SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS

AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Contents

Preface 2
Jan 1999 3
Aug 1999 8
Jan 2000 A 13
Jan 2000 B 18
Aug 2000 23
Aug 2001 28
Jan 2002 34
Jan 2003 37
May 2003 44
Aug 2003 49
Jan 2004 55

1
2 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Preface

The following pages contain solutions to problems from qualifying exams in Anal-
ysis given at Rice University dating back to 1999. The solutions are due to the
combined efforts of Amanda Knecht, Eric Samansky, Steve Wallace, and Charlie
Bingham with great aid from Professors Gao, Hardt, and Jones. These solutions
have been compiled, typeset, and are currently being maintained by Charlie Bing-
ham (bingham@math.rice.edu).

These solutions were last updated on March 1, 2004. A good number of mistakes
were remedied. Most were only minor, but a few solutions were complete nonsense.
In addition January 2004 exam solutions were added. Corrections are still welcome.

Onward to 1999, Year of the Rabbit...

 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 3

Jan 1999

Problem 1: Evaluate
Z
(e2πz + 1)−2 dz
|z|=1

where the integral is taken in the counterclockwise direction.

i
Solution 1: The zeros of e2πz + 1 are 2
and − 2i . The possible poles of the integrand
i
function are thus 2
and − 2i . So let’s determine the residues incurred at the two points
i
respectively. Expanding e2πz in its power series about 2
we obtain:
i −4π 2 i i i
(e2πz + 1) = [−1 − 2π(z − ) + (z − )2 + ... + 1] = (z − )(−2π − 2π 2 (z − ) + ...)
2 2 2 2 2
i i i
⇒ (e2πz + 1)−2 = (z − )−2 (−2π − 2π 2 (z − ) + ...)−2 = (z − )−2 P (z)
2 2 2
Note: here we may observe that 2 is a pole of order 2 since P ( 2 ) = 4π1 2 . Now recall
i i

the binomial formula:

X∞ µ ¶
α k
α
(1 + w) = w
k=0
k
So setting w = (π(z − 2i ) + ...) we have the following:
i i
P (z) = (−2π)−2 (1 + π(z − ) + ...)−2 = (−2π)−2 (1 + π(z − ) + · · ·)−2 =
2 2
1 i
2
=(1 − 2π(z − ) + higher order terms)
4π 2
Recall from above that (e 2πz
+ 1) −2
= (z − 2i )−2 P (z) Hence the residue at i
2
is
( 4π1 2 )(−2π) = − 2π
1
. Since all of the above is unaltered for the calculation of the
residue at z = −i
2
, we may conclude:
Z X
(e2πz + 1)−2 dz = 2πi (residues of (e2πz + 1)−2 with |z| < 1) =
|z|=1

1 1
= 2πi(− − ) = −2i
2π 2π
¤
4 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

(Another) Solution 1:

Problem 2: Give an example of a subset of R having uncountably many connected

components. Can such a subset be open? Closed?

Solution 2: An example of such a subset is the Ternary Cantor Set (the Middle
Thirds Cantor Set). This set is the compliment of a union of open intervals, hence it
is closed. So we need only address the second question: Can such a subset be open?
An open subset of R must contain an open interval (a, b). By density of Q in R, each
open interval (a, b) must contain a rational number. Since Q is countable, we can
have at most a countable number of open connected components. Hence a subset
with uncountable connected components can’t be open. ¤

Problem 3: Let f be a meromorphic function on C which is analytic in a neighbor-

hood of 0. Suppose its Taylor Series at 0 is:

∞
X
ak z k
k=0

With ak ≥ 0 . Let r = min{|z0 | : f has a pole at z0 } < ∞. Prove f has a pole at

z = r where r ∈ R .

Solution 3: By elementary arguments and the hypothesis ak ≥ 0 we obtain the

following:

¯ ¯
∞
X ¯X∞ ¯ X ∞
¯ ∞
¯ X
k ¯ k¯ ¯ k¯
(∗) ak z ≤ ¯ ak z ¯ ≤ ak z = ak | z k |
¯ ¯
k=0 k=0 k=0 k=0
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 5

We know we have a minimum modulus pole on |z| = r, say z0 . Since power series
representations diverge at poles, utilizing (∗) we have
∞
X ∞
X ∞
X
∞ = ak z0k ≤ ak | z0k | = ak rk
k=0 k=0 k=0

This implies the power series diverges at z = r as well. Therefore z = r is a pole. ¤

Problem 4: Let f (x) be R-valued, defined ∀x ≥ 1 , satisfying f (1) = 1 and:

f 0 (x) = 1/(x2 + f (x)2 )

Prove limx→∞ f (x) exists and limx→∞ f (x) ≤ 1 + π/4 .

Solution 4: By the Fundamental Theorem of Calculus, we have:

Z x Z x
0
f (x) = f (1) + f (t)dt = 1 + f 0 (t)dt
1 1
Rx
Now, since f 0 (t) > 0 ∀t, the function 1
f 0 (t)dt is increasing in x and hence has a
limit ∀x. So taking a limit on both sides yields:
Z x
lim f (x) = 1 + lim 1/(t2 + f (t)2 )dt
x→∞ x→∞ 1
0
Furthermore, f (t) ≥ 0 ∀t implies:
Z x Z x
2 2
1/(t + f (t) )dt ≤ 1/(t2 + 1)dt
1 1

Z x
⇒ lim f (x) ≤ 1 + lim 1/(t2 + 1)dt = 1 + (arctan (∞) − arctan (1)) = 1 + π/4
x→∞ x→∞ 1
¤

Problem 5: Suppose f : [0, 1] → R is continuous.

R1
a. Prove that limn→∞ 0 xn f (x)dx = 0
R1
b. Prove that limn→∞ n 0 xn f (x)dx = f (1)
6 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Solution 5: (a) Note that f is continuous on [0, 1], and xn is continuous on [0, 1]∀n.
Hence their product is continuous on [0, 1], a compact subset of R. Therefore xn f is
integrable on [0, 1]. Furthermore, since 0 ≤ xn ≤ 1 on [0, 1], xn f is dominated by f .
Thus by Lebesgue’s Dominated Convergence Theorem, we have:
Z 1 Z 1 Z 1
n n
lim x f (x)dx = lim x f (x)dx = 0dx = 0
n→∞ 0 0 n→∞ 0

√
(b) Let u = xn . Then x = n
u and du = nxn−1 . Substitution into the integral yields
(bringing the constant n inside):
Z 1 √ √
( n u)f ( n u)du
0

Note that the domain of integration is not altered by the change of variables. Since
√
composition of continuous functions preserves continuity, f ( n u) is continuous. So by
the same argument as in (a), we can interchange the limit and the integral s.t.
Z 1 Z 1 Z 1
√ √ √ √
lim ( u)f ( u)du =
n n
lim ( u)f ( u)du =
n n
f (1)du = [f (1)u]10 = f (1) ¤
n→∞ 0 0 n→∞ 0

Problem 6: Let α be a complex number and ² a positive number. Prove that the
1
function f (z) = sin z + z−α
has infinitely many zeros in the strip |Imz| < ².

Solution 6: This problem stinks of smelly cheese. So our first instinct should be to
apply Rouche’s Theorem. Since we know where the zeros of sin x lie, we should apply
1
Rouche to f (z) = sin z + z−α
and g(z) = sin z with appropriate contour.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 7

Let Z = {zi } be the set of zeros of f (z). Let {xi } be the subset of Z on the the R
interval [0, 2π]. Since f is non-constant, there are only finitely many, say N zeros on
the interval. So let U = {Ui }N
i=1 be an open cover of the interval [0, 2π] such that each

Ui does not contain any other zero (on or off the R line). Furthermore, there exists a
contour γ0 contained in U such that γ0 is the oriented boundary of a rectangle with
vertices δi, 2π + δi, −δi, 2π − δi shifted slightly to the right or left to avoid zeros on
the vertical edges of both f and g. We can also take δ < ².

γ0 is a compact subset of C hence it achieves its infimum. By construction, f is non-

zero on γ0 . So let m = minγ0 f (z). Now recall that sin z has period 2π. Therefore we
may shift γ0 sufficiently far to the right to some contour γ1 (or left for that matter)
so that we have ¯ ¯
¯ 1 ¯
¯ ¯
¯ z − α¯ ≤ m

on γ1 . We can do this since α is fixed and m is unchanged by 2π shifting. We may

now apply Rouche’s Theorem to obtain that f and g have the same number of roots
inside γ1 . Note that there is at least one zero of sin z inside γ1 since sin x has period
2π. We may obtain the result be repeating this last step on {γk } the set of 2kπ shifts
of γ1 . ¤
8 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Aug 1999

Problem 1: Prove that, for any family F of upper semi-continuous functions on R,

the function

g(x) = inf{f (x)|f ∈ F}

is upper semi-continuous.

Solution 1: f is upper semicontinuous if and only if for any t > f (x) there exists
a neighborhood of x such that t > f (x) in that neighborhood. Let t > g(x). Then
t > f (x) for at least one function f ∈ F. But then that function has a neighborhood
as mentioned above. There we have t > f (y) ≥ g(y). Hence g is upper semicontinu-
ous by the above equivalence. ¤

Problem 2: Suppose that f is a holomorphic function on {z : |z| < 3R}, f (0) = 0,

MR = sup|z|≤R |f (z)|, and NR = sup|z|≤R |f 0 (z)|.
a. Estimate MR (from above) in terms of NR .
b. Estimate NR (from above) in terms of M2R .

Solution 2: (a) Take any contour γ along a radii of the circle of radius R. Let
z0 ∈ |z| = R. Then:
Z
f (z0 ) = f (0) + f 0 (z)dz ⇒ f (z0 ) ≤ 0 + RNR
γ

Since this holds for arbitrary z0 on |z| = R, we have MR ≤ RNR .

(b) Since f is holomorphic on |z| < 3R, Cauchy’s Integral Formula gives us:
Z
1 f (ζ)
f (z0 ) = dζ
2πi γ ζ − z0
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 9

where γ is the boundary of |z| ≤ 2R positively oriented, and z0 on

1
R f (ζ)
|z| = R. So restricting our attention to |z| = 1 we have f (z) = 2πi γ ζ−z
dζ. Differ-
entiating on both sides we have:

Z Z Z
0 d 1 f (ζ) 1 d f (ζ) 1 −f (ζ)
f (z) = dζ = dζ = dζ
dz 2πi γ ζ −z 2πi γ dz ζ − z 2πi γ (ζ − z)2

Note that the second equality above is justified by the fact that the integral is on
a compact subset of the domain of analyticity of an analytic function. Now taking
absolute values on both sides of the equality, we obtain:
¯ Z ¯ Z ¯ ¯ Z
¯ 1 −f (ζ) ¯ 1 ¯ −f (ζ) ¯ 1 |f (ζ)|
|f (z)| = ¯¯
0
dζ ¯ ≤
¯
¯
¯ dζ ¯ =
¯ |dζ| ≤
2πi γ (ζ − z)2 2π γ (ζ − z)2 2π γ |(ζ − z)|2
2π2RM2R 2M2R
≤ 2
=
2π(2R − R) R

Thus we have NR ≤ 2M2R /R. ¤

Problem 3: Suppose that f (x) is defined on [−1, 1], and that f 00 (x) is continuous.
Show that the series
X∞
1 1
(n(f ( ) − f (− )) − 2f 0 (0))
n=1
n n
converges.

