D with h(z) = g(z) (namely, h(z) = e ). Then one immediately has
h ∈ F and also M = |g (0)| = 2 |h(0)| |h (0)| so that |h (0)| > |g (0)| = |f 0 (z0 )|, which is a contradiction.
0 0 0 0

• Now we can give some simple examples of analytic isomorphisms of various dierent regions.

z−i
• Example: The function f (z) = is an analytic isomorphism from the upper half-plane Re(z) > 0 to the
z+i
z +1
unit disc D. Its inverse f −1 (z) = −i maps the unit disc to the upper half-plane.
z−1
◦ Since f is a fractional linear transformation it suces to observe that f (z) maps the real axis to the unit
t−i
circle, which in turn follows by noting that for t real we have = 1, and that f (i) = 0 lies inside
t+i
the image of f.
◦ This map is often called the Cayley transform.

18
• Example: The function f (z) = z 2 is an analytic isomorphism from the rst quadrant Re(z) > 0, Im(z) > 0
to the upper half-plane.

◦ Thus, composing this map with the inverse of the Cayley transform yields an analytic isomorphism
z2 + 1
z 7→ −i from the rst quadrant to the unit disc.
z2 − 1
◦ Additionally, the inverse map f −1 (z) = z 1/2 = e[Logz]/2 (which is holomorphic on the upper half-plane
since it is only badly behaved at z = 0) is an analytic isomorphism from the upper half-plane to the rst
quadrant.

z+1
• Example: The function f (z) = is an analytic isomorphism of the upper half-disc |z| < 1, Re(z) > 0
1−z
with the rst quadrant.

◦ To see this we simply calculate the action of f on the real axis (mapped to the real axis) and the circle
3 + 4i
|z| = 1 (mapped to the imaginary axis), and observe that f (i/2) = lies in the rst quadrant.
5
z2 + 1
◦ As a consequence, composing this map with the ones above yields an analytic isomorphism z 7→
1 − z2
i 1
from the upper half-disc to the upper half-plane, and an analytic isomorphism z 7→ − (z + ) with the
2 z
unit disc.

• Example: The logarithm f (z) = Log(z) is an analytic isomorphism of the upper half-plane with the strip
0 < Im(z) < π .

◦ This follows immediately from our description of Log(z) = r + iθ for z = reiθ with r>0 and 0 < θ < π.
◦ The logarithm also gives an analytic isomorphism of the full plane with [0, ∞) removed with the strip
0 < Im(z) < 2π .

5.2.4 Functions on the Unit Disc, Harmonic Functions


• We can also classify the analytic automorphisms of the unit disc. To do this we will require a few related
preliminary results about injective maps on the unit disc, all of which are generally referred to as Schwarz's
lemma:

• Proposition (Schwarz's Lemma): Let D be the open unit disc |z| < 1 and let f :D→D be one-to-one and
such that f (0) = 0.

1. We have |f (z)| ≤ |z| for all z ∈ D.


P∞
◦ Proof: Sincef (0) = 0, the power series expansion for f is of the form f (z) = a1 z + n=2 an z n .
P∞ n
◦ Then g(z) = f (z)/z = a1 + n=1 an+1 z is holomorphic on D after removing the removable singu-
larity at 0, and also since |f (z)| < 1 by hypothesis we see that |g(z)| < 1/ |z| for all |z| < 1.

◦ In particular, for |z| < r we have |g(z)| < 1/r. Hence by the maximum modulus principle, since the
maximum modulus of g on |z| ≤ r must occur on the boundary and g is continuous, we see that
|g(z)| ≤ 1/r for |z| ≤ r.
◦ Now taking r → 1 from below yields |g(z)| ≤ 1 for |z| < 1, whence |f (z)| ≤ |z| as desired.
2. If |f (z0 )| = |z0 | for any nonzero z0 ∈ D , then f is a rotation: f (z) = eiθ z for some θ.
◦ Proof: If we have |f (z0 )/z0 | = 1 then we have the equality case for the maximum modulus princi-
ple, which only occurs when the function is constant. Thus f (z)/z is constant, hence must equal
f (z0 )/z0 = eiθ for some θ, so f (z) = eiθ z as claimed.

