Cauchy's Theorem
Cauchy's Theorem
Cauchys Theorem
The following result is one of the cornerstones of the calculus of complex variables:
Theorem (Cauchys Theorem): If a complex function f (z) is analytic in a simply connected region
D of the complex plane, then for any closed curve C lying entirely inside D,
I
f (z) dz = 0.
C
C
D
A corollary of Cauchys theorem is that we can use contour deformation to highlight when
contour integrals are path independent.
C1
ZB
C2
ZA
C3
If f (z) is analytic in the everywhere in the shaded region (i.e. D), then we can apply Cauchys
theorem to the closed contour constructed from going out along C2 and back along C1 , giving,
I
Z
Z
f (z) dz = 0, or
f (z) dz =
f (z) dz.
C2 C1
C2
C1
If the point marked x is a singularity, then we cannot make the same argument for C3 , as f (z)
is not analytic in the required region. Generally
Z
Z
f (z) dz 6=
f (z) dz.
C3
C1
In summary:
R Contours can be deformed across regions where f (z) is analytic, without changing
the value of C f (z) dz, but not across singularities, branch points or branch cuts.
Proof: The following proof of Cauchys theorem requires the additional assumption that f 0 (z) is
continuous. (The additional condition on f 0 (z) is not necessary, but the required proof is then much
more technical. The more general result is often referred to as the Cauchy-Goursat theorem, see e.g.
Ablowitz and Fokas, Complex Variables, pg. 105.)
The proof requires the use of Greens theorem in the plane (GTP) discussed in the following handout.
Expanding z = x + iy and f (z) = u(x, y) + iv(x, y) we have
I
I
(u + iv)(dx + idy)
f (z) dz =
C
C
I
I
v dx + u dy
u dx v dy + i
=
C
C
ZZ
ZZ
v
u v
u
=
dS + i
dS (applying GTP to each real path integral)
x y
x y
S
Z ZS
=
0 dS
(f (z) analytic in S = Cauchy-Riemann eqns. ux = vy , uy = vx )
S
= 0.
Note that for the first integral we are applying GTP (as written on the handout) with P = u
and Q = v. For the second integral P = v and Q = u.
Note that S refers to the two-dimensional region inside the closed contour C. The integrals are
two-dimensional surface integrals over S.
( F) n dS =
C
for any continuously differentiable three-dimensional vector field F(r). Note that n is the normal to
the surface, and is related to the direction of integration around the closed curve C by the right-hand
rule. If we choose F = P (x, y)i + Q(x, y)j then F = (Qx Py )k. If we next choose the surface S
to lie entirely in the (x, y) plane, so that its normal n = k, Stokes theorem reduces to
ZZ
I
Q P
P dx + Q dy.
dS =
x
y
S
C
This is Greens theorem in the plane. The closed curve C is the boundary of the area S, and we
integrate anti-clockwise around it.
y
d
x=g1(y)
Cl
x=g2(y)
Cr
c
x
y
y=f2(x)
C+
C
x
y=f1(x)
a
x
b
Figure illustrating the contour C and region S used in Greens theorem in the plane (left), and the
two different ways the integral is executed during the proof (right panels).
Note: There is a more general (and difficult) proof of Cauchys theorem by Goursat that does not require the
derivative f 0 (z) to be continuous on D.
Proof: Here we aim to prove the result for this special case where S is a continuous region as
illustrated2 .
The proof proceeds by considering the integrals of Qx and Py separately. In the first case, the surface
integral is partially evaluated by performing an x-integral (top right panel in figure), whereas in the
second case it is a y-integral (bottom right panel). It is assumed that the region S is sufficiently simple
that the curve C can be twice be split into two parts, which in each case are described by single-valued
functions ({Cl : x = g1 (y)} and {Cr : x = g2 (y)} in the first instance, and {C : y = f1 (x)} and
{C+ : y = f2 (x)} in the second).
Then we can write
ZZ
ZZ
ZZ
Q P
Q
P
dS =
dS
dS
x
y
S
S x
S y
!
!