Solution 3: Utilizing the Taylor series expansion about 0, we have:

f 00 (0) 2 f 000 (t) 3
f (x) = f (0) + f 0 (0)x + x + x
2 6
0 f 00 (0) 2 f 000 (s) 3
f (−x) = f (0) − f (0)x + x − x
2 6
for some s, t ∈ [−1, 1]. Thus we have the following:
x3 000
f (x) − f (−x) = 2f 0 (0)x + [f (s) + f 000 (t)]
6
10 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Now let x = 1/n to obtain:

2f 0 (0) 1
f (1/n) − f (−1/n) = + 3 [f 000 (s) + f 000 (t)]
n 6n
1
⇒ An := [n(f (1/n)) − f (−1/n)) − 2f 0 (0)] = 2 [f 000 (s) + f 000 (t)]
6n
Since f 000 (x) is continuous on [−1, 1], f 000 (x) is bounded by some M so
M
P∞
[f 000 (s) + f 000 (t)] ≤ 2M . Thus An ≤ 3n 2 ⇒ 1 An converges. ¤

Problem 4: Let f be an analytic function such that f (z) = 1 + z + z 2 + ... for |z| < 1.
Define a sequence of real numbers a0 , a1 , a2 , ... by
X∞
f (z) = an (z + 2)n
n=0
P∞
What is the radius of convergence of this new series n=0 an (z + 2)n ?

1
Solution 4: 1 + z + z 2 + ... = 1/(1 − z) ∀z = n
as n → ∞. So by uniqueness
of power series representations of analytic functions, f (z) = 1/(1 − z) on D. In
fact, f (z) = 1/(1 − z) on the latter’s domain of analyticity, which is C\{1}. Hence
the radius of convergence for the power series representation of f (z) about −2 is
d(−2, 1) = 3. ¤

Problem 5: Suppose that f1 , f2 , ... are nonnegative continuous functions on [0, 1]

R1
with 0 fn (x)dx ≤ M .
a. Show that there exists a point a ∈ [0, 1] with f1 (a) ≤ 2M and f2 (a) ≤ 2M .
b. Does there exist a better estimate? That is, a number N < M so that

inf max{f1 (a), f2 (a)} ≤ 2N

0≤a≤1

for all such f1 , f2 . If so, find the smallest such N . If not, give a counterexample.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 11

c. Show that there always exists an a ∈ [0, 1] so that fn (a) ≤ M

for infinitely many n.

Solution 5
(a) Suppose not. Then f1 (a) + f2 (a) > 2M for all a. Hence we have:
Z 1
[f1 (x) + f2 (x)]dx > 2M
0
R1 R1
⇒ either 0
f1 > M or 0
f2 > M , a contradiction to our hypothesis.

(b) Draw a picture, or: Fix N < M , say M − N = ². So 2M − 2N = 2². Now let:

 (2M − ²) ifx ∈ [0, 21 ]
1
−(2M −²)x (2M −²)( 2 +δ)
f1, ² = + for x ∈ [ 21 , 12 + δ] .
 δ δ
0 otherwise
 1
 (2Mδ−²)x − (2M −²)(
δ
2
−δ)
for x ∈ [ 12 − δ, 12 ]
f2, ² = (2M − ²) forx ∈ [ 12 , 1] .

0 otherwise
Note that f1, ² and f2, ² are are reflections about x = 1/2 of each other. From this
R1 R1
observation we note: 0 f1, ² (x)dx = 0 f2, ² (x)dx. Since the area under these curves
is the area of a rectangle and a right triangle with base; height 21 ; (2M − ²) and δ;
2M − ² respectively, we have the following:
Z 1 Z 1
1 δ 1
f1, ² (x)dx = f2, ² (x)dx = (2M − ²) + (2M − ²) = (2M − ²)(1 + δ)
0 0 2 2 2
We want the above functional restricted to our two functions to be bounded above
by M . i.e. we want 12 (2M − ²)(1 + δ) ≤ M . So we simplify the inequality:
1 1
(2M − ²)(1 + δ) = (2M + 2M δ − ² − ²δ) ≤ M
2 2
⇒ 2M δ − ² − ²δ ≤ 0 ⇒ δ(2M − ²) − ² ≤ 0
Then let δ = ²/(2M −²). We thus have equality above which is sufficient to satisfy the
hypothesis of concern. So substituting the δ value computed above into our definition
of f1, ² and f2, ² yields a pair of counterexample functions for any N < M .
12 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

(c) The statement is false. Consider the following counterexample:


 M (1 − nx) if x ∈ [0, 1/n]
For n ≥ 2 define fn (x) = M (−1 + n2 x) if x ∈ [1/n, 2/n] .

M if x ∈ [2/n, 1]
fn (x)
Now let gn (x) =
1 − n1
R1
Note that ∀n 0
gn (x)dx = M and for each a ∈ [0, 1], gn (a) ≤ M for only
finitely many n. So gn serves as a counterexample to the statement in (c). ¤

Problem 6: Prove that there is no one-to-one conformal map of the punctured disc
G = {z ∈ C : 0 < |z| < 1} onto the annulus A = {z ∈ C : 1 < |z| < 2}.

Solution 6: Suppose there exists such a map, g : D\{0} → A. Then g has an

analytic continuation G on D. G(0) must lie in A (which is the open annulus) since
0 ∈ intD ⇒ G(0) ∈ intĀ by the Open Mapping Theorem. But g onto implies that
there exists some z0 ∈ D\{0} such that g(z0 ) = G(0). Now take open disjoint neigh-
borhoods about z0 and 0, Uz0 and U0 respectively. Push them forward to the range
space. Since G(0) = g(z0 ), G(U0 ) ∩ g(Uz0 ) 6= ∅. Therefore, points in our two domain
neighborhoods share image points. Since G|D\{0} = g, we have a contradiction to
confomality (conformal maps are injective). ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 13

Jan 2000 A

Problem 1: Suppose g(x) = limn→∞ gn (x) for x ∈ [0, 1] where {gn } are positive and
continuous on [0, 1]. Also suppose that:
Z 1
gn dx = 1
0
R1
a. Is it always true that 0
g(x)dx ≤ 1?
R1
b. Is it always true that 0
g(x)dx ≥ 1?

Solution 1: a. Yes. This is a direct application of Fatou’s Lemma:

Z Z
lim inf gn (x)dx ≤ lim inf gn (x)dx
n→∞ n→∞
Z 1
⇒ g(x)dx ≤ 1
0
¤

b. No. Consider the following family of functions on [0, 1]:

 2
 n2 x − ( n2 − n) if x ∈ [ 21 − n1 , 12 ]
2
gn (x) = −n2 x + ( n2 + n) if x ∈ [ 12 , 12 + n1 ] .

0 otherwise
1
The graphs of these functions look like triangles with height n peaks at x = 2
and with bases on [ 21 − 1
n
1
+ n1 ]. The ratio of the base to the height is 2 ∀n but
, 2
R1 R1
limn→∞ fn = δ 1 = 0 almost everywhere. Hence 0 f (x) = 0 δ 1 = 0. ¤
2 2

Problem 2: Let f be C-valued in D(0, 1) such that g = f 2 and h = f 3 are both

analytic. Prove f is analytic in D(0, 1).

Solution 2: Define k = h/g = f 3 /f 2 = f . Now we must be careful to show that

this function k is in fact analytic. The only possible singularities of a quotient of two
14 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

analytic functions is at the zeros of the denominator function. So suppose f 2 (z0 ) = 0,

then f 3 (z0 ) = 0. Let m be the multiplicity of the zero of f 3 at z0 , and n be the
multiplicity of the zero of f 2 at z0 . That is,

f 3 (z) = αm (z − z0 )m + . . . and f 2 (z) = βn (z − z0 )n + . . .

Now (f 3 )2 = f 6 = (f 2 )3 ⇒ 2m = 3n ⇒ m > n. Therefore the limit of k(z) as

z → z0 exists (in fact it is zero). Thus z0 is a removable singularity of k, and so k is
indeed analytic there by the Riemann Removable Singularity Theorem. Since z0 was
arbitrary, we are done. ¤

Problem 3: Construct an open set U ⊂ [0, 1] such that U is dense in [0, 1], the
Lebesgue measure µ(U ) < 1, and that µ(U ∩ (a, b)) > 0 for any interval (a, b) ⊂ [0, 1].

1
Solution 3: Order Q ∩ [0, 1], say {q1 , q2 , ...}. Let Ai be the interval (qi − 2i+2
,
1 1
qi + 2i+2 ). Let Bi = Ai ∩ [0, 1]. Note that Bi is open ∀i and µ(Bi ) ≤ 2i+1 . So we have:
P∞ 1 1 1 1 1 1
1 µ(∪Bi ) ≤ 4 + 8 + ... = 4 /(1 − 2 ) = 2 . Hence µ(∪Bi ) ≤ 2 . Q ⊂ ∪Bi , so for any

open interval (a, b) ⊂ [0, 1], ∃q ∈ Q such that q ∈ (a, b). And by construction, we
have: µ((a, b) ∩ (∪Bi )) > 0. ¤

Problem 4: Let f, g1 , g2 , ... be entire functions. Assume that the kth derivatives at
0 satisfy:
(k)
a. |gn | ≤ |f (k) | for all n and k;
(k)
b. limn→∞ gn (0) exists for all k.
Prove that the sequence {gn } converges uniformly on compact sets and thats its limit
is an entire function.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 15

Solution 4: The {gn } are entire, so they have power series representations at 0:

g1 = α10 + α11 z + α12 z 2 + . . .

g2 = α20 + α21 z + α22 z 2 + . . .
.. .. ..
. . .
gn = αn0 + αn1 z + αn2 z 2 + . . .
.. .. ..
. . .

Also let
f (z) = β0 + β1 z + β2 z 2 + . . .
(k)
be the power series representation of f at 0. Observe that gn (0) = (k!)(αnk )
(k)
and f (k) (0) = (k!)(βk ). By hypothesis (ii), gn (0) converge for each k. There-
fore limn→∞ αnk exists for each k since the sequence is merely a constant multiple
(k!) times the given sequence. Let αi = limn→∞ αni . From our observation above
we have αi ≤ βi . Now we claim that gn (z) converges uniformly on compacta to
P∞ i
g(z) where g(z) = i=0 αi z . That is, given ² > 0 and | z| ≤ R, ∃N such that

n ≥ N ⇒ | g(z) − gn (z)| < ² for all z such that | z| ≤ R. More precisely, restricting
to our compact domain we desire the following:
¯ ∞ ¯
¯X ∞
X ¯
¯ i i¯
n≥N ⇒ ¯ αi z − αin z ¯ < ²
¯ ¯
ı=0 ı=0

And since
¯ ¯ ¯ ¯
¯X∞ ∞
X ¯ ¯X ∞ ¯ X ∞
¯ ¯
¯ i i¯ ¯ i¯ ¯ (αi − αin )z i ¯
¯ αi z − αin z ¯ = ¯ (αi − αin )z ¯ ≤
¯ ¯ ¯ ¯
ı=0 ı=0 ı=0 ı=0

it is sufficient to bound the following by ² for n ≥ N

X∞
| αi − αin | | z|i (∗)
ı=0

So let ² > 0 and R ∈ R be given. Let | z| ≤ R. Now, the power series expansions
of analytic functions converge absolutely in the interior of the function’s domain of
16 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

analyticity. Since f is entire, on | z| ≤ R we have ∃K ∈ N such that the second

inequality holds below.
X∞ ∞
¯ ¯ X ²
¯ βi z i ¯ ≤ | βi | R i <
i=K i=K
4
Substituting the estimate | αi − αin | ≤ 2 | βi | into (∗) we obtain
X∞ X∞
¯ ¯ ¯ ¯
¯ αi z i − αni z i ¯ ≤ 2 ¯ βi z i ¯ < ²
i=K i=K
2
We now have an ²/2-bound for the infinite tail of the sum which is independent of n
and z, so all we need do is bound the first K terms by ²/2 independent of z. Since
the αni converge to the αi ,
²
∃Ni such that n ≥ Ni ⇒ | αi − αni | <
2KRi
K−1
Put N = maxi=0 {Ni }. Then n ≥ N ensures:
K−1
X K−1
X K−1
X ²
i ² i ²
| αi − αni | | z| < i
R = =
i=0 i=0
2KR i=0
2K 2
Combining our two bounds, we may indeed conclude that proved that the limit func-
tion g is entire as well. ¤

Problem 5: Let f be a continuous, strictly increasing function from [0, ∞) onto

[0, ∞) and let g = f −1 be the inverse of f . Prove that:
Z a Z b
f (x)dx + g(x)dx ≥ ab
0 0
for all positive numbers a and b.