3. We have |f 0 (0)| ≤ 1 with equality if and only if f is a rotation.

◦ Proof: As in (1) observe that g(z) = f (z)/z is holomorphic on D after removing the removable
singularity, and that f 0 (0) = a1 = g(0).
◦ By (1) we have |g(z)| ≤ 1 for all z 6= 0, so taking z → 0 yields |g(0)| ≤ 1.

19
◦ Now consider the case where |f 0 (0)| = 1.
If g is nonconstant then g is an open mapping, so in
particular the image of g near 0 would contain an open disc around g(0). But since |g(0)| = |f 0 (0)| = 1
the image of g would include values of absolute value exceeding 1, which contradicts (1).

◦ Hence g must be constant, in which case f is a rotation just as in (2).

• Now we can classify the analytic automorphisms of the unit disc. As in the situation of the extended complex
plane, they all turn out to be fractional linear transformations:

• Proposition (Analytic Automorphisms of the Disc): Let D be the unit disc |z| < 1 and suppose that f : D → D
is an analytic automorphism (i.e., that f is holomorphic and a bijection). If f (α) = 0 then there exists an angle
α−z
θ such that f (z) = eiθ ; conversely all such maps with |α| < 1 and θ real are analytic automorphisms
1 − αz
of the unit disc.

◦ The analytic automorphisms of the disc are often called Möbius transformations

◦ Proof: First we observe that all maps of the given form are analytic automorphisms of the unit disc.
Since these maps are fractional linear transformations it suces to observe that they map the unit circle
to itself, and that they map an interior point to another interior point.

|z − α| 1 |1 − αz|
◦ If |z| = 1 then |f (z)| = eiθ = = 1 since |z| = 1 and |1 − αz| = |1 − αz| since they
|1 − αz| |z| |1 − αz|
are complex conjugates, so f maps the unit circle to itself. Furthermore, since f (α) = 0, f maps the
interior point α to another interior point 0, as required.
α−z
◦ Now suppose that f is an analytic automorphism of the disc and that f (α) = 0. Take g(z) = .
1 − αz
Then g is an analytic automorphism of the unit disc so h = f ◦ g −1 is as well, and h maps 0 to 0.

◦ Since h is an injective map from the disc to itself, by the Schwarz lemma we have |h(z)| ≤ |z| for all z in
the disc. But h−1 is also an injective map from the disc to itself, so h−1 (z) ≤ |z| for all z in the disc,
which upon applying h implies |z| ≤ |h(z)| for all z in the disc.

◦ Hence we have equality: |h(z)| = |z| for all z , so h must be a rotation: h(z) = eiθ z for some θ. Then
iθ α − z
f (z) = (h ◦ g)(z) = e , as claimed.
1 − αz
• One of the main utilities of being able to calculate all of these analytic isomorphisms is that they allow us
to transfer solutions to certain types of equations from one region to another. The connection comes via
harmonic functions:

• Denition: If u(x, y) is a real-valued function of two variables (equivalently, a real-valued function of z = x+iy ),
∂2 ∂2
we say u is harmonic when uxx + uyy = 0. Equivalently, in terms of the Laplacian operator ∆ =
∂x2 + ∂y 2 =
1 ∂2
, u is harmonic when ∆u = 0.
4 ∂z∂z
◦ From the Cauchy-Riemann equations, we can see immediately that if f = u(x, y)+iv(x, y) is holomorphic,
then u and v are both harmonic.

