NWI-WB093: COMPLEX ANALYSIS (2023/24)

WADIM ZUDILIN

Our starting example is fully linguistical: mathematics is famous for giving confus-
ing/misleading names to concepts. Students have to be brave enough to understand
that real, imaginary and complex numbers do not in fact refer to their actual role
in nature. They are simply the nicknames for deep mathematical concepts.
Carl Friedrich Gauss, arguably a great mathematician of all the times, complained
about the real/imaginary terminology:
That this subject [imaginary numbers] has hitherto been surrounded
by mysterious obscurity, is to be attributed largely to an ill adapted
notation. If, for example, +1, −1, and the square root of −1 had been
called direct, inverse and lateral units, instead of positive, negative
and imaginary (or even impossible), such an obscurity would have
been out of the question.
Let us think of another quote by Freeman Dyson, one of the prominent contem-
porary representatives of both mathematics and physics, about the role of complex
numbers in the real world:
One of the most profound jokes of nature is the square root of minus
one that the physicist Erwin Schrödinger put into his wave equa-
tion when he invented wave mechanics in 1926. Schrödinger was a
bird who started from the idea of unifying mechanics with optics. A
hundred years earlier, Hamilton had unified classical mechanics with
ray optics, using the same mathematics to describe optical rays and
classical particle trajectories. Schrödinger’s idea was to extend this
unification to wave optics and wave mechanics. Wave optics already
existed, but wave mechanics did not. Schrödinger had to invent wave
mechanics to complete the unification. Starting from wave optics as
a model, he wrote down a differential equation for a mechanical par-
ticle, but the equation made no sense. The equation looked like the
equation of conduction of heat in a continuous medium. Heat conduc-
tion has no visible relevance to particle mechanics. Schrödinger’s idea
seemed to be going nowhere. But then came the surprise. Schrödinger
put the square root of minus one into the equation, and suddenly it
made sense. Suddenly it became a wave equation instead of a heat
conduction equation. And Schrödinger found to his delight that the
equation has solutions corresponding to the quantized orbits in the
1
2 WADIM ZUDILIN

Bohr model of the atom. It turns out that the Schrödinger equa-
tion describes correctly everything we know about the behavior of
atoms. It is the basis of all of chemistry and most of physics. And
that square root of minus one means that nature works with com-
plex numbers and not with real numbers. This discovery came as a
complete surprise, to Schrödinger as well as to everybody else.
All through the nineteenth century, mathematicians from Abel to
Riemann and Weierstrass had been creating a magnificent theory of
functions of complex variables. They had discovered that the the-
ory of functions became far deeper and more powerful when it was
extended from real to complex numbers. But they always thought
of complex numbers as an artificial construction, invented by human
mathematicians as a useful and elegant abstraction from real life.
It never entered their heads that this artificial number system that
they had invented was in fact the ground on which atoms move. They
never imagined that nature had got there first.
In what follows, we slowly cover some basic ingredients of complex analysis. After
a brief review of complex numbers, we will consider some familiar examples and
problems from other parts of mathematics where the use of these numbers as well
as functions of complex variable simplify solving considerably.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 3

1. Basics
Though we can always think of the complex numbers C = {z = x + iy : x, y ∈ R}
as of R2 = {(x, y) : x, y ∈ R}, the latter is just a normed linear vector space equipped
with addition and multiplication by real scalars. However, we can endow the former
with the additional multiplication operation
(x1 + iy1 )(x2 + yy2 ) = (x1 , y1 ) × (x2 , y2 ) = (x1 x2 − y1 y2 , x1 y2 + x2 y1 )
= (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ),
which is easily checked to be associative and commutative; in this way C becomes
a field.
The coordinates x and y of z = x + iy ∈ C are called the real and imaginary
parts of z, respectively. We write Re z = x and Im z = y. Moreover, we define the
conjugate of z to be z = x − yi.
The modulus (or absolute value) of z coincides with the underlying norm of R2 ,
√ p
|z| = z z = x2 + y 2 ,
and the topology induced by this norm defines the topology of C. In particular, we
can talk about limits, convergence of sequences, continuity of functions f : C → C,
uniform continuity, and more, by using the corresponding definitions for R2 .
One more notion related to the geometric visualisation of complex numbers and
their polar coordinates is the argument of z = x + iy ∈ C. It is only defined for
z 6= 0 as an angle θ ∈ R such that x = |z| cos θ and y = |z| sin θ. The argument
arg z = θ is clearly an example of a multi-valued function, because the angle is only
defined modulo 2πZ; for practical purposes, we will assign a special branch of this
function Arg z, the principal argument of z, to be the unique angle θ in the interval
−π < θ ≤ π (though another standard convention 0 ≤ θ < 2π is occasionally in
place).
Exercise 1.1. Given a polynomial p(z) with real coefficients, show that if z0 is its
root then so is its conjugate z 0 . (In other words, complex roots always appear in
conjugate pairs.) Conclude from this that a polynomial p(z) ∈ R[z] of odd degree
always has a real root.
In fact, it is linguistically correct to speak about the zeros of polynomials p(z)
rather than roots; the terminology ‘root’ refers to a solution of the equation p(z) = 0.
We will say that a zero z0 of a polynomial p(z) has multiplicity m ≥ 1 if p(z0 ) = 0
and p(z)/(z − z0 )m is a polynomial that does not vanish at z0 . (The fact that
p(z)/(z − z0 ) is a polynomial follows from little Bézout’s theorem, also known as the
polynomial remainder theorem.)
One useful way of thinking of complex numbers z = x + iy is through identifying
them with the real two-dimensional vector space of (2 × 2)-matrices
 
x y
,
−y x
4 WADIM ZUDILIN
p
with norm given by the square root of the determinant x2 + y 2 . First note that any
such matrix, for z 6= 0, is invertible because its determinant equals x2 +y 2 = |z|2 6= 0.
Secondly, the matrix addition and matrix multiplication of two such matrices result
in matrices of exactly the same form:
     
x1 y1 x2 y 2 x1 + x2 y1 + y2
+ =
−y1 x1 −y2 x2 −(y1 + y2 ) x1 + x2

and
    
x1 y 1 x2 y 2 x1 x2 − y 1 y 2 x1 y 2 + x2 y 1
= ;
−y1 x1 −y2 x2 −(x1 y2 + x2 y1 ) x1 x2 − y1 y2
and the resulted matrices correspond to the complex numbers z1 + z2 and z1 z2 ,
respectively. This example shows that the field of complex numbers and the set of
matrices
  
x y
: x, y ∈ R
−y x
with the usual addition and multiplication are isomorphic. We will witness many
instances of this isomorphism.

Of course, polynomials p(z) ∈ C[z] and rational functions p(z)/q(z) are typical
examples of elementary (complex) functions, which are continuous (apart from the
zeros of q(z)).

Theorem 1.2. The function

∞
X zn
exp : z 7→ ez =
n=0
n!

is well-defined for every z ∈ C. It is a continuous function and satisfies

exp(z + w) = exp(z) · exp(w).

Proof. To show that the series ∞ zn

P
n=0 n! converges, we will use the Weierstrass M-
n
test. For n ≥ 0, define fn (z) = zn! and Mn = |z|n! . Note that |fn (z)| ≤ Mn on the
n

closed subset A := {z ∈ C | |z| ≤ R} ⊂ C. Moreover, the infinite series

∞ ∞
X X |z|n
Mn =
n=0 n=0
n!

converges and equals e|z| , where the latter is the real-valued (familiar!) exponential
function.
P This shows that the conditions of the Weierstrass M -test are satisfied, so
that ∞ zn
n=0 n! converges, absolutely (and uniformly) on the disk {z ∈ Z | |z| ≤ R}.
As the uniform limit of continuous functions (namely, the partial sums which are
all polynomials), it is continuous on the disk |z| ≤ R; as this argument works for
any choice of R, it is continuous at all complex points.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 5

To check the functional identity, we use the binomial theorem:

∞ ∞ ∞ ∞
z+w
X (z + w)n X 1 X n! j k X X z j wk
e = = k!z w =
n=0
n! n=0
n! j,k≥=0 j! j=0 k=0
j!k!
j+k=n

,
∞ ∞
j X
X z wk
= = ez ew .
j=0
j! k=0
k!

The next exercise is a real analysis recall; the binomial theorem

n   n
n
X n k n−k X n!
(a + b) = a b = ak bn−k
k=0
k k=0
k!(n − k)!
is the driving force again.
Exercise 1.3. Prove that, for any z ∈ C,
 z n
lim 1 + = ez .
n→∞ n
Exercise 1.4. Prove that the following trigonometric and hyperbolic functions, now
defined for complex variable by the series
∞ ∞
X (−1)n z 2n+1 X (−1)n z 2n
sin z = , cos z = ,
n=0
(2n + 1)! n=0
(2n)!
∞ ∞
X z 2n+1 X z 2n
sinh z = , cosh z = ,
n=0
(2n + 1)! n=0
(2n)!
are continuous functions of z ∈ C.
Theorem 1.5 (Euler’s relation). eiz = cos z + i sin z for any z ∈ C.
Proof. This follows from the formal manipulation with the power series. ,
If z = θ is real, then Theorem 1.5 implies that eiθ = cos θ + i sin θ is on the unit
circle, |eiθ | = 1. In particular,
|ez | = |ex+iy | = |ex | · |eiy | = |ex | = ex = eRe z ,
the formula we will use on multiple occasions. Now the polar representation of a
complex non-zero number z ∈ C can be cast in the form
z = |z|ei arg z ,
where any branch of the multi-valued arg z can be used because
ei(θ+2πn) = eiθ e2πin = eiθ (cos 2πn + i sin 2πn) = eiθ
for any n ∈ Z. In other words, the exponential function exp z is periodic in C with
the (pure imaginary) period 2πi.
Exercise 1.6. Prove that the (multi-valued!) argument arg z satisfies
arg(zw) = arg z + arg w (mod 2πZ).
6 WADIM ZUDILIN

Exercise 1.7 (Weierstrass’s approximation theorem). The goal of this exercise it

to prove that a complex-valued continuous function f : [a, b] → C always possesses a
good polynomial approximation on the interval: for any ε > 0 there is a polynomial
p(z) ∈ C[z] such that |f (z) − p(z)| < ε for all z ∈ [a, b]. This was first established
by Weierstrass in 1855; the proof below follows Bernstein’s method.
(a) Reduce the theorem to the one for the interval [0, 1].
(b) Using the binomial theorem, prove the identity
n  
X n k
z (1 − z)n−k (k − nz)2 = nz(1 − z).
k=0
k
(c) For z ∈ [0, 1], n ∈ Z>0 and each δ > 0, define the subset Mn,δ (z) ⊂ {0, 1, . . . , n}
such that
Mn,δ (z) = {k = 0, 1, . . . , n : |k/n − z| < δ}.
Show that
X n 1
z k (1 − z)n−k ≤ 2 .
k nδ
k∈M
/ n,δ (z)

(d) Given n ∈ Z>0 and the continuous function f : [0, 1] → C (hence uniform con-
tinuous!), define its Bernstein polynomial
n    
X n k k
Bn (z) = f z (1 − z)n−k .
k=0
k n
Clearly, this is a polynomial of degree at most n. Define the absolute bound M
of f (x) on [0, 1] and, for each ε > 0, the corresponding δ = δ(ε) > 0 such that
|f (x) − f (y)| < ε/2 whenever |x − y| < δ. Show that
ε 2M
|f (z) − Bn (z)| ≤ + for all z ∈ [0, 1].
2 nδ 2
(e) For the function f : [0, 1] → C from the previous part, use the uniform continu-
ity (δ only depends on ε) and choose n sufficiently large to conclude with the
statement of Weierstrass’s theorem.
One of the principal features of functions is their differentiability. This clearly
extends to the functions f : Ω → C defined on an open set Ω ⊂ C. We say that a
function f is differentiable at a ∈ Ω if
f (z) − f (a) = C · (z − a) + o(z − a) as z → a
for some C ∈ C, where o(z − a) denotes a function of the form ε(z)(z − a) with
ε(z) → 0 as z → a. The principal (linear) part of the increment f (z) − f (a), namely
C(z − a), is called the differential of f (z) at z = a, and the coefficient C is called
the derivative of f (z) at z = a. We let f 0 (a) denote the derivative of f (z) at z = a.
By dividing both sides of the identity by z − a we see that differentiability of f at
z = a is equivalent to
f (z) − f (a)
lim = C.
z→a z−a
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 7

A tradition in complex analysis is calling differentiable functions holomorphic. Fur-

thermore, we say that f : Ω → C is holomorphic on Ω if it is holomorphic (i.e.,
differentiable) at every a ∈ Ω.
One can check following mutatis mutandis the arguments of real analysis that
all the rules of differentiation of complex functions are those familiar from real
analysis. In particular, polynomials are holomorphic (in fact, infinitely many times
differentiable) at every point z ∈ C. Similarly, rational functions p(z)/q(z) are
holomorphic (and infinitely many times differentiable) at all points z with q(z) 6= 0.
A holomorphic function f : C → C is called an entire function. Examples of
entire functions are polynomials in z and the exponential function.
Theorem 1.8. The function f (z) = exp z is entire. Its derivative at z = a is equal
to exp a.
Proof. As in calculus, we write z = a + h and notice that
∞
eh − 1 X hn
=1+ → 1 as h → 0.
h n=1
(n + 1)!
Therefore,
0
f (a) = lim
z→a
f (z) − f (a)
z−a
= lim
h→0
ea+h − ea
h
a
= e lim
h→0
eh − 1
h
= ea . ,
Exercise 1.9. Show that the functions sin z, cos z, sinh z and cosh z are entire and
compute their derivatives.
8 WADIM ZUDILIN

2. Holomorphic functions
Let f : Ω → C be holomorphic at a = a1 + ia2 ∈ Ω. Write f (z) = u(x, y) + iv(x, y)
and b = f 0 (a) = b1 + ib2 for the value of the derivative. The limiting relation
(f (z) − f (a))/(z − a) → b as z → a implies that
f (x + ia2 ) − f (a1 + ia2 )
→ b as x → a1
(x + ia2 ) − (a1 + ia2 )
and
f (a1 + iy) − f (a1 + ia2 )
→ b as y → a2 .
(a1 + iy) − (a1 + ia2 )
At the same time the quotients featured can be given in the form
f (x + ia2 ) − f (a1 + ia2 ) u(x, a2 ) − u(a1 , a2 ) v(x, a2 ) − v(a1 , a2 )
= +i
(x + ia2 ) − (a1 + ia2 ) x − a1 x − a1
and
f (a1 + iy) − f (a1 + ia2 ) u(a1 , y) − u(a1 , a2 ) v(a1 , y) − v(a1 , a2 )
= −i + ;
(a1 + iy) − (a1 + ia2 ) y − a2 y − a2
therefore, the limits above imply the existence of partial derivatives
∂u u(x, a2 ) − u(a1 , a2 ) ∂v v(x, a2 ) − v(a1 , a2 )
(a1 , a2 ) = lim , (a1 , a2 ) = lim ,
∂x x→a1 x − a1 ∂x x→a1 x − a1
∂u u(a1 , y) − u(a1 , a2 ) ∂v v(a1 , y) − v(a1 , a2 )
(a1 , a2 ) = lim , (a1 , a2 ) = lim ,
∂y y→a2 y − a2 ∂y y→a2 y − a2
and the following equalities between them.
Theorem 2.1 (Cauchy–Riemann equations). If f (z) = u(x, y) + iv(x, y) : Ω → C
is holomorphic at a ∈ Ω, then
∂u ∂v ∂u ∂v
(a) = (a) and (a) = − (a). (2.1)
∂x ∂y ∂y ∂x
Furthermore,
∂u ∂v ∂v ∂u
f 0 (a) = (a) + i (a) = (a) − i (a).
∂x ∂x ∂y ∂y
Exercise 2.2. Assume that the functions u(x, y) and v(x, y) have continuous partial
derivatives at the point a ∈ Ω, and these derivatives satisfy the Cauchy–Riemann
equations (2.1). Prove that the function f (z) is holomorphic at a.
Exercise 2.3. Check that the functions
z 7→ z, z 7→ Re z, z 7→ Im z and z 7→ |z|
are not holomorphic on C. At the same time, they are differentiable as functions
from R2 to R2 (when we exclude the origin for the last function).
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 9

At this point, it is important to stress the difference between notions of holomor-

phicity of function f (z) : C → C and differentiability of (u(x, y), v(x, y)) : R2 → R2 .
For the latter, the derivative is a linear operator from R2 to R2 , which can be
identified with the 2 × 2 matrix
 ∂u ∂v

∂x
(a1 , a2 ) ∂x (a1 , a2 )
J(a) = ∂u ∂v
∂y
(a1 , a2 ) ∂y (a1 , a2 )
of the partial derivatives at the point. (This object is clearly 4-dimensional over R.)
For the former, the derivative is simply a complex number b = b1 + ib2 . (This is
2-dimensional!) The Cauchy–Riemann equations (2.1) simply mean that
 
b1 b2
J(a) = ;
−b2 b1
in other words, the holomorphicity imposes the corresponding 2×2 matrix to lie in a
2-dimensional subspace over R, in fact, to represent the very same complex number
b = b1 + ib2 under the isomoprhism described in Section 1 (before Theorem 1.2).
The requirement to fit in the 2-dimensional subspace looks quite restrictive (when
we ask a mapping from R2 to R2 to be holomoprhic, we indeed severely limit the
notion of differentiability); a surprising consequence of the restriction is the existence
of remarkable theory for holomorphic functions, a theory which is not available for
differentiable real mappings. Holomorphic mappings are also natural from geometric
perspectives — something we are going to discuss next.
We will say that a holomorphic function f : Ω → C is regular at a ∈ Ω if f 0 (a) 6= 0
and that it is regular on Ω if f 0 (a) 6= 0 for all a ∈ Ω.
Lemma 2.4. If f : Ω → C is holomorphic at a ∈ Ω and locally invertible at a (in
other words, there exists a functional inverse f −1 in a neighbourhood of the point),
then f is regular at a.
Proof. This is essentially an application of the inverse function theorem. Namely, if
g = f −1 , i.e., g(f (z)) = z for z in a neighbourhood of a, then g is differentiable at

,
b = f (a) with the derivative satisfying g 0 (b)f 0 (a) = g 0 (f (a))f 0 (a) = 1. The latter is
only possible when f 0 (a) 6= 0.
Exercise 2.5. For a regular holomorphic map f on Ω, show that its local functional
inverses f −1 are holomorphic and regular.
Assume that a function f : Ω → Rd on a domain Ω ⊂ Rd is C ∞ (has derivatives of
all orders at any a ∈ Ω) and is locally invertible at any a ∈ Ω. We call f conformal
at a ∈ Ω if for any pair of smooth curves γ1 and γ2 passing through the point a (say,
at t = t1 and t = t2 , respectively) the angle between the tangent vectors γ10 (t1 ) and
γ20 (t2 ) is preserved by f : it is equal to angle between the tangent vectors (f ◦ γ1 )0 (t1 )
and (f ◦ γ2 )0 (t2 ). If f is conformal at every a ∈ Ω, we say that f is conformal on Ω.
Observe that the general chain rule (in Rd ) applies to the composition (f ◦γ)(t) at
t = t0 of a differentiable mapping f at x0 = γ(t0 ) ⊂ Rd with a smooth curve γ at t0 ;
the result is (f ◦ γ)0 (t0 ) = f 0 (x0 )γ 0 (t0 ), where f 0 (x0 ) = Jf (x0 ) : Rd → Rd is a linear
operator. This implies that the angle between the two tangent vectors is preserved
10 WADIM ZUDILIN

if and only if Jf (x0 ) is a linear operator preserving angles. Such linear operators
are composed from rotations in Rd with ‘stretches’ (corresponding to the matrices
±r Id with r ∈ R \ {0}, hence commuting with any other linear transformations).
In R2 , the matrix of any such operator therefore assumes the form
 
cos θ sin θ
±r ,
− sin θ cos θ
which corresponds to the rotation by θ followed by the dilation by r 6= 0 (and
possible reflection in the case of the minus sign). All such matrices (and only these
ones!) have the form
 
b1 b2
B= with det B = b21 + b22 6= 0.
−b2 b1
Thus, the characterisation of holomorphic mappings in Theorem 2.1 brings us to
the following result. (Though the requirement that f is C ∞ plays no role in it, we
will see later that it is valid for a holomorphic function.)
Theorem 2.6. Let f : Ω → C be a regular holomorphic map. Then f is conformal.
Conversely, if f : Ω → C is conformal, then f is holomorphic and regular on Ω.
Another standard application of holomorphic functions is to constructing har-
monic functions on R2 . A (real-valued!) function u(x) = u(x1 , . . . , xd ) : Ω → R is
said to be harmonic on Ω ⊂ Rd if it has second-order partial derivatives ∂ 2 u/∂x2j
on Ω for j = 1, . . . , d and its Laplacian vanishes on Ω:
∂ 2u ∂ 2u
∆u = + · · · + = 0.
∂x21 ∂x2d
Theorem 2.7. Let f : Ω → C be holomorphic and C 2 on Ω, then its real and
imaginary parts are harmonic functions.
Proof. By differentiating the Cauchy–Riemann equations (2.1) we find out that
∂ 2u ∂ 2v ∂ 2u ∂ 2v
(a) = (a) and (a) = − (a)
∂x2 ∂x ∂y ∂ 2y ∂y ∂x
at every a = a1 + ia2 ∈ Ω. It remains to notice the Analysis 2 equality
∂ 2v ∂ 2v
(a) = (a)
∂x ∂y ∂y ∂x
for a C 2 function f . ,
The condition that f is C 2 on Ω may look restrictive; we will soon see that if f
is holomorphic on Ω then it is C ∞ — quite a spectacular bonus for functions which
are a priori known to have just their first derivative.
Exercise 2.8. Let f : Ω → C be holomorphic on a connected domain Ω ⊂ C and
∆ = ∂ 2 /∂x2 + ∂ 2 /∂y 2 the Laplacian. Show that
∆ |f (z)|2 = 4|f 0 (z)|2 .


