Professional Documents
Culture Documents
singularities
Course 414, 200304
February 9, 2004
where, for any choice of r with R1 < r < R2 and r : [0, 1] C given by r (t) = a+r exp(2it)
Z
1 f (z)
an = dz
2i r (z a)n+1
Proof. Recall from 1.9 that a Laurent series has an annulus of convergence. In this case the
annulus of convergence must include R1 < |z a| < R2 .
Turning to the proof, notice that from Cauchys theorem we can conclude that the formula
given for an is independent of r in the range R1 < r < R2 . Fix any z with R1 < |z a| < R2
and choose r1 , r2 with R1 < r1 < |z a| < r2 < R2 . From the winding number version of
Cauchys integral formula (1.30 with = r2 r1 ) we can deduce
Z Z
1 f () 1 f ()
f (z) = d d.
2i r2 z 2i r1 z
The remainder of the proof is quite like the theorem that analytic function in a disk are repre-
sented by power series (Theorem 1.23).
For | a| = r2 , we do exactly as in the power series case
1 1 1 1 1 1
= = za =
z ( a) (z a) a 1 a a1w
za |za|
where w = a
has |w| < r2
= 2 < 1. Hence
n
(z a)n
1 1 X n 1 X za X
= w = = n+1
.
z a n=0 a n=0 a n=0
( a)
1
2 414 200304 R. Timoney
a
and this time w = za is also less than one in modulus (for all on | a| = r1 ). In fact
|w| = r1 /|z a| = 1 < 1. Thus we get another series
n
( a)n ( a)m1
1 1 X n 1 X a X X
= w = = n+1
. = m
.
z z a n=0 z a n=0 za n=0
(z a) m=1
(z a)
Plugging these two series into the the two integrals in the integral formula for f (z) and
exchanging the order of integral and sum (using uniform convergence to justify the exchange)
we get
Z ! Z !
X f () n
X
m1 1
f (z) = n+1
d (z a) + f ()( a) d
n=0 r2 ( a) m=1 r1 (z a)m
Definition 4.2 If an analytic function f is analytic on an open set G C that includes a punc-
tured disk D(a, r) \ {a} of positive radius r > 0 about some a C, then a is called an (isolated)
singularity if a is not itself in G.
The residue of f at an isolated singularity a of f is defined as the coefficient a1 in the
Laurent series for f in a punctured disk about a:
Z
1
res(f, a) = a1 = f (z) dz
2i |za|=
Proof. We apply the winding number version of Cauchys theorem 1.30 to a new chain 1
constructed as follows. Choose > 0 so small that (i) D(j , ) G and (ii) < |j k | for
j 6= k and 1 j, k n. Let j be the circle |z a| = traversed Ind (j ) times anticlockwise
and let
1 = + 1 + 2 + + n .
Chapter 4 open mapping theorem, removable singularities 3
By 1.30, Z
f (z) dz = 0
1
and this means
Z n Z
X Z n
X
0 = f (z) dz + f (z) dz = f (z) dz + 2i(Ind (j ))res(f, j )
j=1 j j=1
Proof. We can apply Theorem 4.3 since Ind (w) = 0 for w C \ G by 1.43.
Corollary 4.5 (Residue theorem, simple closed curve version) Let G C be open and f (z)
a function analytic on G except perhaps for a finite number of (distinct) points 1 , 2 , . . . , n
G. Suppose that is a (piecewise C 1 ) simple closed curve in G \ {1 , 2 , . . . , n }, oriented
anticlockwise and with its interior contained in G. Then
Z X
f (z) dz = 2i res(f, j )
{j:1jn,j inside }
Proof. By definition, the outside of is the unbounded component of C\. Also Ind (w) = 0 for
w outside . Since and its inside is contained in G, w C \ G w outside Ind (w) = 0.
So we can apply the theorem (4.3) to .
By definition of anticlockwise, IndR (j ) = 1 if j is inside and Ind (j ) = 0 if j outside.
Thus the formula of Theorem 4.3 for f (z) dz reduces to the one above in this situation.
4 414 200304 R. Timoney
Remark 4.6 The residue theorem can be used to work out many integrals of analytic functions
along closed curves in C. It is only necessary to be able to work out residues (and winding
numbers, but in many examples winding numbers are easy to find).
To find, for example, the residue of
ez
f (z) =
(z 1)2 (z 2)
at z = 1, we can write it as
1 ez 1
f (z) = 2
= g(z)
(z 1) z 2 (z 1)2
where g(z) is analytic near z = 1. It follows that g(z) has a power series representation in some
disk about 1 (in fact in D(1, 1), the largest disk that excludes z = 2)
X g (n) (1)
g(z) = (z 1)n
n=0
n!
