Complex Analysis (part 3): Complex Integration (by Evan Dummit, 2022, v. 1.

00)

Contents
3 Complex Integration 1

3.1 Integrals on Complex Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

3.1.1 Complex Line Integrals via Riemann Sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3.1.2 Evaluating Complex Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3.1.3 Antiderivatives, The Fundamental Theorem of Calculus . . . . . . . . . . . . . . . . . . . . . 6

3.2 Cauchy's Integral Theorem and Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.2.1 Cauchy's Integral Theorem via Green's Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.2.2 Cauchy's Integral Theorem via Deformation of Contours . . . . . . . . . . . . . . . . . . . . . 10

3.2.3 Integration of Power Series and Laurent Series . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.2.4 Winding Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2.5 Cauchy's Integral Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3.2.6 Higher Derivatives and Series Expansions of Holomorphic Functions . . . . . . . . . . . . . . 21

3 Complex Integration
In this chapter, our goal is to study integration in the complex plane. We extend the notion of a line integral of a
real-valued function along a curve in R2 to dene the line integral of a complex function along a curve in C, and
then characterize various properties of line and contour integrals in C. We then develop Cauchy's integral theorem
and Cauchy's integral formula, which together form one of the most fundamental tools in complex analysis, and
study many of their immediate consequences.

3.1 Integrals on Complex Curves

• We would like to develop a complex analogue to integration, to parallel our complex derivative.

◦ One possibility is simply to generalize the functions we allow ourselves to integrate on an interval [a, b]:
´b
in other words, to dene the integral
a
f (t) dt where now f (t) is a complex-valued function.

◦ We can do this just by writing the usual denition of a real integral using Riemann sums but now allowing
the function f (t) = x(t) + iy(t) to be complex.

◦ However, it is not hard to see that since Riemann sums and real integrals are linear, this simply amounts
´b ´b ´b
to dening
a
[x(t) + iy(t)] dt = a
x(t) dt + i a
y(t) dt.
◦ In particular, by an immediate application of the fundamental theorem of calculus for real-valued func-
tions to the real and imaginary parts, we see that if F (t) is dierentiable with F 0 (t) = f (t), then
´b
a
f (t) dt = F (b) − F (a).

• A far more useful generalization of integration is to integrate a complex function f along a complex curve γ.

1
3.1.1 Complex Line Integrals via Riemann Sums

• Just as with the integral of a real-valued function, we can formally dene complex line integrals as a limit of
Riemann sums (though, just as with integrals of real-valued functions, we will essentially never actually use
Riemann sums to evaluate integrals).

◦ Explicitly, suppose that γ : [a, b] → C is a continuous curve, so that limt→t0 γ(t) = γ(t0 ) for all t ∈ [a, b],
with the appropriate one-sided limits taken at the endpoints, and let f : C → C be a complex function.
◦ For convenience, we will set up Riemann sums using a partition of the interval [a, b] rather than a partition
of the actual geometric curve im(γ).

• Denition: A tagged partition of the interval [a, b] is a set P ∗ = {t0 , t1 , . . . , tn } with a = t0 < t1 < · · · < tn = b,
∗
together with values ti ∈ [ti−1 , ti ] for each 1 ≤ i ≤ n. By applying γ to all of the ti and t∗i , we obtain a tagged
partition of the curve γ .
Pn
• Denition: If P ∗ is a tagged partition of [a, b], the associated Riemann sum is RSP ∗ (f ) = k=1 f (zi∗ )∆zi
where zi∗ = γ(t∗i ) represents the point on γ at which we evaluate f , and ∆zi = γ(ti ) − γ(ti−1 ) represents the
segment on which we are summing. The norm of the partition is dened to be the maximum of the values
|∆zi | for 1 ≤ i ≤ n.
´
• Denition: We say f has integral L on γ, and write
γ
f (z) dz = L, if for every  > 0 there exists a δ > 0
(depending on ) such that for every tagged partition P ∗ with norm(P ∗ ) < δ , we have |RSP (f ) − L| < .

◦ Intuitively, the idea is that we dene the integral as the limit of the value of the Riemann sums as the
norm tends to 0 (i.e., as the pieces of the partition of γ become uniformly small).

◦ By some fairly tedious analysis of renements of partitions , one may prove that if
1 f is continuous on γ
(meaning that for any z0 ∈ γ and any  > 0 there exists a δ > 0 such that |f (z) − f (z0 )| <  for all z ∈ γ
such that |z − z0 | < δ ) and γ is continuously dierentiable, then in fact f is integrable on γ .

• For no additional cost we can weaken slightly the dierentiability requirement of the curves over which we set
up complex line integrals:

• Denition: A continuous curve γ in C is a continuous function γ : [a, b] → C. Explicitly, this means γ(t) =
x(t) + iy(t) for some continuous real-valued functions x(t) and y(t) on the interval [a, b].
• Denition: A continuous curve is closed if γ(a) = γ(b), and it is simple if γ is one-to-one except for having
γ(a) = γ(b).
• Denition: A continuous curve is rectiable if γ is continuously dierentiable except at a nite number of
points; as part of this requirement we also require the one-sided limits limt→a+ γ 0 (t) and limt→b− γ 0 (t) to
exist.

◦ The motivation for the endpoint conditions in the denition of rectiability is to ensure that the fun-
damental theorem of calculus holds on [a, b] for the function γ : explicitly, we have γ(b) − γ(a) =
´ b− 0 ´b
lim→0+ [γ(b − ) − γ(a + )] = lim→0+ a+ γ (t) dt = a γ 0 (t) dt.
◦ If f is continuous and γ is rectiable, then by an appropriate partition renement argument, we see that
´
f is integrable on γ and the integral of f is the sum of f (z) dz over the nitely many pieces where γ
is continuously dierentiable.

◦ It is also easy to see that if γ1 : [a, b] → C

γ2(: [b, c] → C are rectiable with γ1 (b) = γ2 (b), then the
and

γ1 (t) for a ≤ t ≤ b
union γ1 ∪ γ2 : [a, c] → C dened by γ1 ∪ γ2 (t) = is rectiable. Geometrically, this
γ2 (t) for b < c ≤ c
corresponds to gluing γ2 to the end of γ1 .

1 The usual approach for real-valued functions is to consider for a xed P∗ the upper sum (using the selection of tag points
maximizing the Riemann sum for a xed partition P ∗) and the lower sum (minimizing the Riemann sum), then to show that the
upper sum decreases and the lower sum increases when rening a partition, and nally using (uniform) continuity of f on γ to show that
for an appropriate renement the dierence between the upper and lower sums can be made arbitrarily small. Applying the result to
the real and imaginary parts of a complex integral, and making an estimate on the error obtained by the fact that partition renement
does not completely respect taking real and imaginary parts (this is where the continuous dierentiability of γ is needed) yields the
result for complex integrals.

2
 ◦ In most of our actual examples we will integrate over smooth curves, but it is convenient to be able to
use rectiable curves, since for example the boundary of a triangle or rectangle is non-dierentiable at
each of the vertices, but is still rectiable.

• Using the denition in terms of Riemann sums, we obtain various basic properties of complex integrals
analogous to those for real integrals.

• Proposition (Basic Integral Properties): Suppose that γ : [a, b] → C is rectiable and f and g are integrable
on γ. Then the following hold:

´´
1. For any c∈C the scalar multiple cf is integrable on γ
(cf )(z) dz = c γ f (z) dz .
and
γ
´ ´ ´
2. The sum f +g and dierence f −g are both is integrable on γ and (f ±g)(z) dz = γ f (z) dz ± γ g(z) dz .
γ
´
3. If γ and γ2 are rectiable and f is integrable on both, then f is integrable on γ1 ∪γ2 and f (z) dz =
´ 1 ´ γ1 ∪γ2
γ1
f (z) dz + γ2
f (z) dz .
´ ´
4. If −γ is the reverse of γ , dened by (−γ)(t) = γ(a + b − t), then f (z) dz = − γ f (z) dz .
−γ
´b ´b
5. If f is continuous then
a
f (t) dt ≤ a |f (t)| dt.
´c ´b ´c
◦ Note that (3) is the analogue of a f (x) dx = a f (x) dx + b f (x) dx while (4) is the analogue of
´a ´b
f (x) dx = − a f (x) dx.
b ´
◦ The triangle inequality (5) does not hold on arbitrary curves γ , since the second integral γ |f (z)| dz
need not even be real, which is why we only pose it for integrals along the real line.

◦ Proofs: (1) follows by noting that Riemann sums for cf are simply c times a Riemann sum for f.
◦ (2) follows by noting that a Riemann sum for f + g is the sum of a Riemann sum for f with a
Riemann sum for g.
◦ (3) follows by noting that a Riemann sum for f on γ1 ∪ γ2 is a Riemann sum for f on γ1 plus a
Riemann sum on γ2 , provided the transition point t = b is included in the partition. Since we may
freely rene partitions, this assumption causes no issues.

◦ (4) follows by noting that a Riemann sum for −γ is a Riemann sum for γ with all terms scaled by
−1 (using the same tag points but interchanging ti and ti−1 ).
´b
◦ (5) follows by applying the usual triangle inequality in C to a Riemann sum for f (t) dt and
´b a
observing that the result is a Riemann sum for
a
|f (t)| dt.

3.1.2 Evaluating Complex Line Integrals

• We would like to actually evaluate some complex integrals, which is quite unpleasant using the Riemann sum
denition. Conveniently, we can avoid most of the technical annoyances by converting a complex integral to
an integral over a real variable, in the same way that we can convert a line integral on a real plane curve into
a single integral of a real variable.

◦ Explicitly, suppose P is a given partition of [a, b]. Under the assumption that γ(t) is dierentiable, we
have ∆zi = γ(ti ) − γ(ti−1 ) ≈ (ti − ti−1 )γ 0 (ti ).
Pn Pn
◦ Therefore, the Riemann sum k=1 f (zi−1 )∆zi for f (z) is approximately equal to k=1 f (γ(ti−1 )) γ 0 (ti−1 ) (ti −
´b
ti−1 ), which is in turn a Riemann sum for the integral a f (γ(t)) γ 0 (t)dt with tag points t∗i = ti−1 .
◦ This latter integral is much easier to evaluate, since we may just integrate the real and imaginary parts
on [a, b] separately, as we discussed earlier.

◦ As we will show, under the assumption that γ(t) is continuously dierentiable, the dierence between
these two Riemann sums must go to zero as the norm of the partition goes to zero.

• Proposition (Complex Line Integrals): Suppose γ : [a, b] → C is a rectiable complex curve and f (z) is a
´ ´b
continuous function on γ. Then f is integrable on γ and γ f (z) dz = a f (γ(t)) γ 0 (t) dt.

