Sie sind auf Seite 1von 88

MATH2015 Analysis 2, 2009/2010

Notes c K Houston, D L Salinger, J M Speight and J C Wood 2010

Contact Details (Semester 2)


Lecturer: Oce: E-mail: Phone: Web: Dr J M Speight School of Mathematics, 9.19i speight@maths.leeds.ac.uk 343 5169 http://www.maths.leeds.ac.uk/speight/teaching/math2015

All handouts will be made available on the above webpage.

11

Contour Integration

In this section we dene integration along a contour in the complex plane. This is a fairly abstract process, the meaning of which usually takes a little time to understand. Fortunately, it is easy to do as it has similar properties to Riemann integration of one real variable, and you have extensive experience of that. First though we must dene complex Riemann integration. Complex Riemann integration We start by dening, in the obvious manner, the derivative and integral of a complex valued function of a real variable. 11.1 Denition Let f : S C where S R and let u, v : S R be the real and imaginary parts of f (so f (t) = u(t) + iv (t)). Then if both u and v are dierentiable at t S , we say that f is dierentiable at t and dene f (t) = u (t) + iv (t). Similarly if [a, b] S and both u and v are Riemann integrable on [a, b], then we say that f is complex Riemann integrable (CRI) and dene
b b b

f (t)dt =
a a

u(t)dt + i
a

v (t)dt.

41

11.2 Example Let f : R C such that f (t) = e(1+i)t . Then


f (t)dt =
0 0

et eit dt

=
0

et cos t dt + i
0

et sin t dt

=
0

d t (e cos t) + et sin t dt + i dt

d t (e sin t) et cos t dt dt

et cos t

+ i et sin t

et cos t dt + i
0 0

et sin t dt

= e (1) e0 (1) + i(0 0) i

f (t)dt
0

(1 + i)
0

f (t)dt = (e + 1)

f (t)dt =

e + 1 1+i

Phew! That was pretty hard work. It would be nice if we could avoid decomposing f into real and imaginary parts. Luckily, we can, using the Fundamental Theorem of Calculus. 11.3 Proposition (FTC for complex Riemann integrals) Let g : [a, b] C be continuously dierentiable. Then
b a

g (t)dt = g (b) g (a).

Proof: Let g (t) = u(t) + iv (t), u, v : [a, b] R. Then u , v are continuous by assumption, so the integral of g exists, and
b b b

g (t)dt =
a a

u (t)dt + i
a

v (t)dt = [u(b) u(a)] + i[v (b) v (a)] = g (b) g (a)

by 9.2 (2nd form of the FTC for real Riemann integrals). We can use the FTC, along with a few standard antiderivatives, to make short work of many complex Riemann integrals.
42

11.4 Exercise Use standard properties of the function ez (and the denitions of sin z and cos z ) to show that, for any complex constant c, d ct e = cect dt d (ii) sin ct = c cos ct dt d (iii) cos ct = c sin ct. dt (i) Example 11.2 revisited: Since f (t) = e(1+i)t = g (t) where g (t) =
e(1+i)t , 1+i

e
0

(1+i)t

e(1+i)t dt = 1+i

=
0

1 (e 1). 1+i

Standard tricks of real Riemann integration like integration by parts and substitution likewise work equally well for complex Riemann integrals. 11.5 Exercise Compute the following complex Riemann integrals:

(i)
0 0

teit dt et+ie dt

(ii)

[hint: try the substitution s = et ]

Contours 11.6 Denition A contour (also called a path) is a continuous map : [a, b] C which is piecewise continuously dierentiable, i.e. there exist a = a0 < a1 < a2 < < an = b such that (i) |[aj 1 , aj ] is dierentiable, for all j , (ii) is continuous on [aj 1 , aj ], for all j . (The left and right derivatives of at aj may dier.) We say is closed if (a) = (b). 11.7 Examples (i) Straight line from z to w :

: [0, 1] C, (t) = z + t(w z ). Note (0) = z , (1) = w .


43

(ii) Circle of radius r based at the origin: : [0, 2 ] C, (t) = reit . Note this is a closed contour: (0) = (2 ) = r . (iii) Circle of radius r based at z0 : : [0, 2 ] C, (t) = z0 + reit . This is also a closed contour: (0) = (2 ) = z0 + r . (iv) Circular arc of radius r based at z0 : (t) = z0 + reit , 1 t 2 . (Note 0 t 2 gives the circle above). (v) Draw the image of the contour : [1, /2] C, (t) = t + 1, for 1 t 0, eit for 0 t /2.

Is this contour closed? (vi) Consider the contour : [0, 4 ] C such that (t) = eit . Then the image of is the unit circle centred at zero. The contour goes round the circle twice (anticlockwise). This subtlety will be important later. Tip Given a contour, try to draw its image. 11.8 Error There is often confusion between a contour and its image. A contour is not a set of points in the complex plane, it is a map. Consider the contours (ii) and (vi) above, taking r = 1 in (ii). They have the same image, the unit circle. However, the contours are dierent, one maps from [0, 2 ], the other [0, 4 ]. We do use the notation z later, by which we mean z ([a, b]). Strictly speaking, writing z is incorrect because is not a set.

44

Since contours are complex-valued functions of a real variable, we can dierentiate them, etc, with ease. Contour Integration We now come to probably the most important denition in complex analysis: contour integral. It is central to the second half of this module. If you dont understand this section, then the rest of the course will be a complete mystery to you. 11.9 Denition Let D C be an open set, f : D C be a continuous complex function and : [a, b] D be a contour. Then, the integral of f along is
b

f (z ) dz :=
a

f ( (t)) (t) dt.

11.10 Remarks (i) Note that f ( (t)) and (t) are complex-valued functions of a real variable, and hence so is their product. Furthermore, since both f and are continuous, so is f . Since is also continuous (actually thats a lie, but well justify it shortly), the integrand f ( (t)) (t) is continuous. Recall we integrate this complex valued function of a real variable by splitting it into its real and imaginary parts. But these are both continuous by Corollary 4.5. Thus we know that the integral exists (Theorem 8.6: real continuous functions on closed intervals [a, b] are Riemann integrable). (ii) The contour integral is a single complex number. This number depends on the function f and the contour . It does not depend on the variable z , which should be thought of as a dummy integration variable (just as for real denite integrals). To emphasize this point we will often denote the contour integral as f.

11.11 Example Let (t) = t + it2 for 0 t 2 and f (z ) = z . Then, f ( (t)) = f (t + it2 ) = t + it2 (t) = 1 + 2it
2 2

f =
0

(t + it )(1 + 2it)dt =
0 2 0

(t 2t3 + 3it2 )dt

t2 t4 + it3 2 2

= 6 + 8i

45

11.12 Example Let (t) = 2 + it2 for 0 t 1 and f (z ) = z 2 . Then, f ( (t)) = (t)2 = (2 + it2 )2 ,
1 1

(t) = 2it
1

f =
0

f ( (t)) (t) dt =
0 1 0

(2 + it2 )2 2it dt = 2i
0

(4t + 4it3 t5 ) dt 11 i 3

1 = 2i 2t2 + it4 t6 6

= 2i 2 + i

1 6

= 2 +

This is just the sort of example you need to be able to do with ease. 11.13 Remark Note that, since we demand only that be piecewise continuously dierentiable, (t), and hence the integrand f ( (t)) (t) may actually fail to be continuous, but only at a nite number of points a1 , a2 , . . . , an1 . But then, just as for real Riemann integration, we can subdivide the integration range [a, b] into n subintervals [aj 1 , aj ], j = 1, 2, . . . , n, on each of which the integrand is continuous. Then:
b n aj

f ( (t)) (t) dt =
a j =1 aj 1

f ( (t)) (t) dt

by Proposition 8.11. This may seem like technical pedantry, but actually most of the contours we consider in concrete applications will only be piecewise continuously differentiable, so the subdivision procedure just described will be used often in practice. 11.14 Example Let be the contour dened in Example 11.7(v). Find
/2 0 /2

z 2 dz .

z 2 dz =
1

(t)2 (t) dt =
1

(t)2 (t) dt +
0

(t)2 (t) dt

/2

=
1

(t + 1)2 1 dt +
0

e2it ieit dt
/2 0

(t + 1)3 3

1 + e3it 3 1

1 1 1 + (i) 3 3 3 i 3

46

11.15 Remark Suppose f : D C is a complex function such that f (x) is real for x real, for example, sin x. If we take : [a, b] C given by (t) = t for a t b, then
b b

f (z ) dz =
a

f (t) (t) dt =
a

f (t) dt.

Thus, by taking a contour along the real line, we can see that contour integration includes the theory of real integration as a special case. 11.16 Fundamental Example (KNOW THIS!) Take the function dened by f (z ) = (z w )n where n Z, (so for n < 0 the map is not dened at w ). Let be a circle with centre w and radius r > 0, i.e. (t) = w + reit , 0 t 2 . Then,
2

(z w )n dz = =

0 2

( (t) w )n (t) dt = r e
n int

0 2

w + reit w ei(n+1)t dt

. ireit dt

. ire dt = ir
2 0

it

n+1 0

Thus

Note this well: this innocuous looking calculation will be used to devastating eect later! Summary Given f : [a, b] C we dene f (t) = u(t) + iv (t) and a f (t)dt = b i a v (t)dt, where u, v are the real and imaginary parts of f .
b b a

1 dz = 2i and zw

0 n+1 r ei(n+1)t = n+1 2 i 0 1.dt

= 0,

if n = 1,

= 2i, if n = 1.

(z w )n dz = 0 for n = 1.

u(t)dt +

A contour is a continuous map : [a, b] C which is piecewise continuously dierentiable. Straight line from z to w : (t) = z + t(w z ), 0 t 1. The integral of f along a contour is f=

Circle of radius r based at z0 : (t) = z0 + reit , 0 t 2 .


b

f (z ) dz =
a

f ( (t)) (t) dt.

1 dz = 2i, zw

(t) = w + reit , 0 t 2 . (t) = w + reit , 0 t 2 .

(z w )n dz = 0 for n = 1,

47

12

Properties of Contour Integration


b b b

There are a number of well known properties of ordinary integration: [f (x) + g (x)] dx =
a c a b

f (x) dx +
a c

g (x) dx, , R, a < b < c,

f (x) dx =
a b a b a

f (x) dx +
b a

f (x) dx,

f (x) dx = f (x) dx =

f (x) dx,
b 1 (b)

f ((y ))(y ) dy (change of variables).


1 (a)

All of these have analogues in contour integration. We shall now describe them. In the following the functions will be continuous on some open set D C and the contours will be maps into D . 12.1 Proposition (Linearity) If f, g are continuous on D , a contour in D , , C, then f + g = f + g.

The proof follows directly from the linearity property of Riemann integration. What if I produce a contour by traversing one contour after another? 12.2 Denition If 1 : [a, b] C and 2 : [b, c] C are two contours such that 1 (b) = 2 (b), then their join (or sum) is the contour 1 2 : [a, c] C given by (1 2 ) (t) = 1 (t) a t b, 2 (t) b t c.

Note that the join is clearly continuous and piecewise continuously dierentiable. Note also that the range of the join is the union of the ranges of 1 and 2 , (1 2 )([a, c]) = 1 ([a, b]) 2 ([b, c]), which explains our choice of notation. 12.3 Example Consider Example 11.7(v). The contour : [1, /2] C is given by (t) = t + 1, for 1 t 0, eit for 0 t /2.

]C This is the join 1 2 where 1 : [1, 0] C such that 1 (t) = t+1 and 2 : [0, 2 such that 2 (t) = eit . 12.4 Proposition (Integration along a join) For any continuous f f=
1 2 1

f+
2

f.

48

Again, this follows from applying the denition and using a C-RI property: c b + c. a 12.5 Example Find

b a

1 dz , where is the boundary of the square with corners 1 i, 1 + z i, 1 + i, 1 i, starting at 1 i and going anticlockwise. Clearly, we can think of as the join of four straight line segments:

= 1 2 3 4

1 : [0, 1] C, 2 : [1, 2] C, 3 : [2, 3] C, 4 : [3, 4] C,

1 (t) = 1 i + 2it 2 (t) = 1 + i 2(t 1) 3 (t) = 1 + i 2i(t 2) 3 (t) = 1 i + 2(t 3)

dz = z

1 0

1 (t) dt 1 (t) 1

=
0

2i dt 1 + i(2t 1)

= 2i
0 1

1 i(2t 1) dt 1 + (2t 1)2 [s := 2t 1]


1 1

= i
1 1

1 is ds 1 + s2 ds + 1 + s2

= i
1

s ds 1 + s2 [why?]

= i[tan1 s]1 1 + 0 = i 4 4 =i 2

49

Similarly

i (check!). So, 2 dz = z
4 j

j =1

dz = 2i. z

Notice that this coincides with the value of the integral when is any circle centred on the origin, see the Fundamental Example. The contour I get by traversing backwards is called the reverse contour, rev . 12.6 Denition If : [a, b] C is a contour, then its reverse is the contour rev : [a, b] C dened by (rev )(t) = (a + b t). The point is that we do backwards. Instead of starting at (a) we end there: (rev )(a) = (a + b a) = (b), (rev )(b) = (a + b b) = (a). 12.7 Warning Our choice of notation 1 2 for join and rev for reverse is not standard. Many authors (including my predecessor as lecturer for this module) use 1 + 2 for join and for reverse. You should bear this is mind when consulting textbooks on Complex Analysis, or past exam papers. The obvious disadvantage of their choice is that 1 + 2 and already mean something quite dierent, namely (1 + 2 )(t) = 1 (t) + 2 (t), and ( )(t) = (1) (t). 12.8 Proposition (Integration along the reverse contour) . For any continuous f , we have
rev

f =

f.

Proof: Again we apply denitions and use RI properties, but in this case we also need a simple change of variables: let s = a + b t, so dt = ds. Note that (rev ) (t) = (1) (a + b t). We have,
b

f =
rev a b

f ((rev )(t))(rev ) (t) dt f ( (a + b t))(1) (a + b t) dt


a b

=
a

=
a

f ( (s)) (s) (ds)

=
b

f ( (s)) (s) ds
b

f ( (s)) (s) ds
a

f.

50

12.9 Example Let be the unit circle centred at 0, traversed anticlockwise: : [0, 2 ] C, (t) = eit Then rev : [0, 2 ] C, rev (t) = that is, the unit circle traversed clockwise. By the Fundamental Example and Proposition 12.8,
rev

dz = z

dz = 2i z

The analogue of a change of variables for contour integration is reparametrization. 12.10 Denition Let : [a, b] C be a contour and let : [c, d] [a, b] be a function such that (i) is continuously dierentiable , (ii) (s) > 0 for all s [c, d], (iii) (c) = a, and (d) = b. Then the contour : [c, d] C dened by (s) = ( (s)) is called a reparametrization of . 12.11 Example Let : [0, 2 ] C such that (t) = eit . Let : [0, 1] [0, 2 ] such that (s) = 2s. Then : [0, 1] C such that (s) = e2is . We will think of reparametrized contours as equivalent since we have the following: 12.12 Proposition (Reparametrized contours are equivalent) . If is a reparametrization of , then for all continuous f , f=
e

f.

