Section 4.

1 Sequences and Series of Complex Numbers 73

Solutions to Exercises 4.1
1. We have
0 |a
n
| =

e
in

2
n

=
1
n
0 as n .
So a
n
0 by the squeeze theorem or by applying (2), Sec. 4.1.
5. We have cosh in = cos n (see (25), Sec. 1.6). Hence
|a
n
| =

cosh in
n
2

cos n
n
2

1
n
2
0 as n .
So a
n
0 by the squeeze theorem.
9. (a) Suppose that {a
n
} is a convergent sequence and let b
n
= a
n+1
. We want to show that {b
n
}
is a convergent sequence and that lim
n
a
n
= lim
n
b
n
. Let L denote the limit of the sequence
{a
n
}. By denition of convergence, given > 0, we can nd N 0, such that for all n N,
|a
n
L| < .
But n N implies that n + 1 N. So for all n N,
|a
n+1
L| < .
But a
n+1
= b
n
, so for all n N,
|b
n
L| < ,
which implies that {b
n
} is a convergent sequence with same limit as {a
n
}.
(b) Dene a
1
= i and a
n+1
=
3
2+an
. Given that {a
n
} is convergent, then by part (a), we have
lim
n
a
n
= lim
n
a
n+1
. Let L = lim
n
a
n
. Then
L = lim
n
a
n
= lim
n
a
n+1
= lim
n
3
2 + a
n
=
3
2 + L
.
Solving for L, we nd
L =
3
2 + L
L(2 + L) = 3;
L
2
+ 2L 3 = 0, (L + 3)(L 1) = 0, L = 3 or L = 1.
Because the limit of a convergent sequence is unique, we have to decide whether L = 1 or L = 3.
Note that |a
1
| = 1. If we can show that Re a
n
0, then Re L = lim
n
Re a
n
0, and this would
eliminate the value L = 3. Lets prove by induction that Re a
n
0. The statement is clearly true
if n = 1, because a
1
= i and so Re i = 0 0. Now suppose that Re a
n
0 and lets prove that
Re a
n+1
0. For this purpose, we compute
Re a
n+1
= Re
_
3
2 + a
n
_
= 3 Re
_
2 + a
n
(2 + a
n
) (2 + a
n
)
_
=
3
(2 + a
n
) (2 + a
n
)
Re (2 + Re a
n
i Ima
n
) = 3
2
(2 + 2 Re a
n
) 0,
where
2
=
1
(2+an)2+an
=
1
|2+an|
2
. This completes the proof by induction and shows that L = 3,
so L = 1.
74 Chapter 4 Power Series and Laurent Series
13. The series

n=3
3 i
(1 + i)
n
is a constant multiple of a convergent geometric series and so it is
convergent. To nd its sum proceed as follows:

n=3
3 i
(1 + i)
n
= (3 i)
_
1
(1 + i)
3
+
1
(1 + i)
4
+
_
= (3 i)
1
(1 + i)
3
_
1 +
1
1 + i
+
1
(1 + i)
2
+
_
=
3 i
(1 + i)
3

n=0
1
(1 + i)
n
=
3 i
(1 + i)
3

1
1 r
,
where r =
1
1+i
, |r| =
1

2
< 1. So the sum is
S =
3 i
(1 + i)
3

1
1
1
1+i
=
3 i
(1 + i)
2
i
=
1 3i
(1 + i)
2
=
3
2
+
i
2
.
17. The series

n=2
1
(n + i)((n 1) + i)
is absolutely convergent by comparison to the series 2

1
n
2
:
|(n + i)((n 1) + i)| =

(n + i)

(n 1) + i

n(n 1)
1
2
n
2
if n 2;
so

1
(n + i)((n 1) + i)

2
n
2
.
To sum the series, use partial fractions and the terms will telescope, as follows:
1
(n + i)((n 1) + i)
=
1
(n 1) + i)

1
n + i
.
So the Nth partial sum is
s
N
=
N

n=2
1
(n 1) + i)
+
1
n + i
=
1
(2 1) + i)

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

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

1
N + i
=
1
2 + i

1
N + i

1
1 + i
0,
as N . So the sum is S =
1
1+i
=
1i
2
.
21. The series

n=0
_
1 + 3i
4
_
n
is a geometric series with z =
1+3i
4
. Since |z| =
1
4

10 < 1, the series

converges to
1
1 z
=
1
1
1+3i
4
=
4
3 3i
=
4
3
1
1 i
=
2
3
(1 + i).
25. Write
(1 + 2in)
n
n
n
=
_
1
n
+ 2i
_
n
.
Section 4.1 Sequences and Series of Complex Numbers 75
It is now clear that the series is a geometric series with z =
1
n
+2i. Since |z| | Imz| = 2, it follows
that the series diverges.
29. For n 2, we have
cos(in)
e
n
3
=
cosh n
e
n
3
=
e
n
+ e
n
2e
n
3
=
1
2
(e
n
3
+n
+ e
n
3
n
) e
n
3
/2
.
The series is absolutely convergent by comparison with the convergent series

n=1
e
n
3
/2
.
33. The series converges as long as

z
2

< 1 |z| < 2.

37. The series converges as long as

1
2 10z

< 1 1 < |2 10z|

1 <
1
10
|
1
5
z|
10 < |
1
5
z|.
Thus the series converges outside the closed disk of radius 10 and center at
1
5
. The limit in this
region is (use the geometric series minus the rst term, which is 1)
1
1
1
210z
1 =
2 10z
1 10z
1 =
1
1 10z
.
41. If the nth partial sum of the series is s
n
=
i
n
, then the series converges and its limit is
lim
n
s
n
= lim
n
i
n
= 0.
An example of such a series is
1 +

n=1
_
i
n + 1

i
n
_
.
45. The test of convergence in Theorem 12 is really about series with positive real terms. For a
proof, see any calculus book and use the same proof for real series.
49. To establish the addition formula for cosines, we will manipulate the series for cosine, using
several properties of absolutely convergent series and Cauchy products. We have
cos z
1
=

n=0
(1)
n
z
2n
1
(2n)!
; cos z
2
=

n=0
(1)
n
z
2n
2
(2n)!
;
cos z
1
cos z
2
=
n

k=0
(1)
k
z
2k
1
(2k)!
(1)
nk
z
2(nk)
2
(2(n k))!
=
(1)
n
(2n)!
n

k=0
(2n)!
(2k)!(2(n k))!
z
2k
1
z
2(nk)
2
=
(1)
n
(2n)!
n

k=0
_
2n
2k
_
z
2k
1
z
2(nk)
2
.
76 Chapter 4 Power Series and Laurent Series
Similarly,
sin z
1
=

n=0
(1)
n
z
2n+1
1
(2n + 1)!
; sin z
2
=

n=0
(1)
n
z
2n+1
2
(2n + 1)!
;
sin z
1
sin z
2
=
n

k=0
(1)
k
z
2k+1
1
(2k + 1)!
(1)
nk
z
2(nk)+1)
2
(2(n k) + 1)!
=
(1)
n
(2(n + 1))!
n

k=0
(2(n + 1))!
(2k)!(2(n + 1) (2k + 1)))!
z
2k+1
1
z
2(n+1)(2k+1)
2
=
(1)
n
(2(n + 1))!
n

k=0
_
2(n + 1)
2k + 1
_
z
2k+1
1
z
2(n+1)(2k+1)
2
.
Hence
cos z
1
cos z
2
sin z
1
sin z
2
=

n=0
(1)
n
(2n)!
n

k=0
_
2n
2k
_
z
2k
1
z
2(nk)
2

n=0
(1)
n
(2(n + 1))!
n

k=0
_
2(n + 1)
2k + 1
_
z
2k+1
1
z
2(n+1)(2k+1)
2
= 1 +

n=1
(1)
n
(2n)!
n

k=0
_
2n
2k
_
z
2k
1
z
2(nk)
2

n=1
(1)
n1
(2n)!
n1

k=0
_
2n
2k + 1
_
z
2k+1
1
z
2n(2k+1)
2
,
where in the rst series on the right, we wrote the rst term separately, and in the second series on
the right, we shifted the index by changing n to n 1. Combining the two series, we see that the
last displayed sum equals
1 +

n=1
(1)
n
(2n)!
_
n

k=0
_
2n
2k
_
z
2k
1
z
2(nk)
2
+
n1

k=0
_
2n
2k + 1
_
z
2k+1
1
z
2n(2k+1)
2
_
.
But the two inner sums in k add up to
2n

k=0
_
2n
k
_
z
k
1
z
2nk
2
= (z
1
+ z
2
)
n
,
because the rst sum adds the even-indexed terms and the second sum adds the odd-indexed terms.
Hence
cos z
1
cos z
2
sin z
1
sinz
2
= 1 +

