Professional Documents
Culture Documents
Analysis. The problems in the first 8 chapters are suitable for an introductory course at the undergraduate level and cover the following topics: power series, Cauchy's theorem, Laurent series, singularities and merornorphic functions, the calculus of residues, conformal mappings, and harmonic functions. The material in Chapters 9 to 16 is more advanced. The reader will find problems on Schwartz reflection, analytic continuation, Jensen's formula, the Phragmen-Lindeloef theorem, entire functions, Weierstrass products, meromorphic functions, the Gamma function, and the Zeta function. This volume also serves as an independent source of problems with detailed answers beneficial for anyone interested in learning complex analysis. Original drawings, provided by the author, assist students in their mastery of problem-solving techniques and increase their overall comprehension of the subject matter.
o
c::J
i:I r
Rami Shakarchi
,."
~
Z
>-
o
Vl
o r
C -;
6 z Vl
o
..."
;:0
()
o
~
""J
,."
COMPLEX ANALYSIS
><
Z
Vl
>~
Vi
>-
ISBN 0-387-98831-9
www.springer-ny.com
Springer
COMPLEX ANALYSIS
Springer
New York Berlin Heidelberg Barcelona Hong Kong London Milan Paris Singapore Tokyo
Rami Shakarchi
COMPLEX
ANALYSIS
With 46 Illustrations
Springer
Rami Shakarchi Department of Mathematics Princeton University Princeton, NJ 08544-1 ()()() USA
Shakarchi, Rami. Problems and solutions for Complex analysis I Rami Shakarchi. p. cm. "Contains all the exercises and solutions of Serge Lang's Complex analysis""-Pref. Includes bibliographical references. ISBN 0-387-98831-9 (softcover : alk. paper) I. Mathematical analysis-Problems, exercises, etc. 2. Functions of complex variables-Problems, exercises, etc. I. Lang, Serge, 1927. Complex analysis. II. Title. QA30J.S48 1999 5IY.9--dc21 99-13255 Printed on acid-free paper.
ISBN 0-387-98831-9
This book is dedicated to my parents, Mohamed and Mireille Shakarchi, in appreciation for their love and support.
Preface
This book contains all the exercises and solutions of Serge Lang's Complex Analysis. Chapters I through vrn of Lang's book contain the material of an introductory course at the undergraduate level and the reader will find exercises in all of the following topics: power series, Cauchy's theorem, Laurent series, singularities and meromorphic functions, the calculus of residues, conformal mappings and harmonic functions. Chapters IX through XVI, which are suitable for a more advanced course at the graduate level, offer exercises in the following subjects: Schwarz reflection, analytic continuation, Jensen's formula, the Phragmen-Lindelof theorem, entire functions, Weierstrass products and meromorphic functions, the Gamma function and the Zeta function. This solutions manual offers a large number of worked out exercises of varying difficulty. I thank Serge Lang for teaching me complex analysis with so much enthusiasm and passion, and for giving me the opportunity to work on this answer book. Without his patience and help, this project would be far from complete. I thank my brother Karim for always being an infinite source of inspiration and wisdom. Finally, I want to thank Mark McKee for his help on some problems and Jennifer Baltzell for the many years of support, friendship and complicity. Rami Shakarchi Princeton, New Jersey 1999
Contents
Preface
I
vii
1
3
8
9 12
II
Power Series
11.1 11.2 11.3 11.4 11.5 11.6 Formal Power Series Convergent Power Series . . . . . . . . . Relations Between Formal and Convergent Analytic Functions Differentiation of Power Series. . . . . . The Inverse and Open Mapping Theorems ..... Series . . . .
14
14 19 24
28
29 31 36 36 37 42 43
III
45
Contents
IV V
Winding Numbers and Cauchy's Theorem IV.2 The Global Cauchy Theorem . . . . . Applications of Cauchy's Integral Formula V.l Uniform Limits of Analytic Functions V.2 Laurent Series ... V.3 Isolated Singularities . . . . . . . . . Calculus of Residues VI. 1 The Residue Formula . . . . . . VI.2 Evaluation of Definite Integrals Conformal Mappings VII.2 Analytic Automorphisms of the Disc VII.3 The Upper Half Plane . VII.4 Other Examples. . . . . . . . . . VII.5 Fractional Linear Transformations
48 48 51 51 60 66 76 76 93
119
VI
VII
119 122 126 137 146 146 153 159 165 167 175 175 179 179 181 185 185 187
VIII Harmonic Functions VIII.1 Definition . . . . . . . . . . . . . . . . . VIII.2 Examples . . . . . . . . . . . . . . . . . VIII.3 Basic Properties of Harmonic Functions . VIII.4 The Poisson Formula . . . . . . . . . VIII.5 Construction of Harmonic Functions . IX X Schwarz Reflection IX.2 Reflection Across Analytic Arcs The Riemann Mapping Theorem X.l Statement of the Theorem X.2 Compact Sets in Function Spaces Analytic Continuation along Curves XI.l Continuation Along a Curve XI.2 The Dilogarithm ........ Applications of the Maximum Modulus Principle and Jensen's Formula XII. 1 Jensen's Formula ................ XII.2 The Picard-Borel Theorem. . . . . . . . . . . XII.6 The Phragmen-Lindelof and Hadamard Theorems
XI
XII
XIII Entire and Meromorphic Functions XIII. 1 Infinite Products .. XIII.2 Weierstrass Products .....
206 211
Contents
xi 213 214
XIII.3 Functions of Finite Order. . . . . . . . . . . . . . XIII.4 Meromorphic Functions, Mittag-Leffler Theorem . XV The Gamma and Zeta Functions XV. 1 The Differentiation Lemma XV.2 The Gamma Function . XV.3 The Lerch Formula . XV.4 Zeta Functions . . . . .
219
219 223 235 238 241 241 245
XVI The Prime Number Theorem XVI.I Basic Analytic Properties of the Zeta Function XVI.2 The Main Lemma and its Application . . . . .
I
Complex Numbers and Functions
I.l
Definition
in the form x i)
i)
Exercise 1.1.1. Express the following complex numbers x, yare real numbers. (a) (-1 + 3i)-1 (b) (1 + i)(1 (d) (i - 1)(2 (c) (1 + i)i(2 - i) (f) (2i + l)n i (e)(7 + ni)(n + i) (h) (i + 1)(i (g) C.j2i)(n + 3i) Solution. -2n
(a)
+ iy,
where
2)(i
+ 3)
(f)
+ in.
~ci-
(g) -3-J2
+ n-J2i.
Exercise 1.1.2. Express the following complex numbers in the form x x, yare real numbers.
(a) (l
~It
( ) -I g""5
+ i)-I
(b)3li
I
(c) ~~:
(d) 2~i
WI~
foi.
-I I. T -"21.
W~
(c) ~
~_~
1 - i, (f) !
+ 51. (h) 3·
+ !i.
Exercise 1.1.3. Let a be a complex number a tsr What is 'ii? Solution. Let a = a
+ ib.
a
Then a
+ ib
a2
b2
a2
+ 2abi
b2
a - ib
2 so
I. ComplexNumbersand Functions
1a
Moreover,
~ 12 - (a2
b2)2
(a2
+ b2)2
- ib
+ 4a2b2
_ -
(a2
(a2
+ b2i _ I + b2)2 - .
ii = a
Exercise 1.1.4. Let
= a + ib = a.
afJ = ap and
that
a, fJ be two
Solution.
Suppose a
= a + ib
Then
ap
and
= (a -
ib)(c - id)
= ac -
bd - i(ad
+ bc) = afJ,
a+ p = a
- ib + c - id
= (a +
c) - i(b + d)
=a+
fJ.
Exercise 1.1.5. Justify the assertion made in the proof of Theorem 1.2, that the real part of a complex number is ~ its absolute value. Solution. Suppose z = x
+ iy.
Then
x2
x2
+ l,
I Re(z)1 s Izl.
Exercise 1.1.6. If a of a and we write b
= a + ib with a, b real, then b is called the imaginary = Im(a). Show that a - a = 2i Im(a). Show that
Im(a)
part
s I Imto)] s lal.
-a
Solution.
We have a-
a = a + ib
+ ib
= 2ib.
The first inequality is obvious, while the second inequality follows from
Im(a)2 ~ Re(a)2
+ Im(a)2. + i)HlO.
SO
Exercise 1.1.7. Find the real and imaginary parts of (I Solution. so Since (1 +i)2
= -I
Exercise l.l.S.
(a) (b) (c)
Solution. All three inequalities are obtained by writing z applying the triangle inequality. Exercise 1.1.9. Let a = a condition
z - w + w and
+ ib and
z =x
+ iy,
lz - al = c
into an equation involving only x, y, a, b, and c, and describe in a simple way what geometric figure is represented by this equation. Solution. By definition we see that
Iz - al
= c is equivalent
- b)2 = C,
to
J(x - a)2
+ (y
so the above equation describes the circle of radius c centered at a. Exercise 1.1.10. Describe following conditions. (a) Iz - i + 31 = 5 (c) lz - i + 31 ~ 5 (e)Imz>O (g)Rez>O geometrically
(b)
the
°°
Solution. (a) Circle of radius 5 centered at i - 3. (b) Complement of the closed disc of radius 5 centered at i - 3. (c) Closed disc of radius 5 centered at i - 3. (d) Closed disc of radius I centered at -2i. (e) Open upper half plane. (f) Closed upper half plane. (g) Open right half plane. (h) Closed right half plane.
Exercise 1.2.2. Put the following complex numbers in the ordinary form x (a) e3i:rr (b) e2i:rr/3 (c) 3ei:rr/4 (d) ]fe-i:rr/3 (e) e2i:rr/6 (I) e-i:rr/2 (g) e-i:rr (h) e-5i:rr/4 Solution. (a) -I (b)
+ iy.
-4
+i~
(c)
1 1 ~+i (d)
i~
(e)
4 +i~
(f) -i (g)
-I (h) - ~ +i ~
Solution. Suppose a = re'", Then the two solutions to Z2 = a are .jiei(cp/2) and .jiei(7r+cp/2).See Exercise 6. Exercise 1.2.4. Let a
+ bi
+ iy)2
=a
+ ib,
expressing x, y in terms of a and b. Solution. Since (x + iy)Z = x2 - y2 + 2ixy we have x2 - y2 = a and 2xy = b. Taking absolute values we also get x2 + y2 = J a2 + b2. These three equations imply
x2=
and therefore x = fa solves our problem.
a + Ja2
2
+b2 _
and
Ja2 + b2 - a .
2
+ J a 2+ b2
2
and
y = (sign b)y
Exercise 1.2.5. Plot all the complex numbers Z such that graph paper; for n = 2, 3, 4, and 5.
= 1 on a sheet of
Solution. The equation Z2 = 1 has two solutions, 1 and -1 which we plot as stars. The equations Z3 = I has three solutions, I, e27ri/3and e47ri/3which we plot as dots. The equation Z4 = 1 has four solutions, 1, i, -1 and -i which we plot as stars. The equation Z5 = 1 has five solutions, 1, e27ri/5,e47ri/5,e67ri/5and e87ri/5which we plot as dots. (See figure at top of next page.) Exercise 1.2.6. Let a be a complex number #- o. Let n be a positive integer. Show that there are n distinct complex numbers z such that z" = a. Write these complex numbers in polar form. Solution. We use the expressions z = re'" and a = se'", Then the equation z" = a is equivalent to
hence
r" _ei(n(J-cp) = 1. s
Taking absolute values we get r" = s. Moreover nO - cp must be an integral multiple of 21l", so the set of solutions of the equation z" = a is
S = {sl/"ei(;),
sl/"eiU+~),
... , sl/nei(;+(n-l)~)}.
Exercise 1.2.7. Find the real and imaginary parts OJil/4, taking theJourth root such that its angle lies between 0 and n /2. Solution. Since i of ei1f/8• Hence
=e
i1f/2
Re(i and
1/4
= cos "8 =
Jl
+cos~ 2
= V 2 + "4
Exercise 1.2.S. (a) Describe all complex numbers z such that eZ = 1. (b) Let w be a complex number. Let a be a complex number such that ea = w. Describe all complex numbers z such that eZ = w.
e' eiy
Solution.
(a) If z = x + iy where x and y are real, then eZ = 1 is equivalent to = 1. Taking absolute values we see that eX = 1 hence x = 0 and therefore y must be an integral multiple of 21f. Conversely, any complex number of the form 21fki with k E Z is a solution of eZ 1. (b) The number w is nonzero because ea = wand therefore
hence by (a) we must have Re(z) = Re(a) and Im(z) - Im(a) must be an integral multiple of 21f. We can also express the solutions of eZ = w independently of a.
+ iy
= £}
(mod 21f).
and
= ei Im(w)
therefore Re(z) Re(w) and Im(z) - Im(w) is equal to an integral multiple of 21f, as was to be shown. Exercise 1.2.10. (a) If£} is real, show that ei9 +e-i(J cos e = ei9 _ e=" 2i
and
sin £}=
(b) For arbitrary complex z. suppose we define cos z and sin z by replacing £}with z in the above formula. Show that the only values of Zfor which cos Z = 0 and sin z = 0 are the usual real values from trigonometry. Solution. (a) The two formulas follow from ei(J + «" = cos e and ei(J - «:" = cos s
- i sin£} = 2 cos £}
+ i sin£} = 2i sin£}o
this equation
(b) If cos z 0 then by definition we get eiz by e" and writing z x + iy we obtain
+ e-iz = O. Multiplying
-1 = e2i(x+iy) = e-2Ye2ix. Taking absolute values we get y 0 and therefore x n /2 (mod rr ) as was to be shown. The equation sin z = 0 is equivalent to eiz - e-iz 0 hence e2iz = 1. Letting z x + i y and arguing as we did before, we find that x = 0 and y = 0 (mod zr).
Exercise 1.2.11. Prove that for any complex number Z =I 1 we have l+z+"'+z"=--.' Solution. This formula follows from (z - 1)(1 + z Z"+l - 1 z-1
+ ... + z")
=z
1 - z - ... - z"
= Z".+l- 1. Exercise 1.2.12. Using the preceding exercise, and taking real pans, prove: 1 + cos sin[(n + 1/2)8] 2sin(£}/2)
2+
forO <
< 27f.
Solution. The real part of the sum L~=o(eiO)k is 1+cos() +cos 2() + ... +cosns. The formula of the preceding exercise gives ~-,<eiO)k =. ~ elf} - 1
k=O n ei(n+I)II _
= _=_---:-::-:::-eiOj2 _ =
ei(n+lj2)11 e-iOj2 _
e-iOj2
ei(n+lj2)11 e-iOj2 _
= i ----:-~-:---2 sin«()/2)
The imaginary part of e-iOj2 - ei(n+lj2)Ois sin( -() /2) - sin[(n + 1/2)()], whence 1 + cos () + cos 2() +···+cosn and the desired formula drops out. Exercise 1.2.13. Let z, w be two complex numbers such that zw ()
2 sin ()/2
i=
1. Prove that
I l-zw I < 1
Z- w
lIZ ~
z: I =
iflzl = 1 or Iwl = 1.