Proof:

We have,
d

f =
e c d

f ( (s)) (s) ds f ( ( (s)) ( (s)) (s) ds


c b

= =
a

f ( (t)) (t) dt, using t = (s), f.

Thus, it does not matter whether we evaluate the integral for 0 t 2 or (t) = e2it for 0 t 1.
51

1/z dz using (t) = eit

12.13 Remarks (i) If 1 : [a, b] C and 2 : [c, d] C are two contours such that 1 (b) = 2 (c), then we can reparametrize 2 as 2 : [b, d + b c] C given by 2 (s) = 2 (c b + s). Then 1 (b) = 2 (b) so we can form the join 1 2 . Often we abuse notation and write this simply as 1 2 . (ii) If is a simple closed contour (i.e. (a) = (b) and it doesnt intersect itself anywhere else) with no orientation specied, then we assume it is traversed anticlockwise. (iii) If is a reparametrization of , then it has the same range as . The converse is not true. Summary If f, g are continuous on D , a contour in D , , C, then f + g =

f +

g.

If f is continuous on D , its integral along the join 1 2 of two contours in D is f=


1 2 1

f+
2

f.

If : [a, b] C is a contour, then its reverse is the contour rev : [a, b] C, For f continuous, we have
rev

(rev )(t) = (a + b t) f = f.

If is a reparametrization of , then f=
e

f,

52

13

Fundamental Theorem of Calculus for Complex Functions

One of the best theorems in Real Calculus is the Fundamental Theorem of Calculus. We now see this in a contour integration setting. 13.1 Theorem (Fundamental Theorem of Calculus) Let f : D C be a continuous complex function and : [a, b] D be a contour. Suppose there exists a holomorphic function F : D C such that F = f . Then,

f (z ) dz = F ( (b)) F ( (a)).

Proof: Let a = a0 < a1 < < an = b be a dissection of [a, b] such that |[aj 1 , aj ] is continuously dierentiable for all j . Then, f (z ) dz =
b

F (z ) dz F ( (t)) (t) dt
a n aj

= =
j =1 n

F ( (t)) (t) dt
aj 1 aj aj 1

=
j =1 n

(F ) (t) dt

=
j =1

j [(F )(t)]a aj 1 by the usual FTC for RI functions,

= F ( (b)) F ( (a)).

13.2 Example Consider Example 11.12. Then, f (z ) = z 2 , (0) = 2 and (1) = 2 + i. 1 3 z is an antiderivative for f , i.e. F (z ) = f (z ). Then, by the Obviously, F (z ) = 3 FTC, f (z ) dz =

1 ( (1))3 ( (0))3 3 1 3 2 (2 + i)3 = 3 11 = 2 + i. 3

Note that we didnt need to know anything about the contour except its start and end points. How common are antiderivatives for continuous complex functions? Do they always exist? For real continuous functions we know that the Riemann integral
53

F (x) =
a

f (t) dt exists and this is an antiderivative. Fortunately, the analogue

is not true for continuous complex functions, nor even holomorphic ones. Consider this corollary of the FTC and the following example. 13.3 Corollary With the assumptions of the above theorem suppose that is a closed contour. Then f (z ) dz = 0.

Proof:

The denition of a closed contour is that (a) = (b). So, f (z ) dz = F ( (b)) F ( (b)) = 0.

13.4 Example Let f (z ) = 1/z . Then, f is dierentiable on D = C\{0}. Let be the unit circle round the origin, traversed once anti-clockwise. Then, by the Fundamental Example we know 1 dz = 2i. f (z ) dz = z Therefore, if there existed an F : C\{0} C such that F = f , then the corollary would be contradicted. Hence, antiderivatives do not always exist. Property of exponential We can now prove a result we would expect to be true by analogy with real analysis: if a function has zero derivative everywhere, then it is constant. First we need a denition. 13.5 Denition A set D C is called connected if for each pair z, w D , there exists a contour : [a, b] D such that (a) = z and (b) = w . So a set is connected if we can join any pair of points in the set by a curve in the set. 13.6 Examples (i) The open sets D = C and D = C\{0} are both connected.

(ii) The open set D = C\{z : z is real} is not connected. There is no way to construct a contour starting below and nishing above the real line, without crossing that line. 13.7 Theorem Suppose that D is a connected open set, and f : D C is holomorphic, with f (z ) = 0 for all z D . Then, f is constant. Proof: Take any z and w in D . Then, as D is connected, there exists a path : [a, b] D , such that (a) = z and (b) = w . By the FTC f (w ) f (z ) = f ( (b)) f ( (a)) =
54

f =

0 = 0.

Thus f (w ) = f (z ). Since these were general points of D , we deduce that f is constant. This theorem allows us to prove a crucial property of ez we stated but did not prove earlier. 13.8 Theorem For all complex numbers z and w we have ez +w = ez ew . Proof: Let f (z ) = ez ez , where is any complex constant. By the product rule we get f (z ) = ez ez + ez ez = 0 for all z . Thus, by Theorem 13.7, the function is constant. So, for all z and , ez ez = f (z ) = f (0) = e . Let = w + z . Then ew+z = ez ew .

Summary Fundamental Theorem of Calculus: Suppose f : D C is a continuous complex function and there exists F such that F = f . Then, f (z ) dz = F ( (b)) F ( (a)).

Not all functions have an antiderivative. An open set D is called connected if for each pair z, w D , there exists a contour : [a, b] D such that (a) = z and (b) = w .

55

14

The Estimation Lemma

The goal of this section is to show that we can estimate (i.e. bound above) f in terms of the maximum value of |f | on and the length of . Since a contour integral is by denition a complex Riemann integral, we start with these. 14.1 Lemma (Triangle inequality for complex Riemann integrals) If g : [a, b] C is C-RI and |g | is R-RI, then
b a b

g (t) dt

|g (t)| dt.

Proof: If LHS = 0, then the statement is trivial. Hence, assume LHS = 0. Let b b = | a g |/ a g . (Hence || = 1). So,
b b

g (t) dt
a

=
a b

g (t) dt
b

=
a b

Re (g (t)) dt + i
a

Im(g (t)) dt

=
a b

Re (g (t)) dt, because LHS is real, |g (t)| dt, by Theorem 7.11, as Re(z ) |z |,
b

a b

|| |g (t)| dt |g (t)| dt, as || = 1.

=
a

Now, what is the length of a contour ? 14.2 Denition Let : [a, b] C be a contour. The length of , denoted L( ), is dened to be
b

L( ) =
a

| (t)| dt.

This really does measure the length of the curve. Intuitively speaking, (t) is the velocity of a contour, so | (t)| is the speed, and the integral of speed over time gives the length of a path. To convince you, lets compute L( ) for a couple of simple examples: 14.3 Example (a) The straight line from z to w , : [0, 1] C, (t) = z + t(w z ),
56

has length
1 1

L( ) =
0

| (t)| dt =

|w z |dt = |w z |

which is reassuring. (b) The circle of radius r centred on 0, : [0, 2 ] C, has length
2 2 2

(t) = reit ,

L( ) =
0

| (t)| dt =

|ireit |dt =

r dt = 2r
0

which is also reassuring. Were now ready to state and prove one of the most useful lemmas in complex analysis. 14.4 Lemma (Estimation Lemma) Let f : D C be a continuous complex function and : [a, b] D be a contour. Suppose that |f (z )| M for all z . Then

f (z ) dz M L( ).

Proof:

We have,
b

f (z ) dz

=
a b

f ( (t)) (t) dt |f ( (t)) (t)| dt by Lemma 14.1


b

M | (t)|dt

= ML( ).

14.5 Example Show that R ez dz z+1 R1

y iR x -iR
57

where describes the semi-circle from iR to iR in the left half-plane {z : Re(z ) 0}, and R > 1.

Let z , so |z | = R. Then, by the triangle inequality, R = |z | = |z + 1 + (1)| |z + 1| + | 1| = |z + 1| + 1. Hence |z + 1| R 1 and so, provided R > 1, 1 1 | z + 1| R1 Furthermore, if z , then x = Re (z ) 0 so |ez | = |ex+iy | = |ex ||eiy | = ex e0 = 1. Combining this with (), we see that for all z , 1 ez . z+1 R1 Since is a semicricle of radius R, we have that L( ) = R. Hence, by the Estimation Lemma, R ez dz R1 z +1 for all R > 1. ().

14.6 Remark The constant M always exists: on the image of the function |f (z )| will always be bounded because the map t |f ( (t))| is a continuous real function on [a, b], and hence is bounded. So M = sup z {|f (z )|} will do as a bound. Something bigger than this may also be useful.

Summary The length of , denoted L( ), is dened to be


b

L( ) =
a

| (t)| dt.

Suppose that |f (z )| M for all z . Then

f (z ) dz M L( ).
58

15

Uniform convergence of complex functions

15.1 Denition Let S be a subset of C and fn : S C be a sequence of complex functions. We say that the sequence fn converges pointwise to f : S C if for each z C, fn (z ) f (z ). We write fn f for short. The idea is that, associated to each point z S , we have a complex sequence an = fn (z ), and this sequence converges. Its limit will depend on which z we are considering. Hence, the limit is itself a function of z , which we denote f (z ). 15.2 Reminder Recall an a means that for each > 0 there exists N () Z+ such that |an a| < for all n N (). 15.3 Example Consider the sequence of functions fn (z ) = z n . The complex sequence associated with the point z = 0 is 0, 0, 0, 0, 0, . . .

which converges to 0. The sequence associated with the point z = 2i is 2i, 4, 8i, 16, 32i, . . .

which clearly diverges. In fact, |fn (z ) 0| = |z n | = |z |n so for all z with |z | < 1, fn (z ) 0, while for all z with |z | > 1, fn (z ) diverges. (Question: what about when |z | = 1?) Thought of as functions on the whole of C, it follows that the sequence fn does not converge pointwise. However, if we restrict the domain of fn to S = B1 (0) = {z C : |z | < 1} (the open unit disk), then fn converges pointwise to the constant function f : S C, f (z ) = 0. 15.4 Example Consider the sequence of functions fn : C C, fn (z ) = z 2 + nz . n|z |2 + 1

Does this converge pointwise on C? If so, to what? Let z C be xed. Then, if z = 0, z + z 2 /n z 1 2 = fn (z ) = 2 |z | + 1/n |z | z as n . This might lead you to expect that fn (0) diverges, but actually, this is wrong: 0 fn (0) = = 0 0. 1
59

So this sequence of functions converges pointwise to f : C C, f (z ) = z=0 0 z=0


1 z

Note that each of the functions fn is continuous, but their pointwise limit f is not. If we have a sequence of functions fn : S C such that fn f , and a contour : [a, b] S , it would be nice if we could conclude that
n

lim

fn (z ) dz =

[ lim fn (z )]dz =

f.

()

Unfortunately, we cant. 15.5 Counterexample Let S = [0, ) C and fn (z ) = n2 zenz . Note every z in S is real and non-negative. Then I claim fn 0 pointwise on S . Clearly fn (0) = 0 0. For all other z S , = |ez | = ez > 1 and hence |fn (z )| = z n2 e
nz

= n2 n log 0

as n (because exponential decay beats polynomial growth). Hence fn f where f (z ) = 0. Now consider the contour : [0, 1] S , (t) = t.
1

fn =
0

fn (t) 1 dt
1

=
0

n2 tent dt

=
0

ses ds (where s = nt)

= 1 (n + 1)en whereas

(integration by parts)

f=

0 = 0 = 1.

So pointwise convergence of functions is too weak to allow us to swap limits as in (). We need something stronger.

60

15.6 Denition Let S be a subset of C and fn : S C be a sequence of complex functions. We say that the sequence fn converges uniformly on S to f : S C if n = sup {|fn (z ) f (z )| : z S } 0. We write fn f for short. 15.7 Remark It is important to understand the dierence between pointwise convergence and uniform convergence. If fn f on S (pointwise convergence), then for each xed z S , fn (z ) f (z ). Hence, z S, > 0, N Z+ such that n N |fn (z ) f (z )| < . The crucial point is that N can depend on and z . u By contrast, if fn f on S (uniform convergence) then for each > 0 there exists N such that for all n N , sup {|fn (z ) f (z )| : z S } < , and hence |fn (z ) f (z )| < for all z S . That is, in symbols, > 0, N Z+ such that z S, n N |fn (z ) f (z )| < . Note that here N can depend on , but must be independent of z . u Clearly fn f imples fn f . It is easy to see, however, that the converse is false. Example 15.3 revisited: Let S = {z : |z | < 1} and fn (z ) = z n . Weve seen that fn converges pointwise on S to f (z ) = 0. However n = sup {|z n 0| : |z | < 1} = 1 which does not converge to 0. Hence fn does not converge to f uniformly on S . 15.8 Example Let S = {z : |z | < 1} and fn (z ) = nz + 1 . Does fn converge uniformly z+n+1
u

on S ? If so, to what? As pointed out in Remark 15.7, if fn converges uniformly to some function f on S , then it must converge pointwise to the same function f on S . So we start by analyzing the pointwise limit of the function fn (z ). Now, fn (z ) = nz + 1 z+n+1 z + 1/n z 1 + (z + 1)/n

for all z S . Hence fn f , where f : S C such that f (z ) = z .

61

But is the convergence uniform? To nd out, we compute the real sequence n : n = sup nz + 1 z : z S z+n+1 nz + 1 z 2 nz z : |z | < 1 z+n+1 | z 2 + z 1| : |z | < 1 | n + z + 1| | z | 2 + | z | + 1| : |z | < 1 n + 1 |z |

= sup

= sup

sup

3 0. n

Hence, yes, fn converges uniformly on S to f (z ) = z . A nice fact about sequences of uniformly convergent continuous functions is that their limit is also a continuous function. 15.9 Theorem Let fn : S C be a sequence of continuous functions converging uniformly to f : S C. Then f is continuous. Proof: . We wish to show that f is continuous at w for each w S . So, for given w S , choose and x > 0. We must establish that there is some > 0 such that |f (z ) f (w )| < whenever |z w | < . u Now, since fn f , there exists N such that for all n N and all z S , |fn (z ) f (z )| < 3 . By assumption, fN : S C is continuous at w , so there exists > 0 such that |z w | < |fN (z ) fN (w )| < . 3 Hence, |z w | < implies that |f (z ) f (w )| = |f (z ) fN (z ) + fN (z ) fN (w ) + fN (w ) f (w )| |f (z ) fN (z )| + |fN (z ) fN (w )| + |fN (w ) f (w )| < + + = 3 3 3 which completes the proof. Note that uniform convergence is crucial in this theorem. If we have only pointwise convergence, the limit function may fail to be continuous, as we saw in Example 15.4.
62

15.10 Theorem Let fn : D C be a sequence of continuous functions and : [a, b] D be a contour such that fn converges uniformly to f on . Then
n

lim

fn (z )dz =

lim fn (z )dz =

f (z )dz.