Solution 5: Since f is monotone increasing, f −1 is monotone increasing. We have

one of the following f (a) = b, f (a) > b or f (a) < b. However, since (f −1 )−1 = f we
may assume without loss of generality that f (a) ≤ b. Thus we have
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 17

Z Z ÃZ Z ! Z
a b a f (a) b
f (x)dx + g(x)dx = f (x)dx + g(x)dx + g(x)dx =
0 0 0 0 f (a)
ÃZ Z ! µZ Z ¶
a f (a) b b
= f (x)dx + g(x)dx + a dx + (g(x) − a)dx =
0 0 f (a) f (a)
ÃZ Z Z ! Z
a f (a) b b
= f (x)dx + g(x)dx + a dx + (g(x) − a)dx =
0 0 f (a) f (a)
Z b
= (ab) + (g(x) − a)dx ≥ ab
f (a)

Problem 6: Suppose that f (z) is analytic and satisfies f ( z1 ) = f (z) for all z ∈ C\{0}.
a. Write down the general Laurent Expansion for f .
b. Show that the coefficients of this expansion are all real if this f has real values on
the unit circle |z| = 1

Solution 6:
(a) The condition f (z) = f ( z1 ) forces the ith and −ith coefficients to be equal:
∞
X
f (z) = α0 + αi [(z − z0 )i + (z − z0 )−i ]
1

(b) For all k we have the following identity:

Z
1 f (ζ)
(∗) αk = dζ
2πi | ζ|=1 ζ k+1

Since f (z) ∈ R on | z| = 1, trivially f (z) = f (z) there. Also on | z| = 1 we have

1 1 z z
= · = =z
z z z kzk2
18 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Using this fact we obtain the following:

−1
dz = dz
z2
Now conjugating both sides of (∗) and incorporating the above facts yields:
Z Z Z
1 f (ζ) 1 f (ζ) 1 f (ζ) −1
αk = dζ = − dζ = − · dζ =
2πi | ζ|=1 ζ k+1 2πi | ζ|=1 ζ k+1 2πi | ζ|=1 ζ −k−1 ζ 2
Z
1 f (ζ)
= dζ = α−k = αk
2πi | ζ|=1 ζ −k+1
From the above string of equalities, we obtain αk = αk . Hence αk ∈ R for all
k ∈ Z. ¤

Jan 2000 B

Problem 1: Does there exist a function f (z) that is holomorphic near the origin and
that satisfies:
1 1 1
f ( ) = f (− ) = 3 , n = 1, 2, 3, ...?
n n n
Why or why not?

Solution 1: No. By hypothesis, f (z) agrees with z 3 on the set { n1 }∞

1 which has 0

as a limit point. By uniqueness of power series expansions, f (z) = z 3 near zero. But
1 −1
n3
6= n3
. ¤

R∞
Problem 2: Let f : R → R be continuous, with −∞
|f (x)|dx < ∞. Show that there
is a sequence xn ∈ R such that xn → ∞ , xn f (xn ) → 0 and xn f (−xn ) → 0 as n → ∞.

Solution 2: Assume to the contrary that there does not exist such a sequence {xn }.
Then either lim inf x→∞ x | f (x)| > 0 or lim inf x→∞ x | f (−x)| > 0. So either ∃M+ ∈ N
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 19

or ∃M− ∈ N such that

x ≥ M+ ⇒ x | f (x)| ≥ k > 0

⇒ | f (x)| ≥ k/x
or
x ≥ M− ⇒ x | f (−x)| ≥ c > 0

⇒ | f (−x)| ≥ c/x
But then we would have
Z ∞ Z ∞
k
| f (x)| dx ≥ dx = ∞
M+ M+ x
or Z Z
∞ ∞
c
| f (−x)| dx ≥ dx = ∞
M− M− x
contradicting integrability. ¤

Problem 3: Suppose that f (z) = a0 + a1 z + a2 z 2 + ...

and g(z) = b−2 z −2 + b−1 z −1 + b0 + b1 z + b2 z 2 + ...

P P∞ R
where the two series ∞ n
n=0 an z and
n
n=0 bn z converge for |z| < 2. Find Γ f (z)g(z)dz

in terms of the an , bn where Γ is the positively-oriented unit circle.

Solution 3:
f (z)g(z) = a0 b−2 z −2 + a0 b−1 z −1 + a1 b−2 z −1 + ...
Hence by the Residue Formula:
Z
f (z)g(z)dz = 2πi(a0 b−1 + a1 b−2 )
Γ
¤

Problem 4: Suppose fn is a sequence of nonnegative measurable functions on R,

R∞
and limn→∞ −∞ fn2 (x)dx = 0.
20 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

a. Show that the Lebesgue measure λ{x : fn (x) > δ} → 0 as n → ∞ for every δ > 0.
R∞
b. Show that limn→∞ −∞ fn (x)g(x)dx = 0 for every g ∈ L2 (R).
c. Find a sequence gn of nonnegative measurable functions on R such that
R∞
limn→∞ λ{x : gn (x) > δ} = 0 for every δ > 0 AND limn→∞ −∞ gn (x)g(x)dx = 0 for
R∞
every g ∈ L2 (R) BUT −∞ gn2 (x)dx = 1.

Solution 4:
(a) Recall the Chebyshev Inequality for L2 :
kfn k22
λ{x : fn (x) > δ} ≤
δ2
Fix δ > 0 then the RHS → 0 as n → ∞ since kfn k22 → 0 by hypothesis.

(b) Via a basic inequality and Hölder’s Inequality we obtain:

Z ∞
fn (x)g(x)dx ≤ kfn gk1 ≤ kfn k2 kgk2
−∞

Taking a limit as n → ∞ yields:

Z ∞
lim fn (x)g(x)dx ≤ lim kfn k2 kgk2 = 0
n→∞ −∞ n→∞

The last equality follows from the fact that kgk2 < ∞ and the hypothesis on fn .

(c) Define gn : R → R by
1
gn (x) = χ[n,2n] · √ for n = 1, 2, 3, . . .
n
Note that Z ∞
1
gn2 (x)dx = ·n=1
−∞ n
Also note that gn (x)g(x) is dominated by the restriction of g to a compact domain
[n, 2n]. Since L2 ⊂ L1 on compact subsets of R, we obtain the following via Lebesgue’s
Dominated Convergence Theorem (since gn (x)g(x) is dominated by .g(x))
Z ∞ Z ∞ Z ∞
lim gn (x)g(x)dx = lim gn (x)g(x)dx = 0 · g(x)dx = 0 a.e.
n→∞ −∞ −∞ n→∞ −∞
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 21

Problem 5:
a. Prove that if f is a holomorphic map from the unit disk D = {z : |z| < 1} to itself
with f (0) = 0, then |f (z)| ≤ |z| for all z ∈ D.
b. Which of these f admit a point a 6= 0 in D with |f (a)| = |a|?
c. Let h be a holomorphic map of the unit disk D into itself which is not the identity
map of D. Show that h can have at most one fixed point.

Solution 5: Note that parts (a)-(c) are known as the Schwarz Lemma.
(a) Define g : D → D by ½
f (z)/z x 6= 0
g(z) = .
f 0 (0) x = 0
Note that this function is analytic in 0 < | z| < 1 Observe that limz→0 g(z) = f 0 (0) =
g(0). Since f is analytic at 0, f 0 (0) is analytic at 0. Hence our g(z) is analytic in D.
By the maximum principle for analytic functions
| f (z)| 1
| g(z)| ≤ ≤
| z| | z|
1
So | g(z)| is bounded by r
on every closed disk of radius r. Letting r → 1 and applying
the Maximum Principle for analytic functions we obtain the inequality | g(z)| ≤ 1 on
D. Therefore | f (z)| ≤ | z| there.

(b) Suppose there exists such a point. Then g(a) = 1. The Maximum Principle then
implies that | g(z)| is constant in D. Hence f (z) must be a rotation about the origin.
That is, f (z) = kz with | k| = 1.

(c) Suppose h : D → D is not the identity, and suppose h(a) = a and h(b) = b,
a, b ∈ D. Let
a−z
g(z) =
1 − ā0 z
22 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Note that g(a) = 0 and g(0) = a. Furthermore, this mapping is an analytic au-
tomorphism of D such that g −1 = g since | a| < 1. Now consider the composition
f = g ◦ h ◦ g : D → D. After observing that f (0) = 0, we apply the Schwarz Lemma
to obtain | f (z)| ≤ | z|. Furthermore f = g ◦ h ◦ g ⇒ g ◦ h = f ◦ g. Therefore
µ ¶ µ ¶
¡ ¢ ¡ ¢ a−b a−b
f g(b) = g h(b) ⇒ f =
1 − āb 1 − āb
Therefore f satisfies the Schwarz Lemma and has a fixed point other than 0 inside
the disk. Hence f (z) = z. But this in turn implies that h = g ◦ id ◦ g = g ◦ g = id, a
contradiction. ¤

Problem 6: Suppose h(x) = limn→∞ hn (x) where h1 ≥ h2 ≥ h3 ≥ ... is a decreasing

sequence of nonnegative continuous functions on R.
a. Give a specific example of such h and hn where h is not continuous at 0.
b. Prove that, in general, such an h is always upper semi-continuous; that is,

h(a) ≥ lim sup h(x)

x→a

for all a ∈ R.

Solution 6:
(a) Consider the following sequence of functions:
½ nx
e x≤0
hn (x) = .
e−nx x > 0
Note that limn→∞ hn (x) = 0 for x 6= 0 and limn→∞ hn (x) = 1 for x = 0.

(b) Lemma: Let F be any family of upper semicontinuous R → R functions. Then

g(x) = inf{f (x) : f ∈ F} is upper semicontinuous.

Proof of Lemma: f is upper semicontinuous if and only if for any t > f (x) there
exists a neighborhood of x such that t > f (x) in that neighborhood. Let t > g(x).
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 23

Then t > f (x) for at least one function in f ∈ F. But then that function has a
neighborhood as mentioned above. There we have t > f (y) ≥ g(y). Hence g is upper
semicontinuous by the above equivalence.

Now note that hn (x) is a monotone decreasing sequence of bounded continuous func-
tions. Hence inf n→∞ hn (x) = limn→∞ hn (x) = h(x). From our Lemma we have that
h is upper semicontinuous. ¤

Aug 2000

Problem 1: Let f ≥ 0 be a real valued function defined and integrable over a

measurable set E ⊂ R. Prove that given ² > 0, there is a δ > 0, such that
¯Z ¯
¯ ¯
¯ f (x)dx¯ < ²
¯ ¯
A

for every measurable subset A ⊂ E with λ(A) < δ.