(There are many ways of doing this. One way is as follows. First check that you may assume that Z is real, say z = r. For the first inequality you are reduced to proving (r - w)(r - w) < (l - rw)(l - rw). You can then use elementary calculus, differentiating with respect to r and seeing what happens for r 0 and r < 1, to conclude the proof)
Solution.
I
=
II
'
so we may assume that z is a real number 0 ~ r ~ 1. We want to show that (r - w)(r - W) ~ (1 - rw)(l - rw) with = if and only if r 1 or I w I = 1. We expand both sides, make the necessary cancellations and move all the terms to the right to see that the above inequality is equivalent to
hence
Conclude.
- =-e
r
1 Z
1 -ie
Let V = C - D, in other words, V is the complement of the closed unit disc. From the above formula we find we see that the image of D under the inversion is V and conversely, the image of V under the inversion is D. If r = 1, then liz = «" so the image of the unit circle is the unit circle. Exercise 1.3.2. Let f(z) = liz. Describe f in the same manner as in Exercise 1. This map is called reflection through the unit circle. Solution. The reflection is defined on the complex plane minus the origin. If z = re'", then f(z) 1 '(J = -e' r
so if V = C - D, then we see that f(D - (O}) = V and f(V) = D - [O}, If r = I, then f(z) = ei(J so the image of the unit circle is the unit circle. Exercise 1.3.3. Let f(z) = e27riz• Describe the image under f of the set shaded in the figure on the facing page, consisting of those points x + i y with ::s x ::s and y:::: B.
Solution.
If z = x
+ iy,
But s x ::s so -rr s 2rrx s n and y :::: B so -2rry ::s -2rr B. From the above expression we see that the absolute value of f(z) is e-27rY and the argument of f(z) is 2rr x. So the image of the shaded region under f is the closed disc of radius e-2TrB minus the origin. Exercise 1.3.4. Let f(z) = e', Describe the image under f of the following sets: (a) The set of Z = x + i y such that x ::s I and 0 ::s y ::s tt . (b) The set of z = x + i Y such that 0 ::s y ::s n (no condition on x). Solution. If z = x
-4
+ iy,
So the absolute value of f(z) is e' and the argument of f(z) is y. (a) The image of the given set is the closed upper half disc of radius e minus the origin. (b) The image of the given region is the closed upper half plane minus the origin.
1. Then
Ir e 1 = r",
n ni8
and therefore lim an = o. Exercise 1.4.2. If la I > I, does limn--+oo exist? Why? an
Solution. We write a
so Ian I ~ as n ~
= re'"
00
00
so the limit
10
does not exist. Exercise 1.4.3. Show that for any complex number z 1 +z+···+zn lflzl < 1, show that
i= 1, we have
zn+l - 1 = --z- 1
n-+oo
I· ( 1 + z im
Solution.
We have
(z - 1)(1 + z
+ ... + z") = Z + ... + zn + zn+l =zn+l_1, drops out. If lz] < 1, then
n-+oo
so if z
lim zn+l = 0,
so
n-+OO
. I im (1 + z
+ ... + z n ) = ---1 z 1
n-+OO
. hm
1 2• 1+ n z
Show that f is the characteristic function of the set {OJ, that is, f(O) = 1, and f(z) = 0 if z i= O. Solution. have We clearly have f(O) = l. If z In2z
+ 11 ~
1 2 n 1Z1- 1 ~ 2n21zl,
I n2z 1+ 11 :::::n2~zl·
Therefore f(z)
= 0 as was to be shown.
i= 1 show
f(z) that the following limit exists:
n-+oo
1)
continuous?
11
~-(-I)I=I~I~o+ +
zn
1
z"
as n ~
00,
as n ~ 00, so fez) 1. From these results we see that we cannot define fez) when lz] = 1 so as to make f continuous. Exercise 1.4.6. Let fez) = lim --.
n-HX)
I zn + 1 -
zn-1
1 = lz"
+ 11 ~
zn
+ z"
(a) What is the domain of definition of f, that is, for which complex numbers z does the limit exist? (b) Give explicitly the values of f(z)for the various z in the domain of f.
0 as n ~
00
so fez)
o. If lzl
>
I, then
as n -+ 00. We now investigate what happens on the unit circle. Let z = eiIJ with 0 :::::J < ( 2JT. Then I + z" I + eniIJ,so if (J 0 we immediately get f (z) 1/2. If (J i= 0 then eniIJ
fez) = 1 + eniIJ = I
+ e-niIJ'
and since e-niIJ goes around the circle we cannot define f at the points z = eiIJ with (J i= O. So if Q denotes the unit circle minus the point I, we see that the domain of definition of f is the set C - Q. Exercise 1.4.7. Show that the series n-l
(I - zn~(l - zn+l)
?;
00
converges to 1/(1 - zi for Izl < I and to l/z(1 - Z)2 for Izl > 1. Prove that the convergence is uniformfor lel :::::c < I in the first case, and lz! 2: b > 1 in the second. [Hint: Multiply and divide each term by I - z, and do a partial fraction decomposition, getting a telescoping effect.]
Solution. (i) Let U; = z" /(1 - zn)(1 - zn+l) and let D(z) = (l - zn)(1 - zn+l).
Then Un
zn(1 - z) D(z)(l - z)
12
1 1- z
[(Zn -
(1 -
1 ~ z [ - 1 - ~n+1 + 1 ~
L Uk--- z [ ---z - 1 -1 1 -1
n k=1
1] 1 - zn+1
and therefore
Sn(Z) = "" 8(I-Zk)(1-Zk+I)
n
zk-I
1
z(1-z)
[1-
l-z
- --. n l l-z + ·
1]
If lz! < I, then 1/(1 - Zn+l) ~ 1 as n ~ 00 and therefore Sn(Z) ~ 1/(1 - Z)2 as n ~ 00. If Izl > I, then 1/(1 - Zn+I) ~ 0 as n ~ 00 so Sn(Z) ~ l/z(1 - Z)2. (ii) Suppose lz] :::: c < 1. A little algebra and part (i) imply that
ISn(z) - (1 ~
I·
But
s (z) _
n
for all z in the region Izl :::: c < 1. Now cn in the region Izl :::: c < 1. If Izl ::: b > I, then
ISn(z) z(1 ~
11 b(b ~ 1) bn+ -
and bn+1
00,
Izl ::: b
> 1.
f(z)
is holomorphic.
= f(x
+ iy)
= u(x,
y)
+ iv(x, y)
au au ax = ay
au au ay = - ax .
13
We use the notation of the section. Let i; = ~:- i ~~. Then using the CauchyRiemann equations we find that f(z + w) - f(z) - i;w is equal to
au [ u(x
+ h , y + k)
+i
- u(x, y) -
ax h -
au] ay k
[ v(x
+ h, y + k) -
v(x, y) - -h ax
av
- -k
ay
av]
where lim(h.k)-+(o.O) aj(h, k) = 0 for i = 1,2, because both u and v are differentiable. For w =P 0 near zero we let a(w) = -(a.(w)
w
Iwl
+ ia2(W».
Then f(z
+ w)
f is holomorphic at z.
II
Power Series
~?~~
e
cos z
Z/
(g) sinz
Z
(h)
sin z
Solution. (a) eZ sin z = z + Z2 + (- ~ + ~) Z3 + higher terms. (b) (sin z)( cos z) = z + (- ~ - ~) Z3 + higher terms. e'-I Z Z2 z3 ( c ) -z- _ 1 + 2! + If + 4! + hi h er terms. ig (d)
()e (f)
(g)
= 1+ 1 + 2Z 2 1
(! + D z +
+
(
~Z2 I)
-"24
+4
z = ~+ 1 + (~ + ~) z + ~Z2 + (3~)2
t+
3!12!
+ t) Z3 + higher
Exercise 11.1.2. Let f(z) = Lanzn. Define f(-z) = Lan(-z)n = Lan(-l)nzn. We define f(z) to be even if an = 0 for n odd. We define f(z) to be odd if an = 0 for n even. Verify that f is even if and only if f( -z) = f(z) and f is odd if and only if f( -z) = - f(z).
15
Solution. we get
Suppose I(-z)
1 is even.
Lan(-l)nzn
neven
L an(_l)nzn
= LanZn
L anZn = I(z).
neven
hence 2 2:nodd anZn = 0 which implies that an = 0 for all n odd. When 1is even, a similar argument proves the desired statement. Exercise 11.1.3. Define the Bernoulli numbers Bn by the power series
eZ
Prove the recursion formula n!O!
z
-
~ B; n 1 = ~ -;:;Tz .
n=O
Bo
+ (n
BI - 1)!1! + ...
1!(n -I)!
Bn-I
11 0
Then Bo = 1. Compute BJ, B2, B3, B4. Show that B; = 0 ifn is odd Solution. We know that eZ =
i= 1.
2::0
z
z" / n!, so
eZ
z
-1
Since we are multiplying power series we can use the formula given in the text so that
n
(;
k!(n - k
e,
+ I)!
11
ifn
= 0,
ifn > O.
Let m = n + 1 and conclude. Using the above formula we get BI = -1/2, B2 = 1/6, B3 = 0 and B4 = -1/30. To show that B; = 0 if n is odd i= 1 we use Exercise 2. Let f(z) = eZ - 1 +
2·
This eliminates the first term of the power series which defines the Bernoulli numbers, hence f(z) = 2:n#1~zn. Some straightforward computations show that I(z) - I( -z) = z Conclude using Exercise 2.
+z
-1 + eZ + e= - 1 1 1 = O. -e e= +
Z -
16
z ez/2
f:o
z eZ - 1
+ 2 = f:o
z eZ +
B2n 2n (2n)!z .
f as
f(z)
=
(2 2
z
Ll.
eZ _ 1
+1
2 eZ -
1=
2 ez/2
z ez/2
+ e-z/2
- e=t? .
Combining these two results we obtain the desired formula. Replacing z by 2rr iz in the left hand side of the above expression we obtain
01.0
e + ee' - e-l
iO
iO
·0
(J
Exercise 11.1.5. Express the power series for tan z. z/ sin z, z cot Z. in terms of Bernoulli numbers.
22n
n=O
(2n)!
?;(-It
00
f:o
00
B n 2n-l. 2Z
17
Finally we find the power series expansion of z] sin z. The trigonometric formula we use is
2z
If we have a factorization T2 UI
T - U2 = (T - a)(T - a'),
and a
=1=
a',
+ Ba
ln
with suitable A, B. Find A, B in terms of ao, ai, a, a'. Consider the power series
00
F(T) = Lanr. n=O Show that it represents decomposition. a rational function, and give its partial fraction
Solution. The existence of A and B is proved in the next exercise. Since A ao and Aa + Ba' = al we conclude that al - a'ao A = ----:-
+B
a -a'
and
B=
aaO - al a-a'
= A L(aTr n=O A
+ B L(a'Tr
n=O B 1 - a'T
00
=--+ 1 - «T
18
=A
+B
- AOt'T - BOtT .
(I - OtT)(1 - Ot'T)
generally, let ao, ... ,ar-I be given complex numbers. Let numbers such that the polynomial peT) =
r: -
(uITr-1
+ ... + ur)
I, ... , Ar
an = AIOt~
+ ... + ArOt~.
Solution. The reader can find a solution to this exercise in the appendix of Lang's book. Here we give another argument. Let (1:) be the system
ao
al
+ x; = XIOtI + + XrOtr
= XI = XIOt~-1
l
Since
Otl, .•. , Otr are
ar-I
+ XrOt;-1
distinct,
Otr
= D(Oti -Otj)
ii)
r-I Otl
Otr
r-I
is nonzero (this is the Vandermonde determinant), so (1:) has at least one solution, say (A I, ... , Ar). Now let S; be the statement
an =
LAk
k=1
Ot
k·
By construction, n .s N. Then
aN+I
= UlaN
= UI (A
-
+ ... + uraN+I-r
NAN) lOtI
+ ... +
...
+ A r OtN+1 r
where this last equality follows from the fact that P(Oti) = 0 for all 1 ~ i ~ r. So SN+I is true, and by induction we conclude that Sn is true for all n ~ o.
19
in its usual simple form. Solution. We simply use a geometric series ~ an __ a_m_+_l _-_a_ ~ a-I'
n=l
00,
so
L~l an
= aj(l-a)
which
Exercise 11.2.2. Let r be a real number; 0 ::::r < 1. Show that the series Lrnein9 n=O
00 00
and
L
n=-oo
rlnlein9
converge (fJ is real). Express that series in simple terms using the usual formula for a geometric series. Solution. We can use the comparison test because Irei9 I = r and in Exercise I we showed that r" converges. The expression of the series in simple terms is given by the following manipulation and the usual formula for the geometric series,
~ r" ein9 = ~ (rei9)n = I.. ~ ~ l-re,9 n=O n=O The second series converges for the exact same reasons the first series converged. To express the second series in a nice way we split the sum, L rlnlein9 = L rlnlein9 + Lrnein9 n=-OO n=-oo n=l so that
00
00
= Lrne-in9 n=O
00
+ Lrnein9
n=l
00
n=-OO
00
Exercise 11.2.3. Show that the usual power seriesfor log(l +z) or log(l-z)from elementary calculus converges absolutely for [z] < 1. Solution. We have
10g(1 +
z) = z - - + ... +
2
Z2
(_l)n+l
z + ... =
L
00
(_l)n+l
n=l
z.
20
Taking the logarithm and using an elementary limit we find limn->oo(l/n)l/n = 1, so the radius of convergence of the above series is 1. Clearly, the proof and the result is the same for 10g(1 - z). Exercise 11.2.4. Determine the radius of convergence for the following power series. (a) Lnnzn (c) L 2nzn
L 2- z (g) L ~zn
(e)
nn
Solution. In all of this exercise we let an denote the coefficients of the power series which we are dealing with. (a) Since IanII/n = n it follows that the radius of convergence is O. (b) In this case we have lanll/n = lin so the radius of convergence is 00. (c) We have IanII/n = 2 so the radius of convergence is 112. (d) For all large n, 1 :s logn :s n holds so 1 :s lanll/n :s n2/n. Since limn->oo n2/n = 1 (take the logarithm and use an elementary limit) we conclude that the power series has radius of convergence equal to 1. (e) Arguing like in (c) we find that the radius of convergence is equal to 2. (f) The radius of convergence of the power series is 1 because of the limit limn->oo n2/n = 1. (g) By Stirling's formula, n! = nne-nun with limn->oo u~/n = 1. Then we see that lanll/n = (nne:nnun )l/n =e-Iu~/n which implies that the radius of convergence is e. (h) Using the notation of (g) we get la n
II/n
--+ _
33
so the radius of convergence of the power series is 27. Exercise 11.2.5. Let fez) L anZn have radius of convergence r > the following series have the same radius of convergence:
(a)
o. Show
that
L na.z"
(b)
L n2a Z
n
Solution. We are given that L anZn has a strictly positive radius of convergence so it is sufficient to investigate the limit of the term next to an. We show that in all four cases, the limit is 1. We do (c) first. We have logn log(nd/n) = d-n
--+
as n --+ 00, so limn->oo nd/n = 1. Let d = I or d = 2 to get (a) and (b). For (d) we want to find the radius of convergence of the power series L~o(n + 1)an+ Iz",
21
+ 1)lan+ll)lln = ((n +
1)1/(n+l)lan+lll/(n+I»)(n+I)ln ,
L z" In2
I ) lin lim (- 2 n
= n--->oo lim
)2 = I .