Proof: By Theorem 15.9 we know that f is continuous on , and hence the contour integral on the right hand side exists. Now, by the Estimation Lemma, fn f

(fn (z ) f (z ))dz

sup {|fn (z ) f (z )| : z } L( ) 0 by the Squeeze Rule, since fn f on . 15.11 Example Consider the sequence dened in Example 15.8, fn (z ) = nz + 1 z+n+1
u

] S such that (t) = eit sin t. Note that on S = {z C : |z | < 1}, and let : [0, 4 (0) = 0 1 1 ( ) = ei/4 = (1 + i). 4 2 2 We saw that fn (z ) z on S . Hence, by Theorem 15.10,
n u

lim

nz + 1 dz = z+n+1

z dz

z2 2

(/4) (0)

i 1 i/2 e = 4 4

How easy would this have been using direct computation of

fn ?

We can use Theorem 15.10 in the case where f (z ) is a convergent series of continuous functions,

f (z ) =
k =0

u k (z ),

63

for example, a power series, uk (z ) = ak z k . We just need to show that the sequence of partial sums fn (z ) = u0 (z ) + u1 (z ) + + un (z ) converges uniformly on to f (z ). Then n n f = lim
n

fn = lim

uk (z ) dz = lim
k =0

uk (z ) dz
k =0

by the linearity property of contour integrals (for nite sums). Showing directly that the sequence of partial sums converges uniformly is usually very dicult. Luckily, there is a simple sucient (but not necessary) test for uniform convergence due to Weierstrass: 15.12 Proposition (The Weierstrass M Test) Consider a series k =0 uk (z ) where each uk : S C. If there exist real constants Mk 0 such that, for each k N, |uk (z )| Mk and Mk converges, then for all z S uk (z ) converges uniformly on S .

Proof: Let fn = u0 + u1 + + un . We will rst show that fn (z ) converges for each z S . To do this well show that fn (z ) is Cauchy and appeal to the General Principle of convergence (complex sequences converge if and only if they are Cauchy). By assumption, sn = n k =0 Mk converges and hence is Cauchy. So, given > 0, there is some N such that n m N implies
n

| sn sm | =

k =m+1

Mk = Mm+1 + Mm+2 + + Mn < .

But then for any z S , we see that n m N implies


n

|fn (z ) fm (z )| =

k =m+1

uk (z ) |um+1 (z )| + |um+2 (z )| + + |un (z )|

Mm+1 + Mm+2 + + Mn < . Hence fn (z ) is Cauchy for each z S , so fn (z ) f (z ) for some function f : S C. We must now show that the convergence fn (z ) f (z ) is uniform in z . Now

n = sup {|fn (z ) f (z )| : z S } = sup {|


k =n+1

u k (z )| : z S }

sup { as n since

k =n+1

| u k (z )| : z S }

k =n+1

Mk 0

Mk converges. This completes the proof.

64

In practice, we will only ever use the Weierstrass M-Test to justify switching the order of summation and integration for contour integrals of series. So it is convenient to extract the following immediate Corollary: 15.13 Corollary (Term-by-term integration of series) Let : [a, b] D be a contour in an open set D . Let uk : D C be continuous complex functions, k N. Suppose that (i) there exist real constants Mk 0 such that |uk (z )| Mk for all z , and (ii) Then,
k =0

Mk converges.

uk (z ) dz =
k =0 k =0

uk (z ) dz

Proof: By the Weierstrass M test, follows from Theorem 15.10.

uk converges uniformly on D , so the result

k 15.14 Exercise Let f (z ) = k =0 ak z be a power series with radius of convergence R > 0. Let : [a, b] BR (0) be a contour. Prove that

f=
k =0

ak

z k dz .

Summary A sequence of function fn : S C converges uniformly to f : S C if n = sup {|fn (z ) f (z )| : z S } 0. If fn : S C converges uniformly to f : S C and is a contour in S , then
n

lim

fn =

f.
n k =0

This applies to series too. We dene fn (z ) =

uk (z ). Mk converges then

The Weierstrass M test: if |uk (z )| Mk for all z S and uk (z ) converges uniformly on S .

65

16

Winding numbers

Cauchys theorem is one of the most remarkable theorems in mathematics. To state it we need the notions of winding number and interior point. Both these notions are intuitively simple. We start with the winding number. 16.1 Denition Let be a closed contour and w a point not on . The winding number n(, w ) of about w is the net number of times that winds about w , with anticlockwise counted positively. 16.2 Remark We can make this a little more precise as follows. Write : [a, b] C as (t) = w + (t)ei(t) where : [a, b] (0, ) and : [a, b] R are continuous. Note that, since is closed, (a) = (b) and ei(a) = ei(b) . It follows that (b) (a) = 2n, where n Z, and this integer is the winding number of about w . 16.3 Example Lets determine the winding numbers for generic points A, B, C in the three regions of the plane dened by a gure of eight contour (assuming the contour is traversed once only, in the direction indicated). The trick is to imagine projecting (t) w , where w = A, B or C , radially onto the unit circle.

C 11 00

A 11 00 A 11 00

B 11 00

11 00

(a)= (b)

B 11 00

C 11 00

66

There is a useful integral formula for winding numbers. 16.4 Lemma (Winding Number Lemma) Let be a closed contour, w C\ . Then n(, w ) = 1 2i

dz . zw

Proof: If we write (t) = w + (t)ei(t) , then (t) and (t) are continuous, piecewise continuous dierentiable functions on [a, b], and we seek to compute n(, w ) = [(b) (a)]/2 . So, there exists a disection a = a0 < a1 < . . . < am = b of [a, b] such that (t) is continuously dierentiable on each [aj 1 , aj ], and is continuous on [a, b]. Hence, dz = zw =
j =1 m aj 1 aj aj 1 m aj aj 1 aj

j =1 m

(t) dt (t) w (t)ei(t) + i (t)(t)ei(t) dt (t)ei(t) (t) dt + i (t) j =1


m aj

=
j =1 m

(t) dt
aj 1 m

=
j =1

(log (aj ) log (aj 1 )) + i

j =1

((aj ) (aj 1 ))

= log (b) log (a) + i[(b) (a)], since the sums telescope. Now, is a closed contour and so (a) = (b), and hence (a) = (b). Thus n(, w ) = 1 (b) (a) = 2 2i dz . zw

16.5 Corollary Let be a closed contour and w C\ . Then n(rev , w ) = n(, w ). Proof: Follows from 16.4 and the fact that
rev

f =

f for any function f .

16.6 Corollary Let 1 and 2 be closed contours, so that the join can be taken, and w C\(1 2 ). Then n(1 2 , w ) = n(1 , w ) + n(2 , w ) Proof: f. Follows from 16.4 and the fact that
1 2

f =

f+

f for any function

67

An easy method of visual calculation In the example above note that when passing from one region to another via an edge (rather than via a crossing of two lines) the winding number for points in the regions only changes by 1. Imagine passing from region 1 to region 2 along a path which intersects the contour once. Imagine our path is a road and the contour is a river. Then as we cross the river over a bridge, the river ows beneath us either (a) from left to right, or (b) from right to left. In case (a), the winding number increases by 1 as we enter region 2. In case (b), the winding number decreases by 1 as we enter region 2. If you forget this rule, its easy to gure it out by considering the simple case of small closed loops, clockwise and anticlockwise:

0 +1 -1

crossing left to right

crossing right to left

Now, consider the complicated contour image drawn below. If we take any point w completely external to the contour (that is, points in the region containing w with |w | arbitrarily large), we clearly have n(, w ) = 0. Now imagine moving w along a path which enters a region inside the contour. On rst entering, the winding number is either 1 or 1, and we know which by the rule above. Having mapped the border regions, we can press on, visiting each region in turn and determining the winding number by those we already know and the crossing rule.

68

Justication of the method We can justify the method by considering two points w and z that lie on opposite sides of a contour line oriented so that the contour crosses a path from w to z from left to right. Lets parametrize the contour so that it starts and nishes at a point close to z and w . Now, take a loop C round w . Then, provided we orient C the right way, n( C, w ) = n(, z ):

z00 11 C w 00 11

Thus, n(, z ) = n( C, w ) = n(, w ) + n(C, w ) = n(, w ) + 1 This shows that if we pass from w to z , the winding number increases by 1.
69

16.7 Warning We have been using the image of the contour to calculate the winding number of a point. Recall that the contour and its image are dierent. Eectively, we have assumed that the contour is traversed only once. If we go round the contour twice, then the numbers calculated by eye have to be doubled. More generally, if we go round k times, then we multiply the by eye calculations by k . Interior points The method of counting the number of times a contour wraps round a point helps dene the notion of an interior point. 16.8 Denition Let be a closed contour. The interior of is the set of points Int( ) = {w C\ : n(, w ) = 0}, that is, the set of points around which winds a non-zero number of times. The interior of a contour can be quite complicated, as the next example shows. 16.9 Example Lets shade the interior points of the contour in the diagram below:

70

16.10 Example Let (t) = w + Re2it , 0 t 1. Then, Int( ) = {z : |z w | < R}.

Summary The winding number is the number of times a contour wraps round a point in an anticlockwise direction. Let be a closed contour, w C\ . Then n(, w ) = 1 2i

dz . zw

It is easy to calculate a winding number by eye. An interior point is any point with non-zero winding number.

71

17

Cauchys (Fantastic) Theorem

We now come to the fundamental theorem in complex analysis. There is no analogue in real analysis. It has far reaching, deep consequences, and most of what we will prove from now on relies on this theorem. 17.1 Theorem (Cauchys Theorem) Let D C be an open set, and f : D C be a holomorphic function. Let be a closed contour such that and its interior points lie in D . Then, f = 0.

17.2 Remarks (i) This is truly a great theorem. It refers to any open set D in C, any holomorphic function on D , and any contour with all interior points in D . And it says that any integral arising from this is zero. Thus, weak assumptions lead to a strong conclusion. (ii) It is important to note that the interior of lies within D . Consider the function f (z ) = 1/(z w ) and the contour (t) = w + eit , 0 t 2 . Then f is holomorphic on D = C\{w }, and the contour lies in D . However, f=

dz = 2i = 0, zw

by the Fundamental Example. The point is that w Int( ) and f is not holomorphic at w , so Cauchys theorem does not apply. (iii) The proof of Cauchys theorem is too complicated for the moment and we postpone it until later. Note that we cant just use the FTC since we dont know that f has an anti-derivative on D . 17.3 Exercise Using the contours in Examples 6 Question 1, to which of the following integrals does Cauchys theorem apply? (There is no need to evaluate them.) z2 1 |z |2 dz, sin z dz, dz. dz, z1 1 z 2 1 2 3 2 We will see later that Cauchys Theorem allows us to calculate f for closed , even when f fails to be dierentiable at some isolated points in the interior of . The basic idea is developed in the next example. 17.4 Example Let be any closed contour which winds once around 0 (that is n(, 0) = 1), and f (z ) = 1/z . Then f is holomorphic on C\{0}, but 0 Int( ), so f is not holomorphic on Int( ), and we cannot apply Cauchys Theorem directly. Instead, let C be a clockwise circular contour of radius > 0, centred on 0, where is chosen so small that C does not intersect . Now let be any contour from the start/end point of to the start/end point of C . Consider the closed contour = C rev .
72

Im

11 00 00 11 11 00
Re

Im

11 00 1 11 0 00 0 1

Im

11 00
Re

Re

1 11 0 00 0 1

Note that 0 is not in the interior of , so f is holomorphic on and its interior. Hence Cauchys Theorem applies and f = 0.

But 0=

f =
C rev

f=

f+

f+
C

f =

f=
C rev C

f = 2i

by the Fundamental Example. Hence 1 dz = 2i z

for any closed contour winding once (anticlockwise) around 0. We will use this trick of surrounding bad points with small circles, then cutting them out, in a more general setting later, so its important to understand this example thoroughly. Summary Let D C be a domain, and f : D C be a holomorphic function. Let be a closed contour such that and its interior points lie in D . Then,

f = 0.

73

18

Strange Consequences of Cauchys Theorem

We will prove a number of surprising theorems that can be deduced from Cauchys theorem. (i) Cauchys Integral Formula: The value of a holomorphic function at z0 is determined by the values on any contour round z0 . (ii) Liouvilles Theorem: Any holomorphic function bounded on the whole of C is contant. (iii) The Maximum Modulus Principle: The modulus of a holomorphic function on a domain achieves its maximum on the boundary of the domain. (iv) Fundamental Theorem of Algebra: Every complex polynomial has a complex root. Cauchys Integral Formula 18.1 Theorem Let D C be an open set, and f : D C be holomorphic. Let be a closed contour such that and its interior points lie in D . If w D \ , then f (z ) dz = 2i n(, w )f (w ). z w 18.2 Remarks (i) If w is in the interior of , then f (w ) = 1 2i n(, w ) f (z ) dz. zw

This says that the values of f at all points inside are completely determined by those on ! Remarkable! This contrasts with real analysis. (ii) The special case f (z ) = 1 for all z D gives the winding number lemma. Proof of Theorem 18.1: Let r be the circular contour r = w + reit , (0 t 2 ), where r > 0 is suciently small so that r is contained in the interior of . [Step 1] Let be a contour in D \{w } from the start point of to the start point of r .

74

The contour = (rev r ) + + (rev r ) (rev ),


n(,w ) times

where there are n(, w ) copies of (rev r ), has winding number zero about w , so by f (z ) Cauchys theorem applied to on D \{w }, zw
e

+
r rev

f (z ) dz n(, w ) zw

f (z ) dz zw f (z ) dz zw f (z ) dz zw f (z ) dz zw

= 0 = 0 = 0 = n(, w )
r

f (z ) dz zw

().