Hence
e(12)e
and the coefficient of (z 1)1 is the residue. So it is g 0 (1) = (12)2
= 2e.
The residue theorem is also an effective technique for working out certain real integrals. We
will not go into this in any detail, but here are some examples that give the flavour of the methods
used.
R 2 R
Consider x4x+1 dx. One can verify that the integral converges by comparison with 1 x12 dx
R 2
or x22+1 dx. Consider the complex analytic function f (z) = z4z+1 and the closed curve made
up of the real axis from R to R followed by the semicircle of radius R in the upper half plane
(say) from R back to R. We take R large (at least R > 1). By the residue theorem, we can work
out this complex integral and get 2i times the sum of the residues at z = ei/4 and z = e3i/4 .
Chapter 4 open mapping theorem, removable singularities 5
iR
R R
i
These residues are in fact 4i exp(3i/4) and 4
exp(i/4). 2i times their sum is / 2.
So if R denotes the semicircle of radius R, then
R
x2 z2
Z Z
dx + dz =
R x4 + 1 R
4
z +1 2
For large R, the integral over the semicircular part of the contour is at most the length of the
contour (R) times the maximum value of the integrand, or
z2 2
Z
R R
dz 0 as R
R z4 + 1 R4 1
eiz
Z
dz = 0
z
by Cauchys theorem.
6 414 200304 R. Timoney
iR
R R
Thus
R
eix eiz eiz eix
Z Z Z Z
dx + dz + dz + dx = 0
R x z z R x
One can show (but it is a bit trickier this time) that the integral around the big semicircle R tends
to zero as R , while (taking z = exp(i( t)), 0 t as a parametrisation of )
Z
eiz exp(i exp(i( t)))
Z
lim dz = lim (i) exp(i( t)) dt = i
0 z
0 0 exp(i( t))
and
R R
eix eix
Z Z Z
sin x
dx + dx = 2i dx.
R x x x
Taking limits, one finally ends up with
Z
sin x
dx = .
0 x 2
Definition 4.7 Let f (z) be analytic on some open set G C. A point a G is called a zero of
f if f (a) = 0. The point a is called a zero of f of multiplicity m (for m > 0 a positive integer)
if f (a) = 0 and f (j) (a) = 0 for 1 j < m but f (m) (a) 6= 0.
Equivalently, if we look at a power series representation of f is a disk about a, the first term
with a nonzero coefficient is the (z a)m term.
X
m m+1
f (z) = am (z a) + am+1 (z a) + = an (z a)n
n=m
p(z) = bn (z 1 )(z 2 ) (z n )
Thus p(z) has the roots j for 1 j n and we can say p(z) has at most n roots. We cannot
say it has exactly n because there may be repetitions among the j . If we group the like terms
Theorem 4.8 (Argument principle, simple version) Let G C be simply connected and
f : G C analytic and not identically zero in G. Let be an anticlockwise simple closed curve
in G and assume that f (z) is never zero on . Let 1 , 2 , . . . , n be the zeros of f that are inside
and let j have multiplicity mj (1 j n).
Then Z 0
1 f (z)
dz = m1 + m2 + + mn = N
2i f (z)
(= the total number of zeros of f inside counting multiplicities, something we will often denote
by N or N or N,f ).
Proof. There are some aspects of the statement that may require some elaboration. Since G is
simply connected, we can say that if w C \ G, then
Z
1
dz = 0
zw
by Cauchys theorem (in the form 2.3 for a simply connected G using 1/(z w) analytic in
G). Hence Ind (w) = 0 for all w C \ G and so the inside of (where the index is +1) must
be in G as well as itself.
Now together with its inside is a compact subset of C, contained in G. So, by the identity
theorem (Corollary 3.5), there can only be a finite number of zeros of f inside or on . We have
assumed there are non on itself.
Finally, each zero of f has a finite multiplicity (again by the identity theorem 3.1) since G is
connected and f is not identically zero.
These are all points that are implicit in the statement of the theorem (or without which we
would need further assumptions in order for the statement to make sense). The proof of the
theorem is essentially to apply the residue theorem and to show that the residue of f 0 /f at a zero
j is the multiplicity mj .