3
 ◦ Proof: Since f f (γ(t)) γ 0 (t) is integrable on [a, b] since it is continuous,
is integrable as noted earlier and
´b
we just need to show that the Riemann sums for f approach Riemann sums for
a
f (γ(t)) γ 0 (t) dt. It
suces to show that this holds on each of the nitely many subintervals on which γ is continuously
dierentiable.

◦ More precisely, for any >0 we will show that the left-endpoint Riemann sums for the rst integral and
the second integral with suciently small norm are within  of one another.

◦ Since f is continuous on the closed curve γ, the function g(t) = |f (γ(t))| from [a, b] → R is continuous
on the closed interval [a, b], so by the extreme value theorem it is bounded above: then |f | ≤ M on γ for
some M.
◦ Also since γ0 is continuous on [a, b] it is uniformly continuous by the Heine-Cantor theorem applied to
both the real and imaginary parts, so there exists δ >0 such that |s − t| < δ implies |γ 0 (s) − γ 0 (t)| <
/[(b − a)M ].
◦ Now let P = {t0 , . . . , tn } be any partition of [a, b] of norm less than δ and take t∗i = ti−1 .
0
◦ By uniform continuity, the absolute value of the dierence between γ(ti ) − γ(ti−1 ) and
Pn (ti − ti−1 )γ (ti ) is
bounded above by (ti−1 −ti )/[(b−a)M ], so the dierence between the Riemann sum k=1 f (γ(ti−1 ))(γ(ti )−
Pn 0
γ(t
Pn i−1 )) and the Riemann sum k=1 f (γ(t i−1 )) γ (t i−1 ) (t i − ti−1 ) is bounded in absolute value by

k=1 M (ti−1 − t i )/[(b − a)M ] =  .

◦ Hence for any partition of norm at most δ we see that the Riemann sums are within , so taking the limit
shows that the dierence between the integrals is at most . This holds for all  > 0, so the integrals are
equal.
´
• Example: For γ(t) = eit for 0 ≤ t ≤ 2π , evaluate the complex line integral
γ
z −1 dz .

◦ We simply compute γ 0 (t) = ieit and f (z) = z −1 = e−it .

´ ´ 2π ´ 2π
◦ Thus
γ
z −1 dz = 0 e−it · ieit dt = 0 i dt = 2πi .
´
• Example: For γ(t) = eit for 0 ≤ t ≤ 2π , evaluate the complex line integral
γ
z n dz for each integer n 6= −1.

◦ As above we have γ 0 (t) = ieit and f (z) = z n = (eit )n .

2π
´ ´ 2π ´ 2π e(n+1)it 1 1
◦ Thus
γ
z n dz = 0
(eit )n · ieit dt = 0
ie(n+1)it dt = = − = 0 .
n+1 t=0 n+1 n+1
´
• Example: For γ(t) = t2 + it for 0 ≤ t ≤ 1, evaluate the complex line integral
γ
(1 − 2z) dz .

◦ We have γ 0 (t) = 2t + i andf (z) = 1 − 2(t2 + it) = (1 − 2t2 ) − 2it.

´ ´1 ´1
◦ Thus
γ
(1 − 2z) dz = 0
[(1 − 2t2 ) − 2it](2t + i) dt = 0
[(4t − 4t3 ) + (1 − 6t2 )i] dt
1
= (2t − t ) + (t − 2t3 )i
4
t=0
= 1−i .

´
• Example: For γ(t) = (1 + t) + (2 − t)i for 0 ≤ t ≤ 3, evaluate the complex line integral
γ
z 2 dz .

◦ We have γ 0 (t) = 1 − i and f (z) = [(1 + t) + (2 − t)i]2 .

´ ´3 ´3
◦ Thus
γ
z 2 dz = 0 [(1 + t) + (2 − t)i]2 (2t + i) dt = 0 [(1 + 8t − 2t2 ) + (7 − 4t − 2t2 )i] dt
3
= (t + 4t2 − 32 t3 ) + (7t − 2t2 − 23 t3 )i t=0
= 21 − 15i .

´ ´
• Example: For γ(t) = t + 2it for 0 ≤ t ≤ 1, evaluate the line integrals
γ
Re(z 2 ) dz and
γ
Im(z 2 ) dz .

◦ We haveγ 0 (t) = 1 + 2i and Re(z 2 ) = Re(−3t2 + 4it2 ) = −3t2 .

´ ´1
◦ Thus γ Re(z 2 ) dz = 0 (−3t2 )(1 + 2i) dt = −(1 + 2i)t3 |1t=0 = −1 − 2i .
´ ´1
◦ Also as Im(z 2 ) = 4t2 , we see γ Im(z 2 ) dz = 0 (4t2 )(1 + 2i) dt = 34 (1 + 2i)t2 |1t=0 = 4
3 + 83 i .

• Although we refer to the function γ as a curve, in fact the value of the integral does not depend on the
particular parametrization we use for the curve.

4
 ◦ For example, consider the line integral of f (z) = z 2 on the upper half of the unit circle traversed
counterclockwise from z=1 to z = −1.
´ ´π
◦ If we use the parametrization γ1 (t) = eit for 0 ≤ t ≤ π, the integral is
γ1
z 2 dz = 0
(eit )2 · ieit dt =
´π e 3it π
2
0
ie3it dt = =− .
3 t=0 3
√
◦ If instead we use the parametrization γ2 (s) = ei s
for 0 ≤ s ≤ π 2 , which follows the same path but at non-
√ π2
´ ´ π2 √ 1 √ ´ π2 i 3i√s e3i s
2
constant speed, the integral is
γ2
z 2 dz = 0
(ei s )2 ·i √ ei s dt = 0 √ e dt = =− .
2 s 2 s 3 3
t=0
◦ Indeed, once we actually plug in the parametrizations, the two resulting integrals are easily seem to be
√
equal by making the change of variables t= s, which (unsurprisingly) also shows that the parametriza-
tions themselves describe the same curve. This result holds in general:

• Proposition (Reparametrization): Suppose that f (z) is continuous on the rectiable curve γ1 : [a, b] → C, and
γ2 : [c, d] → C is also
´ rectiable and there exists a continuous one-to-one function g : [a, b] → [c, d] such that
´
γ1 = γ2 ◦ g . Then γ1 f (z) dz = γ2 f (z) dz .

◦ Proof: Since both integrals exist it suces to observe that Riemann sums on γ2 yield Riemann sums on
γ1 . But this follows immediately: if P∗ is a tagged partition of [a, b], then applying g yields a tagged
partition Q∗ of [c, d], and the underlying Riemann sums are then equal since γ1 = γ2 ◦ g .

• As a consequence of our reparametrization result, we only need a description of the actual curve γ in C to
evaluate a complex line integral. We may therefore refer to such a contour integral merely by describing the
curve (contour) and function to be integrated on it.

◦ We can parametrize the line segment from z0 to z1 via γ(t) = (1 − t)z0 + tz1 for 0 ≤ t ≤ 1; note γ(0) = z0
and γ(1) = z1 .
◦ r centered at z0 oriented counterclockwise via γ(t) = z0 + reit
We can parametrize the circle of radius
for 0 ≤ t ≤ 2π ; note that γ(0) = γ(2π) = z0 + r . Additionally, if we want only a portion of the circle, or
to have γ go around the circle multiple times, we need only adjust the range of values for t.

´
• Example: Find γ z 2 dz where γ is the counterclockwise boundary of the triangle with vertices −1, 1, and i.

◦ There are three boundary components to parametrize.

◦ For the segment from −1 to 1 we take γ(t) = −1 + 2t for 0 ≤ t ≤ 1. Then γ 0 (t) = 2 and z 2 = (−1 + 2t)2
´1 ´1
so the integral is
0
(−1 + 2t)2 · 2 dt = 0 (2 − 8t + 8t2 ) dt = 2/3.
◦ For the segment from 1 to i we take γ(t) = (1 − t) + it for 0 ≤ t ≤ 1. Then γ 0 (t) = −1 + i and
2 2
´1 ´1
z = [(1−t)−it] so the integral is
0
[(1−t)−it]2 ·(−1+i) dt = 0 [(−1+4t−2t2 )+(1−2t2 )i] = 1/3+i/3.
◦ For the segment from i to −1 we take γ(t) = −t + (1 − t)i for 0 ≤ t ≤ 1. Then γ 0 (t) = −1 − i and
´1 ´1
z 2 = [−t+(1−t)i]2 so the integral is 0
[−t+(1−t)i]2 ·(−1−i) dt = 0 [(1−2t2 )+(1−4t−2t2 )i] = 1/3−i/3.
◦ Therefore the integral around the entire boundary is the sum, which is 4/3 .

´
• Example: Find
γ
Im(z) dz where γ is the curve that goes around the circle |z − i| = 2 three times counter-
clockwise.

◦ The parametrization γ(t) = i + 2eit for 0 ≤ t ≤ 6π winds around the given circle three times counter-
clockwise.

◦ Then γ 0 (t) = 2ieit and Im(z) = Im(i + 2eit ) = 1 + 2 sin t = 1 + ie−it − ieit , so the desired integral is
´ 6π ´ 6π 6π
0
[1 + ie−it − ieit ](2ieit ) dt = 0 [2ieit − 2 + 2e2it ] dt = [2eit − 2t − ie2it ] t=0 = −12π .

5
3.1.3 Antiderivatives, The Fundamental Theorem of Calculus

• It is natural to expect that there should be complex analogues of our familiar properties of derivatives,
antiderivatives, and integrals (e.g., the fundamental theorem of calculus). We collect many of these properties
here:

• Theorem (Properties of Antiderivatives and Line Integrals): Suppose that f (z) is a continuous function on a
region R and γ is a contour in R. Then the following hold:

1. If R is connected and f 0 (z) = 0 on R, then f is constant on R.

◦ Proof: First suppose z, z 0 ∈ R and suppose that there is a continuously dierentiable contour
γ : [a, b] → C that connects them in R.
◦ Then the function g : [a, b] → C given by g(t) = f (γ(t)) is dierentiable so by the chain rule
g 0 (t) = f 0 (γ(t))γ 0 (t) = 0 identically. But this means the real part and imaginary part of g are both
constant (since they are dierentiable real-valued functions with zero derivative), so g is constant.
f (z) = g(a) = g(b) = f (z 0 ).
In particular

◦ Now for general z, z 0 ∈ R since R is connected there exists a path γ = γ1 ∪ γ2 ∪ · · · ∪ γn where γi

is continuously dierentiable and starts at zi−1 and ends at zi , from z1 to z2 . Applying the result
0
above to each γi shows f (z) = f (z1 ) = f (z2 ) = · · · = f (zn ) = f (z ). Thus the value of f is the same
at each point in R, meaning that f is constant.