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

n=0
(1)
n
(2n)!
(z
1
+ z
2
)
n
= cos(z
1
+ z
2
),
as desired.
Note: While this was a good exercise in Cauchy products, it is not the most ecient way to prove
the addition formula for the cosine. For a more elegant proof, see Example 2(b), Sec. 4.6.
Section 4.2 Sequences and Series of Functions 77
Solutions to Exercises 4.2
1. The sequence of functions f
n
(x) =
sin nx
n
converges uniformly on the interval 0 x . To see
this, let M
n
= max|0f
n
(x)| = max|f
n
(x)|, where the maximum is taken over all x in [0, ]. Then
M
n
=
1
n
. Since M
n
0, as n , it follows that f
n
converges uniformly to f = 0 on [0, ]. In
fact, we fact uniform convergence on the entire real line.
5. (a) and (b) First, let us determine the pointwise limit of the sequence of functions f
n
(x) =
nx
n
2
x
2
x + 1
. For x in the interval 0 x 1, we have
f
n
(x) =
nx
n
2
x
2
x + 1
=
nx
n
2
(x
2

x
n
2
+
1
n
2
=
x
n(x
2

x
n
2
+
1
n
2
0, as n .
Does the sequence converge to 0 uniformly for all x in [0, 1]? To answer this question we estimate
the maximum possible dierence between 0 and f
n
(x), as x varies in [0, 1]. For this purpose, we
compute M
n
= max|f
n
(x)| for x in [0, 1]. We have
f

n
(x) =
n n
3
x
2
(1 x + n
2
x
2
)
2
; f

n
(x) = 0 n n
3
x
2
= 0 x =
1
n
;
f
n
(
1
n
) =
1
2
1
n
>
1
2
M
n
>
1
2
.
Since M
n
does not converge to 0, we conclude that the sequence does not converge uniformly to 0
on [0, 1].
(c) The sequence does converge uniformly on any interval of the form [a, b], where 0 < a < b 1.
To see this, pick n so that 0 <
1
n
< a. Then, f
n
(x) < 0 for all a < x (check the sign of f

n
(x) if
1
n
< x. Hence f
n
(x) is decreasing on the interval [a, b]. So, if M
n
= max|f
n
(x)| for x in [a, b], then
0 M
n
|f
n
(a)|. But f
n
(a) 0, by part (a), so thus M
n
0, and so f
n
(x) converges uniformly
on [a, b].
9. The sequence f
n
(z) =
nz + 1
z + 2n
2
converges to 0 for all z, as we now show:
nz + 1
z + 2n
2
=
n(z +
1
n
)
n
2
(
z
n
2
+ 2)
=
z +
1
n
n(
z
n
2
+ 2)
0, as n .
(b) Let M
n
= max|f
n
(z)| for |z| 1, To prove that f
n
converges uniformly, we must show that
M
n
0, as n . We have
|f
n
(z)| =

z +
1
n
n(
z
n
2
+ 2)

|z| +
1
n
n(2
|z|
n
2
)

1 +
1
n
n(2
1
n
2
)
.
This shows that M
n
is smaller than the last displayed expression, which tends to 0 as n . Hence
f
n
does converge to 0 uniformly for all |z| 1.
13. If |z| 1, then

z
n
n(n + 1)

1
n(n + 1)
.
Apply the Weierstrass M-test with M
n
=
1
n(n+1)
. Since

M
n
is convergent, it follows that the
series

n=1
z
n
n(n + 1)
, converges uniformly for all |z| 1.
17. If |z| 2, then

_
z + 2
5
_
n

_
4
5
_
n
.
78 Chapter 4 Power Series and Laurent Series
Apply the Weierstrass M-test with M
n
=
_
4
5
_
n
. Since

M
n
is convergent (a geometric series with
r < 1), it follows that the series

n=0
_
z + 2
5
_
n
converges uniformly for all |z| 2.
21. If 2.01 |z 2| 2.9, then

(z 2)
n
3
n

_
2.9
3
_
n
= A
n
,
and

2
n
(z 2)
n

_
2
2.01
_
n
= B
n
.
Apply the Weierstrass M-test with M
n
= (A
n
+ B
n
). Since

M
n
is convergent (two geometric
series with ratios < 1), it follows that the series

n=0
_
(z 2)
n
3
n
+
2
n
(z 2)
n
_
converges uniformly in
the annular region 2.01 |z 2| 2.9.
25. (a) If |z
1
2
| <
1
6
then
|z| |z
1
2
+
1
2
| |z
1
2
| +
1
2
<
1
6
+
1
2
=
2
3
< 1.
So the series

n=0
z
n
converge uniformly on |z
1
2
| <
1
6
by the Weierstrass M-test with M
n
= (23)
n
.
(b) If |z
1
2
| <
1
2
, then z can get very close to the value 1, where the series

n=0
z
n
does not
converge. So the initial guess is that the series is not uniformly convergent in the region |z
1
2
| <
1
2
.
To prove this assertion, you can repeat the proof given in Example 4; in particular, the inequalities
in (3) and the argument that follows them still hold.
29. (a) Let > 1 be a positive real number. To show that the series
(z) =

n=1
1
n
z
(principal branch of n
z
)
converges uniformly on the half-plane H

= {z : Re z > 1}, we will apply the Weierstrass

M-test. For all z H

, we have
|n
z
| =

e
(x+iy) ln n

= e
xln n
= n
x
> n

.
So

1
n
z

1
n

= M
n
.
Since

M
n
=

1
n

is a convergent series (because > 1), it follows from the Weierstrass M-test
that that

n=1
1
n
z
converges uniformly in H

.
(b) Each term of the series

n=1
1
n
z
is analytic in H = {z : Re z > 1} (in fact, each term is
entire). To conclude that the series is analytic in H, it is enough by Corollary 2 to show that the
series converges uniformly on any closed disk contained in H. If S is a closed disk contained in H,
S is clearly disjoint from the imaginary axis. Let H

( > 0) be a half-plane containing S. By part

(a), the series converges uniformly on H

, consequently, the series converges uniformly on S. By

Corollary 2, the series is analytic in H. (Note the subtilty in the proof. We did not show that the
series converges uniformly on H. In fact, the series does not converge uniformly in H.)
(c) To compute

(z), according to Corollary 2, we can dierentiate the series term-by-term. Write

1
n
z
=
1
e
z ln n
= e
z ln n
.
Section 4.2 Sequences and Series of Functions 79
Using properties of the exponential function, we have
d
dz
1
n
z
=
d
dz
e
z ln n
= ln ne
z ln n
= ln n
1
n
z
.
So, for all z H,