Z d Z g2 (y)
Z b Z f2 (x)
Q
P
=
dx dy
dy dx
c
g1 (y) x
a
f1 (x) y
Z d
Z d
Z b
Z b
=
Q(y, g2 (y)) dy
Q(y, g1 (y)) dy
P (x, f2 (x)) dx +
P (x, f1 (x)) dx
c
c
a
a
Z
Z
Z
Z
=
Q dy
Q dy
P dx +
P dx
Cr
Cl
C+
C
I
I
Z
Z
Z
Z
=
P dx + Q dy,
since
=
+
=
.
C
C+
Cr
Cl
Notice that the expressions on the third line are recognisable as parameterisations of integrals along
Cl , Cr , C+ and C respectively, with y as the parameter in the first two cases and xH in the second.
This permits the step to the fourth line. The final line is explained by recalling that C requires the
contour to be traversed in an anti-clockwise direction.
The proof can be extended to more complicated shapes using surgery techniques involving, for example, cross-cuts
(see later in course).
H
C
f (z)dz = 0
f (z)
,
zw
clearly has a singularity at z = w, marked as an x on the diagram below. However g(z) is analytic
away from w, so we can use Cauchys theorem to deform contours of integration provided we do not
cross z = w.
First, divide up the contour C into two parts Cl (left) and Cr (right) using cross-cuts as illustrated.
C1 C1
Cl
Cl
Cr
Cr
l
C2
C2
Step 1
Clearly g(z) is analytic everywhere inside the closed contour constructed from Cr C1 Cr C2
(where Cr is a semi-circle of radius centred on z = w) as this contour does not contain z = w. Hence
we can deform Cr
Z
Z
Z
Z
=
Cr
Z
=
Cl
+
C2
Z
+
C1
Cr
C1
Z
+
Cl
.
C2
Cl
Cr
C1
Cr
C2
C1
Z
=
+
Cl
C2
g(z) dz,
+
Cl
Cr
C
where we have used the fact that the integrals on the cross-cuts cancel, i.e.,
Z
Z
Z
Z
+
=
+
= 0.
C1
C1
C2
C2
In summary, we have shown that the integral around the closed curve C is equal to the integral around
a small circle, radius , centred on z = w, as illustrated in the diagram.
Step 2: Step 1 holds for any value of the radius of the small circle. Hence we have the freedom to
take the limit 0, and evaluate the integral around the closed contour C in this limit.
To evaluate the integral, we can parameterise C as follows
n
o
C : z() = w + ei , < . .
It is easy to verify that the above parameterisation satisfies |z w| = , i.e. the equation of a circle
radius and centre w, and as varies from to we traverse this circle.
Using this parameterisation we have (substituting g(z) = f (z)/(z w) in the step 1 result)
I
I
f (z)
f (z)
dz = lim
dz
0
z
w
z
w
C
C
Z
f (w + ei )
= lim
iei d
(using dz = iei d)
0
ei
Z
= lim i
f (w + ei ) d
0
Z
= i
f (w) d
(taking the limit 0)
= 2if (w).
Cauchys integral formula is obtained by dividing through by 2i.
Example: Evaluate the integral
I
C
ez
dz,
z1
cos z
dz,
z(z 2 + 9)
where C is a square centred on the origin with a side length of 4, using Cauchys integral formula.
[Answer: 2i/9.]
Remarks:
Note that, if w lies outside of C,
I
C
f (z)
dz = 0,
zw
because Cauchys theorem can be applied (since f (z)/(z w) is then analytic inside C).
Cauchys integral formula reveals something remarkable. If you know f (z) at every point on a
closed contour C, and that f (z) is analytic in a region enclosing C, we can use Cauchys formula
for derivatives to write down f (z) for any z inside C, i.e. we know f everywhere inside C just
from its value on the boundary!
Cross-cuts can be used to extend the validity of Cauchys integral formula to other situations.
Consider the integral
I
f (z)
dz,
C (z w1 )(z w2 )
with w1 , w2 both inside C and f (z) satifying the analyticity conditions for the integral formula.
We cannot use Cauchys integral formula as stated above, because the integrand has two singularities inside C at z = w1 and z = w2 . However, suppose we use cross-cuts to divide the integral
into two as illustrated
C1
C2
w2
w1
Inside the closed contour to the left, C1 , the function g1 (z) = f (z)/(zw2 ) is analytic everywhere.