2. (Uniqueness of Antiderivatives) If F and G are holomorphic on R and F 0 (z) = G0 (z) on R, then F (z) =
G(z) + c for some c ∈ C.
◦ Proof: Apply (1) to F (z) − G(z), which has derivative zero hence is constant on R.
0
3. (Independence of Path) Suppose that F (z)
´ is holomorphic on a region R with F (z) = f (z) and that γ
is any contour in R from z0 to z1 . Then
γ
f (z) dz = F (z1 ) − F (z0 ).
◦ Note that this is the analogue of the second part of the usual real fundamental theorem of calculus:
´b
if F (x) is dierentiable on [a, b] with F 0 (x) = f (x), then
a
f (x) dx = F (b) − F (a).
◦ We refer to this result as independence of path (particularly in the context of real line integrals in
the plane), since it states that the integral of F 0 (z) from z0 to z1 depends only on the values of F (z)
at the endpoints, and not on the specic path of integration between them.

◦ Proof: First suppose that γ is continuously dierentiable, so that γ : [a, b] → R is continuously

dierentiable.
d
◦ Then by the chain rule we have [F (γ(t))] = F 0 (γ(t))γ 0 (t) = f (γ(t))γ 0 (t).
dt
´ ´b ´b d
◦ Therefore
γ
f (z) dz = a f (γ(t))γ 0 (t) dt = a [F (γ(t))] dt = F (γ(b)) − F (γ(a)) by the usual
dt
fundamental theorem of calculus.

◦ For the general result for arbitrary contours, say γ = γ1 ∪ γ2 ∪ · · · ∪ γn where γi is continuously
´
dierentiable and starts atzi−1 and ends at zi , by our result on gluing curves we have γ f (z) dz =
Pn ´ Pn
i=1 γi f (z) dz = i=1 [F (zi ) − F (zi−1 )] = F (zn ) − F (z0 ) as required.

4. (Necessity for Antiderivatives) Suppose that

´ F (z) is holomorphic on a region R with F 0 (z) = f (z) and
that γ is any closed contour in R. Then
γ
f (z) dz = 0.
◦ By changing focus from F (z) to f (z) here, we obtain a necessary condition for the existence
´ of an
F (z) of f (z) on R: namely, for any closed contour γ in R, we must have γ f (z) dz = 0.
antiderivative
´
◦ Proof: By (3) we have γ f (z) dz = F (z1 ) − F (z0 ) where γ starts at z0 and ends at z1 . But z0 = z1
for a closed contour, so F (z1 ) − F (z0 ) = 0.

5. (Existence of Antiderivatives, I) Let

´ R be a connected open set and f (z) be a continuous function on R
such that f (z) dz = 0 for any closed contour in R. Choose any a ∈ R and then for each z ∈ R select
γ ´
any path γ in R from a to z . Then F (z) = γ f (z) dz is well dened and has F 0 (z) = f (z) for all z .
◦ Note that this is the analogue of the rst part of the fundamental theorem of calculus for real
´x
functions, which states that if f is continuous and a ∈ R, then F (x) = a
f (t) dt is dierentiable
and has derivative F 0 (x) = f (x).

6
 ´
◦ However, here we have the extra hypothesis that
γ
f (z) dz = 0 for any path γ , which is necessary
by (4). So the point of this result is that the condition of (4) is both necessary and sucient for an
antiderivative to exist.

◦ Proof: First, to show that F (z0 ) is well dened for each z0 ∈ R , since R is connected (hence path-
connected since it is open) there is at least one contour from a to z0 . Now suppose we have two
paths γ1 and γ2 a to z0 in R.
from
´ ´ ´ ´
◦ Then by setting γ = γ1 ∪ (−γ2 ), we see that γ f (z) dz = γ1 f (z) dz + −γ2 f (z) dz = γ1 f (z) dz −
´ ´ ´
γ2
f (z) dz , and so γ1 f (z) dz = γ2 f (z) dz . Therefore, the value of F (z0 ) in our denition above
does not depend on which path we take from a to z0 : this means F (z0 ) is well dened.

◦ Now for the derivative, since R is open and z0 ∈ R, there exists some r > 0 such that the open
F (z0 + h) − F (z0 ) 1´
disc |z − z0 | < r is contained in R. We have = f (z) dz , where the integral
h h γ
is taken along any path γ from z0 to z0 + h.

◦ Since we are only interested in the limit as h → 0, and R is open, we may restrict attention to
|h| < r. Then the line segment from z0 to z0 + h is fully contained in the open disc |z − z0 | < r, so
we may take γ to be that line segment, which is parametrized by γ(t) = z0 + ht for 0 ≤ t ≤ 1.
F (z0 + h) − F (z0 ) 1´ 1 ´1 ´1
◦ Then we have = γ
f (z) dz = 0
f (z0 + ht) h dt = 0 f (z0 + ht) dt.
h h h
◦ Since f is continuous on the open disc, for any  > 0 there exists δ > 0 such that |f (z0 ) − f (z)| < 
´1
for |z − z0 | < δ . Then by the triangle inequality for integrals we have [f (z0 + ht) − f (z0 )] dt ≤
0
´1
0
|f (z0 + ht) − f (z0 )| dt <  under the hypothesis that |h| < δ .
´1
◦ Therefore, taking h → 0 in 0 f (z0 + ht) dt yields a value within  of f (z0 ) for every  > 0,
´1 F (z0 + h) − F (z0 )
whence limh→0
0
f (z0 + ht) dt = f (z0 ). Thus, we have F 0 (z0 ) = limh→0 =
´1 h
limh→0 0 f (z0 + ht) dt = f (z0 ) as required.
´
• Example: Evaluate the complex line integral γ z n dz for each integer n 6= −1, where γ is the line segment
from 0 to i.

z n+1
◦ Note that for f (z) = z n we have an antiderivative F (z) = .
n+1
´ z n+1 i in+1
◦ Therefore, by applying independence of path, we see that
γ
z n dz = |z=0 = .
n+1 n+1
◦ Note that we could also compute the integral using a parametrization: with γ(t) = it for 0 ≤ i ≤ 1, we
´1 i n+1 n+1
t in+1
have γ 0 (t) = i and f (z) = z n = (it)n , so the integral is
0
(it)n
· i dt = |1
t=0 = , which is of
n+1 n+1
course the same. The appeal of using path-independence is that we see that the value of the integral is
independent of the shape of the actual contour γ.

• Interestingly, the existence (and nonexistence) of antiderivatives depends both on the function f (z) and on
the region R on which we are trying to construct an antiderivative.

• Example: Show that f (z) = z −1 has no antiderivative on the annulus 1/2 < |z| < 2.

◦ As we calculated previously, the integral of f (z) around the contour γ that winds once counterclockwise
around the unit circle is 2πi 6= 0.
◦ Therefore, f (z) cannot have an antiderivative on R, since if it did, the integral around every closed
contour would be zero.

◦ Remark: More generally we can show that f (z) has no antiderivative on any region of the form a < |z| < b
by integrating around an appropriate circle |z| = r for some a < r < b.

• Example: Show that f (z) = z −1 does have an antiderivative on the disc |z + 1| < 1.

◦ One way to construct this antiderivative is to nd a series expansion for f (z) centered at z=1 and then
antidierentiate it term-by-term inside its radius of convergence.

7
 1 P∞ P∞
◦ Letting w = z+1, we want to nd an antiderivative for (w−1)−1 = − = − n=0 wn = − n=0 (z+
1−w
1)n , which converges absolutely for |z + 1| < 1.
P∞ (z + 1)n+1
◦ Then integrating termwise yields the antiderivative F (z) = − n=0 , which also converges
n+1
absolutely for |z + 1| < 1 and is holomorphic there with derivative f (z).
◦ Another approach to constructing this antiderivative would be to select a branch of the complex logarithm
function with a branch cut that does not intersect the disc |z + 1| < 1.
◦ In fact, since the principal complex logarithm Log(z) is holomorphic on C\[0, ∞), it provides such an
antiderivative.

3.2 Cauchy's Integral Theorem and Formula

• We would like to analyze the existence of antiderivatives on regions more cleanly, since as a practical matter
´
our result that a continuous f (z) has an antiderivative on R if and only if
γ
f (z) dz = 0 for all contours in R
is not practical to verify for specic functions f (z).

◦ By necessity, our approach to giving a simpler condition will be somewhat technical, but the short version
is that we will want to impose a topological condition on R known as simple connectivity, and we will
also require f (z) to be holomorphic.

◦ We will begin by motivating the utility of these two conditions using Green's theorem to prove a weaker
version of Cauchy's integral theorem, and then we will provide another approach using deformation of
contours that also illustrates the necessity of these conditions and gives a stronger statement.

◦ We then motivate the second half of our discussion, regarding Cauchy's integral formula, by using our
results to greatly simply evaluations of contour integrals of functions dened as convergent Laurent series.

◦ Finally, we state and prove the full version of Cauchy's integral formula.

3.2.1 Cauchy's Integral Theorem via Green's Theorem

• First, we recall a basic result from topology:

• Theorem (Jordan Curve Theorem): Suppose γ : [a, b] → C is a closed, simple, continuous curve. Then its
complement C\im(γ) has exactly two connected components: its unbounded exterior and its bounded interior,
each of whose boundary is the curve im(γ).

◦ This theorem is quite dicult to prove for general continuous curves, and ultimately the theorem is quite
topological in nature.

◦ For smooth curves (i.e., curves that are continuously dierentiable and where the derivative is everywhere
nonzero), one may give a fairly intuitive procedure for deciding whether a point P is in the interior or
exterior.

◦ Explicitly, by smoothness, there are only nitely many line segments in γ. Now given any point P ∈ C,
draw a ray starting at P in any direction not parallel to any of these line segments, and then count the
number of times the ray crosses the curve (which is necessarily nite by smoothness). If the number
of crossings is even, then P is in the exterior, while if the number of crossings is odd, then P is in the
interior.

◦ To make this argument rigorous, one must then show that the number of crossings does not depend on
the ray chosen, that the regions of odd points and even points are connected, and that there is no
path from an odd point to an even point. We omit the precise details.

• By the Jordan curve theorem, if γ is a simple closed rectiable curve, then it encloses an interior region R,
which by rectiability will always be on the same side (either left or right) of γ as one travels around the
curve.

8
 ◦ By reversing the direction of travel (i.e., by replacing γ with −γ , which as we have noted earlier simply
negates the associated line integral) if needed, we can also assume that γ has counterclockwise orientation,
meaning that R is always on the left of γ.