Note that the LHS of () is independent of r . Hence the RHS of () is a constant function of r , which thus coincides with its limit as r 0, that is,

f (z ) dz = n(, w ) lim r 0 zw

f (z ) dz . zw

[Step 2] It remains to show that


r 0

lim

f (z ) dz = 2if (w ), zw

or, equivalently, that


r 0

lim

We have, f (z ) dz = f (w ) zw

f (z ) dz 2if (w ) = 0. zw

dz f (z ) f (w ) + dz zw r z w r f (z ) f (w ) = 2if (w ) + dz, by winding lemma. zw r f (z ) f (w ) dz zw r f (z ) f (w ) sup 2r, zw z r =

Therefore f (z ) dz 2if (w ) zw

()

75

by the Estimation Lemma. Now limz w f (z ) f (w ) f (w ) < 1 , zw

f (z ) f (w ) exists (its f (w )). Hence, given zw = 1, say, there exists > 0 such that, for all z with 0 < |z w | < , so f (z ) f (w ) < | f (w )| + 1 . zw

Thus, for all r < , RHS() < 2 (|f (w ) + 1|)r , which tends to 0 as r 0. Hence by the Squeeze Rule,
r 0

lim

f (z ) dz 2if (w ) = 0, zw

as claimed. sin z dz , where (t) = 3eit , 0 t 4 . 4 z We can write the integrand in the form f (z )/(z w ) by choosing f (z ) = 1 sin z, 4 and w= 4

18.3 Example Calculate

Then f is holomorphic on and its interior (in fact its holomorphic on the whole complex plane), so by Cauchys Integral Formula sin z dz = 4z f (z ) dz z /4

= 2in(, /4)f (/4)

= 2i (2)

1 sin(/4) = i 4 2

18.4 Example Let (t) = 2 + 2eit , 0 t 2 . Calculate

ez dz. (z 3)(z + 1)

First draw the contour and indicate where the integrand is not holomorphic:

76

Note that f (z ) =

ez is holomorphic on and its interior, and so z+1

f (z ) ez dz = dz (z 3)(z + 1) z 3 = 2i 1 f (3) ie3 . = 2

(by Theorem 18.1)

Liouvilles Theorem The next theorem is also rather surprising. 18.5 Theorem Suppose f is holomorphic on the whole of C, and is bounded, i.e. there exists M such that |f (z )| M for all z C. Then, f is constant. Proof: We want to show the function is constant. This is true if f () = f (0) for an arbitrary . So we need to show |f () f (0)| = 0 for every C. Let C. We choose R 2|| and let (t) = Re2it for 0 t 1 be a circle of radius R around the origin. Then, |f () f (0)| = = = = 1 1 f (z ) f (z ) dz dz , by CIF 2i z 2i z 1 1 1 f (z ) dz 2i z z 1 f (z ) dz 2 z (z ) 1 L( ) by Estimation Lemma M sup 2 z z (z ) | | 1 M .2R 2 R (R | | ) M | | . R | |

But this is true for all R > 2|| and as R , the RHS 0. Since the LHS is independent of R, it must be 0. 18.6 Remark Contrast this with real analysis where being bounded and dierentiable on R certainly does not imply constant. For example, consider sin. This is dierentiable on all R and | sin x| 1 for all x R but it is not constant. Conversely, it follows from Liouvilles Theorem that sin, cos : C C, being holomorphic and nonconstant, must be unbounded. Can you prove this directly (without appeal to Liouvilles Theorem)?

77

The Maximum Modulus Principle 18.7 Theorem Let f : D C be a holomorphic function and be a closed contour such that Int( ) D . If |f (z )| M for all z , then |f (w )| M for all w Int( ). 18.8 Remark The theorem says that the modulus of a function within the interior of a contour is never bigger than the modulus of the function on the contour. In other words the maximum modulus occurs on the boundary of a region. Proof: This proof contains a nice trick. We shall apply CIF to w Int( ) and f (z )k , where k is a natural number: f (w )k = Dene the distance from w to by dist(, w ) = inf {|z w | : z }. Then, obviously |z w | dist(, w ) for all z . So, |f (w )|k since Mk 1 L( ), by the Estimation Lemma, 2 |n(, w )| dist(, w ) |f (w )|k Mk on . |z w | dist(, w ) | f (w )| L( ) 2 |n(, w )| dist(, w )
1/k

1 2in(, w )

f (z ) k dz. zw

Therefore,

M.

Now, we use the fact that limk x1/k = 1 for all x > 0, (this is true because 1 log x) exp(0) = 1). Thus, letting k gives |f (w )| M . x1/k = exp( k Fundamental Theorem of Algebra We are now in a position to prove the Fundamental Theorem of Algebra, (which is in fact not really a theorem of algebra but of analysis!) 18.9 Theorem Every polynomial p(z ) = z n + an1 z n1 + . . . a1 z + a0 , n 1, has a root in C. Proof: Suppose not and derive a contradiction. Since p(z ) = 0 for all z C the function f dened by f (z ) = 1/p(z ) is holomorphic on all of C. Now, for z = 0, p (z ) an1 a0 = 1+ + + n 1 as |z | . n z z z
78

So there exists an R such that |z | R implies that |f (z )| =

p (z ) 1 . This in turn means n z 2

2 1 2 n n for |z | R. p (z ) |z | R

Note that this upper bound applies when |z | = R. Hence, by the Maximum Modulus Principle applied to f with being the standard circle of radius R round the origin, 2 we have |z | < R = |f (z )| n also. R 2 So |f (z )| n on all of C. By Liouvilles theorem, f is constant, and hence so is R p. This is a contradiction, so p has a root.

Summary Cauchys theorem can be used to prove surprising theorems that have no analogues in real analysis. Cauchys Integral Formula gives the value of a holomorphic function at z0 knowing only its values on any contour around z0 . Liouvilles Theorem: Any holomorphic function bounded on the whole of C is contant. The Maximum Modulus Principle: The modulus of a holomorphic function on a domain achieves its maximum on the boundary of the domain. Fudamental Theorem of Algebra: Every complex polynomial has a complex root.

79

19

Holomorphicity of Power Series and Taylors Theorem

In this section we will show that all power series with radius of convergence R > 0 are holomorphic on the disk BR (0), and that the derivative of a power series is itself a power series (the obvious series obtained by termwise dierentiation). It follows immediately that power series are actually innitely dierentiable. It also follows that important functions such as sin, cos and exp, which we dened as power series, are holomorphic and have the expected derivatives. So power series are holomorphic. Rather more surprisingly, the converse is true. That is, every function which is holomorphic on an open set can be written, on a neighbourhood of any point in that set, as a convergent power series! This is Taylors Theorem which, as we will see, is yet another (strange) consequence of Cauchys Theorem. Since power series are innitely dierentiable, we see that, in the complex world, if a function is once dierentiable on an open set, it is innitely dierentiable! This is radically dierent from real analysis, as we will see. Holomorphicity of power series We begin by showing that power series are continuous on their disk of convergence, which follows immediately from the following proposition:
n 19.1 Proposition Let f (z ) = n=0 an z have radius of convergence R > 0 and (0, R). Then the power series converges uniformly to f (z ) on the closed disk B (0) = {z : |z | }.

Proof:

Exercise: use the Weierstrass M Test and Theorem 3.36.

n 19.2 Corollary Let f (z ) = n=0 an z have radius of convergence R > 0. Then f is continuous on BR (0) = {z : |z | < R}.

Proof:

Each partial sum


k

fk (z ) =
n=0

an z n

is a polynomial, hence is continuous. Hence, by Proposition 19.1 and Theorem 15.9, f is continuous on B (0) for all (0, R).
n So f (z ) = n=0 an z is continuous on BR (0). But is it holomorphic? The obvious candidate for its derivative is the series obtained by term-by-term dierentiation1 , i.e. n1 . But, initially, we do do not even know that this sequence converges. n=1 nan z So, we begin by proving that a power series and its obvious derivative have the same radius of convergence, and then show that this obvious derivative really is the derivative of the series.

19.3 Lemma The series f (z ) = radius of convergence.


1 Hopefully,

n=0

an z n and g (z ) =

n=0

nan z n1 have the same

studying maths has taught you that just because something is obvious doesnt mean its true!

80

Proof:

We dene the following subsets of R, A = {|z | : |an z n | converges}, B = {|z | : |nan z n1 | converges},

recalling that sup A and sup B are the radii of convergence of f (z ) and g (z ) respectively. We will prove that (i) for all z with |z | < sup B , f (z ) converges absolutely (so |z | A and hence sup A sup B ); (ii) for all z with |z | < sup A, g (z ) converges absolutely (so |z | B and hence sup B sup A). It follows immediately that sup A = sup B , which is what we seek to prove. (i) Let z C with |z | < sup B . Then |nan z n1 | converges (Theorem 3.36), and |an z n | = |z | |an z n1 | |z | |nan z n1 |,

so |an z n | converges by (naive) comparison with |nan z n1 |. Hence f (z ) converges absolutely. (ii) Let z C with |z | < sup A. Choose (|z |, sup A). Then |an |n converges (Theorem 3.36). Now |nan z
n1

| = |z |

|z | n

|an |n ,

and n(|z |/)n is easily shown to converge (use the ratio test, noting that > |z |), so the real sequence of terms bn = n(|z |/)n must converge to 0. Since bn is convergent, it is bounded, that is, there exists K such that bn < K for all n. Hence |nan z n1 | K |z |1 |an |n , |an n |. Hence g (z ) converges

so |nan z n1 | converges by (naive) comparison with absolutely.

So the termwise derivative converges on BR (0) (and hence is continuous on this disk, and converges uniformly on any closed subdisk). To prove that it really is the derivative of the original power series, we need one more ingredient (which well also use in the proof of Cauchys Theorem, much later). 19.4 Lemma Let D C be open, g : D C be continuous, w D , and hn C be any sequence converging to 0. Consider the sequence of functions gn : [0, 1] C, Proof: We must show that n = sup{|g (w + hn t) g (w )| : 0 t 1}
81

gn (t) = g (w + hn t).

Then gn (t) converges uniformly to the constant function g (w ) on [0, 1].

converges to 0 as n . Let > 0 be given. Since g is continuous at w , there exists > 0 such that |z w | < = | g (z ) g (w )| < ( ).

But hn 0, so there exists N Z+ such that n N implies |hn | < . Hence, for all n N and all t [0, 1] |(w + hn t) w | = t|hn | |hn | < , so by (), n < . Such N exists for any , so n 0.
n 19.5 Theorem Suppose that f (z ) = n=0 an z has radius of convergence R > 0. Then f is holomorphic on BR (0) and f (z ) = n=0 nan z n1 , for all |z | < R.

n1 . By Lemma 19.3 this has radius of convergence Proof: Let g (z ) = n=1 nan z R, by Proposition 19.1 it converges uniformly on B (0) for any (0, R) and by Corollary 19.2 it is continuous on BR (0). Let w BR (0) be xed. Then given any h C suciently small that w + h BR (0), dene the contour h (t) = w + ht, 0 t 1 (a straight line from w to w + h). Then

g =
h h n=1

nan z n1 dz an
n=1 h w +h an [z n ]w n=1

= =

nz n1 dz

by Theorem 15.10 by FTC for Contour Integrals

= f (w + h) f (w ). Hence 1 f (w + h) f (w ) = h h
1

g (z )dz =
h 0

g (w + th)dt.

We wish to evaluate this quoitient in the limit h 0. So, let hn C\{0} be any sequence converging to 0, and gn (t) = g (w + thn ). By Lemma 19.4, gn converges uniformly on [0, 1] to the constant g (w ), so by Theorem 15.10 f (w + hn ) f (w ) = n hn lim
1

g (w )dt = g (w ).
0

Hence f is dierentiable at w and f (w ) = g (w ), as was to be proved. 19.6 Example We know that

zn =
n=0

1 1z
82

for |z | < 1.

How? The k th partial sum is fk (z ) = 1 + z + z 2 + + z k zfk (z ) = z + z 2 + + z k+1 (1 z )fk (z ) = 1 z k+1 1 z k+1 fk (z ) = 1z

which converges to (1 z )1 if |z | < 1. This is a very important power series which we will use repeatedly in the following lectures. Dierentiating both sides gives

nz n1 =
n=0

1 (1 z )2

for |z | < 1

which is not at all obvious. Applying Theorem 19.5 to the power series dening exp, sin and cos, we immediately nd 19.7 Corollary exp, sin and cos are holomorphic on the whole complex plane, and have the expected derivatives, namely d z e = ez , dz d sin z = cos z, dz d cos z = sin z. dz

We have seen that, if f is dened by a power series on the disk of convergence, then so is f , and f is a series. We can dierentiate f term-by-term to get f and so on. This suggests the following: 19.8 Corollary If f (z ) = an z n has radius of convergence R > 0, then f is innitely f (k) (0) for all k . dierentiable on BR (0). Furthermore, ak = k! Proof: Exercise: use induction and Theorem 19.5.
n=0

19.9 Remark Weve considered only power series based at z0 = 0, but all the results clearly apply to power series based at general z0 , that is

f (z ) =
n=0

an (z z0 )n

equally well. We just note that f (z ) is holomorphic on BR (z0 ) if and only if f (z + z0 ) is holomorphic on BR (0).

83

Taylors Theorem Having proved that convergent power series are holomorphic (in fact, innitely differentiable) we turn to the converse question. Our next result shows that every holomorphic function is, at least locally, a convergent power series. 19.10 Theorem (Taylors theorem for complex functions) Suppose f is a holomorphic function on a disk BR (z0 ) for some R > 0. Then there exists a unique set of coecients an C such that

f (z ) =
n=0

an (z z0 )n , for all |z z0 | < R.

Proof: Choose any (0, R). We will nd a power series for f (z0 + h) in terms of h, valid for all h B (0). Dene the closed contour (t) = z0 + eit , 0 t 2 , and note that Int( ) BR (z0 ), so we may apply Cauchys integral formula: for each h with |h| < , f (z0 + h) = 1 2i

f (z ) dz. z (z0 + h) f (z ) = z z0

Now, for all z , |z z0 | = > |h|, so f (z ) f (z ) = z (z0 + h) (z z0 ) 1


h z z 0 n=0

h z z0

=
n=0

f (z )hn (z z0 )n+1

Hence 1 f (z0 + h) = 2i where u n (z ) = Now let Mn = sup |un (z )| = sup


z z

un (z )dz
n=0

f (z )hn . (z z0 )n+1

f (z )hn |h|n |h|n | f ( z ) |} = { sup = C , (z z0 )n+1 n+1 n+1 z

where C is independent of n. Then n=0 Mn converges since |h| < . (Use the ratio test.) Thus, by Corollary 15.13 (the Weierstrass M Test), 1 f (z0 + h) = 2i for all |h| < , where an =

un (z )dz =
n=0 n=0

an hn

1 2i

f (z ) dz. (z z0 )n+1
84

The coecients ostensibly depend on , but, by Corollary 19.8 we see that an = f (n) (z0 )/n! which is unique and independent of . Hence

f (z ) =
n=0

an (z z0 )n

on B (z0 ) for all (0, R), and hence on BR (z0 ). A function which locally coincides with a convergent power series is often called analytic. This explains why some authors use the word analytic to mean complex dierentiable: dierentiable and analytic are equivalent in complex analysis. Analytic and dierentiable are not equivalent in real analysis as we see in a later example. An immediate Corollary of Taylors Theorem is 19.11 Corollary (Innite Dierentiability) Suppose f is a holomorphic function on an open set D C. Then, f is innitely dierentiable on D , i.e. it has derivatives of all orders. Proof: Suppose z0 is any point in D . Since D is open, there exists R > 0 such that BR (z0 ) D , and by Taylors Theorem f (z ) coincides with a convergent power series on BR (z0 ). But power series are ininitely dierentiable (Corollary 19.8). This Corollary is really astounding. It doesnt happen for real functions as the next example illustrates. 19.12 Example Suppose f : R R is dened by f (x) = Then, f is dierentiable with derivative f (x) = 2x x > 0 0 x 0. x2 x > 0 0 x 0.

