Solutions To UMD Analysis Quals

Solutions to UMD Analysis Quals
Sean Kelly
Contents
Chapter 1. Solutions to Quals 5
1. Solutions for August 1999 5
3. Solutions for January 2001 10
3
CHAPTER 1
Solutions to Quals
1. Solutions for August 1999
3. Solution.
1
(a) No.
2
Suppose such a cover exists. Since
m < , its tails go to 0. So for all > 0,

there is an N such that
n=N
m(In) < .
Since each x is in innitely many In, |In
n=N
still covers E. By monotonicity,
m(E) m
_

_
n=N
In
_
n=N
m(In) < .
Since is arbitrary, m(E) = 0, contradicting the fact that E has positive measure.
(b) Yes. Since E has measure 0, for every n there is a countable collection of disjoint
intervals |I
n,k
k=1
such that E
k=1
I
n,k
and
m
_

_
k=1
I
n,k
_
=
k=1
m(I
n,k
) <
1
2
n
.
Then the countable collection of intervals |I
n,k
n,k=1
satises the condition that every
x E is contained in nitely many intervals, and by subadditivity,
n,k=1
m(I
n,k
) =
n=1
k=1
m(I
n,k
) <
n=1
1
2
n
= 1.
4. Solution. (a) We have the fact that if t = limn [an[
1/n
, then the power series
Since f, g are analytic on the disk, their radii of convergence are 1. So
lim
n
[an[
1/n
= t 1 and, lim
n
[bn[
1/n
= s 1.
The H-product f g has coecients anbn satisfying
lim
n
[anbn[
1/n
= (lim
n
[an[
1/n
)(lim
n
[bn[
1/n
) = ts 1
by basic properties of lim. So f g has radius of convergence 1, so is analytic on [z[ 1.
1
Remark: this problem is really just the Borel Cantelli lemma: if
m(In) < , the set of x R

contained in innitely many In has measure zero. Since that would be a one sentence proof, you should
prove it directly.
2
pf from Xuwen
5
6 1. SOLUTIONS TO QUALS
(b) By uniqueness of power series, since (1 z)
1
has expansion
(1 z)
1
=
z
n
it follows that [anbn[ = 1. By assumption, both g A, so limn [bn[
1/n
= s 1. So for all
but nitely many n, [bn[
1/n
1 [an[
1/n
1. Thus limn [an[
1/n
= t 1, so the radius
of convergence of f is exactly 1.
If f had analytic continuation to every point of [z[ = 1, it would be analytic on some
neighborhood of every such z, and therefore on an neighborhood of the unit circle. Thus
its radius of convergence would be strictly greater than 1.
3
7. Solution. Since L
2
[0, 1] is a Hilbert space, the parallelogram law holds. Namely for
all fn,
|fn fm|
2
2
= 2|fn|
2
2
+ 2|fn|
2
2
|fn +fm|
2
2
Taking lim of both sides,
4
lim
n,m
|fn fm|
2
2
= 4 lim
n
|fn|
2
2
lim
n,m
|fn +fm|
2
2
.
Since |fn|2 1 for all n, limn |fn|
2
2
1. So
lim
n,m
|fn fm|
2
2
4 lim
n,m
|fn +fm|
2
2
= 4 lim
n,m
|fn +fm|
2
2
= 4 2 2 = 0.
Since |fn fm|
2
2
is a positive sequence, it follows that its limit exists and is 0, so |fn is
a Cauchy sequence. Since L
2
[0, 1] is complete, |fn converges in L
2
[0, 1].
1. Solution. Remark: This is a thoroughly dumb problem (a) Obvious. Since the En
are disjoint, for x [0, 1] the function n(x) takes the value 1 at most once, and 0 the
rest of the time.
(b) Compute
|fn|2 =
_
_
1
0
_
1
mEn
n
_
2
=
_
1
(
mEn)
2
_
En
1 =
_
1
(
mEn)
2
mEn = 1.
The sequence |fn|2 is thus constantly 1 so converges to 1.
(c) Fix g L
2
[0, 1]. Since we are on a set of nite measure, g L
1
[0, 1]. By monotonicity,
mEn m[0, 1] = 1, so for all x [0, 1]
[fn(x)g(x)[ [1/
mEnn(x)[ [g(x)[ [g(x)[.

3
Alternate proof: Actually, a much stronger statement can be proved. If there exists g A such
that the H-product f g has radius of convergence 1, as the geometric does, then neither f nor g can
be analytically continued to every point on the circle |z| = 1. Suppose WLOG f can be, so as in proof
1, 1/t > 1. Since g A, 1/s 1. But by equation a, the H-product has radius of convergence
1
ts
> 1,
contradicting the assumption that its radius of convergence is exactly 1.
4
Remark: the proof will not work without using lim; there is no reason why limfn2 should exist
2. AUGUST 2000 7
And therefore we are justied applying the LDCT
lim
n
_
1
0
fng =
_
1
0
( lim
n
fn)g =
_
1
0
0 g = 0.
2. Solution. If [z[ > 1, [1/z
2
[ < 1, so z 1 + 1/z
2
maps the set |[z[ > 1 into
|z [ [z 1[ < 1, which is within the region where the principle branch of

is dened:
Picture here
So no funny business with dogbones.
Notice that
_
1 + 1/z
2
is analytic outside of the unit disk, so we should ip it:
Picture here
Let
w = 1/z, so thatdw =
dz
z
2
.
Under this transformation, the curve (1/z) C, a curve contained in the unit disk winding
once CW around 0. Denote by C
this new curve in the CCW orientation, that is,

C
(t) = 1/C(t). Then we get

_
C
z
_
1 +
1
z
dz =
_
1/C
1
w
_
1 +w
2
_
dw
w
2
_
=
_
C
1 +w
2
w
3
dw.
Then
1 +w
2
is analytic and 1/w
3
is analytic everywhere being considered except for a
pole at w = 0. We compute
d
2
dw
2
_
1 +w
2
0
=
d
dw
1/2 2w
1 +w
2
0
=
1 +w
2
w
2
/
1 +w
2
1 +w
2
0
= 1
and thus we have a Laurent expansion
1 +w
2
w
3
= +
1
2!

1
z
+ ,
so by the residue theorem
_
C
z
_
1 +
1
z
dz =
_
C
1 +w
2
w
3
dw = 2i
1
2
= i.
Remark: In my notes I have the answer as i, but typing it up write now I convinced
myself that its i. So either present me or past me is making an error.
3. Solution. (a) The Cantor-Lebesgue function is an increasing function that
satises (0) = 0, (1) = 1, and
(x) = 0 for almost all x [0, 1]. Since it is increasing,