There is a small catch as the statement of the residue theorem we would like to rely upon
(4.5) is stated with a finite total number of singularities. In our case f 0 /f has singularities where
8 414 200304 R. Timoney
f (z) = 0 and there could be infinitely many such points in all of G. What we can do is shrink
G to a smaller open set H that still contains and its interior but where f has only finitely many
zeros.
One way to do that is to construct H so that its closure is compact and contained in G. Let
K denote the union of with its inside (already noted to be a compact subset ofSG). For each
z K we can find a z > 0 so that the closed disk D(z, z ) G. Then K zK D(z, z )
is an open cover of K and so has a finite subcover (K compact) K kj=1 D zj , zj . Take
S
G. By the identity theorem, f has only finitely many zeros in H, hence only finitely many in H.
Thus we can apply the residue theorem (4.5) to get
Z 0 n 0
f (z) X f
dz = 2i res , j
f (z) j=1
f
To complete the proof we show that if f (z) has a zero z = of multiplicity m then
res(f, ) = m. To do this, start with a power series for f in a disk about
X f (k) () X
f (z) = (z )k = ak (z )k
k=m
k! k=m
with am = f (k) ()/k! 6= 0 (and this expansion is valid in some disk |z | < ). We also have
X
0
f (z) kak (z )k1
k=m
Remark 4.9 One might ask why this theorem is called the argument principle. If we consider
the composition of f with the curve we get a new curve f in C \ {0} (because f (z) is never
zero on ). If we compute the index of this curve around the origin, we get
Z Z 0
1 1 1 f (z)
Indf (0) = dw = dz
2i f w 2i f (z)
(as can be seen by writing the integrals in terms of a parameter w = f (z) = f ((t))).
Thus the integral in the theorem is the number of times f (z) goes anticlockwise around the
origin as z goes around .
Theorem 4.10 (Open mapping theorem) Let G C be a connected open set and f : G C
analytic but not constant. Then for each open subset U G the image f (U ) is open in C.
(That means that forward images of open sets are open, while inverse images of open sets are
open by continuity.)
Proof. Fix U G open and w0 f (U ). Thus w0 = f (z0 ) for some z0 U . (This z0 may not
be unique, but fix one.) Now the function f (z) w0 has a zero at z = z0 . As G is connected and
f is not constant (f (z) w0 is not identically zero on G) the identity theorem (3.1) tells us that
this zero has a finite multiplicity m 1.
There must be some > 0 with f (z)w0 never zero for 0 < |z z0 | < (and D(z0 , ) G)
by the identity theorem again. Choose r < , r > 0 with D(z0 , r) U . Then, by the argument
principle (4.8) we must have
f 0 (z)
Z
1
dz = m
2i |zz0 |=r f (z) w0
Proof.
S
(i) Note that G is the union of its connected components G = iI Gi . Now each restriction
of f to a connected component Gi is injectiveSand analytic on the open set Gi . Hence, by
Theorem 4.10, f (Gi ) is open. Hence f (G) = iI f (Gi ) is open.
(ii) If f 0 (z0 ) = 0 for some z0 G, then we can use the arguments of the proof of Theorem 4.10
with m > 1. we find out that there is some > 0 so that for |w f (z0 )| < we have
N (w) = m > 0 solutions for f (z) w = 0 counting multiplicities and only looking at zs
inside a small disk D(z0 , r). We can make this claim as long as r > 0 is small enough and
then > 0 is chosen to depend on r.
Now f 0 (z) analytic but not identically zero in the connected component of z0 in G (reason:
if f 0 was identically zero there, then f (z) would have to be constant there and that would
mean it was not injective). So, by the identity theorem (3.1 applied to f 0 on the connected
component) we can choose r > 0 small enough that f 0 (z) is never zero for 0 < |z z0 | < r.
This means that for this or smaller r there are no multiple zeros of f (z) w because there
are no points were f 0 (z) = 0 except z = z0 and w = w0 . Hence if we take w 6= w0 there
are m > 1 distinct solutions of f (z) w = 0. This contradicts injectivity of f . So f 0 is
never zero.
(iii) Certainly the inverse map f 1 : f (G) G makes sense. To show it is analytic, we work
on each connected component Gi of G separately and consider the restriction fi of f to Gi
and the corresponding inverse fi1 : f (Gi ) Gi (which is the restriction of f 1 to f (Gi )).
In other words, we can deduce the result if we prove it for the case where G is connected
open.