Solution 1: Fix ². Since f ∈ L1 (E), there exists a simple function s such that
0 ≤ s ≤ f and:
Z Z
s(t)dt > f (t)dt − ²/2
E E
Now s assumes a finite number of values, so let M = max{s(x) : x ∈ E}. Then
Z Z Z
f (t)dt = s(t)dt + (f − s)(t)dt
A A A
Z
⇒ f (t)dt < M [λ(A)] + ²/2
A
So let δ = ²/2M , and we have our result. ¤
24 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Problem 2: For each of the following conditions, decide if there is a non-constant,

holomorphic function defined on the whole complex plane with the given property.
a.
realf (z) > 0 ∀z

b.
1 + |z|
|f (z)| < ∀z
log(1 + |z|)

c. A function f which has 0 and ∞ as its only asymptotic values. (α ∈ C is an

asymptotic value of f if there is an unbound path γ ⊂ C s.t. limz→∞, z∈γ f (z) = α)

Solution 2:
¯ ¯
(a) Since f (z) is entire, ef (z) is entire. Now ¯ ef (z) ¯ = eref (z) > 1 by hypothesis. Since
¯ ¯
this function does not have a zero, 1/ef (z) is entire as well. But ¯ 1/ef (z) ¯ < 1 thus
Liouville’s Theorem implies 1/ef (z) is constant which in turn implies that f (z) is con-
stant. So there does not exist such a function. ¤

(b) Consider the following:

1 + |z| 1 1
|f (z)| < ∀z ⇒ |f (z)| < + |z| ∀z
log (1 + |z|) log (1 + |z|) log (1 + |z|)

⇒ |f (z)| < 1 + |z| ∀z such that |z| ≥ (e − 1).

So by the Extended Liouville Theorem, f (z) is a polynomial of degree at most 1. But

we can do better. I.e. our first argument gave up a lot of ground on estimating the
growth of |f (z)|. Suppose f (z) = αz + β, for some α, β ∈ C. |αz + β| ≤ |α||z| + |β|.
If we consider z with large enough modulus, then eventually 1/(log (1 + |z|) will be
smaller than both |α| and |β|. Hence for some M,
1 + |z|
|f (z)| > M ⇒ |f (z)| >
log (1 + |z|)
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 25

Therefore, if an entire function satisfies such an inequality, it must be constant. ¤

(c) Claim: ez satisfies the criterion. First, if we take the paths from the origin to
infinity along the positive x-axis and negative x-axis we obtain the asymptotic values
of ∞ and 0 respectively. Now, suppose α ∈ C is some other asymptotic value. Then
| α| = k, k ∈ (0, ∞). This implies that the asymptotic path must tend towards x = ek
and hence the path approaches ∞ in the imaginary direction. But if so |ez | argument
never approaches a single value. Roughly speaking, along this path ez asymptotically
cycles the circle of modulus ek . Therefore α is not really an asymptotic value. (This
argument could be made more precise by a delta-epsilon argument) ¤
Problem 3: Let g(z) be analytic in the right half-plane {z : real(z) > 0}, with
| g(z)| < 1 for all such z. If g(1) = 0, how large can |g(2)| be?

Solution 3: There exists an analytic isomorphism of the right half-plane with the
1−z 1+z
open unit disk, h : RHP → D given by h(z) = −( 1+z ) with inverse h−1 (z) = ( 1−z ).
So consider g(h−1 (z)) : D → D. g(h−1 (0)) = g(1) = 0 so we may apply Schwarz’s
1
Lemma to our composition of analytic functions. But first note: h−1 : 3
7→ 2. Hence:
1
| g(h−1 (z))| ≤ | z| ⇒ | g(2)| ≤
3
¤

Problem 4: Let f be a C 2 function on the real line. Assume f is bounded with

bounded second derivative. Let

A = sup |f (x)| and B = sup |f 00 (x)|

x∈R x∈R

Prove that
√
sup |f 0 (x)| ≤ 2 AB
x∈R
26 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Solution 4: The Mean Value Theorem states that ∀(a, b) ⊂ R, ∃c ∈ (a, b) such that
f (b)−f (a) 2A
f 0 (c) = b−a
. A simple estimate provides f 0 (c) ≤ b−a
. Applying the MVT in an
f 0 (x)−f 0 (c)
identical manner, (∗∗) picking an x ∈ (a, b) we also have f 00 (y) = x−c
for some
y ∈ (x, c) (Assume x < c WLOG). Thus we have the following:
f 0 (x) − f 0 (c)
f 00 (y) = ⇒ f 0 (x) = f 0 (c) + (x − c)f 00 (y) ⇒
x−c
⇒ | f 0 (x)| ≤ | f 0 (c)| + | x − c| | f 00 (y)|
2A
⇒ | f 0 (x)| ≤ + (b − a)B
b−a
Now at (∗∗) if we select x = b−a 2
, we obtain a slightly better estimate:
2A b−a
⇒ | f 0 (x)| ≤
+ B
b−a 2
Setting l = (b − a) and minimizing the RHS of the inequality, we may obtain a best
2A Bl
estimate. I.e. let F (l) = l
+ 2
Then
r
0 −2A B 2A B 2 A A
F (l) = 2 + = 0 ⇒ − 2 = − ⇒ l = 4 ⇒ l = 2
l 2 l 2 B B
+ 00 4A
This critical value of F (l) is indeed a global minimum for R for F (l) = l3
≥ 0 ∀l.
Plugging this value of l into our inequality yields:
√ √
2A B 2B A √
| f 0 (x)| ≤ √ + √ = 2 AB
2 A 2 B
¤

Problem 5: Let f (x, y) be defined on the unit square

S = {(x, y) ∈ R2 : 0 ≤ x ≤ 1 and 0 ≤ y ≤ 1}

and suppose that f has the following properties:

i. For each fixed x, the function f (x, y) is an integrable function of y on the unit
interval.
∂f
ii. The partial derivative ∂x
exists at every point (x, y) ∈ S, and is a bounded func-
tion on S.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 27

Show that:
∂f
a. The partial derivative ∂x
is a measurable function of y for each x.
b. Z Z
1 1
d ∂f
f (x, y)dy = (x, y)dy
dx 0 0 ∂x

Solution 5:
(a) Consider the following function in the variable y (fix x, h):
f (x + h, y) + f (x, y)
Fh (y) =
h
Since h is a constant, Fh is a measurable function by application of (i). Since a limit
∂f
of measurable functions is also measurable, we obtain ∂x
is measurable in y for each
x by observing:
∂f f (x + h, y) + f (x, y)
(x, y) = lim = lim Fh (y)
∂x h→0 h h→0

(b) All of the following equalities are elementary or definitional except the second,
which is an application of Lebesgue’s Dominated Convergence Theorem justified by
hypothesis (ii).
Z 1 Z 1 Z 1
∂f (f (x + h, y) + f (x, y))dy (f (x + h, y) + f (x, y))dy
(x, y)dy ≡ lim = lim =
0 ∂x 0 h→0 h h→0 0 h
R1 Z 1
0
(f (x + h, y) + f (x, y))dy d
= lim ≡ f (x, y)dy
h→0 h dx 0
¤

Problem 6: Suppose that f (z) is a non-constant holomorphic function on a con-

nected open set U ⊂ C. Suppose that V is an open set such that its closure V̄ is a
28 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

compact subset of U , and suppose that |f (z)| is constant, say k, on the boundary of
V . Show that f has at least one zero in V .

Solution 6: By the maximum principle, we have | f (z)| < | z 0 | , for all z ∈ V , z 0 on

∂V . So | f (z)| < | k| on V . Hence f maps V into the open disk of radius k. Now
suppose there is no point z0 of V such that f : z0 7→ 0. Then 1/f is holomorphic on
V as well. By the maximum principle and compactness of domain, it obtains a max-
imum on ∂V . When 1/f is maximized, f is minimized, hence f obtains a minimum
on ∂V . Therefore, f has constant modulus V ⇒ f is constant there ⇒ f is constant
on U , a contradiction. Whence f indeed has a zero in V . ¤

Aug 2001

Problem 1: Suppose f is a positive bounded measurable function on [0, 1] and

Rx
F (x) = 0 f (t)dt for 0 ≤ x ≤ 1.
a. Show that F is continuous on [0, 1].
b. Show that F is differentiable at almost every point in [0, 1].

Solution 1:
(a) ¯Z Z ¯
¯ x y ¯
| F (x) − F (y)| = ¯¯ f (t)dt − f (t)dt ¯¯
0 0
Let ² be given. By hypothesis, |f | ≤ M . Without loss of generality, let y ≤ x. Then
we have
¯Z Z ¯ ¯Z y Z y Z x ¯
¯ x y ¯ ¯ ¯
¯ f (t)dt − f (t)dt ¯¯ = ¯¯ f (t)dt − f (t)dt + f (t)dt ¯¯ =
¯
0 0 0 0 y
¯Z x ¯
¯ ¯
= ¯¯ f (t)dt ¯¯ ≤ M · | x − y |
y
So let δ = ²/2M . ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 29

(b) This is an application of Lebesgue’s Differentiation Theorem. Since f is bounded,

f ∈ L1loc (R). Then we have F 0 (x) = f (x) almost everywhere. ¤
30 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Problem 2: Suppose f : C → C is a holomorphic function, m and n are integers,

and
dn f
(2 + |z|m )−1
dz n
is bounded on C.
a. Prove that f is a polynomial
b. Estimate the degree of f in terms of m and n.

Solution 2:
(a) f is entire so dn f /dz n is entire. Furthermore, by the Extended Louiville Theorem,
dn f
i.e. since ≤ 2M + M | z|m
dz n
dn f /dz n is a polynomial of degree at most m. Integrating backwards, we have f must
be a polynomial.

(b) In (a) we integrate backwards n times. Therefore deg(f ) ≤ m + n. ¤

Problem 3: Suppose f and g are integrable functions on R. Show that the convo-
R∞
lution h(x) = −∞ f (x − y)g(y)dy is integrable with
Z ∞ Z ∞ Z ∞
h(x)dx ≤ f (x)dx g(x)dx
−∞ −∞ −∞

Solution 3: Consider the absolute value of the given product function. By Fubini’s
Theorem for non-negative measurable functions, we have:
Z ∞ Z ∞ Z ∞ Z ∞
dx | f (y)g(x − y)| dy = dy | f (y)g(x − y)| dx
−∞ −∞ −∞ −∞
R∞
Let (| f | ∗ | g|)(x) denote the convolution −∞ | f (y)g(x − y)| dy. Now observe
Z ∞
LHS = (| f | ∗ | g| )(x)dx
−∞
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 31

Z ∞ Z ∞
RHS = | f (y)| dy · | g(x − y)| dx = kf k1 kgk1
−∞ −∞
Since f, g ∈ L1 , (| f | ∗ | g| )(x) exists a.e. and is integrable. Now note the following
inequalities
¯Z ¯ Z
¯ ∞ ¯ ∞
f ∗ g ≤ | f ∗ g| = ¯¯ f (y)g(x − y)dy ¯¯ ≤ | f (y)g(x − y)dy| = | f | ∗ | g|
−∞ −∞

So incorporating all of the above and using the fact that h ∈ L1 ⇐⇒ |h| ∈ L1 , we
obtain kf ∗ gk1 ≤ kf k1 kgk1 ¤

Problem 4: Find the following integral:

Z ∞
dx
0 x4 +1

1
Solution 4: Let f (z) = .
Consider the following contour integral
z 4 +1
Z Z R Z
1 1 1
dz = dx + dz
γR z4 + 1 4
−R x + 1
4
SR z + 1

where γR is the contour that runs from −R to R on the x-axis then over the half
circle in the upper half-plane of radius R, and SR is the half-circle component of the
contour. By the residue formula, we have
Z X
1
(∗) 4
dz = 2πi (residues of f (z) in the UHP)
γR z + 1
¯ ¯
Now since ¯ z41+1 ¯ ≤ | z|1 4 ≤ R14 the following is true:
Z
1
lim 4
dz = 0
R→∞ S z + 1
R

Since (∗) holds for all R, we obtain via the limit R → ∞ that
Z ∞ X
1
4
dx = 2πi (residues of f (z) in the UHP)
−∞ x + 1