(z) =

n=1
ln n
n
z
.
33. To show that f
n
{f
n
} converges uniformly on , it is enough to show that
max
z
|f
n
(z) f
m
(z)|
can be made arbitrarily small by choosing m and n large. In other words, given > 0, we must
show that there is a positive integer N such that if m, n N, then
max
z
|f
n
(z) f
m
(z)| < .
This will show that the sequence {f
n
} is uniformly Cauchy, and hence it is uniformly convergent by
Exercise 30.
Since each f
n
is analytic inside C and continuous on C, it follows that f
n
f
m
is also analytic
inside C and continuous on C. Since C is a simple closed path, the region interior to C is a bounded
region. By the maximum principle, Corollary 2, Sec. 3.7, the maximum value of |f
n
f
m
| occurs
on C. But on C the sequence {f
n
} is a Cauchy sequence, so there is a positive integer N such that
if m, n N, then
max
zC
|f
n
(z) f
m
(z)| < .
Hence
max
z
|f
n
(z) f
m
(z)| max
zC
|f
n
(z) f
m
(z)| < ,
which is what we want to prove.
The key idea in this exercise is that the maximum value of an analytic function occurs on the
boundary. So the uniform convergence of a sequence inside a bounded region can be deduced from
the uniform convergence of the sequence on the boundary of the region.
80 Chapter 4 Power Series and Laurent Series
Solutions to Exercises 4.3
1. Apply the ratio test: For z = 0,
= lim
n

a
n+1
a
n

= lim
n

(1)
n+1
z
n+1
2n + 1 + 1
(1)
n
2n + 1
z
n

= |z| lim
n
2n + 1
2n + 3
= |z| lim
n
n(2 +
1
n
)
n(2 +
3
n
)
= |z|.
Thus the series converges absolutely if |z| < 1 and diverges if |z| > 1. The radius of convergence is
1, the disk of convergence is |z| < 1, the circle of convergence is |z| = 1.
5. Apply the ratio test: For z =
2
4i
,

n=0
(4iz 2)
n
2
n
.
= lim
n

a
n+1
a
n

= lim
n

(4iz 2)
n+1
2
n+1

2
n
(4iz 2)
n

=
|4iz 2|
2
.
Thus the series converges absolutely if
|4iz 2|
2
< 1 |4i(z
2
4i
)| < 2 |z
1
2
(i)| <
2
4
|z
(i)
2
| <
1
2
.
The series diverges if |z
(i)
2
| >
1
2
. The radius of convergence is
1
2
; the disk of convergence is
|z
(i)
2
| <
1
2
; and the circle of convergence is |z
(i)
2
| =
1
2
.
9. We compute the radius of convergence by using the Cauchy-Hadamard formula
1
R
= limsup
n
_

(1 e
in

4
)
n

= limsup

1 e
in

= 2.
To understand why the limsup is equal to 2, recall that the limsup is the limit of the sup of the
tail of the sequence {

1 e
in

n=N
, as N tends to . The terms e
in

4
take values from the set
{

2
2

2
2
, i, 1}. So the largest value of

1 e
in

is 2, and this value repeats innitely often,

which explains the value of the limsup. Thus, R =
1
2
.
13. Start with the geometric series
1
1 z
=

n=0
z
n
, |z| < 1.
Dierentiate term-by-term:
1
(1 z)
2
=

n=1
nz
n1
, |z| < 1.
Multiply both sides by 2:
2
(1 z)
2
=

n=1
2nz
n1
, |z| < 1.
Section 4.3 Power Series 81
17. We have

n=0
(3z i)
n
3
n
=

n=0
(3(z
i
3
))
n
3
n
=

n=0
(z
i
3
)
n
=

n=0
w
n
(w = z
i
3
)
=
1
1 w
=
1
1 (z
i
3
)

z
i
3

< 1
=
3
3 + i 3z
,
which is valid for

z
i
3

< 1.
21. This exercise is a simple application of Theorem 15, Sec. 4.1, and basic properties of power
series. If

n=0
a
n
(z z
0
)
n
has radius of convergence R
1
> 0 and

n=0
b
n
(z z
0
)
n
has radius of
convergence R
2
> 0, then these series converge absolutely within their radii of convergence. Apply
Theorem 15, Sec. 4.1: If z is inside the disk of convergence of both series; that is, if |z z
0
| < R,
where R is the smallest of R
1
and R
2
, then
_

n=0
a
n
(z z
0
)
n
__

n=0
b
n
(z z
0
)
n
_
=

n=0
c
n
,
where
c
n
=
n

k=0
a
k
(z z
0
)
k
b
nk
(z z
0
)
nk
=
n

k=0
a
k
b
nk
(z z
0
)
n
= (z z
0
)
n
n

k=0
a
k
b
nk
= (a
0
b
n
+ a
1
b
n1
+ + a
n1
b
1
+ a
n
b
0
)(z z
0
)
n
.
Thus, for |z z
0
| < R,
_

n=0
a
n
(z z
0
)
n
__

n=0
b
n
(z z
0
)
n
_
=

n=0
(a
0
b
n
+ a
1
b
n1
+ + a
n1
b
1
+ a
n
b
0
)(z z
0
)
n
.
Cauchy products work nicely with power series. The Cauchy products of two power series,

n=0
a
n
(z
z
0
)
n
and

n=0
b
n
(z z
0
)
n
, centered at z
0
is another power series centered at z
0
,

n=0
c
n
(z z
0
)
n
.
Moreover, the coecient of (z z
0
) in the product series, c
n
, is the nth term in the Cauchy product
of the series

a
n
and

b
n
.
25. (a) In the formula, take z
1
= z
2
=
1
2
, then
[(
1
2
)]
2
= 2(1)
_
2
0
cos
11
sin
11
d = 2
_
2
0
d = 2

2
= ,
so (
1
2
) =

.
(b) In (9), let u
2
= t, 2udu = dt, then
(z) =
_

0
t
z1
e
t
dt =
_

0
u
2(z1)
e
u
2
(2u)du = 2
_

0
u
2z1
e
u
2
du.
(c) Using (b)
(z
1
)(z
2
) = 2
_

0
u
2z11
e
u
2
du2
_

0
v
2z21
e
v
2
dv
= 4
_

0
_

0
e
(u
2
+v
2
)
u
2z11
v
2z21
dudv.
82 Chapter 4 Power Series and Laurent Series
(d) Switching to polar coordinates: u = r cos , v = r sin , u
2
+ v
2
= r
2
, dudv = rdrd; for (u, v)
varying in the rst quadrant (0 u < and 0 v < ), we have 0

2
, and 0 r < , and
the double integral in (c) becomes
(z
1
)(z
2
) = 4
_

0
_
2
0
e
r
2
(r cos )
2z11
(r sin )
2z21
rdrd
= 2
_
2
0
(cos )
2z11
(sin )
2z21
d
=(z1+z2)
..
2
_

0
r
2(z1+z2)1
e
r
2
dr
(use (b) with z
1
+ z
2
in place of z)
= 2(z
1
+ z
2
)
_
2
0
(cos )
2z11
(sin )
2z21
d,
implying (d).
Section 4.4 Taylor Series 83
Solutions to Exercises 4.4
1. According to Theorem 1, the Taylor series around z
0
converges in the largest disk, centered at
z
0
= 0, in which the function is analytic. Clearly, e
z1
is entire, so radius of convergence is R = .
5. According to Theorem 1, the Taylor series around z
0
converges in the largest disk, centered at
z
0
= 2 + i, in which the function is analytic. The function f(z) =
z + 1
z i
is analytic for all z = i. So
the largest disk around z
0
on which f is analytic has radius R = |z
0
i| = |2 + i i| = 2.
9. According to Theorem 1, the Taylor series around z
0
converges in the largest disk, centered at
z
0
= 0, in which the function is analytic. The function f(z) = tan z is analytic for all z =

2
+ 2k.
So the largest disk around z
0
on which f is analytic has radius equal to the distance from z
0
= 0 to
the nearest point where f fails to be analytic. Clearly, then R = |0

2
| =

2
.
13. Arguing as we did in Exercises 1-9, we nd that the Taylor series of f(z) =
z
1 z
around
z
0
= 0 has radius of convergence equal to the distance from z
0
= 0 to the nearest point where f
fails to be analytic. Thus R = 1. (This will also come out of the computation of the Taylor series.)
Now, for |z| < 1, the geometric series tells us that
1
1 z
=

n=0
z
n
.
Multiplying both sides by z, we get, for |z| < 1,
z
1 z
=

n=0
z
n+1
.
17. Because the function is entire, the Taylor series will have an innite radius of convergence. Note
that the series expansion around 0 is easy to obtain:
e
z
=

n=0
z
n
n!
ze
z
= z

n=0
z
n
n!
=

n=0
z
n+1
n!
.
But how do we get the series expansion around z
0
= 1? In the previous expansion, replacing z by
z 1, we get
(z 1)e
z1
=

n=0
(z 1)
n+1
n!
.
The expansion on the right is a Taylor series centered at z
0
= 1, but the function on the left is not
quite the function that we want. Let f(z) = ze
z
. We have
(z 1)e
z1
= e
1
ze
z
e
z1
= e
1
f(z) e
z1
.
So f(z) = e
_
(z 1)e
z1
+ e
z1

. Using the expansion of (z 1)e

z1
and the expansion of e
z1
=

n=0
(z1)
n
n!
, we nd
f(z) = e
_

n=0
(z 1)
n+1
n!
+

n=0
(z 1)
n
n!