it has bounded variation. But
_
1
0
= 0 ,= 1 = (1) (0).
So cannot be absolutely continuous since it would have to satisfy the FTC on [0, 1],
which it does not.
(b)
5
Let > 0. Since g
/2
is absolutely continuous, there exists such that for all disjoint
collections of intervals |[an, bn],
n
[bn an[ <
n
[g
/2
(bn) g
/2
(an)[ < /2.
Further by denition of T
1
0
, for all collections of disjoint subintervals of [0, 1],
n
[(g
/2
f)(bn) (g
/2
f)(an)[ < /2
so by the triangle inequality,
n
[bn an[ <
n
[f(bn) f(an)[
n
[f(bn) f(an) (g
/2
(bn) g
/2
(an))[ +
n
[g
/2
(bn) g
/2
(an)[
< /2 +/2 =
so f is absolutely continuous.
4. Solution. (a) Well write,
f(z) =
n
k=1
k
(z)
m
j=1
1
j(z)
where each
k
, j is the standard form of an analytic isomorphism of the unit disk. In
particular, these functions map the unit circle into itself and their only zeros are at the
points a
k
and bj respectively. Let C denote the curve t f(e
it
). Then
Winding # of f(e
it
) =
1
2i
_
C
1
d =
1
2i
_
||<1
f
()
f()
d = Vararg
|z|=1
(f),
where we are just using the denitions of winding number, path integral, and variation of
argument. But its clear that Vararg
|z|=1
(f) = n m since the m zeros of j become m
poles of f.
(b) Equivalently, we are counting the roots of
n
k=1
k
(z)
m
j=1
1
j(z)
call
= g(z) h(z).
Since the the unit disk is stable under all
k
, j, we have [g(z)[ = 1 = [h(z)[ for all [z[ = 1.
Thus
[g(z) (g(z) h(z))[ = [h(z)[ < 1 = [g(z)[,
so by Rouches theorem g(z) and g(z) h(z) have the same number of zeros in the unit
disk, counting multiplicity. But g(z) clearly has n roots counting multiplicity, so were
done.
5. Solution. I think this is one of those keep doing what you think you should be
doing and you get the answer problems. I have the solution in my notes... dont want to
type it out... god I hate Real.
6. Solution. (a) Remark: Standard problem, with a slight twist: you need to split up
the sum due to accumulations of poles at both 0 and .
5
The proof does not use or need the fact that f is of bounded variation
2. AUGUST 2000 9
Since a compact subset of C is bounded, there is a natural number L so that for all
z K, [z[ < L.
k=L
converges uniformly on compact subsets For k > log L/ log 2+1, so that 2
k1
>
L,
[z 2
k
[ > 2
k
L > 2
k
2
k1
= 2
k1
Well, heres what I have written.. this can be done better....
So for such z,
2
k
(z 2
k
)
2
<
2
k
(2
k1
)
2
= 2
k2k2
= (1/2)
k+2
.
Since
k=L
(1/2)
k+2
< , by the M-test, we get the result.
L
k=
converges uniformly on compact subsets A compact set meeting the require-
ments stays away from 0 and powers of 2. That is, there exists > 0 such that for all
z K, and k Z,
[z 2
k
[ > and, [z[ >
so... (heres just what i have written)
2
k
(z 2
k
)
2
)
<
2
k
2
=
0
k=
2
k
(z 2
k
)
2
converges uniformly by the Weierstrass M-test, since
L
k=
2
k
< .
Now using both steps, we conclude the entire summation converges since each part
does (basically the denition).
To show F is meromorphic you just need to note that the only accumulation point of
the set |2
k
[ k Z is 0; if we remove 0 then theres no accumulation of singularities of
the locally analytic function F, which is what it means to be meromorphic.
The nal identity is just formal, every step valid by uniform convergence,
F(2z) =
k=
2
k
(2z 2
k
)
2
=
k=
2
4
2
k1
(z 2
k1
)
2
=
1
2
F(z).
(b) Suppose not. Dene a function on C |0 by
G(z) =
1
F(z)
.
This function is analytic on the domain being considered, and it has zeros exactly where
F has poles (note the zero 2
k
of G corresponds to the pole 2
k
of F. You also need to
make sure that in fact F has an actual pole of order 2 at each 2
k
and not a removable
singularity).
I claim that 0 is a removable singularity of G. This is because
G(z) =
1
F(z)
=
1
2F(2z)
=
1
2
G(2z)
and so G(z) =
1
2
k
G(2
k
z). Without getting too technical, we can pick some compact
neighborhood, say a large annulus, such that for every [z[ < 1, 2
k
z is in the compact
neighborhood for some k 0. Being continuous, G is bounded on this neighborhood, so
the equality we just proved shows that G is bounded on [z[ < 1, so z = 0 is a removable
singularity.
But now we have a contradiction: G(0) can be dened to make G(z) analytic on a
neighborhood of 0, but G has an accumulation of zeros at 0 corresponding to the accu-
mulation of poles of F at (heres where its importation we veried that F has actual
poles). Thus, G(z) is identically 0, which is absurd.
6
3. Solutions for January 2001
1. Solution. (a) Let > 0. By absolute continuity, there is such that
n
[bn an[ < =
n
[f(bn) f(an)[ <
E has measure 0, so there exist closed intervals |In covering E such that
n
(In) <
Since f is continuous and each In is compact, f attains it minimum at some an In and
maximum at some bn In. Then each [f(an), f(bn)] covers f(In), so the collection of
intervals |[f(an), f(bn)] covers f(E). Since [an, bn] In,
n
[bn an[
n
(In) < =
n
([f(an), f(bn)]) < .
Since was arbitrary, f(E) has measure 0.
(b) Let E be measurable. By inner regularity, there is an F set F such that F E
and m(E F) = 0. Let F =
n
Fn, Fn closed. Each Fn [0, 1] and they are therefore
compact, so each f(Fn) is compact by continuity of f. So,
f(F) = f
_
_
n
Fn
_
=
_
n
f(Fn) = f(F) is F.
By part (a), since E F has measure 0, f(E F) has measure 0. Then f(E) = f(F
(E F)) = f(F) f(E F), so again by inner regularity, f(E) is measurable.
4. Solution. c open. Since u is harmonic, it is continuous, therefore u
1
(c, ) is
open in C.
c nonempty. Suppose there exists c such that u(z) c for all z C. Since C is
simply connected, u is the real part of an entire function f. Since
[e
f(z)
[ = e
Re(f(z))
= e
u(z)
,
6
There is a second (harder) proof. You can also dene,
G(z) =
1
F(1/z)
.
This is what I tried doing rst and it didnt work out because I couldnt show neatly that 0 was a
removable singularity, but I bet it can be done. This version is dierent from the previous one in that
you use Liouville at the end.
4. AUGUST 2001 11
we have [e
f(z)
[ e
c
for all z C. But e
f(z)
is an entire function, being the composition
of two entire functions, so by Liouville it is constant. So f is constant so, therefore, is u.
c simply connected. Let be a Jordan curve contained in c. Then the interior of
, call it U, is a connected domain. Take z0 U, and suppose z0 , c. Then u(z0) c,
while for all z , u(z) > c since c. But this contradicts the minimum modulus
principle for harmonic functions,
7
since u is not constant. Thus U c, that is, is
nullhomotopic.
2. Solution. Since the integrand is a rational function, it is meromorphic; its
singularities are the 2nth roots of 1. Now let = e
i/2n
, so that
2n
= 1. Choose the
path R that is composed of the straight line from 0 to R, the arc from R to R
2
= Re
i/n
,
and the line segment from R
2
back to 0.
Picture here
Then R contains exactly one singularity of the integrand, . Futher, since
d
dz
z
2n
+ 1
= 2n
2n1
,= 0,
it follows that is a simple pole. Thus the residue at is
Res
1
z
2n
+ 1
=
1
d
dz
z
2n
+ 1
=
1
2n
2n1
.
By the residue theorem,
_
R
dz
z
2n
+ 1
= i
_
1
n
2n1
_
=
i
n
as
2n
= 1.
Now parameterize the line segments of R by 1(r) = r, and 3(r) = r
2
, so
1
(r) = 1,
3
(r) =
2
. Then the integrals along 1 and 2 satisfy
_
3
dz
z
2n
+ 1
=
_
0
r=R
2
dr
(r
2
)
2n
+ 1
=
2
_
0
r=R
dr
r
2n
+ 1
=
2
_
R
r=0
dr
r
2n
+ 1
=
2
_
1
dz
z
2n
+ 1
,
thus,
_
R
dz
z
2n
+ 1
=
_
1
dz
z
2n
+ 1
+
_
2
dz
z
2n
+ 1
+
_
3
dz
z
2n
+ 1
= (1
2
)
_
1
dz
z
2n
+ 1
+
_
2
dz
z
2n
+ 1
.
7
Heres a quick proof: again u is the real part of an analytic f, and look at e
f
. Then |e
f
| = |e
u
|.
If u attains its minimum value on the interior, then e
f
attains its maximum as well, so is constant.
On 2, [z[ = R, and [z
2n
+ 1[ 2R
2n
. So if n > 1,
8
2
dz
z
2n1
+ 1
en(2) sup
z
2
1
z
2n
+ 1
R
2
2R
2n
0.
Now we simply take limits to get,
(1
2
)
_