To show that L IIn2 < 00 we can either use the integral test with hex) = I/x2 or use the fact that IIn2 < Iln(n - I) and that the partial sums of L I/n(n - I) are telescopic. Let fn(z) = zn In2 and let D be the closed unit disc. Then on D we have
Exercise 11.2.7. Let a, b be two complex numbers, and assume that b is not equal to any integer s: O. Show that the radius of convergence of the series
"
z:
in some cases.
Solution. First, note that if a is equal to some negative integer, the series has only finitely many terms, and in this case the radius of convergence is 00. Suppose a is never equal to some negative integer. Let
Cn
a(a
+ I)· .. (a + n)
I)···(b+n)
b(b+
L anzn
22
where 8 > O. [Hint: For this problem and the next, use summation by parts.] Solution. Let Tn(z) = LZ=o akzk and Sn(Z) = LZ=o Zk. The summation by parts formula gives n-l Tn(z) = anSn(Z) - L Sk(z)(ak+l - ak) k=O hence if n > m some straightforward computations show that Tn(z) - Tm(z) = an(Sn(Z) - Sm(Z» n-l L (Sk(Z) - Sm(Z»(ak - ak+d k=m+l
Summing a geometric series we find Sn(Z) = (zn+l - 1)/(z - I), so using the assumption that lz] :s 1 and Iz - 11 :::: 8 we get the uniform bound ISn(z)1 :s 2/8 for all n. Therefore ISn(z) - Sm(z)1 :s 4/8 for all m and n. Putting absolute values in the above displayed equation and using the triangle inequality and the fact that {an} is positive and decreasing we get ITn(z)-Tm(z)1 4 4 n-l :san-+L (ak-ak+l) 8 8 k=m+l =
4 =: + 4 8(am+1
an)
4 = 8am+1• Since am --+ 0 as m --+ 00 we conclude that the series L anZn is uniformly convergent in the domain lel :s land [z - 11 :::: 8 of the complex plane. Let L~o anZn be a power series with radius of convergence > 1. Assume that the series L~o an converges. Let 0 :s x < 1. Prove that
x-+ I n-+OO
L~l akxk,
A=
L~l ak and
n
An =
L~=l ai,
Consider
23
We first prove that the sequence of partial sums {sn(x)} converges uniformly for o :s x :s 1. For m < n, applying the summation by parts formula, we get
n
+L
n-I
There exists N such that for k, m > N we have IAk - Am+11 o :s x :s 1 and n, m ~ N we have ISn(x) - sm(x)1
:s
E.
Hence for
:s E + E
=E
n-I
+ E(X
+1
x")
:s 3E
This proves the uniform convergence of {sn(x)}. Now given E, pick N as above. Choose 8 (depending on N) such that if Ix 8, then
11 <
:s If(x)
:s If(x)
- sn(x)1 - sn(x)1
+ ISn(x) + 5E
- sN(x)1
+ ISN(X)
- ANI
+ IAN
- AI
for all n ~ N and Ix - 11 < 8. For a given x, pick n so large (depending on xl) so that the first term is also < E, to conclude the proof. This argument is in the appendix of Lang's book.
Exercise 11.2.10. Let L anzn and L bnzn be two power series, with radius ofconvergence rand s, respectively. of the series: (a) L(an + bn)zn Solution. What can you say about the radius of convergence (b)L anbnzn ?
+ bnllzln :s L
m lanllzl
n=O
+L
n Ibnllzl .
If lzl < miner, s), then L(an + bn)zn converges, so we can say that the radius of convergence of this last series is ~ miner, s). (b) Since Iim sup e.v, :s (limsupun)(limsupvn) it follows that the radius of convergence of the series L an b; z" is ~ r s .
24
Solution. By Exercise 10 we see that the radius of convergence of the series is at least 1/1,81. We contend that this radius of convergence is exactly 1/1,81. Let an = 3an - 5,8n. Then
Exercise 11.2.12. Let {an} be the sequence of real numbers defined by the conditions: ao
1,
al
= 2,
and
an
= an-l + an-2
for n 2: 2.
[Hint: What is the general solution of a difference equation? Cf. Exercise 6 of §l.J Solution. This exercise is a special case of Exercise 13.
Exercise 11.2.13. More generally, let Ul, U2 be complex numbers such that the polynomial
has two distinct roots with lall < la21. Let ao, al be given, and let an = Ulan-l
+ U2an-2
forn
2: 2.
L: an T" ?
Solution. We solve the difference equation as was done in Exercise 6, § 1 of this Chapter, so that we find an = Aa~
+ Bai
where A and B are complex numbers determined by ao, ai, al and a2. Then lanl
= Bla21n
I~ + 11. (::Y
25 <
(b)Letzo define
i= O.Letabeacomplexnumbersuchthatexp(a)
log z =
= zo.Forlz-zol
lzol
C: - 1)
+ a.
eX=L- x" n!
00
n=O
00
log x =
L
n='
(_l)n-' n
(x - l)"
We know that exp log x = x for all x > 0, so if z is real and lz - 11 < 1 then exp log z = z. Combined with a translation to the origin, the uniqueness theorem (Theorem 3.2) implies the desired result. (b) The given inequality implies that Iz/zo - 11 < 1, so by (a) explogz
= (expj
C: - I)).
00
(exp«) = z: zo = z.
Note that we have used the fact that exprz, + Z2) = (exp zj jtexp z-) which can be easily proved using the uniqueness theorem twice. Exercise 11.3.2. (a) Let exp(T) = L::o log(l show that explog(1 t»/ n! and [k,
+ T) = L(-I)k-'Tk
k='
+ T) = 1 + T
= T.
(b) Let h, (T) and h2(T) be formal power series with 0 constant terms. Prove that log«(l + h,(T))(1 + h2(T))) log(l + h,(T)) + 10g(1 + h2(T)). (c) For complex numbers a, f3 show that log(l + T)et a loge 1 + T) and
(I
+ T) = 1 + T.
26
(b) To prove this assertion, we first note that substituting power series in part (a) we find that exp log(l + h(T» = 1 + h(T) for all power series h(T) with 0 constant term. 1l1erefore exp logtj l
+ hl(T»(l + h2(T») = explog(1 + hl(T) + h2(T) + hl(T)h2(T» = 1 + hl(T) + h2(T) + hl(T)h2(T) = (1 + hl(T»(l + h2(T»
= explog(1 = exp(log(1
+ hl(T»explog(1
+ h2(T»
So 10g«1 + hi (T»(1 + h2(T») and by a constant multiple of 2Jr i. Evaluating at 0 we find that this constant is O. (c) If a is real, then for all real T we have the identity log( I + T)et = a log( 1 + T), so when a is real, we have equality.of the two power series in T. Now log(l + T)et is a power series with real coefficients which we may consider polynomials in a, namely 10g(1 + T)et = Lan(a)P where an(a) are real polynomials in a. Similarly, a 10g(1 + T) is also a power series in T whose coefficients are real polynomials in the variable a, say, a 10g(1 + T) = L bn(a)Tn. Then, since an(a) = bn(a) for all real a, this equations also holds true for all complex o. Hence 10g(1 + T)et = a log(l + T) for all complex a. Since (1 + T)et = 1 + het(T) and (1 + T)P = 1 + htl(T) where het(T) and hp(T) are power series with 0 constant term, we get from (b) 10g«1 Since log(l
+ Tnl
+ T)tl)
= 10g(1 + Tr = a 10g(1 + T)
+ 10g(1 + T)P.
+ T)et = a 10g(1 + T)
10g«1
we find that
+ T)et(1 + T)fi)
Exponentiating
eiz
+ e="
2
and sinz
eiz
e-iz
+ e-
IZ
neven
n.
n=O
(2n).
= 2cosz.
Use the same argument to prove the formula for the sine.
27
Exercise 11.3.4. Show that the only complex numbers z such that sin z = 0 are z = ktt , where k is an integer. State and prove a similar statement for cos z. Solution. By the previous exercise, the equation sin z = 0 is equivalent to eiz = e-iz or simply e2iz = 1. Writing z = x + iy where x and yare real we find that the equation sin z = 0 is equivalent to e2ix e-2y = 1. Since e2ix has absolute value 1 it follows that e-2y = 1 so Y = O.Therefore the real part of z must verify e2ix = 1, so x = kit with k E Z. For the cosine the exact same argument shows that cos z = 0 if and only if z = kit /2 where k is an integer. Exercise 11.3.5. Find the power series expansion of f(z) = 1/(z find the radius of convergence. Solution. namely
We use partial fraction decomposition to modify the expression of I, 1 (z+I)(z+2) z+1 z+2 1 1- (-z) 1
21- (-z/2)
The formula for the sum of a geometric series implies that for [z I < 1 we get
00
2: L(-z/2r
n=O
00
L(-1)n( n=O
oo
1- -
2n+1
1)
n z.
2n~l) l/n
= 1, the radius of convergence of the power series is can de defined as the coefficients
f(z) = (1 - 2az
Calculate the first four Legendre polynomials. Solution. Squaring both sides we get 1 21 az
2
+z
(f:
Pn(a)zn)2 n=O
+ Z2) (~dn(a)Zn)
28
II. PowerSeries
=L
00
00
+ L dn(a)zn+2
00
n=O
= do(a)
+ (dl(a)
- 2ado(a»
+ L(dn(a)
n=2
00
- 2adn_1(a)
+ dn(a»zn.
+ dn(a)
=0
which allow us to find the desired Legendre polynomials. We find PI(a) = a, P2(a) = 4(3a2 - 1), P3(a) = 4(5a3 - 3a) and for the fourth polynomial we find P4(a) = i(35a4 - 30a2 + 3).
= 2.tn!D" [(a
2-
1)n].
f. We write
(z - 1»,
fez) = [(z - 1)2 + 2(z - 1) + 1] ( - ~(Z Therefore the beginning of the power series expansion of -1 - (1
- l)k) .
is - 1)3.
+ 2)(z
- 1) - (1
+ 2 + 1)(z - Ii - (1 + 2 + 1)(z
Exercise 11.4.2. Find the terms of order ~ 3 in the power series expansion of the function fez) = (z - 2)f(z + 3)(z + 2) at Z = 1.
+3 -
+ 2·
29
5(
...)
... ) .
and the same method shows that _4_ = ~ (1 _ ~(z - 1) + _!_(z - 1)2- _!_(z - 1)3 + z +2 3 3 32 33 Therefore we find that the beginning of the power series of
f at 1 is
(
5 - - - ( -5 - -4 ) (z - 1) + ( -5 4) (4 3 42 32 43
- -4) 3
3
(z - 1)2 -
-5 - -4) (z - 1)3 . 44 34
Z2n (2n)!
Let
z2n
+ zf'(z)
= 4z2 fez).
Solution. By the ratio test we see that the radius of convergence of the series is 00. Differentiating we find
f (z)
so that
2n 8(n!)2Z
2n-1
and
f"()
Z
=~
82n(2n (n!)2
1) 2n-2 z
30
"t" _ ~
=r
=
2n(2n - 1) 2n _ ~ (n!)2 z -4
=r
2 n 2n , (n!)2z - zf (z)
+ zf'(z)
4?;
00
(:!)2Z2n =
4?;
00
1 2 «n _ 1)!)2Z2n = 4z f(z)
as was to be shown. Exercise 11.5.3. Let f(z) Show that j'(z) = 1/(Z2 =z Z3 Z5 Z7
+ 1).
Solution. Since lim sup(l/n)l/n ing term by term we get f (z) = 1 - z Exercise 11.5.4. Let
, 2
= 1 the radius
~
+z
- ... = ~(-z)
n=O
2n
?;
00
+ zJ'(z) + Z2J(z) = O.
Solution. The ratio test implies that the radius of convergence is 00. Let S(z) = Z2J"(z) + zJ'(z) + Z2J(z). Then differentiating term by term, some elementary manipulations show that
However (Z2 + 4n2) (_l)n 2n (_1)n 2(n+l) (_l)n-1 2n (n !)222n Z = (n !)222n Z - «n _ 1)!)222(n-1) Z so we have a telescopic sum, which implies that S(z) Exercise 11.5.5. For any positive Jk(Z) integer k, let
= 0, as was to be shown.
.
?;
00
31
Solution. The ratio test implies that the radius of convergence is a - -::--:---n - 22n+kn!(n
Let
(-It
+ k)!
Z2
Differentiating term by term we find that the coefficient of z2n+k in the sum Jt(z) + zJ£(z) + (Z2 - k2)J(z) is = an(2n
+ k)(2n + k - 1) + an(2n + k) + an-I - k2an = (4kn + 4n2)an + an-I = -4n(n + k) so Z2Jt(z) + zJ£(z) + (Z2 - k2)h(z) = 0 as was
to
Exercise 11.5.6. (a) For lz - 11 < 1, show that the derivative of the function logz = 10g(1
is l/z. (b) Let Zo i= O. For that f'(z) l/z.
+ (z -
1» = L(_1)n-I--n=1
00
L(-lt-l(z
n=l
00
1)n-l
1 1-(-(z-l»
1 = -. z
L
00
(_1)n-l
n=l
Zo
(z - zot-1
L
00
(_l)n-l
n=l
Zo
(z- Zo
=
11.6
1 zol-(z/zo-l)
- z.
Determine which of the following functions are local analytic isomorphism at the given point. Give the reason for your answer.
=e
at
= O.
= f(z), so f'(O)
Solution. By definition we have f(z) = Lzn/n! and f'(z) 1 i= O. Therefore f is a local analytic isomorphism at O.
= sin(z2) at Z = O.
32
Solution. For all z near zero we have sinrz") = sinCe_Z)2) so there does not exist and open ball around 0 such that the given function is an analytic isomorphism in this ball. Exercise 11.6.3. fez) Solution. We write fez) so that
= (z -
z- 2
= -(1
z near 1 we have
= -(z
- 1) [1
+ (z
- 1)
+ (z -
1)2
+ ... ]
f' (1)
= -1.
1.
Exercise 11.6.4. fez) = (sin Z)2 at Z = O. Solution. For all z near 0 we have (sin Z)2 = (sin -Z)2 and therefore f cannot be a local isomorphism at O. Exercise 11.6.5. fez)
= cos z at Z = Jr.