= e
_

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

n=0
(z 1)
n
n!

= e
_

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

n=1
(z 1)
n
n!

= e
_
1 +

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

= e
_
1 +

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

= e
_
1 +

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

.
84 Chapter 4 Power Series and Laurent Series
21. (a) To obtain the partial fractions decomposition
1
(1 z)(2 z)
=
1
1 z

1
2 z
,
we proceed in the usual way:
1
(1 z)(2 z)
=
A
1 z
+
B
2 z
=
A(2 z) + B(1 z)
(1 z)(2 z)
;
1 = A(2 z) + B(1 z)
Take z = 2 1 = B, B = 1.
Take z = 1 1 = A.
Thus we obtain the desired partial fractions decomposition. Expanding each term in the partial
fractions decomposition around z
0
= 0, we obtain
1
1 z
=

n=0
z
n
, |z| < 1;

1
2 z
=
1
2(1
z
2
)
=
1
2

n=0
_
z
2
_
n
, |
z
2
| < 1, or |z| < 2.
So, for |z| < 1,
1
(1 z)(2 z)
=
1
1 z

1
2 z
=

n=0
(1
1
2
n+1
)z
n
.
(b) We an derive the series in (a) by considering the Cauchy products of the series expansions of
1
1z
and
1
2z
, as follows. From (a), we have
1
1 z

1
2 z
=

n=0
z
n

n=0
z
n
2
n+1
=

n=0
c
n
z
n
,
where c
n
is obtained from the Cauchy product formula (see Exercise 21, Sec. 4.3):
c
n
=
n

k=0
a
k
b
nk
,
a
k
= 1, b
nk
=
1
2
nk+1
,
c
n
=
n

k=0
1
2
nk+1
=
1
2
n+1
n

k=0
2
k
.
(c) To show that the Cauchy product is the same as the series that we found in (a), we must prove
that
1
2
n+1
n

k=0
2
k
= (1
1
2
n+1
).
But this is clear since
n

k=0
2
k
= 1 + 2 + 2
2
+ + 2
n
= 2
n+1
1,
Section 4.4 Taylor Series 85
and so
1
2
n+1
n

k=0
2
k
=
1
2
n+1
(2
n+1
1) = (1
1
2
n+1
).
The radius of the Maclaurin series is 1. This follows from our argument in (a) or from Theorem 1,
since the function has a problem at z = 1.
25. The easiest way to do this problem is to start with the series expansion of f(z) =
1
z 2i
and
then dierentiate it term-by-term, twice. Lets see:
1
z 2i
=
1
2i(
z
2i
1)
=
i
2(
z
2i
1)
=
i
2

1
1
z
2i
=
i
2

n=0
_
z
2i
_
n
=
i
2

n=0
(i)
n
z
n
2
n
,
which is valid for |
z
2i
| < 1 or |z| < 2. Within the radius of convergence, the series can be dierentiated
term by term as often as we wish. Dierentiating once, we obtain
1
(z 2i)
2
=
i
2

n=1
(i)
n
n
z
n1
2
n
.
Dierentiating a second time, we obtain
2
(z 2i)
3
=
i
2

n=2
(i)
n
n(n 1)
z
n2
2
n

1
(z 2i)
3
=
i
4

n=2
(i)
n
n(n 1)
z
n2
2
n
.
All series are valid in |z| < 2. If we shift the index of summation of the series (change n to n + 2),
we obtain the series
i
4

n=0
(i)
n+2
(n + 2)(n + 1)
z
n
2
n+2
=
i
4

n=0
(i)
n
(n + 2)(n + 1)
z
n
2
n+2
.
29. For all z, we have
e
z
2
=

n=0
(z
2
)
n
n!
=

n=0
z
2n
n!
= 1 + z
2
+
z
4
2!
+ .
Hence, for all z,
e
z
2
1 = z
2
+
z
2
2!
+ .
If z = 0, we have
e
z
2
1 = z
2
+
z
4
2!
+
z
6
3!
+ = z
2
(1 +
z
2
2!
+
z
4
3!
+ ),
so, if z = 0,
e
z
2
1
z
2
=
z
2
(1 +
z
2
2!
+
z
4
3!
+ )
z
2
= 1 +
z
2
2!
+
z
4
3!
+ ,
where the series converges for all z. But a power series is analytic in the disk where it converges. So
the series 1 +
z
2
2!
+
z
4
3!
+ =

n=0
z
2n
(n+1)!
is entire. Call g(z) =

n=0
z
2n
(n+1)!
. We just proved that
for z = 0, g(z) =
e
z
2
1
z
2
= f(z). Now for z = 0, we have g(0) = 1 +
0
2
2!
+
0
4
3!
+ = 1 = f(0). Thus
f(z) = g(z) for all z and since g is entire, it follows that f is entire and its Maclaurin series is g(z).
86 Chapter 4 Power Series and Laurent Series
33. (a) The sequence of integers {l
n
} satises the recurrence relation l
n
= l
n1
+ l
n2
for n 2,
with l
0
= 1 and l
1
= 3. As suggested, suppose that l
n
occur as the Maclaurin coecient of some
analytic function f(z) =

n=0
l
n
z
n
, |z| < R. To derive the given identity for f, multiply the series
by z and z
2
, and then use the recurrence relation for the coecients. Using l
0
= 1 and l
1
= 3, we
obtain
f(z) =

n=0
l
n
z
n
= 1 + 3z +

n=2
l
n
z
n
;
zf(z) =

n=0
l
n
z
n+1
=

n=1
l
n1
z
n
= l
0
z +

n=2
l
n1
z
n
;
zf(z) = z +

n=2
l
n1
z
n
;
z
2
f(z) =

n=0
l
n
z
n+2
=

n=2
l
n2
z
n
.
Using the recurrence relation and the preceding identities, we obtain
f(z) = 1 + 3z +

n=2
l
n
z
n
= 1 + 3z +

n=2
_
l
n1
+ l
n2
_
z
n
= 1 + 3z +
zf(z)z
..

n=2
l
n1
z
n
+
z
2
f(z)
..

n=2
l
n2
z
n
= 1 + 3z + zf(z) z + z
2
f(z) = 1 + 2z + zf(z) + z
2
f(z).
Solving for f(z), we obtain
f(z) =
1 + 2z
1 z z
2
.
(b) To compute the Maclaurin series of f, we will use the result of Exercise 22:
1
(z
1
z)(z
2
z)
=
1
z
1
z
2

n=0
(z
n+1
1
z
n+1
2
)
(z
1
z
2
)
n+1
z
n
, |z| < |z
1
|, z
1
= z
2
, |z
1
| |z
2
|.
To derive this identity, start with the partial fractions decomposition
1
(z
1
z)(z
2
z)
=
1
z
1
z
2
_
1
z
2
z

1
z
1
z
_
=
1
z
1
z
2
_
1
z
2
(1
z
z2
)

1
z
1
(1
z
z1
)
_
.
Apply a geometric series expansion and simplify:
1
(z
1
z)(z
2
z)
=
1
z
1
z
2
_
1
z
2

n=0
_
z
z
2
_
n

1
z
1

n=0
_
z
z
1
_
n
_
=
1
z
1
z
2

n=0
_
1
z
n+1
2

1
z
n+1
1
_
z
n
=
1
z
1
z
2

n=0
(z
n+1
1
z
n+1
2
)
(z
1
z
2
)
n+1
z
n
.
Section 4.4 Taylor Series 87
Now, consider the function
1
1 z z
2
=
1
z
2
+ z 1
=
1
(z
1
z)(z
2
z)
,
where z
1
and z
2
are the roots of z
2
+ z 1:
z
1
=
1 +

5
2
and z
2
=
1

5
2
,
arranged so that z
1
| < |z
2
|. These roots satisfy known relationships determined by the coecients
of the polynomial z
2
+ z 1. We will need the following easily veried identities:
z
1
z
2
=

5 and z
1
z
2
= 1.
We will also need the following identities:
z
n
1
(z
1
2) = (1)
n
_
1