So now all we need to do is calculate the residues in the UHP. The resides of f
occur at the singularities of f , which are the zeros of z 4 + 1. Those zeros are
32 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

eπi/4 , e3πi/4 , e5πi/4 , e7πi/4 . Only the first two lie in the UHP. The residues are given
by the following function (evaluated at each singularity)
1 1
=
f 0 (z) 4z 3
Therefore
Z ∞
1 1 1 πi
dx = 2πi( e−3πi/4 + e−πi/4 ) = ( )(e−πi/4 )(1 + e−πi/2 ) =
−∞ x4 +1 4 4 2

πi 1 − i π
=( )( √ )(1 − i) = √
2 2 2
¤

Problem 5:
a. Show that any sequence fn of non-negative integrable functions on [0, 1]
R1
with 0 fn2 dx ≤ n13 must converge to zero almost everywhere.
b. Is there a sequence gn of non-negative integrable functions on [0, 1] satisfying
R1 2
g dx → 0 which does not converge to zero almost everywhere? Explain.
0 n

Solution 5:
(a) The first equality in the following line is a corollary of Lebesgue’s Monotone
Convergence Theorem (using non-negativity).
Z 1 X∞ ∞ Z 1
X ∞
X
2 2 1
( fn (x))dx = fn (x)dx ≤ <∞
0 0 0 0 0
n3

Z 1 X∞ ∞
X
( fn2 (x))dx < ∞ ⇒ fn2 (x) < ∞ a.e. ⇒ fn2 (x) → 0 a.e.
0 0 0

⇒ fn (x) → 0 a.e.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 33

(b) Define gn : R → [0, ∞) by gn (x) = χAn where A1 = [0, 1], A2 = [0, 12 ], A3 =

[ 21 , 1], A4 = [0, 13 ], A5 = [ 13 , 23 ], A6 = [ 32 , 1], A7 = [0, 14 ], ...
Z 1
lim gn (x)dx = lim λ(An ) = 0
n→∞ 0 n→∞

Yet for each point, no matter how much time passes, there will be infinitely many An
that cover it. Hence no point has a limit. Notice that the integral tends very slowly
to 0. This is caused by requiring a covering of [0, 1] for each interval width. ¤

Problem 6: Suppose A = {z ∈ C : 1 < |z| < 2}.

1
a. Is the function z
the unif orm limit of a sequence of polynomials on A. Explain.
1
b. Is the function z−3
the unif orm limit of a sequence of polynomials on A. Explain.

Solution 6:
(a) No. If polynomials converge uniformly for | z| = R then they converge uniformly
1
for all 0 ≤ r ≤ R. But z
blows up at 0 and hence cannot be such a limit.

1
(b) Yes. At z0 = 0, z−3
has a radius of convergence r = 3, the distance to its singu-
larity z = 3. Since the annulus is inside its radius of convergence, the power series for
P
1
z−3
converges absolutely and uniformly on A. Thus let pn (z) = n0 an z n , the partial
sums of the power seires. ¤
34 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Jan 2002

Problem 1: Let Ω = C\{R− }c . Is there a non-constant bounded holomorphic func-

tion on Ω? Justify your answer.

Solution 1: Ω is star-convex (take −1 as the star center) hence simply connected.

Thus the Riemann Mapping Theorem applies to Ω. I.e. Ω is analytically isomorphic
to the unit disk. Therefore there exists an analytic map h : Ω → D. ¤

Problem 2: A function f : R → R is convex if

f ((1 − t)a + tb) ≤ (1 − t)f (a) + tf (b) ∀a, b ∈ R and 0 ≤ t ≤ 1.
a. Prove that a convex function is continuous
b. Prove that a twice differentiable function f is convex if f 00 ≥ 0.

Solution 2:
(a) The convexity condition above is equivalent to
f (b) − f (a) f (c) − f (b)
a<b<c ⇒ ≤
b−a c−b
So let x ∈ R be given, and choose s < l < x < r < t. Then our equivalent convexity
condition implies
f (l) − f (s) f (x) − f (l) f (t) − f (x)
≤ ≤
l−s x−l t−x
and
f (x) − f (s) f (r) − f (x) f (t) − f (r)
≤ ≤
x−s r−x t−r
Therefore
¯ ¯
¯ f (t) − f (x) ¯
either | f (x) − f (l)| ≤ | x − l| ¯¯ ¯ (∗)
t−x ¯
¯ ¯
¯ f (l) − f (s) ¯
or | f (x) − f (l)| ≤ | x − l| ¯¯ ¯ (∗∗)
l−s ¯
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 35

and ¯ ¯
¯ f (t) − f (r) ¯
either | f (r) − f (x)| ≤ | r − x| ¯ ¯ ¯ (∗ ∗ ∗)
t−r ¯
¯ ¯
¯ f (x) − f (s) ¯
or | f (r) − f (x)| ≤ | r − x| ¯¯ ¯ (∗ ∗ ∗∗)
x−s ¯
(*) and (****) are independent of l and r respectively thus left and right continuity
holds in those cases (let l → x and r → x). Using the following estimates we similarly
have left and right continuity for (**) and (***) respectively
½ ¯ ¯¾
¯ f (t) − f (x) ¯
| f (l)| ≤ max | f (s)| , ¯¯ f (x) + s ¯
t−x ¯
½ ¯ ¯¾
¯ f (s) − f (x) ¯
| f (r)| ≤ max | f (t)| , ¯¯ f (x) + t ¯
s−x ¯

(b) DO THIS PART

Problem 3: If f is a nowhere zero, entire holomorphic function, is there necessarily

an entire holomorphic function g such that eg = f . Prove your answer.

Solution 3: Define Z z
f 0 (t)
h(z) ≡ dt
z0 f (t)
Hence h0 (z) = f 0 (z)/f (z). Thus we have
µ ¶0
f ¡ −h ¢0
h
= fe = f 0 e−h + (−h0 )f e−h = e−h (f 0 − f 0 ) = 0
e
And so f /eh is constant, which implies eh = αf . So let g ≡ h + ln α. ¤

Problem 4:
a. Prove that if fn is a sequence of Lebesgue measurable functions on [0, 1] with
36 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

R1 R1
limn→∞ 0
fn2 dx = 0, then limn→∞ 0
|fn |dx = 0.
b. Does it also follow that limn→∞ fn (x) = 0 for a.e. x ∈ [0, 1]?
c. Is the implication in (a) still true if [0, 1] is replaced by the whole real line R?

Solution 4:
R1 R1 1
(a) limn→∞ 0 fn2 dx = 0 ⇒ limn→∞ ( 0 fn2 dx) 2 = 0. I.e. limn→∞ ||f || 2 = 0. Since
||fn ||1 ≤ ||fn || 2 on the unit interval, we have limn→∞ ||f ||1 = 0. ¤

(b) No. Consider the following family of sets, {Ai }∞

i=1 where

A1 = [0, 1], A2 = [0, 21 ], A3 = [ 21 , 1], A4 = [0, 31 ], A5 = [ 13 , 23 ], A6 = [ 23 , 1], ...

R1
Let fn = χAn . Then we have limn→∞ 0 |fn (x)|dx = 0 but limn→∞ fn (x) does not
exist for any x ∈ [0, 1]. ¤

(c) The conclusion does not extend to R. Consider the following counterexample.
µ ¶
1
fn (x) = χ[0,n] ·
n
Note that µ ¶
Z ∞
1 1
lim fn2 (x)dx = lim (n) = lim =0
n→∞ 0 n→∞ n2 n→∞ n

while
Z ∞
lim | fn (x)| dx = 1
n→∞ 0
¤

Problem 5: Does there exist an analytic function mapping the annulus

A = {z : 1 ≤ |z| ≤ 4} onto B = {z : 1 ≤ |z| ≤ 2}

and taking C1 → C1 , C4 → C2 , where Cr is the circle of radius r? Why or why not?

 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 37

Solution 5: Define g(z) = z 2 . Consider g : B → A. g : C1 7→ C1 and g : C2 7→ C4 .

Then:
¯ ¯ ¯ ¯
¯ (g ◦ f )(z) ¯ ¯ z ¯
¯ ¯=1=¯ ¯
¯ z ¯ ¯ (g ◦ f )(z) ¯
But then the maximum modulus principle implies that g ◦ f = c, a constant with
√ √
| c| = 1. In turn this implies that f (z) = c z. But z is not a surjective mapping
√ √
from A to B for z : A ,→ B ∩ U HP . Since | c| = 1 ⇒ c z is some rotation of the
half annulus, we cannot possibily find a surjective map from A to B. ¤

Problem 6: Suppose that f ∈ L1 (R). Find a necessary and sufficient condition such
that
||f + tg||1 − ||f ||1
lim
t→0 t
exists for all g ∈ L1 (R).

Solution 6: (Missing)

Jan 2003

Problem 1: How many roots of the equation f (z) = z 4 + 8z 3 + 3z 2 + 8z + 3 = 0 lie

in the right half plane? Hint: Apply the argument principle to a large half disk.

Solution 1: Let γR be the contour which travels along the circle of radius R in the
right half-plane and then from iR to −iR on the imaginary axis. On the half-circle,
i.e. z = Reit for − π2 ≤ t ≤ π2 , we have:
8 3 8 3
f (z) = f (Reit ) = R4 e4it (1 + + + + ) = R4 e4it (1 + ζ)
Reit R2 e2it R3 e3it R4 e4it
38 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

where |ζ| ≤ 22/R < ² for any ² for R large enough. So the argument of f (Reit ) is
approximately the argument of e4it . As we move along the half-circle, t goes from − π2
to π2 , thus the argument of f (z) changes by approximately 4π along this part of the
contour. Now lets consider the argument of f (z) along the imaginary axis section of
γ. I.e. z = iy where y ∈ [−R, R]. So f (iy) = y 4 − 8iy 3 − 3y 2 + 8iy + 3. Hence:

ref (iy) = y 4 − 3y 2 + 3 and imf (iy) = −8y 3 + 8y

So for R large enough, f (iR)’s real part (quartic) dominates its imaginary part (cu-
bic). Hence for R large enough the change in argument on the vertical segment is
approximately zero. Therefore by the Argument Principle, the number of roots in
1
the RHP is 2π
(4π + 0) = 2. ¤

Problem 2: Suppose that f and g are entire functions, and |f (z)| ≤ |g(z)| for all z
with |z| ≥ 1000. Show that f /g is a rational function.

Solution 2: g is entire but g may have some zeros. But there can only be finitely
many. For if there are infinitely many in Ω = {z : |z| ≤ 1000}, the set of zeros has a
limit point there. So then g = 0 on Ω. Hence g = 0 on C which implies f = 0 on C.
So let {α}N
1 be the set of roots of g in Ω. Let q(z) = (z − α1 )(z − α2 )...(z − αN ). Then

qf /g is entire. Furthermore, by hypothesis | qf /g| ≤ | q| for all z such that |z| ≥ 1000.
The Extended Liouville Theorem implies qf /g is a polynomial, p of degree at most
N . Therefore f /g = p/q is a rational function. ¤

Problem 3: Suppose that f is a nonvanishing holomorphic function on the punc-

tured disk Ω ≡ {z : 0 < |z| < 2}, and limz→0 |f (z)| = +∞.
a. Prove that limz→0 z N f (z) = α for some positive integer N and nonzero complex
number α.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 39

b. Find a formula for α in terms of an integral over the unit circle of some expression
in terms of f and N .
c. Find a formula for N in terms of an integral over the unit circle of some expression
in terms of f and f 0 .

