The Georgia Institute of Technology

Spring 2012
Math 6338 - Real Analysis II
Lecture: Room: Skiles 154 - Time: MWF 3:05pm – 3:55pm

Instructor: Brett D. Wick Office: Skiles 224

Office Phone: 404-864-4311 Office Hours: MWF 11:00am – 12:00pm
Email: [email protected] or by appointment
Webpage: http://people.math.gatech.edu/∼bwick6/teaching/math6338.html

Text: There will be no required text for the course. There are many reasonable texts for
this course that the student could consult. Potential texts include:

• Introductory Functional Analysis with Applications by Kreyszig;

• A Course in Functional Analysis by Conway;

• Functional Analysis by Rudin.

Prerequisite and Description: Math 6338 is an introduction to graduate real analysis.

Topics covered include: Basics of measure theory; Lp spaces: Completeness, Riesz repre-
sentation and the dual of Lp , Hölder’s, Jensen’s, and Minkowski’s inequalities; Topological
spaces, normed vectors spaces, compact spaces and the Tychonoff theorem. Linear func-
tionals, Banach and Hilbert spaces, orthonormal bases and orthogonal projections in Hilbert
spaces. Basic theorems of functional analysis, including the Hahn-Banach, Baire Category,
and open mapping theorems and the uniform boundedness principle.

Prerequisites for the course are Math 4318 (Undergraduate Analysis II) or consent of the
School.

Attendance: Attendance is required for all lectures. The student who misses a class meet-
ing is responsible for any assignments and/or announcements made. Office hours will not
be utilized to re-teach material presented in class. However, questions to better understand
the course are always welcome.

In the event of an absence that will impact your ability to complete your assign-
ments due to travel representing Georgia Tech, you must notify the professor at
least two weeks in advance to arrange an early test or other alternative. Other-
wise, such absences will be treated as personal.

Homework: This course will have weekly homework assignments that will be collected and
graded.

Exams: This course will have two mid-term exams, and a comprehensive final exam. The
exams for the course will take place on:
Exam Dates:
Exam 1 February 13, 2012
Exam 2 April 2, 2012
Final Exam Friday, May 4 2012 2:50pm - 5:40pm
Learning Disabilities: It is the right of any student with a certified learning disability to
request necessary accommodation. Such requests must be made well in advance of the time
that the accommodation is required and a letter of documentation from the ADAPTS office
must be presented at the time of any request.

Academic Honesty: It is expected that all students are aware of their individual respon-
sibilities under the Georgia Tech Academic Honor Code, which will be strictly adhered to in
this class.

Grades: The usual ten-point scale will be used (A: 90-100, B: 80-89, C: 70-79, D: 60-69, F:
0-59), however, if necessary, adjustments will be made to arrive at a standard grade distri-
bution. Grades will be based upon attendance, participation, and exams and homework. An
approximate grade break down will be 40% for homework, 15% for each midterm, and 30%
for the final exam. This is break down may change to arrive at a suitable grade distribution.

Important Dates for Spring 2012:

January 9 First day of classes
January 13 Last day to register and/or make schedule changes
January 16 School Holiday
March 2 Last day to drop or withdraw with a grade of “W”
March 14 Last day to withdraw from school with a grade of “W”
March 19-23 Spring Break
May 4 Final Exam
 Math 6338 : Real Analysis II
Mid-term Exam 1
13 February 2012
Instructions: Answer all of the problems.

1. Let X be a real normed space. Suppose that C ⊂ X is a convex subset which contains 0
and has the property that [
tC = X
t>0
For every x ∈ X define
QC (x) = inf{t > 0 : x ∈ tC}
Show that the map QC : X → R is a sublinear functional. Namely it satisfies the following
properties:
(i) QC (x + y) ≤ QC (x) + QC (y);
(ii) QC (tx) = tQC (x) for all x ∈ X and all t ∈ R+ .

Solution: For x ∈ X, let

S(x) = {t > 0 : x ∈ tC}.
In particular, note that since 0 ∈ C we have that S(0) = (0, ∞), and thus QC (0) = 0.
To prove the result in question, it suffices to show that

(i) S(x + y) ⊃ S(x) + S(y);

(ii) S(λx) = λS(x) for all x ∈ X and all λ > 0.

Using these two conditions and the definition of QC , the result follows.

We now turn to proving (i). Let t ∈ S(x) and s ∈ S(y). Consider the elements u = t−1 x
and v = s−1 y. As u, v ∈ C and C is convex it follows that
1 t s
(x + y) = u+ v ∈ C.
t+s t+s t+s
This means that x + y ∈ (t + s)C, and so t + s ∈ S(x + y). This proves that
S(x + y) ⊃ S(x) + S(y).

