Professional Documents
Culture Documents
Matt Rosenzweig
1
Contents
2
Conformal Mapping, Dirichlets Problem 27
The Riemann Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.1.1 Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.1.1 Exercise 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Elliptic Functions 28
Weierstrass Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.3.2 Exercise 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.3.3 Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.3.3 Exercise 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.3.3 Exercise 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.3.3 Exercise 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.3.3 Exercise 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.3.3 Exercise 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.3.5 Exercise 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3
Chapter 4 - Complex Integration
ez
I I
1 f (z)
1 = f (0) = dz 2i = dz
2i |z|=1 z |z|=1 z
4.2.3 Exercise 1
1. Applying Cauchys differentiation formula to f (z) = ez ,
(n 1)! ez ez
I I
(n1) 2i
1=f (0) = n
dz = n
dz
2i |z|=1 z (n 1)! |z|=1 z
(d) If n < 0, m < 0, then sincen(|z| = 2, 0) = n(|z| = 2, 1) = 1, we have by the residue formula that
I I I
m n m n
(1 z) z = 2ires(f ; 0) + 2ires(f ; 1) = (1 z) z dz + (1 z)m z n dz
|z|=2 |z|= 12 |z1|= 21
4
Using Cauchys differentiation formula, we obtain
"I #
(1 z)|m| z |n|
I I
m n
(1 z) z dz = dz + dz
|z|=2 |z|= 21 z |n| |z1|= 21 (1 z)
|m|
We consider two cases. First, suppose |a| > . Then z(z a)2 is holomorphic on and inside {|z| = }
2
and a lies inside {|z| = }. By Cauchys differentiation formula,
" #
i 2
I
4 2 3
2
|z a| |dz| = 2i 2 (z a) 2z(z a) z= 2 = 2 2 1 2 2
|z|= a a a ( a a)2 a( a a)
2 2
2(2 + |a| ) 2(2 + |a| )
= 2 = 2
(2 |a| )3 (|a| 2 )3
2 2
Now, suppose |a| < . Then a lies outside |z| = , so the function z(z a )2 is holomorphic on and
inside {|z| = }. By Cauchys differentiation formula,
" #
2 2 2 3
i
I
4 2 a
|z a| |dz| = 2i 2 (z ) 2z(z ) = 2 2 12 2
|z|= a a a z=a a (a )2 (a ) a a
2 2 2
2 (a + a ) 2(|a| + 2 ) 2(|a| + 2 )
= 2 2
= 2 = 2
(|a| 2 )2 a a
(|a| 2 )3 (2 |a| )3
4.2.3 Exercise 2
Let f : C C be a holomorphic function satisfying the following condition: there exists R > 0 and n N
n
such that |f (z)| < |z| |z| R. For every r R, we have by the Cauchy differentiation formula that for
all m > n,
n
|z|
I
(m) m! m!
f (a) m+1 |dz| mn
2 |z|=r |z| r
Noting that m n 1 and letting r , we have that f (m) (a) = 0. Since f is entire, for every a C, we
may write
f (n) (a)
f (z) = f (a) + f 0 (a)(z a) + + (z a)n + fn+1 (z)(z a)n+1 z C
n!
where fn+1 is entire. Since fn+1 (a) = f (n+1) (a) = 0 and a C was arbitary, we have that fn+1 0 on C.
Hence, f is a polynomial of degree at most n.
5
Local Properties of Analytical Functions
4.3.2 Exercise 2
Let f : C C be an entire function with a nonessential singularity at . Consider the function g(z) = f z1
at z = 0. Let n N be minimal such that limz0 z n g(z) = 0. Then the function z n1 g(z) has an analytic
continuation h(z) defined on all of C. By Taylors theorem, we may express h(z) as
And hc
n1 cn2 i
lim g(z) + + + c0 = lim f (z) = f (0)
z z n1 z n2 z0
since f is entire. Note that we also obtain that c0 = f (0). Hence, g(z) zcn1 cn2
n1 + z n2 + + c0 (we
are abusing notation to denote the continuation to all of C) is a bounded entire function and is therefore
identically zero by Liouvilles theorem. Hence,
4.3.2 Exercise 4
Let f : C {} C {} be a meromorphic function in the extended complex plane. First, I claim that
f has finitely many poles. Since the poles of f are isolated points, they form an at most countable subset
{pk }k=1 of C. By definition, the function f(z) = f z1 has either a removable singularity or a pole at z = 0.
In either case, there exists r > 0 such that f is holomorphic on D0 (0; r). Hence, {pk }k=1 D(0; r). Since
this set is bounded, {pk }k=1 has a limit point p. By continuity, f (p) = and therefore p is a pole. Since p
is an isolated point, there must exist N N such that k N, pk = p.
Our reasoning in the preceding Exercise 2 shows that for any pole pk 6= of order mk , we can write in a
neighborhood of pk
cmk cmk 1 c1
f (z) = + + + + c0 +gk (z)
(z pk )mk (z pk )mk 1 z pk
| {z }
fk (z)
I claim that h is (or rather, extendsPto) an entire, bounded function. Indeed, in a neighborhoodPof each zk ,
n
h can be written as h(z) = gk (z) i6=k fk (z) and in a neighborhood of z as h(z) = g (z) k=1 fk (z),
1
which are sums of holomorphic functions. h(z) = h z is evidently bounded in a neighborhood of 0 since
1 1 1
the fk z are polynomials and f z f z = g (z), which is holomorphic in a neighborhood of 0. By
Liouvilles theorem, h is a constant. It is immediate from the definition of h that f is a rational function.
6
Calculus of Residues
4.5.2 Exercise 1
Set f (z) = 6z 3 and g(z) = z 7 2z 5 z + 1. Clearly, f, g are entire, |f (z)| > |g(z)| |z| = 1, and
f (z) + g(z) = z 7 2z 5 + 6z 3 z + 1. By Rouches theorem, f and f + g have the same number of zeros,
which is 3 (counted with order), in the disk {|z| < 1}.
So |z| (1, 2). Hence, the equation z 4 6z + 3 = 0 has 3 roots (counted with order) with modulus strictly
between 1 and 2.
4.5.3 Exercise 1
1 1
1. Set f (z) = z 2 +5z+6 = (z+3)(z+2) . Then f has poles z1 = 2, z2 = 3 and by Cauchy integral formula,
(z + 3)1
I
1 1
res(f ; z1 ) = dz = |z=2 = 1
2i |z+2|= 12 (z + 2) z+3
(z + 2)1
I
1 1
res(f ; z2 ) = dz = |z=3 1
2i |z+3|= 12 (z + 3) z+2
1 1
2. Set f (z) = (z2 1) 2 = (z1)2 (z+1)2 . Then f has poles z1 = 1, z2 = 1. Applying Cauchys differentia-
(z 1)2
I
1 1
res(f ; z1 ) = 2
dz = 2(z 1)3 |z=1 =
2i |z+1|=1 (z + 1) 4
(z + 1)2
I
1 1
res(f ; z2 ) = dz = 2(z + 1)3 |z=1 =
2i |z1|=1 (z 1)2 4
3. sin(z) has zeros at k, k Z, hence sin(z)1 has poles at zk = k. We can write sin(z) = (z
zk ) [cos(zk ) + gk (z)], where gk is holomorphic and gk (zk ) = 0. By the Cauchy integral formula,
1
[f 0 (zk ) + gk (z)]
I
1 1
res(f ; zk ) = dz = = (1)k
2i |zzk |=1 (z zk ) f 0 (zk ) + g(zk )
4. Set f (z) = cot(z). Since sin(z) has zeros at zk = k, k Z and cos(zk ) 6= 0, cot(z) has poles at
zk , k Z. We can write sin(z) = (z zk ) [cos(zk ) + gk (z)], where gk is holomorphic and gk (zk ) = 0.