• Next we recall Green's theorem from multivariable calculus:

• Theorem (Green's Theorem): If γ is a simple closed rectiable curve oriented counterclockwise in R2 , and
R
ˆ is the region it encloses, then for any continuously dierentiable functions
¨   P (x, y) and Q(x, y) on R,
∂Q ∂P
P dx + Q dy = − dy dx.
γ R ∂x ∂y

◦ Green's theorem is a natural generalization of the fundamental theorem of calculus to planar regions.

◦ For illustration we will give the proof of Green's theorem for rectangular regions. One may use this
calculation to establish the general proof by observing that both the line integral and the double integral
are compatible with gluing two regions together along a shared boundary, and then showing that a
suitable rectangular approximation of the region R yields close approximations to both the line integral
and the double integral.
´ ´ ´ ´ ´
◦ Proof (for rectangles): For a rectangular region a ≤ x ≤ b, c ≤ y ≤ d,
γ
= γ1 + γ2 + γ3 + γ4 , we have

where γ1 , γ2 , γ3 , and γ4 are the four sides of the rectangle (with the proper orientation), and the function
P dx + Q dy .
to be integrated on each curve is
´ ´ ´b
◦ Setting up parametrizations shows γ1 [P dx + Q dy] + γ3 [P dx + Q dy] = a [P (x, c) − P (x, d)] dx, and
´ ´ ´d
γ2
[P dx + Q dy] + γ4 [P dx + Q dy] = c [Q(b, y) − Q(a, y)] dy .
˜ ∂P ´ b ´ d ∂P ´b
◦ For the double integral we have
R
− dy dx = a c − dy dx = a [P (x, c) − P (x, d)] dx, and
∂y ∂y
˜ ∂Q ´ c ´ b ∂Q ´d
R ∂x
dx dy = d a dx dy = c [Q(b, y) − Q(a, y)] dy .
∂x
´ ˜
 
∂Q ∂P
◦ Comparing the expressions shows immediately that C [P dx + Q dy] = R − dy dx, as de-
∂x ∂y
sired.

• We can convert this statement to a complex analogue by viewing R2 as C and allowing P (x, y) and Q(x, y)
to be complex-valued.

◦ The only additional task is to explain how to interpret line integrals with dierentials dx and dy rather
than dz .
´ ´
◦ An easy approach is simply to write down Riemann sum denitions for f (z) dx and γ f (z) dy , in
´ γ
the same way we wrote a Riemann sum denition for
γ
f (z) dz : the only dierence is that instead of
multiplying the value f (zi∗ ) by zi − zi−1 , we multiply it by xi − xi−1 = Re(zi ) − Re(zi−1 ) or yi − yi−1 =
Im(zi ) − Im(zi−1 ).
´ ´
◦ By following these denitions through, we see immediately that f (z) dz = f (z)[dx + i dy] =
´ ´ γ γ
γ
f (z) dx + i γ
f (z) dy .
◦ Equivalently, and more simply, we have the dierential identity dz = dx + i dy , as one would expect
naturally from z = x + iy .
◦ Furthermore, if we have a parametrization γ(t) = x(t) + iy(t) for a ≤ t ≤ b, then we also have the
´ ´b ´ ´b
expected formulas
γ
f (z) dx = a
f (γ(t)) x (t) dt and γ f (z) dy = a f (γ(t)) y 0 (t) dt.
0

• With these denitions in hand, we can pose the complex version of Green's theorem:

• Corollary (Complex Green's Theorem): If γ is a simple closed rectiable curve oriented counterclockwise in
C, and R is the region it encloses, then for any continuously dierentiable functions
ˆ ¨   P (x, y) and Q(x, y) on
∂Q ∂P
R, we have P dx + Q dy = − dy dx.
γ R ∂x ∂y

◦ The proof follows immediately by breaking P and Q apart into real and imaginary parts and applying
the real version of Green's theorem to each.

9
 • Much more interesting is the result when we apply the complex Green's theorem to a line integral using the
dierential dz and a function f (z) = u(z) + iv(z).

◦ Here, we require that f (z) = u(x + iy) be continuously dierentiable as a function of x and y on R,
which as we have seen is implied by being holomorphic on R.

• Theorem (Cauchy's Integral Theorem, First Version): If γ is a simple closed rectiable curve oriented coun-
terclockwise in
ˆ C, and R is the region it encloses, then for any holomorphic function f (z) on R we have

f (z) dz = 0.
γ
´ ´ ´
◦ Proof: We have
γ
f (z) dz = γ
f (z)[dx + i dy] = γ
f (z) dx + if (z) dy .
∂f
◦ So if we apply the complex Green's theorem with P = f (z) and Q = if (z), then recalling that =
 ∂z
´ ˜
  
1 ∂f ∂f ∂f ∂f
+i = 0 because f is holomorphic, we obtain γ f (z) dx+if (z) dy = R i − dy dx =
2 ∂x ∂y ∂x ∂y
˜ ˜ ∂f ˜

∂f ∂f
2i R +i dy dx = 2i R dy dx = 2i R 0 dy dx = 0.
∂x ∂y ∂z
´
◦ Therefore, γ f (z) dz = 0, as claimed.

• We will remark, as a key point, that although we are only computing the line integral around the boundary
γ of R, in fact we do require f (z) to be holomorphic on all of the interior region R for the theorem to hold.

◦ Clearly this fact is required in the proof, since we need to be able to evaluate the double integral
˜ ∂f
R ∂z
dy dx on R.
◦ However, as the example of f (z) = 1/z with γ the counterclockwise boundary of the unit circle shows, we
cannot even drop the assumption of holomorphicity at a single point in
´ R without making the theorem
false, since
γ
1/z dz = 2πi is nonzero yet f (z) = 1/z only fails to be holomorphic at the single point
z = 0.
◦ If we switch focus from the curve γ to the region R, we can clearly drop the requirement that γ run
counterclockwise, since reversing the direction will simply scale the integral by −1, it will still be zero.
So the regions R on which the theorem applies are obtained by lling in the interior of a simple closed
rectiable curve.

3.2.2 Cauchy's Integral Theorem via Deformation of Contours

• Intuitively, we can describe the regions R in our statement of Cauchy's integral theorem above as the ones
that have no holes in them. We would like to formalize this notion, which we do using homotopy.

◦ For convenience, by rescaling the parametrization, we will assume that all of our curves are parametrized
as γ : [0, 1] → C.

• Denition: Suppose that γ0 : [0, 1] → R and γ1 : [0, 1] → R are continuous closed curves in a region R. We say
that a function h : [0, 1] × [0, 1] → R is a homotopy in R from γ0 to γ1 when h is continuous, h(s, 0) = h(s, 1)
for all s, and h(0, t) = γ0 (t) and h(1, t) = γ1 (t) for all t ∈ [0, 1]. If there exists some homotopy in R from one
curve to another, we say the curves are homotopic in R.

◦ A homotopy is a continuous deformation of γ0 into γ1 : the idea is, roughly, that the homotopy describes
how to slide γ0 through the plane so that it becomes γ1 . Explicitly, we describe this deformation
via the intermediate curves γs : [0, 1] → C γs (t) = h(s, t) for each 0 ≤ s ≤ 1. The condition
where
h(s, 0) = h(s, 1) is then simply that each intermediate curve γs is also closed.

◦ It is not hard to verify that being homotopic in R is an equivalence relation on continuous curves in R.
Additionally, whether or not two curves are homotopic will depend on the region R in addition to the
curves.

10
 ◦ Example: If R = C and γ1 : [0, 1] → C is any continuous closed curve, then γ1 is homotopic to the
trivial curveγ0 (t) = 0 for all 0 ≤ t ≤ 1, via the homotopy h(s, t) = sγ1 (t). We can see that h(s, t) is
continuous (since it is the product of two continuous functions), that h(s, 0) = sγ1 (0) = sγ1 (1) = h(s, 1),
and that h(0, t) = 0 = γ0 (t) and h(1, t) = γ1 (t), so it is indeed a homotopy.

◦ As a consequence, since homotopy is an equivalence relation, we see that any two continuous closed
curves are homotopic in C.
◦ It may therefore seem that our denition of homotopic curves is rather trivial, but that is only because
there are no obstacles to deforming curves in C. If we remove points from C, we can create obstacles
that prevent two curves from being homotopic.

◦ Example: If R = C\{0} and γ1 is the curve winding once counterclockwise around the unit circle, then
γ1 γ0 (t) = 1 for all 0 ≤ t ≤ 1. This result is not so easy to prove
is not homotopic to the trivial curve
2
rigorously , but the point is that since γ0 contains 0 in its interior, any continuous deformation into
another curve γ1 that does not contain 0 in its interior must have an intermediate curve γs in R that
passes through 0, which is not allowed since R does not contain 0.

• For nice regions, such as the interior R of a continuous curve (per the Jordan curve theorem), every continuous
closed curve in R is homotopic to a point (i.e., a constant curve). These are the regions we wish to discuss:

• Denition: If R is a region in C, we say R is simply connected if R is connected and every continuous curve
γ : [0, 1] → R is homotopic to a point.

◦ Intuitively, the idea is that if we draw any curve in R, we may simply shrink the curve (without leaving
R) until it is a single point. This prevents the region from having any holes, since a curve drawn around
such a hole could not be shrunk to a point.

◦ As is often shown as part of the Jordan curve theorem, if R is the interior (or exterior) of a region whose
boundary is a continuous closed curve, then R is simply connected. In particular, every region to which
we can apply Green's theorem (and thus the Cauchy integral theorem we proved using Green's theorem)
is simply connected.

• We will now give another proof of Cauchy's integral theorem that relies on deformation of contours. The main
engine for this approach is the following very useful fact:

• Theorem (Deformation of Contours): Suppose R is a region such that

´ γ1 and γ2 ´are closed rectiable curves
in R that are homotopic in R and f (z) is holomorphic on R. Then
γ1
f (z) dz = γ2 f (z) dz .

◦ For arbitrary homotopies this result requires stronger techniques using approximations. We will give a
proof in the simpler case where the homotopy function h(s, t) is twice continuously dierentiable in s
and t and f 0 (z) is continuous.

◦ We recall Leibniz's rule for dierentiation under the integral sign, which allows us to change the order of an
integral and a partial derivative, analogously to how Clairaut's theorem allows us to interchange the order
of partial derivatives (when those associated partial derivatives are continuous) and Fubini's theorem
allows us to interchange the order of multiple integrals (when the integrand is absolutely integrable).