5
2
_
n
_
2 +
1 +

5
2
_
= (1)
n
_
1

5
2
_
n
_
5 +

5
2
_
= (1)
n

5
_
1

5
2
_
n+1
.
Similarly,
z
n
2
(2 z
2
) = (1)
n
_
1 +

5
2
_
n
_
2 +
1 +

5
2
_
= (1)
n
_
1 +

5
2
_
n
_
5 +

5
2
_
= (1)
n
_
1 +

5
2
_
n

5
_
1 +

5
2
_
= (1)
n

5
_
1 +

5
2
_
n+1
.
We are now ready to derive the desired Maclaurin series. We have
1
1 z z
2
=
1
z
2
+ z 1
=
1
(z
1
z)(z
2
z)
=
1

n=0
z
n+1
1
z
n+1
2
(1)
n+1
z
n
=
1

n=0
(1)
n+1
_
z
n+1
1
z
n+1
2
_
z
n
=
1

5
=

5
..
(z
1
z
2
)(1) +
1

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

n=1
(1)
n
_
z
n+1
1
z
n+1
2
_
z
n
;
2z
1 z z
2
=
2

n=0
(1)
n
_
z
n+1
1
z
n+1
2
_
z
n+1
=
2

n=1
(1)
n
_
z
n
1
z
n
2
_
z
n
.
88 Chapter 4 Power Series and Laurent Series
So
f(z) =
1 + 2z
1 z z
2
= 1 +
1

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

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

n=1
_
_
_
(1)
n
(1)
n

5
_
_
_
1 +

5
2
_
n+1
+
_
1

5
2
_
n+1
_
_
z
n
_
_
_
= 1 +

n=1
_
_
_
1 +

5
2
_
n+1
+
_
1

5
2
_
n+1
_
_
z
n
.
Thus
l
n
=
_
1 +

5
2
_
n+1
+
_
1

5
2
_
n+1
, n 0.
37. We use the binomial series expansion from Exercise 36, with =
1
2
. Accordingly, for |z| < 1,
(1 + z)
1
2
=

n=0
_
1
2
n
_
z
n
,
where, for n 1,
_
1
2
n
_
=
1
2
(
1
2
1) (
1
2
n + 1)
n!
=
1
2
3
2

(2n3)
2
n!
= (1)
n1
1
2
3
2

(2n3)
2
n!
= (1)
n1
1
2
n
n!
1 3 5 (2n 3) 2 4 (2n 2)
2 4 (2n 2)
= (1)
n1
1
2
n
n!
(2n 2)!
2 1 2 2 2 (n 1)
= (1)
n1
1
2
n
n!
(2n 2)!
2
n1
1 2 (n 1)
= = (1)
n1
1
2
n
n!
(2n 2)!
2
n1
(n 1)!
= (1)
n1
1
2
n
n!
(2n 2)!
2
n1
(n 1)!
2n(2n 1)
2n(2n 1)
= (1)
n1
1
2
n
n!
(2n)!
2
n
n!(2n 1)
(1)
n1
(2n)!(2n 1)
2
2n
(n!)
2
= (1)
n1
(1)
n1
2
2n
(2n 1)
_
2n
n
_
.
Thus
(1 + z)
1
2
=

n=0
(1)
n1
2
2n
(2n 1)
_
2n
n
_
z
n
, |z| < 1.
41. (a) and (b) There are several possible ways to derive the Taylor series expansion of f(z) = Log z
about the point z
0
= 1 + i. Here is one way. Let z
0
= 1 +i, so |z
0
| =

2. The function Log z is

analytic except on the negative real axis and 0. So it is guaranteed by Theorem 1 to have a series
expansion in the largest disk around z
0
that does not intersect the negative real axis. Such a disc,
as you can easily verify, has radius Imz
0
= 1. However, as you will see shortly, the series that we
obtain has a larger radius of convergence, namely |z
0
| =

2 (of course, this is not a contradiction
to Theorem 1).
Section 4.4 Taylor Series 89
Consider the function Log z in B
1
(z
0
), where it is analytic. (The disk of radius 1, centered at z
0
is contained in the upper half-plane.) For z B
1
(z
0
), we have
d
dz
Log z =
1
z
. Instead of computing
the Taylor series of Log z directly, we will rst compute the Taylor series of
1
z
, and then integrate
term-by-term within the radius of convergence of the series (Theorem 3, Sec. 4.3). Getting ready
to apply the geometric series result, we write
1
z
=
1
z
0
(z
0
z)
=
1
z
0
_
1
z0z
z0
_ (z
0
= 0)
=
1
z
0

1
1
z0z
z0
=
1
z
0

n=0
_
z
0
z
z
0
_
n
=
1
z
0

n=0
(1)
n
_
z z
0
_
n
z
n
0
,
where the series expansion holds for

z
0
z
z
0

< 1 |z
0
z| < |z
0
|.
Thus the series representation holds in a disk of radius |z
0
| =

2, around z
0
. Within this disk, we
can integrate the series term-by-term and get
_
z
z0
1

d =
1
z
0

n=0
(1)
n
z
n
0
_
z
z0
_
z
0
_
n
d =
1
z
0

n=0
(1)
n
(n + 1)z
n
0
_
z z
0
_
n+1
.
Reindexing the series by changing n + 1 to n, we obtain
_
z
z0
1

d =

n=1
(1)
n+1
nz
n
0
_
z z
0
_
n
|z z
0
| < |z
0
| =

2.
Now we have to decide what to write on the left side. The function Log z is an antiderivative of
1
z
in the disk of radius 1, centered at z
0
. (Remember that Log z is not analytic on the negative real
axis, so we cannot take a larger disk.) So, for |z z
0
| < 1, we have
_
z
z0
1

d = Log

z
z0
= Log z Log z
0
.
Thus, for |z z
0
| < 1, we have
Log z = Log z
0
+

n=1
(1)
n+1
nz
n
0
_
z z
0
_
n
,
even though the series on the right converges in the larger disk |z z
0
| <

2.
90 Chapter 4 Power Series and Laurent Series
Solutions to Exercises 4.5
1. Apply (5) with w = z
2
and get
1
1 + z
2
=

n=1
1
(z
2
)
n
, 1 < |z
2
|;
equivalently,
1
1 + z
2
=

n=1
(1)
n+1
z
2n
, 1 < |z|.
It is worth doing this problem without appealing to formula (5). Since

1
z
2

< 1, we can use a

geometric series in
1
z
2
as follows. We have
1
1 + z
2
=
1
z
2
Expand using
a geometric series.
..
1
1
1
z
2
=
1
z
2

n=0
(
1
z
2
)
n
=
1
z
2

n=0
(1)
n
z
2n
=

n=0
(1)
n
z
2(n+1)
=

n=1
(1)
n1
z
2n
,
where in the last series we shifted the index of summation by 1. Note that (1)
n1
= (1)
n+1
, and
so the two series that we derived are the same.
5. We have
Log (1 + w) =

n=1
(1)
n1
n
w
n
, |w| < 1.
Put w =
1
z
, then
Log (1 +
1
z
) =

n=1
(1)
n1
n
_
1
z
_
n
, |
1
z
| < 1;
equivalently
Log
_
1 +
1
z
_
=

n=1
(1)
n1
n
1
z
n
, 1 < |z|.
9. Simplify the function by using sin 2a = 2 sin a cos a, and get
sin
1
z
cos
1
z
z
=
1
z
1
2
sin
2
z
=
1
2z
sin
2
z
.
Start with the Maclaurin series for sin w: for all w,
sin w =

n=0
(1)
n
(2n + 1)!
w
2n+1
.
For z = 0, put w =
2
z
,
sin
2
z
=

n=0
(1)
n
(2n + 1)!
_
2
z
_
2n+1
=

n=0
(1)
n
2
2n+1
(2n + 1)!z
2n+1
.
Section 4.5 Laurent Series 91
To get the desired expansion, multiply by
1
2z
: for z = 0,
1
2z
sin
2
z
=
1
2z

n=0
(1)
n
2
2n+1
(2n + 1)!z
2n+1
=

n=0
(1)
n
2
2n
(2n + 1)!z
2(n+1)
.
13. Applying a partial fractions decomposition, we nd that
z
(z + 2)(z + 3)
=
3
z + 3