Consider the second claim (ii). Suppose that t ∈ S(λx), then we have that λx ∈ tC, or
equivalently, λ−1 tx ∈ C, which means that λ−1 t ∈ S(x), and so

t = λλ−1 t ∈ λS(x).

This proves the inclusion S(λx) ⊂ λS(x). To see the other inclusion, suppose that
s ∈ λS(x), then there must exist a t ∈ S(x) such that s = λt. Since t = λ−1 s ∈ S(x),
this means that x ∈ λ−1 sC, or λx ∈ sC, so s ∈ S(λx), i.e. λS(x) ⊂ S(λx).
2. Let X be a real normed space, and let C ⊂ X be a convex open set which contains 0. If
x0 ∈
/ C, then show that there exits a bounded linear functional ϕ̃ : X → R such that
(i) ϕ̃(x0 ) = 1;
(ii) ϕ̃(v) < 1 for all v ∈ C.
Hint: Apply Problem 1.

Solution: The idea is to apply the Hahn-Banach Theorem with the sublinear functional
from Problem
S 1. First note that if C
S is a convex open set which contains 0, then we have
X = t>0 tC. Note that we have t>0 tC ⊂ X. Now let x ∈ X, we need to show that
x ∈ ρC from some value of ρ > 0. Note that the map t → tx is continuous. Since C is
a neighborhood of 0, there exists some ρ > 0 such that for all t ∈ [−ρ, ρ] we have that
tx ∈ C, in particular, ρx ∈ C or x ∈ ρ−1 C. We now can apply Problem 1.

Let Y = Rx0 = {rx0 : r ∈ R}. This is clearly a linear subspace of X. Define ϕ : Y → R

by ϕ(rx0 ) = r. Then, we have that ϕ(x0 ) = 1, and the idea will be to extend the
functional ϕ to all of X. We next note that

ϕ(y) ≤ QC (y) ∀y ∈ Y.

If y = rx0 with r < 0, then this is immediate since ϕ(y) = r < 0 ≤ QC (y) since
QC (y) ≥ 0 always. If y = rx0 with r ≥ 0, then we have that

ϕ(y) = r ≤ rQC (x0 ) = QC (y)

here we used that since x0 ∈

/ C, we must have that QC (x0 ) ≥ 1.

So we have that ϕ is a linear functional with ϕ(y) ≤ QC (y) for all y ∈ Y . By the
Hahn-Banach Theorem we can find an extension ϕ̃ : X → R such that ϕ̃|Y = ϕ and
ϕ̃(x) ≤ QC (x) for all x ∈ X. Now we obviously have that ϕ̃(x0 ) = ϕ(x0 ) = 1. And we
also have that ϕ̃(v) < 1 for all v ∈ C.

It remains to prove that ϕ̃ is bounded, and since ϕ̃ is linear to do so it will be enough to

prove that it is continuous at the origin. Namely, we need to prove that for  > 0 there
is some open set U containing the origin such that

|ϕ̃(u)| <  u ∈ U .

Take U = C ∩ −C, which is clearly open and contains the origin. Note that for all
u ∈ U we have that ±u ∈ C, or equivalently, −1 (±u) ∈ C. This implies that

QC (±u) < 

and since ϕ̃(u) ≤ QC (u) for all u, we have that |ϕ̃(u)| <  for all u ∈ U .

3. Let X be a normed vector space over the real numbers. Suppose that A, B ⊂ X are
non-empty convex sets with A open and A ∩ B = ∅. Show that there exists a bounded linear
functional φ : X → R and a real number α such that
φ(a) < α ≤ φ(b) ∀a ∈ A, b ∈ B.
Hint: Apply Problem 2 with C = A − B + b0 − a0 .

Solution: Fix points a0 ∈ A and b0 ∈ B and define the set

C = A − B + b0 − a0 = {a − b + b0 − a0 : a ∈ A, b ∈ B}

It is easy to see that C is convex and contains the origin. We also have that
[
C= (A + b0 − a0 )
b∈B

so C is open. Consider the vector x0 = b0 − a0 , and since A ∩ B = ∅, we have that

x0 ∈/ C. Now apply Problem 2, to find a bounded linear map ϕ̃ such that ϕ̃(x0 ) = 1 and
ϕ̃(v) < 1 for all v ∈ C. Now compute that

1 > ϕ̃(a − b + b0 − a0 ) = ϕ̃(a) − ϕ̃(b) + ϕ̃(b0 ) − ϕ̃(a0 )

Using that ϕ̃(b0 ) − ϕ̃(a0 ) = 1 we have that

ϕ̃(a) < ϕ̃(b) a ∈ A, b ∈ B.