This derivative is not dierentiable at 0, so f is not innitely dierentiable. When computing a Taylor series, we can use the formula an = f (n) (z0 )/n! to nd the coecients, just as in real analysis. Working in the complex plane gives us extra insight, however. 19.13 Example What is the radius of convergence of the Taylor series of the function f (z ) = based at z = 0?
85

ez

1 +2

Solution: we could try to compute the Taylor expansion explicitly by using an = f (n) (0)/n!, then use the ratio test to nd R, but thats hard work. I nd that f (z ) = 1 1 1 z z2 + 3 9 54

(check it!) but the expression for f (3) (z ) is suciently nasty to stop me going any further. So if we dont know the power series, how on earth can we nd its radius of convergence? Well, note that f is holomorphic on the whole of C except where ez = 2, that is where ex+iy = 2ei Equating moduli, we see that ex = 2, x = log 2

Equating arguments, we see that eiy = ei , y = + 2k, k Z.

Hence, f is holomorphic on the set D = C\{log 2 + (2k + 1) : k Z}.

The closest points to 0 which are not in D are log 2 i . Hence, f is holomorphic on the disk BR (0) for R = (log 2)2 + 2 , but not for R any larger than this. It follows from Taylors Theorem that the radius of convergence of the Taylor series is R = (log 2)2 + 2 . Note that the same conclusion holds, even if we are thinking of f as a real function of a real variable. Some facts about power series We now use Taylors Theorem to prove some facts about series which might otherwise be dicult to demonstrate.
n n 19.14 Theorem Suppose f (z ) = n=0 an z for all |z | < R1 and g (z ) = n=0 bn z for all n |z | < R2 . Then, f (z )g (z ) = (f g )(z ) = n=0 cn z for all |z | < min{R1 , R2 }, where cn = n k =0 ak bnk .

Proof: Let R = min{R1 , R2 }. Then, f and g are complex dierentiable for all |z | < R. The product of two dierentiable functions is dierentiable by the product
86

rule. So, f g is dierentiable. By Theorem 19.10 it has a series expansion for |z | < R, and by Corollary 19.8, we get

(f g )(z ) =
n=0

(f g )(n) (0) n z . n!

Now we can apply Leibnitzs rule


n

(f g )

(n)

(0) =
k =0 n

n! f (k) (0)g (nk)(0) k !(n k )! n! f (k) (0) g (nk)(0) k ! (n k )! ak bnk .

=
k =0

= n!
k =0

Therefore,

(f g )(z ) =
n=0

cn z , (|z | < R), where cn =

ak bnk .
k =0

19.15 Remark This theorem justifes the usual applied mathematicians trick of working to order n. For example, to order 2, if

f (z ) =
n=0

an z n = a0 + a1 z + a2 z 2 + ,

g (z ) =
n=0

bn z n = b0 + b1 z + b2 z 2 +

then wed expect f (z )g (z ) = (a0 + a1 z + a2 z 2 + )(b0 + b1 z + b2 z 2 + ) = a0 b0 + (a0 b1 + a1 b0 )z + (a0 b2 + a1 b1 + a2 b0 )z 2 + and these are precisely the rst 3 terms of the series given by the theorem. 19.16 Corollary Suppose f and g are power series with positive radii of convergence. If g (0) = 0, then f /g has a power series expansion at 0 with positive radius of convergence. Proof: Dene h(z ) = 1/g (z ). By the dierentiability of g and the fact that g (0) = 0, there is an open disk B (0) on which h is dierentiable. Thus, h has a power series expansion, and the result follows from the theorem above applied to f h. 19.17 Remark The precise value of the radius of convergence will depend on where g is zero.

87

19.18 Example Express f (z ) =

ez as a power series about z = 2. 1z Solution: we must nd power series ez = g (z ) = an (z 2)n and 1 = h(z ) = 1z an = bn (z 2)n . e2 n! bn = (1)n+1

Now

g (n) (z ) = ez h(n) (z ) =

g (n) (2) = e2

n! (1 z )n+1

h(n) (2) =

n! (1)n+1

Hence f (z ) =
n=0

cn (z 2)n , where
n

cn =
k =0

(1)

2 nk +1 e

k!

= (1)

n+1 2

k =0

(1)k . k!

The rst few terms in this power series are: e2 e2 ez = e2 + 0(z 2) (z 2)2 + (z 2)3 + 1z 2 3 Summary Power series are holomorphic on their disk of convergence. The derivative of a power series is another power series, obtained by termwise dierentiation. Power series are ininitely dierentiable. If a complex function is holomorphic on an open set, then at every point it has a power series expansion valid on some open disk. Holomorphic functions are innitely dierentiable. If S1 and S2 are power series with radii of convergence R1 and R2 , then S1 S2 is a power series with radius of convergence min{R1 , R2 }. If S is a power series at z0 with positive radius of convergence, then 1/S has a power series expansion at z0 provided S (z0 ) = 0.

88

20

Laurent Expansions

Hopefully it is becoming clear that in order to compute f around a closed contour , it is crucial to understand the behaviour of f (z ) at the points in Int( ) where it fails to be holomorphic. We have seen (Taylors theorem) that if f is dierentiable on an open set containing z0 , we can write it as a power series in (z z0 )

f (z ) =
n=0

an (z z0 )n ,

|z z0 | < ,

and the coecients an are unique. In this section we will show that if f is dierentiable on B (z0 ) except at z0 itself, we can still write it as a power series in (z z0 ) with unique coecients, but now the power series may contain negative powers of (z z0 ),

f (z ) =
n=

an (z z0 )n ,

0 < |z z0 | <

( ).

Such a series is called the Laurent expansion of f about z0 . We start with a quick denition of double sided series: 20.1 Denition By n= an = nZ an we mean the sum of the pair of ordinary series n=0 an and n=1 an . We say that n= an converges (to s+ + s ) if n=0 an converges (to s+ ) and a converges (to s ). n n=1 So we now know what the expression in () means. We want to show that any function holomorphic on a punctured disk can be written as a double sided power series, as in (), centred on the puncture. 20.2 Theorem (Existence of the Laurent expansion) Let f be a function which is holomorphic on the punctured disk A = {z : 0 < |z z0 | < R}. Then

f (z ) =
n=

an (z z0 )n ,

z A,

where an =

1 2i

with a circle of radius (0, R), centred on z0 .

f (w ) dw (w z0 )n+1

Proof: Without loss of generality, we may assume z0 = 0, so that A = BR (0)\{0}. Fix z A and choose r1 , R1 such that 0 < r1 < |z | < R1 < R. Dene closed contours and as shown in the diagram below. Then, since z Int() and z / Int( ), f (z ) = 1 2i 1 0 = 2i f (w ) dw wz f (w ) dw wz by Cauchys Integral Formula by Cauchys Theorem.

89

z
R R1

r1

So f (z ) is the sum of these two integrals. But in this sum, the contributions from the straight line segments cancel, so, denoting by the (anticlockwise) circle of radius centred on 0, 1 f (w ) f (w ) 1 f (z ) = dw dw. 2i R1 w z 2i r1 w z For all w R1 , |z/w | < 1, so 1 1 1 1 = = wz w 1 (z/w ) w Similarly, for all w r1 , |w/z | < 1, so 1 1 1 1 = = wz z 1 (w/z ) z Hence 1 f (z ) = 2i = 1 2i
R1 n=0 R1 n=0 n=0 n=0

zn . wn

wn . zn

1 f (w ) n z dw + n +1 w 2i 1 f (w ) n z dw + n +1 w 2i

r1 n=0 1

f (w ) (n+1) z dw w n

f (w ) m z dw m+1 r1 m= w

where m = (n + 1). Its not hard to show, using the Weierstrass M Test (Theorem 15.12) that both these series converge uniformly on their respective contours, so by Theorem 15.10, we may interchange the order of summation and integration. Then

f (z ) =
n=0

1 2i

R1

f (w ) dw z n + n +1 w n=
90

1 2

r1

f (w ) dw z n . n +1 w

It seems as though the coecients of our expansion depend on r1 , R1 and hence on z (recall r1 < |z | < R1 by denition). However, I claim that

f (w ) dw w n+1

is independent of , provided (0, R). To see this, choose any pair 1 , 2 with 0 < 1 < 2 < R, and let (t) = t, 1 t 2 (see diagram). Then = 1 rev (2 ) rev ( ) is a closed contour whose interior lies in A, so by Cauchys Theorem

f (w ) dw = 0 w n+1 f (w ) f (w ) dw = dw. n +1 n+1 w 2 w

Hence, in our expansion for f (z ) we can replace the contours r1 and R1 by any single contour with (0, R). This completes the proof. 20.3 Theorem (Uniqueness of the Laurent expansion) Let f be holomorphic on the punctured disk A = {z : 0 < |z z0 | < R} and suppose f (z ) =
nZ

bn (z z0 )n

for all z A. Then bn = an , as dened in the statement of Theorem 20.2. Proof: Again we may assume z0 = 0 without loss of generality. We will compute an as dened in Laurents Theorem, and demonstrate that an = bn . So, choose any (r, R). Then an = = 1 2i

f (w ) 1 dw = n +1 w 2i

bm w mn1 dw
mZ

1 bm 2i mZ

w mn1 dw

91

where we have used the Weierstrass M Test (Theorem 15.12) and Theorem 15.10 again to justify swapping the order of integration and summation (check it!). By the Fundamental Example, we know that 1 dw = wp 0 p = 1 , 2i p = 1

so the only nonzero term in the above double sided series is the m = n term, which is just 2i. The result follows. Theorems 20.2 and 20.3 can be combined to give a clean statement called Laurents Theorem: 20.4 Corollary (Laurents Theorem) Let f be holomorphic on the punctured disk A = {z : 0 < |z z0 | < R}. Then there exists a unique set of coecents an C, n Z, such that f (z ) = an (z z0 )n
nZ

for all z A. 20.5 Denition Let f be a function holomorphic on a punctured disk BR (z0 )\{z0 }. Then the power series f (z ) = an (z z0 )n
nZ

valid on BR (z0 )\{z0 }, whose existence and uniqueness is guaranteed by Laurents Theorem is called the Laurent expansion of f about z0 . Furthermore: If an = 0 for all n < 0, we say that f has a removable singularity at z0 . If an = 0 for all n < 1 and a1 = 0, we say that f has a simple pole at z0 . More generally, if there is some N Z+ such that an = 0 for all n < N and aN = 0, we say that f has a pole of order N at z0 . Note a simple pole is a pole of order 1. If no such N exists (that is, the set {n Z : an = 0} is unbounded below) then we say that f has an essential singularity at z0 In all cases, the coecient a1 in the Laurent expansion is called the residue of f at z0 , denoted res(f, z0 ). 20.6 Remark (i) If f has a removable singularity at z0 we can extend f to a holomorphic function on the whole disk BR (z0 ) by dening f (z0 ) = a0 , so

f (z ) =
n=0

an (z z0 )n

for all |z z0 | < R.

This explains the name removable singularity.

92

(ii) Note that uniqueness of the Laurent coecients an is crucial for these denitions to make sense (else the classication of a singularity z0 would depend on which expansion we used). (iii) Although Theorem 20.2 gives a formula for the coecients an as contour integrals, we will never calculate them like that. Since they are unique, if we can, by any means, come up with a Laurent expansion valid in a neighbourhood of z0 , then the coecients in this expansion are the ones we want. 20.7 Example (i) f (z ) = 1/z is holomorphic on C\{0}. It has an expansion at 0 with an = 0 for n = 1, a1 = 1. Hence f has a simple pole with residue 1 at 0. (ii) f (z ) = ez /z 3 is also holomorphic on C\{0}. Its Laurent expansion at 0 is, 1 ez = z3 z3 which has an =
n=0

1 1 1 1 z zn = 3+ 2+ + + + n! z z 2z 6 24

1 1 for n 0, a1 = , a2 = a3 = 1, and aj = 0 for (n + 3)! 2 1 j 4. Hence f has a pole of order 3 (a triple pole) with residue at 0. 2 1 is holomorphic on C\{i, i}. Lets compute its Laurent expansion (iii) f (z ) = 2 z +1 about i. (Its expansion about i is similar and is left as an exercise). Since f is holomorphic on B2 (i)\{i}, this expansion will be valid on the punctured disk B2 (i)\{i} also (so unlike the previous two examples, the expansion is not valid on the whole domain of f ). Dene w = z i, so we seek an expansion of f (z ) = f (i + w ) in powers of w : f (z ) = 1 (z i)(z + i) 1 = w (w + 2i) 1 1 = w 2i(1 + w/2i) 1 1 = 2iw 1 (iw/2)

1 = 2iw

(iw/2)n
n=0

(by Example 19.6)

=
n=0

in1 n1 w 2n+1 im (z i)m (m = n 1)

2m+2 m=1

93

So f has a simple pole of residue (iv) f (z ) =

1 at i. 2i

sin z is holomorphic on C\{0}. Its Laurent expansion is z

f (z ) =
n=0

1 z 2n+1 (1)n = z (2n + 1)!

n=0

(1)n

z 2n . (2n + 1)!

This expansion has no negative powers, so f has a removable singularity at 0. If we dene f (0) = a0 = 1, we can extend f to a function holomorphic in the whole plane. As you can see, computing Laurent expansions can be quite hard work. Luckily we usually only want to know the order of the pole z0 and its residue, and we will develop ways of extracting this information without computing the whole expansion. Before embarking on that endeavour, well do a quick example to illustrate why, for the purposes of contour integration, we care about poles and residues. 20.8 Example Let (t) = eit , 0 t 2 . How can we compute

ez dz z2

We cant use Cauchys Theorem as the integrand is not holomorphic at 0. We cant use Cauchys Integral Formula because ez /z is not holomorphic at 0 either. The Fundamental Theorem of Calculus doesnt help as we cant see any obvious antiderivative (and in fact, there is no antiderivative, as well see). Instead, lets compute the integrands Laurent expansion about 0 (which is guaranteed to be valid on the whole of C\{0}), and do the interchange trick to integrate this term-by-term2 . ez 1 f (z ) = 2 = 2 z z

n=0

zn zn = n! n=2 (n + 2)!

f =

zn dz. n=2 (n + 2)!

2 Obviously,

we must check that we can do this!

94

Now z implies |z | = 1, so 1 zn = =: Mn (n + 2)! (n + 2)! 1 = e < so the series converges uniformly on by the k ! n=2 k =0 Weierstrass M Test. Hence, by Theorem 15.10, say, and Mn = f=

1 (n + 2)! n=2

z n dz.

But as we argued in the proof of Theorem 20.3, only the n = 1 term in this series is non-zero, by the Fundamental Example. Hence, f=

1 (1 + 2)!

z 1 = 2i.

Notice that only the 1/z term in the Laurent expansion mattered. By the Fundamental Example, all the other terms were irrelevant. This is the key to understanding why poles and their residues are important. In fact we will develop a method for computing f which requires us to know only the residues of f at all its poles in the interior of . As an aside, note that the integral of this f around this closed contour is non-zero, so the fact that we were unable to see an antiderivative of f was not due to lack of imagination: none exists! Summary A function holomorphic for 0 < |z z0 | < R has a Laurent expansion

f (z ) =
n=

an (z z0 )n

valid for 0 < |z z0 | < R. The coecients an are unique. We say that f has a pole of order N at z0 , if aN = 0 and an = 0 for all n < N . A pole of order 1 is a simple pole. The coecient a1 in the Laurent expansion is called the residue of f at z0 .