2
z + 2
.
In the annulus 2 < |z| < 3, we have

2
z

< 1 and

z
3

< 1. So to expand
3
z+3
, factor the 3 in the
denominator and youll get
3
z + 3
=
3
3(1 +
z
3
)
=
1
1 (
z
3
)
,
where

z
3

< 1. So we can apply a geometric series expansion and obtain:

3
z + 3
=
1
1 (
z
3
)
=

n=0
_
z
3
_
n
=

n=0
(1)
n
z
n
3
n
.
This series expansion is valid for |z| < 3; so, in particular, it is valid in the annulus 2 < |z| < 3. To
expand
2
z+2
, in the annulus 2 < |z| < 3, because

2
z

< 1, we divide the denominator by z and get

2
z + 2
=
2
z(1 +
2
z
=
2
z
1
1
2
z
.
Expand, using a geometric series, which is valid for |2/z| < 1 or 2 < |z|, and get
2
z + 2
=
2
z
1
1
2
z
=
2
z

n=0
(
2
z
)
n
=

n=0
(1)
n
2
n+1
z
n+1
=

n=1
(1)
n1
2
n
z
n
,
which is valid for 2 < |z|. Hence, for 2 < |z| < 3,
z
(z + 2)(z + 3)
=
3
z + 3

2
z + 2
=

n=0
(1)
n
z
n
3
n

n=1
(1)
n1
2
n
z
n
=

n=0
(1)
n
z
n
3
n
+

n=1
(1)
n
2
n
z
n
.
17. First, derive the partial fractions decomposition
z
2
+ (1 i)z + 2
(z i)(z + 2)
= 1 +
1
z i

2
z + 2
.
(The rst step should be to reduce the degree of the numerator by dividing it by the denominator.
As in Exercise 13, we handle each term separately, the constant term is to be left alone for now.
92 Chapter 4 Power Series and Laurent Series
In the annulus 1 < |z| < 2, we have

1
z

< 1 and

z
2

< 1. So to expand
1
zi
, factor the z in the
denominator and youll get
1
z i
=
1
z(1
i
z
)
=
1
z
1
1
i
z
,
where

i
z

< 1 or 1 < |z|. Apply a geometric series expansion: for 1 < |z|,
1
z i
=
1
z
1
1
i
z
=
1
z

n=0
_
i
z
_
n
=

n=0
i
n
z
n+1
=

n=1
i
n1
z
n
.
To expand
2
z+2
, in the annulus 1 < |z| < 2, because

z
2

< 1, we divide the denominator by 2 and

get
2
z + 2
=
2
2(1 +
z
2
=
1
1
z
2
.
Expand, using a geometric series, which is valid for |z| < 2, and get
2
z + 2
=
1
1
z
2
=

n=0
(
z
2
)
n
=

n=0
(1)
n
z
n
2
n
.
Hence, for 1 < |z| < 2,
1 +
1
z i

2
z + 2
= 1

n=1
i
n1
z
n
+

n=0
(1)
n
z
n
2
n
.
21. The function f(z) =
1
(z 1)(z + i)
has isolated singularities at z = 1 and z = i. If we
start at the center z
0
= 1, the closest singularity is i and its distance to z
0
is

2. Thus f(z) is
analytic in the disk of radius

2 and center at z
0
= 1, which is the annulus |z +1| <

2. Moving
outside this disk, we encounter the second singularity at z = 1. Thus f(z) is analytic in the annulus

2 < |z + 1| < 2, and has a Laurent series representation there. Finally, the function is analytic in
the annulus 2 < |z + 1| and so has a Laurent expansion there.
We now derive the three series expansions. Using a partial fractions decomposition, we have
f(z) =
1
(z 1)(z + i)
=
A
z 1

A
z + i
,
where A =
1
2

i
2
=
1
2
(1 i). We have, for |z + 1| < 2,
1
z 1
=
1
2 + (z + 1)
=
1
2
1
1
z+1
2
=
1
2

n=0
_
z + 1
2
_
n
.
For |z + 1| <

2, we have

z+1
1i

< 1, and so
1
z + i
=
1
(i 1) + (z + 1)
=
1
1 i
1
1
z+1
1i
=
1
1 i

n=0
_
z + 1
1 i
_
n
.
Section 4.5 Laurent Series 93
Thus, for |z + 1| <

2, we have
f(z) =
1
(z 1)(z + i)
=
A
z 1

A
z + i
=
A
2

n=0
_
z + 1
2
_
n
+
A
1 i

n=0
_
z + 1
1 i
_
n
=
1 i
4

n=0
_
z + 1
2
_
n
+
1
2

n=0
_
z + 1
1 i
_
n
.
For

2 < |z + 1|, we have

1i
z+1

< 1, and so
1
z + i
=
1
(i 1) + (z + 1)
=
1
z + 1
1
1
1i
z+1
=
1
z + 1

n=0
_
1 i
z + 1
_
n
=

n=0
(1 i)
n
(z + 1)
n+1
=
1
1 i

n=1
(1 i)
n
(z + 1)
n
.
So, if

2 < |z + 1| < 2, then
f(z) =
1
(z 1)(z + i)
=
A
z 1

A
z + i
=
1 i
4

n=0
_
z + 1
2
_
n

1 i
2

n=0
(1 i)
n
(z + 1)
n+1
.
Finally, for 2 < |z + 1|, we have
1
z 1
=
1
2 + (z + 1)
=
1
z + 1
1
1
2
z+1
=
1
z + 1

n=0
_
2
z + 1
_
n
=
1
2

n=1
2
n
(z + 1)
n
.
So, if 2 < |z + 1|, then
f(z) =
1
(z 1)(z + i)
=
A
z 1

A
z + i
=
1 i
2

n=0
2
n
(z + 1)
n+1

1 i
2

n=0
(1 i)
n
(z + 1)
n+1
=
1 i
4

n=1
2
n
(z + 1)
n

1
2

n=1
(1 i)
n
(z + 1)
n
.
25. In this problem, the idea is to evaluate the integral by integrating a Laurent series term-
by-term. This process is justied by Theorem 1, which asserts that the Laurent series converges
absolutely and uniformly on any closed and bounded subset of its domain of convergence. Since
a path is closed and bounded, if the path lies in the domain of convergence of the Laurent series,
then the series converges uniformly on the path. Hence, by Corollary 1, Sec. 4.2, the series can be
94 Chapter 4 Power Series and Laurent Series
dierentiated term-by-term. We now present the details of the solution. Using the Maclaurin series
of sin z, we have for all z = 0,
sin
1
z
=

n=0
(1)
n
(2n + 1)!
z
(2n+1)
.
Thus
_
C1(0)
sin
1
z
dz =
_
C1(0)
_

n=0
(1)
n
(2n + 1)!
z
(2n+1)
_
dz =

n=0
(1)
n
(2n + 1)!
_
C1(0)
z
(2n+1)
dz.
We now recall the important integral formula: for any integer m:
_
C)
z
m
dz =
_
2i if m = 1,
0 otherwise,
where C is any positively oriented simple closed path containing 0 (see Example 4, Sec. 3.4). Thus,
_
C1(0)
z
(2n+1)
dz =
_
2i if n = 0,
0 otherwise.
Hence all the terms in the series

n=0
(1)
n
(2n + 1)!
_
C1(0)
z
(2n+1)
dz
are 0, except the term that corresponds to n = 0, which is equal to 2i. So
_
C1(0)
sin
1
z
dz = 2i.
29. We follow the same strategy as in Exercise 25 and use the series expansion from Exercise 5.
We have
_
C4(0)
Log
_
1 +
1
z
_
dz =
_
C4(0)
_

n=1
(1)
n1
n
1
z
n
_
dz
=

n=1
(1)
n1
n
_
C4(0)
1
z
n
dz
= 2i,
where we have used the fact that
_
C4(0)
1
z
n
dz = 2i if n = 1 and 0 otherwise.
33. (a) Consider the power series in (1), where the coecients are given by (2). Since f(z) and
1
(zz0)
n+1
are analytic in the annulus A
R1,R2
(z
0
), the path in the integral dening the coecients
in (2), C
R
(z
0
), can be replaced by any positively oriented path C
r
(z
0
), where R
1
< r < R
2
. The
reason for this is that C
r
(z
0
) and C
R
(z
0
) are mutually deformable in A
R1,R2
(z
0
). Fix z, r
1
and
r
2
, such that z A
R1,R2
(z
0
) is arbitrary but xed, R
1
< r
1
< |z z
0
| < r
2
< R
2
. Let M be the
maximum value of |f()| for in the closed annular region r
1
| z
0
| r
2
. Note that because
f is analytic, hence continuous, in this annular region, M is nite. Use the circle of radius r
2
to
evaluate the coecients in (2) and estimate as follows:
|a
n
| =