By Cauchys integral formula,
I 1
1 cos(z) [cos(zk ) + gk (z)] cos(zk )
res(f ; zk ) = dz = =1
2i |zzk |=1 (z zk ) cos(zk ) + gk (zk )
5. It follows from (3) that f (z) = sin(z)2 has poles at zk = k, k Z. We remark further that
gk (z) = cos(zk )(z zk )2 + hk (z), where hk (z) is holomorphic. By the Cauchy differentiation formula,
2
gk0 (zk )
I
1 [cos(zk ) + gk (z)]
res(f ; zk ) = dz = 2 =0
2i |zzk |=1 (z zk )2 (cos(zk ) + gk (zk ))3
7
1
6. Evidently, the poles of f (z) = z m (1z)n are z1 = 0, z2 = 1. By Cauchys differentiation formula,
(1 z)n
(n + m 2)! n+m2
I
1
res(f ; z1 ) = m
dz = =
2i |z|= 21 z (n 1)!(m 1)! m1
4.5.3 Exercise 3
1cos(2)
(a) Since a + sin2 () = a + 2 = 2 [(2a + 1) cos(2)], we have
Z Z Z Z 0
2 d 2 d dt d
=2 = =
0 a + sin2 () 0 (2a + 1) cos(2) 0 (2a + 1) cos(t) (2a + 1) + cos( )
Z
d
=
0 (2a + 1) + cos( )
where we make the change of variable R= , and the last equality follows from the symmetry of
dx
the integrand. Ahlfors p. 155 computes 0 +cos(x) = 2 1 for > 1. Hence,
Z
2 d
2 =p
0 a + sin () (2a + 1)2 1
(b) Set
z2 z2 z2
f (z) = = =
z 4 + 5z 2 + 6 (z 2 + 3)(z 2 + 2) (z 3i)(z + 3i)(z 2i)(z + 2i)
For R >> 0,
1 : [R, R] C, 1 (t) = t; 2 : [0, ] C, 2 (t) = Reit
and let be the positively oriented closed curve formed by 1 , 2 . By the residue formula and applying
eiz
the Cauchy integral formula to z+ai to compute res(f ; ai),
Z
f (z)dz = 2ires(f ; 3i) + 2ires(f ; 2i)
R3
Z
f (z)dz |R2 3| |R2 2| 0, R
2
Hence,
Z
Z
x2
Z
x2 x2 dx ( 3 2)
2 = dx = ( 3 2) =
0 x4 + 5x2 + 6 x4 + 5x2 + 6 0 x4 + 5x2 + 6 2
8
(e) We may write
cos(x) eix
= Re
x2 + a2 (x2 + a2 )
eiz
So set f (z) = z 2 +a2 , which has simple poles at ai. First, suppose that a 6= 0. For R >> 0, define
1 : [R, R] C, 1 (t) = t; 2 : [0, ] C, 2 (t) = Reit
and let be the positively oriented closed curve formed by 1 , 2 . By the residue formula,
ei(ai) ea
Z
f (z)dz = 2ires(f ; ai) = 2i =
(2ai) a
Z Z iR[cos(t)+i sin(t)] Z iR cos(t) R sin(t)
e it
e e it
f (z)dz =
Re dt = Re dt
2 0 R2 e2it + a2
0 R2 e2it + a2
Z
ReR sin(t) R
2 2
dt 2 0, R
0 R a R a2
since eR sin(t) 1 on [0, ]. Hence,
Z
1 eix ea
Z
cos(x)
= Re dx =
0 x2 + a2 2 x2 + a2 2a
If a = 0, then the integral does not converge.
log(z) 3
(h) Define f (z) = (1+z 2 ) , where we take the branch of the logarithm with arg(z) [ 2 , 2 ). For R >> 0,
define
1 1 it 1
1 : [R, ] C, 1 (t) = t; 2 : [, ] C, 2 (t) = e ; 3 : [ , R] C, 3 (t) = t; 4 : [0, ] C, 4 (t) = Reit
R R R
and let be the positively oriented closed curve formed by the i .
Z log |R|1 + 3
R(log |R| + 3
Z
2 1 2 )
f (z)dz 1 dt 2 1|
0, R
2 0
R2 1 R |R
Z Z
|log |R| + it| R(log |R| + )
f (z)dz 21
Rdt 0, R
4
0 R R2 1
By the residue formula and applying the Cauchy integral formula to f (z)/(z + i) to compute res(f ; i),
2
Z
log(z)
f (z)dz = 2ires(f ; i) = 2i |z=i = 2i 2 =
(z + i) 2i 2
Hence,
Z R1 Z R Z R1 Z R Z R1
2 log(tei )
Z Z
log(t) log(|t|) log(t) 1
= f (z)dz+ f (z)dz = 2
dt+ 2
dt = 2
dt+ 2
dt+ dt
2 1 3 R 1 + t R
1 1 + t R 1 + t 1
R
1 + t R 1 + t2
Z R Z R Z R
log(t) 1 log(t) 2
=2 + dt = 2 +
1
R
1 + t2 1
R
1 + t2 1
R
1 + t2 2
R 1
where weve used 0 1+t2 dt = limR arctan(R) arctan(0) = 2 . Hence,
Z R Z
log(t) log(t)
2
dt = 0 dt = 0
1
R
1 + t 0 1 + t2
9
Harmonic Functions
4.6.2 Exercise 1
Let u : D0 (0; ) R be harmonic and bounded. I am going to cheat a bit and assume Schwarzs theorem
for the Poisson integral formula, even though Ahlfors discusses it in a subsequent section. Let
Z 2
1 rei z
Pu (z) = Re i u(rei )d
2 0 re + z
denote the Poisson integral for u on some circle of fixed radius r < . Since u is continuous, Pu (z) is a
harmonic function in the open disk D(0; r) and is continuous on the boundary {|z| = r}. We want to show
that u and Pu agree on the annulus, so that we can define a harmonic extension of u by setting u(0) = Pu (0).
Define
g(z) = u(z) Pu (z)
and for > 0 define
|z|
g (z) = g(z) + log 0 < |z| r
r
Then g is harmonic in D0 (0; r) and continuous on the boundary. Furthermore, since u is bounded by
hypothesis and Pu is bounded by construction on D(0; r), we have that g is bounded on D(0; r). g (z) is
harmonic in D0 (0; r) and continuous on the boundary since both its terms are. Since log r1 |z| , z
0, we have that
lim sup g (z) < 0
z0
Hence, there exists > 0 such that 0 < |z| g (z) 0. Since g is harmonic on the closed annulus
{ |z| r}, we can apply the maximum principle. Hence, g assumes its maximum in {|z| = } {|z| = r}.
But, g (z) 0 |z| = , by our choice of , and since u, Pu agree on {|z| = r}, we have that g (z) = 0 |z| = r.
Hence,
g (z) 0 0 < |z| r
Letting 0, we conclude that g(z) 0 0 < |z| r, which shows that u Pu on the annulus. Applying
the same argument to h = Pu u, we conclude that u = Pu on 0 < |z| r. Setting u(0) = Pu (0) defines a
harmonic extension of u on the closed disk.