So we assume from now on that G is connected. By the Open Mapping Theorem 4.10,
forward images f (U ) of open subsets of G are open. But this means that f 1 is continuous
because the inverse image under the inverse function
(f 1 )1 (U ) = f (U )
Chapter 4 open mapping theorem, removable singularities 11
Example 4.12 Since the exponential map exp : C C is analytic, if G C is any open set
where exp is injective (equivalently where it is not possible to have z1 , z2 G, z1 6= z2 and
z1 = z2 + 2n, n Z) then the restriction exp |G of the map to G has an inverse
(exp |G )1 : exp(G) G C
which is analytic. This will be a branch of log w for w exp(G) since exp (exp |G )1 (w) =
ww exp(G).
If we take G to be the strip G = {z C : < =(z) < } we have
and so the inverse function in this case is the principal branch Log w.
for some > 0 (that is if f is bounded in some punctured disc about a), then there exists an
analytic extension of f (z) to include z = a.
That is, if f : G C is analytic on G C open and a C \ G satisfies the hypotheses, then
there exists g : G {a} C analytic with g(z) = f (z)z G (and g(a) = limza f (z)).
12 414 200304 R. Timoney
where Z
1 f (z)
an = dz
2i |za|=r (z a)n+1
(any 0 < r < ). Let M = sup0<|za|< |f (z)| and estimate
1 |f (z)| M
|an | (2r) sup n+1
r n+1
= M rn
2 |za|=r |z a| r
We have this estimate for all small r > 0. If n < 0 (and so n > 0), let r 0+ to get |an | = 0
for all n < 0. Thus the Laurent series for f is in fact a power series
X X
f (z) = an (z a)n = n = 0 an (z a)n (0 < |z a| < )
n=
Proof. : If there is an extension, then limza f (z) exists in C and so limza (z a)f (z) = 0
(limit of a product).
: If limza (z a)f (z) = 0, we can repeat the estimate in the proof of the above Theo-
rem 4.13, with small changes. First fix > 0 and choose r > 0 small enough that |z a| r
|(z a)f (z)| < . Then we get,
1 |f (z)| |(z a)f (z)|
|an | (2r) sup n+1
= r sup n+2
r n+2 = n+1
2 |za|=r |z a| |za|=r |z a| r r
for all sufficiently small r > 0. Now if n < 1, then n + 1 > 0 and so letting r 0+ we get
an = 0 (n 2). For n = 1 we get |a1 | . Since > 0 is arbitrary, this means we must
have a1 = 0 also. Thus the Laurent series is a power series as before.
Definition 4.15 An isolated singularity z = a of an analytic function f (z) is called a pole of f
of order p if the Laurent series for f in a punctured disc about a has the form
ap ap+1 X
f (z) = + + = an (z a)n
(z a)p (z a)p1 n=p
Chapter 4 open mapping theorem, removable singularities 13
with p > 0 and ap 6= 0. (This last condition is to ensure that the term with (z a)p is really
there.)
An isolated singularity z = a of f is called a pole of f if it is a pole of some order p > 0.
An isolated singularity z = a of f is called an essential singularity of f if it is neither a pole,
nor removable.
Thus the Laurent series for f in a punctured disc about an essential singularity z = a has the
form
X
f (z) = an (z a)n (0 < |z a| < )
n=
where there are infinitely many n < 0 with an 6= 0. By contrast, for a removable singularity
z = a all the negative coefficients vanish (an = 0n < 0) and for a pole there is a nonzero finite
number of n < 0 with an = 0.
lim |f (z)| =
za
Proof. : If z = a is a pole, then the Laurent series for f in a punctured disk 0 < |z a| <
gives
X X 1
f (z) = an (z a)n = (z a)p an (z a)n+p = p
g(z)
n=p n=p
(z a)
P
where g(z) = n=p an (z a)n+p is analytic for |z a| < , p is the order of the pole and
g(a) = ap 6= 0. It follows that
|g(z)|
lim |f (z)| = lim =
za za |z a|p
: If limza |f (z)| = , then there exists > 0 with |f (z)| > 1 for 0 < |z a| < .
Thus g(z) = 1/f (z) is analytic in the punctured disc D(a, ) \ {0} and also bounded by 1 there
(|g(z)| 1 for 0 < |z a| < ). Thus by Theorem 4.13, g(z) can be defined at z = a to make
it analytic. In fact g(a) = limza g(z) = limza 1/f (z) = 0. The analytic function g must have
a zero of some finite multiplicity m > 0 at z = a (by the identity theorem 3.1 applied to g(z) on
D(a, ). Hence the power series for g about a is of the form
X
X
n m
g(z) = bn (z a) = (z a) bn (z a)nm = (z a)m h(z)
n=m n=m
with h(0) = bm 6= 0 and h analytic in a disc about a. Thus there is a r > 0 so that h(z) 6= 0
for all z D(a, r) and 1/h(z) is analytic in D(a, r). In D(a, r) we must have a power series
expansion
1 X
= cn (z a)n
h(z) n=0
14 414 200304 R. Timoney
and so
1 1 1 X
n
X
f (z) = = = cn (z a) = cn (z a)nm
g(z) (z a)m h(z) (z a)m n=0 n=0
is a Laurent series for f in 0 < |z a| < r. Thus f has a pole (of order m) at z = a.