◦ Explicitly, Leibniz's rule says that if f (x, t) and its x-partial fx (x, t) are continuous in x and t, then
d h´ b i ´
b
f (x, t) dt = a fx (x, t) dt.
dx a
◦ Proof (Special Case): Suppose that γ1 : [0, 1] → R and γ2 : [0, 1] → R are closed rectiable curves in R
and h(s, t) is a twice continuously dierentiable homotopy from γ1 to γ2 .
´1
◦ Consider the function I(s) = 0
f (h(s, t)) ht (s, t) dt: we will show I 0 (s) = 0.
◦ Since f , h, and ht are all continuously dierentiable in all variables, so is the product f (h(s, t)) ht (s, t).
2 A direct approach is as follows: rst, the region C\{0} can be continuously deformed onto the onto the unit circle via the homotopy
h(s, reiθ ) = (1 − r cot(sπ/2))eiθ , so it suces to show that the path winding once around the unit circle is not homotopic to a constant
map when the region R is just the unit circle itself. To show this consider the set S of s ∈ [0, 1] such that −1 6∈ γs . Then since h(s, t) is
continuous, S is closed and also S c must be closed (these follow by using /2 arguments in either direction). But since 0 ∈ S and 1 ∈ S c
this would contradict the connectedness of the interval [0, 1], which is a contradiction. Another less ad hoc approach is to use properties
of covering spaces, or equivalently to show that the path winding once around the unit circle is a generator of the fundamental group
π1 (S 1 ) ∼
=Z of the circle S1 while the constant path corresponds to the identity element.

11
 ◦ Thus, by Leibniz's dierentiation rule, the product and chain rules, Clairaut's theorem applied to obtain
hst = hts , and the fundamental theorem of line integrals, we have

ˆ 1  ˆ 1
d ∂
I 0 (s) = f (h(s, t)) ht (s, t) dt = [f (h(s, t)) ht (s, t)] dt
ds 0 0 ∂s
ˆ 1
= [f 0 (h(s, t)) hs (s, t)ht (s, t) + f (h(s, t)) hst (s, t)] dt
0
ˆ 1
= [f 0 (h(s, t)) ht (s, t)hs (s, t) + f (h(s, t)) hts (s, t)] dt
0
ˆ 1
∂
= [f (h(s, t))hs (s, t)] dt
0 ∂t
= f (h(s, 1))hs (s, 1) − f (h(s, 0))hs (s, 0)

and now nally this last expression is zero because the function γs : [0, 1] → R with γs (t) = h(s, t) is a
closed curve, so h(s, 1) = h(s, 0) and hs (s, 1) = hs (s, 0).
◦ Thus I 0 (s) = 0 0 < s < 1, so I is constant on [0, 1] hence in particular we have I(0) = I(1).
for
´ ´1 ´1 ´
◦ Finally, since γ1 f (z) dz = 0 f (γ1 (t)) γ10 (t) dt = 0 f (h(0, t)) ht (0, t) dt = I(0) and similarly γ2 f (z) dz =
´1 ´1 ´ ´
0
f (γ2 (t)) γ20 (t) dt = 0 f (h(1, t)) ht (1, t) dt = I(1), we obtain γ1 f (z) dz = γ2 f (z) dz as claimed.

• Using this result we can evaluate many contour integrals by transforming the shape into one where the integral
is easier to calculate via a parametrization.

◦ As an actual practical matter, the easiest closed curve to parametrize is a circle, since we do not need to
break it into several components like we would for a triangle or rectangle.

´
• Example: Evaluate
γ
z −1 dz where γ(t) = 2 cos(t) + 4i sin(t) for 0 ≤ t ≤ 2π .

◦ We see that f (z) = z −1 is holomorphic on R = C\{0}, so we may freely deform the contour as long as
we avoid the point z = 0.

◦ Since the contour is simply the counterclockwise ellipse x2 /4 + y 2 /16 = 1, we may simply stretch it
2 2
continuously until it becomes the counterclockwise unit circle x + y = 1, without causing the curve to
pass through z = 0.
´ ´
◦ Thus, if γ2 is the boundary of the unit circle, we see
γ
z −1 dz = γ2
z −1 dz = 2πi as we have previously
calculated.

´ z
• Example: Evaluate
γ
dz where γ is the counterclockwise boundary of the square with vertices
z2 − 2z + 1
±3 ± 3i.

◦ We see that f (z) is holomorphic on R = C\{1} so we may freely deform the contour as long as we avoid
z = 1.
◦ A quick sketch shows that we can continuously shrink the contour until it becomes the circle |z − 1| = 2
without passing through z = 1.
◦ We can parametrize this circle as γ(t) = 1 + 2eit for 0 ≤ t ≤ 2π , with γ 0 (t) = 2ieit . Then the integrand is
z 1 + 2e it
1 ´ z ´ 2π 1 −2it
f (z) = = = (e−2it +2e−it ), so γ 2 dz = 0 (e +2e−it )(2ieit ) dt =
z 2 − 2z + 1 (2eit )2 4 z − 2z + 1 4
´ 2π i −it
0 2
(e + 2) dt = 2πi .

• In the situation where R is simply connected, every continuous closed curve is homotopic to a point. Applying
this fact to our deformation of contours result immediately yields Cauchy's integral theorem (again), but with
a less restrictive hypothesis:

• Corollary (Cauchy's Integral Theorem, Second Version): If γ is a closed rectiable curve and
ˆ R is a simply

connected region, then for any holomorphic function f (z) on R we have f (z) dz = 0.
γ

12
 ◦ Proof: Suppose γ is a closed rectiable curve and R is a simply connected region. Then by denition, γ
is homotopic to a point: namely, a constant curve γ0 .
´ ´
◦ By the deformation of contours theorem, we then have
γ
f (z) dz = γ0 f (z) dz and this last integral is
zero since the contour is trivial. (Indeed, all of its associated Riemann sums are even zero.)

• As a consequence of this stronger version of Cauchy's integral theorem, we also see that every holomorphic
function on a simply-connected region has an antiderivative:

• Corollary (Existence of Antiderivatives, II): Suppose R is a simply connected open region and f (z) is holo-
morphic on R. Then there exists a holomorphic function F (z) such that F 0 (z) = f (z) on R.
ˆ
◦ Proof: By Cauchy's integral theorem, we have f (z) dz = 0 for all closed rectiable curves in R.
γ
◦ Since f (z) is continuous, by our previous result on existence of antiderivatives, for any xed
´ a ∈ R and
rectiable path γ0 from a to z0 (which exists since R is connected), the function F (z0 ) = γ0
f (z) dz has
0
F (z) = f (z) on R.

• By deforming contours in more complicated ways, and also by simplifying integrands, we can often reduce
dicult contour integrals to simpler ones that can be evaluated directly.

´ 4z − 4
• Example: Evaluate
γ
dz where γ(t) = 1 + 2eit for 0 ≤ t ≤ 2π .
z 2 − 2z
◦ We could simply plug in the parametrization, but the resulting rational function in sin t and cos t is
rather messy and unpleasant to evaluate (though of course, it is possible).

◦ Additionally, the integrand is holomorphic on R = C\{0, 2} so we may freely deform the contour as long
as we avoid both of those points.

◦ Now imagine pinching the circle in the middle along Re(z) = 1 until it reaches a gure eight shape,
and then round out the sides to form two circles:

◦ Since contour integrals are additive on curves, it suces to calculate the integral on the curve γ1 (the
counterclockwise unit circle) and γ2 (the counterclockwise circle |z − 2| = 1).
4z − 4 2 2
◦ Furthermore, from partial fraction decomposition we have = + , so we need only calculate
z 2 − 4z z z−2
´ 1 ´ 1 ´ 1 1 ´
γ1
dz , γ1 dz , γ2 dz , . and
γ2
z z−2 z z−2
´ 1 ´ 1
◦ Using our previous calculations we have γ1 dz = 2πi and also γ2 dz = 2πi (since this integral is
z z−2
the same as the rst one after substituting w = z − 2).
1 1
◦ Furthermore, since is holomorphic on the region enclosed by γ2 , and is holomorphic on the
z z−2
´ 1 ´ 1
region enclosed by γ1 , we also have dz = γ2 dz = 0.
γ1 z − 2 z
´ 4z − 4 ´ 2 ´ 2 ´ 2 ´ 2
◦ Putting all of this together shows that γ 2 dz = γ1 dz + γ1 dz + γ2 dz + γ2 dz =
z − 2z z z−2 z z−2
2(2πi) + 2(2πi) = 8πi .

13
3.2.3 Integration of Power Series and Laurent Series

• Our next goal is to discuss integration of functions dened by convergent Laurent series. We might hope that
we could simply integrate such series term by term, and that is indeed the case inside their (punctured) disc
of convergence:

• Proposition (Integration and Laurent Series): Suppose γ is a rectiable curve in the region R. Then the
following hold:

´ ´
1. If the continuous functions fn (z) converge uniformly to f (z) on R, then limn→∞ γ
fn (z) dz = γ
f (z) dz .
◦ Similar to many other results involving uniform convergence, this result allows us to interchange the
integral and the limit.
´b
◦ We will also use a fact about lengths of curves: formally, the length of γ is dened as L= a
|γ 0 (t)| dt.
0
Since γ is continuous except at a nite number of points, the length of γ is nite.

◦ Proof: As we have shown previously, the uniform limit of continuous functions is continuous, so f (z)
is continuous and hence integrable.

◦ Now let  > 0 and suppose that γ : [a, b] → C has length L. By uniform convergence there exists N

such that |fn (z) − f (z)| < for all z ∈ R and all n ≥ N .
L
´ ´ ´b ´b
◦ Then γ fn (z) dz − γ f (z) dz = a [fn (γ(t)) − f (γ(t))]γ 0 (t) dt ≤ a |fn (γ(t)) − f (γ(t))| |γ 0 (t)| dt <
´b  0  ´b 0
a L
|γ (t)| dt = |γ (t)| dt =  by the triangle inequality.
L a
´ ´ ´
◦ Thus by the denition of limit, since γ fn (z) dz − γ f (z) dz <  for all n ≥ N , we have limn→∞ γ fn (z) dz =
´
γ
f (z) dz as claimed.
P∞ n
2. If the Laurent series f (z) = n=−k an (z − a) has positive radius of convergence R, and
´ γ is a rec-
tiable closed curve contained entirely within the punctured disc 0 < |z − a| < R, then f (z) dz =
P∞ ´ n
γ
n=−k an γ (z − a) dz .

◦ Proof: Since γ : [0, 1] → C is continuous, the function g(t) = |γ(t) − a| is continuous on [0, 1] hence
by the extreme value theorem it attains its minimum and maximum values.