4.6.2 Exercise 2
If f : = {r1 < |z| < r2 } C is identically zero, then there is nothing to prove. Assume otherwise. Since
the annulus is bounded, f has finitely many zeroes in the region. Hence, for R, the function
is harmonic in \ {a1 , , an }, where a1 , , an are the zeroes of f . Applying the maximum principle to
g(z), we see that |g(z)| takes its maximum in . Hence,
log |z| + log |f (z)| = g(z) max { log(r1 ) + log(M (r1 )), log(r2 ) + log(M (r2 ))} z \ {a1 , , an }
log(r) + log (M (r)) max { log(r1 ) + log(M (r1 )), log(r2 ) + log(M (r2 ))}
We now find R such that the two inputs in the maximum function are equal.
r1 M (r2 )
log(r1 ) + log(M (r1 )) = log(r2 ) + log(M (r2 )) log = log
r2 M (r1 )
1
M (r2 )
Hence, = log M (r1 ) log rr21 . Exponentiating both sides of the obtained inequality,
log rr2
M (r2 ) M (r1 )
M (r) exp log(M (r2 )) + log = exp log(M (r2 ) + log
M (r1 ) log rr12 M (r2 )
10
= M (r1 ) M (r2 )1
1
r2 r2
where = log r log r1 . I claim that equality holds if and only if f (z) = az , where a C, R.
It is obvious that equality holds if f (z) is of this form. Suppose quality holds. Then by Weierstrasss extreme
value theorem, for some |z0 | = r, we have
r
1
M (r1 ) z0 f (z0 ) = r1 M (r1 )
|f (z0 )| = M (r) =
r
But since the bound on the RHS holds for all r1 < |z| < r2 , the Maximum Modulus Principle tells us that
z f (z) = a C r1 < |z| < r2 . Hence, f (z) = az . But is an arbitrary real parameter, from which the
claim follows.
4.6.4 Exercise 1
We seek a conformal mapping of the upper-half plane H+ onto the unit disk D. lemma The map given by
1+z
(z) = i
1z
is a conformal map of D onto H+ and is a bijective continuous map of D onto R {}, where 1 7 . Its
inverse is given by
wi
1 (w) =
w+i
Proof. The statements about conformality and continuity follow from a general theorem about the group of
linear fractional transformations of the Riemann sphere (Ahlfors p. 76), so we just need to verify the images.
For z D,
2
1+z 1z 1 |z|
Im((z)) = Im i = 2 >0
1z 1z |1 z|
since |z| < 1. Furthermore, observe that Im((z)) = 0 z D. In particular, (1) = . For w H+ ,
2
(w)2 = w i w + i = |w| 2Im(w) + 1 < 1
1
w+i wi 2
|w| + 2Im(w) + 1
by hypothesis that Im(z) > 0. Furthermore, observe that 1 (w) = 1 Im(w) = 0.
is harmonic in H+ . Fix w0 D and let x0 + iy0 = z0 = 1 (w0 ). Let Pw0 () denote the Poisson kernel. We
apply the change of variable t = 1 (ei ) to obtain
z0 i 2
2
ti
1 d 1 1 z0 +i 1 t+i 1
2
|z0 + i| |z0 i|
2
((t + i) (t i))
2
Pw0 () = 2 =
2 dt 2 z0 i ti 2 ti 2 2 2
z0 +i t+i (z0 i) ti 2 |t + i|
t+i (z0 + i)
t+i
2 y0 2 y0 1 y0
= = =
|(z0 i)(t + i) (t i)(z0 + i)|2 2 |z0 t|2 (x0 t)2 + y02
11
Hence, Z
1 y
PU (x, y) = U (t)dt
(x t)2 + y 2
+
is a harmonic function in H . Furthermore, since the value of PU (z) for |z| = 1 is given by U (z) at the points
of continuity and since 1 (D) = R {}, we conclude that
4.6.4 Exercise 5
I couldnt figure out how to Rshow that log |f (z)| satisfies the mean-value property for z0 = 0, r = 1 without
first computing the value of 0 log sin()d.
R R
By symmetry, it follows that the improper integral log |sin()| d exists and therefore 0 log |sin()| d
2
R R
exists. Again by symmetry, 02 log(sin())d = 02 log cos()d, hence
Z Z Z Z
2 1 2 1 2 1 1 2
log sin()d = log sin() cos()d = log sin(2) d = sin(2)d log(2)
0 0 2 2 0 2 2 0 4
1
Z
= sin()d log(2)
4 0 4
R R
where we make the change of variable = 2 to obtain the last equality. Since 0 log sin()d = 2 02 log sin()d,
we conclude that Z
log sin()d = log(2)
0
We now show that for f (z) = 1 + z, log |f (z)| satisfies the mean-value property for z0 = 0, r = 1. Observe
that
1 1 1
log 1 + ei = log (1 + cos())2 + sin2 () = log 1 + 2 cos() + cos2 () + sin2 () = log |2 + 2 cos()|
2 2 2
1 1 + cos()
= log |2| + log
2 2
Substituting and making the change of variable 2 = ,
Z 2 Z 2 Z Z
i 1 2 1 + cos(2)
log cos2 ()d
log 1 + e d =
log 2 + log cos () d = 2 log 2+
log
d = 2 log 2+
0 0 2 0 2 0
By symmetry, integrating log cos2 () over [0, ] is the same as integrating log sin2 () over [0, ]. Hence,
Z 2 Z Z
log 1 + ei d = 2 log 2 + log sin2 () d = 2 log 2 + 2
log |sin()| d = 0
0 0 0
12
4.6.4 Exercise 6
Let f : C C be an entire holomorphic function, and suppose that z 1 Re(f (z)) 0, z . By Schwarzs
formula (Ahlfors (66) p. 168), we may write
Z
1 +z d
f (z) = Re(f ()) |z| < R
2i ||=R z
Let > 0 be given and R0 > 0 such that R R0 , Re(fz(z)) < . Let R be sufficiently large that R > R2 > R0 .
4.6.5 Exercise 1
Let f : C C be an entire holomorphic function satisfying f (R) R and f (i R) i R. Since f (R) R,
f (z) f (z) vanishes on the real axis. By the limit-point uniqueness theorem that
f (z) = f (z) z C
Since f (iR) iR, f (z) + f (z) vanishes on the imaginary axis. By the limit-point uniqueness theorem that
f (z) = f (z) z C
Combining these two results, we have
f (z) = f (z) = f (z) = f (z) z C
4.6.5 Exercise 3
Let f : D C be holomorphic and satisfy |f (z)| = 1 |z| = 1. Let : C {} C {} be the linear
fractional transformation
zi
(z) =
z+i
+
Consider the function g = 1 f : H C. By the maximum modulus principle, |f (z)| 1 |z| 1.
+ +
Hence, g : H H . Since |f (z)| = 1 |z| = 1, 1 (f (z)) R |z| = 1. Hence, f(R) R. By the Schwarz
Reflection Principle, g extends to an entire function g : C C satisfying g(z) = g(z). Define
f = g 1 : C C
Then f is meromorphic in C since has a pole at z = i and 1 has a pole at z = 1. In particular,
f has finitely many poles. We proved in Problem Set 1 (Ahlfors Section 4.3.2 Exercise 4) that a function
meromorphic in the extended complex plane is a rational function, so we need to verify that f doesnt have
an essential singularity at . But in a neighborhood of 0,
1 + z1
1 z+1
f =g i =g i
z 1 z1 z1
which is evidently a meromorphic function. Alternatively, we note that |z| 1, f(z) 1 since g maps
H onto H . So the image of f in a suitable neighborhood of is not dense in C. The Casorati-Weierstrass
theorem then tells us that f cannot have an essential singularity at .