1
2i
_
Cr
2
(z0)
f()
( z
0
)
n+1
d

1
2
2r
2
M
r
n+1
2
=
M
r
n
2
.
Section 4.5 Laurent Series 95
For r
1
< |z z
0
| < r
2
, let =

zz0
r2

, then 0 < 1. Thus

a
n
(z z
0
)
n

M
r
n
2
|z z
0
|
n
M
n
,
and so

n=0
a
n
(z z
0
)
n
converges absolutely, by comparison with the series M

n
. Since z is
arbitrary in A
R1,R2
(z
0
), we conclude from Lemma 1, Sec. 4.3, that the series

n=0
a
n
(z z
0
)
n
converges absolutely for all |z z
0
| < R
2
and uniformly on any subdisk |z z
2
| r
2
< R
2
.
Note that this proof shows that the series in (1) denes an analytic function in the disk B
R2
(z
0
).
Were it not for the other series with negative powers in (1), the function f(z) would be analytic in
B
R2
(z
0
).
(b) The proof in this part is very similar to the proof in part (a). In fact, we will use the notation
from (a). Fix z, r
1
and r
2
, such that z A
R1,R2
(z
0
) is arbitrary but xed, R
1
< r
1
< |z z
0
| < r
2
<
R
2
. Let M be the maximum value of |f()| for in the closed annular region r
1
| z
0
| r
2
.
Use the circle of radius r
1
to evaluate the coecients in (2) and estimate as follows: for n < 0,
|a
n
| =

1
2i
_
Cr
1
(z0)
f()
( z
0
)
n+1
d

1
2
2r
1
M
r
n+1
1
=
M
r
n
1
= Mr
n
1
.
Equivalently, for n 1,
|a
n
| Mr
n
1
.
For r
1
< |z z
0
| < r
2
, let =

r1
zz0

, then 0 < 1. Thus, for n 1,

a
n
(z z
0
)
n

Mr
n
1
|z z
0
|
n
= M
n
,
and so

n=0
an
(zz0)
n
converges absolutely, by comparison with the series M

n
. Since z is arbi-
trary in A
R1,R2
(z
0
), we conclude that the series

n=0
an
(zz0)
n
converges absolutely for all A
R1,R2
(z
0
).
To show that the series converges uniformly on closed subsets of A
R1,R2
(z
0
), it suces to show that
it converges uniformly on any closed subannulus A
r1,r2
(z
0
), where R
1
< r
1
< r
2
< R
2
. For all z in
A
r1,r2
(z
0
), we have |
an
(zz0)
n
|
|an
|r
n
1
. From the previous part, we know that the series

n=1
an
r
n
1
converges absolutely, since the point z
0
r
1
is in the annulus A
R1,R2
(z
0
) (to see this, compute
|z
0
(z
0
r
1
)| = r
1
, and R
1
< r
1
< R
2
). Thus, if we take M
n
=
|an|
r
n
1
, we can apply the Weierstrass
M-test and infer that the series

n=0
an
(zz0)
n
converges uniformly on A
r1,r2
(z
0
), as desired.
96 Chapter 4 Power Series and Laurent Series
Solutions to Exercises 4.6
1. The function f(z) = (1 z
2
) sin z has zeros at 1 and k, where k is an integer. All these zeros
are simple zeros. For z = 1, write f(z) = (z 1)(z + 1) sin z = (z 1)(z). Since (1) = 0, we
conclude that z = 1 is a simple zero. A similar argument applies to z = 1. For the remaining
zeros, which are the zeros of sinz, see Example 1.
5. We will use the fact that the zeros of sin z are all isolated simple zeros. At z = 0, we have
sin z = zz, where 0 = 0. So,
f(z) =
sin
7
z
z
4
= z
3

7
(z)
7
(0) = 0.
This shows that if we dene f(0) = 0, then f(z) has a zero of order 3 at 0. All other zeros of f
occur at the zeros of sinz, k, k = 0, and they are of order 7.
9. We have
cos z = 1
z
2
2!
+
z
4
4!

z
6
6!
+
So
f(z) = 1
z
2
2
cos z =
z
4
4!
+
z
6
6!
+ = z
4
_

1
4!
+
z
2
6!
+
_
.
Note that (z) =
1
4!
+
z
2
6!
+ is a power series that converges for all z. Thus (z) is entire.
Moreover, (0) =
1
4
= 0. Thus f(z) = z
2
(z), where (z) is analytic and (0) = 0, implying that
f(z) has zero of order 2 at 0.
13. Clearly, the function
f(z) =
1 z
2
sinz
+
z 1
z + 1
has isolated singularities at 1 and k, where k is an integer. These singularities are all simple
poles. To prove the last assertion, it is easier to work with each part of the function separately.
First, show that
1z
2
sin z
has a simple pole at the zeros of sinz, which follows immediately from the
fact that the zeros of sinz are simple zeros. Second, show that
z1
z+1
has a simple pole at z = 1,
which follows immediately from the fact that 1 is a simple zero of z +1. Now to put the two terms
together, you can use the following fact:
If f(z) has a pole of order m at z
0
and g(z) is analytic at z
0
, then f(z) + g(z) has a pole of
order m at z
0
.
This result is easy to prove using, for example, Theorem 8.
17. Write
z tan
1
z
= z
sin
1
z
cos
1
z
.
The problem points of this function are at 0 and at the zeros of the equation cos
1
z
= 0. Solving, we
nd
1
z
=

2
+ k z = z
k
=
2
(2k + 1)
, k an integer.
Since, as k , z
k
0, the function f(z) is not analytic in any punctured disk of the form 0 < |z|.
Thus 0 is not an isolated singularity. At all the other points z
k
, the singularity is isolated and the
order of the singularity is equal to the order of the zero of cos z at z
k
. Since the zeros of cos z are
all simple (this is very similar to Example 1), we conclude that f(z) has simple poles at z
k
.
21. We have an isolated singularity at 0, which is clearly an essential singularity. To see this, we
consider the Laurent series expansion of f(z):
1
z
sin
1
z
=
1
z

_
1
z

1
3!z
3
+
1
5!z
5

_
=
1
3!z
3

1
5!z
5
+ .
Section 4.6 Zeros and Singularities 97
25. Determining the type of singularity of f(z) =
1
z+1
at is equivalent to determining the type
of singularity of
f
_
1
z
_
=
1
1
z
+ 1
=
z
1 + z
at z = 0. Since f
_
1
z
_
has a removable singularity at 0, we conclude that f has a removable singularity
at z = . Note that this is consistent with our characterization of singularities according to the
behavior of the function at the point. Since f(z) 0 as z , we conclude that f has a removable
singularity and may b redened to have a zero at .
29. Arguing as in Exercise 25, it follows that sin
1
z
has a removable singularity at and may be
redened to have a zero at .
33. (a) Suppose that f is entire and bounded. Consider g(z) = f(
1
z
). Then g is analytic at all
z = 0. So z = 0 is an isolated singularity of g(z). For all z = 0, we have |g(z)| = |f(
1
z
)| M < ,
where M is a bound for |f(z)|, which is supposed to exist. Consequently, g(z) is bounded around 0
and so 0 is a removable singularity of g(z).
(b) Since f is entire, it has a Maclaurin series that converges for all z. Thus, for all z, f(z) =
sum

n=0
a
n
z
n
. In articular, we can evaluate this series at
1
z
and get, for z = 0 ,
g(z) = f(z) =

n=0
a
n
z
n
.
By the uniqueness of Laurent series expansion, it follows that this series is the Laurent series of g.
But g has a removable singularity at 0. So all the terms with negative powers of z must be zero,
implying that g(z) = a
0
and hence f(z) = a
0
is a constant.
37. (a) If f has a pole of order m 1 at z
0
, then
f(z) =
1
(z z
0
)
m
(z),
where is analytic at z
0
and (z
0
) = 0. (See (6), Sec. 4.6.) So if n is a positive integer, then
[f(z)]
n
=
1
(z z
0
)
mn