Proof. Fix > 0 small and put S = f (D(a, ) \ {a}). If the closure of S is not all of C, choose
w0 C \ S. As C \ S is open there is a disc D(w0 , ) C \ S with radius > 0. Hence,
for z D(a, ) \ {a} we have |f (z) w0 | > . Thus g(z) = 1/(f (z) w0 ) is analytic in the
punctured disc D(a, ) \ {a} and bounded by 1/ there. Therefore it has a removable singularity
at z = a and
lim g(z) C
za
and so limza f (z) = w0 + g(a) C exists. Thus f has a removable singularity at z = a, again
a contradiction to the hypotheses.
This S must be dense in C.
and this was also called the Casorati-Weierstrass theorem. There is a way we can relate the two
versions of the theorem.
We say that a function f (z) has an isolated singularity at infinity if g() = f (1/) has an
isolated singularity at = 0. That means there is some R 0 so that f (z) is analytic for |z| > R
and g() is analytic for 0 < || < 1/R.
We say that a function f (z) with an isolated singularity at has a removable singularity at
if g() = f (1/) has a removable singularity at = 0. Similarly, we say that f has a pole
Chapter 4 open mapping theorem, removable singularities 15
and then
X
g() = f (1/) = an n
n=0
lim nj = K G.
j
Then H or G \ H.
The case H is not possible since H open would then imply D(, r) H for some
r > 0. Thus f analytic on D(, r) and this implies there are no poles of f in D(, r) nj /
D(, r)j. This contradicts being the limit.
On the other hand the case G \ H is also impossible. If G \ H, then is an isolated
singularity of f and there must be a punctured disc D(, r)\{} H. Now limj nj =
j0 such that j j0 implies nj D(, r). As nj is a pole of f , this forces nj = j j0
and contradicts the choice of the n as distinct.
Corollary 4.21 If f is meromorphic on G C open then we can list all the poles of f in a finite
or infinite sequence 1 , 2 , . . ..
Proof. If G = C, let Kn = D(0, n) and if G 6= C let
1
Kn = {z G : dist(z, C \ G) and |z| n}.
n
Here dist(z, C \ G) = inf wC\G |z w|. Clearly dist(z, C \ G) 0 (if G 6= C and when G = C
we could perhaps interpret it as ).
Now Kn is clearly bounded (Kn D(0, n)) and Kn is closed because its complement is
[ 1
C \ Kn = {w C : |w| > n} D w,
n
wC\G
Now by Lemma 4.20, there can only be a finite number of poles of f in K1 (or possibly
none). We can list the poles in K1 as a finite list 1 , 2 , . . . n1 . (Take n1 = 0 if there are no
poles in K1 .) Now there are also a finite number of poles in K2 . Let n1 +1 , n1 +2 , . . . , n2 be
those poles in K2 but not in K1 . In general let nj +1 , nj +2 , . . . , nj+1 be those poles in Kj+1 not
already in K1 K2 Kj .
In this way, we have constructed a complete list 1 , 2 , . . . of the poles of f in G.
Remark 4.22 We will have further use for these Kn and they have additional useful properties.
It is clear from the way they are defined that Kn Kn+1 for each n. In fact Kn is contained in
the interior of Kn+1 because z Kn implies
1 1
D z, Kn+1
n n+1
[
G= Kn .
n=1
Proof. Let H G be the subset where f is analytic. We cannot immediately apply the identity
theorem for analytic functions (3.1) to f on H as we have not assumed H connected.
Now if z1 , z2 H, then there is a continuous path [made up of finitely many straight line
segments] in G and joining z1 to z2 (connected open sets are path connected). The path is a
compact subset of G and so passes through at most a finite number of points of G \ H (by
Lemma 4.20). Around any such point a, there is a punctured disk D(a, r) \ {a} H and this
allows us to divert the path around a. After a finite number of such diversions, we end up with a
path in H from z1 to z2 .