 NWI-WB093: COMPLEX ANALYSIS (2023/24) 11
Pn
Conclude from this that if f1 , . . . , fn are holomorphic functions and k=1 |fk (z)|2 is
harmonic, then all fk are constant.
Exercise 2.9. Show that f (z) = z exp(z) is holomorphic on C. Deduce that the
function u(x, y) = ex (x cos(y) − y sin(y)) is harmonic on R2 .
Lemma 2.10. Let f : D → C be holomorphic on an open disk D = D(a, r), centred
at a of radius r > 0. If f 0 (z) = 0 for all z ∈ D then f (z) is constant on D.
Proof. Connect an arbitrary point z ∈ D with the centre a by the line segment
γ(t) = (1 − t)a + tz, where t ∈ [0, 1]; of course, the entire path is in D because of
the convexity of the latter. Differentiating along the interval we get
d
f (γ(t)) = f 0 (γ(t))γ 0 (t) = 0 for all t ∈ (0, 1),
dt
since f 0 (γ(t)) = 0 by the hypothesis. By the mean-value theorem (applied to the

,
real and imaginary parts of f ), we have that f (z) − f (a) = f (γ(1)) − f (γ(0)) = 0,
so that f (z) = f (a). This is valid for any z ∈ D implying the result.
In the remaining part of this section we will assume that Ω is a connected open
set in C.
Theorem 2.11. Let f : Ω → C be holomorphic on a connected open subset Ω ⊂. If
f 0 (z) = 0 for every z in Ω, then f (z) is constant on Ω.
Proof. We fix a point a ∈ Ω. Take an arbitrary z ∈ Ω. Since Ω is open and
connected, it is arcwise connected: the points a and z can be connected by a contin-
uous γ : [0, 1] → Ω. Since [0, 1] is compact and γ continuous, its trajectory γ([0, 1])
is compact. At each point of w of the trajectory choose a disk Dw centred at w,
which is entirely in Ω. The induced covering of the (compact) trajectory by the
disks admits a finite subcover; call it Dw1 ∪ · · · ∪ Dwn , with the convention that
Dw1 covers a = γ(0), the disk Dwn covers z = γ(1) and Dwj ∩ Dwj−1 6= ∅ for
j = 1, . . . , n − 1 (in other words, any two successive disks overlap). By Lemma 2.10,
f is constant on each of the disks, hence on the union Dwj ∪Dwj−1 of two consecutive
disks (because of the coincidence on their overlapping intersection). Now, induction

,
on j = 1, . . . , n − 1 shows that the constant is same for all the disks in the cover,
hence f (z) = f (γ(1)) = f (γ(0)) = f (a).
Exercise 2.12. Prove that the conclusion of Theorem 2.11 does not hold in general
when the domain Ω is not connected.
Some corollaries of Theorem 2.11 are as follows:
(a) if holomorphic maps f and g on Ω satisfy f 0 = g 0 , then f − g is constant on Ω;
(b) if a holomorphic map f takes only real values, or only imaginary values, or if
|f | is constant, then f is constant.
For part (b), observe that if f 0 (a) 6= 0 at some point a ∈ Ω, then the image f (D)
of some neighbourhood D = D(a, ε) is an open subset C (in particular, of real
dimension 2), therefore, cannot lie entirely in R, or iR, or {w ∈ C : |w| = R} for
some R > 0. Thus, f 0 (a) = 0 for all a ∈ Ω is the only available option.
12 WADIM ZUDILIN

Exercise 2.13. Use the Cauchy-Riemann equations to prove that the image of
non-constant holomorphic map f : Ω → C is not contained in R.
Exercise 2.14. Show that f : Ω → C is a polynomial if and only if its nth derivative
f (n) (z) vanishes in Ω for some n ∈ Z>0 .
Exercise 2.15. For a holomorphic map f : Ω → C, show that the following are
equivalent:
(i) f is constant;
(ii) Re f and Im f are constant;
(iii) f 0 = 0;
(iv) |f | is constant.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 13

3. Power series and analytic functions

We have already seen that power series may be used to define holomorphic func-
tions: this is true for polynomials (which are always represented by finite power
series) but also for the exponential function exp(z) and its relatives — trigonometric
and hyperbolic sine and cosine.
We say that a function f : Ω → C is analytic at a ∈ Ω if there is real number R > 0
and a sequence of complex numbers a0 , a1 , . . . such that Pthe open disc D(a, R) :=
{z ∈ Z | |z−a| < R} is contained in Ω and the power seriesP ∞ n
n=0 an (z−a) converges
∞ n
to the function f (z) on D(a, R) ⊂ Ω, i.e., the identity n=0 an (z − a) = f (z) is
valid for all z ∈ D(a, R). If this property is valid for all a ∈ Ω, we say that f is
analytic (on Ω).
By dividing a polynomial f (z) successively by z − a, we can write it in the form
Pd n
n=0 an (z − a) (this finite series converges in C); this shows that any polynomial
is analytic on C. For the exponential function exp(z), which we defined by the
convergent (on C) power series
∞
X zn
exp(z) = ,
n=0
n!
we can use its functional equation (Theorem 1.2) in the form ez = ea · ez−a to
conclude with
∞
X ea
exp(z) = an (z − a)n , where an = for n = 0, 1, 2, . . . .
n=0
n!
This shows that the exponential function is analytic on C as well.
Exercise 3.1. Verify that sin z, cos z, sinh z and cosh z are analytic on C.
The following three results are stated (and also proven) in Section 1.7 of Moskowitz’s
book.
Lemma 3.2. For z ∈ D(0, 1) = {z : |z| < 1}, the (geometric) series
∞
X
zn
n=0

converges to the rational function 1/(1 − z).

Proof. For the partial sums we have
N
X 1 − z N +1
zn = .
n=0
1−z

,
N +1
If |z| < 1 then |z| → 0 as N → ∞, hence z N +1 → 0 as well. Therefore, taking
the limit as N → ∞ in the equality above, we arrive at the desired claim.
Lemma 3.3 (Abel’s lemma). If a power series ∞ n
P
n=0 an z converges at some z0 ,
then it converges absolutely in the open disk |z| < |z0 |.
14 WADIM ZUDILIN
P∞ n
Proof. Without loss of generality, z0 6= 0. Now, the convergence of n=0 an z0
implies that sequence of complex numbers an z0n tends to 0 as n tends to ∞. This
means that there exists an integer n0 such that |an | |z0 |n ≤ 1 for all n > n0 . Define
C = max{1, |a0 |, |a1 z0 |, |a2 z02 |, . . . , |an0 z0n0 |}.
Then, the estimate |an | |z0 |n ≤ C holds for all n = 0, 1, 2, . . . . Now, for every
complex number z satisfying |z| < |z0 |, we obtain
∞ ∞ ∞  n
X
n
X
n |z|n X |z|
|an z | = |an z0 | · n
≤C ,
n=0 n=0
|z 0 | n=0
|z 0 |

the latter series converging in view of Lemma 3.2 because of |z|/|z0 | < 1. ,
Theorem 3.4. Assume that a power series ∞ n
P
n=0 an (z − a) converges in a disk
D(a, R) = {z : |z − a| < R}, where 0 < R ≤ ∞. Then it converges absolutely and
uniformly in any closed disk D(a, r) = {z : |z − a| ≤ r} with r < R.
Proof. Clearly, by shifting the variable, it is sufficient to establish the statement
when a = 0. Since r < R, the P absolute convergence inn D(a, r) follows from
Lemma 3.3. This implies that ∞ n=0 Mn with Mn = |an |r converges. But then

,
n
|an z | ≤ Mn for all z ∈ D(a, r) and all n, and the uniform convergence follows from
the Weierstrass M-test (see Proposition 1.7.3 in Moskowitz).
Exercise 3.5. Denote
M = lim sup |an |1/n ∈ [0, ∞].
n→∞

Take R = 1/MP (so that R = ∞ when M = 0 and R = 0 when M = ∞). Prove

that the series ∞ n
n=0 an (z − a) converges for all z ∈ C if R = ∞; it converges for
all z ∈ D(a, R) if R > 0 is finite, and diverges for all z 6= a if R = 0.
Hint. Observe that Theorem 3.4 implies that if
 X∞ 
n
R = sup r ≥ 0 : |an |r converges ,
n=0
P∞

,
n
then the series n=0 an (z −a) converges for |z −a| < R and diverges for |z −a| > R.

Theorem 3.6. Assume that P a function f : Ω → C is analytic at a ∈ Ω and rep-

resented
P∞ by the power series ∞ n 0
n=0 an (z − a) in the disk D(a, R). Then f (z) =
n−1
n=1 nan (z − a) and the series converges in the same disk.
In other words, a function that is analytic at a is automatically holomorphic at it
(in fact, it possesses derivatives of all orders there!) — we do not need to appeal to
the Cauchy–Riemann equations for analytic functions.
P∞ n
Proof.
P∞ Without loss of generality, we may assume that a = 0. The
√ series n=0 an z
and n=1 nan z n−1 have the same radius of convergence, because n n → 1 as n → ∞.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 15

Take z ∈ D(0, R) and pick r > 0 such that |z| < r < R. Let ε > 0 be arbitrary.
Then there is N = N (ε) such that for the tail of the convergent series
X ε
n|an |rn−1 < .
n>N
3
For any h ∈ D(0, r − |z|) we have w = z + h ∈ D(0, r) and by the triangle inequality
∞ ∞ 
wn − z n

f (w) − f (z) X n−1
X
n−1
− nan z = an − nan z
w−z n=1 n=1
w − z
∞
X
= an (wn−1 + wn−2 z + · · · + wz n−2 + z n−1 − nz n−1 )
n=1
N
X
≤ an (wn−1 + · · · + wz n−2 + z n−1 − nz n−1 )
n=1
X
+ |an | (|w|n−1 + · · · + |w| |z|n−2 + |z|n−1 + n|z|n−1 )
n>N
N
X 2ε
< an (wn−1 + · · · + wz n−2 + z n−1 − nz n−1 ) + .
n=1
3
It remains to notice that by choosing w sufficiently close to z, that is, |w − z| < δ
for some δ = δ(ε) > 0, we can make the absolute value of the remaining sum to be
less than ε/3. Therefore,
∞
f (z) − f (w) X
− nan z n−1 < ε
z−w

,
n=1

implying the statement.

This theorem gives another proof of the holomorphicity of the exponential func-
tions (and of its trigonometric and hyperbolic relatives).
Exercise 3.7. (a) For z ∈ D(0, 1), compute the limit
∞
X
nz n−1 .
n=1

(b) Show that for each k > 0, there exists a polynomial Pk (t) with integer coefficients
of degree k − 1 such that
∞
X Pk (z)
nk z n−1 = .
(1 − z)k+1

,
n=1

Proof. (b) Use induction on k = 1, 2, . . . . This statement is due to Euler.

16 WADIM ZUDILIN

4. Contour integrals and the Cauchy theorem

Let Ω be a connected domain in C. A contour (or path) is a continuous mapping
γ : [a, b] → Ω of a real interval. In what follows, we will impose a stricter requirement
on countours to be piecewise smooth; in other words, apart of a finitely many points
a = t0 < t1 < · · · < tn = b the function γ(t) is differentiable and its derivative is
continuous (meaning that this is valid for its real and imaginary parts). With this
setup we can always think of the integral over t ∈ [a, b] as
Z b X n Z tj
= ,
a j=1 tj−1

where each integral in the sum is considered as the improper Riemann integral when
the values of the integrand at the endpoints are undefined. To each such contour
we assign its trajectory γ([a, b]) and length
Z b
`(γ) = `(γ([a, b])) = |γ 0 (t)| dt
a

the latter formula is obtained in Analysis 2 for more general contours in Rd (Garling’s
Theorem 17.2.8). Finally, for a continuous function f : Ω → C define its contour
integral
Z Z b
f (z) dz = f (γ(t))γ 0 (t) dt;
γ a
the real and imaginary parts of the integral on the right-hand side are produced by
integrating the real and imaginary parts of the integrand.
Exercise 4.1. Assume that for another (piecewise smooth) contour γ̃ : [ã, b̃] → Ω we
have the trajectory γ̃([ã, b̃]) coinciding with γ([a, b]) and γ̃(ã) = γ(a), γ̃(b̃) = γ(b).
Verify that
Z b̃ Z b
0
f (γ̃(s))γ̃ (s) ds = f (γ(t))γ 0 (t) dt.
ã a
In other words, the definition above does not depend on the choice of parameterisa-
tion of the contour.

,
Hint. Use the implicit function theorem to conclude that s = s(t) is a piecewise
smooth bijection of [a, b] onto [ã, b̃]; then apply the chain rule.
Exercise 4.2. Given a ∈ C and r > 0, consider the contours
(a) γ1 (t) = a + re2πit , where t ∈ [0, 1]; (b) γ2 (t) = a + re4πit , where t ∈ [0, 1];
2
(c) γ3 (t) = a + re−2πit , where t ∈ [0, 1]; (d) γ4 (t) = a + re2πit , where t ∈ [0, 1].
Which of these are different parameterisations of the same contour?
Lemma 4.3 (Fundamental inequality). For a continuous function f : Ω → C and
contour γ : [a, b] → Ω, consider
M = sup |f (z)| = max |f (γ(t))| and L = `(γ) = `(γ([a, b])).
γ a≤t≤b
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 17

Then Z
f (z) dz ≤ M L.
γ

Proof. This is a particular case of the mean-value inequality for the Riemann integral

,
from Analysis 2 (Garling’s Theorem 18.3.3), which is not hard to verify by writing
the estimates for the upper Riemann sums involved.
Now notice that for a contour γ : [a, b] → C the contour (−γ) : [−b, −a] → C
defined by the rule (−γ)(t) = γ(−t) makes perfect sense with the resuling integral
Z Z
f (z) dz = − f (z) dz.
−γ γ

More generally, we have a natural additive operation γ = γ1 ∪ γ2 (in this order!) for
two contours in Ω such that the ending point of γ1 coincides with the starting point
of γ2 : Z Z Z
f (z) dz = f (z) dz + f (z) dz.
γ γ1 γ2
Then −γ is nothing but the additive inverse of γ. Observe that each contour is
equipped with the orientation, and for closed contours we can talk about positive
(or negative) orientation depending on whether the enclosed area is on the left (or
on the right) of the contour. For a closed positively oriented contour, for example,
the unit disk |z| = 1, the standard convention is using symbol
I
f (z) dz.
γ

As an example, let us compute the integral of f (z) = 1/z defined on the domain
Ω = C× = C \ {0} over the unit disk |z| = 1. The latter can be parameterised by
γ(t) = e2πit , where t ∈ [0, 1]. Therefore,
I Z 1 Z 1 Z 1
1 −2πit 2πit −2πit 2πit
dz = e d(e ) = 2πi e × e dt = 2πi dt = 2πi.
|z|=1 z 0 0 0

Exercise 4.4. Compute the integral of f (z) = z n (on the domain C× ) over the unit
disk for any (positive and negative) integer n.
Given a continuous function f : Ω → C, it is convenient (as in real analysis) to
call F (z) its primitive (if exists!) when F (z) is holomorphic on Ω and F 0 (z) = f (z)
in the domain. Why primitives are important? Applying the Fundamental Theorem
of Calculus (the Newton–Leibniz formula) to the real and imaginary parts we find
out that Z Z b
f (z) dz = f (γ(t))γ 0 (t) dt = F (γ(b)) − F (γ(a)).
γ a
In particular, we see from this result that the integral does not depend on the
contour γ but only on its end-points γ(a) and γ(b), if the primitive of the function
f (z) exists. In the latter case, we automatically get the zero value of the integral for
a closed contour γ in Ω. In particular, our example above shows that the function
f (z) = 1/z does not have a primitive in the domain C× .
18 WADIM ZUDILIN

Theorem 4.5 (Cauchy–Goursat). Let f : Ω → C be a holomorphic function on a

connected domain Ω ⊂ C. Take a closed triangle T ⊂ Ω and consider its boundary
γ = ∂T as a contour in Ω traversed in positive direction. Then
I
f (z) dz = 0.
γ

Proof. Denote I
I= f (z) dz ≥ 0.
γ
S4
Drawing the three midparallels divide the starting triangle T into four, T = j=1 Tj ,
as shown on the figure:

Denote γj = ∂Tj for j = 1, 2, 3, 4 their boundaries transversed in positive direction.

Then
I X4 I
f (z) dz = f (z) dz,
γ j=1 γj

because the integrals along the midparallels compensate each other. By the triangle
inequality (same name but a different triangle is meant!) we have
I X4 I
I= f (z) dz ≤ f (z) dz
γ j=1 γj

implying that I
1
f (z) dz ≥ I
γj 4
for the boundary of at least one of the four smaller triangles T1 , T2 , T3 , T4 . Take
T 0 = T and T 1 to be (any) one of the triangles, for which the latter inequality
happens; set γ 0 = ∂T 0 and γ 1 = ∂T 1 . Apply the above procedure to the triangle
T 1 to get a smaller triangle Tj with boundary γj = ∂Tj , for which
I I
1 1
f (z) dz ≥ f (z) dz ≥ 2 I,
γj 4 γ1 4
denote T 2 = Tj , γ 2 = γj , and repeat the process for T 2 , etc. As a result, we obtain
a chain T 0 ⊃ T 1 ⊃ T 2 ⊃ · · · of embedded triangles with respective boundaries
γ 0 , γ 1 , γ 2 , . . . such that
I
k 1 k 1 1
diam T = k diam T, `(γ ) = k `(γ) and f (z) dz ≥ k I
2 2 γk 4
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 19

for k = 1, 2, .T. . . In view of diam T k → 0 as k → ∞ and by completeness of C, the

intersection k≥0 T k ⊂ Ω consists of a single point a ∈ Ω. By the hypothesis, the
function f (z) is holomorphic at a, hence representable in the form

f (z) = f (a) + f 0 (a)(z − a) + g(z)(z − a),

where g(z) → 0 as z → a. Observe that the latter equality defines g(z) to be

a continuous function on Ω, because the functions f (z) and z − a are continuous
there. The function f (a) + f 0 (a)(z − a) has a primitive in Ω (equal to f (a)(z − a) +
f 0 (a)(z − a)2 /2) implying that

I I
f (z) dz = g(z)(z − a) dz ≤ sup |g(z)(z − a)| × `(γ k )
γk γk γk

≤ sup |g(z)| × diam T k × `(γ k ),

Tk

where Lemma 4.3 was used. With the previous data,

I
k 1 1
I≤4 f (z) dz ≤ 4k sup |g(z)| ×
k
diam T × k `(γ) = diam T `(γ) sup |g(z)|
γk Tk 2 2 Tk
→ diam T `(γ)|g(a)| = 0 as k → ∞.