◦ Since γ cannot pass through 0, the minimum must be some positive value rmin > 0, and since γ is
contained within the disc |z − a| < R the maximum must be some value rmax < R.
Pd
◦ On the closed disc rmin ≤ |z − a| ≤ rmax , as we have shown, the partial sums fd (z) = n=−k ad (z −
d
a) are continuous and converge uniformly to f.
´ Pd
◦ Thus by (1), the integral of their limit f (z) is equal to the limit limd→∞ [ n=−k an (z − a)n ] dz ,
P∞ ´ γ
which equals n=−k an γ
(z − a)n dz since the sum is nite.

3. If γ is any rectiable closed curve in the punctured disc 0 < |z − a| < R and n is any integer not equal
´
to −1, then γ (z − a)n dz = 0. If n ≥ 0 then the result also holds if γ passes through z = a.
(z − a)n+1
◦ Proof: For n 6= −1 we see that (z − a)n has an antiderivative dened for all z 6= a (if
n+1
n ≥ 0, z = a).
of course, this function is also dened at
´n
◦ But then
γ
(z − a) dz = 0 by the fundamental theorem of line integrals, since γ is closed.
P∞ n
4. If the Laurent series f (z) = n=−k an (z − a) has positive radius of convergence ´R, and γ is a rectiable
´
closed curve contained entirely within the punctured disc 0 < |z − a| < R, then f (z) dz = a−1 γ (z −
γ
a)−1 dz .
◦ Proof: This follows immediately by plugging (3) into (2).

• Quite interestingly, we see that the value of the integral of a function dened by a convergent Laurent series
depends only on the coecient a−1 of the series along with the value of the integral of (z − a)−1 around the
contour.

◦ As noted previously, this behavior is intimately linked to the fact that z −1 has no antiderivative on a
punctured disc of positive radius around 0.

14
 ◦ By deforming contours appropriately, we can evaluate any contour integral of this form.

◦ γ is the circle |z − a| = r traversed once counterclockwise, then by using

The main idea is to observe that if
´ ´ 2π 1 ´ 2π
it
the parametrization z = a + re for 0 ≤ t ≤ 2π , we see
γ
(z − a)−1 dz = 0 it
ireit dt = 0 i dt = 2πi,
re
regardless of the value of r .
´
◦ Thus, we may complete our evaluation of γ f (z) dz by deforming the contour into one that traverses a
circle centered at z = a some number of times.

◦ We will content ourselves only to give some motivating examples at the moment, since we aim ultimately
to generalize this discussion to all holomorphic functions, rather than just those dened locally by a
convergent Laurent series.

´ ez
• Example: Find
γ z4
dz where γ(t) = eit for 0 ≤ t ≤ 2π .

ez −4 −3 z −2 z −1 1
◦ As a Laurent series centered at z = 0, we have = z + z + + + +···, which converges
z4 2! 3! 4!
for all z 6= 0. Note also that γ traverses the unit circle once counterclockwise.

´ ez 1 ´ −1 1 1
◦ Thus, by our discussion above, we see that
γ z3
dz = z dz = (2πi) = πi .
3! γ 3! 3
´
• Example: Find
γ
csc(2z) dz where γ(t) is the counterclockwise boundary of the square with vertices ±1, ±i.

1 1 1 2
◦ As a Laurent series we have csc(2z) = = = z −1 + z + · · · , which
sin(2z) (2z) − (2z)3 /3! + · · · 2 3!
converges for all z 6= 0 since the series for sin(2z) converges for all z .
´ 1 ´ −1
◦ Therefore, we have
γ
csc z dz = z dz .
2 γ
◦ It is easy to see from the description that we may continuously deform γ into the counterclockwise
boundary γ2 of the unit circle.
´ 1 ´ −1 1´ 1
◦ Thus, we see
γ
csc(2z) dz = γ
z dz = γ
z −1 dz = (2πi) = πi .
2 2 2 2
´ cos z
• Example: Find
γ z4
dz where γ(t) is the closed polygonal path with successive vertices 1, −1 + i, −1 − i,
5 + i, −7 + 3i, −8 + 4i, 4 − 11i, and then nally returning to 1.
cos z 1 1
◦ As a Laurent series we have = z −4 − z −2 + − · · · , which converges for z 6= 0.
z4 2! 4!
◦ ´Since the coecient of z −1 is zero we see that the integral is also 0 , without even needing to evaluate

γ
z −1 dz .
´ cos z
• Example: Find
γ z3
dz where γ(t) is the closed polygonal path with successive vertices 1, −1 + i, −1 − i,
5 + i, −7 + 3i, −8 + 4i, 4 − 11i, and then nally returning to 1.
cos z 1 1
◦ As a Laurent series we have = z −3 − z −1 + z − · · · , which converges for z 6= 0.
z3 2! 4!
´ −1
◦ Now we do need to evaluate
γ
z dz . By drawing the contour, we can see that crosses itself, and winds
around 0 twice. By separating the contour into the two separate pieces and then deforming each of the
´ ´
resulting simple loops into the unit circle, we can see that
γ
z −1 dz = 2 γ2
z −1 dz = 4πi.
´ cos z 1 ´ −1
◦ Hence by our results we have
γ z3
dz = − z dz = −2πi .
2! γ

15
3.2.4 Winding Numbers

• Our calculations with integrating Laurent series indicated that for any f (z) described by a convergent Laurent
series centered at z = z0 and any closed γ that lies within its disc of convergence, we can reduce the calculation
´ ´ 1
of
γ
f (z) dz to the calculation of
γ z−z
dz .
0

◦ Our goal is to show that this same sort of reduction can be made for arbitrary holomorphic functions.

´ 1
• First, we will study the integrals
γ
dz in more detail, where γ is any closed contour not passing through
z − z0
z0 .

◦ As we have shown several times, if γ is any circle of positive radius centered at z = z0 traversed once
1 ´
counterclockwise, the integral
γ z−z
dz has value 2πi. More generally, if γ traverses the circle n times
0
´ 1
counterclockwise (where n could be 0 or negative), then
γ z−z
dz = n · 2πi.
0
◦ Furthermore, since the value of the contour integral is invariant under continuous deformation of the
contour, if we have any contour that is homotopic in C\{0} to a circular path that traverses the unit
´ 1
circle n times counterclockwise, then
γ
dz = n · 2πi.
z − z0
1
◦ Intuitively, the antiderivative of , in its most general possible sense, should be the complex loga-
z − z0
rithm log(z − z0 ): but this is a multivalued function, so it does not make sense, strictly speaking, to use
it to evaluate the integral using the fundamental theorem of line integrals.
´ 1
◦ Nonetheless, formally, if γ : [a, b] → C, we would want to write something along the lines of γ
dz =
z − z0
log(γ(b) − z0 ) − log(γ(a) − z0 ). Then since γ(a) = γ(b), if we allow each of the logarithms to take any of
their possible values, the possible values of the dierence are the integer multiples of 2πi.
◦ This calculation, although not rigorous, suggests that we should expect the actual value of this integral
´ 1
γ
dz to be an integer multiple of 2πi.
z − z0
◦ In fact, if we instead work with a specic branch of the logarithm (e.g., the principal logarithm Log(z)),
we can pin down more precisely what this multiple of 2πi represents.

◦ Specically, the branch cut of the logarithm will separate γ into some nite number of pieces. Then we
´ 1
can sum the contributions made to
γ
dz on all of the pieces of γ separated by the branch cut, and
z − z0
´b 1
then keep track of the dierence in the value of
a
dz and the value Log(b − z0 ) − Log(a − z0 ) each
z − z0
time the curve γ crosses over the branch cut. Because of the nature of the discontinuity, this dierence
will be +2πi if γ crosses the branch cut upwards and −2πi if γ crosses the branch cut downwards.

◦ Tallying up over all of the pieces shows that the integral diers from 0 by an integer multiple of 2πi
given by the total number of upward crossings minus the number of downward crossings of γ across
the branch cut.
1 ´ 1
◦ In this way, we see that the value
γ
dz , roughly speaking, is counting the number of times
2πi z − z0
the contour γ winds around the point z = z0 in the counterclockwise direction.

◦ We now make this notion precise:

• Proposition (Winding Numbers): Suppose γ is any closed contour not passing through z 0 ∈ C. Then the
1 ´ 1
value dz is always an integer; we refer to this integer as the winding number of γ around z0 and
2πi γ z − z0
denote it as Wγ (z0 ).

◦ Proof: By deforming the contour appropriately we may assume that γ is continuously dierentiable
(simply smooth any nondierentiable points, making sure to avoid z0 while doing so). So suppose
´ 1 ´ b γ 0 (t)
γ : [a, b] → C is continuously dierentiable: then
γ z−z
dz = a dt.
0 γ(t) − z0

16
 ´x γ 0 (t)
◦ Now dene the function F (x) = a
dt for a ≤ x ≤ b. By the fundamental theorem of calculus,
γ(t) − z0
γ 0 (x)
we have F 0 (x) = .
γ(x) − z0
d  F (x)
g(x) = eF (x) (γ(x)−z0 )−1 : (γ(x) − z0 )−1 = eF (x) F 0 (x)(γ(x)−

◦ Consider the function its derivative is e
dx
z0 )−1 − eF (x) (γ(x) − z0 )−2 γ 0 (x) = 0 using the formula for F 0 (x) above.
0
◦ Therefore g (x) is identically zero for a ≤ x ≤ b, so g(x) is constant on [a, b]. In particular, g(b) = g(a),
so that eF (b) [γ(b) − z0 ]−1 = eF (a) [γ(a) − z0 ]−1 .
´ a γ 0 (t)
◦ But since γ(b) = γ(a) and F (a) = a dt = 0, we immediately see that eF (b) [γ(b) − z0 ]−1 =
γ(t) − z0
´ b γ 0 (t)
eF (a) [γ(a)−z0 ]−1 implies eF (b) = 1, and so by properties of exponentials this means F (b) = a dt =
γ(t) − z0
´ 1
γ z−z
dz is an integer multiple of 2πi, as claimed.
0
◦ Remark: The motivation for examining the function eF (x) (γ(x)−z0 )−1 is that F (x) should be the complex
logarithm log(γ(x) − z0 ): working with eF (x) allows us to avoid the issues with having a multivalued
F (x)
logarithm, and e should simply be γ(x) − z0 , so it is natural to try to establish the desired result by
showing e
F (x)
(γ(x) − z0 )−1 has zero derivative since this function should just be identically 1.

• Intuitively, the winding number of a contour around a point is the net number of times the contour winds
counterclockwise around that point. We can calculate winding numbers visually:

◦ For the contour γ1 drawn above, the winding number around P is 1, the winding number around Q is 1,
and the winding number around R is 0.