+ x) so f is not a
Solution. For all real x near 0 we have cosor - x) = cosor local analytic isomorphism at O. Exercise 11.6.6 (Linear Differential Equations). Prove:
Theorem. Let ao(z), ... , ak(Z) be analytic functions in a neighborhood of o. Assume that ao(O) :f O. Given numbers co, ... , Ck-l, there exists a unique analytic function f at 0 such that
D" f(O) = Cn for n = 0, ... , k - 1 and such that ao(z)Dk fez)
+ al (z)Dk-l
fez)
+ ... + ak(z)f(z)
= O. f byaformalpower
[Hint: First you may assume adr) = 1 (why?). Thensolvefor series. Then prove that this formal series converges.]
Solution. The solution of this exercise is in the appendix of Lang's book but for the sake of completeness we will repeat the argument here. Since a(O) :f 0 we must have a(z) :f 0 in a neighborhood of 0 which shows after dividing through by ao that we may assume ao = 1. We change notation a little to be in accordance with the appendix of the book. In fact, we prove the following Theorem. Let p be an integer j; 2. Let go, ... , gp-l be power series with complex coefficients. Let ao, ... , ap-l be given complex numbers. Then there exists a unique power series f (T) = L an T" such that DP f = gp_lDP-l f
+ ... + gof.
If go, ... , gp-l converge in a neighborhood of the origin, then so does f. The coefficient an (n ::: p) of f will be determined inductively and uniquely. Then we prove that the power series L anzn converges in a neighborhood of the
33
DP f(T)
= Ln(n n=p
- I) ... (n - p
+ l)anTn-p,
DP f(T)
= ~)m m=O
00
+ p)(m + p
- 1)··· (m
+ l)am+pTm.
Similarly, for every positive integer s with 0 ::::s ::::p - 1 we have DS f(T) = ~)k k=O
+ s)(k + s -
1)··· (k
+ l)ak+sTk. + 1) for s
2: 1
It will be useful to use the notation [k, s] = (k + s)(k + s - 1)· .. (k and [k, s] = 1 if s = O. Next we write down the power series for each gs, say
s. = Lbs.jTj.
j=O Then gsD f(T)
S
00
=L
cs,mTm
where cs,m
(1)
Hence once we are given ao, ... , ap-l we can solve inductively for am in terms of ao, ... , am-l and the coefficients of gl, ... , gs by the formula am+p
CO,m+
... + Cp-l,m
[m, p]
= L.....
~"
L..... --ak+s
[k, s]
s,j
(2)
which determineszn.L, uniquely in tennsof ao, ... , am+p-l, bs,l, ... , bs,m' Hence we have proved that there is a unique power series satisfying the differential equation. Assuming that the power series go, ... , gp_1 converge, we must now prove that f(T) converges. We select a positive number K sufficiently large 2: 2 and a positive number B such that laol ::::K, ... , lap-Ii and for all s = 0, ... , p - 1 and all j we have Ibs,j ::::K
I :::: K
e'.
(3)
The standard m-th root test for convergence then shows that f(T) converges. We note that the expression (1) for Cs,m and hence (2) for am+p have positive coefficients as polynomials inao, al, ... and the coefficients bs,j of the power series
34
gs' Hence to make our estimates, we may avoid writing down absolute values by replacing bs,j by K e', and we may replace ao,... , ap_1 by K. Then all values am+p (m ~ 0) are positive and we want to show that they satisfy the bound in (3). Observe first that for 0 ::::k :::: m and 1 ::::s ::::p - 1 we always have [k,s] < 1. [m, p] Hence the fraction [k, s]/[m, p] will be replaced by 1 in the following estimates. Now first we estimate am+p with m = O. Then k + j = 0, so k = j = 0 and p-I ap ::::L L ak+sbs,j s=ok+j=O
s pK2
as desired. Suppose by induction that we have proved (3) for all integers > 0 and < n. Then p-I a n+p _ '" <~ '" a k+sb·s.] _ p '" 2n-lpnK2Kk-IBkKBj < ~ ~ s=ok+j=n k+j=n ::::2n-1 pn+1 L K3 Kk-I Bk e! k+j=n
s 2n-1
r= K2 L
=0
and
I'(z)
= g(f(z».
[Hint: Again find a formal solution, and then prove that it converges.] Solution. Again, the proof is in the appendix of Lang's book, but we repeat the argument for sake of completeness. Let geT) = L bkTk and write I(T) with unknown coefficients I(T) = LamTm. m=1
00
35
= 'L(n
+ l)an+IP'
2:)n
n=O
l)an+I T" = bo
+ bvf(T) + b2!(Ti
+ ....
= bo, and
where P; is a polynomials with positive integer coefficients. In particular, starting with al = bo, we may then solve inductively for an+1 in terms of ai, ... ,an for n ~ 1. This proves the existence and uniqueness of the power series f(T). Assume next that g(T) converges. We must prove that f(T) converges. Let Bk be positive numbers such that IbkI ~ Bi, and such that the power series G(T) = 'L BkTk converges. Let F(T) be the solution of the differential equation F'(T) = G(F(T)), and let F(T) = 'L AmTm, with AI> (n 0 and lad ~ AI. Then , Bn; AI, ... , An),
with the same polynomial Pn. Hence lan+d ~ An+h and if F(T) converges so does f(T). Since g(T) converges, there exists positive numbers K, B such that Ibkl~KBk We let for all z
ee
Oi I, ....
s, = K Bk. Then
G(T)
=K
L BkTk = 1-KBT ,
00
k=O
and so it suffices to prove that the solution F(T) of the differential equation F'(T) = K 1- BT
converges. This equation is equivalent with F'(T) = K which we can integrate to give F(T) = KT By the quadratic formula, we find F(T)
+ BF(T)F'(T), + B-F(Ti -. 2
+ ....
I - (1 - 2K BT)I/2 B = KT
We then use the binomial expansion which we know converges. This concludes the proof.
III
Cauchy's Theorem, First Part
IlL 1
Lemma. Let S be a subset of an open set U. Then S is closed on U if and only if the complement of Sin U, that is, U - S is open. In particular, if S is both open and closed in U, then U - S is also open and closed in U. Solution. Suppose S is closed, and let W E U - S. Since U is open, U - S is not empty. Then for some n the ball of radius 1/ n centered at w is contained in U and this ball does not intersect S, for otherwise w is in the closure of S which contradicts the fact that S is closed and w E S. Hence U - S is open. Conversely, suppose that the set U - S is open, and let Z E U and Z E S, where S denotes the closure of S. If z is not in S then there exists a ball centered at z which is contained in U - S. So z is not adherent to S, a contradiction which ends the proof.
Exercise 111.1.2. Let U be a bounded open connected set, Un} a sequence of continuous functions on the closure of U, analytic on U. Assume that Un} converges uniformly on the boundary of U. Prove that {fn} converges uniformly on U. Solution. We use Cauchy's criterion for uniform convergence. Let 1I·llau and 11'llu denote the sup norm on the boundary of U and the sup norm on U respectively. Let E > 0 and choose N such that for all n, m > N we have Ilfn - fm lIau < E. By the maximum modulus principle, the function fn - fm attains its maximum on the boundary of U, so we have II fn - fm II u < E for all n, m > N which implies the uniform convergence of the sequence Un} on U.
37
Exercise 111.1.3. Let a I, ... , an be points on the unit circle. Prove that there exists a point Z on the unit circle so that the product of the distances from Z to the a j is at least 1. (You may use the maximum modulus principle.)
Solution. Consider the map
f : D(O, 1) -+ C given by
n
fez) = (z - al)'"
The closed unit disc is compact and the function f is continuous on D(O, 1) and analytic in the open unit disc. We have If(O)1 = I and If(ak)1 = 0 for all k 1, ... , n which implies that f is not constant. By the maximum modulus principle, f attains its maximum at a point z of the unit circle, so If(z)1 2: If(O)1 = 1.
(z - zotdz
for all integers n, positive or negative. (b) Suppose f has a power series expansion fez) =
L ak(Z ()()
ZO)k
k=-m
which is absolutely convergent on a disc of radius > R centered at zo. Let C R be the circle of radius R centered at ZOo Find the integral
JCR
Solution. (a) Suppose n
f(z)dz.
n+
=f
so
l
".
(z -
zor dz = O.
If n
= -1
-= c Z - zo
dz
1
0
y'(t) dt = yet) - Zo
1
0
". ie" r
-.-dt re"
3~
L~=-mak(Z
1
CR
f(z)dz
= 2rria_l.
= eZ from -3 to 3 taken along a semicircle. Is this integral different from the integral taken over the line segment between the two points? Solution. Since f' I, the function f is its own primitive. Therefore the integral is independent of the path and is equal to f(3) - f( -3).
+ t2
Solution.
1.
Jl.
Exercise 111.2.4. Find the integral of each one of the following functions over
each one of the curves in Exercise 3. (a) f(z) = Z3 (b) f(z) Z (c) f(z) l/z
= =
(I~i)4
(b) (i) i + 1/2 (ii) =xi (iii)rri (iv) 2 + (c) (i) 10g.J2 + i~ (ii) -rri (iii) n i (iv) log
¥.
,,/5 + i arctan(I/2)
39
(at from the point i to the point -i + 2, taken along the straight line segment, and (b) from 0 to 1 + i along the parabola y = x2• Solution. We see that the function g defined by g(z) = 4ez2 is a primitive for zez2 so for (a) the integral is equal to g( -i
+ 2) -
+ i)
1 - g(O) = 2(e
2'
I-
1).
sin zdz
= - cos z is a holomorphic
1)
primitive
sin
where zo is a fixed complex number, and c is a fixed real number> O. (Draw a picture.) Let a = zo + x and a' = Zo - x, where x is real positive. Find hm
x~o
. i( z -1a
o
-----
z - a'
1)
dz.
(Draw the picture.) Warning: The answer is not Of Solution. The picture is the following:
40
Since a'(t)
= ic we have
1(
<7
-- 1
z- a
- -- 1) z - a'
d:
= ie
11
-1
Reducing to the same denominator we find that this last integral is equal to = ie
2x
-1 -t2e2 - x2
-2ie dt = -X
11
-1
1
(te/x)2
+1
dt
-1
-4i arctantc/x).
x---+o
1(_1_ - _1_,)
<7
z-
z-
d.: = -2ni.
>
lim
8---+00
1 (1 +
8 -8
-t
ix
- -t - ix
1)
dt.
Solution. We write
--
t+ix
--
-2ix
t2+X2
-2i
t=-Lx
X (t/x)2+1
----::,-----
-»
x
[x arctan(t/x)]~8
= -4i arctan(B/x).
8---+00 -8
1 (_1_. +
8
IX
__ t - 1_. )dt
IX
= -2ni.
41
= yea + b -
lyly-
F=-
ly
r F.
t))y'(a
F=
la
-F(y(a
+b -
+b -
t)dt.
Changing variables u = a
+b -
t we obtain =-
F=
lb
r F(y(u))y'(u)du
ly
r F,
Exercise 111.2.10. Let [a, b] and [c, d] be two intervals (not reduced to a point).
Show that there is a function get) = rt + s such that g is strictly increasing, g(a) = c and g(b) d. Thus a curve can be parametrized by any given interval.
:=~
and s
=c-
lib F(t)dtl
:::: ib
IF(t)ldt.
(Hint: Let P = [a = aD,aI, ... , an = b] be a partition of [a, b]. From the definition of integrals with Riemann sums, the integral ib F(t)dt is approximated by the Riemann sum
L F(ak)(ak+l
k=O
n-l
- ak)
L IF(ak)l(ak+l
k=O
n-l
- ad·
The proof is concluded by using the triangle inequality.] Solution. Given E > 0 there exists 8 > 0 such that for any partition of size < 8 we have the following two inequalities:
(1)
42
lIb
lib
SO for all FI ~
IFI - ~ IF(adl(ak+1
- adl <
(2)
I~
F(ak)(ak+1 - ak)1
+E
IF(ak)l(ak+1
- ak)
+E
lib
E
FI ~ ~
IF(adl(ak+1
- ak)
+E ~
Ib IFI +
2E.
r: I ~ r:
F
IlL 5
Exercise 111.5.1. A set S is called star-shaped if there exists a point Zo in S such that the line segment between zo and any point z in S is contained in S. Prove that a star-shaped set is simply connected, that is, every closed path is homotopic to a point. Solution. Consider a closed path y : [a, b] ~ S. Then by hypothesis we see that we have a homotopy Fy : [a, b] x [0, 1] ~ S defined by Fy(t, u) = (1 - u)y(t)
+ uzo.
Hence every closed path in S is homotopic to zo which implies that S is simply connected. Exercise 111.5.2. Let U be the open set obtained from C by deleting the set of real numbers ~ O. Prove that U is simply connected. Solution. By the previous exercise, it is sufficient to show that U is star-shaped. Let zo = -1. A picture shows that U is star-shaped, but here is a formal argument. Given z = x+iyconsiderthesegment[z, Zo]parametrizedbyy(t) = tz+(l-t)zo. The imaginary part of y(t) is given by ty so we see at once that the line segment is contained in U. Exercise 111.5.3. Let V be the open set obtainedfrom numbers ~ O. Prove that V is simply connected. Solution. C by deleting the set of real
43
Exercise 111.5.4. (a) Let U be a simply connected open set and let f be an analytic function on U. Is feU) simply connected? (b) Let H be the upper half-plane, that is, the set of complex numbers z = x + i Y such that y > O. Let fez) = e27riz. What is the image f(H)? Is f(H) simply connected?
Solution. We first solve (b). If z = x fez)
+ iy ; then
we see that
= e-27rYei
(2rrx)
If y > 0, then 0 < e-2rry < 1 so the image of H is the open unit disc minus the origin which is not simply connected. This shows that the image of a simply connected set under an analytic map need not be simply connected. This answers (a) because we see that the image of a simply connected set need not be simply connected.
3;
(h) loge -1 - i) =
4 log 2 + i 5;
Exercise 111.6.2. Compute the values of the same expressions as in Exercise 1 (except if) and (g)) when the open set consists of the plane from which the negative
real axis has been deleted. Then take -rr < () < tt . Solution. (b)log(-i) (a) log i if = loge3irr/2 =
(c) 10g(-1 + i) = (d) i' = e-rr/2 (e) (-iY = err/2 (h) log( -1 - i)
-if
44
Exercise 111.6.3. Let U be the plane with the negative real axis deleted. Let y > O. Find the limit
y--->O
lim [log(a
+ iy)
-log(a
- iy)]
where a > 0, and also where a < O. Solution. First suppose a > O. Let ry = (a2 + y2)1/2. Then for small y > 0 we can write a + iy = ryeiEy and a - iy = rye-iEy with 0 < Ey < n12. Hence log(a and since
+ iy)
- log(a - iy)
log ry
+ ie; -
log ry - i( -Ey)
2iEy
Ey ~
0 as y ~ 0 we find that
y--->O
lim[log(a
+ iy)
-log(a
- iy)] = O.
Now suppose a < O. In this case we can write a ryei(-tr+Ey) so that arguing as above we get log(a therefore
y--->O
+ iy
= ryei(tr-Ey) and a - iy =
+ iy)
lim [log(a
+ iy)
-log(a
- iy)]
= hi.