13
Chapter 5 - Series and Product
Developments
N N N
X 1 X 1 X 1 X 1
z , < <
z
n n=1 n
z n Re(z 0 ) n Re(z0)
n=1 n=1 n=1
Hence, N
X a N 1
n 1 1 X 1 1
M + M + M
nz |N zz0 | |nzz0 | (n + 1)zz0 nzz0
n=m n=m
Observe that
1 1
log(n+1)(zz ) 1 Z log(n+1)
e log(n)(zz0 ) = et(zz0 ) dt
(n + 1)zz0 nzz0 = e
0
z z0 log(n)
14
log(n+1)
|Re(z) Re(z0 )| log(n+1)(Re(z)Re(z0 ))
Z
1
et(Re(z)Re(z0 )) dt = e log(n)(Re(z)Re(z0 ))
|z z0 | |z z0 |
e
log(n)
1 1
e log(n)(Re(z)Re(z0 )) e log(n+1)(Re(z)Re(z0 )) =
nRe(z)Re(z0 ) (n + 1)Re(z)Re(z0 )
Since this last expression is telescoping as the summation ranges over n, we have that
N
X an M M 1 1
+ Re(z)Re(z ) + M Re(z)Re(z ) Re(z)Re(z )
n=m nz N Re(z)Re(z0 )
m 0 N 0 m 0
2M M m Re(z)Re(z0 ) 4M
+ 1 Re(z)Re(z ) 0, m
mRe(z)Re(z0 )
mRe(z)Re(z0 ) N m 0
PN an
Hence, the partial sums of n=1 nzz0
are Cauchy and therefore converge by completeness.
P an P
Corollary 3. If n=1 nz converges for some z = z0 , then n=1 annz conveges uniformly on compact subsets
of {Re(z) Re(z0 )}.
Proof. Let K {Re(z) Re(z0 )} be compact. Since
P z 7 Re(z) is continuous, there exists z1 K such that
Re(z) Re(z1 ) z K. Since Re(z1 ) Re(z0 ), n=1 nazn1 converges. The proof of the preceding lemma
shows that we have a uniform bound
N
X a 4M 4M
n
Re(z )Re(z )
n=m nz mRe(z)Re(z0 )
m 1 0
where M depends only on z0 . The claim follows immediately from the M -test and completeness.
P
= n=1 annz converges if we take z R>0 (the well-known alternating series), we
Since the series f (z)P
have by the lemma that n=1 annz converges Re(z) > 0. We now show that this series is holomorphic on the
region {Re(z) > 0}.
Define a sequence of functions (fN )N N by
N
X (1)n+1
fN =
n=1
nz
It is clear that fN is holomorphic, being the finite sum of holomorphic functions. Set = {z C : Re(z) > 0}
and let K be compact. Since the fN are just the partial sums of the series, we have by the corollary to
the lemma that fN f uniformlyP on K. By Weierstrasss theorem, f is holomorphic in .
To see that (1 21z )(z) = n=1 annz on {Re(z) > 1}, observe that
N N N N N
1z
X 1 X (1)n+1 X 1 X 1 X (1)n+1 X 1
(1 2 ) = 2 = 2
n=1
nz n=1 nz n=1
nz n=1
(2n)z n=1 nz N <n2N
nz
n is even
P 1
Since n=1 nz is absolutely convergent, we see that by taking N sufficiently large, the RHS can be made
less than for > 0 given.
5.1.2 Exercise 2
12
Differentiating 1 2z + z 2 with respect to z, we obtain
2 2z
p1 () = 3 |z=0 =
2(1 2z + z 2 ) 2
12
To compute higher order Legendre polynomials, we differentiate 1 2z + z 2 and its Taylor series to
obtain the equality
z X
n1 z 2
X
3 = nPn ()z = (1 2z + z ) nPn ()z n1
(1 2z + z 2 ) 2 1 2z + z 2
n=1 n=1
15
Hence,
X
X
X
X
X
Pn ()z n Pn ()z n+1 = nPn ()z n1 2nPn ()z n + nPn ()z n+1
n=0 n=0 n=0 n=0 n=0
Invoking elementary limit properties and using the fact that a function is zero if and only if all its Taylor
coefficients are zero, we may equate terms to obtain the recurrence
5.1.2 Exercise 3
Observe that
sin(z) 1 X (1)n 2n+1 X (1)n 2n
= z = z
z z n=0 (2n + 1)! n=0
(2n + 1)!
So, sin(z)
z 6= 0 in some open disk about z = 0. Hence, the function z 7 log sin(z)
z is holomorphic in an open
disk about z = 0, where we take the principal branch of the logarithm. Substituting,
P (1)n 2n m
sin(z)
sin(z)
X 1 n=0 (2n+1)! z
log = log 1 1 =
z z m=1
m
1 2 1 4 1 6
m
X
3! z 5! z + 7! z [z 8 ]
=
m=1
m
1 2 1 4
Set P (z) = 3! z 5! z . Then
6
P (z) + [z 8 ] P (z)2 + [z 8 ] P (z)3 + [z 8 ]
sin(z) z
log = + + +
z 7! 1 2 3
2
z4 z6
4
2z 6 z6
z 1 z 8
= + + + + [z ]
3! 5! 7! 2 (3!)2 (3!)(5!) 3(3!)3
1 1 4 1 6
= z2 z z + [z 8 ]
6 180 2835
16
2
where we expand nz 2 using the geometric series. Since both series are absolutely convergent, we may inter-
change the order of summation to obtain
!
2
X X 1 2k
X
z cot(z) = 1 2z 2(k+1)
z = 1 2 (2k)z 2k
k=0 n=1
n k=0
X Bk
= iz + (2iz)k
k!
k=0
P k
P k
where we may use the geometric expansion since k=1 (2iz) (k+1)!
1
k=1 |2z| < 1 (|z| < 2 ), and the
change in the order of summation is permitted since the series are absolutely convergent. According to
Ahlfors, the numbers Bk are called Bernoulli numbers, the values of which one can look up. Since the two
series representations for z cot(z) are equal, the coefficients must agree. Hence,
1 (2i)2 B2 2
(2) = =
2 2! 6
1 (2i)4 B4 16 4 4
(4) = = 60 =
2 4! 6 90
1 (2i)6 B6 32 6 6 6
(6) = = = =
2 6! 42 6! 21 45 945
5.2.1 Exercise 2
We first observe that
X 1
z 3 n3
n=
P
converges absolutely, being comparable to n=1 n13 . For z 6= 0, we may write (after some laborious compu-
tation, which can be found at the end of the solutions)
1 1 1 A B C
= 2 4 = 2 4 = + i 4 + i 2
z 3 n3 i 3 i
(z n)(z ne )(z ne ) 3 i i
(z n)(e z n)(e z n)
3 3 z n ze 3 n ze 3 n
where 2 4
e3i e3i 1
C= 2
B = 2
A= 2
3z 3z 3z
Pm 1
Ahlfors p. 189 shows that limm m zn = cot(z), 0 < |z| < 1. Hence, for 0 < |z| < 1,
m m 2 m 4 m
X 1 1 X 1 e3i X 1 e3i X 1
lim = lim + lim + lim
m
m
z 3 n3 3z 2 m m z n 3z 2 m m zei 2
3 n 3z 2 m m zei 4
3 n
2 2 4 4
cot(z) e 3 i cot(e 3 i z) e 3 i cot(e 3 i )
= + +
3z 2 3z 2 3z 2
17
5.2.2 Exercise 2
In what follows, we will restrict ourselves to z D(0; 1). For n Z0 , define
n
Y
n i
Pn (z) = (1 + z) 1 + z 2 1 + z 2 = 1 + z2
i=1
n+1
First, I claim that (1 z)Pn (z) = (1 z 2 ). Suppose the claim is true for some n, then
n+1 n+1 n+1
n+2
(n+1)+1
(1 z)Pn+1 (z) = [(1 z)Pn (z)] (1 + z 2 ) = (1 z 2 )(1 + z 2 ) = 1 z 2 = 1 z2
The base case is trivial, so the result follows by induction. Therefore,
Pn (z) 1 1 2n+1
|(1 z)Pn (z) 1| = |z| 0, n
1z 1 |z|
1
since |z| < 1. Since 1|z| , |z| are bounded on any compact subset of D(0; 1), we remark that the convergence
is uniform on compact subsets of D(0; 1).