x=0
dx
x
2n1
+ 1
=
i
n
which simplies down to
_

x=0
dx
x
2n1
+ 1
=
i
n(
1
)
=
i
n 2i Im()
=
/2n
sin(/2n)
3. Solution. Remark: there are two ways to prove the integration by parts formula.
Recall that in calculus, integration by parts is proved from the product rule for derivatives,
so you can try and recreate that proof. You can also prove it directly from Fubinis
theorem.
Proof 1: For all x [0, ),
[G(x)[ =
_

x
g(y) dy
_

x
[g(y)[ dy
_

0
[g(y)[ dy < .
So G(x) is bounded, as is F(x). The product of an integrable function with a bounded
function is integrable, so f(x)G(x) and g(x)F(x) are integrable.
4. Solution. Note that it is not possible to prove that h is continuous!
9
Step one.
Notice that since g is bounded on [z[ > 1, it has a removable singularity at . That is,
[g(1/z)[ is bounded on the punctured disk |[z[ < 1, z ,= 0, so can be made to be analytic
on |[z[ < 1.
Step two. Dene k(z) = g(1/z). Then k is an analytic function on [z[ < 1.
10
Fix
z C. By the denition of h and k,
lim
r1
k(rz) = lim
r1
g(1/rz) = lim
r1
+
g(rz) = h(z)
since 1/z = z.
Step three. Let u(z) = Re(f(z) k(z)) and v(z) = Im(f(z) +k(z)). Then u and v are
the real and imaginary parts of analytic functions, so are harmonic function on [z[ < 1.
Then
lim
r1
u(rz) = Re( lim

r1
f(z) k(z)) = Re(h(z) h(z)) = 0

and
lim
r1
v(rz) = Im( lim

r1
f(z) +k(z)) = Im(h(z) +h(z)) = 0.

8
n = 1?
9
example
10
the easiest way to see this is to show k satises the Cauchy-Riemann equations
5. JANUARY 2002 13
So both u and v can be extended continuously to be 0 on the unit circle. By the maximum
and minimum modulus principles for harmonic functions,
u 0 and v 0
so
Re(f(z)) = Re(k(z)) = Re(g(1/z) and Im(f(z)) = Im(k(z)) = Im(g(1/z)
And thus f(z) = g(1/z) on [z[ < 1. If g is nonconstant, g(1/z) is not an analytic function
since it fails Cauchy-Riemann,
11
contradicting the fact that f is analytic. Thus g is
constant, and hence so is f.
6. Solution. Step one. Suppose f has an innite number of zeros in [z[ < 1. By
Bolzano-Weierstrass, they must have an accumulation point. They cannot accumulate to
the boundary, given the condition that f is arbitrarily large near the boundary. They
cannot accumulate to the interior, since then f would be identically zero. Thus f has
nitely many zeros (of nite order).
Step two. Let the zeros of f be z0, . . . , zn with orders m0, . . . , mn. Then,
g(z) =
n
k=0
(z z
k
)
m
k
f(z)
is an analytic function in [z[ < 1. But
lim
|z|1
[g(z)[ = lim
|z|1
n
k=0
(z z
k
)
m
k
f(z)
= 0.
By the maximum modulus principle, g is identically zero, which means that f is not nite
on [z[ < 1.
1. Solution. L
2
[0, 1] is a Hilbert space with inner product
(f, g) =
_
1
0
fg
To say the sequence |n is orthonormal means
(i, j) = ij =
_
0 i ,= j
1 i = j
So
(n(x))
2
=
1
n
2
_
_
n
i=1
2
i
(x) + 2
i=j
i(x) j(x)
_
_
11
Another argument: g(1/z) is not analytic because g is conformal (a.e.) while 1/z is anti-
conformal (a.e.), so their composition is anti-conformal.
and
|n(x)|
2
L
2 =
1
n
2
_
1
0
_
_
n
i=1
2
i
(x) + 2
i=j
i(x) j(x)
_
_
=
1
n
2
n
i=1
(i, i) +
B
0
2
n
2
i=j
(i, j)
=
1
n
2
n
n
0.
Thus n 0 in L
2
([0, 1]). Since |f|
L
1 c |f|
L
2 for some constant c, n 0 in
L
1
([0, 1]). Convergence in mean implies convergence in measure, so we are done.
1. Solution. (a) Let > 0. There is a 1 > 0 such that if |[cn, dn] is any collection of
subintervals of [f(a), f(b)],
[cn dn[ < 1
[g(cn) g(dn)[ <

by absolute continuity of g. By absolute continuity of f, there is a 2 > 0, depending on
,1, such that
[an bn[ < 2
[f(an) f(bn)[ < 1.

Since f is strictly increasing, it maps intervals to intervals. Thus if |[an, bn] is a collection
of subintervals of [a, b], then |[f(an), f(bn)] is a collection of subintervals of [f(a), f(b)]
and
[anbn[ < 2
[f(an)f(bn)[ < 1
[g(f(an))g(f(bn))[ < .
So g f is absolutely continuous.
(b) An absolutely continuous function is dierentiable almost everywhere. Further, a
function is dierentiable at a point only if it is dierentiable on an open interval containing
that point. Thus at all points where g f is dierentiable, we can just use the normal
chain rule from calculus.
(c) Since f is strictly increasing, is derivative is bounded on [a, b]. The product of an
integrable function with a bounded function is integrable, so (f(x)) f
(x) is integrable.
Let g(z) =
_
f(a)
x(y) dy. Then g is absolutely continuous. So by part (b),
(g f)
(x) =
d
dx
_
f(x)
f(a)
(y) dy = (f(x)) f
(x).
By part (a), since g f is absolutely continuous, the FTC applies, so integrating both
sides gives
_
b
a
(f(x))f
(x) dx =
_
b
a
(gf)
(x) dx = g(f(b))g(f(a)) =
_
f(b)
f(a)
(y) dy
B
0
_
f(a)
f(a)
(y) dy
6. AUGUST 2002 15
4. Solution. Since g is entire and has no zeros, the function g
/g is entire. Pick any

point z0 C and set w = g(z0), w ,= 0. For any z Z, choose a path : z0 z and
dene,
f(z) =
_
()
g()
d.
since C is simply connected, by the homotopy form of Cauchys theorem, f is a well dened
function, and as its the antiderivative of the entire function g
/g, it is also entire.