Exercise 111.6.4. Let U be the plane with the positive real axis deleted. Find the limit
y--->O
lim [log(a
+ iy)
-log(a
- iy)]
where a < 0, and also where a > O. Solution. We argue as we did in the previous exercise. When a > 0, write a + i y = ryeiEy and a - iy = ryei(2tr-Ey) so that
y--->O
lim [log(a
+ iy)
-log(a
- iy)] = =Ln i. iy
= ryei(tr+E
y)
and therefore
lim [log(a
- iy)] = O.
Exercise 111.6.5. Over what kind of open sets could you define an analytic function ZI/3, or more generally z'!" for any positive integer n? Give examples, taking the open set to be as "large" as possible. Solution. We can define an analytic function z'!" over any simply connected open set not containing the origin. Indeed, on such a set the log is well defined. Then we define
45
and assume that fez) =I- Ofor all Z E U. Show that there exists an analyticfunction g on U such that g2 = f. Does this last assertion remain true if2 is replaced by an arbitrary positive integer n? Solution. We show that the assertion is true for an arbitrary n > O. It is shown in the text that we can define log fez) on U. Define gn by gn(Z)
f(d1n
= e~ logf(z).
Z = x + iy with y > O. Let «p(z) = e21CizProve that «p(U) is the open unit disc . from which the origin has been deleted.
Exercise 111.6.7. Let U be the upper half plane, consisting of all complex numbe rs
Solution.
If z = x
+ iy
Since x ranges over Rand y > 0 it is clear that «p(U) is the open unit disc minus the origin.
real axis from the complex numbers. For an integer m ~ I define L_m(z) = (IOgz _ (1
+ ~ + ... + ~))
:m!.
Show that L'-m(z) = L-m+l(Z), and that L'-l(Z) = log z. Thus L_m is an m-fold integral of the logarithm. Solution. Using the rule for differentiating a product we find that
L'-m(z) = ~zm Z m!
+ (IOgz
-
(.!_ + logz
m ,
+ ~ + ... +
2
m
(m -
-z
+ (log z -
1)
= logz
III.7
z
Exercise 111.7.1. Find the integrals over the unit circle f, sin z d : (C) r cos(Z2) dz
y
Jy
46
Solution. (a) We use the local Cauchy formula on D(O, 1). Let !(z) zo = O. Then
!(zo)
I = -2.
tt t
1
y
= cos z and
-dz.
cos z
Therefore
1,y
cos Z Z
dz
= 2n i,
1,y 1,y
Z
(b) The same argument as in (a) shows that (c) Arguing like in (a) we find
sinzdz = O.
Z
coS(z2)dz = 2ni.
Exercise Ill.7 .2. Write out completely the proof of Theorem 7.7 to see that all the steps in the proof of Theorem 7.3 apply. Solution. Let Zo E U and zo not on y. Since the image of y is compact there is a minimum distance between zo and points on y. Select 0 < R < dist(zo, y) and take R also small enough that the closed disc D(zo, R) is contained in U. Select o < s < R. We write ~ ~ z = ~~ Zo
= ~~ Zo
(I + ; =:: + (; =:: Y + .. -) .
zol .::::s because
1;
(I _ 1.L.lQ.) ~-Zo
= ::
=
I.:::: ~
< 1.
The function g is continuous on y so it is bounded. By Theorem 2.4 of Chapter III, we can integrate term by term and we find
!(z)
L an(z
00
- zn)"
n=O
where an-
ly (~ _ zo)n+l
= n!
g(~)
~.
I, (~ zo)n+l -
g(n
d~
thereby concluding the proof of Theorem 7.7. Exercise 111.7.3. Prove Corollary 7.4, that is, prove: Corollary. Let f be an entire junction, and let II f II R be its sup norm on the circle of radius R. Suppose that there exists a constant C and a positive integer k such that
47
.:s k.
Solution. Let n > k. The estimate lanl .:s IlfiIR/ R" of the coefficients in the power series expansion of f gives in our particular case the inequality
lanl .:s
C Rk/ Rn
which holds for all large R. Letting Ri-v co we get that an 0 for all n > k. So in the power series expansion of f, all terms of degree > k are equal to 0, whence f is a polynomial of degree .:s k.
IV
Winding Numbers and Cauchy's Theorem
IV.2
f t--+ f'/ f (where f is holomorphic) sends products to sums. (b) If P(z) = (z - a,)· .. (z - an), where a" ... ,an are the roots, what is P' / P? (c) Let y be a closed path such that none of the roots of P lie on y. Show that -. 1 Solution.
bTl
1
y
(P I / P)(z)dz
= W(y, a,)
+ ... + W(y,
an).
(a) The product rule for differentiation implies that (fg)' f'g fg' --=-+-=-+-, fg fg fg f' f g' g
so the association f t--+ f' / f sends products to sums. (b) Let us write cJ>(f) = f' / f. Part (a) implies that cJ> (Il fk) fk = z - as, then (f~/fk)(Z) = 1/(z - ak) and therefore P' cJ>(P)(z) = -(z) P (c) By definition 1 W(y, ak) = -2.
JTI
= L cJ>(fk)so if
= L --.
n
k='
Z-
ak
y Z-
dz --,
ak
49
= _1_.
2Jrl
t1 .ss: t:
k=1
YZ-
ak
k=1
W(y, ad
as was to be shown.
Exercise IV.2.2. Let fez) (z - zo)mh(z), where h is analytic on an open set U, and h(z) i= 0 for all z E U. Let y be a closed path homologous to 0 in U, and such that Zo does not lie on y. Prove that
Solution. We have
--=--+-.
fez) z - Zo
f'(z)
h'(z) h(z)
The hypothesis that h does not vanish on U implies that h' / h is holomorphic on U and therefore by Cauchy's theorem its integral along y is O. So
1 2n I
.1
Y
ff'((z»dz
Z
21 nI
.1
I
Y Z-
__!!!__dz
Zo
= W(y,
zo)m.
Exercise IV.2.3. Let U be a simply connected open set and let z 1, ... , Zn be points of U. Let U* = U - {z 1 , ••• , Zn} be the set obtained from U by deleting the points Zl, ... , Zn. Let f be analytic on U*. Let Yk be a small circle centered at Zk and let
ak
= 2Jr
-1-.1
Yk
f(~)d~.
Let h(z) fez) - L ak/(Z - zr). Prove that there exists an analytic function H on U* such that H' = h. Solution. Fix a point define
W
to a point
ZE
U*
Hy(z) = [ h(nd~. We claim that this function is independent of the path chosen from w to z. Indeed, suppose TJ is another path from w to z. Then we have Hy(z) - HlI(z) =
h(nd~,
where /1- is a closed curve in U*. Since U is simply connected, the path /1- is homologous to 0 in U. By Cauchy's theorem, we see that if mk denotes the winding number of /1- with respect to u, then
1
/L
h(nd~
= I:mk1
h(nd~.
Yk
50
By construction, we have
h(Od~ =0
which proves our claim. So we may use the notation H to denote the function defined above. The standard argument then shows that H is analytic and that H' h. This concludes the proof.
v
Applications of Cauchy's Integral Formula
V.I
U and f'(zo)
=I O.
-1
-
1 d c f(z) - f(zo) z,
where C is a small circle centered at Zoo Solution. For z near Zo we can write f(z) - f(zo) with al = f'(zo) =I O. So f(z) - f(zo) = al (z - zo)
= a, (z (z - zo)
zo)
+ a2(z -
ZO)2+ ...
(1 + :~
1
+ .. -)
which after inverting the expression in parenthesis implies that on a small disc around Zo the function given by f(z~=f(zo) is analytic, thus by Cauchy's formula We get
1
27ri
dz c f(z) - f(zo)
= al =
1
!'(zo)·
real interval [a, b] states that a continuous junction can be uniformly approximated by polynomials. Is this conclusion still true for the closed unit disc, i.e., can every continuous function on the disc be uniformly approximated by polynomials?
52
Solution. Since polynomials are holomorphic, the uniform limit of a sequence of polynomials is holomorphic. However, not every continuous function on the disc is holomorphic (z f-+ Z for example), so the conclusion of Weierstrass' theorem is false for the closed unit disc. Exercise V.l.3. Let a > O. Show that each of the following series represents a holomorphic function: (a) L~ 1 e-an2z for Re(z) > 0;
e-an2z and z
= x + iy
If x = Re(z) 2: c, then e-an2x s e-an2c :::e-anc and the geometric series L(e-aC)n converges for c > 0 so we conclude that the series L e-an2z converges uniformly for Re(z) 2: c. Clearly, the functions fn are holomorphic so the series L e-an2z defines a holomorphic function on Re(z) > O. (b) Let fn(z) = e-anz /(a+n)2. The functions fn areholomorphic and for Re(z) > 0 we have le-anzl < 1 which implies that Ifn(z)1 ::: l/n2. The convergence of L l/n2 implies that the series L e-anz /(a + n)2 defines a holomorphic function for Re(z) > O. (c) Let fn(z) = 1/(a + n)z. Then Ifn(z)1 = 1/(a + nY, so if Re(z) 2: 1 + E, for E > 0, we get that Ifn(z)1 ::: l/n 1+(. Since L l/n IH converges, we conclude that the function L 1/(a + n)Z is holomorphic for Re(z) > 1. Exercise V.1.4. Show that each of the two series converges uniformly on each closed disc iz] ::: c with 0 < c < I:
and
n=1
For all sufficiently large n, 1 - cn 2: 1/2, so for all large n we have lan(z)1 :::Znc", The ratio test implies at once that L nc" converges, so the series L nz" /(1 - z") converges uniformly on each closed disc [z] ::: c with 0 < c < 1. Consider the series L z" / (l - zn)2, and let an(z) = zn/ (1 - Zn)2. Estimating lan(z)1 as we did for the first series we find that for large n the inequality lan(z)1 :::4cn holds. But L c" converges, so disc [z] ::: c with 0 < c < 1.
L z" /(1-
53
Exercise V.I.S. Prove that the two series in Exercise 4 are actually equal. [Hint: Write each one in a double series and reverse the order of summation.] Solution. We use the fact that l~zn = L~o(zn)k to get
n=l j=l
j=l n=l
. ~(znt]
00
= ~ zn ~(k
00
+ 1)(zn)k
L L(k + 1)(zn)k+l = L L
n=l
iev.
k=O
n=l j=l
It is now clear that both series are equal. Exercise V.1.6 (Dirichlet Series). Let {an} be a sequence of complex numbers. Show that the series L an/ns, ifit converges at allfor some complex s, converges absolutely in a right half-plane Re(s) > (To,and uniformly in Re(s) > (To+ E for every E > O. Show that the series defines an analytic function in this half plane. The number (Tois called the abscissa of convergence. Solution. Suppose that the series converges for some so. Let (To= Re(so) + 1 and suppose that Re(s) > (To+ E. The series L an/nso converges, so for all large n we have the inequality lan/nsol
lanl/nRe(so) < 1
which implies that for all large n we have lan/nsl ::::lanl/nO"oH = lanl/nRe(so)+I+€::::1/n1+€. Conclude. Exercise V.I.7. Let f be analytic on a closed disc D of radius b > 0, centered at ZOo Show that
~f[
nb
iD
f(x
+ iy)dydx
= f(zo).
[Hint: Use polar coordinates and Cauchy's formula. Without loss of generality, you may assume that zo = O. Why?] Solution. Let g(z) =
[ f i:
f(x
+ iy)dxdy =
f[
g(x
+ iy)dxdy,
iD(O,b)
54
and g(O) = f(zo), so we may assume that Zo = O. If 0 < r < band C; denotes the circle centered at the origin of radius r , Cauchy's formula implies f(O) = We parametrize C; by re'" with () f(O) 1 = -2' n
i
-1-.1
2m
c,
f(O
~
at;
lDr
0
= -21 n
lDr
0
f(reiO)d().
We can now multiply both sides of the above equality by r and integrate with respect to r from 0 to b. Interchanging the order of integration we obtain f(O)
Jo
t' rdr
= _1_
2n Jo Jo
(2TC
f(reiO)rdrd().
Evaluating the integral on the left and changing variables in the integral on the right (from polar to rectangular), we get after some simplifications f(O) = ~
it
J(
JD(O,b)
f(x
+ iy)dydx,
which is the desired formula. Exercise V. 1.8. Let D be the unit disc and let S be the unit square, that is, the set of complex numbers Z such that 0 < Re(z) < 1 and 0 < Im(z) < 1. Let f : D ~ S be an analytic isomorphism such that f(O) = (1 + i)/2. Let u, v be the real and imaginary parts of f respectively. Compute the integral
(au)2 ax
+ ay
(au)2
= 1)./,
J Is
dxdy
J Iv
so the desired integral is equal to the area of the unit square namely
Exercise V.l.9. (a) Let f be an analytic isomorphism on the unit disc D, and let f(z) = L:anzn
n=l
00
55
Prove that
00
areaf(D)
= n Lnlanl2.
n=1
(b) Suppose that f is an analytic isomorphism on the closed unit disc D, and that If(z)1 ~ 1 iflzl = 1, and f(O) = O. Prove that area feD) ~ n . Solution. (a) Applying the change of variable formula together with the CauchyRiemann equations we find area feD) =
fr
I
Jf(D)
dxdy
fr
JD
If'(z)12dxdy.
r = Jo Jo
thus
If'(reiO)12
= 0 hence
The desired formula drops out after evaluation of the last integral. (b) We are given that If(eiO)1 ~ 1 for all real B. So
2tc
If(ei(J)12dB
Since f(O) = 0, the constant term of the power series of fat r21C that Jo em'0 dB 0 for all nonzero integers n, we find that
21C iO 2 If(e )1 dB
1
00
21C ~
2 lanl dB
= 2n
2 lanl .
Therefore combined with the previous inequality we get result now follows from part (a)
00
L::I Ian 12
> 1. The
areaf(D)
= n Lnlanl2
n=1
~nL
n=1
lanl2 ~ n,
56
J JD
that
fez) = I>nzn.
n=O
Prove that _1
27r
f J[
=
If(z)IZdxdy
f:
n=O
Ian12 /(2n
+ 2).
Z
1"(
!vlf(Z)IZdXdY
1 l'
Z
1"(
If(reie)12rdrdB
1'1
If(reie)lzrdBdr.
so that
as was to be shown. For the next exercise, recall that a norm II . II on a space offunctions associates to each function f a real number > 0, satisfying the following conditions: Nt. We have IIfll = 0 if and only iff = o. N 2. If c is a complex number, then IIcfll = lei IIfll. N 3. II f + gil s II f II + IIgII· Exercise V.1.11. Let A be the closure of a bounded open set in the plane. Let f, g be continuous functions on A. Define their scalar product (f, g) =
fi
f(z)g(z)dydx
and define the associated L z-norm by its square, IIfll~ = Show that
fi
s
If(z)1 dydx.
fi
If(z)ldydx.