Solution 3:
(a) Consider g(z) ≡ 1/f (z). Since f is non-vanishing on Ω, g : Ω → C is holomorphic.
Furthermore by hypothesis limz→0 = 0. Hence 0 is a removable singularity for g. So
g may be extended to a holomorphic function G : Ω ∪ {0} → C given by:
½
g(z) on Ω
G(z) = .
0 at 0
Hence for some unique positive n we have that G(z) = z N · Q(z) where Q is holo-
morphic on Ω ∪ {0} and Q(0) 6= 0. G(z) is non-vanishing on Ω implies that Q(z) is
non-vanishing on Ω ∪ {0}. It thus follows that 1/Q(z) is a holomorphic function on
Ω ∪ {0}. Therefore on Ω ∪ {0}
∞
X
1/Q(z) = βj z j
j=0

For 0 < | z| < 2

1
f (z) =
G(z)
1
= z −N ·
Q(z)
̰ !
X
= z −N · βj z j
j=0
∞
X
= βj+N z j
j=−N

By uniqueness of Laurent expansions, this series agrees with f on Ω so it must indeed

be it’s Laurent expansion. Our desired result follows by observing that
à ∞
!
¡ N ¢ X
lim z · f (z) = lim z N · βj+N z j = β0
z→0 z→0
j=−N
40 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Note: This proof was tailored from the development in Function Theory of One Com-
plex Variable by Greene & Krantz. Their development of meromorphic functions is
excellent.

(b) Let γ be the curve that traverses the unit circle once in the positive orientation.
Cauchy’s Integral Formula states
Z
f (z)dz = 2πi · Res 0 (f )
γ

Recall that Res 0 (f ) is defined as the i = −1 coefficient in the Laurent expansion at

0 for f . So to find α in the statement (β0 in the proof of part (a)), we need only take
the following integral
Z
1
z N −1 f (z)dz
2πi γ

(c) The following is a consequence of the Reside Theorem. Assume f is meromorphic

in U , with finite number of zeros and poles there. Let a1 , a2 , . . . , ar be the zeros in
U and b1 , b2 , . . . , bn be the poles in U . Also assume that γ is a closed 0-homologous
curve in U , then
Z Ã r n
!
f0 X X
= 2πi mj ordaj f − mk mulbk f
γ f j=1 k=1

where mi is the winding number relative γ at each point ai (likewise for the poles).
In this case, we have that inside D(0, 2) f has no zeros (non-vanishing), has precisely
one singularity at 0, and γ winds about 0 once. Hence the formula above becomes:
Z 0
f
= 2πi(− mul 0 f )
γ f

Solving for mul 0 f we obtain

Z
−1 f0
N = mul 0 f =
2πi γ f
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 41

Problem 4: Calculate
Z 1/n
n
lim dx
n→∞ 0 1+ n 2 x2 + n 6 x8
Show all your work and justify all steps.

Solution 4: Let y = nx, then dy = ndx so

Z 1/n Z 1
n 1
A = lim 2 2 6 8
dx = lim 1 8 dy
n→∞ 0 1+n x +n x n→∞ 0 1 + y 2 +
n2
y

1 1
1 8 is dominated by 2
∈ L1 ([0, 1])
1+ y2 + n2
y 1+y
So Lebesgue’s Dominated Convergence Theorem implies
Z 1
1 π π
A= lim 2
dy = − 0 =
0 n→∞ 1 + y 4 4
¤

Problem 5: Suppose f is a nonnegative measurable function defined on Rk . Let

R
a(n) = Rk f n , for n = 1, 2, 3, ...
P
a. Suppose that the infinite series a(n) converges. Show that f < 1 almost every-
where and that f /(1 − f ) is in L1 .
b. Prove the converse.

Solution 5:
(a) First suppose that f ≥ 1 on a set B ⊂ Rk with λ(B) = ². Then
X∞ ∞ Z
X X∞ Z X∞
n n
a(n) = f ≥ f ≥ ²=∞
n=1 n=1 Rk n=1 B n=1
42 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Which is a contradiction to the hypothesis. Hence f < 1 a.e.. Using this result we
obtain that ∞
f X
= fn
1−f n=1
Since f is non-negative, as a consequence of Lebesgue’s Monotone Convergence The-
orem we have
Z Z ∞
X ∞ Z
X ∞
X
f n n
= f = f = a(n) < ∞
Rk 1−f Rk n=1 n=1 Rk n=1
1
Therefore f /(1 − f ) ∈ L .

(b) The above arguments are reversible, i.e.

Z Z X ∞ X∞ Z X∞
f n n
∞> = f = f = a(n)
Rk 1 − f Rk n=1 n=1 R
k
n=1
¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 43

Problem 6:
a. Suppose f : R → R is a smooth function with f (0) = 0, f (−x) = f (x), and
|f 00 (x)| ≤ 3 for all x ∈ R. Find the best bound M so that |f (x)| ≤ M for all
x ∈ [−2, 2].
b. Suppose that fn is a sequence of functions satisfying all the conditions on f from
(a). Prove that fn contains at least one subsequence that converges uniformly on
every bounded subset of R to a continuous function.

Solution 6:
(a) First note that the hypothesis f (−x) = f (x) and f (0) = 0 implies that f 0 (0) = 0.
Now by the Fundamental Theorem of Calculus we have
Z x Z x Z xµ Z t ¶
0 0 0 00
f (x) = f (0) + f (t)dt = f (t)dt = f (0) + f (s)ds dt
0 0 0 0
Z x µZ t ¶
00
= f (s)ds dt
0 0
Taking absolute value on both sides we obtain:
Z x µZ t ¶ Z x µZ t ¶ Z x
00 3x2
| f (x)| ≤ | f (s)ds| dt = 3ds dt = 3tdt =
0 0 0 0 0 2
Hence any function satisfying the hypothesis is absolutely bounded by 6 on [−2, 2].
This bound is shown to be a best estimate by the function 3x2 .

(b) Let F be a family of functions that satisfy the hypothesis, and suppose f ∈ F.
On any interval [−R, R] we have the following
¯Z y Z x ¯ ¯Z y ¯ Z y
¯ ¯ ¯ ¯
| f (y) − f (x)| = ¯¯ f 0 (t)dt − f 0 (t)dt¯¯ = ¯¯ f 0 (t)dt¯¯ ≤ | f 0 (t)| dt ≤
0 0 x x
Z y
≤ 3Rdt = 3R | y − x|
x
The above inequlality implies that F is equicontinuous. For given an ² > 0 we may set
δ = ²/6R to guarantee that | y − x| < δ ⇒ | f (y) − f (x)| < ² for all f ∈ F. Hence
the Arzela-Ascoli Theorem implies that on every interval [−R, R], F has a uniformly
convergent subsequence (which converges to a continous function). We may find a
44 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

convergent subsequence on R via the following diagonal argument. Let Fn1 be the
convergent subsequence on [−1, 1], Fn2 be the convergent subsequence on [−2, 2], . .
. , Fnk be the convergent subsequence on [−k, k], . . . Now define a new subsequence
Fn by setting Fi = Fii . From the construction it is clear that Fn converges uniformly
on all bounded subsets of R. ¤

May 2003

Problem 1: Suppose that x1 > 0 and xn+1 = (2 + xn )−1 for n = 1, 2, 3, ... . Prove
that the sequence xn converges and find its limit.

Solution 1: First we may find the fixed points for the mapping by setting the
following equality

xn = xn+1 ⇒ xn = (2 + xn )−1 ⇒ x2n + 2xn − 1 = 0

√
So we obtain that the fixed points for the map are −1 + − 2. Since x1 > 0 and the
√
map preserves positivity, the possible limit point is the fixed point −1 + 2. We need
only prove that the map indeed converges. Let M (xn ) = (2 + xn )−1 . Now observe:
¯ ¯
¯ 1 1 ¯¯
¯
| M (xn+1 ) − M (xn )| = ¯ − =
2 + xn+1 2 + xn ¯
¯ ¯
¯ xn − xn+1 ¯ 1
=¯¯ ¯ ≤ | xn+1 − xn |
4 + 2xn + 2xn+1 + xn xn+1 ¯ 4

The inequality follows from the fact that xn , xn+1 are positive ∀n (since x1 > 0). So
now we invoke Banach’s Fixed Point Theorem to obtain that iteration of the mapping
does converge to our fixed point. ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 45

Problem 2: Evaluate Z ∞
log x
dx
0 x2 + a 2
where a > 0.

Solution 2: Cr as usual denotes the circle of radius R. Let γ1 be the contour that
traverses CR from z = R to z = −R in the positive orientation (counter-clockwise).
Let γ2 be the contour that traverses Cδ from z = −δ to z = δ in the negative orienta-
1
tion. Suppose at all stages to come that δ < 2
and R > 2. Let γ3 be the straight-line
contour on the x-axis from δ to R, and γ4 be the straight-line contour on the x-axis
from −R to −δ.

Now consider the contour integral

Z Z Z Z Z
= + + +
γ γ1 γ4 γ2 γ3

By the Residue Theorem, we know that

Z X
f (z)dz = 2πi (residues of f in the U HP )
γ

Now put
log z
f (z) =
+ a2 z2
R
First we show that γ1 f (z)dz −→ 0 as R → ∞
Z Z π Z π
log Reit it log R + it
f (z)dz = 2 2it 2
iRe dt = a2
idt
0 R e +a it
γ1 0 Re + Reit

Note that the RHS tends to 0 as R approaches ∞.

R
Next we will show that γ2 f (z)dz −→ 0 as δ → 0.
Z Z π Z π ¡ ¢
log (δeit ) it (log δ) δeit + δeit (it)
f (z)dz = 2 2it + a2
iδe dt = idt
γ2 0 δ e 0 δ 2 e2it + a2
Z π ¡ ¢ Z π ¡ it ¢
(δ log δ) eit + iteit e + iteit
= idt = (δ log δ) idt
0 δ 2 e2it + a2 0 δ 2 e2it + a2
46 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

R
Therefore limδ→0 (δ log δ) = 0 ⇒ γ2
f (z)dz = 0.

Since the Residue Theorem result holds for all δ, R, taking the respective limits as we
did above, we obtain that
Z µZ −δ Z ∞ ¶
log (−x) + iπ log (x)
f (z)dz = lim dx + dx
γ δ→0 −∞ x2 + a 2 δ x2 + a2
Z ∞ Z ∞
log (x) iπ
=2 2 2
dx + dx
0 x +a 0 x + a2
2

R
Substituting 2πi(Resf (ai)) for γ f (z)dz we obtain that
Z ∞ µ Z ∞ ¶
log (x) 1 iπ
(∗) 2 2
dx = 2πi(Resf (ai)) − dx
0 x +a 2 0 x + a2
2

We have
Z ∞ Z
iπ iπ ∞
1 iπ ¡ ¡ ¢¢ ¯¯∞ iπ π iπ 2
2 2
dx = 2 ¡ x ¢2 dx = arctan x/a ¯ = −0=
0 x +a a 0 +1 a 0 a 2 a
a

And
µ ¶ ¯¯
1 ¯ log a + iπ
Resf (ai) = log (z) · ¯ =
2z ¯ 2ai
z=ai

Substituting these two values into (∗) we have

Z ∞ µ ¶
log (x) 1 π log a + iπ 2 iπ 2 π log a
dx = − =
0 x2 + a2 2 a a 2a

R∞
Problem 3: Suppose f : R → R is differentiable and −∞
|f (x)|dx < ∞.
a. Show that the Lebesgue measure of {x : |f (x)| > t} approaches 0 as t → ∞.
R∞
b. Show that the additional assumption −∞ |f 0 (x)|dx < ∞ implies that f (x) → 0 as
x → ∞.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 47

Solution 3:
(a) Recall the Chebyshev Inequality for L1 (R):
kf k1
λ{x : |f (x)| > t} ≤
t
Taking the limit as t → ∞ we obtain the desired inequality.