Thus H is path connected and so connected.
Now we can apply the identity theorem to f on H and conclude that f (z) = 0z H. This
rules out any poles a G \ H (where limza |f (z)| = ). So H = G and f 0 on G.
Remark 4.24 We can define sums and products of meromorphic functions on G C open but
the definition requires a small bit of care. If f, g are two meromorphic functions on G, then they
are analytic on two different open subsets Hf G and Hg G.
18 414 200304 R. Timoney
We define the sum f + g and the product f g on Hf Hg in the obvious way (f g)(z) =
f (z)g(z) and (f + g)(z) = f (z) + g(z) on Hf Hg . It is not always the case that every point of
G \ (Hf Hg ) = (G \ Hf ) (G \ Hg ) is a pole of f g or f + g.
For example, if f (z) = 1/z and g(z) = z, G = C, Hf = C \ {0} and Hg = . However
f (z)g(z) = 1 on Hf Hg and has no singularity (r a removable singularity) at z = 0.
If we took f (z) = 1/(z 1) + 1/(z 2) and g(z) = 1/z 1/(z 1), then we find that
Hf = C \ {1, 2}, Hg = C \ {0, 1}, (f + g)(z) = 1/z and this has pole at z = 0 and z = 2, but
none at z = 1.
In general though all points of G \ (Hf Hg ) are either poles or removable singularities of
f g. So the product makes sense as a meromorphic function. (Similarly for f + g.) [Exercise:
verify.]
Corollary 4.25 If G C is connected and f (z) is meromorphic on G but not identically 0, then
1/f is meromorphic on G with poles where f (z) = 0 if we define 1/f to be 0 at poles of f .
Proof. Let H G be the open subset where f is analytic and Zf = {z H : f (z) = 0}. Let
Pf = G \ Hf be the set of poles of f .
Zf is clearly closed in H and so H \ Hf is open. 1/f (z) is analytic at all points of H \ Hf .
If a Zf then thereP is a disk of positive radius D(a, r) H where f has a power series
representation f (z) = n=0 an (z a)n (|z a| < r). PBy Theorem 4.23, not all the coefficients
an can be zero though f (a) = a0 = 0. Thus f (z) = n=m an (z a)n with m 1 and am 6= 0.
We can then write f (z) = (z a)m g(z) with g(z) = nm
P
n=m an (z a) analytic in
D(a, r) and g(a) 6= 0. So there is some with 0 < r and g(z) never 0 in D(a, ). Hence
Zf D(a, ) = {a}. Also 1/f (z) = (z a)m (1/g(z)) is analytic in the punctured disc
D(a, ) \ {a} and has an isolated singularity z = a which is a pole of order m.
At points b Pf , limzb |f (z)| = (by Proposition 4.16) and so limzb 1/f (z) = 0. Thus
1/f has a removable singularity at z = b (where it should be assigned the value 0).
Also H Pf is open because H is open and b Pf D(b, ) \ {b} H for some > 0,
which implies D(b, ) H Pf .
So 1/f is now analytic on the open set H Pf = G \ Zf and has isolated singularities at
points of Zf that are all poles.
Remark 4.26 If G C is open, let M (G) denote all the meromorphic functions on G. We have
operations of addition and multiplication on M (G) and since constant functions are in M (G),
we can multiply elements of M (G) by complex scalars (same as multiplying by a constant).
These operations make M (G) a commutative algebra over C. That means a vector space
over C (addition and multiplication by complex scalars) where multiplication is possible (and
certain natural rules are satisfied such as associativity and distributivity). Also the algebra M (G)
has a unit element (the constant function 1 has the property that mutliplying it by any f M (G)
gives f ). Commutativity means f g = gf .
When G is connected we also have the possibility of dividing by nonzero elements. This
makes M (G) a commutative division algebra over C. In particular it is a commutative division
ring (forget the vector space structure) and these are called fields. As M (G) contains (a copy of)
the field C (in the form of the constant functions), it is an extension field of C.
Chapter 4 open mapping theorem, removable singularities 19
Theorem 4.27 (Argument principle, simple meromorphic version) Let G C be simply con-
nected and f a meromorphic function on G with finitely many zeros 1 , 2 , . . . , k and finitely
many poles 1 , 2 , . . . , ` . (We allow k = 0 or ` = 0 when there are no zeros or no poles.)
Say mj is the multiplicity of j as a zero of f (1 j k) and pj is the order of the pole j
(1 j `).