We can apply Cauchys integral formula to C1 giving
I
f (z)
2if (w1 )
dz = 2ig1 (w1 ) =
.
(z
w
)(z
w
)
w1 w2
1
2
C1
Similarly, g2 (z) = f (z)/(z w1 ) is analytic everywhere inside the closed contour C2 , giving
I
f (z)
2if (w2 )
dz = 2ig2 (w2 ) =
.
w1 w2
C2 (z w1 )(z w2 )
Since the integrals along the cross-cuts cancel, we have
I
I
I
f (w1 ) f (w2 )
f (z)
dz =
+
= 2i
.
w1 w2
C1
C2
C (z w1 )(z w2 )
The above procedure can be extended to integrands with an arbitrary number of simple singularities inside C.
(n)
dn f
1 dn
(w) =
(w)
=
dwn
2i dwn
f (z)
1
dz =
zw
2i
dn
f (z) n
dw
1
zw
dz.
()
1
zw
=
1
.
(z w)2
and from there it is easy to show by induction (assume true for n, prove for n + 1) that
dn
1
n!
=
.
dwn z w
(z w)n+1
Inserting the above expression into () gives the formula for derivatives.
Remarks:
The above analysis has revealed something remarkable; if f (z) is analytic inside C then all of its
derivatives exist, and can be evaluated everywhere inside C from knowledge of f (z) on C alone.
(This is very different from real calculus, for which a function may be differentiable only a finite
number of times).
The integral formula for derivatives leads directly to a proof of the formula for Taylor series
expansions (see following handout).
Example: Evaluate the integral
I
C
cos z sin z
dz,
(z + i)4
where C is any simple closed contour enclosing z = i, using Cauchys integral formula for derivatives.
[Answer:
3 (sinh(1)
+ i cosh(1)).]
Taking the differential inside the integral is a legitimate operation here, but really requires more technical justification.
See Ablowitz and Fokas, Complex variables, pg. 91, for a more rigorous treatment.
(x a)2 00
(x a)n (n)
f (a) + ... +
f (a) + ...
2!
n!
Complex variable methods provide a powerful technique for deriving and formally justifying Taylor
series. A statement of Taylors theorem is as follows:
Statement: If a single-valued function f (z) is analytic everywhere in a circular region |z z0 | R
of the complex plane then f (z) can be expanded as
f (z) =
bn (z z0 )n ,
where
bn =
n=0
f (n) (z0 )
,
n!
and the series will converge uniformly within the region (|z z0 | < R).
Im z
C
R
z0
Re z
1
2i
I
C
f (w)
1
dw =
wz
2i
I
C
f (w)
1
dw =
(w z0 ) (z z0 )
2i
I
C
f (w)
w z0
1
(z z0 )
(w z0 )
1
dw.
Now we know that for z anywhere inside the circle |z z0 | < |w z0 | = R, so we can use the geometric
series formula to write
I
1
f (w) X z z0 n
f (z) =
dw.
2i C w z0
w z0
n=0
gives
I
X
X
1
f (w)
f (n) (z0 )
n
f (z) =
dw
(z
z
)
=
(z z0 )n ,
0
2i C (w z0 )(n+1)
n!
n=0
n=0
using Cauchys integral formula for derivatives.
Although this operation is legitimate for convergent infinite sums it really requires more justification. There is in
fact an explicit theorem proving the required result.... it demonstrates that the integral of a function that is the limit
of a sequence of functions is equal to the limit of the sequence of integrals of the functions in tbe sequence (see e.g.
Ablowitz and Fokas, Complex Variables, pg. 111.)
X
zn
ez =
n=0
sin z =
(1)n
n=0
cos z =
n!
z2 z3
+
+ ...
2!
3!
z 2n+1
(2n + 1)!
=z
z3 z5
+
...
3!
5!
z 2n
(2n)!
=1
z2 z4
+
...
2!
4!
(1)n
n=0
=1+z+
[Answer:
1
z2
about z = 1.
n=0
Laurent Series: Laurent series (discussed on the following handout) extend the idea of Taylor series
to allow functions f (z) to be expanded about points z0 where there is singularity. The Laurent series
takes the form
f (z) =
A2
A1
An (z z0 )n = ... +
+
+A0 + A1 (z z0 ) + A2 (z z0 )2 + ....