Thus, I = 0, the required statement. ,

We say that Ω is a star-shaped domain about a ∈ Ω (the point a is then declared
to be a centre) if for any z ∈ Ω the line segment [a, z] joining a to z lies entirely
in Ω.
The Cauchy–Goursat theorem has the following application to such domains.

Lemma 4.6. Suppose that f : Ω → C is holomorphic on a star-shaped domain with

centre a. Then
Z
F (z) = f (ζ) dζ
[a,z]

is a holomorphic primitive of f (z).

Proof. Take any z ∈ Ω and consider the disk around it which is entirely in Ω. For
any w from this disk, we have [z, w] ⊂ Ω, so that the triangle with vertices a, z, w
lies entirely in Ω:
20 WADIM ZUDILIN

w
z Ω

By Theorem 4.5
Z Z Z
F (w) − F (z) = f (ζ) dζ − f (ζ) dζ = f (ζ) dζ
[a,w] [a,z] [z,w]
Z Z
= f (z) dζ + (f (ζ) − f (z)) dζ
[z,w] [z,w]
Z
= f (z)(w − z) + (f (ζ) − f (z)) dζ.
[z,w]

Now observe that the first term is linear in w, hence holomorphic, while Lemma 4.3
applies to the second term to estimate its absolute value by M L with
M = sup |f (ζ) − f (z)| → 0 as w → z

,
[z,w]

and L = |w − z|. Thus, F 0 (z) = f (z).

Theorem 4.7 (Cauchy integral theorem). Let f : Ω → C be a holomorphic function
on a star-shaped domain Ω ⊂ C. Then
I
f (z) dz = 0
γ
for any closed contour in Ω.
Proof. Take F (z) to be the holomorphic primitive of f (z) constructed in Lemma 4.6.
Then I
f (z) dz = F (z(b)) − F (z(a)) = 0,

,
γ

because z(b) = z(a).

Remark 4.8. One can give a short proof of the Cauchy integral theorem if one is
willing to use Green’s theorem. to do.
Exercise 4.9. Explain why the Cauchy theorem is not applicable to the function
f (z) = 1/z on the domain C× .
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 21

5. Applications of the Cauchy theorem

Consider the function f (z) = 1/z on the star-shaped domain Ω = C \ (−∞, 0],
choose a = 1 as its centre. By Lemma 4.6 the function
Z
dζ
F (z) =
[1,z] ζ

is a holomorphic primitive of f (z) on Ω. We call this function F (z) (the principal

branch of) the logarithm and denote by Log z; by definition it is only defined for
complex numbers z = reiθ with r > 0 and θ = Arg z ∈ (−π, π).
Using the integral calculated in our example preceding Exercise 4.4 we find out
that Z −1+0·i I
dζ dζ
F (−1 + 0 · i) − F (−1 − 0 · i) = = = 2πi,
−1−0·i ζ |ζ|=1 ζ
so that the function F (z) cannot be defined at z = −1 (or at any other z ≤ 0) as con-
tinuous — it does not extend continuously over (−∞, 0]. It is however holomorphic
on Ω = C \ (−∞, 0].
Exercise 5.1. Use the contour

1
−1

and the Cauchy integral theorem to give another proof of

Log(−1 + 0 · i) − Log(−1 − 0 · i) = 2πi.
Furthermore, for z = reiθ we obtain
Z z Z r Z reiθ
dζ dζ dζ
F (z) = Log z = = +
1 ζ 1 ζ r ζ
Z θ it Z θ
d(re )
= log r + = log r + i dt = log r + iθ,
0 reit 0
so that Re Log z = log |z| and Im Log z = Arg z.
Lemma 5.2. The following real integral is valid :
Z ∞
2 2
e−πx e−2πiξx dx = e−πξ .
−∞
22 WADIM ZUDILIN
2
In other words, the Fourier transform of the Gaussian distribution f (x) = e−πx is
the function f (ζ) itself.

x
Proof. We first consider the case ξ = 0:
Z ∞ 2 Z ∞ Z ∞
−πx2 −πx2 2 2 +y 2 )
e dx = e dx · e−πy dy = e−π(x dx dy
−∞ −∞ −∞
R2

(pass to the polar coordinates x = r cos θ, y = r sin θ and use dx dy = r dr dθ)

Z ∞ Z 2π Z ∞
−πr2 1 2 2 ∞
= e r dr θ dθ = 2π · e−πr d(πr2 ) = e−πr 0
=1
0 0 2π 0
implying Z ∞
2
e−πx dx = 1 (5.1)
−∞
because the integral is positive; this proves the formula for ξ = 0. (This integral
was also encountered in Analysis.) Without loss of generality we can further assume
that ξ > 0, as otherwise we replace ξ with −ξ and make the change x 7→ −x in the
original integral.
2
Take Ω = C and f (z) = e−πz . The closed rectangular contour below depends on
the parameter R > 0:
III
−R + iξ R + iξ
iξ
II
IV

−R I R

We have
Z Z R
2
f (z) dz = e−πx dx,
I −R
Z Z R Z R
−π(x+iξ)2 πξ 2 2
f (z) dz = − e dx = −e e−πx e−2πiξx dx.
III −R −R
On part II of the contour we get
2 2 2
M = max |e−πz | = max e−π Re(z ) = max e−π Re(R+it)
0≤t≤ξ
−π(R2 −t2 ) −π(R2 −ξ 2 )
= max e =e → 0 as R → ∞,
0≤t≤ξ

hence Z
f (z) dz ≤ M L = M ξ → 0 as R → ∞.
II
Likewise Z
f (z) dz → 0 as R → ∞.
IV
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 23

At the same time

Z Z Z Z
f (z) dz + f (z) dz + f (z) dz + f (z) dz = 0
I II III IV

by the Cauchy theorem. Substituting our findings into this equality and letting
R → ∞ we conclude with
Z ∞ Z ∞
−πx2 πξ 2 2
e dx − e e−πx e−2πiξx dx = 0.
−∞ −∞

It remains to apply equation (5.1). ,

Lemma 5.3. The following integral is valid :
Z ∞
1 − cos x π
2
dx = .
0 x 2
Proof. This time we deal with the function f (z) = (1 − eiz )/z 2 and use the following
contour, which depends on two parameters ε > 0 and R > ε:

iR
II

IV
III I
−R −ε ε R

A choice of the corresponding star-shaped domain Ω is C with exclusion of the

negative part of the imaginary axis. We have
Z R Z −ε
1 − eix 1 − eix
Z Z
f (z) dz + f (z) dz = dx + dx
I III ε x2 −R x2
Z R Z R Z R
1 − eix 1 − e−ix 1 − cos x
= 2
dx + 2
dx = 2 dx,
ε x ε x ε x2
hence Z  ∞
1 − cos x
Z Z
lim f (z) dz + f (z) dz =2 dx.
ε→0
R→∞ I III 0 x2
For the integral along II we use the fundamental inequality (Lemma 4.3) with
1 − eiz 1+1
M = max 2
≤
|z|=R z R2
and L = πR to conclude that
Z
f (z) dz → 0 as R → ∞.
II
24 WADIM ZUDILIN

To compute the remaining integral along IV, notice that eiz = 1 + iz + z 2 g(z) for
some analytic function g(z) at 0 (in particular, holomorphic in a neighbourhood of
z = 0) with the corresponding holomorphic primitive G(z). Then
−iz
Z Z Z
f (z) dz = 2
dz+ g(z) dz = (−i)(−πi)+G(ε)−G(−ε) → −π as ε → 0,
IV IV z IV
where the first integral in the sum is computed as in the example preceding Ex-
ercise 4.4. It remains to make the limit passage as R → ∞ and ε → 0 in the
sum Z Z Z Z
f (z) dz + f (z) dz + f (z) dz + f (z) dz = 0,

,
I II III IV
which follows from the Cauchy theorem.
The following simple observation belongs to real analysis.
Lemma 5.4 (Jordan’s inequality). For M > 0,
Z π
π
e−M sin t dt < .
0 M
Proof. First note that
Z π Z π/2 Z π Z π/2
−M sin t −M sin t −M sin t
e dt = e dt + e dt = 2 e−M sin t dt,
0 0 π/2 0

since sin t = sin(π − t) for 0 ≤ t ≤ π/2. Second, sin t ≥ 2t/π for 0 ≤ t ≤ π/2 from
the convexity of the sine function:

0 π/2 π

Therefore, e−M sin t ≤ e−2M t/π and

Z π/2 Z π/2
−M sin t π π
e dt ≤ e−2M t/π dt = (1 − e−M ) < ,
2M 2M
,
0 0

and the result follows.

Lemma 5.5 (Jordan’s lemma). Assume that f (z) is a continuous function in the
domain Im z ≥ 0, |z| ≥ R0 for some R0 > 0, while the contour γR for each R ≥ R0
is the half-circle Reit , where 0 ≤ t ≤ π:

γR

−R R
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 25

If
max |f (z)| → 0 as R → ∞,
z∈γR

then for any m > 0 we have

Z
lim eimz f (z) dz = 0.
R→∞ γR

Notice that the result is not implied by Lemma 4.3, because with
M = max |eimz f (z)| = max |f (z)|
z∈γR z∈γR

and L = πR we do not necessarily get M L → 0 as R → ∞.

Exercise 5.6. Give an example of continuous function f (z) on the upper half-plane
H := {z | Im z > 0}, for which
MR = max |f (z)| → 0 as R → ∞,
z∈γR

but πRMR → ∞.
Proof of Lemma 5.5. Indeed, we have
Z Z π
imz it
e f (z) dz ≤ eimRe f (Reit )iReit dt
γR 0
Z π
≤ e−mR sin t R dt · max |f (z)|
0 z∈γR
π
≤ · max |f (z)| → 0 as R → ∞,
m z∈γR
where Jordan’s inequality was applied. ,
As an example of use of Jordan’s lemma we compute the following classical inte-
grals.
Lemma 5.7 (Fresnel’s integrals). The following integrals are valid :
Z ∞ Z ∞ √
2 2 2π
cos(x ) dx = sin(x ) dx = .
0 0 4
2
Proof. We take f (z) = eiz and the contour as the boundary of the following sector:

III

II

π/4
0 I R
26 WADIM ZUDILIN

Then
Z Z R Z R Z ∞ Z ∞
2 2 2
f (z) dz = cos(x ) dx + i sin(x ) dx → cos(x ) dx + i sin(x2 ) dx,
I 0 0 0 0
Z Z π/4 Z π/4
it )2 2 (cos 2t+i sin 2t)
f (z) dz = R ei(Re eit dt = R eiR eit dt
II 0 0
Z π/4 Z π/2
2 R 2 R π
≤R e−R sin 2t
dt = e−R sin t
dt ≤ · →0
0 2 0 2 2R2
by Lemma 5.4 and
R √
1 + i R −x2
Z Z Z
i(eπi/4 x)2 πi/4 1+i π
f (z) dz = − e d(e x) = − √ e dx → − √ ·
III 0 2 0 2 2
by (5.1), as R → ∞. On the other hand, the Cauchy integral theorem implies
Z Z Z
f (z) dz + f (z) dz + f (z) dz = 0;

,
I II III

letting R → ∞ in this equality we conclude with the evaluations claimed.

Exercise 5.8 (Dirichlet’s integral). Prove that
Z ∞
sin x π
dx = .
0 x 2

,
Hint. Use f (z) = eiz /z and the contour from the proof of Lemma 5.3; estimating of
the integral along II requires an application of Jordan’s lemma.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 27

6. The Cauchy integral formula

Throughout the section, a domain Ω is understood as an open simply connected
set in C. This is the convention we motorously use.
Theorem 6.1 (Cauchy integral formula). Given a domain Ω ⊂ C and a point
a ∈ Ω, let a closed disk D(a, r) = {z : |z − a| ≤ r} lie entirely in Ω. If f : Ω → C is
a holomorphic function, then
I
1 f (ζ)
f (z) = dζ
2πi |ζ−a|=r ζ − z
for any z ∈ D(a, r).
Proof. Since Ω is open, we can choose a radius R > r such that D(a, R) ⊂ Ω (and,
clearly, D(a, R) ⊃ D(a, r)). We replace Ω with a smaller domain Ω̂ = D(a, R) \ {cut
from z away from a}; this is star shaped(!) with centre a. Consider the contour
γ = I ∪ II ∪ III ∪ IV as on the picture:

II
IV
z
III
a r

I
R

Here parts II and IV of the contour follow two different banks of the cut (in opposite
directions); since f (ζ)/(ζ −z) is holomorphic on them (in particular, it is continuous
there and assumes the same values on each bank) we have
Z Z
f (ζ) f (ζ)
dζ + dζ = 0.
II ζ − z IV ζ − z

At the same time,

Z Z Z Z
f (ζ) f (ζ) f (ζ) f (ζ)
dζ + dζ + dζ + dζ = 0
I ζ −z II ζ − z III ζ − z IV ζ − z

by Cauchy’s theorem (Theorem 4.7) and

Z I
f (ζ) f (ζ)
dζ = dζ.
I ζ −z |ζ−a|=r ζ − z
28 WADIM ZUDILIN

If ε > 0 stands for the radius of part III of the contour, then
Z I
f (ζ) f (ζ)
dζ = − dζ
III ζ − z |ζ−z|=ε ζ − z
f (ζ) − f (z)
I I
1
=− dζ − f (z) dζ
|ζ−z|=ε ζ −z |ζ−z|=ε ζ − z
→ 0 − f (z) · 2πi as ε → 0,
where we use the estimates |(f (ζ) − f (z))/(ζ − z)| ≤ |f 0 (z)| + 1 and fundamental
inequality (Lemma 4.3) with M = |f 0 (z)|+1, L = 2πε for analysing the intermediate

,
integral. By combining the equalities obtained and considering the limit as ε → 0,
the Cauchy integral formula follows.
Exercise 6.2 (mean-value theorem for holomorphic functions). Let f be holomor-
phic on a disk D(a, R) and continuous on its closure D(a, R). Show that for any
r ≤ R we have Z 2π
1
f (a) = f (a + reit ) dt.
2π 0
Theorem 6.3. Under the hypotheses of Theorem 6.1,
dn
I
n! f (ζ)
n
f (z) = dζ
dz 2πi |ζ−a|=r (ζ − z)n+1
for any z ∈ D(a, r) and any n = 0, 1, 2, . . . . In particular, the derivatives of any
order of a holomorphic function f (z) exist and are holomorhic.
Proof. The statement follows by induction on n = 0, 1, 2, . . . : the differentiation

,
under the integral sign with respect to z is legitimate, because the function g(z) =
(ζ − z)−(n+1) is a differentiable function of z ∈ D(a, r).
Exercise 6.4 (Cauchy inequalities). Let f (z) be holomorphic on a domain contain-
ing the disk D(a, r). Show the inequality
n!
|f (n) (a)| ≤ · sup |f (z)|
rn |z−a|=r
for the nth derivative, where n = 0, 1, 2, . . . .
Hint. Use Theorem 6.3 with z = a and the fundamental inequality. ,
Exercise 6.5. Assume that Ω is star-shaped and a holomorphic function f : Ω → C
does not vanish on Ω. Prove that there exists a holomorphic function g : Ω → C
such that f (z) = eg(z) for all z ∈ Ω.
Deduce from this that for a holomorphic non-vanishing on Ω function f (z) (as
above) there is a holomorphic nth root f (z)1/n for any n ∈ Z \ {0} (that is, there is
a holomorphic h(z) such that h(z)n = f (z) for all z ∈ Ω).

,
Hint. Verify that f 0 (z)/f (z) is holomorphic on Ω and consider its holomorphic prim-
itive.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 29

Exercise 6.6. Recall the definition of a harmonic function (preceding Theorem 2.7).
Prove that a function u(x, y) is harmonic on a (simply connected) domain Ω ⊂ R2
if and only if there exists a holomorphic function f (z) on Ω such that u = Re f .
Hint. Notice that Theorem 6.3 implies that f (z) is of class C ∞ at each z ∈ Ω. ,
Exercise 6.7. Let u(x, y) be harmonic on a domain Ω ⊂ R2 containing the closed
disk {(x, y) : (x − a1 )2 + (y − a2 )2 ≤ r2 }. Prove that
Z 2π
1
u(a1 , a2 ) = u(a1 + r cos t, a2 + r sin t) dt.
2π 0
Theorem 6.8. Let f : Ω → C be holomorphic on Ω ⊂ C, and for a point a ∈ Ω the
disk D(a, r) ⊂ Ω. Then the power series
∞
X f (n) (a)
an (z − a)n with an =
n=0
n!
converges to f (z) on D(a, r). In particular, the function f (z) is analytic at a, hence
also analytic on Ω.
Proof. First of all, from the Cauchy inequalities (Exercise 6.4)
√
n
p
n M
|an | ≤ , where M = sup |f (z)|,
r |z−a|=r

so that
p
n 1
|an | ≤ ,
lim sup
n→∞ r
hence the radius R of convergence of the series in question satisfies R ≥ r (recall
Exercise 3.5). This means that the series indeed converges on D(a, r).
Now for z ∈ D(a, r) use the Cauchy integral formula:
I I
1 f (ζ) 1 f (ζ)
f (z) = dζ = dζ
2πi |ζ−a|=r ζ − z 2πi |ζ−a|=r (ζ − a) − (z − a)
I
1 f (ζ) 1
= · dζ
2πi |ζ−a|=r ζ − a 1 − (z − a)/(ζ − a)
(expand the second factor into the convergent geometric series, using |z − a| <
|ζ − a| = r for z within the disk)
∞ I
X 1 f (ζ)
= n+1
dζ · (z − a)n
n=0
2πi |ζ−a|=r (ζ − a)

,
∞
X
= an · (z − a)n .
n=0

Recall that a function f holomorphic on entire C is called entire. The theorem

we have just proved implies that any such function is automatically analytic on C,
so that the analiticity of the exponential function discussed in Section 3 is not an
apparent fact and holds for entire functions in general.
30 WADIM ZUDILIN

Theorem 6.9 (Liouville). An entire bounded function f is constant.

Proof. By the hypothesis, |f (z)| ≤ M for all z ∈ C. For any a ∈ C, the case n = 1
of the Cauchy inequalities (see Exercise 6.4) reads |f 0 (a)| ≤ M/r for any r > 0,

,
which is only possible when f 0 (a) = 0. Now, f 0 (a) = 0 for any a ∈ C implying
f (z) is constant (Theorem 2.11).
The following result is known by the name ‘fundamental theorem of algebra’.
Theorem 6.10 (Gauss). A polynomial P (z) = z n +a1 z n−1 +· · ·+an of degree n ≥ 1
wth complex coefficients has a zero in C.
Proof. Suppose that P (z) 6= 0 for all z ∈ C, so that f (z) = 1/P (z) is holomorphic
on C. Furthermore, f (z) → 0 as z → ∞, hence |f (z)| ≤ 1 for all |z| > R with a
suitable choice of R; this means that the function f (z) is bounded on C:
n o
|f (z)| ≤ max 1, sup |f (z)| .
|z|≤R

,
By Liouville’s theorem f (z) is constant, which is seen to be 0 because of the limit
at infinity. But then P (z) = 1/f (z) is not defined; contradiction.
Exercise 6.11. Prove that P (z) in Theorem 6.10 admits factorisation (z − z1 ) ×
(z − z2 ) · · · (z − zn ) for some z1 , z2 , . . . , zn ∈ C, not necessarily distinct.
Theorem 6.12 (identity theorem). Let f be holomorphic on a domain Ω. Assume
that there is a point a ∈ Ω and a sequence z1 , z2 , . . . , zn , . . . of distinct points in Ω
converging to a as n → ∞ such that f (zn ) = 0 for all n = 1, 2, . . . . Then f (z) = 0
identically on Ω.
In particular, if for two holomorphic functions f and g on Ω we have f (zn ) = g(zn )
for all elements of the sequence as above, then f (z) and g(z) coincide in Ω.
Proof. The proof is similar in spirit to the proof of Theorem 2.11. We first show it
for Ω = D(a, r), and then use the argument of the theorem to extend the equality
to a general simply connected domain Ω ⊂ C.
In the case Ω = D(a, r) we use Theorem 6.8 to write
X∞
f (z) = an (z − a)n for all z ∈ Ω,
n=0
(n)
where an = f (a)/n! . Since f (z) is continuous and zn → a as n → ∞, we
conclude that 0 = f (zn ) → f (a), hence f (a) = 0. This means that a0 = f (a) = 0.
If a0 = a1 = · · · = am−1 = 0 has been already established, write
X∞ X∞
n
f (z) = an (z − a) = an (z − a)n = (z − a)m h(z),
n=0 n=m

where h(z) = am + am+1 (z − a) + · · · is analytic (hence holomorphic) at z = a.