57
Show that f t-+ IIf III is a norm on the space of continuous functions on A, called the L '<norm. This is just preliminary. Prove: (a) Let 0 < s < R. Prove that there exist constants CI, C2 having the following property. If f is analytic on a closed disc D of radius R, then
g}
+ f3fj(g,
g).
Note that ex
= Ilgll~. Substituting
os
+ IIgll~(f,
g}(f, g}.
= l(f,
Ilgllilllll~.
Conclude the proof by considering both cases Ilg112 0 and Ilg112 O. =1= We now show that II. 112 a norm. Clearly, IIfl12 ::: 0 for all f. Suppose that f is is not identically zero. Then by continuity, If(z)1 > /) > 0 for all z in some ball, so 111112 O. The second condition is obvious, Finally for the triangle inequality > we use the Schwarz inequality. We have
g)
+ (g,
f)
+ (g,
g)
+ (g,
+ gll~ s
IIfll~
+ 21(f,
g}1 + IlglI~
The triangle inequality follows from taking square roots on both sides. Finally, II . III is a norm. The first two conditions are obvious, and the triangle inequality follows from the triangle inequality for the standard absolute value of complex numbers. (a) These two estimates show that the integral norms dominate the sup norm. To prove the first inequality, we proceed as follows. Fix a real number ro such that s < ro < R. There exists a constant /) > 0 such that [z - ~ I > /)for all z in the closed disc of radius s and all ~ on a circle of radius r with ro < r < R. In fact,
58
we may take 0 = ro - s. Then, applying Cauchy's formula we obtain fez) = _1_. { 2;rn
Jc,
fen I; -
dr.
Writing I; = re'", putting absolute values and using the above observation we get If(z)l:::: _1_
21<0
Jo
{21f If(reilJ)lrde.
Integrating both sides with respect to r from ro to R, and using polar coordinates we find
(R - ro)lf(z)1
:::: ( C
Jro<IZI<R
If
I :::: C ~
JD(O,R)
If
which completes the proof. The second inequality follows from applying the Schwarz inequality taking one function to be If I and the other to be 1. (b) Let K be a compact subset of U. For each point z in K, choose rz > 0 such that D(z, 2rz) cU. Then UZEK Drz, rz) is an open cover of K from which we can select a finite subcover Uf=, D(Zi,ri). By part (a) we have Ilfllr; :::: C;llfI12,2r; for some constant Clearly IlfI12,2r; :::: Ilfl12,u so Ilfllr; :::: C; IlfI12,u' If Un} is L2-Cauchy, then given E > 0 there exists N such that IIfn - fm 112,u < E for all n, m > N. So Ilfn - fm Ilr; < CiE for all i = 1, ... , P and therefore Ilfn - fm 11K< CE for some positive constant C (actually, C = L C; will do). This proves that Un} is uniformly Cauchy on K and therefore Un} converges uniformly on compact subsets of U.
c..
Exercise V.I.12. Let U, V be open discs centered at the origin. Let f = fez, w) be a continuous function on the product U xV, such that for each w the function Z t-+ fez, w) andfor each z the function w t-+ fez, w) are analytic on U and V, respectively. Show that f has a power series expansion fez, w) = LamnZmwn which converges absolutely and uniformly for [z I :::: r and I wi:::: r, for some positive number r. [Hint: Apply Cauchy's formula for derivatives twice, with respect to the two variables to get an estimate of the coefficients amn.} Generalize to several variables instead of two variables. Solution. Select R > 0 such that D(O, R) c U and D(O, R) C V. Fixing w, Cauchy's formula implies fez, w) = _1_. ( f(l;, w) dt; 2;rn JCR I; - z for z
E D(O, R).
59
for w
f(z,
w) =
(_1_)2 r r
2rri lcRlcR
Since f is continuous, we can apply Fubini's theorem to transform the iterated integral into an integral over C R x CR. We get f(z,
w)
(_1_)2 r
Zn i
This series is absolutely and uniformly convergent whenever Izl ~ rand Iwl Since f is bounded on C R x C R we can integrate term by term and we get
f(z, w) =
r.
m,n=O
L
00
amnZmwn
where
la I < _1_llfllcRxcR
mn 4rr2
Rn+m+2
4rr2 R2 =
IlflbxCR,
Rn+m
where II . II denotes the sup norm. The same argument shows that for n variables, we have the expansion
f(ZI, ... , Zn) =
L'"
;1
Lail'''inzill ill
...
z~
60
V.2
Laurent Series
Exercise V.2.I. Prove that the Laurent series can be differentiated term by term in the usual manner to give the derivative of f on the annulus. Solution. We use the notation of the section and of Theorem 2.1. Write f = f+ + r>. Look at f+(z) = L~Oakzk and let fn+(z) = L~=oakzk. Then fn+ ~ I" uniformly on a slightly smaller annulus, namely on the annulus s ~ Izl ~ s. So (fn+)' ~ (f+), uniformly on s ~ lz] ~ S and therefore we can differentiate term by term. The same argument shows that we can differentiate term by term to obtain
tr:».
r:
Exercise V.2.2. Let f be holomorphic on the annulus A, defined by 0< r ~ lzl ~ R. Prove that there exist functions fi, h such that fl is holomorphic for is holomorphicfor Izl :::: rand f=/I+h on the annulus. Solution. Write f(z) L~-oo anzn. Let fl (z) = L~o anzn and h(z) = n L~I a_nC • We show that fl is holomorphic for Izl ~ R. Since it is holomorphic in the annulus we must only consider the case when Izl ~ r. But then, if r < S < R, we have lanznl ~ lanlrn ~ lanlSn and by assumption the series Ln>O IanI S" converges so this proves that fl is holomorphic for lz] ~ R. The same argument shows that h is holomorphic for lel :::: r. Exercise V.2.3. Is there a polynomial P (z) such that P (z)e I/z is an entire function? Justify your answer. What is the Laurent expansion of el/z for [z] :f:. O? Solution. We know that eZ
Iz I ~ R, h
= L~
el/z
f:~(~)n
n=On.
Z
zn/ n !, so
If we have a polynomial P(z) = adzd + ... + ao, then the Laurent expansion of P(z)el/z near zero will have terms of the form z where CN :f:. 0 and N > O. To see this, note that the coefficient of l/ZN in the expansion of P(z)el/z is
cz
CN
ao = N!
+ (N+l)!
al
+ ... + (N+d)!
:f:. 0, then
we see that we can
ad
= (N
+ r)!
[ar+1 a; + N
+ r + 1 + ... + (N + r + 1) ...
ad]
(N
+ d)
61
and therefore this coefficient is nonzero for all large N. This proves that there is no polynomial P(z) such that P(z)el/z is entire. Exercise V.2.4. Expand the function fez)
1+
z
Z3
(a) in a series of positive powers of z, and (b) in a series of negative powers of z: In each case, specify the region in which the expansion is valid. Solution. (a) For
lel
< 1, we have
fez) =
so that fez) = L:o( _l)n Z3n+l. (b) For Izl > 1, we write fez) =
Y + .. -)
whence fez) = L:o(-It(1/z)3n+2. Exercise V.2.S. Give the Laurent expansions for the following functions: (a) zl(z + 2)for [z] > 2 (b) sin I/zfor z =1= 0 (c) cos I/zfor z =1= 0 (d) (z~3)for [z] > 3 Solution. (a) z~2 ,= I+~/Z = L:o(2/z)n. . ",,00 (_I)" 1 (b) sm II z = L..n=O(2n+I)!zz.+T' (c) COS11i-;;'L:;O~o ~;:~ ~L~:-(d) _1__ !_I__ ",,00 (_3)" z-3 - z 1-3/z - L..n=O Z"+I.
\
Exercise V.2.6. Prove the following expansions: (a) eZ = e + e L:I ~(z - l)" (b) liz = L:o(-It(z - l )" for Iz - 11 < 1 (c) I/~2 = 1 + L:I(n + I)(z + It for Iz + 11 < Solution. (a) By definition,
ez-I
1
1
L -(z n=On!
00
It = 1 + L -(z
00
n=1
n!
- l)".
However ez-1 eZe-l• (b) For [z - 11 < 1 we can use the formula for the geometric series,
+ (z -
1) -
_ ~(_I)n(
f:o
_ I)"
62
+ I)'
and therefore
00
so -liz
= L~o(Z 1
+ It
00
= 1 + L~I(Z
+ l)"
2" = I:>(z
Z n=1 which proves the formula.
+ l)n-1
= 1 + L(n n=1
+ 1)(z + l)n
Exercise V.2.7. Expand (a) cos z, (b) sin z in a power series about n12.
1 ..:__-_:_...:_-
(z - tt 12)2n+1 (2n
+ I)!
= cos(z
n 12), so
00 (z n12)2n "'( - l)n,- _-~_ ~ n=O (2n)!
. smz=
Exercise V.2.S. Let fez) = (Z-I;(Z-2). Find the Laurent series for f: (a) In the disc lz] < 1. (b) In the annulus 1 < Izi < 2. (c) In the region 2 < lzl.
-1
+ --.
z- 2
2 1- (zI2) = -2
1
1 00 (z)n
2~
00
(z)n
2.
-1 -1 1 -1 1 z- 1= 1 - (1I z) = ~ n=Ozn
jZ
f=
63
and _1 so
z- 2
= -1
f: (~)n
n=O
lzl. then
so
00
-1
(z - 1)(z - 2) = ~ zn+! Exercise V.2.9. Find the Laurent seriesfor(z+ (b) [z] > 1. .~ Solution. (a) Let fez) = (z
+ ~~
00
(2)n
~
in the region (a)
1)/(z-l) ---
Izl
< 1;
+ ...)
f in the region
which gives, after expansion, the desired Laurent expansion for [z] < 1. (b) For Izl > 1 we have fez) = ~ z
1 - (1/z)
1 (1 + ~) (1 + ~ + _!_ + ...) . z z
= Z2
Expanding the above expression gives the desired Laurent expression of [z] » 1. Exercise V.2.10. Find the Laurent series for l/z2(1 - z) in the regions: (a) 0 < lz] < 1; (b)lzl > 1. Solution. fez) (a) We write
f for
= --- 1 Z2
III
-(1
Z2
11
Z3 1 - (l/z)
-1 (1 + ~ + _!_
Z3 z
Z2
Exercise V.2.ll.
64
Solution. We have the factorization Z2 + 1 = (z + i)(z - i), so the singularities of f are at i and -i. Since 11 - i I = 11 + i I = .../i, the radius of convergence of the power series expansion of fat z = 1 is .../i. Using partial fractions we get the expression f(z) i =2
[I
1
-.
Z+l
- -.
Z-l
I]
Since we want the power series expansion at z = 1 we must make the following transformation,
f(z) so that
i[
+ 1 (1 + ~~:)-
1 1-
i(1 +
1]
~=:)
n
00 i (_l)n f(Z)=~-2-
(1
=
(i+l)n-(I-i)n
1)
(z-I).
Exercise V.2.12. Find the Laurent expansion of f(z) for 1 < 1 (z _ 1)2(z
+ 1)2
lel < 2.
Solution. Write
).
Conclude. Exercise V.2.13. Obtain the first four terms of the Laurent series expansion of f(z) validforO < eZ = Z(Z2 + 1)
Izl < 1.
z 1 - (-Z)2 z 1 z 52 = - + 1 - - - -z + .... z 26
= -I
1+ z
Z2 Z3) + - + - + ... 2! 3!
(l - Z2
+ Z4 + ...)
65
periodic
Exercise V.2.14. Assume that / is analytic in the upper half plane, and that / is of period 1. Show that / has an expansion of the form
/= where
L Cne27rinZ,
00 -00
c; =
11
/(x
+ iy)e-27rin(x+iY)dx,
f* on a disc
for any value of y > O. [Hint: Show that there is an analytic function from which the origin is deleted such that j*(e27riZ) = f(z). What is the Laurent series for f*? Abbreviate q = e27riz.J
Solution. Let Zo be a point in the unit disc minus the origin. Let z be another point on the unit disc minus the origin and let y be a path joining zo and z. Let gy(z) =
zc.v
di; i;
+ logzo
for some branch of the logarithm. For different paths y joining Zo and z and different branches of the logarithm, the function gy differs by integral multiples of 2n i, Since f is periodic we see that we can define an analytic function on the unit circle minus the origin by j*(q) = f (2~;<gy(q») .
The coefficients in the Laurent expansion of f* are given by _1 an - 2n i Changing variables 2n x coefficient is an =
_1 -2n
127r f*(reiO)
0
rneniO
de.
e and writing r
11
11
f*(e27ri(x+iY»e-27rin(x+iY)dx.
= L~ooanqn.
But q
+ iy)
11
f(x
+ iy)e-27rin(x+iY)dx.
00
Exercise V.2.IS. Assumptions being as in Exercise 14, suppose in addition that there exists Yo > 0 such that f(z) = f(x + iy) is bounded in the domain y ~ Yo. Prove that the coefficients Cn are equal to 0 for n < O. Is the converse true? Proof? Solution. Suppose that y > Yo and let B be a bound for we have the estimate Icnl S S
10 10
If(x
+ iy)1
1e- 1l"in(x+iY)Idx
2
Be21l"nYdx =
e>»,
If n < 0, letting y ~ 00 we see that Cn 0, as was to be shown. Now we show that the converse is true. Suppose that c; = 0 for all n < O. Then using the notation of Exercise 14 we have an = 0 for all n < O. Therefore J* is bounded in a neighborhood of the origin, that is there exists 0 < Band 0 < C < 1 such that for alllql < C we have 1J*(q)1 < B. But since Iql = le21l"izl = e-21l"Y and J*(e21l"iZ) = f(z) we conclude that If(z)1 < B whenever y > 2"; log C. This ends the exercise.
on
a ) L....m=O n!(n+z)
,",00
(_I)"
(b)
(c)
I::I I::I
z2~n2
+ I::O,n=-oo + ~] [z~n
Solution. (a) Let f(z) = I::o n~(,.~"z)'We contend that f is meromorphic on C with simple poles at the negative integers. Let R > 0, and select an integer N such that N > 2R. We write f(z)
The first sum exhibits the poles of order 1 at the negative integers of absolute value S N. The second sum defines a holomorphic function for lz] < R. Indeed, for lz] < Randn > Nwehave
and I: proved.
I n!(n
(_l)n
+ z)
I<
1 - n!(n -Izl)
1 <- n!R
(b) Let I(z) = L~! z2~n2 and let S = {... , -2i, -i, i, 2i, ... }. We contend that 1is meromorphic on C with simple poles at points of S. Let R > 0, N > 2R and write I(z) =
L
N
1
-2--2
n=! Z
+n
n=N+! Z
L
00
1
-2--2'
+n
Since Z2 +n2 = (z+in)(zin), the first sum defines a meromorphicfunction with simple poles ±in for n ::::N. If [z] < R the second sum defines a holomorphic function. Indeed, with lz] < Rand n > N we have the estimates
::r
N
I (z
+ n)2
+ n~! "
00
I (z
+ n) 2 •
The first sum exhibits the poles. If lz] < R the infinite sum defines a holomorphic function. Indeed, if lz] < Rand n > N we have
I (z ~ n)21 .s (n -\ZI)2
I (z ~
.s
(n ~ R)2'
and the denominator satisfies (n - R)2 = n2(1 - R/n)2 2: n2(l - 1/2)2 so n)21 ::::~.