Now compute,
d
dz
exp(f(z)) =
g
(z)
g(z)
exp(f(z))
so
d
dz
exp(f(z))
g(z)
=
g
(z)
g(z)
exp(f(z)) g(z) g
(z) exp(f(z))
(g(z))
2
= 0.
Thus the entire function
exp(f(z))
g(z)
is constant. What constant? Well, by Cauchys theorem
f(z0) = 0, so
exp(f(z0))
g(z0)
=
1
w
thus
g(z) = wexp(f(z)),
and if we let be any complex logarithm of w, we get that f is the required function.
12
6. Solution. (a) Let [q
(0)[ = < 1. Choose > 0 such that 1 > +

call
= . Since q
is dierentiable at 0, there exists > 0 such that
[z[ <
q(z) q(0)
z 0
q(z)
z
< + =
by the denition of dierentiability. So q maps the disk [z[ < into itself and [q(z)[ < [z[.
Thus
[q
(n)
(z)[ <
n
[z[ <
n
.
Since < 1, the geometric series
n
is nite.
13
Thus the tails of the series converge
to zero, so
|Sn Sm| = sup
|z|<
k=m+1
q
(k)
(z)
sup
|z|<
n
k=m+1
[q
(k)
(z)[
n
k=m+1
k

n
0.
Therefore the sequence is uniformly Cauchy, hence converges uniformly to an analytic
function on the disk [z[ < .
(b) Step 1. Suppose there exists r such that on [z[ < r,
sup
|z|<r
[Sn(z)[
n
.
12
Remark: This is a standard complex problem. Its done in Lang. Whatever your solution is, it
is important that is uses the topology of the complex plane! Pick a dierent Riemann surface, say the
punctured complex plane. The function g(z) = z is nonzero on the entire punctured complex plane;
however it does not have a well dened logarithm. The point is that you need simply connectedness.
13
Therefore Sn converges uniformly to an analytic function by the Weierstrass M-test. What
follows is basically a proof of the theorem
That is, the sequence of uniform norms |Sn| is bounded. Thus by Bolzano-Weierstrass
there is a subsequence, say |Sn
j
|, that is Cauchy. This subsequence of functions therefore
converges uniformly on the disk [z[ < r to some analytic function, say h, and further
S
n
j
h
.
Step 2. By the chain rule,
(q
(k)
)
(0) = q
(q
(k1)
(0)) q
(q(0)) q
(0) = (q
(0))
k
,
since q(0) = 0. So,
S
n
j
(0) =
n
j
k=1
(q
(k)
)
(0) =
n
j
k=1
(q
(0))
k
.
Since [q
(0)[ > 1, the sequence S
n
j
(0) . But S
n
j
(0) h
(0), which must be a nite

number.
6. Solution. Suppose otherwise, [z[ = 1 [P(z)[ < 1. Then for all [z[ = 1,
1 = [z
n
[ > [P(z)[ = [(z
n
P(z)) z
n
[
By Rouches theorem, since z
n
, P are analytic on [z[ < 1, z
n
and z
n
P(z) have the same
number of zeros counting multiplicity inside the disk. But 0 is a zero of multiplicity n of
z
n
, while z
n
P(z) = an1z
n1
+ +a1z +a0 is a polynomial of degree (n 1), so has
at most (n 1) roots counting multiplicity.
6. Solution. (a) Let ( be the family of maps | h, where (z) =
z1
z+1
.
Picture here
First, notice that is an analytic isomorphism from C (, 0] D. Therefore, ( is
a normal family if and only if is.
14
But ( is a normal family because it is uniformly
bounded by 1.
(b) Let hn such that hn(0) 0. Let gn = hn.
Claim: gn 1 uniformly on compact subsets of D.
Pf: Suppose not. Then there is a compact set K D, > 0, and subsequence gn
k
such that,
sup
zK
[gn
k
(z) + 1[ > .
But gn
k
is itself a normal family, so there is a subsequence gn
k
j
that converges uniformly
on K to some analytic function g : D D. By continuity, lim
n
gn(0) = (0) = 1, so
g(0) = 1. Since g maps into the unit disk, by the maximum modulus principle, g must
be identically 1. So gn
k
j
converges uniformly to 1, that is, there is a J such that for
all j J,
sup
zK
[gn
k
j
(z) + 1[ <
14
??why?
9. AUGUST 2004 17
contradicting the fact that the whole sequence gn
k
is uniformly bounded away from 1.
So gn converges to 1 uniformly, and composing with
1
, it follows that hn converges
to 0 uniformly.
15
2. Solution. This is a standard qual problem. The answer is
/9
sin /9
.
Years where this problem appears:
August 2008 number 2
3. Solution. (a) Since fn C
1
[0, 1], the FTC holds, so for all x [0, 1],
fn(x) =
_
x
0
f
n
by (a)
Now by (b) all the derivatives f
n
are dominated by a function in L
1
[0, 1] [by basic calc],
so f
n
L
1
[0, 1] and we can use the LDCT: since f
n
h by (c),
lim
n
fn(x) = lim
n
_
x
0
f
n
=
_
x
0
h
call
= f(x).
By the Lebesgue Dierentiation Lemma, f(x) is an absolutely continuous function. There-
fore all that remains to show is that the convergence is uniform.
(b) Since both f
n
h pointwise and |f
n
|1 |h|1, it follows that f
n
h in L
1
[0, 1] as
well.
16
That is, given > 0,
_
1
0
[f
n
h[ < for suciently large n, so
[fn(x) f(x)[ =
_
x
0
f
n

_
x
0
h
_
x
0
f
n
h
_
x
0
[f
n
h[
_
1
0
[f
n
h[ <
Since the inequality does not depend on x, the convergence is uniform by denition.
4. Solution. (a) Let u : C R be harmonic. Dene a complex function
f = uy +iux.
That is, f is basically the gradient of u. Since u is harmonic it is C
2
, f is continuous with
continuous real partials. Further,
(uy)x = (ux)y since u is C
2
(uy)y = (ux)x since
2
u = 0.
Therefore by the converse to Cauchy-Riemann, f is an entire function.
15
??why?
16
cite
(b) Now pick any C. Then f is also an entire function, and nonconstant if u is
nonconstant. By Picards theorem, it misses at most one point, in particular it hits the
imaginary axis. Thus, for some z, Re(f(z)) = 0. But,
Re(f(z)) = Re uy(z) Im ux(z).
Thus for this z,
Re
Im
=
ux(z)
uy(z)
.
The thing on the left is the slope of , and the thing on the right is the slope of u at z.
5. Solution. (a) Notice that fn(x) 0 for all x, n. Suppose that fn(x) fn1(x), the
base case already assumed. Then
fn+1(x) =
__
x
0
fn
_
1/2
__
x
0
fn1
_
1/2
since fn nonnegative
= fn(x).
Thus by induction, |fn(x) is monotonically decreasing and a nonnegative sequence.
Therefore it has a nonnegative limit, namely its inmum.
(b) Since f0 L
1
and for all x [0, 1], fn(x) f0(x), by the LDCT
lim
n
_
x
0
fn =
_
x
0
lim
n
fn =
_
x
0
f.
By continuity of

,
__
x
0
f
_
1/2
= lim
n
__
x
0
fn
_
1/2
= lim
n
fn+1(x) = f(x).
(c) Part (b) says that f is integrable, since its integral on [0, 1] is just (f(1))
2
. Now
_
x
0
f is absolutely continuous by the Lebesgue lemma so in particular it is continuous. So
f(x) =
_
_
x
0
f is continuous. By the old FTC from elementary calculus if f is continuous,
then
_
x
0
f is C
1
and d/dx
_
x
0
f = f(x), everywhere.
17
But also
_
x
0
f = (f(x))
2
, so by the
chain rule
f(x) = d/dx
_
x
0
f = 2f
(x)f(x)
Thus if f(x) > 0 we cancel f(x) from both sides to get f
(x) =
1
2
.
(d) The solution to this dierential equation with the constraint f(x) > 0 is f(x) =
x
2
+c
for some c 0.
6. Solution. The map z
z+1
z1
is an analytic isomorphism from the right half plane
to the unit disk.
Picture here
17
This may seem like much, but I think you must say all this... all the theorems from analysis are
true almost everywhere, and this problem wants you to prove it everywhere. In particular, the lebesgue
dierentiation lemma wouldnt work
10. JANUARY 2005 19
Let g = f. Then g maps D to D, and g(0) = (f(0)) = (1) = 0. So by the
Schwarz lemma for all z D, [z[ [g(z)[. Since f(z) =
1
(g(z)) =
1+g(z)
1g(z)
, we get
[f(z)[ =
1 +g(z)
1 g(z)
1 +[z[
[1 g(z)[
by the triangle inequality, and since 1 [z[ [g(z)[, by the reverse triangle inequality,
[1 g(z)[ 1 [z[, so
[f(z)[
1 +[z[
[1 g(z)[