n
(z) =
1
(z z
0
)
mn
(z),
where is analytic at z
0
and (z
0
) = 0. Thus [f(z)]
n
has a pole at z
0
of order mn if n > 0. If
n < 0, then
[f(z)]
n
= (z z
0
)
mn
1

n
(z)
= (z z
0
)
mn
(z),
where is analytic at z
0
and (z
0
) = 0. Thus [f(z)]
n
has a zero at z
0
of order mn if n < 0.
(b) If f has an essential singularity at z
0
then |f(z)| is neither bounded nor tends to innity at z
0
.
Clearly, the same holds for |[f(z)]
n
| = |f(z)|
n
: It is neither bounded nor tends to near z
0
. Thus
[f(z)]
n
has an essential singularity at z
0
.
41. Suppose that f and g are analytic in a region and fg is identically zero in . Suppose that
g(z) is not identically 0 in and let z
0
be such that g(z
0
) = 0. Because g is continuous and g(z
0
) =,
we can nd a neighborhood of z
0
, B
r
(z
0
), such that g(z) = 0 for all z B
r
(z
0
). Since f(z)g(z) = 0
for all z , it follows that f(z) = 0 for all z B
r
(z
0
). Thus, the zeros of f are not isolated, and
so, by Theorem 3, f is identically 0.
45. Suppose that p(z) is a polynomial such that |p(z)| = 1 for all |z| = 1. Since |p(z)| = 1 for
all |z| = 1, the polynomial has no zeros on the unit circle. Let m be the degree of the polynomial
and let a
1
, a
2
,. . . ,a
n
denote the zeros of p(z) inside the unit disk, counted according to multiplicity.
98 Chapter 4 Power Series and Laurent Series
Then n m. For each j = 1, . . . , , n, consider the linear fractional transformations
aj
(z) =
zaj
1ajz
.
We know form Example 3, Sec. 3.7, that |
j
(z)| = 1 for all |z| = 1. Let (z) =
n
j=1

j
(z) denote
the product of the
j
, where each
j
is repeated according to the multiplicity of the zero. We have
|(z)| = 1 for all |z| = 1. Moreover, has exactly n zeros at the a
j
. (If some of the a
j
s are repeated,
then the order of the zero at a
j
is equal to the number of times we repeat a
j
.)
Consider
g(z) =
p(z)
(z)
= p(z)
n
j=1
1 a
j
z
a
j
z
.
Factoring out the zeros in p(z) and canceling the factors (z a
j
) in the numerator with the same in
the denominator in g(z), we see that g(z) has removable singularities at a
j
, hence it can be redened
to be analytic at these points, and so we will consider g(z) to be analytic in B
1
(0). Moreover, g(z) has
no zeros in B
1
(0), and for |z| = 1, we have |g(z)| =
|p(z)|
|(z)|
= 1. By the maximum modulus principle
(Corollary 3(ii), Sec. 3.7), it follows that g(z) = A is constant in B
1
(0), and since |g(z)| = 1 for
|z| = 1, it follows that |A| = 1. As a consequence, we have
p(z) = A
n
j=1
a
j
z
1 a
j
z
.
Thus
p(z)
n
j=1
(1 a
j
z) = A
n
j=1
(a
j
z).
On the right side we have a polynomial of degree n. On the left side, we have a polynomial p(z)
of degree m n, multiplied by a polynomial of degree equal to the number of nonzero a
j
s. It
is clear that such an equality is impossible unless m = n and all a
j
= 0; otherwise the degree of
the polynomial on the left becomes strictly greater than the degree of the polynomial on the right.
Consequently, p(z) = A()
n
z
n
= Bz
n
, where |B| = 1.
Section 4.7 Harmonic Functions and Fourier Series 99
Solutions to Exercises 4.7
1. (a) The graph of the boundary function
f() =
_
+ if 0,
if 0 < < ,
is a triangular wave that repeats every 2-units. See Figure 8, Sec. 7.2.
(b) The solution of the Dirichlet problem with boundary function f() is given by
u(r, ) = a
0
+

n=1
r
n
(a
n
cos n + b
n
sin n) (0 r < 1),
where a
n
and b
n
are the Fourier coecients of f, given by formulas (6)(8). However, since the
formula for f is given on the interval [, ], it is more convenient to use equivalent formulas that
are obtained by changing the interval of integration in (6)(8) from [0, 2] to [, ]. More precisely,
we can use
a
0
=
1
2
_

f() d,
a
n
=
1

f() cos n d (n 1),

b
n
=
1

f() sin n d (n 1).

The reason is that the preceding integrands are 2-periodic, and so the values of the integrals are
the same over any interval of length 2. (See Theorem 1, Sec. 7.1) (c) Since f is even, we have
a
0
=
1
2
_

f() d =
2
2
_

0
f() d =
1

_

0
( ) d =

2
.
Similarly, since f() cos n is even,
a
n
=
1

f() cos n d =
2

_

0
( ) cos n d
=
2

cos n
n
2
( )
sin n
n
_

0
=
2

_
1
n
2

cos n
n
2
_
.
Now using the fact that f() sin n is odd, we obtain
b
n
=
1

f() sin n d = 0 (n 1).

(d) Plugging the coecients into the series, we obtain the solution for 0 < r < 1,
u(r, ) =

2
+

n odd
4
n
2
r
n
cos n =

2
+
4

k=0
r
2k+1
(2k + 1)
2
cos(2k + 1).
(e) We know that the solution of the Dirichlet problem inside the unit disk approaches the boundary
function as r 1. Hence lim
r1
u(r, ) = f(). Now assuming that lim
r1
u(r, ) = u(1, ), we
obtain u(1, ) = f(), or
f() =

2
+

n odd
4
n
2
cos n =

2
+
4

k=0
1
(2k + 1)
2
cos(2k + 1).
100 Chapter 4 Power Series and Laurent Series
5. (a) Starting with the solution that we derived in Example 1 and using the series in Exercise 4,
we obtain
u(r, ) = 50 +
200

k=0
1
2k + 1
_
r
2
_
k
sin(2k + 1) = 50 +
200

k=0
1
2k + 1

k
sin(2k + 1),
where =
r
2
, 0 < 1. So by Exercise 5,
u(r, ) = 50 +
200

1
2
_
tan
1
_
sin
1 cos
_
+ tan
1
_
sin
1 + cos
_
_
= = 50 +
100

_
tan
1
_
r sin
2 r cos
_
+ tan
1
_
r sin
2 + r cos
_
_
= 50 +
100

_
tan
1
_
y
2 x
_
+ tan
1
_
y
2 + x
_
_
,
where we have used the relations x = r cos and y = r sin .
(b) Let 0 < T < 100 be a given temperature. Suppose further that T = 50. Let (x, y) be a point
inside the disk x
2
+ y
2
= 4 such that u(x, y) = T. Then
T = 50 +
100

_
tan
1
_
y
2x
_
+ tan
1
_
y
2+x
__
,

100
(T 50) = tan
1
_
y
2x
_
+ tan
1
_
y
2+x
_
.
Apply the tangent to both sides and use the identity tan(a +b) =
tana+tan b
1tana tanb
. To simplify notation,
write A = tan
_

100
(T 50)
_
. Then
tan
_

100
(T 50)
_
=
y
2x
+
y
2+x
1
y
2x
y
2+x
,
A =
4y
x
2
+ y
2
+ 4
,
x
2
+ y
2
+ 4 = 4Ay, x
2
+ y
2
+
4
A
y = 4,
x
2
+
_
y +
2
A
_
2
=
4
A
2
4.
This is the equation of a circle centered at (0, y
1
), where
y
1
=
2
A
=
2
tan
_

100
(T 50)
_ = 2 cot
_

100
(T 50)
_
,
and radius
R =
_
4
A
2
4 = 2
_
1
A
2
12
_
cot
2
_

100
(T 50)
_
1
= 2
_
csc
2
_

100
(T 50)
_
= 2 csc
_

100
(T 50)
_
.
Note that the cosecant is positive for
_

100
(T 50)
_
, with 0 < T < 100. So there is no need to use
absolute values when evaluating the square root.