2
(z
z
)
(z
z
)
0
0
n=
|
{z
}
Principal Part
where the n < 0 terms in the sum are known as the principal part, as shown.
Note: A function may have more than one Laurent series, each of which is valid in a different annular
region Ri < |z z0 | < Ri+1 in which f (z) is analytic (i = 1, 2, 3, .. etc.).
In practice we usually find Laurent series using similar techniques to the Taylor series expansions
above.
Example: Find the Laurent series expansions of
(a)
ez
,
z4
(b) ze1/z ,
(c)
1
,
z(z 1)
about z = 0.
[Answers: (a)
X
z n4
n=0
(c)
X
n=0
n1
n!
(b)
X
n=0
1
, both valid everywhere in C except z = 0.
n!z n1
X
n=0
1
z n+2
f (z) =
An (z z0 )n ,
Laurent Series
n=
where
1
An =
2i
I
C
f (w)
dw,
(w z0 )n+1
< n < .
and C is any simple closed contour enclosing the inner boundary of the annular region |z z0 | = R1 .
Im z
C2
C1
Cin
C out
R1
z0
R2
Re z
z
)
+
f
(w)(w
z
)
dw
0
0
2i C2 (w z0 )(n+1)
2i C1
(z z0 )n+1
n=0
n=0
I
X
1
f (w)
dw (z z0 )n .
=
(n+1)
2i
C (w z0 )
n=
The final line results from the fact that by Cauchys theorem we can deform both C1 and C2 back onto
any simple contour C that encircles the annulus once, proving our result. The terms in the second
(C1 ) sum can then be written as the n < 0 terms in the final sum.
1
.
sin z1
The most important property of a singularity is its residue. Complex methods are sometimes referred
to as residue calculus, which should give you an idea of how important the following definition is!
Definition: The residue of an isolated singularity of an analytic function f (z) at z = z0 is the
coefficient (A1 ) of the (z z0 )1 term in the Laurent series valid in the immediate vicinity of the
singularity.
There are (at least) two alternative methods for finding residues:
Method 1: Use the methods above to find the Laurent series for f (z) about z0 , and simply read off
the coefficient of (z z0 )1 (i.e. A1 ).
Method 2: To find the residue at a pole of order m, use the formula
"
#
d(m1)
1
m
(z z0 ) f (z)
A1 = Res{f (z); z = z0 } =
(m 1)! dz (m1)
Note that for m = 1 (a simple pole) the formula reduces to
h
i
A1 = Res{f (z); z = z0 } = (z z0 )f (z)
z=z0
Pole of order m.
z=z0
Proof of Method 2: The residue formula for A1 can be verified by inserting the Laurent series for
a pole of order m
X
f (z) =
An (z z0 )n ,
n=m
"
#
1
d(m1)
m
(z z0 ) f (z)
(m 1)! dz (m1)
z=z0
#
"
(m1)
X
1
d
An (z z0 )n+m
(m 1)! dz (m1) n=m
z=z0
"
#
X
1
(n + m)!
An
(z z0 )n+1
(m 1)!
(n + 1)!
n=1
z=z0
1
A1 (m 1)!
(m 1)!
(Evaluating at z = z0 ; n = 1 gives only non-zero term.)
= A1 .
Examples: Use both of the two methods above to
(a) Find the residue of f (z) = (z/(z 1))2 at z = 1,
(b) Find the residue of f (z) = 1/(ez 1)2 at z = 0,
(c) Find the residue of f (z) = e2z tan z at z = 3/2,
[Answers: (a) 2, (b) 1, (c, harder!) e3 .]
The residue theorem can be viewed as an extension of Cauchys theorem to the case where C
includes a finite number of singularities (recall that for Cauchys theorem to hold f (z) must be
analytic everywhere inside C).
The residue theorem provides a remarkable means of evaluating many contour integrals that we
dont know how to do. Instead of trying to integrate f (z) along C, all we need to do is find all
of the singularities zj inside C and calculate the residues of f (z) there.