1 +[z[
1 [z[
.
Remark: a better but similar problem from the Michigan qual: if f maps D into |z
C [ 1 < Re z < 1, and f(0) = 0, show for all z D,
[ Imf(z)[
2
log
1 +[z[
1 [z[
.
1. Solution. Suppose that such a sequence does not exist. Then there exists n N
such that for all x n, [xf(x)[ >
1
n
, since otherwise, we can pick such an xn for each n,
and this sequence will satisfy the required property. So for all x n, [f(x)[
1
nx
. But by
comparison,
_

n
[f[
_

n
1
nx
=
since
1
x
, L
1
(n, ) for any n, contradicting the fact that f L
1
(0, ).
2. Solution. (a) Choose the contour integral below:
Picture here
and select the branch of

such that along 1, log(z z
2
) has a small imaginary
part. Specically , write
log(z z
2
) = log
1
z + log
2
(1 z)
and pick log
i
dened by arg
i
where
arg
1
z (2, 0) arg
2
z (, ).
Then by continuity, as 1 approaches the real axis, and z 1, log
1
z approaches ln [z[
and log
2
(1 z) approaches ln [1 z[, so
z z
2
approaches exp(
1
2
(ln [z[ + ln [1 z[)),
which will be exactly
x x
2
on the real axis. Along 3, things are dierent. For z 3,
log
1
z approaches ln [z[ 2i and log
2
(1z) approaches ln [1z[. So
z z
2
approaches
exp(
1
2
(ln [z[ +ln [1 z[) i) = exp(
1
2
(ln [z[ +ln [1 z[)). Thus, as 1, 3 approach the
real axis, taking into account the reverse orientation of 3, both
_
1
and
_
3
approach the
real integral we are trying to compute.
(b) Our choice of

is clearly locally analytic, but it might not be well dened: the only
way this would happen is if the function is not continuous as we loop around the slit. But
as we follow , by the time weve made a loop weve picked up a 2i from log
2
and a
+2i from log
1
, which end up canceling. Thus
z z
2
is analytic C [0, 1].
(c) We now compute
_
. Let w =
1
z
. Then dz =
1
w
2
dw, and the integral becomes
_
z z
2
z + 2
dz =
_
1
w
2
_
1
w

1
w
2
1
w
+ 2
dw =
_
w 1w
2
w(1 + 2w)
dw
where
is oriented in the wrong (CW) direction. Then the integrand has isolated singu-
larities at w = 1
1
2
and w = 0.
4. Solution. f also denes a complex function,
18
which by factoring,
f(z) =
1
(1 +z
2
) +z
4
(1 +z
2
)
=
1
(1 +z
2
)(1 +z
4
)
,
we see that f is meromorphic with precisely six simple poles: at i and the fourth roots
of 1.
Picture here
It is a (nontrivial) fact that the radius of convergence of
n
an(z 2)
n
is also the radius
of the largest disk centered at 2 on which f is analytic.
19
Since the nearest singularities
of f to 2 are obviously e
i/4
=
2/2 + (
2/2)i, and
[2
2/2 + (
2/2)i[ =
_
(2
2/2)
2
+
1
2
=
_
5 2
2
it follows that the radius of convergence is
_
5 2
2. By Cauchy-Hadamard,
lim
n
[an[
1/n
= radius of convergence =
_
5 2
2.
6. Solution. This problem would take a couple pages to get all the details right. So
for right now Im going to leave most of them out. Heres how you should think of the
problem.
Simplify the region. Think of regions on the complex plane as stretchy regions on a
sphere. So this region D is just all of the sphere with two circles removed. Well if you
think about it, thats just an annulus (almost). In fact you can nd the map explicitly,
but Ill leave the details out:
There is a map : C C that maps D isomorphically to a region

D
which is the unit disk with a smaller disk cut out of the interior. The
map is a bilinear transformation.
18
without making a fuss, well just prove the required fact for the complex function f. If you
want to be thorough, you need to observe that our new complex function f(z) will restrict to be the
original function f(x) on the real line; by uniqueness, its power series will be the same as the power
series for the original (real) function, and therefore have real coecients, and also anything we prove
for the coecients (radius of convergence) of f(z) will also be true of the coecients (resp. radius of
convergence) of f(x). But why bother saying all that?
19
Lets talk about this a bit. Since the problem is pretty trivial except for this fact, you are perhaps
supposed to know how to prove it.
Note this statement is false for real analytic functions. The simplest example is f(x) = 1/(x
2
+1);
it is analytic at 0 and you can compute the radius of convergence of its power series to be 1. However,
f is analytic on the entire real line, even though it cannot be represented by the power series centered
at 0 outside of |x| < 1.
Insert proof here.
11. AUGUST 2005 21
Extend the map. Notice weve reduced the problem to showing that any function
f, analytic and 1:1 on

D and satisfying

f(
D) =

D, is a bilinear transformation (set
f = f
1
). To do this well extend to a map on the entire unit disk. First, since
f is analytic and 1:1, it takes the boundary to the boundary. So by composing with
a proper bilinear map, we may assume that it takes the outer circle to the outer circle.
It is a theorem that we can extend the map continuously to the inner circle (and it will
be an injection there). The circle is an analytic arc and we can therefore apply Schwarz
reection to get an analytic, 1:1 map on the unit disk with a smaller disk removed.
[Now heres where you get bogged down in details.] As you apply Schwarz reection
ad innitum, you get an analytic, 1:1 map on the punctured unit disk (the puncture wont
be in the center). Finally, we note that this puncture is obviously a removable singularity,
since were mapping inside the unit disk and therefore bounded.
20
Classify the map. Therefore we reached our goal: weve found a map g which is an
analytic isomorphism of the unit disk. All analytic isomorphisms of the unit disk are
bilinear transformations.
21
Therefore g is a bilinear transformation. Its restriction to