1 ´ ´ 1 ´ 1
◦ As a consequence, we therefore have dz = 2πi, γ1
γ1
dz = 2πi, and γ1 dz = 0.
z−P z−Q z−R
◦ For the contour γ2 drawn above, the winding number around P is 1, the winding number around Q is
−1, and the winding number around R is 0.
´ 1 ´ 1 ´ 1
◦ As a consequence, we therefore have
γ2 z − P
dz = 2πi, γ2 dz = −2πi, and γ2 dz = 0.
z−Q z−R
◦ For the contour γ3 drawn above, the winding number around P is 3, the winding number around Q is 2,
the winding number around R is 1, and the winding number around S is 0.
´ 1 ´ 1 ´ 1
◦ As a consequence, we therefore have
γ3 z − P
dz = 6πi, γ3 dz = 4πi, γ3 dz = 2πi, and
z−Q z−R
´ 1
γ3 z − S
dz = 0.

• We would also expect, based on geometric intuition, that winding numbers should not change as we vary the
point z0 , as long as we do not cross the contour γ, which we can make precise as follows:

• Proposition (Properties of Winding Numbers): Suppose γ is a closed contour in C.

17
 1 ´ 1
1. For z0 not lying on γ, the function Wγ (z0 ) = γ
dz is continuous in z0 .
2πi z − z0
´
 
1 1
◦ Proof: We want limz0 →a Wγ (z0 ) = Wγ (a), which is equivalent to limz0 →a γ
− dz =
z − z0 z−a
0.
◦ Suppose γ : [0, 1] → C. Then since the function g(t) = |γ(t) − a| is continuous on the closed interval
[0, 1], it attains its minimum value M . Note that M > 0 because a does not lie on γ .
◦ For |z0 − a| < M/2 by the triangle inequality we see that |γ(t) − z0 | > M/2 for all t ∈ [0, 1].
1 1 a − z0 |z0 − a| 2
◦ Therefore, for such z0 , we have − = ≤ = 2 |z0 − a|.
z − z0 z−a (z − z0 )(z − a) (M/2)M M
M M2
◦ Now let  > 0 and suppose γ has length s. Then for δ = min( , ), for |z0 − a| < δ
2 2s
´

1 1
we see that |z0 − a| < M/2 so by the calculations above we have − dz ≤
γ z−z z−a
0
1 1 2 2 M2
s · maxz∈γ − ≤ s · 2 |z0 − a| < s · 2 ·  = , as required.
z − z0 z−a M M 2s
´
 
1 1
◦ Therefore limz0 →a
γ z−z
− dz = 0 and so limz0 →a Wγ (z0 ) = Wγ (a) as claimed.
0 z−a
2. If S is a connected region not intersecting the image of γ, then the function Wγ (z) is constant on S.
◦ As an immediate consequence, if γ is a Jordan curve, then the winding number Wγ (z) is constant
on the interior of γ.
◦ Proof: As we showed in (1), Wγ (z) is a continuous function of z. By our proposition above on
winding numbers, Wγ (z) is integer-valued.

◦ Now let a, b be any points in S : since S is connected consider any continuous path in S that joins
them. The winding number Wγ (z) is then a continuous integer-valued function along this path, but
such a function must be constant by the intermediate value theorem (if it took distinct integer values
n1 n2 , then it would also
and take every other real value between n1 and n2 ).
◦ Therefore Wγ (a) = Wγ (b) for any a, b ∈ S , so Wγ (z) is constant on S .

3. If S is a connected unbounded region not intersecting the image of γ, then Wγ (z) is identically zero on
S.
◦ As an immediate consequence, if γ is a Jordan curve, then the winding number Wγ (z) is identically
zero on the exterior of γ.
◦ S is unbounded it contains z0 with arbitrarily large absolute values.
Proof: If

◦ Suppose γ : [0, 1] → C has length s. Since g(t) = |γ(t)| is continuous on [0, 1] it attains its maximum
value M .
1 1
◦ Then by the triangle inequality we have ≤ , yielding the simple estimate |Wγ (z0 )| ≤
z − z0 |z0 | − M
s 1
· .
2πi |z0 | − M
◦ As |z0 | → ∞ this quantity tends to zero, so since it is integer-valued, we must have Wγ (z0 ) = 0 for
suciently large |z0 |.

◦ But then by (2), since Wγ (z) is constant on S and Wγ (z0 ) = 0 for some z0 ∈ S , we have Wγ (z) = 0
for all z ∈ S .

3.2.5 Cauchy's Integral Formula

´
• As we have seen, if f (z) is holomorphic on a simply connected region R, then
γ
f (z) dz = 0.
n
P∞
◦ We have also seen that for a function represented by a Laurent series n=−k an (z − a) with
f (z) =
positive radius of convergence R, and γ is a rectiable closed curve contained entirely within the punctured
´ ´
disc 0 < |z − a| < R, then
γ
f (z) dz = a−1 γ (z − a)−1 dz .
◦ Since the given integral is simply
´ 2πi times the winding number Wγ (a) we see that the value of the
integral
γ
f (z) dz is given by the comparatively simple formula 2πia−1 Wγ (a).

18
 ´
◦ In particular, for any contour winding once counterclockwise around z = a, we see
γ
f (z) dz = 2πia−1 ,
1 ´
or, equivalently, a−1 = f (z) dz .
2πi γ
◦ In other words, we can calculate the coecient of (z − a)−1 in the Laurent series expansion of f (z) by
1 ´
evaluating the integral f (z) dz .
2πi γ
◦ But since (z − a)d f (z) has the same Laurent series as f (z) except with terms shifted forward by d, we
d
can simply replace f (z) by (z − a) f (z) to obtain the −d − 1th coecient in the Laurent expansion of
f (z).
◦ In particular, for d = −1, we see that the coecient of (z − a)0 , which is to say, the constant term, is
1 ´ f (z)
given by dz : but this constant term is simply f (a).
2πi γ z − a
◦ Therefore, we see that for any function f (z) with a convergent Laurent series expansion centered at
1 ´ f (z)
z = a, we have the formula f (a) = dz , where γ is any contour winding once counterclockwise
2πi γ z − a
around z = a (for instance, any circle of small positive radius centered at z = a).

• Our goal now is to establish this formula, which is known as Cauchy's integral formula, for an arbitrary
holomorphic function.

• Theorem (Cauchy's Integral Formula, First Version): Suppose f is holomorphic on a simply connected region
and let γ be the counterclockwise boundary of the region. Then for any z0 in the interior of the region, we
1 ´ f (z)
have f (z0 ) = dz .
2πi γ z − z0
f (z)
◦ Proof: Since f (z) is holomorphic on the region, is holomorphic everywhere inside the region
z − z0
except at z = z0 .
◦ Since z0 is in the interior of the region, it is some positive distance away from the region's boundary. By
our results on deformation of contours, we may deform the contour γ to be a counterclockwise circle γr
of some suciently small radius r>0 centered at z0 without changing the value of the integral.
´ 1 1 ´ f (z0 )
◦ From our previous calculations, we know that
γr
dz = 2πi, so rearranging yields dz =
z − z0 2πi γr z − z0
f (z0 ).
´ f (z) − f (z0 )
◦ Therefore to establish the desired formula, it suces to show that
γr
dz = 0.
z − z0
f (z) − f (z0 )
◦ Since f is holomorphic, if we dene g(z) = for z 6= z0 with g(z0 ) = f 0 (z0 ), then g is
z − z0
continuous at z = z0 (and elsewhere).

◦ Now let Mr be the maximum value of |f (z) − f (z0 )|

γr (which exists since f is a continuous function
on
f (z) − f (z0 ) |f (z) − f (z0 )| Mr
and γr is a closed curve), so that for z ∈ γr we have = ≤ .
z − z0 |z − z0 | r
◦ Additionally, since f is continuous at z = z0 , we have limr→0 Mr = 0.

´ f (z) − f (z0 ) Mr
◦ Then we have the simple estimate
γr
dz ≤ 2πr · = 2πMr , which goes to zero as r → 0.
z − z0 r
´ f (z) − f (z0 )
◦ Therefore we have limr→0+
γr
dz = 0. But by our results on deformation of contours, the
z − z0
value of the integral is the same on each γr , so the value must be zero for all γr .
´ f (z) − f (z0 ) ´ f (z) − f (z0 ) 1 ´ f (z) 1 ´ f (z0 )
◦ Thus we have
γ
dz = γr dz = 0 and so dz = dz =
z − z0 z − z0 2πi γ z − z0 2πi γr z − z0
f (z0 ), as claimed.

• We will remark that by essentially the same proof, we can generalize Cauchy's integral formula to arbitrary
closed contours.

19
• Corollary (Cauchy's Integral Formula, Second Version): Suppose f is holomorphic on a simply connected
´ f (z)
region R, let γ be any closed contour in R, and let z0 be any point in the interior of R. Then
γ
dz =
z − z0
2πi · Wγ (z0 ) · f (z0 ).

◦ Proof: Using the same estimation argument as in the proof of the rst version of Cauchy's inte-
´ f (z) ´ f (z0 )
gral formula, we see that
γ
dz = γ dz . But the second integral is simply f (z0 ) times
z − z0 z − z0
´ 1
γ
dz = 2πi · Wγ (z0 ) by our results on winding numbers.
z − z0
• Although Cauchy's integral formula may seem to lack great utility at rst glance, as we will see, it is really
quite a powerful statement.

◦ Intuitively, the formula is an averaging result, which we can see very explicitly in the situation where γ
is the counterclockwise boundary of the circle of radius r centered at z = z0 .
◦ Using the parametrization γ(t) = z0 + reit for 0 ≤ t ≤ 2π , which has γ 0 (t) = ireit , we obtain
1 ´ f (z) 1 ´ 2π f (z0 + reit ) 1 ´ 2π
dz = ireit dt = f (z0 + reit ) dt.
2πi γ z − z0 2πi 0 reit 2π 0
◦ This last integral is the average value of f (z) on the circle |z − z0 | = r, in the usual sense of the average
value of a function being its integral on an interval divided the length of the interval.

◦ Thus, in short, Cauchy's integral formula says that if f is holomorphic, then the average value of f on
any circle is simply equal to the value of f at the center of the circle.

◦ More generally, and much more powerfully, it says that the values of a holomorphic function f on the
boundary of a simply connected region completely determine the values of f on the interior of the region.