Let be an anticlockwise simple closed curve in G \ {1 , 2 , . . . , k , 1 , 2 , . . . , ` } (that is
in G but not passing through any zeros or poles of f ).
Then
Z 0 k `
1 f (z) X X
dz = mj Ind (j ) pj Ind (j )
2i f (z) j=1 j=1
We will often write the right hand side as N P (or Nf Pf or Nf Pf, ). The first sum
counts the number of zeros of f according to multiplicity and the winding number of around
them and the second sum does a similar thing for poles and their orders.
Proof. The idea is similar to the proof of Theorem 4.8 for the case of analytic functions. We
use the residue theorem and show that the residue of f 0 /f at z = j is mj (the same as before)
and the residue of f 0 /f at z = j is pj . Note that f 0 /f is analytic on G except for isolated
singularities at the zeros j and poles j .
In a punctured disc about a pole z = j , f has a Laurent series
X
f (z) = an (z j )n (0 < |z j | < r)
n=pj
X
= (z j )pj an (z j )n+pj
n=pj
= (z j )pj g(z)
Here g(z) is analytic in |z j | < r and g(j ) = apj 6= 0 and so there exists a positive r
so that g(z) is never zero in the disc D(j , ). In the punctured disc 0 < |z j | < we have
f 0 (z) = pj (z j )pj 1 g(z) + (z j )pj g 0 (z)
f 0 (z) pj (z j )pj 1 g(z) + (z j )pj g 0 (z)
=
f (z) (z j )pj g(z)
1 pj g(z) + (z j )g 0 (z)
=
z j g(z)
1
= h(z)
z j
Remark 4.28 Instead of assuming that G is simply connected in the theorem above (4.27) we
could assume that G is connected and that satisfies one of the following restrictions:
(c) a simple closed anticlockwise curve in G with its inside also contained in G. In this case
the expression for N P in the theorem can be simplified because we are in the situation
where Ind (z) = 1 for z inside (and zero for z outside ). So we get
Z 0
1 f (z) X X
dz = mj pj
2i f (z)
1jk,j inside 1j`,j inside
To remove the necessity to assume that f has only finitely many zeros and poles we need an
improved version of the Residue Theorem.
Theorem 4.29 (Residue theorem, final version) Let G C be open and suppose f is analytic
in G except for isolated singularities. (That is assume there is H G open so that f : H C
is analytic and f has an isolated singularity at each point of G \ H.) Suppose is a (piecewise
C 1 ) curve in G that does not pass through any singularity of f (so it is in fact a curve in H) with
the property that Ind (w) = 0 for all w C \ G.
Then Z X
f (z) dz = 2i res(f, a)Ind (a)
a a singularity of f
Though the sum appears potentially infinite we will show that there can be at most finitely
many nonzero terms in the sum. Excluding the zero terms we are left with a finite sum and we
mean the finite sum.
Proof. To establish first the point about the finiteness of the number of nonzero terms in the sum,
let K = {z C : Ind (z) 6= 0}. Now K G since G and w C \ G Ind (w) =
0w / K. (Thus C \ G C \ K or K G.)
Also K is compact since it is closed and bounded. K is bounded because Ind is zero on
the unbounded component of C \ (which includes {z C : |z| > R} if R is big enough that
D(0, R)). K is also closed because Ind is constant on connected components of C \
which implies that C \ K = {z C : Ind (z) = 0} is a union of connected components of C \
(and these components are all open). Being closed and bounded in C, K is compact.
Now there can only be a finite number of singularities of f in K by an argument similar to the
proof in Lemma 4.20 (which was for meromorphic f ). [Here is the idea: If there were infinitely
many singularities in K we could find an infinite sequence (an ) n=1 of distinct singularities in K.
Chapter 4 open mapping theorem, removable singularities 21
Then we could find a subsequence (anj ) j=1 converging to a limit a K. The function f can
neither be analytic at a nor have an isolated singularity there.]
We now let H be the open subset of G where f is analytic and a1 , a2 , . . . , an the singularities
of f inside K. Let G1 = H {a1 , a2 , . . . , an }. Then G1 is open, K G1 and f is analytic on
G1 except for a finite number of singularities. is a curve in H G1 with Ind (w) = 0 for all
w C\G1 (since such w are not in K). The earlier version of the residue theorem (Theorem 4.3)
applies to f on G1 and implies the result.
Remark 4.30 The argument principle can now be extended to meromorphic functions with po-
tentially infinite numbers of zeros and poles. We need to avoid non-isolated zeros which
would mean f identically zero on some connected component of G by the identity theorem and
then the curve could not be in such a component because we insist that f is never zero on .