The equalities 0 = f (zn ) = (zn − a)m h(zn ) translates into h(zn ) = 0 and imply

,
that am = h(a) = 0. This establishes am = 0 for all m, hence f (z) = 0 for all
z ∈ D(a, r).
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 31

Exercise 6.13. Complete the proof for an arbitrary general simply connected do-
main Ω.
The identity theorem has many important consequences. To get a taste for its
strength we indicate the following results.
Theorem 6.14. If f (z) is holomorphic on Ω and it is not identically zero, then all
its zeros are isolated. In particular, if the domain is bounded, then there are only
finitely many zeros of f (z) on the domain.
Theorem 6.15. Let p(w1 , . . . , wk ) be a polynomial in k complex variables, while
f1 (z), . . . , fk (z) be k holomorphic functions on a domain Ω containing a real interval
[a, b]. Assume that p(f1 (x), . . . , fk (x)) = 0 for all x ∈ [a, b]. Then p(f1 (z), . . . ,
fk (z)) = 0 for all z ∈ Ω.
This last property allows us to conclude from the equality ex+y = ex ·ey for all real
x, y that ez+w = ez ew for all z, w ∈ C, as well as to extend the known trigonometric
identities (like sin2 x + cos2 x = 1) from real equalities to the ones valid for complex
variable.
Theorem 6.16 (maximum modulus principle). The maximum of the modulus of
a holomorphic function f : Ω → C on a disk D(a, r) = {z : |z − a| ≤ r} ⊂ Ω is
attained on the boundary C(a, r). In other words, if f (z) is not constant and
Mr (a) = max |f (z)|,
z:|z−a|≤r

then |f (z)| < Mr (a) for all z ∈ D(a, r).

Proof. The Cauchy integral formula (Theorem 6.1)
I
1 f (ζ)
f (z) = dζ for z ∈ D(a, r),
2πi |ζ−a|=r ζ − z
applied for z = a, and fundamental inequality imply that
1 f (ζ)
|f (a)| ≤ max · length C(a, r) = max |f (ζ)|.
2π ζ∈C(a,r) ζ − a ζ∈C(a,r)

Assume on the contrary to the claim that the maximum Mr (a) of |f (z)| is attained
at some b ∈ D(a, r). Take a disk D(b, r0 ) ⊂ D(a, r). The choice of b guarantees
that |f (b)| ≥ |f (z)| for all z ∈ C(b, r0 ), while our argument above implies that
|f (b)| ≤ |f (z)| for all such z. Therefore, |f (z)| = |f (b)| for all z ∈ C(b, r0 ). As this
is true for any choice of disk D(b, r0 ) ⊂ D(a, r), by varying the radius r0 we conclude
that |f (z)| = |f (b)| for all z ∈ D(b, r1 ) for a choice of r1 > 0. By Exercise 2.15 this

,
means that f (z) = f (b), a constant, for all z ∈ D(b, r1 ), hence also f (z) is constant
for all z ∈ Ω; a contradiction.
One corollary of this result is that, for a non-constant holomorphic function
f : Ω → C, its absolute value |f (z)| does not have a local maximum on Ω.
Exercise 6.17. (a) Show that the statement of Theorem 6.16 remains true if we
replace the modulus |f (z)| with Re f (z) and Im f (z).
32 WADIM ZUDILIN

(b) Prove that a non-constant harmonic function u(x, y) on a simply connected

domain Ω ⊂ R2 does not have local extrema in Ω.
Exercise 6.18. Prove that for a non-constant holomorphic function f : Ω → C,
which does not vanish on Ω, its absolute value |f (z)| does not have a local minimum
on Ω.
Theorem 6.19 (Schwarz’s lemma). If f (z) is holomorphic on D(0, 1), it has value
f (0) = 0 at zero and satisfies |f (z)| ≤ 1, then it satisfies a stronger inequality
|f (z)| ≤ |z| in D(0, 1). Furthermore, if |f (a)| = |a| for some a ∈ D(0, 1), a 6= 0,
then |f (z)| = |z| for all z ∈ D(0, 1) and f (z) = eiθ z, for some θ ∈ R, is a rotation.
Proof. The Taylor series expansion f (z) = a0 + a1 z + a2 z 2 + · · · converges (at least)
in D(0, 1), because f (z) is analytic on the disk. We also have a0 = f (0) = 0.
Therefore, g(z) = f (z)/z = a1 + a2 z + · · · is also an analytic function on D(0, 1). If
z ∈ C(0, r) for some 0 < r < 1, then
|f (z)| 1 1
|g(z)| = ≤ = .
|z| |z| r
By Theorem 6.16 we therefore have, for any z ∈ D(0, r),
1
|g(z)| ≤ max |g(ζ)| ≤ ;
ζ∈C(0,r) r
letting r → 1 we conclude that |g(z)| ≤ 1 for all z ∈ D(0, 1). But this means exactly
that |f (z)| ≤ |z| for all z ∈ D(0, 1).
Now assume that |f (a)| = |a| for some a ∈ D(0, 1), a 6= 0. This means that
|g(a)| = 1 and, since a is an internal point of the unit disk, this can only happen

,
when g(z) = c, a constant, such that |c| = 1. Writing c = eiθ we complete the proof
of Schwarz’s lemma.
Exercise 6.20. Show that in Theorem 6.19 we have |f 0 (0)| ≤ 1 for the derivative
of f (z) at zero and that |f 0 (0)| = 1 is only possible for a rotation f (z).
The identity theorem also serves as a natural tool for analytic continuation of
holomorphic functions.
Principle (for uniqueness) of analytic continuation. If f (z) and g(z) are
holomorphic on a (simply connected) domain Ω and coincide on a subdomain Ω0 ⊂ Ω
then f (z) = g(z) for all z ∈ Ω.

,
Proof. Indeed, this follows from the identity theorem (Theorem 6.12) by choosing
a ∈ Ω0 and a sequence of distinct points z1 , z2 , . . . ∈ Ω0 that converges to a.
To illustrate how the principle of analytic continuation works, let us have a look
at some concrete examples.
Example 6.21. Consider a function f (z) given by the (geometric) power series
1 + z + z 2 + . . . . The latter converges inside the unit disk D(0, 1) and defines there
an analytic function; the series diverges for |z| > 1. Consider, on the other hand, the
rational function fˆ(z) = 1/(1 − z) which is holomorphic on Ω = C \ {1}. Its Taylor
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 33

series expansion at z = 0 coincides with the one above, so that f (z) = fˆ(z) for all z
in some simply connected component of Ω. If there is another analytic function that
continues f (z) from the unit disk to Ω, then it should also coincide with fˆ(z); because
the latter is analytic in Ω (and not just in some its simply connected subdomain)
we therefore have an analytic continuation of f : D(0, 1) → C to C \ {1}.
Example 6.22. A similar trick can be played with the function f (z) given in the
unit disk by the power series expansion
∞
z2 z3 z4 X zn
z− + − + ··· = (−1)n−1 .
2 3 4 n=1
n
To extend the latter from D(0, 1) to a larger domain notice that
d
f (z) = 1 − z + z 2 − z 3 + · · ·
dz
and the latter continues analytically to 1/(1 + z) on Ω = C \ {−1} by the previous
example. Therefore,
Z z Z 1+z
ˆ dζ dζ
f (z) = = = log(1 + ζ)
0 ζ +1 1 ζ
provides us with an explicit formula for analytic continuation of the original function
f (z), this time only to a simply connected subdomain of Ω, for example, to Ω0 =
C \ (−∞, −1]. As seen before no holomorphic primitive for 1/(1 + z) exists on Ω.
Example 6.23. This is a somewhat more advanced situation, which shows how
linear differential equations can be used for analytic continuation of their solutions.
For generic complex parameters a, b, c ∈ C, consider the following homogeneous
linear differential equation
d2 y c − (a + b + 1)z dy ab
2
+ − y = 0.
dz z(1 − z) dz z(1 − z)
By taking any z0 ∈ C to be different from 0, 1 we see that the knowledge of a0 = y(z0 )
and a1 = y 0 (z1 ) allows us to compute the value of the second derivative y 00 (z0 ) with
the help of the equation. But we can also differentiate the equation (the coefficients
are rational functions holomorphic on C \ {0, 1}) to compute the values of the third,
fourth, etc., derivatives at z = z0 . This gives us a power series expansion
a0 + a1 (z − z0 ) + a2 (z − z0 )2 + a3 (z − z0 )3 + · · · ,
which converges within the disk centred at z0 of radius r = min{|z0 |, |1 − z0 |} — the
distance to the nearest point in {0, 1}. By the construction, the series satisfies the
differential equation. Because the construction depends on the choice of a0 and a1 ,
we in fact get a solution space spanned by two linearly independent solutions (for
example, corresponding to the choices a0 = 1, a1 = 0 and a0 = 0, a1 = 1). At the
same time the solution space of any second-order linear homogeneous differential
equation is always spanned by two independent solutions. Therefore, if we apply
the procedure at two different point z0 , ẑ0 we will get two pairs of solutions y1 , y2
34 WADIM ZUDILIN

and ŷ1 , ŷ2 (as power series in z − z0 and z − ẑ0 , respectively) such that y1 and y2 are
linear combinations of ŷ1 , ŷ2 with constant coefficients:
   
y1 ŷ
=M 1
y2 ŷ2
for some matrix M ∈ GL2 (C) with det M 6= 0 (the determinant is non-zero, because
we can also write ŷ1 and ŷ2 in terms of y1 , y2 , so that M −1 exists). If the disks D(z0 , r)
and D(ẑ0 , r̂) overlap (where r̂ = min{|ẑ0 |, |1−ẑ0 |}), then the equality defines analytic
continuation of y1 and y2 to the domain D(z0 , r)∪D(ẑ0 , r̂). By applying this strategy
several times (moving from ẑ0 to another z̃0 and so on) we can analytically continue
the original pair of analytic solutions of the differential equation to any other point
in C \ {0, 1}. The uniqueness of such analytic continuation will be only managed
when we continue to a simply connected domain, for example, to C \ (−∞, 1] or
C \ [0, ∞).
Later on we will discuss solutions of the differential equation at points z = 0, 1
(and ∞) and also the monodromy of the equation — connection between solutions
spaces when analytic continuation is executed along two different paths in C\{0, 1}.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 35

7. Laurent series and singularities

We first look for a generalisation of the Cauchy integral formula
I
1 f (ζ)
f (z) = dζ for z ∈ D(a, r),
2πi |ζ−a|=r ζ − z
to the annulus domain (not simply connected!) {z : r1 < |z − a| < r2 }.
Theorem 7.1. Given a domain Ω ⊂ C and a point a ∈ C (not necessarily a ∈ Ω),
assume that the closed annulus {z : r1 ≤ |z − a| ≤ r2 } with r1 < r2 lies entirely in Ω.
If f : Ω → C is a holomorphic function, then
I I
1 f (ζ) 1 f (ζ)
f (z) = dζ − dζ
2πi |ζ−a|=r2 ζ − z 2πi |ζ−a|=r1 ζ − z
for any z with r1 < |z − a| < r2 .
Proof. The proof is similar to the proof of Theorem 7.1; we only make another cut
from the circle |z − a| = r2 to the circle |z − a| = r1 as shown on the picture:

z
r1
a r2

Application of the Cauchy theorem after the limit as ε → 0 for the radius of circle
surrounding the point z leads to
I I
f (ζ) f (ζ)
dζ − 2πif (z) − dζ = 0,
|ζ−a|=r2 ζ − z |ζ−a|=r1 ζ − z

from which the required formula follows. ,

Theorem 7.2 (Laurent series expansion). Under the hypotheses of Theorem 7.1 we
have
X∞
f (z) = an (z − a)n ,
n=−∞

where
I
1 f (ζ)
an = dζ for n ≥ 0
2πi |ζ−a|=r2 (ζ − a)n+1
36 WADIM ZUDILIN

and
I
1 f (ζ)
an = dζ for n < 0.
2πi |ζ−a|=r1 (ζ − a)n+1
The series — known as the Laurent series — converges for z within the annulus r1 <
|z − a| < r2 .
Proof. We use the strategy from our proof of Theorem 6.8. When |z−a| < r2 = |ζ−a|
we get
I I
1 f (ζ) 1 f (ζ)
dζ = dζ
2πi |ζ−a|=r2 ζ − z 2πi |ζ−a|=r2 (ζ − a) − (z − a)
I
1 f (ζ) 1
= · dζ
2πi |ζ−a|=r2 ζ − a 1 − (z − a)/(ζ − a)
∞ I
X 1 f (ζ)
= n+1
dζ · (z − a)n
n=0
2πi |ζ−a|=r2 (ζ − a)
∞
X
= an (z − a)n ,
n=0

where an is given by the formulae in the statement, for n ≥ 0. Here also |an | ≤ M2 /r2n
with M2 = max|ζ−a|=r2 |f (ζ)| by the fundamental inequality (Lemma 4.3) implying
that
lim sup |an |1/n ≤ 1/r2 ,
n→∞

hence the convergence of the resulting series ∞ n

P
n=0 an (z − a) for |z − a| < r2 .
On the other hand, when |z − a| > r1 = |ζ − a| we obtain
I I
1 f (ζ) 1 f (ζ)
− dζ = dζ
2πi |ζ−a|=r1 ζ − z 2πi |ζ−a|=r1 (z − a) − (ζ − a)
I
1 f (ζ) 1
= · dζ
2πi |ζ−a|=r1 z − a 1 − (ζ − a)/(z − a)
I ∞
1 f (ζ) X 1
= · (z − a)m dζ
2πi |ζ−a|=r1 z − a m=0 (ζ − a)m
∞ I
X 1
= f (ζ)(ζ − a)m dζ · (z − a)−m−1
m=0
2πi |ζ−a|=r1
∞
X −1
X
−m−1
= a−m−1 (z − a) = an (z − a)n ,
m=0 n=−∞

where again an are as in the statement, this time for n < 0. Now |an | = |a−m−1 | ≤
M1 r1m+1 with M1 = max|ζ−a|=r1 |fP(ζ)| by Lemma 4.3, hence |a−m−1 |1/m → r1 imply-
ing the convergence of the series ∞ m=0 a−m−1 (z − a)
−m−1
for |z − a| > r1 .
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 37

It remains to combine the two series expansions to the formula deduced in Theo-
rem 7.1:
I I
1 f (ζ) 1 f (ζ)
f (z) = dζ − dζ
2πi |ζ−a|=r2 ζ − z 2πi |ζ−a|=r1 ζ − z
∞
X −1
X ∞
X
n n
= an (z − a) + an (z − a) = an (z − a)n ,

,
n=0 n=−∞ n=−∞

where the series converges if |z − a| < r2 and |z − a| > r1 .

Example 7.3. A particular case of Theorem 7.2, when a = 0 and r1 < 1 < r2 , can
be stated as follows:
∞
X
f (z) = an z n converges for |z| = 1,
n=−∞

with the related formulae for an . By taking F (t) = f (eit ) and choosing the radii
r1 → 1− and r2 → 1+ in the formulae we find out that
X∞
it
F (t) = f (e ) = an eint converges for t ∈ R,
n=−∞

where
Z Z π Z π
1 f (ζ) 1 −inθ 1
an = n+1
dζ = iθ
f (e )e dθ = F (θ)e−inθ dθ.
2πi |ζ|=1 ζ 2π −π 2π −π

The expansion is known as the Fourier series; the corresponding functions t 7→ eint ,
where n ∈ Z, form an orthonormal basis of the space L2 ([−π, π], (2π)−1 dt).
We now move on to classification of isolated singularities of analytic functions,
with the ultimate goal to extend the Cauchy integral formula to the residue sum
theorem, which is one of the principal tools in complex analysis.
Consider a function f : D× (a, r) → C holomorphic on the punctured disk D× (a, r)
= {z : 0 < |z − a| < r}, assuming that the function is not defined (or not differen-
tiable) at z = a. In this case we say that f (z) has an isolated singularity at z = a.
By Theorem 7.2 the Laurent series development
X∞
f (z) = an (z − a)n
n=−∞

converges for 0 < |z − a| < r. We distinguish three possible cases:

(i) a removable singularity, when an = 0 for all n < 0;
(ii) a pole of order m ∈ Z>0 , when a−m 6= 0 and an = 0 for all n < −m;
(iii) an essential singularity, when an 6= 0 for infinitely many n < 0.
In case (i), we can define f (z) to be a0 at z = a; then the power series
∞
X ∞
X
n
f (z) = an (z − a) = an (z − a)n
n=−∞ n=0
38 WADIM ZUDILIN

converges within the disk D(a, r) and defines an analytic function there. This means
that there is actually no singularity of the function f (z) at z = a, hence the name
‘removable’. One example to illustrate the principle is the function f (z) = (sin z)/z,
which is a priori not defined at z = 0, but if we let it to be 1 there and use the
power series
∞
X (−1)n z 2n
n=0
(2n + 1)!
at z = 0, then f (z) becomes analytic on C. This means that the singularity z = 0
is removable.
In case (ii), we can think of f (z) as of the reciprocal 1/g(z) of a holomorphic
function g : D(a, r) → C, which has a zero of order m at z = a. Indeed, writing
g(z) = bm (z − a)m + higher order terms, where bm 6= 0, we see that
1 b−1
m 1
= m
·
g(z) (z − a) 1 + c1 (z − a) + c2 (z − a)2 + · · ·
b−1
m
= · (1 − c1 (z − a) + (c21 − c2 )(z − a)2 + · · · ),
(z − a)m
which is a Laurent series development that starts from (z − a)−m . For example,
f (z) = 1/(z 5 + z 2 ) has a pole of order 2 at z = 0.
Exercise 7.4. Show that all singularities of a rational function f (z) are (isolated
and) poles.
Exercise 7.5. List all poles of the rational function f (z) = 1/(z 5 + z 2 ) and their
respective orders.
A standard illustration for case (iii) is the function
∞
1/z
X z −n
e = .
n=0
n!
This also means that z = ∞ is an essential isolated singularity of ez in view of the
following definition. We say that the point at infinity is an isolated singularity of
f (z) if the function g(z) = f (1/z) is holomorphic in a disk 0 < |z| < r for some
r > 0 and has an isolated singularity at z = 0; the classification of the singularity
of f (z) at z = ∞ is precisely the same as that of g(z) at z = 0.
Finally, the residue of a holomorphic function f : D× (a, r) → C at the isolated
singularity z = a is defined by
I
1
Res f (z) = f (ζ) dζ,
z=a 2πi |ζ−a|=ε
where ε is any in the interval 0 < ε < r. According to Theorem 7.2 we have
Resz=a f (z) = a−1 , the coefficient of (z − a)−1 in the Laurent series.
To define the residue at infinity of a function
X∞
f (z) = an z n
n=−∞
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 39

with the series convergent for |z| > R, consider

∞
X
ĝ(z) = f (1/z) = bn z n
n=−∞

where, clearly, bn = a−n for all n ∈ Z. The latter series converges in the punctured
disk 0 < |z| < 1/R. Then
I I
1 1
Res f (z) = − f (ζ) dζ = f (1/ζ) d(1/ζ)
z=∞ 2πi |ζ|=R1 2πi |ζ|=1/R1
I
1 ĝ(ζ)
=− dζ = −b1 = −a−1
2πi |ζ|=1/R1 ζ 2
for any R1 > R, where the choice of the minus sign at the begining is due to the
positive traversal around infinity (that is, a negative one around the origin). The
formula can be also stated in the form
f (1/z)
Res f (z) = − Res .
z=∞ z=0 z2
Exercise 7.6. Find the residues of the rational function f (z) = 1/(z 5 + z 2 ) at all
its poles and at infinity.
Theorem 7.7 (Riemann removability theorem). Suppose that f (z) is holomorphic
on a punctured disk D× (a, r) and bounded near z = a, that is, |f (z)| ≤ M for all z
in the neighbourhood 0 < |z − a| < ε for some ε > 0. Then z = a is a removable
singularity of f (z).
Proof. Recall that f (z) = ∞ n ×
P
n=−∞ an (z − a) on D (a, r), where
Z
1
an = a−m−1 = f (ζ)(ζ − a)m dζ
2πi |ζ−a|=r1
for n = −m − 1 ≤ −1. Here the radius r1 > 0 can be chosen to be any r1 < r. Since