(b) The additional hypothesis implies that f 0 (x) → 0 as x → ∞. Since f is

differentiable, the Fundamental Theorem of Calculus gives us: f (x) = f (x − 1) +
Rx
x−1
f 0 (t)dt. Thus we obtain:
¯Z x ¯ Z x
¯ ¯
¯
|f (x)| ≤ |f (x − 1)| + ¯ 0 ¯
f (t)dt¯ ≤ |f (x − 1)| + | f 0 (t)| dt
x−1 x−1

Now taking a limit yields the result:

Z x
lim |f (x)| ≤ lim (|f (x − 1)| + | f 0 (t)| dt) = 0 + 0 · 1 = 0
x→∞ x→∞ x−1

Note: the last equality follows from the application of Lebesgue’s Dominated Conver-
gence Theorem. We can apply it here because f 0 (t) is a continuous function integrated
over a compact domain. One could alternatively view the limit as x → ∞ on the inte-
gral as a limit of a family of continuous functions defined on [0, 1]. I.e. gn : [0, 1] → R
given by gn (x) = f 0 (x + (n − 1)). The application of LDCT from this viewpoint is
more easily justified. ¤

2
Problem 4: Find all entire functions f which satisfy Re(f (z)) ≤ |z|
for |z| > 1.
Prove your answer.

¯ ¯ ¯ ¯
Solution 4: Since f is entire, ef is entire. Furthermore, ¯ ef (z) ¯ = ¯ eRef (z) ¯ ≤ e2 for
|z| ≥ 1. The maximum modulus principle implies that ef (z) ≤ e2 on D(0, 1). So ef is
entire and bounded on C. By Liouville’s Theorem ef is constant which implies f is
constant. ¤
48 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

R∞
Problem 5: Suppose that f is a bounded measurable function on R and −∞
|g(x)|dx <
∞. Prove that Z
lim f (x)[g(x + t) − g(x)]dx = 0
t→0

Solution 5: Suppose f is bounded on R by M , we may write:

Z ¯Z ¯ Z
¯ ¯
¯
f (x)[g(x+t)−g(x)]dx ≤ ¯ f (x)[g(x + t) − g(x)]dx¯¯ ≤ |f (x)[g(x+t)−g(x)]|dx ≤
Z
≤ M |[g(x + t) − g(x)]|dx
Now since g ∈ L1 by hypothesis, we have continuity of translation. I.e. if we take the
limits on both sides of the string of inequalities above, we obtain:
Z Z
lim f (x)[g(x + t) − g(x)]dx ≤ lim M | [g(x + t) − g(x)]| dx =
t→0 t→0
Z
= M lim | [g(x + t) − g(x)]| dx = M · 0 = 0
t→0
¤

Problem 6:
a. State Rouche’s Theorem.
b. State Schwarz’s Lemma.
c. Suppose f is holomorphic in the unit disk |z| < 1 with |f (z)| ≤ 1 and f (0) = 0.
Prove that for any integer n ≥ 1, f (z) − 2n z n has precisely n zeros (counting multi-
plicity) in the disk |z| < 21 .

Solution 6:
(a) Rouche’s Theorem: Let γ be a closed path homologous to 0 in U and assume that
γ has an interior. Let h, g be analytic on U , and | h(z) − g(z)| < | h(z)| for z on γ.
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 49

Then h and g have the same number of zeros in the interior of γ.

(b) Schwarz’s Lemma: If f (z) is analytic for | z| < 1 and satisfies the conditions
| f (z)| ≤ 1, f (0) = 0, then | f (z)| ≤ | z| and | f 0 (0)| ≤ 1.

1
(c) Let γ be the boundary of | z| < 2
oriented in the positive direction. Let h(z) =
−2n z n , g(z) = f (z) − 2n z n . Then | h(z) − g(z)| = | f (z)|. Applying the Schwarz
1 1
Lemma we obtain that | f (z)| ≤ 2
on | z| = 2
. Since | h(z)| = 2n−1 ≥ 1 for
n = 1, 2, 3, . . . , Rouche’s Theorem implies that h(z), g(z) have the same number
of zeros inside γ. Therefore, f (z) − 2n z n has precisely n zeros inside | z| = 21 . ¤

Aug 2003

Note: the solutions to this exam’s problems were posted by Professor Hardt following
the exam. Permission has been given to reproduce his solutions here. Any mistakes
below are certainly due to error in reproduction.

Problem 1: Suppose 0 < α < 2.

R ∞ xα
a. The principal value integral of 0 x−x 3 dx is

Z 1−² Z ∞
xα xα
lim dx + dx
²→0 0 x − x3 1−² x − x
3

Why is this definition necessary?

b. Compute this integral.
c. Show that the answer you obtained in (b) is in agreement with the change of
1
variables x = y
in the integral.
50 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Solution 1:
(a) Here it is necessary to consider a limit (called an improper integral) because the
xα
function x−x3
is not integrable near x = 1. It is locally integrable near 0 and ∞.
The choice of 1 − ² and 1 + ² for approximating integration limits is the simplest (the
principal value) as opposed to other choices such as 1 − ², 1 + ²2 .

(b) Here one can apply the Cauchy integral formula to the principal branch f (z)
z α
of − z(z−1)(z+1) on the domain Ω² in the upper half plane bounded by the 4 oriented
intervals [− 1² , −1 − ²], [−1 + ², −²], [−², ²], [1 + ², 1² ], the upper counterclockwise ori-
ented semicricle C1/² of radius 1² , and 3 clockwise oreiented upper semicircles D²−1 ,
D²0 , D²+1 of radius ² centered at −1, 0, 1 respectively. Ae ² → 0, the integral over the
³ ´
1 1 1+α−3
big semicricle C1/² approaches 0 because there |f (z)|( ² ) ≤ C · ² → 0. The
integrals over D²+1 and D²−1 approach −πi times the residues of f at +1 and −1,
xα
which are − 12 and − 12 eπiα . On the positive X-axis f (z) = x−x3
and on the negative
eπiα |x|α
X-axis f (z) = x−x3
. Changing variables x 7→ −x for the integrals on the negative
X-axis now gives
Z Z 1−² Z 1/²
πi πiα πiα xα
0= f (z)dz = o(²) + (1 + e ) + (1 − e )( + ) 3
dx
Ω² 2 ² 1+² x − x

Letting ² → 0 the principal value integral equals

Z ∞
xα πi eπiα + 1 π πα
3
dx = ( ) πiα−1 = cot ( )
0 x−x 2 e 2 2
¤
(c) Consider the following
Z ∞ Z 0 −α Z ∞ −α
xα y −dy y
I(α) = dx = 1 1 = dy =
0 x − x3 ∞ y − y3 y
2
0 y − y1
Z ∞ 2−α Z ∞ 2−α
y y
= 3
dy = − dy = −I(2 − α)
0 y −y 0 y − y3
and cot π(2−α)
2
= cot (π − πα
2
) = − cot ( πα
2
) agrees. ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 51

Problem 2: Suppose that f (x, y) is continuous on the plane and that there is a finite
M so that |f (x, y) − f (x, z)| ≤ M |y − z| for all x, y, z ∈ R.
a. Show that, for any x ∈ R, the partial derivative ∂f ∂y
(x, y) exists for almost all y ∈ R.
R 1 R 1
b. Prove that dyd
0
f (x, y)dx = 0 ∂f∂y
(x, y)dx.
R y 2
d
c. Express dy 0
f (x, y)dx in terms of integrals of f and ∂f∂y
.

Solution 2:
(a) For any x ∈ R, the function f (x, ·) is Lipschitz, hence absolutely continuous. So
∂f
the partial derivative ∂y
(x, y) exists for almost all y ∈ R.

(b) For each y ∈ R, x ∈ [0, 1], and sequence ²i → 0 let

f (x, y + ²i ) − f (x)
gi (x, y) =
²i
Then | gi (x, y)| ≤ M for all i. So Lebesgue’s Dominated Convergence Theorem implies
that Z Z Z
1 1 1
lim ²−1
i [ f (x, y + ²i )dx − f (x, y)dx] = lim gi (x, y)dx =
i→∞ 0 0 i→∞ 0
Z 1 Z 1
∂f
= lim gi (x, y)dx =
(x, y)dx
0 ∂y
0 i→∞
Since the RHS is independent of the sequence ²i → 0, one finds that the derivative
d
R1
dy 0
f (x, y)dx exists and equals the RHS.

R y2
(c) Express dy d
0
f (x, y)dx in terms of integrals of f and ∂f ∂y
. Letting F (s, t) =
Rs
0
f (x, t)dx, we see from the Fundamental Theorem and (b) that
Z s
∂F ∂F ∂f
(s, t) = f (s, t) and (s, t) = (x, t)dx
∂s ∂t 0 ∂y
So we use the chain rule to compute
Z y2
d d ∂F 2 ∂y 2 ∂F 2 ∂y
f (x, y)dx = F (y 2 , y) = (y , y) + (y , y)
dy 0 dy ∂s ∂y ∂t ∂y
Z y2
2 ∂f
= 2yf (y , y) + (x, y)dx ¤
0 ∂y
52 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Problem 3:
a. Show that the direct analog of Rolle’s theorem does not apply to holomorphic
functions. Do this by exhibiting an entire holomorphic function f such that
f (0) = f (1) and yet f 0 (z) never takes the value 0.
b. Suppose f is a holomorphic function on the unit disk {|z| < 1}. Show that f
must be constant if f (ai ) = f (bi ) for two sequences ai , bi of positive real numbers
that satisfy the inequalities:

0 < ... < ai+1 < bi+1 < ai < bi < ... < a1 < b1 < 1

Solution 3:
(a) ez doesn’t vanish and ez+2πi = ez . So we can rotate the domain by 90o and rescale
by letting f (z) = e2πiz . Then f (0) = 1 = f (1) and f 0 (z) = 2πie2πiz 6= 0.

(b) Both monotone sequences converge to some some real number c with 0 ≤ c < 1.
Writing f = u + iv we find from Rolle’s Theorem, points ai < ci < bi so that
∂u
(c )
∂x i
= 0. Since ci → c, we deduce from the real analyticity of u(·, 0) ≡ 0 and so u
is constant on the X-axis. Similarly v is also constant on the X-axis. But then the
holomorphic function f being constant on the X-axis, must itself be constant. ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 53

Problem 4: Suppose 0 < M < ∞ and, for each positive integer j,

fj : [0, 1] → [−M, M ] is a monotone increasing function. Prove that there is a subse-
quence fj 0 and a countable subset A of [0, 1] so that fj 0 (t) converges, as j 0 → ∞, for
every t ∈ [0, 1]\A.

Solution 4: Suppose Q∩[0, 1] = {a1 , a2 , . . . }. A subsequence fα1 (1) (a1 ), fα1 (2) (a1 ), . . .
of the bounded sequence of numbers f1 (a1 ), f2 (a1 ), . . . converges to a number f (a1 ).
Inductively, choose a subsequence fαj (1) (aj ), fαj (2) (aj ), . . . of the sequence
fαj−1 (1) (aj ), fαj−2 (2) (aj ), . . . convergent to a number f (aj ).
Let j 0 = αj (j) and f (x) = supai <x f (ai ) = lim²→0 supx−²<ai <x f (ai ). Then f is mono-
tone increasing and the set A of discontinuities of f is at most countable. To see that
limj→∞ fj 0 (x) = f (x) for any x ∈ (0, 1) \ A, we choose, for ² > 0, numbers ai < x < aĩ
so that f (aĩ ) − ² < f (x) < f (ai ) + ², and then J so that | fj 0 (ai ) − f (ai )| < ² and
| fj 0 (aĩ ) − f (aĩ )| < ² for j ≥ J. For such j it follows that
f (x) − 2² < f (ai ) − ² < fj 0 (ai ) < fj 0 (x) < fj 0 (aĩ ) < f (aĩ ) + ² < f (x) + 2².
Thus | fj 0 (x) − f (x)| < 2². ¤

Problem 5:
a. Is there a nonconstant real function h that is continuous on the closed disk
{z : |z| ≤ 1}, harmonic on the open disk {z : |z| < 1}, and vanishes on the up-
per unit semi-circle?
b. Is there a nonconstant complex function f that is continuous on the closed disk
{z : |z| ≤ 1}, holomorphic on the open disk {z : |z| < 1}, and vanishes on the upper
unit semi-circle?