D
is

f and by uniqueness of complex analytic functions,

f is also bilinear.
1. Solution. Since f is bounded, by replacing it with f/|f| we may assume WLOG
that [f(x)[ 1, since f is integrable if and only if f/|f| is. Dene the set
En = |x [
1
n
4
[f(x)[ >
1
(n + 1)
4
.
20
To get a good grasp of this problem, try the following problem with is basically the same thing:
Regions in the complex plane have an equivalence relation: we call U, V C conformally equivalent
if theres an analytic isomorphism : U V . The Riemann mapping theorem can be interpreted as
saying all simply connected domains are in the same conformal equivalence class.
Fact: Two annuli are conformally equivalent if and only if the ratio
big radius
small radius
is equal for both annuli. Also, all analytic automorphisms of annuli are bilinear
transformations.
21
There are three objects whose analytic automorphisms you must memorize:
All analytic isomorphisms f : C C are linear (f(z) = az + b, a, b C). This is a corollary
of Casorati-Weierstrass.
All analytic isomorphisms f : {upper half plane} {upper half plane} are given by
f(z) =
az + b
cz + d
where
a b
c d
SL2(R).
All analytic isomorphisms f : {unit disk} {unit disk} are given by
f(z) =
z
1 z
where || < 1. This is a corollary of the Schwarz Lemma.
Then En |x[[f(x)[ > (n + 1)
4
, so mEn C/
_
(n + 1)
4
= C(n + 1)
2
. Further on
the set En, [f(x)[ 1/n
4
, so
_
En
[f[
1
n
4
C(n + 1)
2
= C(
1
n
2
+
2
n
3
+
1
n
4
).
The disjoint union of all the En is the support of f, so by the Monotone Convergence
Theorem,
_
R
[f[ =
_
nEn
[f[ =
n
_
En
[f[
n
C(
1
n
2
+
2
n
3
+
1
n
4
) < ,
since the three series on the right converge by the p-test. So f is integrable.
2. Solution. (a) Let an be the coecient in the power series. Then
lim
n
an+1
an
= lim
n
(3n + 3)!
(3n)!

(2n + 2)!
(2n)!

n)!
(n + 1)!

(4n)!
(4n + 4)!
= lim
n
(3n + 3)(3n + 2)(3n + 1)(2n + 2)(2n + 1)
(n + 1)(4n + 4)(4n + 3)(4n + 2)(4n + 1)
=
3
3
2
2
4
4
= (3/4)
3
.
By the ratio test, this power series converges for all z such that [z[ < (3/4)
3
=
(4/3)
3
= R.
(b) Let [z[ = R = (4/3)
3
Notice that if k > n,
(kn)! (kn)(kn 1) (kn n + 1)
. .
n factors
(kn n)
n
,
while clearly (kn)
n
(kn)!. So
an
(2n)
n
(n)
n
(n)
n
(4n)
n
=
_
1
2
_
n
and so,
[anz
n
[ = an[z[
n
_
1
2
_
n
_
4
3
3
3
_
n
=
_
32
27
_
n
n
.
So the series
n
anz
n
cannot converge since the modulus of the terms do not converge to
0.
3. Solution. Let G(x) =
_
x
0
[f(t)[
2
dt. By the Lebesgue lemma, G
(x) = [f(x)[
2
. Now
suppose that G(b) > 0, b a, where a is given in the hypothesis. Since [f(x)[
2
0, G is
increasing, so G > 0 on the set [b, ). Therefore 1/G is a continuous function dened on
[b, ), with derivative satisfying
_
1
G
_
(x) =
G
(x)
G
2
(x)
=
[f(x)[
2
(
_
x
0
[f(t)[
2
)
2
1
11. AUGUST 2005 23
by hypothesis. But it is not possible to have a positive continuous function decreasing
more than linearly.
22
So we get that for all b a,
G(b) = 0 =
_
b
0
[f(t)[
2
dt
Since b was arbitrary and [f(t)[
2
is nonnegative, by Fatous lemma
_
0
[f(t)[
2
dt = 0.
Hence [f(t)[
2
= 0 almost everywhere, so the same must be true of f(t).
4. Solution. We rst look at the image of the domain D under the map z 4/z. This
Mobius transformations sends circles through the origin to lines and all other circles to
circles. So the image of the circles [z 2[ = 2 and [z 1[ = 1 will be some lines. Notice
that z 4/z takes the real line to itself. Further, it is a conformal map at z = 1 and
z = 2. Since the circles [z 2[ = 2 and [z 1[ = 1 are perpendicular to the real line, their
images will be perpendicular to the real line, i.e., be the vertical lines x = 4/4 = 1 and
x = 4/2 = 2, resp.
Picture here
Since each Cn gets mapped to a circle, call it C
n
, and 4/z is conformal except at 0,
each Cn preserves its tangency to the circles [z 2[ = 2 and [z 1[ = 1, so each C
n
will
be a circle tangent to the vertical lines x = 1 and x = 2, so we get the picture above.
Let s = 4/t. The exterior of the circle [z[ = t gets mapped to the interior of the
circle [z[ = s under the map 8/z. So if we let N0(s) be the number of C
n
contained in the
interior of the circle [z[ = s, we get N(t) = N0(s).
Now let B1(s) and B2(s) be the regions below, specically,
B1(s) = |x +iy [ [y[
_
s
2
4
and
B2(s) = |x +iy [ [y[ s.
If N1(s) and N2(s) represent the number of C
n
contained in B1(s) and B2(s), we have
the inequality
N1(s) N0(s) N2(s)
Now each C
n
has diameter 1 and B2(s) has height 2s. It isnt too hard to see that B2(s)
contains exactly 2[[s
1
2
[] + 1 circles, so we get the inequality
2s 1 N2(s) 2s
Picture here
5. Solution. (a) No. as a counterexample, xsin(1/x) is absolutely continuous on [, 1]
for all > 0 and continuous at 0, but it is not absolutely continuous on [0, 1].
22
By the mean value theorem, there is a c [b, b + n] such that
1
G
(c) =
(1/G)(b + n) (1/G)(b)
(b + n) b
(1/G)(b)
n
n
0
which contradicts the bound on the derivative.
(b) Yes. Suppose f is of bounded variation on [0, 1]. Then f
exists almost everywhere

and is integrable. In particular, for every > 0, there exists > such that,
mE <
_
E
[f
[ < .
Let |[ai, bi]i be a countable collection of disjoint intervals such that
i
[bi ai[ <
MISSING
6. Solution. Pf 1: Since [z[ 0 [z[
1/2
0 [f(z)[ 0, f has a removable
singularity at z = 0, so WLOG f(0) = 0. Let z0 C, [z0[ = R. Then by Cauchys
formula, if C2R = |z [ [z[ = 2R,
[f
(z0)[ =
1
2
_
C
2R
f(z)
(z z0)
2
dz
1
2
2(2R) sup
|z|=2R
[f(z)[
[z z0[
2
.
Since [z z0[ R for [z[ = 2R, and given [f(z)[ [z[
1/2
, we get
[f
(z0)[ 2
R
R
2
R
1/2
=
2
R
1/2
0
as R . Thus f
is bounded. But f being entire implies f
is entire, so by Liouville,
f
is a constant. Thus f is linear. But no nontrivial linear function can satisfy the given
growth condition on [f(z)[ [z[
1/2
, so we must have f 0.
23
Pf 2: One can show that f has to be linear in another manner. Again, f has a
removable singularity at 0, so WLOG f(0) = 0. If [z[ = 1, [f(z)[ [z[
1/2
= 1. Thus
f(D) = D where D is the unit disk, so by Schwarzs lemma, z D [f(z)[ [z[. By
basic algebra, [z[ 1 [f(z)[ [z[. Now let g(z) = f(z)/z. Since f is continuous and
f(0) = 0, g is entire. Further we have shown that g is bounded by 1 on the complex plane,
so by Liouville, g is constant, i.e., f is linear. Again, this contradicts the growth condition
on f unless it is identically 0.
1. Solution. The function
x
n
[0,n]
attains its maximum at x = n, at which point its value is n/n = 1. Thus for all x,
[
x
n
[0,n]
[ 1, so
x
n
[0,n]
f(x)
f(x).
Since f L
p
(R), by the LDCT,
lim
n
1
n
_
n
0
xf(x) dx = lim
n
_
R
x
n
[0,n]
f(x) dx =
_
R
f(x) lim
n
x
n
[0,n]
dx =
_
R
f(x)0 dx = 0.
2. Solution. (a) Let A be divided up into three overlapping sections as shown at right,
call them A1, A2, A2.
Picture here
23
Remark: one can repeat this argument with any growth condition of the form |f(z)| |z|
p
and
Cauchys formula for a suitable choice of nth derivative and show that f has to be a polynomial of
degree depending on p.
12. JANUARY 2006 25
On each Ai, since it is simply connected, there exists an analytic function fi such
that
Re fi = u on Ai.
This gives us three (probably) dierent functions. However they all dier by an imaginary
constant, for example on A1 A2,
Re(f1 f2) = u u = 0.
By Cauchy-Riemann, any holomorphic function with constant real part has constant imag-
inary part.
24
Applying this to the holomorphic function f1 f2 on A1 A2, we can write
f1 = f2 +iK2. Similarly
MISSING
3. Solution. (a) Let y = nx, so dy = ndx and
_
|x|
nk(nx) dx =
_
|y|n
k(y) dy.
By the MCT, since k 0,
lim
m
_
m
m
k(y) dy =
_
R
k(y) dy.
As n , n , and thus by additivity
lim
n
_
|y|n
k(y) dy = lim
n
__
R
k(y) dy
_
n
n
k(y) dy
_
= 1 1 = 0.
(b) Again making the change of variables y = nx, the integrand becomes k(y)g(y/n) dy.
Since g is bounded, the integrand is dominated by the integrable function k(y), so by the
LDCT
lim
n
_
R
k(y)g(y/n) dy =
_
R
lim
n
k(y)g(y/n) dy =
_
R
k(y)g(0) dy = g(0),
where the penultimate step follows from the fact that g is continuous.
25
5. Solution. (a) The function x f(x, y) is decreasing by (i) and has continuous
partial by (iii), f1(x, y) 0, so f1(x, y) is a nonnegative function. Thus by Tonelli (two
times)
_