,
|a−m−1 | ≤ M r1m+1 ≤ M r1 by Lemma 4.3 for any r1 < ε and m ≥ 0, we conclude
that |an | = |a−m−1 | = 0 for all such m.
Theorem 7.8 (Casorati–Sokhotski–Weierstrass). Suppose that f (z) is holomorphic
on a punctured disk D× (a, r) and has an essential singularity at z = a. Then the
image f (D× (a, ε)) is dense in C for all 0 < ε < r.
Proof. Suppose, on the contrary, that there are δ > 0 and b ∈ C such that |f (z)−b| ≥
δ for all z within a punctured disk 0 < |z − a| < ε. Take g(z) = 1/(f (z) − b).
Then g(z) is holomorphic on D× (0, ε) and satisfies |g(z)| ≤ 1/δ. By the Rie-
mann
P∞ removability theorem, the singularity of g(z) at z = a is removable: g(z) =
n
n=0 an (z − a) . Choose m such that am 6= 0 and an = 0 for all n < m; in other
words, m is the order of zero at z = a of the holomorphic function g(z). Then
f (z) = b + 1/g(z) is holomorphic, if m = 0, or has a pole of order m, if m > 0,

,
at z = a. This contradicts the hypothesis that z = a is an essential singularity
of f (z).
40 WADIM ZUDILIN

Theorem 7.8 was published by Weierstrass in 1876 (in German) and by Sokhotski
in 1868 in his master thesis (in Russian). So it was called Sokhotski’s theorem in
the Russian literature and Weierstrass’s theorem in the Western literature. The
same theorem was published by Casorati in 1868, and by Briot and Bouquet in the
first 1859 edition of their book “Theorie des fonctions doublement periodiques, et
en particulier, des fonctions elliptiques”. However, Briot and Bouquet removed this
theorem from the second 1875 edition.
Picard proved a much stronger statement known as Picard’s Big Theorem. Namely,
if f (z) is holomorphic on a punctured disk D× (a, r) and has an essential singularity
at z = a, then then on any punctured neighborhood of a, the function f (z) takes on
all possible complex values, with at most a single exception, infinitely often. The
proof of this theorem goes a bit beyond the scope of this course.
Exercise 7.9. Suppose that z = a is an isolated singularity of a function f (z)
holomorphic in its neighbourhood Ω. Show that a is an essential singularity if and
only if there exist two sequences {zn } and {wn } in Ω both converging to a as n → ∞,
with f (zn ) → 0 and f (wn ) → ∞.
Exercise 7.10. (a) Prove that an entire function has a removable singularity at
infinity if and only if it is a constant.
(b) Prove that an entire function has a pole at infinity of order m if and only if it
is a polynomial of degree m.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 41

8. Residue sum theorem and applications

A set S ⊂ Ω ⊂ C is said to be discrete if for any z ∈ Ω there is a neighbourhood
D(z, ε) ⊂ Ω such that D(z, ε) ∩ S is either empty or contains one point (the latter
situation means that z ∈ S). In other words, a discrete set is made up only of
isolated points.
Exercise 8.1. Show that any discrete subset of a compact K ⊂ C contains finitely
many points.
This result means that for a bounded domain Ω and a contour γ in it, a discrete
subset may have only finitely many points inside the contour. Indeed, the interior
of γ together with the boundary γ itself is a compact set in C.
Theorem 8.2 (residue sum theorem). Let f : Ω → C be given on a star-shaped
domain Ω with centre a. Assume that f (z) is holomorphic on Ω \ S and has isolated
singularities at the points of a discrete subset S ⊂ Ω. Let γ be a closed positively
oriented contour in Ω with {z1 , . . . , zn } ⊂ S inside it and all other points from S
outside. Then I n
1 X
f (ζ) dζ = Res f (z).
2πi γ k=1
z=zk

Proof. Applying the Cauchy integral theorem (Theorem 4.7) to the contour

we obtain I n I
1 X 1
f (ζ) dζ − f (ζ) dζ = 0
2πi γ k=1
2πi |ζ−zk |=ε
for ε > 0 sufficiently small (such that the disks D(zk , ε) do not overlap and all lie
inside γ). On the other hand,

,
I
1
f (ζ) dζ = Res f (z).
2πi |ζ−zk |=ε z=zk
42 WADIM ZUDILIN

Notice that the Cauchy integral theorem is itself a particular case of Theorem 8.2
when S is empty. The Cauchy integral formula corresponds to the choice S = {z}
and function f˜(ζ) = f (z)/(ζ − z), where f (z) is holomorphic on Ω; in this case f˜(ζ)
is holomorphic on Ω \ S with the residue Resζ=z f˜(ζ) = f (z).
Lemma 8.3. For a > 1, the following real integral is valid :
Z 2π
dt 2π
=√ .
0 a + cos t a2 − 1
Proof. Take z = eit , hence dz = ieit dt = iz dt and
Z 2π I I
dt dz 2 dz
= −1
= 2
.
0 a + cos t |z|=1 iz(a + (z + z )/2) i |z|=1 z + 2az + 1
√ 2
2
Since a > 1, among the two √ zeros −a ± a − 1 of the quadratic polynomial √ z +
2az+1 we have z1 = −a+ a2 − 1 inside the unit disk |z| = 1 and z2 = −a− a2 − 1
outside it. Then
1 1 1 1 1
Res 2 = Res = = = √ ,
z=z1 z + 2az + 1 z=z1 (z − z1 )(z − z2 ) z − z2 z=z1 z1 − z2 2 a2 − 1
hence I
1 dz 1
= √ ,
2πi z2 + 2az + 1 2 a2 − 1
,
|z|=1

and the result follows.

Exercise 8.4. Give necessary and sufficient conditions on the real parameter a for
convergence of the improper integral
Z ∞ −ax
e dx
x
.
−∞ 1 + e

Compute it using the residue sum theorem.

Hint. Take Ω = C, f (z) = e−az /(1 + ez ) and the contour γ to be the rectangle with

,
vertices at ±R and ±R + 2πi. Apply the residue sum theorem and then compute
the limit as R → ∞ using the fundamental inequality.
Lemma 8.5. For |p| < 1, the following evaluation is valid :
Z ∞ p
x dx π
= .
0 1 + x2 2 cos(πp/2)
Proof. First notice that the improper integral in question converges at 0 when p >
−1 and at ∞ when p < 1, so that the assumed conditions are indeed required to
insure its convergence.
Take
zp ep log z
f (z) = = ,
1 + z2 1 + z2
where log z = log |z| + iθ for 0 < θ < 2π. Then for the contour
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 43

II

i
I R
IV III
−i

we obtain
R
xp
Z Z
2πip
f (z) dz = (1 − e ) dx,
I∪III ε 1 + x2
while
Rp
Z
f (z) dz ≤ × 2πR → 0 as R → ∞
II R2 − 1
since p + 1 < 2, and
εp
Z
f (z) dz ≤ × 2πε → 0 as ε → 0
IV 1 − ε2
since p + 1 > 0. The only singularities of f (z) within the contour are single poles at
z = ±i, with residues
ep log z epπi/2 ep log z e3pπi/2
Res f (z) = lim = and Res f (z) = lim = .
z=i z→i z + i 2i z=−i z→−i z − i −2i
Applying the residue sum theorem and letting R → ∞, ε → 0 in the result we arrive
at
1 − e2πip ∞ xp epπi/2 − e3pπi/2
Z
pπi πp
dx = = −e · sin
2πi 0 1 + x2 2i 2
implying

,
Z ∞
xp sin(πp/2) sin(πp/2) π
2
dx = −2πi 2πip pπi
=π = .
0 1+x (1 − e )/e sin(πp) 2 cos(πp/2)
Exercise 8.6. For real parameters 0 < m < n, compute the integral
Z ∞ m−1
x
dx.
0 1 + xn
Hint. Consider f (z) = e(m−1) log z /(1+z n ) and the sector of the circle of radius R > 0

,
going along the arc from R to Re2πi/n , for which the origin is excluded by an arc of
radius ε > 0.
44 WADIM ZUDILIN

Lemma 8.7. For 0 < a < 1, the following improper integral takes place:
Z ∞ a−1
x π
J(a) = dx = .
0 1+x sin πa
Proof. We take
e(a−1) log z
f (z) = , where log z = log |z| + iθ for 0 < θ < 2π,
1+z
and the contour γ on the picture

II

−1 I R
IV III

involves arcs of the circles |z| = ε and |z| = R for 0 < ε < 1 < R. The only
singularity of f (z) inside the contour is a simple pole at z = −1 with residue
Res f (z) = lim (1 + z)f (z) = eπi(a−1) .
z=−1 z→−1

The condition 0 < a < 1 implies that the integrals along both arcs tend to 0 as
ε → 0 and R → ∞, respectively. For the line integrals we get
Z R+i0 Z R a−1
x
f (z) dz = dx
ε+0i ε 1+x

for the upper bank and

Z ε−0i R
xa−1
Z
2πi(a−1)
f (z) dz = −e dx.
R−i0 ε 1+x
Therefore, as R → ∞ and ε → 0 we arrive at

π=
e−πi(a−1) − eπi(a−1)
2i
J(a) = − sin π(a − 1) J(a) = sin πa J(a). ,
Exercise 8.8. (a) Let f (z) be a holomorphic function on a disk D(a, r) with a zero
of order k at z = a. Show that there is a holomorphic function g(z) on D(a, ε),
where ε ≤ r, such that
f 0 (z) k
= + g(z).
f (z) z−a
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 45

(b) Let f (z) be a holomorphic function on a punctured disk D× (a, r) with a pole of
order m at z = a. Show that there is a holomorphic function g(z) on D(a, ε),
where ε ≤ r, such that
f 0 (z) −m
= + g(z).
f (z) z−a
(c) Assume that f (z) is a (non-constant) function on a simply connected domain
Ω ⊂ C, whose singularities on Ω are all isolated and not essential. Let γ be
a positively oriented piecewise smooth closed contour in Ω, with no zeroes or
poles of f (z) on it. Show that
I 0
1 f (z)
dz = Z(f ) − P (f ),
2πi γ f (z)
where Z(f ) = Zγ (f ) and P (f ) = Pγ (f ) are the numbers of zeroes and poles,
respectively, inside γ counted each with their order (aka multiplicity).
(d) (Rouché’s theorem) Let Ω ⊂ C be a simply connected domain, γ a simple
piecewise smooth closed contour in Ω, and f (z), g(z) two holomorphic functions
on Ω. Assume that |f (z)| > |g(z)| > 0 for all z ∈ γ. Prove that the number of
zeroes of f (z) and f (z) + g(z) (counted with multiplicity!) inside the contour γ
is the same.
(e) Does the statement in (d) remain necessarily true if the assumption is replaced
with |f (z)| ≥ |g(z)| > 0 (the inequality is nonstrict)?
(f) Show that for a polynomial P (z) = an z n + · · · + a1 z + a0 ∈ C[z] of degree n > 1,
all its zeroes lie within the disk DR (0), where the radius R is any quantity
satisfying
 
|an−1 | + · · · + |a1 | + |a0 |
R > max 1, .
|an |
Hint. (a) + (b) Write f (z) = (z − a)k h(z) and f (z) = (z − a)−m h(z), respectively,

,
and show that the logarithmic derivative g(z) = h0 (z)/h(z) is holomorphic in a
neighbourhood D(a, ε) for some 0 < ε < r.
Exercise 8.9. In this exercise we keep the notation ι : C → C for the complex
conjugation: ι : z 7→ z.
(a) For a function f (z) holomorphic on a domain Ω ⊂ C, define the function g(z)
by g(z) = f (z). Prove that g(z) is holomorphic on ιΩ.
(b) (Schwarz reflection principle) Let Ω be a domain in the upper half-plane {z :
Im z > 0} whose boundary contains a real interval I = (a, b). Assume that
f : Ω ∪ I → C is a function holomorphic on the domain Ω, continuous on Ω ∪ I
and satisfying f (x) ∈ R for all x ∈ I. Define the function g : Ω ∪ ιΩ ∪ I → C by
the formula (
f (z) if z ∈ Ω ∪ I,
g(z) =
f (z) if z ∈ ιΩ.
Show that g(z) is holomorphic on Ω ∪ ιΩ ∪ I.
46 WADIM ZUDILIN

(c) Assume that the series f (z) = ∞ n

P
n=0 an z converges in the disk |z| < r and that
f (x) is real-valued for −r < x < r. Prove that all coefficients an are real and
f (z) = f (z) for all z ∈ D(0, r).
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 47

9. Infinite products and entire functions

Recall that an entire function is an analytic function on C; if we allow a function to
have singularities of finite order (that is, poles), it is called a meromorphic function.
In this part we give analogues of polynomial-type factorisation for entire functions
as well as of partial-fraction decomposition for meromorphic functions.
We first review the concept of partial-fraction decomposition of rational functions,
something we are familiar with from calculus and algebra. Such decomposition is
unique and involves polynomials and terms of the form cjk /(z − zj )k , where the
numbers zj are the poles of a function in consideration, k are positive integers and
cjk are certain constants (from C). For example,
1 1
1 −1 0 1 2 2
= + 2+ 3+ + . (9.1)
z 3 (z 2 + 1) z z z z+i z−i
Here we see that the sum of all residues (at finite points) is equal to (−1)+ 12 + 12 = 0.
In fact, it is easy to control the sum for a general meromorphic function f (z) that
has an isolated singularity at ∞.
Exercise 9.1. (a) Show that if a meromorphic function f (z) has an isolated singu-
larity at ∞, then there are only finitely many poles of it on C.
(b) For a meromorphic function f (z) with poles at z1 , . . . , zn and an isolated singu-
larity at ∞, verify that
Xn
Res f (z) = − Res f (z).
z=zk z=∞
k=1
Hint. (a) Recall that this means that there is a neighbourhood of ∞, that is, an
exterior |z| > R of circle, where f (z) has no singularities. Then use Exercise 8.1

,
and the fact that isolated singularities form a discrete subset of C.
(b) Use the residue sum theorem and definition of the residue at infinity.
In the above example (9.1), the residue at infinity is seen to be 0 because
f (1/z) 1 1 z3
− = − · = − = −z 3 + O(z 4 ),
z2 1/z 3 · (1 + 1/z 2 ) z 2 1 + z2
so that the coefficient of z vanishes.
There are analogues of representation (9.1) for more general meromorphic func-
tions; instead of giving this general result here, we will restrict ourselves to discussing
an important example, the function cot πz.
The function
cos πz
cot πz =
sin πz
has period 1 and simple poles at the points z = n where n ∈ Z with residues
cos πz 1
Res cot πz = Res cot πz = lim z = .
z=n z=0 z→0 sin πz π
Take any complex a ∈ / Z and consider the function
cot πz
f (z) = .
z−a
48 WADIM ZUDILIN

This has simple poles at z = n for n ∈ Z with residues

1 1
Res f (z) = lim (z − n)f (z) = Res cot πz =
z=n z→n n − a z=n π(n − a)
and also a simple pole at z = a with residue
Res f (z) = lim (z − a)f (z) = cot πa.
z=a z→a

We take γ to be a positively oriented rectangle with vertices ±(N + 21 ) ± iN , where

N > |a| is large.
We form the integral I
1 cot πz
IN (a) = dz.
2πi γ z − a
By the residue sum theorem
N N
1 X 1 1 X 1
IN (a) = cot πa + = cot πa − .
π n=−N n − a π n=−N a − n

The change of variable z 7→ −z brings us to

I
1 cot πz
IN (a) = − dz
2πi γ z + a
(the contour is not affected!), so that
 I I  I
1 1 cot πz 1 cot πz a cot πz
IN (a) = dz − dz = dz.
2 2πi γ z − a 2πi γ z + a 2πi γ z 2 − a2
To estimate the latter integral on the upper (similarly, lower) side z = iN + t, t ∈ R,
of the rectangle we observe that
eπi(iN +t) + e−πi(iN +t) e−πN eπit + eπN e−πit
| cot πz| = = → 1 as N → ∞,
eπi(iN +t) − e−πi(iN +t) e−πN eπit − eπN e−πit
because |e±πit | = 1 and eπN → ∞. In particular,
| cot πz| ≤ 2 for N ≥ N0
on the horizontal sides of the contour. For the vertical sides, observe that eπi(N +1/2) =
(−1)N i, so that for z = N + 12 + it we get
e−πt (±i) + eπt (∓i) e−πt − eπt e2πt − 1
| cot πz| = = = ≤ 1.
e−πt (±i) − eπt (∓i) e−πt + eπt e2πt + 1
We also have |z 2 − a2 | ≥ N 2 − |a|2 for z ∈ γ, while the length of the contour (that is,
the perimeter of rectangle) is equal to (2N ) + (2N + 1) + (2N ) + (2N + 1) = 8N + 2.
Therefore,
|a| 2
|IN (a)| ≤ · 2 · (8N + 2) → 0 as N → ∞.
2π N − |a|2
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 49

This implies that

N  N  
1 X 1 1 1 X 1 1
cot πa = lim = + lim +
π N →∞ n=−N a − n π a N →∞ n=1 a − n a + n
 ∞ 
1 1 X 2a
= + .
π a n=1 a2 − n2
On using the fact that the resulting series is bounded from above by
∞
X 1
C
n=1
n2
on any compact domain Ω ⊂ C \ Z, where C only depends on Ω, we conclude with
the following theorem.
Theorem 9.2. The partial-fraction decomposition
∞
1 X 2z
π cot πz = +
z n=1 z 2 − n2
is valid for all z ∈
/ Z. The series converges absolutely and uniformly on every
compact set in C \ Z.
Notice that
d
π cot πz = log sin πz
dz
is the logarithmic derivative of sin πz, while the right-hand side in Theorem 9.2 is
the derivative of ∞
z2
X  
log z + log 1 − 2 .
n=1
n
To legalise the integration
∞
z2
X  
log sin πz = C + log z + log 1 − 2 ,
n=1
n
we first perform it on compacts in the half-plane Re z > 0, where the convergence
is uniform. It implies, after exponentiation, that
∞ 
z2
Y 
sin πz = cz 1− 2 .
n=1
n
Here both sides define entire functions, so that the equality must be true for all
z ∈ C. Finally, we see that c = π by dividing the both sides of equality by z and
taking the limit as z → 0.
Theorem 9.3. We have the following infinite product for the sine function,
∞ 
z2
Y 
sin πz = πz 1− 2 ,
n=1
n
which is valid for all complex z.
50 WADIM ZUDILIN

Exercise 9.4 (Euler). Prove that for the values (of Riemann’s zeta function)
∞
X 1
ζ(2k) = , where k = 1, 2, . . . ,
n=1
n2k
we always have ζ(2k)/π 2k ∈ Q; in other words, ζ(2k) is a rational multiple of π 2k .
In particular, show that
π2 π4
ζ(2) = and ζ(4) = .
6 90
In the formula obtained in Theorem 9.3 we may write
z2
  
z z
1− 2 = 1− 1+ for n = 1, 2, . . . .
n n n
However, this factorisation does not permit to write the product in the form
∞  
Y z
1− ,
n=−∞
n
n6=0

since this product diverges already for real values of z > 0. Nevertheless, the
convergence can be repaired by the following modification.
We start from the divergent series
∞  
X z
log 1 −
n=1
n
and construct a convergent series from it. We restrict z to the disk |z| ≤ R and
choose N ≥ 2R. Then, for every n ≥ N ,
   2  3
z z 1 z 1 z
log 1 − =− − − − ··· ,
n n 2 n 3 n
hence
   2  3
z z 1 z 1 z
log 1 − + ≤ + + ···
n n 2 n 3 n
 2  3
1 |z| 1 |z|
≤ + + ···
2 n 3 n
1 |z|2 1 R2
≤ ≤ ;
2 n2 1 − |z|/n n2
this implies that the series
∞    
X z z
log 1 − +
n=N
n n
converges absolutely and uniformly in the disk |z| ≤ R. Then each of the products
∞   −N   ∞  
Y z z/n Y z z/n Y z −z/n
1− e and 1− e = 1+ e
n=N
n n=−∞
n n=N
n
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 51

converges absolutely and uniformly in the disk |z| ≤ R, hence the same is true for
their product
∞  
Y z z/n
1− e .
n=−∞
n
n6=0

If we combine in pairs those factors for which the indices k have opposite signs,
we see that the product has the same value as the product
∞ 
z2
Y 
1− 2 .
n=1
n

Thus, the product representation in Theorem 9.3 of the function sin πz can be
written as
∞  
Y z z/n
sin πz = πz 1− e .
n=−∞
n
n6=0

Exercise 9.5. Use Theorem 9.3 to give a product representation of ez − 1, with all
factors linear.