◦ This is quite a special property of holomorphic functions, since for general nonholomorphic functions,
it is not possible to reconstruct their values on the interior of a region using only their values on the
boundary.
2
◦ For example, the nonholomorphic function f (z) = 1 − zz = 1 − |z| is identically zero on the circle |z| = 1
yet is nonzero everywhere inside the circle. So since the function 2f (z) has the same values on the circle
|z| = 1 yet dierent values everywhere in the interior, it is not possible to reconstruct f (z) for |z| < 1
solely from its values on |z| = 1, since any such procedure cannot distinguish between f (z) and 2f (z).

• The most immediate application of Cauchy's integral formula is to evaluate contour integrals with less eort:

´ ez ´ ez
• Example: Find
γ z
dz and
γ
dz where γ is the counterclockwise boundary of the square with vertices
z−1
±2 ± 2i.
´ f (z)
◦ Note that each integral is of the form
γ
dz with f (z) = ez and γ the counterclockwise boundary
z − z0
of a simply connected region.

◦ Thus by Cauchy's integral theorem, the rst integral equals 2πi · f (0) = 2πi , while the second integral

equals 2πi · f (1) = 2πei .

◦ Note that we could also evaluate these integrals by deforming the contour γ to an appropriate circle and
ez 1 ez e
then using the Laurent series expansions = z −1 + 1 + z + · · · and = e(z − 1)−1 + e + z + · · · ,
z 2! z−1 2!
which again yield the two values 2πi and 2πi · e.

´ ez
• Example: Find
γ
dz where γ is the counterclockwise boundary of the unit circle.
z(z − 2)

ez
◦ Note that is holomorphic for all z 6= 0, 2, so since z = 2 is not inside the unit circle, the integral
z(z − 2)
´ f (z) ez
is of the form
γ z−z
dz with f (z) = , z0 = 0, and γ the counterclockwise boundary of a simply
0 z
connected region on which f (z) is holomorphic.

20
 ◦ Thus by Cauchy's integral theorem, the integral equals 2πi · f (0) = −πi .

´ zez
• Example: Find
γ
dz where γ is the curve traversing the unit circle twice clockwise.
4z + i
zez ´ 1 zez
◦ Note that is holomorphic for all z 6= −i/4, so if we rewrite the integral as ·
γ 4 z + i/4
dz , it is
4z + i
´ f (z)
now in the form
γ
dz with f (z) = zez , z0 = −i/4, and γ a closed contour with winding number
z − z0
−2 around z0 .
◦ Thus by Cauchy's integral theorem, the integral equals 2πi · f (−i/4) · (−2) = −πe−i/4 .

´ sin z + ez
• Example: Find
γ z(z − 1)
dz where γ is the closed curve plotted below:

sin z + ez sin z + ez sin z + ez

◦ First, we can use partial fraction decomposition to write = − , so we may
z(z − 1) z−1 z
´ sin z + ez ´ sin z + ez
equivalently evaluate
γ
dz and γ dz separately.
z z−1
´ f (z)
◦ Each integral is of the form
γ z−z
dz where f (z) = sin z + ez is holomorphic everywhere and γ does
0
not pass through z0 .
◦ From the plot we can see that γ has winding number 3 around z = 0 and winding number 2 around
z = 1.
´ sin z + ez ´ sin z + ez
◦ Therefore, we have
γ
dz = 2πi · Wγ (0) · f (0) = 6πi and
γ
dz = 2πi · Wγ (1) · f (1) =
z z−1
4πi(sin 1 + e).
´ sin z + ez ´ sin z + ez ´ sin z + ez
◦ Hence we see γ dz = γ dz − γ dz = 4πi(sin 1 + e) − 6πi .
z(z − 1) z−1 z

3.2.6 Higher Derivatives and Series Expansions of Holomorphic Functions

• We motivated Cauchy's integral formula by explaining how, if γ winds once counterclockwise around z0 , the
1 ´ f (z)
integral dz calculates the constant term in a Laurent series expansion for f (z).
2πi γ z − z0
◦ By changing the power of z − z0 , we can compute other coecients in the Laurent expansion, and thus
obtain values of the various derivatives of f (z) z = z0 .
at

P∞ f (z) P∞
◦ Explicitly, if f (z) = n=0 an (z − z0 )n , then
d+1
= n=−d an (z − z0 )n−d−1 whose coecient of
(z − z0 )
(z − z0 )−1 is ad .
1 ´ f (z) f (d) (z0 )
◦ Therefore, integrating around γ yields
γ d+1
dz = ad = .
2πi (z − z0 ) d!

21
 d! ´ f (z)
◦ We therefore obtain the formula f (d) (z0 ) = dz , valid whenever f (z) is analytic at
2πi γ (z − z0 )d+1
z = z0 and for γ a circle of suciently small radius to lie inside the disc of convergence for the power
series of f at z = z0 .
◦ We would now like to prove that this formula holds for any holomorphic function f (z). In fact, we can
prove a stronger statement:

• Theorem (Dierentiation Via Integration): Suppose R is an open region.

1. Let γ be a closed contour in R and let f be a continuous function on γ . Then for all z0 not lying on γ ,
´ f (ζ) ´ f (ζ)
the function g(z0 ) = γ
dζ is analytic on R\γ , and furthermore g (n) (z0 ) = n! γ dζ for
ζ − z0 (ζ − z0 )n+1
each n ≥ 0.
◦ Proof: Let z0 ∈ R\γ . Then as in the proof of Cauchy's integral formula, since z0 is in the interior
of the region, it is some positive distance r1 z0 does not
away from the region's boundary, and since
lie on the closed contour γ γ.
it is some positive distance r2 away from

◦ We show that f has a power series expansion on the disc of radius r = min(r1 , r2 ) centered at z0 .
1 1 1
◦ So let 0 < s < r and observe the geometric series expansion = · =
ζ −z ζ − z0 1 − − z0 z
ζ − z0
P∞ 1 n
n=0 (z − z0 ) .
(ζ − z0 )n+1
◦ For ζ ∈ γ , this geometric series converges absolutely and uniformly for |z − z0 | ≤ s since its common
z − z0 s s
ratio is ≤ ≤ < 1.
ζ − z0 r2 r
◦ Therefore, because f (ζ) is bounded on γ since it is continuous, we see that the partial sums of
P∞ f (ζ) g(ζ)
n=0 (z − z0 )n converges absolutely and uniformly for |z − z0 | ≤ s to for each
(ζ − z0 )n+1 ζ −z
ζ ∈ γ.
◦ Hence by our results on uniform convergence and integrals, we may switch the order of the sum and
integral to see

ˆ ˆ X
∞
f (ζ) f (ζ)
f (z) = dζ = n+1
(z − z0 )n dζ
γ ζ −z γ n=0 (ζ − z0 )
∞ ˆ
X f (ζ)
= n+1
(z − z0 )n dζ
n=0 γ (ζ − z0 )
∞ ˆ 
X f (ζ)
= n+1
dζ (z − z0 )n
n=0 γ (ζ − z0 )

P∞ ´ f (ζ)
which is of the form g(z) = n=0 an (z − z0 )n for an = γ
dζ . This is the desired power
(ζ − z0 )n+1
series expansion for g and so g is analytic as claimed.

◦ The values of the derivatives follow immediately from the formula for the terms an and the fact that
(n) an
g (z0 ) = .
n!
2. The function f (z) is holomorphic on R if and only if it is analytic on R.
◦ Proof: We have previously shown analytic functions are holomorphic.

◦ Conversely, supposef (z) is holomorphic on R and let z0 ∈ R.

◦ Then by (1), if we take γ to be a counterclockwise circle of radius r > 0 inside R centered at z0 ,
´ f (ζ)
then the function
γ ζ −z
dζ is analytic on R\γ .
´ f (ζ)
◦ But by Cauchy's integral formula, γ dζ = 2πi f (z), so in particular f is analytic at z0 as
ζ −z
required.

22
 3. Suppose R is simply connected with counterclockwise boundary γ and f
R. Then f is is holomorphic on

(n) n! ´ f (z)
innitely dierentiable on R and for any z0 in the interior of R we have f (z0 ) = dz
2πi γ (z − z0 )n+1
for each n ≥ 0.

◦ Proof: By (2), if f is holomorphic then f is analytic, and analytic functions are innitely dierentiable
as we have previously shown.
1
◦ For the formulas observe as in (2) that f = g via the Cauchy integral formula, where g is the
2πi
function from (1), and use the formulas for the derivatives in (1).
P∞
4. Suppose that f is holomorphic on the closed disc |z − z0 | ≤ r. Then the power series f (z) = n=0 an (z −
z0 )n for f centered at z = z0 has radius of convergence at least r.
1 ´ f (z)
◦ Proof: By (2) we know that f is analytic at z0 and (3) gives an = dz where we
2πi γ (z − z0 )n+1
may take γ |z − z0 | = r.
to be the counterclockwise circle

◦ Since f γ , say with |f (z)| ≤ M .

is continuous it is bounded on
f (z) |f (z)| M
◦ Then for |z − z0 | = r we have n+1
= n+1 ≤ n+1 , so since γ has length 2πr we obtain
(z − z0 ) r r
1 ´ f (z) 1 M M
the estimate |an | = dz ≤ · 2πr · n+1 = n .
2πi γ (z − z0 )n+1 2π r r
1/n
◦ Thus we see lim supn→∞ |an | ≤ 1/r and so the radius of convergence of the power series is at
least r, as claimed.
´
5. (Morera's Theorem) If f is continuous on the region R and
γ
f (z) dz = 0 for all closed contours γ in R,
then f is holomorphic on R.
◦ Recall that we previously established the converse of this theorem (if f is holomorphic then its
integral is zero on all closed contours).
´
◦ f is continuous and γ f (z) dz = 0
Proof: By our earlier result on existence of antiderivatives, since
0
on all closed contours, f possesses an antiderivative F on R, with F holomorphic and F (z) = f (z).

◦ But then by (3) since F is holomorphic it is innitely dierentiable. In particular, F is twice

0
dierentiable, meaning that F = f is dierentiable, which is to say that f is holomorphic.

• In addition to the rather unexpected fact that we can compute the derivative of a holomorphic function via
integration, the fact that every holomorphic function is innitely dierentiable is also quite unexpected.

◦ This state of aairs stands in stark contrast to the situation with real-valued functions, since for example
the function f (x) = xn+2/3 is dierentiable n times at x = 0 but not n + 1 times: the (n + 1)st derivative
tends to −∞ as x → 0− and to +∞ as x → 0+)

• In the next chapter we will discuss many additional applications of Cauchy's integral formula that expand on
the results we have obtained so far.

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

Copyright notice: This material is copyright Evan Dummit, 2015-2022. You may not reproduce or distribute this
material without my express permission.

23