x
_

0
f1(t, y) dy dt =
__
{(t,y) dtt>x,y>0}
f1(t, y) dt dy =
_

0
_

x
f1(t, y) dt dy.
Since x f(x, y) is C[1], by the (freshman calculus) FTC, for all b x,
_
b
t=x
f1(t, y) dt =
_
x
t=b
_

t
f(t, y)
_
dt = f(x, y) f(b, y).
24
If f(z) = u(x, y) + iv(x, y) then ux = uy = 0 = vy = vx = 0
25
Remark: this is probably not the proof that was intended, since part (a) suggests that the -
denition of continuity is supposed to be used for part (b). As a modication to part (b), try solving
the following using part (a): If g L
2
(R) and continuous on R, but not necessarily bounded, and k is
also square integrable, show the same result holds.
By (i), as b , f(b, y) 0. We obtain
_

x
_

0
f1(t, y) dy dt =
_

0
f(x, y) lim
b
f(b, y) dy =
_

0
f(x, y) dy = F(x).
(b) By the Lebesgue lemma, if g(t) L
1
(a, b), then G(x) =
_
x
a
g(t) dt is dierentiable
almost everywhere and G
(x) = g(x). Let g(t) =

_
0
f1(t, y) dy, which must exist a.e.
by part (a). Also by part (a), for all x > 0, g is integrable with respect to t on the interval
(x, ). So picking any a (x, ), if we let G(x) =
_
a
x
g(t) dt, G
(x) = g(x). Since

_
a
g(t) dt is constant with respect to x, its derivative is 0. So we get,
F
(x) =
d
dx
_

x
g(t) dt = g(x) =
_

0
f1(t, y) dy for almost all x > 0.
3. Solution. Since (x
p
+
1
x
p
)f L
2
(0, ), (x
p
+
1
x
p
)
2
f
2
= x
2p
f
2
+ 2f
2
+ x
2p
f
2
L
1
(0, ). Since this is the sum of three nonnegative functions, it follows that each one is
integrable on the same domain. So
x
2p
f
2
L
1
(0, ) x
p
f L
2
(0, )
x
2p
f
2
L
1
(0, ) x
p
f L
2
(0, ).
Now we break up the integral over f into the sets (0, 1] and (1, ) and compute each
separately. Over (0, 1], we have that x x
p
is square integrable, so by H olders,
_
1
0
f(x) dx =
_
1
0
x
p
x
p
f(x) dx
__
1
0
x
2p
dx
_
1/2
__
1
0
x
2p
f
2
(x) dx
_
1/2
... UNFINISHED?
1. Solution. Let gn(x) =
x
n
1+x
n
. We write
lim
n
_

0
gnf = lim
n
__
1
0
gnf +
_

1
gnf
_
and evaluate each integral separately.
Case: x (0, 1). Then 0 < x
n
< 1, so gn(x) < 1. Thus gnf is dominated by the
integrable function f, so by the LDCT and the fact that x
n
n
0,
lim
n
_
1
0
gn(x)f(x) =
_
1
0
lim
n
x
n
1 +x
n
f(x) =
_
1
0
0 f(x) = 0.
Case: x (1, ). In this case x
n
> x
n+1
, so
1
gn(x)
=
1 +x
n
x
n
=
1
x
n
+ 1 >
1
x
n+1
+ 1 =
1
gn(x)
14. JANUARY 2007 27
thus gn is monotonically increasing with respect to n. Further as n , x
n
, so
gn(x) 1. By the MCT,
lim
n
_

1
gn(x)f(x) =
_

1
lim
n
gn(x)f(x) =
_

1
f(x).
2. Solution. Let be a primitive 8th root of -1. Since z
8
+ 1 =
(z
j
), we are
looking for a partial fraction decomposition of the form,
f(z) =
a1
z
+
a2
z
2
+ +
a7
z
7
for some constants a1, . . . , a7.
Now let j be the small circle centered at j shown at right, so j contains no other
power of . Then f(z) has a simple pole at .
_
1
f
_
(z) =
z
7
(8z
7
) (z
8
+ 1)(7z
6
)
(z
7
)
2
,
since (
j
)
8
= 1,
_
1
f
_
(
j
) =
8
14j
(0)(7
6j
)
14j
= 8.
Now f has a simple pole at
j
, so,
26
Res
j
1
(1/f)
=
1
(1/f)
(
j
)
=
1
8
.
Therefore by the residue theorem,
_
j
f(z) = 2i(Res
j ) =
i
4
.
For i ,= j,
P
i
(z)
z
i
is analytic inside
j
, so,
i
4
=
_
j
f(z) =
i
_
j
ai
z
i
=
_
j
aj
z
i
= aj
giving our partial fraction decomposition.
3. Solution. Let
p
=
p +q
q
, q
=
p +q
p
and, r
=
pq
p +q
Then we have the equalities,
1
p
+
1
q
= 1,
1
r
+
1
r
= 1, p
= p and, q
= q
Notice that ([f[
r
)
p
= [f[
r
= [f[
p
, so [f[
r
L
p
[0, 1]. Similarly, [g[

q
L
q
[0, 1].
By H olders inequality applied to the conjugate pair p
, q
,
__
1
0
[f[
r
[g[
r
_
1/r
__
1
0
[f[
r
_
1/p
__
1
0
[g[
r
_
1/q
=
__
1
0
[f[
p
_
1/p
__
1
0
[g[
q
_
1/p
<
26
do this a better way without using this dumb formula. I hate all these formulas... theyre
impossible to remember
so fg L
r
[0, 1], and |fg|

r
|f|p|g|q. So by H olders inequality applied to the conju-
gate pair r, r
, (fg)h is integrable and,