(We also require that f has no poles on ).
Since has to be in one connected component of G in any case, we can just assume that
G is connected, f meromorphic on G and no poles or zeros of f on . Then we must make
suitable assumptions about (such as piecewise C 1 closed curve in G with Ind (w) = 0 for
all w C \ G).
The argument principle will then state
f 0 (z)
Z
1 X X
dz = multf (a)Ind (a) orderf (b)Ind (b) = Nf Pf
2i f (z)
aG,f (a)=0 bG,b pole of f
where multf (a) means the multiplicity of a as a zero of f and orderf (b) means the order of the
pole b of f .
Theorem 4.31 (Rouches theorem) Suppose f and g are meromorphic functions on a con-
nected open G C and is a piecewise C 1 closed curve in G with
Then
Nf Pf = Ng Pg
where X X
Nf = multf (a)Ind (a), Pf = orderf (b)Ind (b)
aG,f (a)=0 bG,a pole of f
and similarly for Ng and Pg .
22 414 200304 R. Timoney
Proof. We could note that the hypotheses (c) actually implies (b) because if there were any poles
of either f or g on then the inequality (c) would not make sense and if there were zeros of
either f or g on then the strict inequality could not hold.
Dividing across by |f (z)| we can rewrite (c) as
1 g(z) < 1 g(z) D(1, 1)
f (z) f (z)
The essence of the proof is that this means Log(g(z)/f (z)) (the principal branch of the log)
makes sense for z and gives an antiderivative for g 0 (z)/g(z) f 0 (z)/f (z). To make this
antiderivative argument work we need the log on some open set that contains .
Say f is analytic on Hf G open (with poles on G \ Hf ) and g is analytic on Hg \ G open
(with poles on G \ Hg ). Let Zf = {z Hf : f (z) = 0} = the zeros of f . Then g(z)/f (z) is
certainly analytic on (Hf \ Zf ) Hg , an open set that contains . By continuity of g/f , the set
It follows then that the integral of this is zero around the closed curve , that is
Z 0
g (z) f 0 (z)
Z 0 Z 0
g (z) f (z)
0= dz = dz dz
g(z) f (z) g(z) f (z)
Thus
g 0 (z) f 0 (z)
Z Z
dz = dz
g(z) f (z)
and so by the argument principle
2i(Ng Pg ) = 2i(Nf Pf )
Example 4.32 (i) We can use Rouches theorem ( 4.31) to reprove the fundamental theorem
of algebra in yet another way. If p(z) = an z n +an1 z n1 + +a1 z +a0 is a polynomial of
degree n 1 (so that an 6= 0), then the earlier proof started by showing that for |z| = R > 0
large enough we have
1
|an1 z n1 + + a1 z + a0 | < |an ||z|n
2
(We still need this part of the earlier proof and most proofs of the theorem need this part.)
If we take f (z) = an z n , g(z) = p(z) and the circle |z| = R traversed once anticlockwise,
Chapter 4 open mapping theorem, removable singularities 23
then we can apply Rouches theorem (with G = C because both functions are entire) since
we have
1
|f (z)g(z)| = |(an1 z n1 + +a1 z +a0 )| < |an ||z|n < |an z n | = |f (z)| for |z| = R.
2
Thus we conclude
Nf P f = Ng P g
or Nf = Ng (since there are no poles in this case). Now Nf = n since f (z) = 0 has only
the solution z = 0 and that has multiplicity n (and Ind (0) = 1). So g(z) has Ng = n zeros
(counting multiplicities) inside |z| = R (for all large R). It follows that if n 1 then p(z)
has a zero.
(ii) Show that if > 1 then the equation z ez = 0 has exactly one solution in the right
half plane <z > 0.
Solution. This is meant to illustrate the difficulty of applying Rouches theorem as we have
only one function here (need another) and no curve . We take g(z) = z ez the
function we want the information about and f (z) = z a simpler function to analyse.
We select as our curve any closed semicircle of radius R > + 1 in the right half plane,
oriented anticlockwise. That is is the semicircle |z| = R in <z 0 plus the segment of
the imaginary axis from iR to iR. We take G = C as our function f and g are entire.
For z we have
|f (z) g(z)| = |ez | = e<z e0 = 1.
For z on the semicircular part of we have
|f (z)| = | z| |z| > R > 1 |f (z) g(z)|.
For z on the imaginary axis we have
p
|f (z)| = | z| = | iy| = 2 + y 2 > 1 |f (z) g(z)|.