5.2.3 Exercise 3
note that even though the function z 7 z is not entire for any branch choice, the function f (z) =
First,
cos( z) is. Indeed, substituting into the definition of cos(z),
X (1)n 2n
f (z) = ( z)
n=0
(2n)!
Since changing the choice of branch only results in a sign change, we see that ( z)2n = z n , and therefore
X (1)n n
f (z) = z
n=0
(2n)!
which is evidentlyan entire function, being
a power series with infinite radius of convergence. Observe that
2
(2n+1)
f (z) has zero set 2 : n Z . Since sin(z + 2 ) = cos(z), cos(z) can be written as
z + 21
1 Y z+ 1 1 Y 2n 1 2z 1 z
2
cos(z) = z + 1 e n = z+ e 2n + n
2 n 2 2n 2n
n6=0 n6=0
!
z2
Y
z 2n 1 z 1 z
+n Y 1
= 1 1 1 2n1 e 2n = 1 2 1 (2n1)2
2 2
2n 2
2 n=1 4n
n6=0 4
Using the infinite product representation of sin(z), we have
2
!
sin 2
2 Y
2
Y 4
= = 1 2 2 = 1 2
2 n=1
n n=1
n
Hence,
!
2 2
Y z Y z
cos(z) = 1 cos(z) = 1
(2n1)2 2
(2n1)
n=1 4 n=1
2
Since
2
X (2n 1) 2 X 1
= <
n=1
2 4 n=1 (2n 1)2
1
P
being comparable to n2 , we see that f (z) is an entire function of genus zero.
18
5.2.3 Exercise 4
Let f (z) be an entire function of genus h. Let {an 6= 0}nN denotes the (at most countable) set of nonzero
zeroes of f and hc denote the genus of the canonical product. We may write
z 2 hc
e an + 2 ( an ) ++ hc ( an )
Y z 1 z 1 z
f (z) = z m eg(z) 1
n=1
an
n=1
an
2 2hc +1 2 2hc +1
z z
2m g (z 2 ) z + 1 z ++ 2hc1+1 zan z
+1 z
++ 2hc1+1 zan
Y
=z e 1 1+ e an 2 an e an 2 an
n=1
an an
where weve chosen some branch of the square root. If we define b1 = a1 , b2 = a1 , . Then
2 z 2 2hc +1
e bn + 2 ( bn ) ++ 2hc1+1 ( bzn )
z 1 z
f(z) = f (z 2 ) = z 2m eg(z )
Y
1
n=1
bn
the breaking up of the product being justified since the individual products converge absolutely by virtue of
X 1 X 1
2hc +1+1
= hc +1
<
|bn | |an |
I claim that the genus of f is bounded from below by h. If h = 0, then there is nothing to prove; assume
otherwise. If h = deg(g(z)) > 0, then h deg(g(z 2 )) > h; so assume that h = hc . We will show that the
genus hc of the canonical product associated to (bn ) is bounded from below by 2hc . Suppose hc < 2hc . Since
an and therefore bn by continuity, we have that for all n sufficiently large |bn | > 1. So it suffices
to consider the case hc = 2hc 1. Then
X 1 X 1 X 1
> = 2hc 1+1
= hc
hc +1
n=1 |bn | n=1 |bn | n=1 |an |
But this shows that the genus of the canonical product associated to (an ) is at most hc 1, which is obviously
a contradiction. Taking f to be a polynomial shows that this bound is sharp.
I claim that the genus of f is bounded from above by 2h + 1. Indeed, 2h + 1 2 deg(g(z)) = deg(g(z 2 )), and
we showed above that hc 2hc + 1 2h + 1. This bound is also sharp since we can take
Y z 2
Y z z
f (z) = 1 f (z ) = 1 en
n=1
n2 n
n6=0
f (z) is clearly an entire function of genus 0, and the genus of the canonical product associated to (n)nZ is
1, from which we conclude the genus of f (z 2 ) is 1 .
5.2.4 Exercise 2
Using Legendres duplication formula for the gamma function (Ahlfors p. 200),
1
2
1 1 12 16 1 1 1 2
= 2 2 + = 2
3
6 6 6 2 3 3
Applying the formula (z)(1 z) = sin(z) (Ahlfors p. 199), we obtain
12 2
1 sin 31
2
1 1 1 3 1
= 2 3 =2 3
6 3 3 3
19
5.2.4 Exercise 3
It is clear from the definition of the Gamma function that for each k Z0 ,
(
1 + kz (z) k 6= 0
f (z) =
z(z) k=0
extends to a holomorphic function in an open neighborhood of k. We abuse notation and denote the extension
also by 1 + kz (z) and z(z). lemma For any k Z>0 ,
(z + k)
(z) = Qk z
/Z
j=1 (z + j 1)
Proof. Recall that (z) has the property that the (z + 1) = z(z). We proceed by induction. The base case
is trivial, so assume that (z) = Qk (z+k)
(z+j1)
for some k N. Then
j=1
Let k Z0 . Then
(1 kz )(z) (z k) (z)
Z Z Z
1 1 1
res (; k) = (z)dz = z dz = dz
2i |zk|= 12 2i |zk|= 21 1 k 2i |zk|= 21 zk
z
Since the function 1 + k (z) extends to a holomorphic function in a neighborhood of k, by Cauchys
integral formula,
(z k) (z) (1)k
Z
1
dz = (z k) (z)|z=k =
2i |zk|= 21 zk |k|!
where use the preceding lemma to obtain the last equality. Thus,
(1)k
res (; k) = k Z0
|k|!
5.2.5 Exercise 2
Lemma 5. Z
1
log dx =
0 1 e2x 12
log(1z)
Proof. Let 1 >> > 0. Consider the function z , which has the power series representation
log(1 z) 1 X 1 n X 1 n1
= z = z |z| < 1
z z n=1 n n=1
n
20
with the understanding that the singularity at z = 0 is removable. Since the convergence is uniform on
compact subsets, we may integrate over the contour : [0, 1 ] C, (t) = t term by term, Thus,
1
log(1 z) log(1 t) 2
Z Z X 1 n
X 1
dz = dt = (1 ) = ]
z 0 t n=1
n2 n=1
n2 6
xn
P
since the function f (x) = n=1 n2 is left-continuous at x = 1. Hence,
1
log(1 t) 2
Z X 1
dt = =
0 t n=1
n2 6
which gives Z Z
1
log 1 e2x dx =
log dx =
0 1 e2x 0 12
For x R>0 , Stirlings formula (Ahlfors p. 203-4) for (z) tells us that
1
(x) = 2xx 2 ex eJ(x)
where Z
1 x 1
J(x) = log d
0 2 + x2 1 e2
The preceding lemma tells us that
1 1 x2 1 1
Z Z
1 1 1 1 1
J(x) = 2 2
log 2
d log 2
d = =
x 0 x + 1e x 0 1e x 12 12x
x2
where weve used 0 < x2 + 2 1 . Set
(x) = 12xJ(x)
x2
It is obvious that (x) > 0 and (x) < 1 since x2 + 2 < 1 almost everywhere, and therefore the preceding
inequality is strict. We thus conclude that
1 (x)
(x) = 2xx 2 ex e 12x 0 < (x) < 1
5.2.5 Exercise 3
2
Take f (z) = ez , and for R >> 0, define
1 : [0, R] C, 1 (t) = t; 2 : [0, ] C, 2 (t) = Reit ; 3 : [0, R] C, 3 (t) = (R t)ei 4
4
and let be the positively oriented closed curve defined by the i .