|fgh|1 |fg|
r
|h|r |f|p|g|q|h|r.
4. Solution. (a) We rewrite the indexing and coecients so that g looks like a normal
power series,
g(z) =
n=1
bnz
n
where bn = 0 unless n = 2
m
for some m, in which case bn =
1
m!
. Then the nonzero terms
of [bn[
1/n
are a decreasing positive sequence and therefore have a limit, which will be the
lim of [bn[
1/n
. By continuity,
lim
n
[bn[
1/n
= lim
m
[
1
m!
[
1/2
m
= exp
_
lim
m
1
2
m
ln(
1
m!
)
_
= exp
_
lim
m
ln(m!)
2
m
_
For all m > 1, m! > m, so since ln is increasing, ln(m!) ln(m), so nally by LHopitol,
lim
m
ln(m!)
2
m
lim
m
ln(m)
2
m
= lim
m
1/m
ln 2 2
m
= 0.
Thus lim
n
[bn[
1/n
= exp(0) = 1, so the radius of convergence of the power series is 1.
(b) For m 2
n
, (e
2ik/2
n
z)
m
= z
m
, so the power series of g is
g(e
2ik/2
n
z) =
m1
j=1
(e
2ik/2
n
z)
j
j!
+
j=m
z
j
j!
=
_
m1
j=1
(e
2ik/2
n
z)
j
j!

m1
j=1
z
j
j!
_
+g(z)
call
= p(z) +g(z)
(c) Suppose g has an extension from D, so there exists a path leaving D which has a
neighborhood where g is analytic. So crosses D, so there exists D, with disk of
radius such that g is analytic on the ball of radius about . Pick any z D. By the
density of |e
2ik/2
n
in the unit circle, there are a k, n such that [e
2ik/2
n
z[ < /2.
But by part b, on the ball of radius about e
2ik/2
n
, g(z) = g(e
2ik/2
n
z) p(z). Since
p is entire, and e
2ik/2
n
z lands inside a neighborhood where we assumed g was analytic,
g is analytic at z.
(d) If g an analytic extension, g would be analytic on the entire unit circle by part c, and
would therefore have radius of convergence r > 1, contradicting part a.
4. Solution. This is a standard problem from Complex Dynamics; of course if you
havent seen complex dynamics it looks strange. The three steps of this proof are three
very important techniques to memorize.
15. SOLUTIONS FOR AUGUST 2007 29
The iterates form a normal family on the unit disk. Write g = , where
(z) = z
2
, , (z) =
z
1 z
and = 1/2. Then it is a standard result that an analytic isomorphism of the unit
disk, while maps the unit disk into itself. Thus g maps the unit disk into itself so its
iterates are uniformly bounded. By Montels theorem, |g
n
forms a normal family on the
unit disk.
Near 1, the iterates converge pointwise to 1. In the language of complex dynamics, 1
is an attracting xed point (z s.t. g(z) = z and [g
(z)[ < 1) and this is generally true of

all attracting xed points. Heres why. First its easy to check that g is analytic near 1
and that 1 is in fact a xed point. A direct computation shows
[g
(1)[ =
2
3
Then near 1, g(z) = 1 +
2
3
(z 1) +O((z 1)
2
), and now notice that g
n
(z) = 1 +(
2
3
)
n
(z
1) +O((z 1)
2
). Even closer to 1, this behaves like 1 + (
2
3
)
n
(z 1), which since [
2
3
[ < 1,
converges to 1 as n .
If a normal family converges pointwise to a constant function, then it converges nor-
mally.
27
We claim that g
n
1 normally on the whole unit disk. Suppose not. Then some
subsequence stays away from 1; formally, there is a compact set K, > 0, and subsequence
|g
n
k
such that for all k,
sup
zK
[g
n
k
(z) 1[ > ()
A subset of a normal family is normal, so there is a further subsequence |g
n
k
j
converging
normally on the unit disk to some function, say
G(z) = lim
n
g
n
k
j
(z)
The normal limit of analytic functions is analytic (Cauchys formula) so G is analytic near
1.
28
But we know g
n
k
j
(z) 1 on a possibly small neighborhood of 1, thus G is identically
1 near 1, and since its analytic,
G(z) = 1 for all [z[ < 1
But now we have a contradiction since we cant have both g
n
k
j
(z) 1 and (*), that is
since g
n
k
j
(z) 1 there exists an index J such that for all j > J,
lim
zK
[g
n
k
j
(z) 1[ < 1.
Remark: theres another approach to this problem that combines the rst two steps.
What you can do is show that for all real numbers near 1, the iterates of g converge
27
This fact is related to a result known as Vitalis theorem, but you should know how to prove it.
We say converges normally to mean uniformly on compact subsets. See the problem is that if youre
a normal family, you only know that a subsequence converges normally, so why does the whole family
converge normally? This comes up a lot.
28
technically all we know is that its analytic on the intersection of some neighborhood of 1 and
the unit disk, but that doesnt matter.
pointwise to 1. For this you just need standard real analysis. Now we just need to slightly
modify the last step... instead of convergence to a function thats identically 1 on a whole
neighborhood, its only identically 1 on a sliver of the real line near 1. But recall that an
analytic function is uniquely dened by its values on a set that accumulates; certainly this
sliver of the real line accumulates to 1, and we still get that the function G = limg
n
k
j
has
to be identically 1. Proceed the same as before.
5. Solution. By additivity,
_
R
[f[ =
_
|x|1
[f[ +
_
|x|>1
[f[.
Now we evaluate each integral separately.
(
_
|x|1
[f[) In this case, the constant function
(1,1)
L
2
(1, 1), and since f
L
2
(R), by the Cauchy-Schwarz inequality,
_
|x|1
[f[
__
1
1
1
2
_
1/2
__
1
1
[f[
2
_
1/2
=
2
__
1
1
[f[
2
_
1/2
2 |f|2
where |f|2 is the norm of f in L
2
(R).
(
_
|x|>1
[f[) By basic calculus, 1/x
2
is integrable on the set [x[ > 1, so 1/x L
2
(R
[1, 1]), and satises
_
|x|>1
1
x
2
= 2
_

1
1
x
2
= 2
_
lim
b
1
x
b
1
= 2.
So again by Cauchy-Schwarz,
_
|x|>1
[f[
_
_
|x|>1
1
x
2
_
1/2
_
_
|x|>1
[fx[
2
_
1/2
=
2 |fx|2.

Solutions To UMD Analysis Quals

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Solutions To UMD Analysis Quals

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Solutions To UMD Analysis Quals

Uploaded by

Copyright:

Available Formats

Solutions to UMD Analysis Quals

m < , its tails go to 0. So for all > 0,

m(In) < , the set of x R

mEnn(x)[ [g(x)[ [g(x)[.

this new curve in the CCW orientation, that is,

(t) = 1/C(t). Then we get

(x) = 0 for almost all x [0, 1]. Since it is increasing,

u(rz) = Re( lim

f(z) k(z)) = Re(h(z) h(z)) = 0

v(rz) = Im( lim

f(z) +k(z)) = Im(h(z) +h(z)) = 0.

[cn dn[ < 1

[g(cn) g(dn)[ <

[an bn[ < 2

[f(an) f(bn)[ < 1.

/g is entire. Pick any

/g, it is also entire.

(0)[ = < 1. Choose > 0 such that 1 > +

(0)[ > 1, the sequence S

(0), which must be a nite

f, analytic and 1:1 on

exists almost everywhere

is bounded. But f being entire implies f

(x) = g(x). Let g(t) =

(x) = g(x). Since

[0, 1]. Similarly, [g[

[0, 1], and |fg|

, (fg)h is integrable and,

(z)[ < 1) and this is generally true of

You might also like