Conclude. (d) Let I(z) = L~! n!;;~~~2)' We contend that f is a meromorphic function with simple poles at the points of S where S is defined as in (b). Note that the solutions of the equation sin nz = are {... , -~n, -"n, 0, i, so the set of zeros of sin nz and Z2 + n2 are disjoint for all n 2: 1. Let R > 0, N > 2R and write
2; , ... },
sin nz
sin nz +n2)
oo
n=N+1
n!(z2 +n )
2'
The first sum exhibits the poles. If Izl < R, we claim that the second sum defines a holomorphic function. Indeed, the power series expansion of the sin implies 1 sin nzl :::: n1zl :::: enR and the ratio test shows that enR In! converges (actually to e
68
0) as n
I n !(Z2 + n2)
sin nz
C ~
1 I(Z2 + n2)1
for all n :::: . The estimate given in (b) for the fraction 1/I(z2 + n2)1 concludes the 1 argument. This proves our contentions. (e) Let f(z) = ~+ I::O,n=-oo [z~n + ~].We contend that f is meromorphic on C with simple poles at the integers. First we have f(z) = ~ +
Z
i:
f:
The expression in the first parenthesis exhibits the poles, while the expression in the second parenthesis defines a holomorphic function for Izl < R. Indeed, the absolute value of the general term of the series is
I Z2 However,
2z
n2
I
~
2 R n2 1 - Iz/ n 12 •
_1:'n2 1
>
I- ~
N2 -
>
1- ~
whenever n > N, so we see that the infinite sum defines a holomorphic function for Izl < R. This proves our contention. Exercise V.3.2. Show that the function f(z) =
?;
00
Z2
n2z2
+8
is defined and continuous for the real values of z. Determine the region of the complex plane in which this function is analytic. Determine its poles. Solution. Let fn(z) = n2::+8' The for all real z the inequality n2z2 + 8 :::: n2z2 holds, thus 0 ~ fn(z) ~ 1/n2. Now each partial sum I:~=! fk is continuous, and the inequalities show that the sequence of partial sums converges uniformly to f. Hence f is well defined for real z and f is continuous. Let S denote the set of zeros of n2z2 + 8 union the origin, that is S = {i.J8/n :n
E
Z} U {OJ.
We contend that f is analytic on the complement of S. Let 8 > 0 be small, and let R > 0 be so large that S is contained in the disc of radius R centered at the origin.
69
Suppose Izl < Rand lz] > 8, so that we stay away from the origin. Choose N so large that 8/ N2 < 82/2. Write
The first sum exhibits the poles. The infinite sum defines a holomorphic function on the region lzl < Rand Izl > 8. Indeed, for n > N we have
8 (82 = n2
:2 ) ~ n2 (82 _
I
8;)
I
n2
Since the series L 1/ converges, the infinite sum defines a holomorphic function in the region Izl < Rand [z] > 8. The above results being true for all large R, and all small 8, our contention is proved. Exercise V.3.3. Show that the series
00
L (z+~)n
n=l
Z-l
defines an analytic function on a disc of radius 1 centered at -i. Solution. Let 0 < of radius s. Then z
s < 1. Suppose that z belongs to the open disc centered at -i + i ::::: sand Iz - i I ~ I - 2i I - Iz + i I ~ 2 - s.
Therefore
I z-i
z-- i I +
- 2-s·
<--s
Since 0 < s < 1 we have 0 < s/(2 - s) < 1, and the series
L(2~S)n
n2:i
L (z+~)n
n2:1 Zl
70
Exercise V.3.4. Let {Zn} be a sequence of distinct complex numbers such that 1 L IZnl Prove that the series
--3
converges.
OO( L
n=1
(z -
Z )2 n
Z2 n
1)
defines a meromorphic function on C. Where are the poles of this function? Solution. Let fez) = L::1 (z-~n)2 and let R > 0. We now show that f defines a meromorphic function on the open disc D(O, R) or radius R centered at the origin. The hypothesis that L:1/lznl3 converges implies that IZnl ~ 00 as n ~ 00. So there are only finitely many n such that Zn E D(O, R). Select N so large that for all n > N we have IZn 1 > 2R. Each Zn E D(O, R) is a pole of order 2. Write fez) =
*),
n=1
i:(
(Z -
Zn)
2-
~)
Zn
n=N+1
f: (
(Z -
Zn)
2-
~)
zn
The first sum exhibits the poles in D(O, R), so it is sufficient to show that the second sum defines a holomorphic function on D(O, R). We have the estimates 1
I (z -
Zn)2 -
= IZn 13 1
1 1-~+2zl 1t _ 112
":;:'!!....---;;--;-
<--
- IZnl3
where B is some large positive constant. By hypothesis this completes the proof.
L:1/IZn 13 converges,
so
Exercise V.3.S. Let f be meromorphic on C but not entire. Let g(z) = ef(z). Show that g is not meromorphic on C. Solution. Since f is not entire it has at least one pole, say at zoo In a neighborhood of zo we can write
(Z - zo)mfez)
= p(z)
p(z)
Zo
+ (Z )m
zo)mh(z)
=(
Z-
+ h(z),
71
But
eh(z)
ef(z)
is not meromorphic
Exercise V.3.6. Let f be a nonconstant entire function, i.e., a function analytic on all of C. Show that the image of f is dense in C. Solution 1. In the spirit of this chapter we use the Casoratti- Weierstrass theorem. Suppose f is an entire function whose image is not dense in C. Then there exists a complex number a and a positive number s such that If(z)-al > s for all complex z. Write f(z) = L~o anzn and suppose that there are infinitely many nonzero terms in this expansion. Then, for all z =I- 0 let g(z) = f(l/z). We see that g has an essential singularity at 0 so by the Casoratti-Weierstrass theorem for some z near 0 we have Ig(z) -al < s, hence If(1/z) -al < s. This contradiction implies that the power series expansion of f can have only finitely many terms. Then the fundamental theorem of algebra guarantees that f is constant. This contradicts the hypothesis. Solution 2. Suppose there exists a complex number a and a positive number s such that If(z) - al > s, for all complex z. Then the function g(z) = 1/(f(z) - a) is entire and bounded, so by Liouville's theorem, g is constant. Hence f is constant, again contradicting the hypothesis. Exercise V.3.7. Let f be meromorphic on an open set U. Let
be an analytic isomorphism. Suppose that g1(zo) = wo, and f has order n at Woo Show that f 0 g1 has order n at Zoo In other words, the order is invariant under analytic isomorphisms. [Here n is a positive or negative integer.] Solution. We can write
00
f(w)
= Lam(w
m=n
- wo)m
for all W near WooAlso we have g1(z) = Wo + bi (z - zo) + b2(z - ZO)2 + ... for all z near ZOo We assume that g1 is an analytic isomorphism so bi =I- O. Put W = g1(z) in the expression of f at WOo Since bi =I- 0 we see that the composite f 0 g1 also has order n at Zo. Exercise V.3.S. A meromorphic function f is said to be periodic with period W if f(z + w) = f(z)for all z E C. Let f be a meromorphicfunction, and suppose f is periodic with three periods WI, WZ, W3 which are linearly independent over the rational numbers. Prove that f is constant. [Hint: Prove that there exist elements W which are integral linear combinations of WI, W2, W3 and arbitrarily small in absolute value.] The exponential function is an example of a singly periodic function. Examples of doubly periodic functions will be given in Chapter XlV.
72
Solution. Following the hint, we prove that there exists integral linear combinations of WI, W2 and W3 which are arbitrarily close to 0 (and not 0 since the three periods are linearly independent over the rational numbers.) Let WI = WI/W3 and W2 = W2/W3. It suffices to show that for every positive integer N, we can find integers m I, m2 and n such that
Iml WI
+ m2w2
nl s c] N,
We use a modification of the pigeon hole principle (see figure on page 73). Fix N, and choose a positive integer B such that: 11m(wi)1 < B for i = 1,2.
Finally, choose an integer M such that M > 4 B N2. Consider the box: R = {z
E
11m(z)] S 2BM}.
For each pair of integers (m I, m2), let {m IWI + m2w2} denote the representative of m I WI + mz ii12mod Z with real part between 0 and 1. If 0 < m I, mz S M, then Os Re ({ml WI 1m ({ml WI so {m I WI
+ m2w2})
=1=
< 1 and
+ m2w2})
= mllm
(w))
+ m21m
(W2).
+ m2w2}
+ m2w2}
=1=
{m; WI
+ m;w2}
because WI, W2 and W3 are linearly independent over the rationals. For 0 < m I, mz S M, we get M2 distinct points in R. But we can write R as a union of squares with side I/N. There are 4BMN2 such squares in R. By assumption M2 > 4BMN2, so two points {m;wl + m;w2} and {m'[wl + m~w2} belong to the same small square. This proves that there exists integers m I, mz and n such that Imlwl + m2w2 - nl S -J2/N. This proves that there exists integral linear combinations of WI, W2 and W3 which are arbitrarily close to O. Let z., be a point where f is holomorphic. Then f(zo+w) = f(Zo)forarbitrarily small W =1= 0, so f is constant in a neighborhood of Zoo Since the set of poles of f is discrete with no points of accumulation, we conclude that f is constant outside of the set of poles. By the theorem on removable singularities we conclude that f is constant on C. Exercise V.3.9. Let f be meromorphic on C, and suppose lim If(z)1 = Izl-+oo Prove that f is a rational function. a finite number of poles.)
00.
Solution. Since limlzl-+oo If(z)1 = 00, the quantity If(z)1 is well defined for all large lzl. This implies that the poles of f are contained in some disc of large radius. Hence f can only have a finite number of poles. Let P be a polynomial
73
'-8M
~R
·
•
. ,
...
i '
1'1 0
['YtI
·
~
,.
with zeros where f has poles and such that fl (z) = f(z)P(z) is entire. Let g(z) = fl(l/z) defined for z 1= o. If fl is not a polynomial, then g has an essential singularity at O. However, Ig(z)1 -+ 00 as lz] -+ 0, which contradicts the theorem of Casoratti- Weierstrass. So fl must be a polynomial, and we are done. Exercise V.3.10 (The Riemann Sphere). Let S be the union ofC and a single point denoted by 00, and called infinity. Let f be afunction on S. Let t = liz, and define get) = f(llt) fort 1= 0, 00. We say that f has an isolated singularity (resp. ismeromorphic resp. is holomorphic) at infinity if g has an isolated singularity (resp. is meromorphic resp. is holomorphic) at O. The order of g at 0 will be called the order order of f at infinity. If g has a removable singularity at 0, and so can be defined as a holomorphic function in a neighborhood of 0, then we say that f is holomorphic at infinity. We say that f is meromorphic on S, if f is meromorphic on C and is also meromorphic at infinity. We say that f is holomorphic on S if f is holomorphic on C and is also holomorphic on at infinity. Prove: The only meromorphic functions on S are the rational functions, that is, the quotients of polynomials. The only holomorphic function on S are the constants. If f is holomorphic on C and has a pole at infinity, then f is a polynomial. In this last case, how would you describe the order of f at infinity in terms of the polynomial?
74
Solution. We prove the last two assertions first. Suppose I is holomorphic on S. Then g has a removable singularity at 0, which implies that in a neighborhood of 0, the function g is bounded. This implies that I is bounded outside some disc and therefore I is bounded on C. By Liouville's theorem we conclude that I is constant. Now suppose I is holomorphic on C, and that I has a pole at infinity. The function I has a certain power series expansion I(z) = anzn, and therefore
g(t) = ~an
00
(l)n t
L~o
Since g has a pole at 0, we conclude that the power series of I at has only finitely many nonzero terms, thus I is a polynomial and deg I = - ord., g. Finally, suppose that I is a meromorphic function on S. Since S is compact, I can have only finitely many poles in C, say {(Zi, ni)}, i = 1, ... , m where n, is the order of the pole Zi. Then cp = I(z)
D(z M
i=1
Zit;
is holomorphic on C, and cp is either holomorphic on S or has a pole at infinity. In all cases, we conclude that cp is a polynomial, hence I is a rational function as was to be shown. Exercise V.3.11. Let I be a meromorphic function on the Riemann sphere, so a rational function by Exercise 8. Prove that
Lordp
p
1= 0,
00.
where the sum is taken over all points P which are either points of C, or P =
Solution. In this exercise, all sums and products are finite. By the previous exercise, we know that I is a rational function. We write
I,
ord»
1= Lmj
j
- Lni.
i
I at infinity.
=K
We have
fl·t-mj(1-tWj)mj
-::::J:--_-:--_-:--_
fli t-ni(l
tZi)nj
75
Lj
i
mj
+ Li
- Lmj
j
ru . Hence
= Lmj
j
- Lni
+ Lni
i
=0
as was to be shown. Exercise V.3.12. Let Pi(i = 1, ... , r) be points of C or such that
r 00,
Lmi=O.
i=1
i = 1, ... , r
if Pi-Pi.
Solution. We may assume without loss of generality that the points Pi are pairwise disjoint and that m, i- 0 for all i, Suppose first that Pi i- 00 for all i, Let zt be the complex number which determines Pi. Then
r
f(z) = n(z
i=1
- Zi)m;
is meromorphic on S and satisfies the desired property. Indeed, if P = Pi, then ord» f = m., if Pi-Pi for all i, and P E C, then ord» f = O.Finally, if P = 00 the previous exercise implies that ord » f = - L m, = O. Now suppose without loss of generality that PI = 00. Let
r
f(z) = n(Z-Zi)m;.
i=2
Then if i is not equal to 1 and P = Pi, we have ord» f = mi. If Pi-Pi and P E C, then ord» f o. Finally, if P = PI we have
for all i,
=-
Lmi
i#1
= mI·
VI
Calculus of Residues
VI. I
Exercise VI.1.t. Find the residue of the following function at 0: (Z2 Solution. resz=o f = 1. Exercise VI.1.2. Find the residue ofthefollowingfunction Solution. resz=o
at 0:
(Z2
= 1.
+ 2).
Solution. resz=o
= O.
+ 1)lz(z3
- 5).
Solution. resz=of = -liS. Exercise VI.1.S. Find the residue of the following function at 0: (sin zj/z". Solution. resz=o
f f
= -1/3!.
Exercise VI.t.6. Find the residue of the following function at 0: (sin z)1 Z5. Solution. resz=o
= O.
Exercise VI.1.7. Find the residue of the following function at 0: (sin zj/z''.
77
Solution. resz=o f = 1/5!. Exercise VI.1.S. Find the residue of the following function at 0: (sin Z)/Z7. Solution. resz=o f
= O.
= 1.