Z Z Z
4 R cos(2t)iR sin(2t) 4
it
eR cos(2t) Rdt
f (z)dz =
e Rie dt
2 0 0
d
Since cos(2t) is nonnegative and cos(2t) 2t (this is immediate from dt cos(2t) = 2 sin(2t) 2 on [0, 4 ])
for t [0, 4 ], we have
Z Z
4 4 1 R
eR cos(2t) Rdt e2Rt Rdt =
e 2 1 0, R
0 0 2
21
Since f is an entire function, by Cauchys theorem,
Z Z Z Z
0= f (z)dz = f (z)dz + f (z)dz + f (z)dz
1 2 3
and letting R ,
Z Z R
Z R Z
x2 i (Rt)2 ei 2 i i(Rt)2 i 2
e dx = lim e 4 e dt = lim e 4 e dt = e 4 eiy dy
0 R 0 R 0 0
R 2
where we make the substitution y = R t. Substituting 0 ex dx = 21 ,
Z Z Z
2 2 ix2 i
cos(x )dx i sin(x )dx = e dx = e 4 = i
0 0 0 2 2 2 2 2
Equating real and imaginary parts, we obtain the Fresnel integrals
Z
cos(x2 )dx =
0 2 2
Z
sin(x2 )dx =
0 2 2
Entire Functions
5.3.2 Exercise 1
We will show that the following two definitions of the genus of an entire function f are equivalent:
1. If
m g(z)
Y z Ph 1
( azn )
j
f (z) = z e 1 e j=1 j
n=1
an
where h is the genus of the canonical product associated to (an ), then the genus of f is max (deg(g(z)), h).
If no such representation exists, then f is said to be of infinite genus.
2. The genus of f is the minimal h Z0 such that
P
z h j
e j=1 j ( an )
Y 1 z
m g(z)
f (z) = z e 1
n=1
an
where we g(z) is a polynomial of degree h1 . Hence, f is of finite genus h2 with respect to definition (2) and
h2 h1 .
Now suppose that f has finite genus h2 with respect to definition (2). Reversing the steps of the previous
argument, we attain that f has finite genus h1 with respect to definition (1), and h1 h2 . It follows
immediately that definitions (1) and (2) are equivalent if f has finite genus with respect to either (1) and
(2), and by proving the contrapositives, we see that (1) and (2) are equivalent for all entire functions f .
22
5.3.2 Exercise 2
lemma Let a C and r > 0. Then
Proof. By the triangle inequality and reverse inequality, we have the double inequality
Hence, inf |z |a|| |r |a|| and sup |z |a|| r + |a|. But these values are attained at z = r and z = r,
respectively.
By Weierstrasss extreme value theorem, |f | and |g| attain both their maximum and minimum on the
circle {|z| = r} at zM,f , zM,g and zm,f , zm,g , respectively. The preceding lemma shows that zM,g = r and
zm,g = r. Consider the expression
z m Q 1 z Y
f (z)
= n=1 an
= |z an |
n=1 |an ||
|z
g(z) m Q z
1
z n=1 |an |
we have that
f (zm,f ) f (zm,f ) Y |zm,f an | Y |zm,f an |
= = =1
g(zm,g ) g(r)
n=1
|r |an || |z
n=1 m,f
an |
Hence, |f (zm,f )| |g(zm,g )|.
5.5.5 Exercise 1
Let be a fixed region and F be the family of holomorphic functions f : C with Re(f (z)) > 0 z .
I claim that F is normal. Consider the family of functions
G = g : C : g = ef for some f F
Hence, G is uniformly bounded on compact subsets of and is therefore a normal family. Fix a sequence
(fn )nN F, and consider the sequence gn = efn . (gn ) has a convergent subsequence (gnk ) which converges
to a holomorphic function g on compact sets (Weierstrasss theorem). Since gnk is nonvanishing for each k,
g is either identically zero or nowhere zero by Hurwitz theorem. If g is identically zero, then it is immediate
that fnk tends to uniformly on compact sets. Now, suppose that g is nowhere zero. g(K) D \ {0}
is compact by continuity. By the Open Mapping Theorem, for each z K, there exists r > 0 such that
D(g(z); r) g() D \ {0}. The disks D(g(z); 41 r) form an open cover of g(K), so by compactness,
n
[ n
[ n
[
g(K) D(g(zi ); 41 ri ) D(g(zi ); 21 ri ) D(g(zi ); ri ) D \ {0}
i=1 i=1 i=1
23
On each D(g(zi ); ri ), we can choose a branch of the logarithm such that log(z) is holomorphic on D(g(zi ); ri ),
and in particular uniformly continuous on D(g(zi ); 21 ri ). For each i, choose i > 0 such that
Set = min1in i , choose k0 N such that k k0 |gnk (z) g(z)| < z K. Then for 1 i n,
k k0 log efnk (z) log(g(z)) < z g 1 D(g(zi ); 21 ri )
Sn
Since K i=1 g 1 D(g(zi ); 21 ri ) , we conclude from the uniqueness of limits that fnk (z) converges to
Suppose in addition that {Re(f ) : f F} is uniformly bounded on compact sets. I claim that F is then
locally bounded. Let K be compact, and let L > 0 be such that Re(f )(z) L z K f F. Then
f (z)
e = eRe(f (z)) eL z K f F
Hence, g = ef : f F is a locally bounded family, and therefore its derivatives are locally bounded. Since
Re(f ) > 0 f F, we have that
|f 0 (z)| f 0 (z)ef (z) = |g 0 (z)|
which shows that {f 0 : f F} is aSlocally bounded family. Since K is compact, there exist z1 , , zn K
n
and r1 , , rn > 0 such that K i=1 D(zi ; r2i ) and D(zi ; ri ) . By Cauchys theorem,
Z r r
i i
f (z) = f 0 (z)dz z D zi ; |f (z)| Mi ri z D zi ;
[zi ,z] 2 2
where [z, zi ] denotes the straight line segment, and Mi is a uniform bound for {f 0 : f F} on D(zi ; 21 ri ).
Setting M = max1in Mi and r = max1in ri , we conclude that
|f (z)| M r z K f F
Normal Families
5.5.5 Exercise 3
Let f : C C be an entire holomorphic function. Define a family of entire functions F by
F = {g : C C : g(z) = f (kz), k C}
Fix 0 r1 < r2 . I claim that F is normal (in the sense of Definition 3 p. 225) in the annulus
r1 < |z| < r2 if and only if f is a polynomial.
24
2|f 0 (z)|
is locally bounded. Since g(z) = f (kz) for some k C, it suffices to show that 1+|f (z)|2 is bounded on C.
The function F (z) given by
2 f 0 z 1
2 a1 z 2n + 2a2 z 2n1 + + nan z n+1
F (z) = 2 = 2n 2
1 + |f (z 1 )| |z| + |a0 z n + a1 z n1 + + an |
Now suppose that F is normal in r1 < |z| < r2 . If f is bounded, then were done by Liouvilles theorem.
Assume otherwise. Let (fn )nN F be a sequence given by
where n , n . Since F is normal, (fn ) has a subsequence (fnk )kN which either tends to ,
uniformly on compact subsets of {r1 < |z| < r2 }, or converges to some limit function g in likewise fashion.