95

21

Zeros and Poles of Functions

In all the examples weve considered, if f (z ) fails to be holomorphic at z0 , it is because f (z ) = p(z )/q (z ) where q (z0 ) = 0. So to investigate poles of holomorphic functions, it seems natural to investigate zeros of functions rst. 21.1 Denition A holomorphic function f : D C has a zero at z0 D if f (z0 ) = 0. By Taylors Theorem, f has a Taylor expansion valid in some neighbourood of z0 , that is, there exists > 0 such that

f (z ) =
n=0

an (z z0 )n ,

for all z B (z0 ), and the an are unique. We say f has a zero of order m at z0 if a0 = a1 = = am1 = 0 but am = 0. 21.2 Remark Since the Taylor coecients an = f (n) (z0 )/n!, f has a zero of order m at z0 if and only if f (j ) (z0 ) = 0 for all 0 j < m and f (m) (z0 ) = 0. In particular, if f (z0 ) = 0 and f (z0 ) = 0, then f has a zero of order 1. 21.3 Examples (i) The function f (z ) = z 2 has a zero of order 2 at 0.

(ii) The function f (z ) = z (z + 2i)3 has a zero of order 1 at 0 and one of order 3 at 2i. (iii) The function f (z ) = sin z has zeros at k , for all k Z. These are all of order 1, since f (k ) = cos k = 0 for all k Z. (iv) The function f (z ) = 1 cos z has zeros at 2k , for all k Z. f (z ) = sin z , f (z ) = cos z , so f (2k ) = sin 2k = 0 while f (2k ) = cos 2k = 1. Hence every zero is of order 2. 21.4 Exercise Find the zeros, and their orders, of: (i) z 2 + 9, (ii) ez 2 1, (iii) (z 2 + 1)2 , (iv) zez ,
2

(v) z 3 + 1.

21.5 Lemma Suppose that f : D C is a holomorphic function with a zero of order m at z0 . Then there exists R > 0 and a unique holomorphic function g : BR (z0 ) C such that f (z ) = (z z0 )m g (z ), for all z BR (z0 ), and g (z0 ) = 0. Proof: First, if such a g exists, it is clearly unique since g (z ) = f (z )/(z z0 )m on BR (z0 )\{z0 }, and is holomorphic, hence continuous, at z0 , so g (z0 ) = limz z0 g (z ), which is uniquely determined by the values of g on BR (z0 )\{z0 }. By Taylors Theorem, there exists R > 0 such that

f (z ) =
n=m

an (z z0 )n ,

for all z BR (z0 ).

96

Let g (z ) =

n=m

an (z z0 )

nm

=
n=0

an+m (z z0 )n .

This converges absolutely on BR (z0 ), by comparison with the Taylor expansion of f (m) (z ) (check it!). Hence g is holomorphic on BR (z0 ) by Theorem 19.5. Clearly f (z ) = (z z0 )m g (z ) and g (z0 ) = am = 0 as required. The next result establishes a link between the order of a pole of a quotient of holomorphic functions p(z )/q (z ) and the order of a zero of q . 21.6 Lemma Suppose that f (z ) = p (z ) , where q (z )

(i) p is holomorphic, and p(z0 ) = 0, (ii) q is holomorphic, and has a zero of order N at z0 . Then, f has a pole of order N at z0 . By Lemma 21.5, we have q (z ) = (z z0 )N r (z ) where r (z ) is holomorphic p (z ) is holomorphic at z0 , so by Theorem 19.10 it has with r (z0 ) = 0. Then, g (z ) = r (z ) a power series expansion: Proof:

g (z ) =
n=0

an (z z0 )n , valid for |z z0 | < R, some R > 0.

Thus, f (z ) = = p (z ) q (z )

p (z ) (z z0 )N r (z ) g (z ) = (z z0 )N

1 an (z z0 )n N (z z0 ) n=0 a0 a1 a2 = + + + ... N N 1 (z z0 ) (z z0 ) (z z0 )N 2 a0 = g (z0 ) = p(z0 ) = 0, r (z0 )

But,

because p(z0 ) = 0 by assumption and r (z0 ) = 0 by denition. So as a0 = 0, we deduce that f has a pole of order N at z0 .

97

21.7 Error The above lemma gives a method for nding a pole and its order. The definition of a pole is that given in Denition 20.5. It is not the multiplicity of the denominator in an expression. Do not confuse the denition of an object with a process by which we nd the object. 21.8 Example sin z has a pole of order 2 at z = 3, since sin is (z 3)2 holomorphic and sin 3 = 0, and (z 3)2 is holomorphic with a zero of order 2 at 3. z+3 has a simple pole at z = k for each k Z. We have (ii) The function f (z ) = sin z p(z ) = z + 3 and q (z ) = sin z , so p(k ) = k + 3 = 0, q (k ) = 0. We know each of these zeros has order 1 (by Remark 21.2) since q (k ) = cos k = (1)k = 0. (i) The function 1 + ez (a) , z (z 2 1) 1+z (b) 3 , z 2z 2 ez 1 (c) (It aint 5!) z5
2

21.9 Exercises Find the poles and their orders for

There is an obvious generalization of Lemma 21.6 which deals with the case where p, q both have zeros at z0 . 21.10 Lemma Suppose that f (z ) = p (z ) , where q (z )

(i) p is holomorphic, and has a zero of order N1 at z0 , and (ii) q is holomorphic, and has a zero of order N2 > N1 at z0 . Then, f has a pole of order N2 N1 at z0 . Proof: By Lemma 21.5, we have p(z ) = (z z0 )N1 r1 (z ) and q (z ) = (z z0 )N2 r2 (z ) where r1 (z ), r2 (z ) are holomorphic on B (z0 ) for some > 0 and r1 (z0 ) = 0, r2 (z0 ) = 0. 1 (z ) is holomorphic on BR (z0 ) for some R > 0 (Corollary 19.16) and Hence g (z ) = r r2 (z ) so has a Taylor expansion

g (z ) =
n=0

an (z z0 )n

valid on this disk. Note that g (z0 ) = r1 (z0 )/r2 (z0 ) = 0. Now, for all z BR (z0 )\{z0 }, g (z ) = an (z z0 )n(N2 N1 ) f (z ) = N N 2 1 (z z0 ) n=0 a0 a1 = + + . N N 2 1 (z z0 ) (z z0 )N2 N1 1 Since a0 = 0, the result follows.

98

21.11 Example Classify the singularity of f (z ) =

sin(1 cos z ) . z4 Clearly f is holomorphic except at 0. Let p(z ) = sin(1 cos z ) and q (z ) = z 4 . Then z = 0 is a zero of q of order N2 = 4.But p(0) = sin(1 1) = sin 0 = 0 also, so we cannot conclude that 0 is a pole of f of order 4. We must determine the order of 0 as a pole of p(z ). Now p (z ) = sin z cos(1 cos z ) p (0) = 0 p (z ) = cos z cos(1 cos z ) sin2 z sin(1 cos z ) p (0) = 1 = 0.

Hence 0 is a zero of p of order N1 = 2, and so, by Lemma 21.10, 0 is a pole of f of order N2 N1 = 2.

Summary A holomorphic function f : D C has a zero at z0 if f (z0 ) = 0. By Taylors theorem f (z ) =


n=1

an (z z0 )n

for all z with |z z0 | < R for some R. Then the order of the zero z0 is N if a1 = = aN 1 = 0 but aN = 0. Suppose that f (z ) = p (z ) , where q (z )

(i) p is holomorphic, and p(z0 ) = 0, (ii) q is holomorphic, and has a zero of order N at z0 . Then, f has a pole of order N at z0 . Suppose that f (z ) = p (z ) , where q (z )

(i) p is holomorphic, and has a zero of order N1 at z0 , and (ii) q is holomorphic, and has a zero of order N2 at z0 . Then, f has a pole of order N2 N1 at z0 .
99

22

Rough Guide to Calculating Residues

Residues are an important part of any course on Complex Analysis. They will allow us to calculate easily otherwise complicated integrals without doing any integration! This section contains the main methods for calculating the residues of poles. You should know them (and their proofs, of course!). Method 1: Simple Poles If f has a simple pole (i.e. order 1) at w then res(f, w ) = lim (z w )f (z ).
z w

Proof:

As f has a simple pole at w we have

f (z ) =
n=1

an (z w )n ,

and this series converges on BR (w )\{w } for some R > 0. Consider

g (z ) =
n=0

an1 (z w )n .

On BR (w )\{w }, this coincides with (z w )f (z ), and hence converges. Clearly it also converges at z = w (to a1 ). Since g (z ) converges on the whole disk BR (w ), it is holomorphic on this disk, by Theorem 19.5, and hence continuous at (in particular) w . Thus lim (z w )f (z ) = lim g (z ) = g (w ) = a1 .
z w z w

22.1 Examples

(i) The function f (z ) =

sin(z ) has a pole of order 1 at z = 3, so z3


z 3

res(f, 3) = lim(z 3)f (z ) = lim (z 3)


z 3

sin(z ) = lim sin(z ) = sin(3). z 3 z3

(ii) The function f (z ) =

z has a pole of order 1 at z = 0, so 1 cos z


z 0

res(f, 0) = lim z = 2 lim

z 4 (z/2)2 z2 = lim = lim 1 cos z z 0 2 sin2 (z/2) z 0 2 sin2 (z/2) z/2 sin(z/2)
2

z 0

z/2 = 2 lim z 0 sin(z/2)

= 2.

100

Method 2: Some quotients (Really good method!) Suppose f (z ) = Then, res(f, w ) = p (w ) . q (w ) p (z ) with p and q holomorphic, p(w ) = 0, q (w ) = 0 and q (w ) = 0. q (z )

Proof: By Remark 21.2, q has a zero of order 1 at w , so f has a simple pole at w by Lemma 21.6. Hence res(f, w ) = lim (z w )f (z ), by Method 1,
z w

= lim (z w )
z w

= = =

p (z ) q (z ) q (z ) lim p(z ) z w zw q (z ) q (w ) lim p(z ) , as q (w ) = 0, z w zw limz w p(z ) q (z ) q (w ) limz w zw p (w ) , by denition of dierentiation. q (w )

22.2 Remark The conditions imply that f has a simple pole, so the method will only work in this case. 22.3 Examples 1 has a pole at z = 1. We have p(z ) = 1 1 z3 p(1) 1 1 and q (z ) = 1 z 3 so q (1) = 3 12 = 3. Thus res(f, 1) = = = . q (1) 3 3 (i) The function f (z ) = 2z 2 has a pole at z = ei/4 (and more besides, but well 1 + z4 ignore them). Let p(z ) = 2z 2 and q (z ) = 1 + z 4 . Then p(ei/4 ) = 2ei/2 = 0; q (z ) = 4z 3 so q (ei/4 ) = 4e3i/4 = 0. So res f, e
i/4

(ii) The function f (z ) =

p(ei/4 ) 2ei/2 ei/4 1 = i/4 = 3i/4 = = (1 i) 2. q (e ) 4e 2 4

101

Method 3: Poles of order N We now generalise Method 1 (which is the case N = 1 in the following). Suppose that f has a pole of order N at w . Then d N 1 1 lim N 1 (z w )N f (z ) . res(f, w ) = z (N 1)! w dz

Proof:

By assumption

f (z ) =
n=N

an (z w )n

and this series converges on BR (w )\{w }. Consider

g (z ) =
n=0

anN (z w )n .

This coincides with (z w )N f (w ), and hence converges, on BR (w )\{w }. It also converges at z = w . Hence g (z ) converges on BR (w ), so is holomorphic on the disk, by Theorem 19.5, and by Corollary 19.8, g (n) (w ) g (n) (z ) = lim z w n! n! for all n. For the last equality we have used the fact that g is holomorphic on the disk, hence smooth, so all its derivatives are dierentiable and hence continuous. Applying this in the case n = N 1, we see that bn = anN = a1 = lim 1 d N 1 g (N 1) (z ) = lim N 1 (z w )N f (z ) . z w (N 1)! (N 1)! z w dz

22.4 Examples

(i) Let f (z ) =

z = i. Hence,

z+i zi

. By Lemma 21.6 this has a pole of order 2 at


2

d21 1 lim 21 res(f, i) = (2 1)! z i dz d (z + i)2 dz = lim 2(z + i) = 1. lim


z i z i

(z i)

z+i zi

= 2(i + i) = 4i. (ii) The function f (z ) = ee /z 2 has a pole of order 2 at z = 0, (again by Lemma 21.6). So, d ee d ez 1 z e = lim ez .ee = e. res(f, 0) = lim z 2 2 = lim z 0 dz z 0 1! z 0 dz z
102
z z

Method 4: Direct expansion We can use this as a last resort. We expand functions as power series, etc., and then calculate coecients. 22.5 Example Let f (z ) = zez . This has a pole of order 3 at z = 1. So, expand (z 1)3 about w = 1: (i.e. let z = w + h and take h as our variable), f (1 + h) = = = = = So res(f, 1) = 3e . 2 (1 + h)e1+h (1 + h 1)3 e.eh (1 + h) 3 h h2 h3 e (1 + h ) 1 + h + + + ... h3 2! 3! e 3 2 1 + 2h + h2 + h3 + . . . 3 h 2 3 2e 3e 2e e + 2+ + + ... 3 h h 2h 3

Summary Simple pole: res(f, w ) = limz w (z w )f (z ). Some quotients: Suppose p(w ) = 0, q (z ) = 0 and q (w ) = 0. Then, res Pole of order N : res(f, w ) = p ,w q = p (w ) . q (w )

1 d N 1 lim N 1 (z w )N f (z ) . (N 1)! z w dz

Calculate expansions directly, equate coecients, etc.


103

23

Cauchys Residue Formula

The next theorem shows that we can calculate certain complex integrals by merely calculating the residues and winding numbers at poles. This gives us a very simple method for calculating integrals that may, at rst sight, seem intractable. 23.1 Theorem (Cauchys Residue Formula) Let a closed contour and f be a function holomorphic on Int( ) except for poles at p1 , . . . , pm Int( ). Then
m

f (z ) dz = 2i
j =1

n(, pj ) res(f, pj ).

Proof:

Consider the following sketch.

1 1 p
1

C1 C2 p2 2 C3 p3 2

C4 p
4

We can surround each pole by a disk of radius > 0, chosen small enough so that Each disk lies in the interior of . f (z ) has a convergent Laurent expansion on each punctured disk B (pj )\{pj }. The disks dont overlap, B (pj ) B (pk ) = if j = k . Let Cj be the circle of radius /2 > 0 centred on pj , taken clockwise. There exist contours j staying within Int( ) which join each Cj to the contour (see diagram). Lets denote by j the part of from the start point of j to the start point of j +1 . So = 1 m . Next, we dene a contour by = 1 n(, p1 )C1 rev (1 ) 1 2 n(, p2 )C2 rev (2 ) 2 . . . m n(, pm )Cm rev (m ) m . By n(, pj )Cj we mean Cj traversed n(, pj ) times if n(, pj ) > 0, and rev (Cj ) traversed n(, pj ) times if n(, pj ) < 0. By construction, this is a path such that f is

104

holomorphic on Int( ). So by Cauchys Theorem we have f = n(, p1 )


e m C1

f = 0. But f+ f
1 m

f + n(, p2 )
C2

f + + n(, pm ) f.