,
Hint. Write the function as ez/2 f (z) and use the infinite product expansion of f (2iz).

As we already know from Exercise 6.5, if f (z) is an entire function without zeros
on C then f (z) = eg(z) for some entire function g(z).
The following chain of statements explains more generally the structure of entire
functions that have zeros on C. The particular instance of the product formula for
sin πz serves as an illustration to this general result.
Exercise 9.6. (a) If an entire function f (z) vanishes at finitely many nonzero points
z1 , . . . , zN (multiple entries on the list correspond to the multiplicity of the cor-
responding zero), then
N  
g(z)
Y z
f (z) = e 1−
n=1
zn

for some entire function g(z).

(b) Define P0 (z) = 0 and
m
X zn
Pm (z) =
n=1
n
for m ≥ 1 (this is a truncation of the Taylor series for − log(1 − z)). Give a
(reasonably sharp) estimate for
| log(1 − z) + Pm (z)|
inside a disk of radius r, where r < 1.
52 WADIM ZUDILIN

(c) For a sequence of nonzero (not necessarily distinct) complex numbers z1 , z2 , . . .

ordered according to their absolute values,
0 < |z1 | ≤ |z2 | ≤ · · · ,
assume that |zn | → ∞ as n → ∞. Show that one can choose a sequence of
nonnegative integers m1 , m2 , . . . such that for any R ≥ 1 there is N = N (R) ≥ 1
for which the sum
∞     
X z z
log 1 − + Pmn
n=N
zn zn
converges absolutely (hence uniformly) inside the disk |z| ≤ R.
(d) Show that for a sequence z1 , z2 , . . . of complex numbers and the related sequence
m1 , m2 , . . . of nonnegative integers from part (c), the infinite product
∞  
Y z
1− ePmn (z/zn )
n=1
zn

is an entire function.
(e) (Weierstrass product formula) Suppose that an entire function f (z) has zero
z = 0 of multiplicity M ≥ 0 and zeros z1 , z2 , . . . (a possibly infinite set), each
occurring according to its multiplicity; without loss of generality they are or-
dered:
0 < |z1 | ≤ |z2 | ≤ · · · ≤ |zn | ≤ · · · .
Then there exist a sequence of nonnegative integers m1 , m2 , . . . and an entire
function g(z) such that
∞  
g(z) M
Y z
f (z) = e z 1− ePmn (z/zn ) .
n=1
zn
The choice of sequence m1 , m2 , . . . and function g(z) is not unique; in our example
for sin πz we use the optimal choice m1 = m2 = · · · = 1, in which case g(z) = π
is constant. We also remark that the Weierstrass product formula can be used to
represent a meromorphic function with given zeros and poles as a quotient of two
entire functions.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 53

10. Gamma function

Our first example of a non-elementary function is the Gamma function Γ(z) that
can be introduced and characterised in several ways; for example, this is a mero-
morphic function which interpolates the factorial: Γ(n + 1) = n! for n = 0, 1, 2, . . . .
Historically, the Gamma function was first defined by Euler as the limit of a certain
product from which one can identify it with the improper integral
Z ∞
e−t tz−1 dt.
0
The construction of the Gamma function involves use of a certain constant, known
in mathematics as the Euler (or Euler–Mascheroni ) constant,
γ = 0.57721566490153286060651209008240243104215933593992 . . . .
The following result is discussed in Analysis 1 course.
Exercise 10.1 (Euler’s constant γ). The sequence
1 1 1
1+ + + · · · + − log(n + 1)
2 3 n
converges to a finite limit. The limit is called γ.
Hint. First show that the sequence
Z 1 Z 1 
t dt 1 1 1 n+1
an = = − dt = − log
0 n(n + t) 0 n n+t n n

,
is positive and bounded above by 1/n2 . Then apply the comparison test to the series
P∞
n=1 an (which is precisely the sequence in consideration).

By the results in Section 9 (in particular, Exercise 9.6), the Weierstrass product
∞  
Y z −z/n
z 1+ e
n=1
n
defines an entire function whose only zeros are at z = 0, −1, −2, . . . . (Roughly
speaking, this is ‘half of the sine product’.)
The Gamma function is defined by the equation
∞  
1 γz
Y z −z/n
= ze 1+ e (10.1)
Γ(z) n=1
n
where γ is Euler’s constant (from Exercise 10.1). It follows from this definition that
Γ(z) is a meromorphic function with simple poles at z = 0, −1, −2, . . . .
Example 10.2. Let us find the values Γ(1) and Γ0 (1).
In a compact neighbourhood of z = 1 we have
∞   
X z z
− log Γ(z) = log z + γz + − + log 1 + .
n=1
n n
54 WADIM ZUDILIN

Substitution z = 1 results in
∞    ∞
X 1 1 X
− log Γ(1) = γ + − + log 1 + =γ− an = 0
n=1
n n n=1

for the sequence an defined in the proof of Exercise 10.1; hence Γ(1) = 1.
Because of the uniform convergence of the series in the compact neighbourhood
of z = 1, we can also differentiate it termwise to obtain
∞ 
Γ0 (z)

1 X 1 1/n
− = +γ+ − + .
Γ(z) z n=1
n 1 + z/n
Taking z = 1 we get
∞ 
Γ0 (1)

X 1 1
− =1+γ+ − + =γ
Γ(1) n=1
n n+1
(the intermediate series telescopes), hence Γ0 (1) = −γ.
Theorem 10.3 (Euler’s formula for the Gamma function). For z ∈ C\{0, −1, −2, . . . },
∞  z  −1
1Y 1 z
Γ(z) = 1+ 1+ .
z n=1 n n

Proof. Comparing the required formula with (10.1), it is sufficient to show

∞   ∞  −z  
γz
Y z −z/n Y 1 z
e 1+ e = 1+ 1+
n=1
n n=1
n n
on any compact in C \ (−∞, 0] and then use the analytic continuation to C \
{0, −1, −2, . . . }. The equality in turn reduces to computing the limit, as N → ∞,
of
N   N  −z  
Y z −z/n Y 1 z
γz + log 1+ e − log 1+ 1+
n=1
n n=1
n n
N N   XN 
X z X 1 1
= γz − +z log 1 + = γz − z − log(N + 1)
n n n

,
n=1 n=1 n=1

(the sum of logarithms telescopes); this is seen to be 0 by Exercise 10.1.

Example 10.4. The resulting formula can be given in the following alternative form
(also due to Euler):
N −1 z  −1
1 Y 1 z
Γ(z) = lim 1+ 1+
z N →∞ n=1 n n
1 N z · 1 · 2 · · · (N − 1)
= lim
z N →∞ (z + 1)(z + 2) · · · (z + N − 1)
1 · 2 · · · (N − 1)
= lim N z.
N →∞ z(z + 1)(z + 2) · · · (z + N − 1)
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 55

We are ready to establish some important properties of the Gamma function.

Theorem 10.5 (functional equation). The Gamma function satisfies the difference
equation
Γ(z + 1) = zΓ(z).
Proof. By Theorem 10.3, if z is not a negative integer,
N N
(1 + n1 )z+1 (1 + n1 )z
 
Γ(z + 1) 1 Y 1 Y
= lim lim
Γ(z) z + 1 N →∞ n=1 1 + z+1 n
z N →∞ n=1 1 + nz

,
N
z Y (1 + n1 )(z + n) N +1
= lim = z lim = z.
z + 1 N →∞ n=1 z + n + 1 N →∞ z + N + 1

Exercise 10.6. Use Example 10.2 to show that Γ(n + 1) = n! for n = 0, 1, 2, . . . .

Theorem 10.7 (reflection formula). The Gamma function satisfies the relation
π
Γ(z)Γ(1 − z) = .
sin πz
Proof. By the definition of the Gamma function and by Theorem 9.3,
∞   ∞  
1 1 2
Y z −z/n Y z z/n z sin πz
· = −z 1+ e · 1− e =− .
Γ(z) Γ(−z) n n π

,
n=1 n=1

Since Γ(1 − z) = −zΓ(−z), the desired formula follows.

√
Exercise 10.8. Prove that Γ( 21 ) = π.
Exercise 10.9. Prove that the so-called digamma function Ψ(z) = Γ0 (z)/Γ(z) sat-
isfies
Ψ(1 − z) − Ψ(z) = π cot πz.
Theorem 10.10 (multiplication theorem). The Gamma function satisfies the rela-
tion
 1  2  n − 1
Γ(z)Γ z + Γ z+ ···Γ z + = (2π)(n−1)/2 n−nz+1/2 Γ(nz).
n n n
Sketch of proof. Using Euler’s formula from Example 10.4 one verifies that the ex-
pression
 1  2  n − 1
nnz Γ(z)Γ z + Γ z+ ···Γ z +
f (z) = n n n
nΓ(nz)
is independent of z; thus, one can compute it for a particular value of z. Since
f (1/n) > 0 and
 1 2 n−1
Y j   j π n−1 (2π)n−1
f = Γ Γ 1− = = ,
n n n π 2π (n − 1)π n
j=1 sin sin · · · sin
n n n
the theorem follows.
56 WADIM ZUDILIN

The auxiliary trigonometric identity

π 2π (n − 1)π n
sin sin · · · sin = n−1
n n n 2
can be shown as follows. Note that the set {e2πij/n : j = 0, 1, . . . , n − 1} exhausts
all roots of unity of degree n, that is,
n−1
Y tn − 1
(t − e2πij/n ) = = tn−1 + · · · + t + 1.
j=1
t−1

Put t = 1 in this equality:

n−1
Y n−1
Y
2πij/n n−1
n= (1 − e ) = (−1) (e2πij/n − 1)
j=1 j=1
n−1
Y
= (−1) n−1
eπij/n (eπij/n − e−πij/n )
j=1
n−1
Y
= (−1) n−1 πi(1+2+···+(n−1))/n
e (eπij/n − e−πij/n )
j=1
n−1 n−1
n−1 πi(n−1)/2
Y πj  n−1 n−1 n−1
Y πj
= (−1) e 2i sin = (−1) i (2i) sin
j=1
n j=1
n
n−1
Y πj
= 2n−1 sin ,
j=1
n

and the result follows. ,

Example 10.11. Taking n = 2, we have the duplication formula,
2z−1
 1 √
2 Γ(z)Γ z + = π Γ(2z).
2
With the help of the Weierstrass M-test for improper integrals, one finds out that
the integral
Z ∞
Γ1 (z) = e−t tz−1 dt
0
converges uniformly on any half-plane Re z ≥ ε, hence it defines an analytical func-
tion of z on Re z > 0. The integral is called the Eulerian integral of the second
kind.
Theorem 10.12. If Re z > 0, then
Z ∞
e−t tz−1 dt = Γ(z).
0

We first establish an auxiliary inequality.

 NWI-WB093: COMPLEX ANALYSIS (2023/24) 57

Exercise 10.13 (Bernoulli’s inequality). Show that if 0 ≤ α ≤ 1, then (1 − α)n ≥

1 − nα.
This is, in fact, a particular case of general Bernoulli’s inequality (presumably fa-
miliar from Calculus): (1 + β)n ≥ 1 + nβ for any integer n ≥ 0 and any real β ≥ −1.

Hint. Using induction when nα < 1; it is also obvious when nα ≥ 1. ,

Lemma 10.14. For n a positive integer and real t, 0 ≤ t ≤ n, we have
n
e−t t2

−t t
0≤e − 1− ≤ .
n n

Proof. Using the Maclaurin series for ex and (1 − x)−1 we have the inequalities

1
1 + x ≤ ex ≤ for 0 ≤ x < 1.
1−x

Writing t/n for x, we conclude that

 n  n
t t −t t
e ≥ 1+ and e ≥ 1− ,
n n

and so
 n   n 
−t t −t t t
0≤e − 1− =e 1−e 1−
n n
n  n  n 
t2
   
−t t t −t
≤e 1− 1+ 1− =e 1− 1− 2 .
n n n

Now, applying the inequality of Exercise 10.13 with α = t2 /n2 we get

n
t2 t2

1− 1− 2 ≤ ;
n n

this together with the estimates above imply the required inequalities. ,
Proof of Theorem 10.12. For a positive integer n, denote
Z n n
t
Π(z, n) = 1− tz−1 dt.
0 n

Writing t = nτ we have
Z 1
z
Π(z, n) = n (1 − τ )n τ z−1 dτ.
0
58 WADIM ZUDILIN

Since Re z > 0 and n is a positive integer, by repeated integrations by parts we find

that
Z 1
n 1
Z
n z−1 1 z n
1
(1 − τ ) τ dτ = τ (1 − τ ) + (1 − τ )n−1 τ z dτ = · · ·
0 z 0 z 0
Z 1
n(n − 1) · · · 1
= τ z+n−1 dτ
z(z + 1) · · · (z + n − 1) 0
n(n − 1) · · · 1
=
z(z + 1) · · · (z + n − 1)(z + n)
1 · 2 · · · (n − 1) n
= · .
z(z + 1) · · · (z + n − 1) n + z
By Euler’s representation from Example 10.4 we conclude that Π(z, n) → Γ(z) as
n → ∞, hence
Z n n
t
Γ(z) = lim 1− tz−1 dt.
n→∞ 0 n
Write
Z n   n  Z ∞ 
−t t z−1 −t z−1
Γ1 (z) − Γ(z) = lim e − 1− t dt + e t dt .
n→∞ 0 n n

Now Z ∞
lim e−t tz−1 dt = 0
n→∞ n
as a tail of the absolutely convergent integral for Γ1 (z). In turn, Lemma 10.14
implies that
Z n  n  Z n −t Re z+1
−t t z−1 e t
e − 1− t dt ≤ dt
0 n 0 n
1 ∞ −t Re z+1
,
Z
Γ1 (Re z + 2)
< e t dt = → 0 as n → ∞.
n 0 n
The name Eulerian integral of the first kind was given by Legendre to the integral
Z 1
B(p, q) = xp−1 (1 − x)q−1 dx.
0

In this integral, the real parts of p and q are supposed to be positive to insure
convergence; and xp−1 , (1 − x)q−1 are to be understood as principal values e(p−1) log x
and e(q−1) log(1−x) (when 0 < x < 1, both log x and log(1 − x) assume real values).
Example 10.15. On writing 1 − x for x, we have
B(p, q) = B(q, p).
Also, integrating by parts,
Z 1 1 1
xp (1 − x)q
Z
q
xp−1 (1 − x)q dx = + xp (1 − x)q−1 dx,
0 p 0 p 0
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 59

so that
q
B(p, q + 1) = B(p + 1, q).
p
Furthermore, by integrating
xp−1 (1 − x)q−1 = xp (1 − x)q−1 + xp−1 (1 − x)q
we deduce that
B(p, q) = B(p + 1, q) + B(p, q + 1).
Combining the two final identities we get
q
B(p, q + 1) = B(p, q).
p+q
The next result, discovered by Euler, connects the Eulerian integral of the first
kind with the Gamma function.
Theorem 10.16. If Re p > 0 and Re q > 0, the following identity is valid :
Γ(p)Γ(q)
B(p, q) = .
Γ(p + q)
Proof. Write
Z ∞ Z ∞ Z R
−t p−1 −x2 2p−1 2
Γ(p) = e t dt = 2 e x dx = 2 lim e−x x2p−1 dx
0 0 R→∞ 0
and, similarly, for Γ(q) to conclude that
x
Γ(p)Γ(q) = 4 lim f (x, y) dx dy,
R→∞
[0,R]2

where
2 2
f (x, y) = e−(x +y ) x2p−1 y 2p−1 .
Denote by SR the circular sector x2 + y 2 ≤ R, x ≥ 0, y ≥ 0, and by TR the part of

x x x
the square [0, R]2 without the sector SR . Then

f (x, y) dx dy − f (x, y) dx dy = f (x, y) dx dy

[0,R]2
x SR
x TR
x
≤ |f (x, y)| dx dy ≤ |f (x, y)| dx dy − |f (x, y)| dx dy → 0
TR [0,R]2 [0,R/2]2

as R → ∞, because both of the latter integrals have the same finite limit
Z ∞
Z ∞ Z ∞ Z ∞
−x2 2 Re p−1 2
|f (x, y)| dx dy = e x dx · e−y y 2 Re q−1 dy
0 0 0 0
1
= Γ(Re p) Γ(Re q)
4
as R → ∞.
x
Now pass to the polar coordinates x = r cos θ, y = r sin θ in the integral
f (x, y) dx dy
SR
60 WADIM ZUDILIN

using dx dy = r dr dθ and the fact that the sector SR is continously mapped onto
the domain 0 ≤ r ≤ R, 0 ≤ θ ≤ π/2 by the change of variables. We deduce that
Z R Z π/2
2
Γ(p)Γ(q) = 4 lim e−r (r cos θ)2p−1 (r sin θ)2q−1 r dr dθ
R→∞ 0 0
Z ∞ Z π/2
−r2 2p+2q−1
=4 e r dr cos2p−1 θ sin2q−1 θ dθ
0 0
Z π/2
= 2Γ(p + q) cos2p−1 θ sin2q−1 θ dθ.
0
2
Finally, another change x = cos θ leads to

,
Z π/2
2 cos2p−1 θ sin2q−1 θ dθ = B(p, q).
0

Exercise 10.17. Establish the following evaluation of trigonometric integral:

1 Γ( m2 )Γ( n2 )
Z π/2
1
cosm−1 x sinn−1 x dx = B m2 , n2 = .
0 2 2 Γ( m+n
2
)
Give closed forms of the integral when m and n are positive integers.
Hint. Use the substitution cos2 x = t. ,
Although the asymptotic behaviour of the meromorphic function Γ(z) as |z| → ∞
(along different directions) goes far beyond the scopes of present lectures, it is an
important part of the theory of the Gamma function. Without giving details (which
are tedious enough), we state here only the final result.
Theorem 10.18. If | arg z| ≤ π − δ for a fixed positive δ > 0, then
 1 √
log Γ(z) = z − log z − z + log 2π + f (z),
2
where f (z) → 0 as |z| → ∞ inside the domain.
Example 10.19. A particular case of Theorem 10.18, when z is restricted to the
positive real axis, is known in the literature as Stirling’s formula:
 x x √

Γ(x + 1) = 2πx 1 + o(1) as x → ∞.
e
This asymptotics finds many applications in combinatorics, since many combinato-
rial characteristics can be expressed through the factorial function n! = Γ(n + 1).
Exercise 10.20. Compute the
leading term in the asymptotic behaviour of the
2n
central binomial coefficient n as n → ∞.
Hint. One economical way of solving the problem is through the observation that
Γ(n + 12 )
,
 
2n 2n
=2 .
n Γ( 12 ) Γ(n + 1)
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 61

11. Hypergeometric function

P∞ n
Long time ago, in Lemma 3.2 we discussed the geometric series n=0 z . Another
way to think of it is by saying that a series ∞
P
n=0 na is geometric if the quotient of its
two consecutive terms an+1 /an is constant. More generally, such a series is said to be
hypergeometric if the quotient an+1 /an is a rational function of index parameter n.
Example 11.1. A motivation behind the definition is that the power series for all
standard elementary functions from high school, including the exponential, logarith-
mic functions, sine and cosine, are hypergeometric series.
The definition of hypergeometric series means that
an+1 P (n)
=
an Q(n)
for some polynomials P (n), Q(n) ∈ C[n]. By collecting the zeros of those polynomi-
als and separating the leading coefficient we can further write
an+1 (n + α1 )(n + α2 ) · · · (n + αp )
=z .
an (n + β1 )(n + β2 ) · · · (n + βq )
Now
n−1
Y (α1 + j)(α2 + j) · · · (αp + j)
a1 a2 an n
an = a0 · · ··· = a0 z .
a0 a1 an−1 j=0
(β 1 + j)(β 2 + j) · · · (βq + j)
This in turn implies that a general hypergeometric series has the form
∞
X (α1 )n (α2 )n · · · (αp )n n
a0 z ,
n=0
(β1 )n (β2 )n · · · (β q )n

where the convenient notation

(
1 if n = 0,
(a)n =
a(a + 1) · · · (a + n − 1) if n ≥ 1,
for the product of n consecutive numbers starting from a is introduced. (The latter
is known as Pochhammer’s symbol, or shifted factorial, because in case a = 1 we
clearly get the factorial (1)n = n!). Notice that with the help of the Gamma function
(Theorem 10.5) we can write
Γ(a + n)
(a)n = for n = 0, 1, . . . .
Γ(a)
One can naturally normalise the series by a0 = 1, so that the constant term is
equal to 1 (we used the same normalisation for the geometric series). The resulting
object
∞
X (α1 )n (α2 )n · · · (αp )n n
z
n=0
(β1 )n (β2 )n · · · (βq )n
is a formal series, whose convergence is still to be determined.
Exercise 11.2. Show that the hypergeometric series above
62 WADIM ZUDILIN

(a) converges for any z ∈ C if p < q;

(b) converges inside the unit disk |z| < 1 if p = q (this generalises Lemma 3.2);
(c) diverges for z 6= 0 if p > q.