◦ Inversely, if u(x, y) U , then in fact u is the real part of a


is harmonic on a simply connected region
holomorphic function. Explicitly, if we dene h = ux − iuy ,
h is holomorphic since it satises the
then
Cauchy-Riemann equations, and so since U is simply connected it has an antiderivative f = p + iq . Then
f 0 = px − ipy so px = ux and py = uy , meaning that p and u have the same partial derivatives hence are
equal up to a constant. By subtracting the constant we obtain a holomorphic function f whose real part
is u, as desired.

◦ For a given u = Re(f ), the imaginary part v = Im(f ) is determined up to an additive constant, and is
called a harmonic conjugate of u.
◦ If u is a harmonic function on V and we have an analytic isomorphism f : U → V for some other region
V , then the composition g = u ◦ f is also a harmonic function on U : explicitly, one may compute using
2
the chain rule that ∆g = (∆u) |f | = 0.
◦ Therefore, composition with an analytic isomorphism f :U →V allows us to transfer between harmonic
functions on U and harmonic functions on V.

20
• As a consequence, if we wish to construct a particular harmonic function (e.g., one satisfying a particular
boundary-value problem) on U, we may equivalently solve that problem on V and then transfer it to U.

◦ One wide class of such functions arises from studying thermal ow: if u(x, y) represents temperature,
then in a steady state temperature distribution in the plane (i.e., constant over time), the heat equation
∆u = ut is equivalent to requiring that u be harmonic.

◦ Boundary-value problems naturally arise in this context in the guise of arranging heating elements around
the boundary of a region (whose temperatures are assumed known and constant): the goal is then to
determine the temperature distribution inside the region.

◦ Another class of such functions arises from studying standing 2-dimensional standing waves: if u(x, y)
represents wave intensity, then in a standing wave pattern (constant over time), the wave equation
∆u = utt requires that u be harmonic. (Indeed, the connection with standing wave patterns is the
reason for the name harmonic function.)

◦ In either situation, if the region is simply connected and bounded, then by the Riemann mapping theorem,
it is analytically isomorphic to the open unit disc. Therefore, if we can solve the corresponding boundary-
value problem for the unit disc, we can transfer back to obtain the solution on the original region.

◦ But since harmonic functions are simply the real parts of holomorphic functions, we are equivalently
seeking to reconstruct a holomorphic function f given its values on the boundary of the disc, and this
is precisely done by Cauchy's integral formula. We need only rewrite it in a way that only involves the
real part of f.

• Theorem (Poisson's Formula): Suppose u is harmonic on the unit disc. Then for any r < 1 we have
ˆ 2π
1 − r2 u(eit )
u(reiθ ) = dt.
2π 0 1 + r2 − 2r cos(t − θ)

◦ Proof: Suppose u = Re(f ) where f


is holomorphic. If γ is the boundary of the unit disc, the Cauchy
1 ´ f (z)
integral formula says that f (z0 ) = dz .
2πi γ z − z0
1 ´ f (z)
◦ By Cauchy's formula again, we also have γ
dz = 0 since 1/z0 is not inside the disc.
2πi z − 1/z0
1 ´ 1 ´
 
1 1 1 − z0 z0
◦ Subtracting yields f (z0 ) = γ
f (z) − dz = γ
f (z) dz .
2πi z − z0 z − 1/z0 2πi (z − z0 )(1 − zz0 )
1 ´ 1 − z0 z0
◦ Parametrizing via γ(t) = eit for 0 ≤ t ≤ 2π yields f (z0 ) = γ
f (eit )ieit dt =
2πi (e − z0 )(1 − eit z0 )
it

1 ´ 1 − |z0 |
2
f (eit ) dt and now since the integrand is f (eit ) times a real number, extracting real parts
2π γ |eit − z0 |2
1 ´ 1 − |z0 |
2
yields u(z0 ) = u(eit ) dt. Rearranging yields the desired formula.
2π γ |eit − z0 |2

Well, you're at the end of my handout. Hope it was helpful.


Copyright notice: This material is copyright Evan Dummit, 2022. You may not reproduce or distribute this material
without my express permission.

21

You might also like