Exercise VI.1.9. Find the residue of the following function at 0: eZ / z. Solution. res.so
Exercise VI.1.10. Find the residue of the following function at 0: eZ / Z2. Solution. resz=o f = 1. Exercise VI.l.11. Find the residue of the following function at 0: eZ / Z3.
Solution. resz=o f = 1/2!. Exercise VI.1.12. Find the residue of the following function at 0: eZ / Z4. Solution. resz=o f = 1/3!. Exercise VI.l.13. Find the residue of the following function at 0: z-210g(l
+ z).
Solution. resz=o f = 1. Exercise VI.l.14. Find the residue of the following function at 0: eZ / sin z.
Solution. resz=o f = 1. Exercise VI.1.1S. Find the residue of the followingfunction Solution. reSz=1 f = 1/6. Exercise VI.1.16. Find the residue of the following function at 1: (Z3 - 1)(z 2)/(Z4 - 1)2. . Solution. reSz=1 f = 32/42, because (Z3 - l)(z + 2) (Z4 - 1)2 (z - 1)(1 + z + Z2)(Z + 2) (z - 1)2(1 + z + Z2 + Z3)2' at 1: 1/(Z2 -l)(z
+2).
+
Exercise VI.l.17. Factor the polynomial residue at 1 ofl/(zn -1). Solution. If we let () = 27T / n, then
zn - 1 = (z - 1)(z - eili)(z - ei21i)... (z _ ei(n-l)Ii). To compute the residue we can differentiate that
78
Exercise VI.I.IS. Let z" ... , z; be distinct complex numbers. Let C be a circle around z, such that C and its interior do not contain Zj for j > 1. Let
f(z) Find
= (Z -
l
Since
f~Z)dZ.
Solution.
if j = 1 if j > 1 the residue formula implies that the desired integral is equal to 2rri resZ=ZI 1/ f). ( Hence
l
Solution.
In
c f(z)
--dz
= 2rri
j=2 Zl -
n---.
1
Zj
Jc (Z4
We have the factorization resz=i We could also differentiate we conclude that
Z4 Z4 -
1
-
+ 1)(z
- i)(z
+ i),
so
I
Z4 _
4i'
-1
Z
1 and evaluate at
(Z4 ~
I/Z
= Zn i
1
(~il)
= -2 .
rr
Jc
2 Z -
3z
+5
dz,
4r-----
~----~C--~
10
79
3z
+ 5 are
.
3-iv'IT 2
The only singularity of 1/ p(z) in the rectangle is at z" and to compute the residue we write p(z) (z - z,)(z - Z2). Therefore, by the residue formula (and being careful about the orientation of C) we find that
1
(b) Let p(z) = Z2
1 --dz e p(z)
-~i
Z2
-~
vii
171.
+ z + 1. Then,
z, =
-1+i~ 2
-1-i~
Clearly, none of the roots of p lie in the interior of C, so 1/ p is holomorphic in this region and therefore
Je
(c) Let p(z) = Z2
-
r _l_dz p(z)
and
= O.
+ 1. Then
z, =
l+i~ 2
1-i~ 2
The point z, is the only singularity of 1/ p in the interior of C so writing p(z) = (z - z,)(z - Z2), the residue formula gives
Je
1 p(z)dz
-2;ri
=
-2;r
z, - Z2 = ~.
(b) Let P(z) be a polynomial of degree :s n - 1, and let a" ... , an be distinct complex numbers. Assume that there is a partial fraction decomposition of the form
Prove that
-an)
80
1
(Z - z.)· .. (Z -
z-)
= --+"'+--.Z Z - z,
a,
an Zn
These two expressions for f allow us to compute the residue at Zj in two different ways. The first formula gives
resz=Zj fez) =
k#j Zj - Zk
n---,
1
aj=n--·
k#j Zj - Zk
(b) Arguing in the exact same way as in (a) we find that Cj = P(aj) . ki'j (aj - ak)
Exercise VI. 1.22. Let f be analytic on an open disc centered at a point zo. except at the point itself where f has a simple pole with residue equal to an integer n. Show that there is an analytic function g on the disc such that f = g' I g, and g(z) = (z - zo)" h(z) where h is analytic and h(zo)
=f. O.
(To make life simpler, you may assume zo = 0.) Solution. Assume zo = 0, and let D* = D - {O} denote the punctured disc centered at the origin. Let W E D* and define gy(z) = i~y f(nd~ where the integral is taken on the path y C D* that joins w to z. Suppose we are given two paths y, and Y2 from w to z. Then the residue formula applied to the closed path y, Y2-l combined with the fact that f has integer residue at 0 implies that gYI (z) and gn (z) differ by an additive integer multiple of 21l" i, Hence g(z) = exp(gy(z)) is well defined (independent of the path y) on D*. Now we show that we can extend g to be analytic at O. It suffices to show that g is bounded near the origin, so that 0 becomes a removable singularity. Suppose z is close to 0 and let y be a path joining w and z. If n fez) = -
+ ao + alZ + ...
f(nd~
w,y
J',
0
81
so g(z) = znh(z) where h is analytic on D. Thus g is bounded near the origin and we can extend g to be holomorphic on the whole disc. Therefore f = g' / g on D* and since f has a simple pole with residue n at 0 we conclude that g has order n at 0, namely there exists an analytic function h such that g(z) = z" h(z) and h(O) =f. O. Exercise VI.1.23. Let f be afunction which is analytic on the upper half plane, and on the real line. Assume that there exist numbers B > 0 and c > 0 such that
If(~)1~ ~
for all
r Prove thatforany
_1_.1
2Jrl
00
-00
f(t) dt. t- z
[Hint: Consider the integral over the path shown on the figure, and take the limit as R ~ oo.]
The path consists of a segment from - R to R on the real axis, and the semicircle as shown. (b) By using a path similar to the previous one, but slightly raised over the real axis, and taking a limit, prove that the formula is still true if instead of assuming that f is analytic on the real line, one merely assumes that f is continuous on the line, but otherwise satisfies the same hypotheses as before.
SR
Solution. Let z be a point in the upper half plane. Choose R so large that z belongs to the interior of SR. Then by Cauchy's integral formula we have f(z) = _1_.
2m
f(~)
This formula holds for all large R. Let can write 2Jrif(z) =
SR ~ -
dr.
J[-R,R)
f(l;) dt;
~-
fst
f(~) d~. t; - z
82
It suffices to show that the last integral goes to 0 as R have the estimate
II II
st
t; -
f(~)
dt;
I~- z]
zd~
st ~ -
f(S)
II II
<
::::
By hypothesis we
- st ~ - z
2: RI2 so 2B
I f(S) I d~
- st I~n~ 2B JrR
<
zl
.u.
s; ~d~::::
RRc
Rc'
---+ 00.
2BJr
which proves that the integral over the upper semi circle goes to 0 as R Hence 2Jrl'f() Z = I' 1m
Rr+oo
1
00 -00
f(~) --
[-R,R]
t; -
dr~,
Z
and therefore
f(z)
_1_.1
2Jrl
f(t) dt. t- z
-III.
, 1ft.
+ i In
to -n +iln.
2 m'f() Z = I' nn
n-'>oo
f(S) --~,
L; ~ -
and
f(~)
Ln~-Z
JR
dt = ( fn(t)dt
JR
n] and
+ iln) + l'I n - Z .
83
and the following estimates show that the fn's are uniformly bounded by an integrable function Ifn(t)I:::: If(t.+ i/n)1 It+l/n-zl and It :::: B It+i/nnt+i/n-zl It I
+ i/nl
so that for all large It I we get Ifn(t)1 ::::Itlc+1 for some positive constant K. We can apply the dominated convergence theorem to obtain I f(z) = -. 27rl
foo
-00
f(t) -dt. t-
Exercise VI.l.24. Determine the poles and find the residues of the following functions. (a) 1/ sin z (b) 1/(1 - eZ) (c) z/(l - cos z).
Solution. (a) The function 1/ sin z has poles at the points and the derivative of sin z, is cos z so resz=krr -.= (-1) . sin z (b) The poles of 1/(1 - eZ) are the points 27rik where k 1 - eZ is =e" so resz=2rrik-= -1. 1 - eZ (c) The solutions of 1 - cos z = 0 are the complex numbers and z/( 1 - cos z) has poles at these points. Moreover
1
E
z = ktt where k
Z,
Z. The derivative of
z = 27rk with k
Z,
z
1 - cos z = z/2!
1 - z3/4!
+ ... '
if k = 0, if k
so resz=2rrk
1- cos z
12 0
# O.
84
Solution.
Let
I(z)
= cos
1 cos:Izl=1
dz,
Exercise VI.l.26. Find the integrals, where C is the circle of radius 8 centered at
the origin.
(a) si~zdz (c) r ..!.±Ldz (e)
Ie
(b)
Ie
Je I-e'
I~~~zdz
Solution. (a) The function 1/ sin z has poles at the numbers n k where k residue at 7rk is ( _I)k (see Exercise 24), so
Z. The
1
1 - cos z 2
e sm z
-.-I-dz = 27ri.
(b) The zeros of I - cos z are located at the points z = 27rk with k E Z. We first compute the residue at O. To do so, we use the power series expansion of the cosine:
and we see that the residue of 1/(1 - cos z) at the origin is O. By periodicity we conclude that the residues of 1/(1 - cos z) at the points 27rk are all 0 and therefore
1
I
e 1 - cos z
d:
= O.
E
Z.
I = I,
resz=-2"i
= -(I
- 27ri)
and
reSz=2"i
= -(I + 27ri).
(d) Let I = -
Ie tan zdz:
1
Then
e 1- e
1 +z d :
Z
= =tsn i.
85
2rri
L (number
of zeros of
f in
C).
The zeros of f are at the points kn /2 with k odd. Therefore, interior of C which gives
tanzdz = -12rri.
(e) Let fez) (1 + z)/(I - sin z). The poles of f are precisely at the points z = kit /2 where k is an odd integer. Since sin z = cos(z - rr/2) we see that fez) =
1+ z I - cos(z - n /2)
Arguing like in (b) we see that the residue of and therefore ( I +.z }cl-smz
dz
= o.
Let f be holomorphic on and inside the unit circle, Izl :s I, except for a pole of order 1 at a point zo on the circle. Let f = L anzn be the power series for f on the open disc. Prove that
Exercise VI.l.27.
=zo.
+ higher
c
terms.
z - Zo
= Lanzn + ~ L zo
(_:_)n = L
zo
(an
n:!) Zo
r.
~ 0 and
The power series of g has radius of convergence> therefore an/an+! ~ Zoo zea-z has a single solution with lz]
1 so an
+ c/z;;+!
:s I, which
Solution. Let fez) = zea-z and g(z) fez) - 1. Then If(z) - g(z)1 = 1 and if Izl = 1 with z = x + iy ; x, y E R, then
If(z)1 = ea-x•
ISO
But since a > l we have l j'(z)] > l whenever Iz] = landthereforel/(z)-g(z)1 < II(z) Ion the circle. By Rouche's theorem we conclude that the equations zea-z = 1 and zea-z = 0 have the same number of solutions in the closed unit disc. Since the second equation has only one solution, we conclude that ze":" = 1 has only one solution in the closed unit disc. Let I(x) xe":", Then 1(0) 0 and 1(1) > I, so the solution of ze":" I is real and positive.
Exercise VI.1.29. Let U be a connected open set, and let D be an open disc whose closure is contained in U. Let I be analytic on U and not constant. Assume that the absolute value III is constant on the boundary 01 D. Prove that I has at least one zero in D. [Hint: Consider g(z) I(z) - I(zo) with zo E Ir.]
Solution. Fix any Zo E D and let g(z) = I(z) - I(zo). Clearly, we have I/(z) - g(z)1 = I/(zo)l. The maximum modulus principle applied to I in D, combined with the fact that II I is constant on the boundary of D and that I is not constant implies I/(zo)1 < I/(z)1 for all z on the boundary of D. By Rouche's theorem, I and g have the same number of zeros in D. Clearly, g has at least one zero, namely zo. so I has at least one zero in D. This concludes the proof. Exercise VI.1.30. Let I be a function analytic inside and on the unit circle. Suppose that I/(z) - z] < [z] on the unit circle. (a) Show that 1/'(1/2)1 ::s 8. (b) Show that I has precisely one zero inside the unit circle. Solution. (a) By Cauchy's integral formula
1'(1/2) -
-1-1 Zn i
c (~ - 1/2)2 ~
E
I(~)
ar
C and
::s
= 8.
one zero inside the unit
I has precisely
01 the polynomial
Z+ I
36z57
+ 71 Z4 + Z3 -
87
(b) of radius 2, centered at the origin. (c) Determine the number of zeros of the polynomial 2z5 in the annulus 1 :::: lz] :::: 2. Solution. (a) Let g(z) be the polynomial given in the exercise, and let f(x) If [z] = 1, the triangle inequality implies Ig(z) - f(z)1 ::::1 + 36 = 71z4•
-
6z2
+z + 1=0
+ 1 + 1 + 1 < 71,
so on the unit circle we have Ig(z) - f(z)1 < If(z)l. By Rouche's theorem we conclude that g(z) = 0 has 4 zeros inside the unit circle. (b) Let g(z) be the polynomial given in the exercise and let f(z) Z87. Then if lz] = 2, we have
+ 71
24
+ 23 + 2 + 1 ::::287
= If(z)l,
so by Rouche's theorem we conclude that g(z) = 0 has 87 zeros inside the circle of radius 2 centered at O. (c) Let g(z) = 2z5 -6z2 +z+ 1, fl(z) = -6z2 and h(z) = 2z5• We denote by Dr the open disc of radius r centered at the origin. Rouche's theorem applied to g and [: shows that g has 5 zeros in D2 and no zero on the boundary of D2. Applying Rouche's theorem to g and fl we find that g has two zeros in DI and no zeros on the boundary of DI. Therefore, g has 3 zeros in the annulus 1 :::: Izl :::: 2. Exercise VI.l.32. Let f, h be analytic on the closed unit disc of radius R, and assume that f(z) =I- 0 for z on the circle of radius R. Prove that there exists E > 0 such that f(z) and f(z) + Eh(z) have the same number of zeros inside the circle or radius R. Loosely speaking, we may say that f and a small perturbation of f have the same number of zeros inside the circle. Solution. Let
E
- Eh(z). Then
Ig,(z) - f(z)1 :::: Ih(z)l. E There exists 8 > 0 such that If(z)1 > 8 for all z on the boundary of C (the circle of radius R) because If I is continuous on this circle and never 0 by hypothesis. Since h is continuous on the same circle, there exist EO > 0 so small that Eolh(z)1 < 8 whenever z E C. By construction, we have Ig,o(z) - f(z)1 < If(z)1 for all z
E
C.
Rouche's theorem guarantees that EOverifies the desired property. Exercise VI.1.33. Let f(z) anzn + ... + ao be a polynomial with an =I- O. Use Rouche's theorem to show that f(z) and anZn have the same number of zeros in a disc of radius R for R sufficiently large. Solution. Select Ro so large that
lan-II Ro
laol