Hint. The convergence of the series can be determined by the ratio test. ,
The general hypergeometric series viewed as a function of single complex variable
z is known in anaysis as a generalised hypergeometric function. There is a quite
advanced theory of such functions. In order to get a taste to it and witness its
importance, in this final part of the course we take a more careful look at the classical
particular case when p = q = 2, α1 = a, α2 = b, β1 = 1 and β2 = c. Thus, the
Euler–Gauss hypergeometric function (without the adjective ‘generalised’) is defined
as
∞
X (a)n (b)n
F (a, b; c; z) = zn
n=0
n!(c)n
a·b a(a + 1) · b(b + 1) 2
=1+ z+ z
1·c 1 · 2 · c(c + 1)
a(a + 1)(a + 2) · b(b + 1)(b + 2) 3
+ z + ··· .
1 · 2 · 3 · c(c + 1)(c + 2)
Exercise 11.3. Show the following expressions of elementary functions through the
hypergeometric function:

log(1 + z) = zF (1, 1; 2; −z),

(1 + z)−a = F (a, b; b; −z),
ez = lim F (1, b; 1; z/b),
b→∞
arcsin z = zF ( 21 , 12 ; 32 ; z 2 ),
arctan z = zF ( 21 , 1; 23 ; −z 2 ).

Denote
(a)n (b)n
an =
n!(c)n
the nth coefficient of the hypergeometric series F (a, b; c; z), so that
∞
X
F (a, b; c; z) = an z n
n=0

is a representation of the hypergeometric function within the unit disk |z| < 1.
Because of the relation

(1 + n)(c + n) · an+1 = (a + n)(b + n) · an for n = 0, 1, 2, . . . ,

 NWI-WB093: COMPLEX ANALYSIS (2023/24) 63

we have
     X∞
d d d d (a)n (b)n n
z z +a z + b F (a, b; c; z) = z z +a z +b z
dz dz dz dz n=0
n!(c)n
∞ ∞
X (a)n (a + n) · (b)n (b + n) n
X (a)n+1 (b)n+1
=z z = z n+1
n=0
n!(c)n n=0
n!(c)n
∞ ∞
X (a)n (b)n X (a)n (b)n · n(c + n) n
= zn = z
n=1
(n − 1)!(c) n−1 n=0
n!(c) n
  X ∞
d d (a)n (b)n n
= z z +c−1 z
dz dz n=0
n!(c) n
  
d d
= z z + c − 1 F (a, b; c; z).
dz dz
Theorem 11.4. The hypergeometric function F (a, b; c; z) satisfies the differential
equation
      
d d d d
z z +a z +b − z z + c − 1 y = 0;
dz dz dz dz
in equivalent form,
d2 y dy
z(1 − z) 2 + (c − (a + b + 1)z) − aby = 0.
dz dz
Note that the differential equation allows us to continue the hypergeometric func-
tion F (a, b; c; z) analytically to the complex plane with the cut (that is, the ‘im-
passable barrier’) from 1 to +∞ along the real axis. This was discussed earlier in
Example 6.23.
Exercise 11.5. Show that the function
y(z) = z 1−c F (a + 1 − c, b + 1 − c; 2 − c; z)
satisfies the differential
d2 y dy
z(1 − z) 2
+ (c − (a + b + 1)z) − aby = 0.
dz dz
(Note that F (a, b; c; z) satisfies the same differential equation.)
Theorem 11.6 (Pochhammer’s integral). If Re c > Re b > 0 and |z| < 1, then
Z 1
Γ(c)
F (a, b; c; z) = xb−1 (1 − x)c−b−1 (1 − zx)−a dx. (11.1)
Γ(b)Γ(c − b) 0
Note that for a = 0 the integral on the right-hand side reduces to Euler’s integral
of the first kind B(b, c − b).
Proof. The conditions Re b > 0 and Re(c − b) > 0 ensure convergence of the integral
Z 1
I(a, b; c; z) = xb−1 (1 − x)c−b−1 (1 − zx)−a dx.
0
64 WADIM ZUDILIN

Furthermore, for |z| < 1,

∞
−a
X (a)n
(1 − zx) = z n xn .
n=0
n!
Therefore,
∞
1X
(a)n z n b+n−1
Z
I(a, b; c; z) = x (1 − x)c−b−1 dx
0 n=0 n!
∞
X (a)n z n Z 1
= xb+n−1 (1 − x)c−b−1 dx
n=0
n! 0
∞
X (a)n z n Γ(b + n)Γ(c − b)
=
n=0
n! Γ(c + n)
Γ(b)Γ(c − b)
= F (a, b; c; z),
Γ(c)
and the result follows. ,
If z varies in a neighbourhood of a point z0 satisfying | arg(1 − z0 )| < π (in other
words, z0 ∈/ [1, +∞)) and x ∈ [0, 1], then
|(1 − zx)−a | = | exp(−a log(1 − zx))| = exp − Re(a log(1 − zx))




= exp − Re a · Re log(1 − zx) + Im a · Im log(1 − zx)


= exp − Re a · log |1 − zx| + Im a · arg(1 − zx)
= |1 − zx|− Re a exp(Im a · arg(1 − zx))
is bounded by an absolute constant. Therefore, for Re c > Re b > 0 the integral in
(11.1) is an analytic function of z ∈ C \ [1, +∞). This means that formula (11.1)
provides us with the analytic continuation of F (a, b; c; z) from the disk |z| < 1 to
the (star-like) domain C \ [1, +∞).
As another corollary of Theorem 11.6 and the theory of analytic continuation we
obtain
Theorem 11.7 (Gauss). If Re c > Re(a + b), then
Γ(c)Γ(c − a − b)
F (a, b; c; 1) = .
Γ(c − a)Γ(c − b)
Proof. The result follows, whenever Re c > Re b > 0 and Re(c − a − b) > 0, by
taking the limit z → 1 in Theorem 11.6 and using the beta integral evaluation in
Theorem 10.16:
Z 1
Γ(c)
F (a, b; c; 1) = xb−1 (1 − x)c−b−a−1 dx
Γ(b)Γ(c − b) 0
Γ(c) Γ(b)Γ(c − a − b)
= · .
Γ(b)Γ(c − b) Γ(c − a)
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 65

,
To get rid of restriction Re c > Re b > 0, note that the formula is valid for Re(c −
a − b) > 0 and use the theory of analytic continuation.
Example 11.8. When a is a negative integer −r, the theorem becomes
r  
X r (b)n (c − b)r
(−1)n = F (−r, b; c; 1) = ,
n=0
n (c)n (c)r

the result known as the Chu–Vandermonde summation. Problem 4.11 from the
problem sheet states that
r     
X p q p+q
= ;
n=0
n r − n r
this is an immediate consequence of the Chu–Vandermonde theorem.
Example 11.9. If we take z = −1, a = 1 + b − c, the integral in Theorem 11.6
becomes Z 1
xb−1 (1 − x2 )c−b−1 dx,
0
which can be evaluated in terms of the Gamma functions, and we obtain
Γ(1 + b − a)Γ(1 + 21 b)
F (a, b; 1 + b − a; −1) = .
Γ(1 + b)Γ(1 + 12 b − a)
Exercise 11.10. Evaluate F (a, 1 − a; c; 12 ) in terms of Gamma functions.
Hint. Use the integral in Theorem 11.6 with the choice b = 1 − a, z = 21 , and take
(1 − x)2 as the new variable. ,
Our next task is to give a connection between hypergeometric functions of z and
1 − z. By changing z into 1 − w in the differential equation
d2 y dy
z(1 − z) 2
+ (c − (a + b + 1)z) − aby = 0
dz dz
of Theorem 11.4, we obtain the equation
d2 y dy
w(1 − w) 2
+ ((a + b + 1 − c) − (a + b + 1)w) − aby = 0
dw dw
which has the same shape. Hence its independent solutions can be given in terms
of hypergeometric functions (see Exercise 11.5):
F (a, b; a + b − c + 1; w) and wc−a−b F (c − a, c − b; 1 + c − a − b; w).
This means that in the region for which |z| < 1 and |1 − z| < 1 we have a relation
of the form
F (a, b; c; z) = CF (a, b; a + b − c + 1; 1 − z)
+ D(1 − z)c−a−b F (c − a, c − b; 1 + c − a − b; 1 − z),
66 WADIM ZUDILIN

where C and D are constants. The constants can be found by putting z = 0 and
z = 1 (this makes use of Theorem 11.7). We thus obtain, if Re(c − a − b) > 0 and
Re c > 0,
Γ(a + b + 1 − c)Γ(1 − c) Γ(1 + c − a − b)Γ(1 − c)
z=0: 1=C +D ,
Γ(a + 1 − c)Γ(b + 1 − c) Γ(1 − a)Γ(1 − b)
Γ(c)Γ(c − a − b)
z=1: = C.
Γ(c − a)Γ(c − b)
Our findings can be summarised as follows.
Theorem 11.11. If Re(c − a − b) > 0 and Re c > 0, then
Γ(c)Γ(c − a − b)
F (a, b; c; z) = F (a, b; a + b − c + 1; 1 − z)
Γ(c − a)Γ(c − b)
Γ(c)Γ(a + b − c)
+ (1 − z)c−a−b F (c − a, c − b; 1 + c − a − b; 1 − z).
Γ(a)Γ(b)
We now list some other transformations and summations of the hypergeometric
function.
Theorem 11.12. If |z| < 1 and Re z < 12 , then
 
−a −z
(1 − z) F a, b; c; = F (a, c − b; c; z).
1−z
Proof. The coefficient of z r on the left is
r r
X (a)n (b)n (−1)n (a + n)r−n (a)r X (b)n (−r)n (a)r (c − b)r
· = =
n!(c)n (r − n)! r! n=0 n!(c)n r!(c)r

,
n=0

by the Chu–Vandermonde theorem, and the formula is proved.

Exercise 11.13. Iterate Theorem 11.12 twice to obtain Euler’s transformation
F (a, b; c; z) = (1 − z)c−a−b F (c − a, c − b; c; z).
Writing the latter result as
(1 − z)a+b−c F (a, b; c; z) = F (c − a, c − b; c; z)
and equating the coefficients of z r in the corresponding power series, we deduce
r
X (a)n (b)n (c − a − b)r−n (c − a)r (c − b)r
= .
n=0
n!(c)n (r − n)! r!(c)r
Hence r
X (a)n (b)n (c − a − b)r (−r)n (c − a)r (c − b)r
=
n=0
n!(c)n (1 + a + b − c − r)n r! r!(c)r
which can be stated in terms of the (more general!) hypergeometric series
  X ∞
a, b, c (a)n (b)n (c)n n
3 F2 z = z
d, e n!(d)n (e)n
n=0
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 67

as follows.
Theorem 11.14 (Pfaff–Saalschütz). We have
 
a, b, −r (c − a)r (c − b)r
3 F2 1 = .
c, 1 + a + b − c − r (c)r (c − a − b)r
The Pfaff–Saalschütz theorem sums the 3 F2 series when z = 1, a+b+c+1 = d+e
and one of the numerator parameters is a negative integer (in other words, the series
terminates). It can be therefore stated in the form
 
a, b, c Γ(d)Γ(1 + a − e)Γ(1 + b − e)Γ(1 + c − e)
3 F2 1 = ,
d, e Γ(1 − e)Γ(d − a)Γ(d − b)Γ(d − c)
provided that a, b or c is a negative integer and d + e = a + b + c + 1.
Theorem 11.15 (Kummer). Inside the loop of the curve |4z| = |1 − z|2 which
surrounds the origin, we have
 
−a a 1+a −4z
F (a, b; 1 + a − b; z) = (1 − z) F , − b; 1 + a − b; .
2 2 (1 − z)2
Proof. The right-hand side of the required formula is analytic inside
√ the indicated
loop and can therefore be expanded in powers of z when |z| < 3 − 2 2. It becomes
∞
X ( a2 )r ( 1+a − b)r
2
(−4z)r (1 − z)−a−2r ,
r=0
r!(1 + a − b)r

and the coefficient of z n is

n n
X ( a )r ( 1+a − b)r (−4)r (a + 2r)n−r
2 2 (a)n X ( 1+a 2
− b)r (a + n)r (−n)r
=
r=0
r!(1 + a − b)r (n − r)! n! r=0 r!(1 + a − b)r ( 1+a 2 r
)
 1+a 
(a)n 2
− b, a + n, −n
= 3 F2 1
n! 1 + a − b, 1+a
2
(a)n (b)n
= ,
n!(1 + a − b)n
√

,
by Theorem 11.14. The desired formula is therefore proved when |z| < 3 − 2 2, and
the complete result follows by analytic continuation.
Exercise 11.16. Give another proof of Theorem 11.15 by verifying that both sides
satisfy the same differential equation and then comparing the first coefficients of the
related power series expansions.
Example 11.17. Now let z = −1, and we find that
a 1 + a 
F (a, b; 1 + b − a; −1) = 2−a F , − b; 1 + a − b; 1 .
2 2
The series on the right can be summed by Theorem 11.7, so that we again arrive at
the formula in Example 11.9.
68 WADIM ZUDILIN

Finally, we give a complex integral representation of the hypergeometric function

known as Barnes’ contour integral. It shows that the hypergeometric function is the
Mellin integral transform of a certain quotient of Gamma functions.
Let z ∈ C \ [0, +∞). Consider the contour integral
Z
Γ(a + s)Γ(b + s)Γ(−s)
IL = (−z)s ds.
L Γ(c + s)
The principal value (−z)s = exp{s(log |z|+arg(−z))} is an (entire) analytic function
of parameter s ∈ C, and the Gamma function Γ(s) does not vanish on C and has
only simple poles at s = 0, −1, −2, . . . as singular points. Therefore, the integrand
has poles at
s = −a − n, s = −b − n, s = n, n = 0, 1, 2, . . . .
Fix a positive integer N > | Im a| + | Im b| and choose L to be an (almost) rectangle
with vertices iN , −iN , N + 21 − iN and N + 12 + iN , including inside the poles at
s = 0, 1, . . . , N of Γ(−s) but none of the poles of Γ(a + s) and Γ(b + s):

By the residue sum theorem,

N
X Γ(a + s)Γ(b + s)Γ(−s)(−z)s
IL = 2πi Res .
n=0
s=n Γ(c + s)
To compute the residues of Γ(−s) at s = 0, 1, . . . , N , we apply the reflection formula
(Theorem 10.7),
π
Γ(−s)Γ(1 + s) = − .
sin πs
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 69

Then
π 1 (−1)n−1 1
Γ(−s) = (−1)n−1 · = + O(1)
sin π(s − n) Γ(1 + s) s − n Γ(1 + n)
(−1)n−1 1
= + O(1) for s → n,
n! s−n
so that the required sum of residues is
N N
X Γ(a + n)Γ(b + n)z n Γ(a)Γ(b) X (a)n (b)n z n
IL = −2πi = −2πi .
n=0
Γ(c + n) n! Γ(c) n=0 (c)n n!
On the other hand,
Z Z +iN Z N +1/2−iN Z N +1/2+iN Z iN
IL = =− + + +
L −iN −iN N +1/2−iN N +1/2+iN
= −IN + J− + J + J+ .
Our nearest goal is to show that J− → 0, J+ → 0 and J → 0 as N → ∞. Then
by the last two formulae
Z +i∞ ∞
Γ(a + s)Γ(b + s)Γ(−s) s Γ(a)Γ(b) X (a)n (b)n z n
(−z) ds = lim IN = 2πi ,
−i∞ Γ(c + s) N →∞ Γ(c) n=0 (c)n n!
so that the following result is true.
Theorem 11.18 (Barnes). For any z ∈ C \ [0, +∞) with |z| < 1,
Z +i∞
Γ(c) 1 Γ(a + s)Γ(b + s)Γ(−s)
F (a, b, c; z) = (−z)s ds.
Γ(a)Γ(b) 2πi −i∞ Γ(c + s)
It is not hard to show that the integral on the right in Theorem 11.18 converges for
every z ∈ C \ [0, +∞) (not necessarily |z| < 1) and represents an analytic function.
This means that the Barnes’ integral can be used as definition of the hypergeometric
function. For example, one of the consequences of this definition is the formula
Γ(a)Γ(b) Γ(a)Γ(b − a)
F (a, b, c; z) = (−z)−a F (a, 1 + a − c, 1 + a − b; 1/z)
Γ(c) Γ(c − a)
Γ(b)Γ(a − b)
+ (−z)−b F (b, 1 + b − c, 1 + b − a; 1/z),
Γ(c − b)
which is true whenever | arg(−z)| < π and all Gamma factors are well defined.
Proof. To estimate the integrals J− , J+ and J, we write the integrand
Γ(a + s)Γ(b + s) π
− · · (−z)s
Γ(c + s)Γ(1 + s) sin πs
and apply Theorem 10.18 about the asymptotic behaviour of the Gamma function:
Γ(a + s)Γ(b + s)
= |s|Re(a+b−c−1) e− Im(a+b−c−1)·arg s 1 + o(1)


Γ(c + s)Γ(1 + s)
< AN Re(a+b−c−1) ,
70 WADIM ZUDILIN

where the constant A > 0 does not depend on N . The estimate is valid at each of the
intervals [−iN, N + 1/2 − iN ], [iN, N + 1/2 + iN ] and [N + 1/2 − iN, N + 1/2 + iN ].
Let us estimate the integrand on the interval [−iN, N + 1/2 − iN ]. On writing s
for x − iN in the integral J− we find that
π 2π 2πe−πN 2πe−πN
= π(N +ix) −π(N +ix)
= πix −π(2N +ix)
≤ −2πN
< 4πe−πN ,
| sin πs| |e −e | |e − e | 1−e
and also
|(−z)s | = |z|x eN arg(−z) < |z|x e(π−ε)N
where ε > 0. Since |z|x ≤ 1 for |z| < 1, we thus deduce
Z N +1/2
|J− | < AN Re(a+b−c−1) 4πe−πN |z|x e(π−ε)N dx
0
 
−εN Re(a+b−c−1) 1
< 4πe AN N+ ,
2
and the same estimate holds true for J+ as well. Finally, e−εN tends to 0 faster than
N Re(a+b−c−1) increases (to infinity). Consequently, J− → 0 and J+ → 0 as N → ∞.
For the integral J, set s = N + 21 + iy where −N ≤ y ≤ N . Since
π π
= < 2πe−π|y|
| sin πs| cosh πy
and
|(−z)s | = |z|N +1/2 e−y arg(−z) ,
we find in this case that
Z N
|J| < AN Re(a+b−c−1) 2πe−π|y| |z|N +1/2 e−y arg(−z) dy
−N
Z N
< 2πA|z| N +1/2
N Re(a+b−c−1)
e−εy dy
−N
N +1/2 Re(a+b−c−1)
< 2πA|z| N · 2N.