Fix > 0 small and consider the compact subset {r1 + |z| r2 }. If fnk g, then I claim that f
is bounded on C, which gives us a contradiction. Indeed, fix z0 C. Since (fnk ) converges uniformly on
{r1 + |z| r2 }, (fnk ) is uniformly bounded by some M > 0 on this set. Let |nk (r1 + )| |z0 |. By
the Maximum Modulus Principle, |f (z)| is bounded on the disk D(0; |nk (r1 + )|) by some |f (w)| for some
w on the boundary. Hence,
and therefore,
|f (z)| inf |f (w)| > M
nk0 r|w|nk r
Since z was arbitrary, we conclude that |f (z)| > M |z| rnk0 . Since M > 0 was arbitrary, we conclude
that f has a pole at .
5.5.5 Exercise 4
Let F be a family of meromorphic functions in a given region , which is not normal in . By Ahlfors
Theorem 17 (p. 226), there must exist a compact set K such that the expression
2 |f 0 (z)|
(f )(z) = 2 f F
1 + |f (z)|
25
is not locally bounded on K. Hence, we can choose a sequence of functions (fn ) F and of points (zn ) K
such that
2 |fn0 (zn )|
2 % , n
1 + |fn (zn )|
Suppose for every z , there exists an open disk D(z; rz ) on which F is normal, equivalently (f ) is lo-
cally bounded. Let Mz > 0 bound (f ) on the closed disk D(z; 21 rz ). The collection D(z; 21 rz ) : z K
forms an open cover of K. By compactness, there exist finitely many disks D(z1 ; 21 r1 ), , D(zn ; 21 rn )
such that
[n
K D(zi ; 21 ri ) and i = 1, , n |(f )(z)| Mi z D(zi ; 21 r) f F
i=1
|(f )(z)| M z K f F
This is obviously a contradiction since limn (fn )(zn ) = +. We conclude that there must exist z0
such that F is not normal in any neighborhood of z0 .
26
Conformal Mapping, Dirichlets
Problem
u u
0 < f 0 (z0 ) = (z0 ) = (z0 ) = g 0 (z0 )
x x
6.1.1 Exercise 2
Suppose now that is symmetric with respect to z0 (i.e. z 2z0 z ). I claim that f satisfies
Define g : D by g(z) = f (2z0 z). Clearly, g is conformal, being the composition of conformal maps,
and g(z0 ) = 0. Furthermore, by the chain rule, g 0 (z0 ) = f 0 (z0 ) > 0. We conclude from the uniqueness
statement of the Riemann Mapping Theorem that g(z) = f (z) z .
27
Elliptic Functions
Weierstrass Theory
7.3.2 Exercise 1
Let f be an even elliptic function periods 1 , 2 . If f is constant then there is nothing to prove, so assume
otherwise. First, suppose that 0 is neither a zero nor a pole of f . Observe that since f is even, its zeroes
and poles occur in pairs. Since f is elliptic, f has the same number of poles as zeroes. So, let a1 , , an ,
and b1 , , bn denote the incongruent zeroes and poles of f in some fundamental parallelogram Pa , where
ai 6 aj mod M, bi 6 bj mod M i, j and where we repeat for multiplicity. Define a function g by
n
!1
Y (z) (ak )
g(z) = f (z)
(z) (bk )
k=1
and where is the Weierstrass p-function with respect to the lattice generated by 1 , 2 . I claim that g
is a holomorphic elliptic function. Since (z) (ak ) and (z) (bk ) have double poles at each z M
for all k, g has a removable singularity at each z M . For each k, (z) (bk ) has the same poles as
and is therefore an elliptic function of order 2. Since bk 6= 0 and is even, it follows that (z) (bk ) has
zeroes of order 1 at z = bk . From our convention for repeating zeroes and poles, we conclude that g has
a removable singularity at bk . The argument that g has removable singularity at each ak is completely
analogous. Clearly,
g(z + 1 ) = g(z + 2 ) = g(z) for z
/ ai + M bi + M M
so by continuity, we conclude that g is a holomorphic elliptic function with periods 1 , 2 and is therefore
equal to a constant C. Hence,
n
Y (z) (ak )
f (z) = C
(z) (bk )
k=1
Since f is even, its Laurent series about the origin only has nonzero terms with even powers. So if f vanishes
or has a pole at the origin, the order is 2m, m N. Suppose that f vanishes with order 2m. The function
given by
f(z) = f (z) (z)m
is elliptic with periods 1 , 2 . f has a removable singularity at z = 0, since (z)k has a pole of order 2k at
z = 0. Hence, we are reduced to the previous case of elliptic function, so applying the preceding argument,
we conclude that
n n
Y (z) (ak ) C Y (z) (ak )
f(z) = C f (z) =
(z) (bk ) (z)m (z) (bk )
k=1 k=1
28
7.3.2 Exercise 2
Let f be an elliptic function with periods 1 , 2 . By Ahlfors Theorem 5 (p. 271), f has the same number
of zeroes and poles counted with multiplicity. Let a1 , , an , b1 , , bn denote the incongruent
Pn zeroes and
poles of f , respectively, where
Pn we repeat for multiplicity. By Ahlfors Theorem p. 271, k=1
Pn b k ak M , so
replacing a1 by a01 = a1 + k=1 bk ak , we may assume without loss of generality that k=1 bk ak = 0.
Define a function g by
n
!1
Y (z ak )
g(z) = f (z)
(z bk )
k=1
n
!1 n 1 !1
Y (z ak + 1 ) Y (z ak )e1 (zak + 2 )
g(z + 1 ) = f (z + 1 ) = f (z) 1
(z bk + 1 ) (z b )e1 (zbk + 2 )
k
k=1 k=1
n
!1
1
Pn
ak bk
Y (z ak )
=e k=1 f (z) = g(z)
(z bk )
k=1
n
!1 n 2 !1
Y (z ak + 2 ) Y (z ak )e2 (zak + 2 )
g(z + 2 ) = f (z + 2 ) = f (z) 2
(z bk + 2 ) (z b )e2 (zbk + 2 )
k
k=1 k=1
n
!1
2
Pn
ak bk
Y (z ak )
=e k=1 f (z) = g(z)
(z bk )
k=1
By continuity, we conclude that g(z + 2 ) = g(z) z C. Since g is an entire elliptic function, it is constant
by Ahlfors Theorem 3 (p. 270). We conclude that for some C C,
n
Y (z ak )
f (z) = C
(z bk )
k=1
7.3.3 Exercise 1
Fix a rank-2 lattice M C and u
/ M . Then
(z u)(z + u)
(z) (u) =
(z)2 (u)2
Proof. I first claim that the RHS is periodic with respect to M . Let 1 , 2 be generators of M and let
1 2 2 1 = 2i. For z
/ M,
1 1 1
(z + 1 u)(z + 1 + u) (z u)e1 (zu+ 2 ) (z + u)e1 (z+u+ 2 )
(z u)(z + u)e21 (z+ 2 )
2 2
= 1 =
21 (z+ 21
(z + 1 ) (u) (z)2 e21 (z+ 2 ) (u)2 (z)2 (u)2 e )
(z u)(z + u)
= = f (z)
(z)2 (u)2
29
The argument for 2 is completely analogous. The RHS has zeroes at u and a double pole at 0. Hence, by
the same reasoning used above, we see that
(z u)(z + u)
(z) (u) = C for some C C
(z)2 (u)2
To find conclude that C = 1, we first note that (z) (u) has a coefficient of 1 for the z 2 term in its
Laurent expansion. If we show that the Laurent expansion of the f (z) also has a coefficient of 1 for the z 2 ,
then it follow from the uniquenuess of Laurent expansions that C = 1.