Cm

0 =
j =1

n(, pj )
Cj

+
1 m

Note that the integrals over all the j cancel. So, as = 1 m we get,
m

f=
j =1

n(, pj )
rev (Cj )

f.

It remains to compute rev (Cj ) f for each j . Let Nj be the order of pole pj . Recall we have a Laurent expansion for f valid on a punctured disk containing Cj , so

f=
rev (Cj ) rev (Cj ) n=N j

an (z pj )n dz

().

Now |z pj | = /2 on Cj , so each term in the series has modulus bounded above n by Mn = |an |(/2)n . But n=0 an (z pj ) has radius of convergence at least , and hence converges absolutely for |z pj | = /2. Hence

Mn = nite sum +
n = Nj n=0

|an (/2)|n

converges. Hence, by the Weierstrass M Test (Theorem 15.12), the Laurent series

n = Nj

an (z pj )n

converges uniformly on Cj to f . Hence, by Theorem 15.10, we may interchange the order of integration and summation in ():

f=
rev (Cj ) n = Nj

an
rev (Cj )

(z pj )n dz = 2ia1

since only the n = 1 term contributes, by the Fundamental Example. Now recall that a1 = res(f, pj ), by denition. There we have it, one of the best theorems in mathematics! 23.2 Example Evaluate the integral 1 dz, z 2 + (1 i)z i

where is a square with sides of length 4 centred at the origin, oriented anti-clockwise.
105

Solution: rst nd the integrands poles: f (z ) = z2 1 1 = , + (1 i)z i (z i)(z + 1)

has simple poles at z = 1 and z = i. Both these lie inside the contour:

Im

i -1
Re

We can use Method 1 to nd their residues: res(f, 1) = 1 1 = z 1 z 1 z i 1+i 1 1 res(f, i) = lim(z i)f (z ) = lim = . z i z i z + 1 1+i lim (z (1))f (z ) = lim

We have n(, 1) = n(, i) = 1, so by Cauchys Residue Formula

1 dz = 2i ( res(f, 1) + res(f, i)) z 2 + (1 i)z i 1 1 = 2i + 1+i 1+i = 0.

Actually, we could have deduced this without computing any residues! To see this, let R be the circle of radius R > 1, that is R (t) = Reit , 0 t 2 . Since R winds around the poles of f the same number of times as does, it follows from Cauchys Residue Formula that f = 2i( res(f, 1) + res(f, i)) =
R

()

for all R > 1.

106

i -1

R
x

The RHS of equation () is independent of R, so it follows that of R (provided R > 1!), and hence f = lim
R

f is independent

f.
R

Now, for all z R , |z | = R, so | f (z )| = 1 1 . |z + i||z 1| (R 1)2 2R 1 = 2 (R 1) (R 1)2

Hence, by the Estimation Lemma, |


R

f | L(R ) f = 0,

lim

whence the result follows.


(z ) where p and q are polynomials and deg p deg q 2. Let 23.3 Exercise Let f (z ) = p q (z ) z1 , z2 , . . . , zm be the poles of f . Prove that m

res(f, zj ) = 0.
j =1

Summary Let be a closed contour and f be holomorphic on Int( ) except for poles at{p1 , . . . , pm } Int( ). Then
m

f (z ) dz = 2i
j =1

n(, pj ) res(f, pj ).

107

24

Evaluation of Real Integrals

We shall now see that complex analysis allows us to solve, in a simple manner, problems involving real functions that are not easy to solve with real methods. Integrals over the whole real line Lets say we want to evaluate the improper integral
a 0

f (x)dx = lim
R

a 0

f (x)dx
0 R

lim

f (x)dx.
b

()

Clearly, since R = R + 0 , if the sum of limits () exists (i.e. if each of the two individual limits exists), it coincides with
R R

lim

f (x)dx
R

().

Note that the converse is false! (An obvious counterexample is f (x) = x, for which () does not exist, while () = 0.) So if the improper integral () exists, we can thus use () to calculate it. This allows us to employ an immensely powerful trick. Note that
R

f (x)dx
R

can easily be reinterpreted as a contour integral, provided f (x) extends naturally to a complex function f (z ). Let R : [R, R] C such that R (t) = t. Then
R R f (R (t))R (t)dt =

f (z )dz =
R R

f (t)dt.
R

Now let R : [0, ] C be the semicircular arc from R to R in the upper half plane, R (t) = Reit . Then R = R R is a closed contour, so we have lots of slick methods to compute f=
R R

f+
R

f,

e.g. Cauchys Theorem, Cauchys Integral Formula, Cauchys Residue Formula.

Im

iR

The contour R

-R

Re

108

Imagine we can compute this. Now take the limit R . On the LHS we have a known number. On the RHS we have the integral we want,

f (x)dx = lim

f,
R

plus the limit of R f . Our strategy will be to show that this last limit vanishes by using the Estimation Lemma (note that |z | = R on R , so we just need the integrand to get small suciently fast at large |z |). Time for an example.

24.1 Example Compute

1 dx. (x2 + 1)2 1 1 = 2 2 + 1) (z i) (z + i)2

The integrand extends to f (z ) = (z 2

which is holomorphic on the whole of C except at i and i where it has poles of order 2.
y

i
x

-i

Consider the closed semicircular contour R , as dened above, with R > 1. Of the two poles, only i Int(R ), so we will need only res(f, i). By Method 3, this is res(f, i) = lim 1 2 1 d 2 d (z i)2 f (z ) = lim = lim = = . z i dz (z + i)2 z i (z + i)3 z i dz (2i)3 4i

Hence, by Cauchys Residue Formula, f = 2i res(f, i) =


R

. 2

Note that this is actually independent of R, as we should expect. Hence


(x2

1 dx = + 1)2 =

lim

f
R

f
R

lim

2
109

( )

But for all z R , |z | = R, so | f (z )| = so


R

|z 2

1 1 , 2 2 + 1| (R 1)2

by the Estimation Lemma, and hence

L(R ) R = 2 2 2 (R 1) (R 1)2 f = 0.
R

lim

Substituting into (), we see that


1 dx = . (x2 + 1)2 2

Could you have computed this integral using only real methods? 24.2 Example Heres a slightly more subtle example:

x2

cos x dx. 2x + 2

The trick here is to dene f (z ) = eiz eiz = z 2 2z + 2 (z 1)2 + 1

which is holomorphic on C except for simple poles at z = 1 + i and z = 1 i. Youll see why shortly.

Im

R
1+i

R R
The residue of f at 1 + i is res(f, 1 + i) =
1-i

Re

eiz i = e1+i z 1+i 2z 2 2 lim


110

by Method 2. Dening R as before, we see that for all R > f = 2i res(f, 1 + i) = e1+i .
R

2,

Now, for z R , |z | = R and y = Im(z ) 0, so | f (z )| = |ei(x+iy) | ey 1 1 = . 2 2 2 | z 2 z + 2| |(z 1) + 1| | z 1| 1 (R 1)2 1 f| L(R ) R = 0 as R . (R 1)2 1 (R 1)2 1
R

Hence, by the Estimation Lemma | Thus lim g = e1+i .


R

( )

So what? Well
R

lim

g=
R

g (x)dx =

cos x + i sin x dx x2 2x + 2

so the integral we seek is just the real part of ():


x2

cos x dx = Re(ei1 ) = cos 1. 2x + 2 e

Try computing that with only real methods! Note that by equating imaginary parts we get the bonus information that

x2

sin x dx = Im(ei1 ) = sin 1, 2x + 2 e

which we werent even looking for! What a method! 24.3 Remark The obvious approach to Example 24.2 would be to dene cos z g (z ) = 2 z 2z + 2

and compute R g . The problem with this is that | cos z | is unbounded on the upper 1 iz half plane (i.e. the region with Imz > 0). This is because cos z = 2 (e + eiz ) and |eiz | = |ei(x+iy) | = |eix ||ey | = ey

which grows large when y is large and positive. So you cant use the Estimation Lemma to show lim | g| = 0
R R

(indeed, this equation isnt true!), and knowing the real integral you want.
111

g does not allow you to deduce

Integrands of the general type f (z ) = p (z ) eiaz q (z )

where p, q are polynomials and a is a real constant occur frequently in complex analysis. It is rather tiresome to have to estimate R f afresh for each of these using the Estimation Lemma. We will now prove Jordans Lemma, which gives conditions on p, q and a which guarantee that the contribution from the large semicircular arc vanishes as R . To do this, well need the following Lemma, whose proof was a homework exercise: 24.4 Lemma (Polynomial Growth Lemma) Let p be a polynomial of degree n. Then there exist constants R > 0 and c2 c1 > 0 such that c 1 | z | n | p (z )| c 2 | z | n for all |z | > R. 24.5 Lemma (Jordans Lemma) Suppose p and q are polynomials, a is a real number, and R (t) = Reit , 0 t . If (i) a 0 and deg(p) deg(q ) 2, or (ii) a > 0 and deg(p) deg(q ) 1 then
R

lim

p(z ) iaz e dz = 0. q (z )

(z ) iaz Proof: Let f (z ) = p e , m be the degree of p(z ) and n be the degree of q (z ). By q (z ) the Polynomial Growth Lemma, there exist constants R1 > 0 and c1 , c2 > 0 such that

| p (z )| c 2 | z | m

and

| q (z )| c 1 | z | n

for all |z | > R1 . Hence, if R > R1 then for all z R , |z | = R > R1 , so c2 p (z ) . q (z ) c1 R n m Consider case (i): n m 2 and a 0. For all z R , y = Im(z ) 0, so |eiaz | = eRe(iaz ) = eay 1, since a 0. Hence, for all z R , if R > R1 , | f (z )| so by the Estimation Lemma, f L(R ) c2 c2 = 0 n m c1 R c1 Rnm1
112

c2 c1 R n m

as R

since n m 2. Case (ii) is more subtle. Instead of using the Estimation Lemma we must use the triangle inequality for complex Riemann integrals:

f
R

=
0

f (Reit )iReit dt R
0 2t

|f (Reit )|dt = R 2 c2 c1 Rnm1


2

p(Reit ) aR sin t e dt q (Reit )

c2 c1 R n m

eaR sin t dt =

eaR sin t dt.

], sin t Now for all t [0, 2

(draw a graph!), so

f
R

2 2 c2 2 c2 e2aRt/ dt < n m 1 c1 R c1 Rnm1 0 c2 2 c2 = 0 = n m 1 c1 R 2aR ac1 Rnm

e2aRt/ dt (exists as a > 0)


0

as R , since n m 1. 24.6 Exercise Use Jordans Lemma and the method of Example 24.2 to evaluate

cos 137x dx x2 + 1

and

x sin 2x dx x2 + 1

If the poles of a function are very symmetrically placed on the plane, it often pays to exploit this symmetry and use a more subtle contour than R . 24.7 Example Compute 1 dx. 10 1 + x The obvious method is to use the semicircular contour R as before, with f (z ) = 1 . This will work, but is quite heavy going. Note that f has 10 poles, 1 + z 10 1 + z 10 = 0 z 10 = 1 = ei z = e(2k+1)i/10 , k Z,

of which 5 lie in the interior of R . So we would have to compute the residues res(f, ei/10 ), res(f, e3i/10 ), res(f, e5i/10 ), res(f, e7i/10 ), res(f, e9i/10 ), using this method. Heres a cleverer method. First note that, since f (x) is even,

1 dx = 2 1 + x10

1 dx 1 + x10

so its enough to compute


R

lim

f
R

113

where R (t) = t, 0 t R. Now let R (t) = ei/5 R (t). Geometrically this is the line R rotated /5 anticlockwise about the origin. Now
R bR R f (R (t))R (t)dt = ei/5 0 f (R (t))R (t)dt = ei/5

f=
0

f
R

, be the arc joining the since f (R (t)) = f (R (t)). Now let R (t) = Reit , 0 t 5 ends of R and R , and dene the closed contour R = R R rev (R ).

Then f is holomorphic on R Int(R ) except for a single simple pole at ei/10 , whose residue is 1 1 ei/10 res(f, ei/10 ) = = = , 10z 9 z =ei/10 10ei9/10 10 by Method 2. Hence, by Cauchys Residue Formula, f = i i/10 e , 5

for all R > 1. Now f=


R R

f+
R

bR

f = (1 ei/5 )

f+
R R

f,

For all z R , |z | = R, so | f (z )| = and hence


R

1 1 10 10 |1 + z | R 1
R 5 10 R

114

as R . Hence

i i/10 1 e = (1 ei/5 ) dx, 5 1 + x10 0 which can be solved for the integral we need. Note that

1 ei/5 = ei/10 (ei/10 ei/10 ) = 2iei/10 sin So


0

. 10

1 1 dx = 1 + x10 10 sin 10 1 1 dx = 1 + x10 5 sin 10

and nally

Trigonometric integrals Cauchys Residue Formula also gives a very slick way to compute integrals of the form
2

f (cos , sin ) d.
0

24.8 Theorem Let (t) = eit , for 0 t 2 . Then


2 0

f (cos , sin ) d = i

z + z 1 z z 1 , 2 2i

z 1 dz.

Proof:

Let z = ei = (). Then cos =

() = iei . Thus, i f

z + z 1 z z 1 and sin = . We also have 2 2i

z + z 1 z z 1 , 2 2i

z 1 dz = i
2

f (cos , sin )ei iei d


0

=
0

f (cos , sin ) d.

24.9 Remark We can change the limits of integration by letting (t) = eit where a t a+2 a + 2 and a R. So we get integrals of the form a f (cos , sin ) d. A common example is a = , so we get .
2

24.10 Example Evaluate the real integral


0

1 dt. 13 + 12 cos t

115

Solution: Applying the theorem we get, for (t) = eit , 0 t 2 ,


2 0

1 dt = i 13 + 12 cos t = i

1 dz z (13 + 6 (z + 1/z )) 1 dz. 2 6z + 13z + 6

Let g (z ) =

6z 2

1 . This has poles at 3 and 2 . 2 3 + 13z + 6

Im

-3 _ 2

-2 _ 3

Re

lies within the unit circle given by . Hence, we calculate the Of these, only 2 3 residue for this pole: res g, 2 3 = 1 12z + 13 =
z =2/3

12

2 3

1 1 = . 5 + 13 + 13

Therefore, by Cauchys Residue Theorem,


2 0

1 1 2 dt = i 2i = . 13 + 12 cos t 5 5

116

Summary To compute

f (x)dx, for suitable f :

(i) Dene R = R + R by R (t) = t for R t R, and R (t) = Reit for 0 t . Take R large enough to include all the poles of f that lie in the upper half-plane. (ii) Calculate
R

f (z ) dz using Cauchys Residue Theorem.