Solution 5:
(a) The Posson Integral Formula shows that, for any continuous function g on the
unit circle, one may find a harmonic function on the open ball which is continuous on
54 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

the closed ball and has boundary values g. So it suffices to choose any nonconstant g
which vanishes on the upper semi-circle.

(b) There is a conformal map from the unit disk to the upper half plane. This takes
the upper semi-circle to an interval on the X-axis. Composing with this conformal
map thus gives a holomorphic map on the upper half plane which vanishes on this
interval. Schwarz reflection about this interval then extends this function to be a
holomorphic function whose domain contains the interval and vanishes on the inter-
val. The identity theorem implies that this function, and hence the original function,
must vanish identically. ¤

Problem 6: Assume that f (x) is a Lebesgue measurable function on R. Prove the

function defined on R2 by F (x, y) = f (x − y) is Lebesgure measurable.

¡ ¢
Solution 6: We need to show that F −1 (a, b) is measurable in R2 for any interval
√
(a, b) ⊂ R. Note that F = f ◦ P where P (x, y) = x + y. Also note that P = p ◦ 2 · φ
where φ is a 45o rotation of the plane and p(x, y) = x. So
¡ ¢ √ ¡ ¡ ¢¢
F −1 (a, b) = ( 2 · φ)−1 p−1 [f −1 (a, b) ]
¡ ¢
E = f −1 (a, b) is measurable in R by the measurability of f ,and
p−1 (E) = E × R is measurable by the definition of Lebesgue measure as a product
measure. Moreover, since Lebesgue measurability is preserved under rotation and
¡ ¢ √
homothety F −1 (a, b) = ( 2 · φ)−1 (E × R) is measurable. ¤
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 55

Jan 2004

Note: the solutions to this exam’s problems were posted by Professor Hardt following
the exam. Permission has been given to reproduce his solutions here. Any mistakes
below are certainly due to error in reproduction.

Problem 1:
(a) Classify all enitre functions f : C → C such that
| f (z)|
sup <∞
z∈C 1 + | z|4
(b) Classify all entire functions g : C → C such that
| g(z)|
inf >0
z∈C | z|4

Solution 1: The function f ( z1 ) has an isolated singularity at 0. If this singularity

is removable, then f is bounded and so constant by Louiville’s Theorem, which is
one possibility. If it had a transcendental singularity at 0, then z 4 f ( z1 ) would also
have a transcendental singularity at 0 and be unbounded, contradicting the growth
assumption on f at ∞. We see that f ( z1 ) must have a pole at 0 so that f is nec-
essarily a polynomial. Also we see that the degree of f is at most 4, and any such
polynomial satisfies the hypothesis. Thus f (z) = a0 + a1 z + a2 z 2 + a3 z 3 + a4 z 4 for
some a0 , . . . , a4 ∈ C.

Again g( z1 ) cannot have a transcendental singularity at 0 because then z 4 g( z1 ) would

be arbitrarily close to zero for some points z near 0. So again g is a polynomial. But
now the condition implies that g can vanish only at the origin. So, by the Fundamen-
tal Theorem of Algebra, g(z) = az m . The condition inf z∈C | a| | z|m−4 > 0 requires
that m − 4 ≥ 0 (for z near 0) and m − 4 ≤ 0 (for z near ∞). So g(z) = az 4 with
56 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

a 6= 0. ¤

Problem 2: Suppose that fn : R → R is differentiable function for every positive

integer n, M = supn,x | fn0 (x)| < ∞ and that f (x) = limn→∞ fn (x) ∈ R exists for all
x ∈ R.
(a) Show that the functions fn are uniformly bounded on each interval [a, b] ⊂ R.
(b) Is f continuous on R? Prove or find a counterexample.
(c) Is f differentiable on R? Prove or find a counterexample.

Solution 2:
(a) Since f (a) = limn→∞ fn (a), N = supn | fn (a)| < ∞. Then for any x ∈ [a, b] the
Fundamental Theorem of Calculus gives the uniform bound
¯Z x ¯
¯ ¯
| fn (x)| ≤ | fn (a)| + ¯¯ fn0 (t)dt¯¯ ≤ N + M | b − a|
a

(b) Yes, as in (a) the Fundamental Theorem of Calculus implies that for −∞ < x <
y < ∞,
Z y
| f (y) − f (x)| = lim | fn (y) − fn (x)| ≤ lim sup | fn0 (t)| dt ≤ M (y − x)
n→∞ x

(c) Not necessarily. One easily obtains an example with f (x) = | x| and the graph of
fn (x) being obtained by slightly rounding the graph of | x|. ¤

Problem 3: Compute the (improper) integral

Z ∞
sin x
dx
0 x(x2 + 1)
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 57

Solution 3: This improper integral exists as limR→∞ IR where

Z R "Z Z R#
−1/R
sinx 1 sinx
IR = 2
dx = + 2
dx
1/R x(x + 1 2 −R 1/R x(x + 1
¯ ¯
because ¯ sinx
x
¯ ≤ 1 and 21 is integrable on [0, ∞). We want to use the Cauchy
x +1
Integral Formula, buw we need to choose the f (z) so that the integral on the extra
boundary curve will approach 0 as the domain gets larger. [Warning: | sinz| ≤ 1 is
not always true for complex z.] One thing that works is to note that
sinx eix
= Im
x(x2 + 1 x(x2 + 1)
for x real and take
eiz
f (z) =
z(z 2 + 1)
on the domain ΩR in the upper half-plane bounded by the 4 curves
1 1 1
[−R, − ], γR = { eiθ : π ≥ θ ≥ 0}, [ , R], ΓR = {Reiθ : 0 ≤ θ ≤ π}
R R R
2
ei 1
Inside ΩR , f (z) has a single pole at z = i with residue i(i+i)
= − 2e . Thus, Cauchy’s
Residue Formula gives
µ µ ¶¶ Z Z Z
π 1
− = Im 2πi − = Im f (z)dz = 2IR + Im f (z)dz + Im f (z)dz
e 2e ∂ΩR γR ΓR
¯ ¯
¯ iReiθ ¯ ¯¯ −R sin θ ¯¯
On ΓR , ¯ e ¯= e ≤ 1 because sin θ ∈ [0, 1]. So we see that
¯Z ¯
¯ ¯
¯ f (z)dz ¯ ≤ 1 πR −→ 0 as R → ∞
¯ ¯ R3
ΓR

Finally Z Z
1 1
f (z)dz = − f (z)dz = − (2πi) Res0 f = −πi(1)
γR 2 ∂B1/R 2
So taking imaginary parts,
1h π i π
lim IR = − + π = (1 − e−1 )
R→∞ 2 e 2
¤
58 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Problem 4:
(a) In the unit disk {z ∈ C : | z| < 1} how many solutions are there to the equation
z 8 − 5z 3 + z = 2?

(b) In the radius-2 disk {z ∈ C : | z| < 2} how many solutions are there to the
same equation z 8 − 5z 3 + z = 2?

Solution 4:
(a)We apply Rouche’s Theorem with f (z) = z 8 − 5z 3 + z − 2 and g(z) = −5z 3 on the
unit disk noting that for | z| = 1,
¯ ¯
| f (z) − g(z)| = ¯ z 8 + z − 2¯ ≤ | z|8 + | z| + 2 = 1 + 1 + 2 = 4 < 5(1)3 = | g(z)|

Thus in the unit disk, f (z) has the same number of zeros as g(z) (counting multiplic-
ities), namely 3. So the equation z 8 − 5z 3 + z = 2 has 3 solutions in the unit disk.

(b) Here we use the same f but now take g(z) = z 8 and note that for | z| = 2 one has
¯ ¯
| f (z) − g(z)| = ¯ −5z 3 + z − 2¯ ≤ 5(2)3 + 2 + 2 = 44 < (2)8 = | g(z)|

So the equation z 8 − 5z 3 + z = 2 has 8 solutions in the radius-2 disk. ¤

Problem 5:
(a) Suppose that f is integrable on [0, 1]. Show that there exists a sequence of positive
numbers an ↓ 0 so that limn→∞ an | f (an )| = 0.
R1
(b) Let fn be a sequence of functions integrable on [0, 1] with supn 0
| fn (x)| dx < ∞.
Does there exist a subsequence fnk of fn and a sequence of positive numbers bk ↓ 0
and so that limk→∞ bk | f (bk )| = 0. If so, prove it. If not, find a counterexample.

Solution 5:
(a) If this were false, then ² = lim inf x→0 x | f (x)| > 0, and there would exist a positive
 SOLUTIONS TO QUALIFYING EXAM PROBLEMS IN ANALYSIS 59

²
δ so that x | f (x)| ≥ 2
whenever 0 < x ≤ δ. But then

Z 1 Z δ Z δ
²
| f (x)| dx ≥ | f (x)| dx ≥ dx = ∞
0 0 0 2x

contradicting the integrability of f .

(b) As Professor Jones pointed out, a stronger result is true. One need only assume
that each fn is integrable and one doesn’t need to pass to a subsequence fnk for the
P∞ R1
conclusion. Here we first choose αk ↓ 0 so that k=1 αk 0 | fk (x)| dx < ∞, and
P
apply (a) to the integrable function f (x) = ∞ k=1 αk | fk (x)| to find points am ↓ 0

so that limm→∞ am f (am ) = 0. Passing to a subsequence we can make this sequence

converge as fast as we want. In particular we can choose inductively amk ↓ 0 so that
amk f (amk ) ≤ αk2 . Letting bk = amk , we conclude that

bk f (bk ) ≤ bk αk−1 f (bk ) ≤ αk−1 αk2 = αk −→ 0 as k → ∞

Problem 6: Suppose 1 ≤ p ≤ ∞, f ∈ Lp ([0, 1]), and h(t) is the Lebesgue measure

of the set {x ∈ [0, 1] : | f (x)| > t} for 0 ≤ t < ∞.
R∞
(a) Show that 0 h(t)dt < ∞ if 1 < p ≤ ∞.
(b) Is this still true for p = 1? Prove or find a counterexample.

Solution 6: Here this is true for p = 1. Since Holder’s inequality implies that
Lp ([0, 1]) ⊂ L1 ([0, 1]), we only need prove the case p = 1 and part (a) follows.

For this, we use Fubini’s Theorem with the characteristic function of the subgraph

A = {(x, y) : 0 ≤ x ≤ 1, 0 ≤ y < | f (x)|}

60 AS GIVEN BY THE RICE UNIVERSITY MATHEMATICS DEPARTMENT

Let λ denote the 1-dimensional Lebesgue measure. By Fubini’s Theorem, A is 2-

dimensional Lebesgue measureable function with 2-dimensional measure
Z 1 Z 1
| A| = λ{y : (x, y) ∈ A}dx = | f (x)| dx < ∞
0 0
But slicing the other way shows that
Z ∞ Z ∞ Z ∞
h(y)dy = λ{| f (x)| > y}dy = λ{x : (x, y) ∈ A}dy = | A| < ∞
0 0 0

One can obtain an alternative proof of (a) but not (b) by using Chebyshev’s Inequality
to see that Z 1
p 1
h(t) = λ{x ∈ [0, 1] : | f (x)| > t } ≤ p p
| f (x)|p dx
t 0
So
Z ∞ Z ∞ µZ 1 ¶Z ∞
p
h(t)dt ≤ 1 + h(t)dt ≤ 1 + | f (x)| dx t−p dt < ∞
0 1 0 1
¤