Proof: The proof of the residue theorem follows from several results we have seen already.
Step 1a: Note that the contour C can be divided up as shown in Fig. 8, using cross-cuts along which
the integrals will cancel, into N closed contours C1 , ..., CN , each containing just one of the singularities,
so that
I
I
I
I
=
+
+... +
.
C
C1
C2
CN
Step 1b: Each of the contours C1 , C2 ,..., CN can then be further deformed, using cross-cuts exactly
as in step 1 of the proof of Cauchys integral formula (see earlier), to small contours C1 , C2 ,..., CN
that consist of circles radius around their respective singularities, as shown in Fig. 8. As seen earlier,
for each of the N closed contours,
I
I
=
.
Cj
Cj
Step 2: The integrals around each of the Cj contours can now be evaluated by exploiting the fact
that f (z) will have a Laurent expansion valid in the region immediately surrounding each singularity.
(Note that this is where the restriction to a finite number of singularities in C becomes important in
the proof. We would not be able to use the Laurent series if one of the singularities was not isolated.)
Then for any Cj we have
I
I
f (z) dz =
Cj
Cj
X
n=
An (zzj ) dz =
X
n=
I
An
Cj
Cauchys residue theorem follows from adding up the contribution from each contour Cj .
(Note that we have used the earlier result that for C a circle,
I
0
for n 6= 1,
n
z dz =
2i
for
n = 1.
C
It is easily verified that this result is unchanged if the circle is centered on z = zj instead of z = 0. We
have also assumed that the sum and the integral can be interchanged; see the comment on the Taylor
series handout.)
C2
C1
C3
Step 1a
C1
C2
Step 1b
C3
Figure 1: Illustrating the two steps of deformation of the contour C in the proof of the residue theorem.
First the single contour is divided using cross-cuts into N contours Cj (j = 1, ..., N ) each of which
contains a single singularity. Second, the contours Cj are deformed to circles Cj of radius centered
on each singularity.
Example: Evaluate
I
C
ez
dz,
z(z 2)
for
C = C1 the circle |z 2| = 1.
C = C2 the circle |z| = 4.
C = C3 the circle |z 2 2i| = 1.
[Answers: e2 i (C1 ), (e2 1)i (C2 ),
0 (C3 ). ]
Z
a
1
dt
1 + t2
is evaluated using the substitution t = tan (giving tan1 b tan1 a). Complex substitutions are
also possible, and can lead to the evaluation of otherwise difficult / impossible integrals by the residue
theorem.
Consider real trigonometric integrals of the form
Z 2
F (cos , sin ) d
0
where F is a real-valued function of two variables. A complex subsitution can be made using
z = ei .
As varies from 0 to 2, z will traverse the unit circle in the complex plane. In other words, we have
mapped the contour of integration from the interval [0, 2] on the real line to the unit circle in the
complex plane. Noting that we have
dz
dz
= iei , we can write d =
,
d
iz
to make the substitution. Using the complex definitions of cos and sin in terms of z, the substitution
gives
Z 2
I
1
1
1
1
1
F (cos , sin ) d =
g(z) dz where g(z) =
F
z+
,
z
,
iz
2
z
2i
z
0
C
and where C is the unit circle. Provided g(z) satisfies the conditions of the residue theorem (which
will ultimately depend on F ), we can evaluate the integral by summing up the residues of g(z) inside
the unit circle.
The method will usually work (in principle) when F is a rational polynomial function
e.g.
F (x, y) =
x2 y + x + y + 1
.
xy+4
The method will typically also work if F is additionally a function of cos n, sin n (n integer),
i.e. for
F = F (cos , sin , cos 2, sin 2, cos 3, sin 3, ...),
as we can use cos n = 21 (z n + z n ) etc.
Integrals of the form
Z
F (cos n) d
0
can be evaluated by noting that cos (n) = cos n and therefore that
Z
Z
1
F (cos n) d.
F (cos n) d =
2
0
The latter integral maps onto the unit circle as above. Other symmetries can sometimes be used
to evaluate other trigonmetric integrals on [0, ].
Examples: Use residue calculus to evaluate
Z 2
cos 2
d,
5 4 cos
0
[Answers:
6,
cos
d.