(iii) Show that R f (z ) dz 0 as R , using the Estimation Lemma (or Jordans Lemma, if applicable). (iv) Relate f (x) dx to R f (z ) dz . To compute
2 0

f (cos , sin )d:

(i) Let (t) = eit , for 0 t 2 . (ii) Then


2 0

f (cos , sin ) d = i

z + z 1 z z 1 , 2 2i

z 1 dz.

(iii) Use Cauchys Residue Formula.

117

25

Advanced techniques of contour integration

Indentation around a simple pole Sometimes a naive application of contour integration methods leads us to consider integrals where the integrand has one or more poles on the contour itself. Strictly speaking, such contour integrals are ill-dened, since the integrand isnt continuous on the contour. If were careful, we can sometimes indent the contour by a circular arc skirting around the pole, then take the limit as the radius of this arc goes to zero. The next lemma is the key technical result. 25.1 Lemma (Indentation Lemma) Let f have a simple pole at w C and (t) = w + eit , 1 t 2 . Then
0

lim

f = i(2 1 ) res(f, w ).

Proof:

Let b = res(f, w ), and note that b dz = b zw lim


2 1

1 ieit dt = i(2 1 )b eit b zw

so it suces to show that


0

f (z )

dz = 0.

Choose and x > 0. By Method 1 we know that b = lim (z w )f (z ),


z w

so there exists > 0 such that for all 0 < |z w | < , , |(z w )f (z ) b| < 2 1 that is, f (z ) b zw b < . zw (2 1 )|z w | L() sup{ f (z ) (2 1 ) b : z } zw

Hence, by the Estimation Lemma, for all (0, ),

f (z )

dz

= . (2 1 )

Notes: (i) In the case where is closed (2 1 = 2n , n Z+ ) this is just a special case of Cauchys Residue Formula.
118

(ii) The lemma only works if w is a simple pole. If w is a pole of order N 2, the limit lim
0

f (usually) doesnt exist because f (z ) blows up too fast as z

approaches w .

25.2 Example Compute the improper integral

sin x dx. x 0 First, what does this improper integral really mean? Answer:
0

sin x dx = lim lim R 0 x

sin x dx x

assuming this double limit exists. Since the integrand is an even function,
R

sin x 1 dx = x 2

sin x dx + x

sin x dx x

1 = Im 2

f+
1 2

where f (z ) = eiz /z and 1 , 2 are the contours depicted below. Let (t) = eit , 0 t (a semicircular arc of radius centred on 0, traversed anticlockwise). Then = 1 rev ( ) 2 R is a closed contour.

R rev() 1 2

Now f is holomorphic except at 0 where it has a simple pole. Since 0 is not on or in its interior, we know from Cauchys Theorem that f =0

for all 0 < < R. Hence f+


1 2

f=

f,
R

so 2
0

sin x dx = Im lim 0 x

f lim

f
R

119

By Jordans Lemma,

f 0 as R , and by the Indentation Lemma, f = i res(f, 0) = i lim zf (z ) = i.


0 z 0

lim

Hence

sin x dx = . x 2

Exact summation of (some) series Using the comparison tests you learned in Analysis 1, its not hard to show that series such as 1 n2 , and n2 n6 + 4 n=1 n=1 converge. But to what? In this section well show how contour integration can answer such questions. The contour of interest, call it N , is a square, with vertices 1 1 (N + 2 )(1 + i), (N + 1 )(1 + i), (N + 2 )(1 i) and (N + 1 )(1 i), traversed once 2 2 anticlockwise, where N is a positive integer. We begin by proving that the function cot z is bounded on N .
(N+ 1 ) (-1+i) 2 (N+ 1 )(1+i) 2

(N+ 1 ) (-1-i) 2

(N+ 1 ) (1-i) 2

25.3 Lemma (cot boundedness lemma) Let N Z+ and N be the contour depicted above. Then there exists a constant C > 0, independent of N , such that | cot z | C for all z N .

Proof: First note that cot z is an odd function, so it suces to prove the inequality for all z on the top horizontal and right vertical sides of N . Now | cot z | = eiz + eiz |eiz | + |eiz | . eiz eiz ||eiz | |eiz ||
1

1 ), so On the top horizontal side of N , z = x + i(N + 2

|eiz | = e(N + 2 )

and
120

|eiz | = e(N + 2 ) .

Hence | cot z | e(N + 2 ) + e(N + 2 ) e


) (N + 1 2
1 1

1 (N + 2 )

1 = coth( (N + )) 2

t where coth t = cosh . Its easy to show that the derivative of coth t is negative for all sinh t t 0, so coth t is a decreasing function, and hence

3 | cot z | coth 2 for all z on the top edge of N . 1 On the right vertical side of N , z = (N + 2 ) + iy , so cot z =
1 1 ) + iy ) ) cos iy sin (N + 2 ) sin iy cos( (N + 1 cos (N + 2 2 = 1 1 1 sin( (N + 2 ) + iy ) sin (N + 2 ) cos iy + cos (N + 2 ) sin iy sin iy = cos iy

) = 0 and sin (N + 1 ) = (1)N . Hence, since cos (N + 1 2 2 | cot z | = | tan iy | = | tanh y | 1


3 }, the lemma is for all z on the right vertical side of N . Taking C = max{1, coth 2 proved.

25.4 Example We can now prove that Consider the function


n=1

1 2 = . n2 6

cot z . z2 You showed, in a homework exercise, that this has simple poles at z = n Z, n = 0, and a triple pole at z = 0. Further, you showed that f (z ) = res(f, n) =
1 n2 2 3

n = 0, n = 0.

Now for each N Z+ consider N f . The interior of N contains the poles at {n Z : N n N }. By Cauchys Residue Formula,
N 1

f = 2i
N j =N

res(f, j ) = 2i
j =N

2 1 + j2 3

j =1

1 j2

1 2 = 2i + 2 3 j2 j =1

Hence
N

j =1

2 1 1 = + 2 j 6 4i
121

f
N

().

Now, for all z N , |z | > N and | cot z | C , C constant (independent of N ) by the cot boundedness lemma, so | f (z )| Hence, by the Estimation Lemma, | f | L(N )
N

C . N2

C 4(2N + 1)C = 0 2 N N2

as N , so limN

f = 0. Hence

j =0

1 2 = . j2 6

25.5 Remark Many other series n=1 (n) can be handled using the same trick. We need to be an even function ((z ) = (z )) which decays at large |z | faster than |z |1 . Then n=1 (n) can be deduced from (z ) cot z dz.
N

Alternating series can also be treated similarly. To compute considers (z ) cosec z dz.
N

n n=1 (1) (n),

one

25.6 Exercise Let a > 0 be a positive constant. By considering cot z dz z 2 + a2 coth a 1 a (a)2
2 6

show that

n=1

2 1 = n2 + a2 2

Note that the expression on the right tends to expect.

in the limit a 0, as one would

122

Summary Improper contour integrals where a simple pole lies on the contour can be handled by indentation. For suitable functions (n), the series

(n)
n=1

and
n=1

(1)n (n)

can be summed exactly, by considering the contour integrals (z ) cot z dz


N

and
N

(z ) cosec z dz

where N is a certain square contour.

123

26

Proof of Cauchys Theorem

In this course we have proved a long chain of powerful and fundamental theorems (Cauchys Integral Formula, Liouvilles Theorem, The Maximum Modulus Principle, The Fundamental Theorem of Algebra, Taylors Theorem, Liouvilles Theorem and Cauchys Residue Formula). They all rely on Cauchys Theorem, which we stated but have not yet proved. The aim of this nal section is to ll in this gap. First, lets recall the statement of Cauchys Theorem: Cauchys Theorem Let D C be an open set and f : D C be holomorphic. Let be a closed contour such that and its interior points lie in D . Then f = 0. We begin by proving the result for a triangle. By a triangle, we shall mean a closed triangular region, including its boundary. Given a triangle T we shall denote its boundary, thought of as a closed contour traversed once anticlockwise, T . We will repeatedly use the following geometric fact, which you should be able to see by drawing a picture: given any pair of points z, w T , 1 |z w | length of longest side of T L(T ). 2 26.1 Lemma (Nested Triangles Lemma) Suppose that T1 T2 T3 . . . is a nested sequence of triangles which are shrinking, in the sense that L(Tn ) 0 as n . Then n=1 Tn = . That is, there exists a complex number w such that w Tn for all n Z+ . Proof: For each n Z+ choose zn Tn . The rst step is to prove that this sequence is Cauchy. Fix any > 0. Since L(Tn ) 0, there exists p Z+ such that L(Tp ) < . Then, for all n, m p, zn Tn Tp and zm Tm Tp , so |zn zm | length of longest side of Tp < . 2

Hence zn is Cauchy. By the General Principle of Convergence (a complex sequence converges if and only if it is Cauchy), zn converges, to w say. I claim that w Tn for all n Z+ . Assume this is false. Then there exists p Z+ such that w C\Tp . Now C\Tp is an open set, so there exists > 0 such that B (w ) (C\Tp ). But for all n p, zn Tn Tp , so zn / B (w ). That is, |zn w | . This contradicts the fact that zn w . 26.2 Theorem (Cauchys theorem for a triangle) Let f be holomorphic on an open set containing a triangle T . Then T f = 0. Proof: Let I = T f . Dissect T into four similar triangles T (1) , T (2) , T (3) , T (4) by joining the midpoints of the sides of T (see diagram). Note that T (j ) each have length exactly one half L(T ).

124

T
T (2) T (4) T (3)

(1)

Then I=
T

f=
j =1 T (j )

as the internal contributions cancel and the external contributions add up. So
4

|I | Hence there exists j such that

f .
j =1 T (j )

1 f |I |. 4 T (j ) Set T1 = T (j ) . Repeat, quadrisecting T1 . We obtain a shrinking nested sequence of triangles T1 T2 . . . such that
Tn

1 4

|I |

and

L(Tn ) =

1 2

L(T ).

By Lemma 26.1 there exists z0 there exists > 0 such that 0 < |z z0 | <

Tn . Take > 0. Since f is holomorphic at z0 , f (z ) f (z0 ) f (z0 ) < . z z0

Choose m so large that L(Tm ) < . Then for all z Tm , |z z0 | < L(Tm )/2 < , so |f (z ) f (z0 ) f (z0 )(z z0 )| |z z0 | L(Tm ). Therefore, by the Estimation Lemma
Tm

f (z ) f (z0 ) f (z0 )(z z0 ) dz L(Tm )L(T ) =

1 4

L(T )2 .

1 Now f (z0 ) + f (z0 )(z z0 ) has an antiderivative, e.g. f (z0 )z + 2 f (z0 )(z z0 )2 , so by the Fundamental Theorem of Calculus (Theorem 13.1) its integral round any closed contour is zero. Therefore,

Tm

f (z ) f (z0 ) (z z0 )f (z0 ) dz =

Tm

f (z ) dz

1 4

|I |.

Thus, |I | L(T )2 . But this is true for all > 0. Hence I = 0.


125

26.3 Denition A subset D C is star shaped if there exists c D such that for all z D , the straight line segment joining c to z lies in D . The point c is called a centre of D (note it may not be unique). 26.4 Example (i) D = C is star-shaped (any c C is a centre).

(ii) D = C\{x R : x 0} is star shaped. For example c = 1 is a centre (in fact any positive real number is a centre). (iii) D = C\{0} is not star shaped. To see this, note that given any c D , z = c is also in D , but the straight line segment joining z and c contains 0, which is not in D . 26.5 Theorem (Existence of antiderivatives on star-shaped domains) Let D C be an open star shaped set, and f : D C be holomorphic. Then there exists a holomorphic function F : D C such that F = f . Proof: Let c be a centre of D , and for each z D denote by z the straight line contour starting at c and ending at z . For each z D , dene F (z ) by F (z ) =
z

f.

This is well-dened since z D (D s star shaped) and f is holomorphic on D . We claim that the function F : D C so dened is holomorphic, with derivative f . To see this, x z D and consider the triangular contour depicted below.

D z

rev( ) z+h c

If |h| is suciently small, then all points on , the segment from z to z + h, are in D (since z D and D is open), and hence, since D is star shaped, and its interior lie

126

in D . Hence, by Cauchys Theorem for Triangles, f = 0

F (z + h) F (z )

f = 0
1

F (z + h) F (z ) = F (z + h) F (z ) = h

f (z + th)h dt
0 1

f (z + th)dt,
0

where we have used the parametrization (t) = z + th, 0 t 1. Now the family of functions fh : [0, 1] C, fh (t) = f (z + th), converges uniformly to the constant function f0 (t) = f (z ) as h 0, by Lemma 19.4. Hence, by Theorem 15.10 F (z + h) F (z ) = h0 h lim
1

f (z ) dt = f (z ).
0

Recall that if f : D C has an antiderivative F and : [a, b] D is a closed contour, then

f = F ( (b)) F ( (a)) = 0

by the Fundamental Theorem of the Calculus (Theorem 13.1). So Cauchys Theorem in the case where f is holomorphic on a star shaped open set containing follows immediately from Theorem 26.5. Proof of Cauchys Theorem: Let K = Int( ); note that this is a subset of D . Let = dist(K, C\D ). Draw a mesh of squares on C of side /4 and let Q1 , . . . , QN j = Qj K and j = Q j . Then each closed contour be the ones meeting K , let Q j lies in an open square in D , by our choice of . But squares are star shaped, so f has an antiderivative on j , by Theorem 26.5. Hence j f = 0 by the Fundamental Theorem of the Calculus (Theorem 13.1) since j is closed. But
N

f=
j =1 j

f =0

since contributions from interior edges cancel.

127

Advice on exam preparation


This is a 20 credit module in pure mathematics. Hence, you are expected to be able to recall and understand the statement of every theorem, proposition, lemma, method etc. in the course. Furthermore, with a few exceptions which we will list below, ALL THE PROOFS ARE EXAMINABLE! Most of the proofs are, if you understand the basic denitions, quite straightforward. Just ask yourself: (i) What do I know (i.e. what do the hypotheses tell me)? (ii) What am I supposed to deduce? If you can sort that out, a sensible proof strategy usually suggests itself. Of course, there may be a clever idea required, and its a good idea to learn (memorize) what the clever idea is. Usually, the clever idea can be summed up in a picture. It is probably not a good idea to try to learn the proofs verbatim without really understanding them. Its extremely unlikely that you will get down on paper exactly what we wrote in the notes, and youre quite likely to write complete nonsense. The scope of the paper will be similar to last year. As always, a substantial portion of the marks can be gained by accurately recalling denitions. Of course, if you dont know the denitions, the theorems, examples etc. are pretty meaningless and youre very unlikely to pass the exam. So our number one piece of advice is LEARN ALL THE DEFINITIONS! A few of the proofs in the course are too intricate to be examinable. Heres a complete list: Theorem 8.4 (the Integrability Criterion) Theorem 8.5 All results from Semester 1 in the Appendix (numbered 0.x) Theorem 15.12 Theorem 17.1 (i.e. all proofs in section 26) Theorem 18.1 Theorem 19.10 Theorem 20.2 Theorem 23.1 Part (ii) of Lemma 24.5 We emphasize once again that the accurate statement and application of all these results certainly is examinable.

128

Das könnte Ihnen auch gefallen