zu 2 z+u + 1 ( z+u )2
zu + 12 ( zu
) 1 z+u
(z 2 u2 )
Q Q
(z u)(z + u) 6=0 1 e e 2
6=0
2 2
= 2
(z) (u) z 1 z 2
Q
z 2 (u) 2
6=0 1 z e + 2 ( )
zu 2 Q 2
Q
1 zu
e
1
+2 ( zu
) 1 z+u
e
z+u 1
+2 ( z+u
)
6=0 6=0
= Q 2 2
e+2()
z 1 z
z
(u)2 6=0 1
| {z }
g1 (z)
zu 2 Q 2
2
Q
1 zu
1
+2 ( zu
) 1 z+u
z+u 1
+2 ( z+u
)
1 u 6=0 e 6=0 e
+ 2 2 2
z
Q
e+2()
z 1 z
z
(u)2 6=0 1
| {z }
g2 (z)
Observe that both g1 (z) and g2 (z) are holomorphic in a neighborhood of 0 since we have eliminated the
double pole at 0. Hence, the coefficient of the z 2 in the Laurent expansion of f (z) is given by g2 (0). But
since is an odd function, it is immediate that g2 (0) = 1.
7.3.3 Exercise 2
With the hypotheses of the preceding problem,
0 (z)
= (z u) + (z + u) 2(z)
(z) (u)
Proof. For z 6= u+M , we can choose a branch of the logarithm holomorphic in a neighborhood of (z)(u).
Taking the derivative of the log of both sides and using the chain rule,
0 (z)
log((z u)) + log((z u)) log((u)2 (z)2 )
=
(z) (u) z
0 (z u) 0 (z + u) 2 0 (z)
= + = (z u) + (z + u) 2(z)
(z u) (z + u) (z)
0 (w)
where weve used (w) = (w) w C (Ahlfors p. 274).
7.3.3 Exercise 3
With the same hypotheses as above, for z 6= u + M ,
1 0 (z) 0 (u)
(z + u) = (z) + (u) +
2 (z) (u)
Proof. Since the last term has a removable singularity at z = u + M , by continuity, we may also assume that
z 6= u + M . First, observe that by replacing switching u and z in the argument for the last identity, we have
that
0 (u)
= [(u z) + (z + u) 2(u)] = (z u) (z + u) + 2(u)
(z) (u)
30
0 (z)
where weve used the fact that (z) is odd and therefore (z) = (z) is also odd. Hence,
0 (z) 0 u
= ((z u) + (z + u) 2(z)) ((z u) (z + u) + 2(u)) = 2(z + u) 2(z) 2(u)
(z) (u)
7.3.3 Exercise 4
By Ahlfors Section 7.3.3 Exercise 3,
0 (z) 0 (u)
1
(z + u) = (z) + (u) +
2 (z) (u)
Differentiating both sides with respect to z and using 0 (w) = (w) w C \ M , we obtain
Applying the same arguments as above except taking u to be variable, we obtain that
Hence,
2 2
6((z)2 (u)2 ) (0 (z) 0 (u))2 0 (z) 0 (u)
1 1
2(z+u) = (z)+(u)+ = 2(z)+2(u)
2 (z) (u) ((z) (u))2 2 (z) (u)
2
0 (z) 0 (u)
1
(z + u) = (z) (u) +
4 (z) (u)
7.3.3 Exercise 5
Using the identity obtained in the previous exercise, we have by the continuity of that
" 2 #
1 0 (z) 0 (u)
(2z) = lim (z + u) = lim (z) (u) +
uz uz 4 (z) (u)
0 (z)0 (u)
!2 2
00 (z)
1 zu = 2(z) + 1
= lim (u) (z) +
uz 4 (z)(u) 4 0 (z)
zu
31
7.3.3 Exercise 7
Fix u, v
/ M such that |u| =
6 |v|, and define a function f : C \ M C by
(z) 0 (z)
1
f (z) = det (u) 0 (u) 1 = 0 (z)((u) (v)) + 0 (u)((z) (v)) + 0 (v)((z) (u))
(v) 0 (v) 1
where we use Laplace expansion for determinants. By our choice of u, v and the fact that the Weierstrass
function is elliptic of order 2, B 6= 0. Hence, f (z) is an elliptic function of order 3 with poles at the lattice
points of M . Since the determinant of any matrix with linearly dependent rows is zero, f has zeroes at u, v.
Since f has order 3, it has a third zero z, and by Abels Theorem (Ahlfors p. 271 Theorem 6),
uv+z 0 mod M z = v u
We conclude that
0 (z)
(z) 1
det (u) 0 (u) 1 = 0
(u + z) 0 (u + z) 1
7.3.5 Exercise 1
Since is invariant under (2) and \ (2) is generated by the linear fractional transformations 7 + 1
and 7 1 , it suffices to show that J( + 1) = J( ) and J( 1 ) = J( ). Recall that satisfies the
functional equations
( ) 1
( + 1) = and = 1 ( )
( ) 1
So,
Observe that 3
4 1 ( ) + ( )2 4 (( ) ei 3 )3 (( ) ei 3 )3
J( ) = =
27 ( )2 (1 ( ))2 27 ( )2 (1 ( ))2
So, J( ) assumes the value 0 on 1 ei 3 . Since is a bijection on 0 , J( ) has two zeroes, each
g2 = 4(e1 e2 + e1 e3 + e2 e3 ) = 0
2 2
for = ei 3 . So using the identity for J( ) proved below, J(ei 3 ) = 0. Using the invariance of J( ) under
, we see that J(ei 3 ) = 0.
J( ) assumes the value 1 on 1 ({1 , , 6 }), where the i are the roots of degree 6 the polynomial
32
Since e1 + e2 + e3 = 0 (see below for argument), we have e1 = e2 and therefore
e3 e2 1
(i) = =
e1 e2 2
Since each point in H+ is congruent modulo 2 to a point in 0 , maps this fundamental conformally
onto C \ {0, 1}, and J( ) is invariant under , we conclude J( ) assumes the value 1 at = i, 1 + i, i+1 2 . I
claim that these are these are the only possible points up to modulo 2 congruence. Suppose J( ) = 1 for
/ i, 1 +ni, i+1
2 . If
o we let S 1 , , S6 denote the complete set of mutually incongruent transformations, then
2
/ ei 3 , ei
since 3 (otherwise J( ) = 0), S1 , , S6 0 are distinct, hence the (Si ) are distinct
roots of p(z), and we obtain that p(z) has more than 6 roots, a contradiction. Moreover,
this argument shows
that the polynomial p(z) has three roots, which by inspection, we see are given by 1, 12 , 2 .
e3 e2
Substituting = e1 e2 , we have
3
4 1 (e3 e2 )(e1 e2 )1 + (e3 e2 )2 (e1 e2 )2
J( ) =
27 (e3 e2 )2 (e1 e2 )2 (e1 e3 )2 (e1 e2 )2
3
4 (e1 e2 )2 (e3 e2 )(e1 e2 ) + (e3 e2 )2
=
27 (e3 e2 )2 (e1 e3 )2 (e1 e2 )2
3
4 e21 2e1 e2 + e22 e3 e1 + e3 e2 + e2 e1 e22 + e23 2e3 e2 + e22
=
27 (e3 e2 )2 (e1 e3 )2 (e1 e2 )2
Since
0 = (e1 + e2 + e3 )2 = e21 + e22 + e23 + 2e1 e2 + 2e1 e3 + 2e2 e3 e21 + e22 + e33 = 2 (e1 e2 + e1 e3 + e2 e3 )
33