2 + 3 cos
- 3 .]
(These can also be evaluated by the (real) half-angle substitution t = tan 2 . I hope you agree that the
residue method is faster!)
I=
f (x) dx.
Integrals of this type can sometimes be evaluated by residue calculus as follows. First consider the
semi-circular contour illustrated.
iR
CR
Cl
R
R
It should be clear that in the limit R we will have Cl I. A necessary condition for the residue
R
theorem to be of use is that the integral on the semi-circle portion CR 0 in the limit R . Then
the residue theorem gives
Z
Z
I
X
I + 0 = lim
f (z) dz + lim
f (z) dz = lim
f (z) dz = 2i
Res {f (z); z = zj },
R C
l
R CR
R C
where the sum will be over the residues at all of the poles zj in the upper half-plane.
Under what conditions does
Z
lim
R CR
f (z) dz = 0?
First parameterise the semi-circular integral path CR , as {CR : z = Rei , 0 } and change
variables to giving
Z
Z
lim
f (z) dz = lim
f (Rei ) iRei d
R CR
R 0
f (Re ) iRe
|f (Rei )|R d
0 as R ,
provided that R|f (Rei )| 0 for all 0 . Therefore a necessary condition that the part of the
integral on the semi-circular contour CR disappears is
lim R|f (Rei )| = 0 for all 0 .
()
Example: Find
Z
I=
x2
1
dx,
+ a2
a .]
TheseR can also often be evaluated using the semi-circular contour. The key question is again: when
does CR 0?
Theorem (Jordans Lemma): Consider the function
g(R) = Max[0,] |f (Rei )|.
If the limit of g(R) as R is zero, i.e.
lim g(R) = 0
then
Z
lim
R CR
f (z)eiaz dz = 0.
R CR
R C
= 2i
Res {f (z)eiaz ; z = zj },
where the sum will again be over the residues at all of the poles zj in the upper half-plane.
Example: By expressing I in terms of a Fourier-type integral and using a suitable contour in the
complex plane find
Z
cos kx
I=
dx,
(k > 0).
2 + a2
x
0
[Answer:
eka
2a .
Note that the condition on f (z) for Jordans lemma to hold is a weaker condition than the
condition () required for
Z
lim
f (z) dz = 0, which was
lim R|f (Rei | = 0 for all 0 .
R CR
In other words, Fourier integrals (with k > 0) can be obtained for a wider class of functions f (x)
(using the semi-circular contour) compared to direct integrals of the function itself (note that
these correspond to the Fourier integral with k = 0).
What if k < 0 (or equivalently eikx appears in the integrand with k > 0)? In this case
R we can
use the semi-circular contour closed in the lower half-plane instead in order to force CR 0.
The integral is then equal to (2i times) the sum of the residues in the lower half-plane instead.
Exercise: Check this by working through the proof of Jordans lemma for the k < 0 case.
Example: By combining the results obtained whilst solving the examples above, with a new
calculation for the k < 0 case, show that
Z
eikx
e|k|a
dx
=
.
x2 + a2
a
Jordans lemma gives a sufficient condition for the integral on the semi-circle to vanish. It is possible
that the integral might still vanish for functions that do not satisfy the Jordans lemma condition.)
Theorem (Jordans Lemma): Consider the function
g(R) = Max[0,] |f (Rei )|.
If g(R) converges uniformly to zero as R , i.e. lim g(R) = 0 then
R
Z
lim
R CR
f (z)eiaz dz = 0.
where CR is the semi-circular contour of radius R closed in the upper-half of the complex plane.
Proof: First parameterise the semi-circular integral path CR , as {CR : z = Rei , 0 } and
change variables to giving
Z
Z
i
lim
f (z)eiaz dz = lim
f (Rei ) eiaRe iRei d
R CR
R 0
(since sin 2
on 0 /2),
g(R)
1 eaR 0
as R by assumption.
a
R
The modulus5 of CR is less than or equal to a function that clearly tends to zero as R . Therefore
R
the integral CR must also tend to zero as R .
5
two graphs meet at the endpoints of the interval, and as the graph of sin is convex on [0, /2] it is seen to lie above the
straight line everywhere else on the interval.