,
Using |z| < 1 we conclude that this time |z|N +1/2 tends to 0 faster than N Re(a+b−c−1)
increases to infinity. Thus, J → 0 as N → ∞, and the theorem follows.
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 71

12. Monodromy of the hypergeometric differential equation

We now turn our attention to general solutions of the hypergeometric equation
d2 u du
E(a, b, c) : z(1 − z) 2 + (c − (a + b + 1)z) − abu = 0
dz dz
in Theorem 11.4 (recall also Example 6.23).
It follows from Cauchy’s theorem that we can assign two linearly independent
analytic solutions of E(a, b, c) in a neighbourhood of each point z0 ∈ C \ {0, 1}. Fix
such a point z0 and a pair of linearly independent solutions u1 , u2 in its neighbour-
hood. The functions u1 , u2 are not single-valued in the domain Z = C \ {0, 1}. If
γ is a closed path (loop) in Z that starts and ends at z0 ,

then the analytical continuation of the solutions u1 and u2 along γ leads to another
pair of linearly independent solutions γu1 and γu2 of the original equation E(a, b, c).
Since the pair u1 , u2 forms a basis in the solution space of the linear equation, the
new solutions can be expressed as linear combinations of the former with constant
coefficients:      
u1 γu1 u
γ = = M (γ) 1
u2 γu2 u2
for some invertible (because the inverse continuation along γ should bring us back
to the pair u1 , u2 ) matrix M (γ) ∈ GL2 (C). This matrix M (γ) is called the circuit
matrix of the pair u1 , u2 along γ.
If γ1 and γ2 are two paths that start and end at z0 , we can define the composition
path γ1 · γ2 to be the one following first γ1 and then γ2 .
Lemma 12.1. We have M (γ1 · γ2 ) = M (γ1 )M (γ2 ).
Proof. Indeed,
      
u1 u1 u
(γ1 · γ2 ) = (γ2 ) γ1 = (γ2 )M (γ1 ) 1
u2 u2 u2

,
   
u1 u
= M (γ1 )γ2 = M (γ1 )M (γ2 ) 1 .
u2 u2
If a path γ1 (with fixed starting and endpoint in z0 ) is continuously deformed in Z
into a path γ2 , then M (γ1 ) = M (γ2 ), and we call such paths γ1 and γ2 equivalent.
The fundamental group of Z with a base point z0 is the related group π1 (Z, z0 ) of
equivalence classes of such paths. By Lemma 12.1 the induced map
π1 (Z, z0 ) 3 γ 7→ M (γ) ∈ GL2 (C)
72 WADIM ZUDILIN

defines a homomorphism of the fundamental group into the matrix group of invert-
ible linear transformations. The homomorphism is called the monodromy represen-
tation of the differential equation E(a, b, c).
Of course, the monodromy representation depends on the choice of the pair u1 , u2
and the base point z0 . If either of them is changed then we obtain a new monodromy
representation, which is just conjugate to the earlier one, so that the differential
equation determines the congugacy class of a representation π1 (Z, z0 ) → GL2 (C).
The monodromy group of the hypergeometric differential equation E(a, b, c) is now
defined as the image of the monodoromy representation in GL2 (C), and one can also
define the projective monodromy group as the image of the monodromy group under
the natural embedding GL2 (C) → P GL2 (C) = GL2 (C)/C∗ . The conjugacy class of
the monodromy group in GL2 (C), as well as the conjugacy class of the projective
monodromy group in P GL2 (C), are determined uniquely by the differential equation.
We now briefly record the Schwarz theory — an arithmetic (modular) interpre-
tation of solutions to the hypergeometric differential equation E(a, b, c) associated
to its (projective) monodromy group; this is addressed in a greater detail in the
monograph [1].
As discussed above (see also Exercise 11.5), if 2 − c 6= 0, −1, −2, . . . then the
function z 1−c F (a + 1 − c, b + 1 − c, 2 − c; z) is a solution of the hypergeometric
equation E(a, b, c).
The restrictions on c are not severe, especially in the case c = 1 (of special
arithmetic interest!), when the newer solution merges the original F (a, b, c; z). For
such situation we can appeal to Frobenius’s method, which simply means treating c
as a variable in a neighbourhood of c = 1 and using the fact that a linear combination
of solutions of E(a, b, c), in particular,
z 1−c F (a + 1 − c, b + 1 − c, 2 − c; z) − F (a, b, c; z)
,
c−1
happens to be a solution as well. Taking then the limit as c → 1 on the basis of
L’Hôpital’s rule we conclude that
d 1−c

z F (a + 1 − c, b + 1 − c, 2 − c; z) − F (a, b, c; z)
dc c=1
∞  n+1−c
(a)n (b)n z n

d X (a + 1 − c)n (b + 1 − c)n z
= −
dc n=0 (2 − c)n n! (c)n n! c=1
∞ n−1 
(a)n (b)n z n
 
X X 1 1 2
=− log z − + − (12.1)
n=0
(1) n n! k=0
a + k b + k 1 + k
is a solution of E(a, b, 1). Here we used d(z n+1−c )/dc = − log z × z n+1−c and
d d
(a + 1 − c)n = (a + 1 − c)n × log(a + 1 − c)n
dc dc
n−1
X 1
= −(a + 1 − c)n × .
k=0
a + 1 − c + k
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 73

Observe that this solution of E(a, b, 1) has a logarithmic singularity at z = 0.

The technology also allows us to write down hypergeometric solutions at other
singularities z = 1 and ∞ of E(a, b, c). The substitutions z 7→ 1 − z and z 7→
1/z transform the differential equation E(a, b, c) into E(a, b, a + b − c + 1) and
E(a, a − c + 1, a − b + 1), respectively (the last one is multiplied by z a without
affecting its solutions).
Theorem 12.2. If 0 < c < 2 and c = 1, then
f01 (z) = F (a, b, c; z) and f02 (z) = z 1−c F (a + 1 − c, b + 1 − c, 2 − c; z)
are linearly independent solutions of E(a, b, c) (in a neighbourhood of z = 0). In the
case c = 1, the second entry should be replaced with the one in (12.1).
If c − a − b is not an integer, then the functions
f11 (z) = F (a, b, a + b + 1 − c; 1 − z) and
f12 (z) = (1 − z)c−a−b F (c − a, c − b, c + 1 − a − b; 1 − z)
form a pair of linearly independent solutions of E(a, b, c) (in a neighbourhood of
z = 1). In the case c − a − b = 0, the second solution should be replaced with
F (a, b, 1 − ε; 1 − z) − (1 − z)ε F (ε + b, ε + a, ε + 1; 1 − z)
f12 (z) = lim .
ε→0 ε
Finally, if a − b ∈
/ Z, then the functions
f∞1 (z) = (−1/z)a F (a, 1 + a − c, a − b + 1; 1/z) and
f∞2 (z) = (−1/z)b F (b, 1 + b − c, b − a + 1; 1/z)
form a pair of linearly independent solutions of E(a, b, c) (in a neighbourhood of
z = ∞).
Exercise 12.3. What the second solution should be replaced with, in the case
a = b?
If we assume that the parameters a, b and c of the hypergeometric equation
E(a, b, c) are real, then the pairs of solutions in Theorem 12.2 are defined on the
domains
C \ (−∞, 0] ∪ [1, +∞), C \ (−∞, 1] and C \ [0, +∞),
respectively; furthermore, they are all real-valued on the remaning real intervals of
their definition. Since all the three pairs form solution of the same second order
linear differential equation, they are related by linear transformations in a common
domain of definition, for example, in the half-plane Z+ = {z ∈ C : Im z > 0}:
(f01 , f02 ) = (f11 , f12 )M 10 , (f01 , f02 ) = (f∞1 , f∞2 )M ∞0
etc., for some matrices M 10 , M ∞0 , . . . ∈ GL2 (C), which are called connection matri-
ces.
The Schwarz map is defined by
f = f0 : Z+ 3 z 7→ f01 (z) : f02 (z) ∈ P1 .
74 WADIM ZUDILIN

Since f is real-valued on (0, 1), the image (f (0), f (1)) of the interval also belongs to
the real line and forms a boundary of f (Z+ ). Similarly, the maps
f1 : Z+ 3 z 7→ f11 (z) : f12 (z) ∈ P1 , f∞ : Z+ 3 z 7→ f∞1 (z) : f∞2 (z) ∈ P1
send (1, +∞) and (−∞, 0) onto real intervals (f1 (0), f1 (∞)) and (f∞ (∞), f∞ (0)),
respectively. On the other hand, the maps f = f0 , f1 and f∞ are connected to each
other by linear fractional transformations (induced by the connection matrices), and
those translate lines into lines and circles. Thus, the image of the union (∞, 0) ∪
(0, 1) ∪ (1, ∞) under f is a boundary of triangle whose sides are line intervals and
arcs. The triangle f (Z+ ) is known as the Schwarz triangle:

The angles in the Schwarz triangle can be easily computed. Indeed, the explicit
formulae for the solutions involved imply that, around z = 0, we have
(
Cz 1−c = C0 exp((1 − c) log z) if c 6= 1,
f0 (z) ∼
C log z if c = 1,
hence the angle between the line intervals (∞, 0) and (0, 1) at z = 0 translates under
the map f0 to the angle π|1 − c| at the vertex f (0) of f (Z+ ). The linear fractional
transformations connecting f , f1 and f∞ are conformal (they do not change the
angles); therefore, the explicit expressions for f1 (z) and f∞ (z) lead to the angles
π|c − a − b| at f (1) and π|a − b| at f (∞).
(The zero angle simply means that the corresponding sides of the triangle are tan-
gent.) If each angle of the Schwarz triangle is less than π, then f (z) bijectively maps
the upper half-plane Z+ onto the interior of the Schwarz triangle.
Theorem 12.4. If a, b, c ∈ R and
|1 − c| < 1, |c − a − b| < 1 and |a − b| < 1,
then the Schwarz map f is a one-to-one correspondence between the upper half-plane
and the (interior ) of the Schwarz triangle.
The last tool to use is Schwarz’s reflection principle, which is applied to the map
f defined on Z+ and to the intervals (−∞, 0), (0, 1) and (1, +∞). The map can
be extended to the half-plane Z− = {z ∈ C : Im z > 0} through any of these
three intervals by reflections. Then the same reflection principle can be used to
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 75

extend from Z− to Z+ through any of the same three intervals, etc. To witness this
reflection development, fix a point z0 ∈ Z+ and start a loop γ from z0 passing to the
half-plane Z− through (−∞, 0) and returning to Z+ through (0, 1). The image of γ
under f is a path in P1 starting at f (z0 ) of the initial triangle T = f (Z+ ), passing
through the side (arc) (f (∞), f (0)) of T into the mirror image T 0 = f (Z− ) (with
respect to the arc), followed by passing through the side (arc) (f (0), f (1)) of the
triangle T 0 into the mirror image T 00 of T 0 (with respect to the latter arc):

The corresponding map γf obtained as the analytical continuation of f along γ

is a linear fractional transformation of f , that is,
 
af + b a b
γf = for some M = ∈ GL2 (C).
cf + d c d

The above method allows us to continue analytically the Schwarz map f from Z+
to Z− , then from Z− to Z+ as many times as desired. The images under participating
reflections are conformal images of the original Schwarz triangle f (Z+ ) as in the
following figure:

If an even number of reflections performed, then we obtain a linear fractional

transformation of f by the argument above. The transformation belongs to the
projective monodromy group Γ ⊂ P GL2 (C) of the underlying hypergeometric dif-
ferential equation E(a, b, c), and all such transformations exhaust the group Γ.
To achieve an aesthetically pleasing situation, we should avoid overlaps of type we
have in the figure above. This means that we need all three angles of the Schwarz
76 WADIM ZUDILIN

triangle to be integral multiples of π:

1 1 1
|1 − c| = , |c − a − b| = , |a − b| = , where p, q, r ∈ {2, 3, . . . } ∪ {∞}.
p q r
The corresponding choices of parameters a, b, c of the hypergeometric equation lead
to modular interpretations of the equation. For example, the case a = b = 21 and
c = 1 corresponds to all the angles being zero; the corresponding tessellation of the
unit disk by triangles is

References
[1] M. Yoshida, Hypergeometric functions, my love, Modular interpretations of configuration
spaces, Aspects Math. E32 (Friedr. Vieweg & Sohn, Braunschweig, 1997).
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 77

13. Rational approximations to the logarithm

As a non-analytic application of the hypergeometric function, we consider the
problem of irrationality of
1 1 1 (−1)n−1
log 2 = 1 − + − + · · · + + ··· .
2 3 4 n
Note that for a seemingly similar number — the Euler–Mascheroni constant γ (de-
fined in Exercise 10.1) — the irrationality question remains open.
For each n = 0, 1, 2, . . . define the series
∞
X (t − 1)(t − 2) · · · (t − n) t
Fn (z) = z,
t=1
t(t + 1) · · · (t + n)
which converges in the region |z| ≤ 1, z 6= 1. The summand is the rational function
(t − 1)(t − 2) · · · (t − n)
Rn (t) = ;
t(t + 1) · · · (t + n)
it possesses the partial-fraction decomposition,
n
X Ak
Rn (t) = .
k=0
t+k
Lemma 13.1. We have
  
n−k n+k n
Ak = (−1) , k = 0, 1, . . . , n.
k k
In particular, all Ak are integers.
Proof. Indeed,

(−k − 1)(−k − 2) · · · (−k − n)
Ak = R(t)(t + k) t=−k
=
(−k)(−k + 1) · · · (−2) · (−1) · 1 · 2 · · · (−k + n)

=
(−1)n (n + k)!/k!
(−1)k k!(n − k)!
for k = 0, 1, . . . , n. ,
Let us transform the original series using the partial-fraction decomposition:
n ∞ n ∞
X X z t+k X X zl
Fn (z) = Ak z −k = Ak z −k
k=0 t=1
t+k k=0 l=k+1
l
n ∞ k
zl X zl
X X 
−k
= Ak z −
k=0 l=1
l l=1
l
n n k
X
−k
X X z −(k−l)
= − log(1 − z) · Ak z − Ak .
k=0 k=0 l=1
l
Introducing the notation
n n k
X
−k
X X z −(k−l)
Qn (z) = − Ak z , Pn (z) = Ak ,
k=0 k=0 l=1
l
78 WADIM ZUDILIN

we thus obtain the following result.

Lemma 13.2. The representation
Fn (z) = Qn (z) log(1 − z) − Pn (z)
holds, where Qn (z) and Pn (z) are polynomials in 1/z of degree at most n. Moreover,
for z = a/b with nonzero integers a, b, we have
   
n a n a
a Qn ∈Z and dn a Pn ∈ Z.
b b
Here and in what follows dn denotes the least common multiple of the numbers
1, 2, . . . , n; we also set d0 = 1. One of the consequences of the prime number theorem
is
lim d1/n
n = e.
n→∞
The following simpler bound (which can be obtained by completely elementary meth-
ods) is sufficient for our purposes.
Lemma 13.3. dn ≤ 4n for n = 0, 1, 2, . . . .
We will now give a different representation of the quantity Fn (z), as a hypergeo-
metric function. Note that
Γ(t)/Γ(t − n) Γ(t)2
Rn (t) = = ;
Γ(t + n + 1)/Γ(t) Γ(t − n) Γ(t + n + 1)
furthermore, summation in
∞
X (t − 1)(t − 2) · · · (t − n) t
Fn (z) = z
t=1
t(t + 1) · · · (t + n)
can be started from t = n + 1, since Rn (t) = 0 for t = 1, 2, . . . , n. Therefore,
∞ ∞
X Γ(t)2 t
X Γ(ν + n + 1)2
Fn (z) = z = z ν+n+1
t=n+1
Γ(t − n) Γ(t + n + 1) ν=0
Γ(ν + 1) Γ(ν + 2n + 2)
∞
n+1 Γ(n + 1)2 X (n + 1)2ν ν
=z z
Γ(2n + 2) ν=0 ν! (2n + 2)ν
Γ(n + 1)2
= z n+1 F (n + 1, n + 1, 2n + 2; z).
Γ(2n + 2)
Applying now Theorem 11.6 we obtain
Lemma 13.4.
1
xn (1 − x)n
Z
n+1
Fn (z) = z dx.
0 (1 − zx)n+1
For z real in the interval −1 ≤ z < 1, the integrand is positive for x ∈ (0, 1). In
particular, this implies that, for any z ∈ [−1, 1), z 6= 0, we have Fn (z) 6= 0.
Set  
x(1 − x)
M (z) = max .
0≤x≤1 1 − zx
 NWI-WB093: COMPLEX ANALYSIS (2023/24) 79

Clearly, computation of the maximum, for any fixed z, reduces to solving a quadratic
equation.
Lemma 13.5. For z ∈ [−1, 1), the estimate
|Fn (z)| ≤ | log(1 − z)| · |z|n M (z)n
holds.
Proof. For our (non-negative) integrand we have
xn (1 − x)n 1
n+1
≤ M (z)n ·
(1 − zx) 1 − zx
implying Z 1
n n dx
|Fn (z)| ≤ |z| M (z) · z .
0 1 − zx
It remains to use the equality

,
Z 1 ∞
dx X zt
z = F0 (z) = = − log(1 − z).
0 1 − zx t=1
t
Lemma 13.6. Suppose that α ∈ R and we have a sequence of fractions pn /qn 6= α
with pn , qn ∈ Z such that qn α − pn → 0 as n → ∞. Then α is irrational.
Proof. Assume, on the contrary, that α = p/q for some p ∈ Z and q ∈ N. By the
hypothesis, q · (qn α − pn ) tends to 0 as n → ∞. On the other hand, the number

,
q · (qn α − pn ) = pqn − qpn is a positive integer, so that |q · (qn α − pn )| ≥ 1. The
contradiction shows that α ∈ / Q.
Setting qn = an dn Qn (z) and pn = an dn Pn (z) for z = a/b ∈ [−1, 1) with a > 0 and
applying Lemmas 13.2–13.6 we obtain the following general result.
Theorem 13.7. Suppose that, for z = a/b ∈ [−1, 1), z 6= 0, we have 4a|z|M (z) < 1
where  
x(1 − x)
M (z) = max .
0≤x≤1 1 − zx
Then the number α = log(1 − z) is irrational.
Proof. Indeed, the estimate for the nonzero linear forms qn α − pn assumes the form

n
|qn α − pn | ≤ Cdn a|z|M (z)

n
< C 4a|z|M (z) → 0 as n → ∞,
(in the notation C = | log(1 − z)| > 0). ,
Knowledge about the asymptotic behaviour of pn α − qn and qn as n → ∞ allows
one to prove the irrationality not only in the qualitative form, but also quantitatively;
that is, to establish estimates of the form
p C
α− > δ , for some δ > 1,
q q
for approximations of α by rationals p/q.
80 WADIM ZUDILIN

Theorem 13.8. The quantity log 2 is irrational.

Proof. Apply Theorem 13.7 with z = −1. To compute the maximum M (z) we find
that
d x(1 − x) 1 1 1 (1 − x2 ) − (x + x2 ) − (x − x2 )
log = − − =
dx 1+x x 1−x 1+x x(1 − x)(1 + x)
2
1 − 2x − x
= ,
x(1 − x)(1 + x)
√
hence the maximum is attained at x = 2 − 1:
√ √
x(1 − x) ( 2 − 1)(2 − 2) √ 1
M (−1) = √
= √ = ( 2 − 1)2 = 0.17157287 . . . < .
1 + x x= 2−1 2 4
Thus, log 2 is irrational. ,