Set α = inf b∈B ϕ̃(b), then we have that

ϕ̃(a) ≤ α ≤ ϕ̃(b) a ∈ A, b ∈ B.

So, to prove the desired result, we need to simply show that

ϕ̃(a) < α a ∈ A.

Suppose to the contrary that there exists a a1 ∈ A such that ϕ̃(a1 ) = α. Since the map
t → a1 + tx0 is continuous, there is some  > 0 such that

a1 + tx0 ∈ A ∀t ∈ [−, ].

In particular, we have that

ϕ̃(a1 + x0 ) ≤ α,
which is equivalent to α +  ≤ α, which is impossible.

4. Suppose that X is a normed linear space over C. Show that the set of all bounded linear
functionals on X, denoted by X ∗ , is a Banach space with norm given by
|f (x)|
kf kX ∗ = sup .
x6=0 kxkX
 Solution: It is easy to see that k · kX ∗ is actually a norm. The real issue here is com-
pleteness of the space X ∗ . So suppose that {fn } is a Cauchy sequence in X ∗ . Namely,
kfn − fm kX ∗ → 0 as n and m get very large. Fix x ∈ X, and observe that

|fn (x) − fm (x)| ≤ kxkX kfn − fm kX ∗

which implies that {fn (x)} is a Cauchy sequence in C. We can then define

f (x) = lim fn (x)

n→∞

since C is complete. Since limits are linear and each fn is linear we have that f is a linear
functional on X. It only remains to show that f is a bounded linear functional. For any
x ∈ X with x 6= 0 we have that for  > 0 that

|f (x) − fN (x)| < kxkX

when n ≥ N . Also, since fn is Cauchy in X ∗ we have that kfn k is a bounded sequence,

i.e. supn kfn kX ∗ = C < ∞. Thus we have that

|f (x)| ≤ |f (x) − fN (x) + fN (x)| ≤ ( + C)kxkX

And since  was arbitrary, we have that |f (x)| ≤ CkxkX , and so f is a bounded linear
functional. This then proves that X ∗ is complete and hence a Banach space.

5. Let H be a Hilbert space over R. Show that H is isometrically isomorphic to H ∗ .

Solution: This is essentially a Corollary of the Riesz Representation Theorem. We

need to find the map that carries H ∗ to H. So, let f ∈ H ∗ , then f is a bounded linear
functional on H, and so by the Riesz Representation Theorem, we have that there exists
a unique z ∈ H such that
f (x) = hx, ziH
and moreover kf kH ∗ = kzkH . Now define a map from ı : H ∗ → H given by ı(f ) = z
where z is the unique element obtained by the Riesz Representation Theorem. We first
observe that ı is linear. If f, g ∈ H ∗ then we have that

ı(f + rg) = zf +rg ı(f ) = zf ı(rg) = zrg

and for any x ∈ H, hx, zf +rg iH = (f + rg)(x) = f (x) + rg(x) = hx, zf iH + r hx, zg iH , so
we have that ı(f +rg) = ı(f )+rı(g), and ı is linear. We also have that ı is injective, since
if ı(f ) = ı(g) = w, then we have that f (x) = hx, wiH = g(x) for all x ∈ H, and so f = g.
It is also true that ı is surjective, since we have that the linear functional fu (x) = hx, uiH
has the property that ı(fu ) = u. So ı : H ∗ → H is an isomorphism between the linear
spaces H ∗ and H. But, the Riesz Representation Theorem then gives that

kıf kH = kzkH = kf kH ∗ .
6. Let X and Y be Banach spaces and let T : X → Y be bounded and linear. Show that
there is a constant c > 0 such that kT xkY ≥ ckxkX for all x ∈ X if and only if ker T = {0}
and Ran T is closed.

Solution: Suppose that kT xkY ≥ ckxkX for all x ∈ X. Then if we have x ∈ ker T we
have that
0 = kT xkY ≥ ckxkX
which gives that x = 0, and so ker T = {0}. Suppose that y ∈ Ran T , and let yn ∈ Ran T
with yn → y. Since yn ∈ Ran T , we have that yn = T xn . Note that xn is a Cauchy
sequence since
kyn − ym kY = kT xn − T xm kY ≥ ckxn − xm kX .
Since X is complete, then we have that xn → x. But, since T is bounded, hence contin-
uous, we have that T xn → T x so T x = y, and so Ran T ⊂ Ran T , and so Ran T is closed.

Now suppose that ker T = {0} and Ran T is closed. Note that since T is bounded and
linear and ker T = {0} we have that T is injective. Similarly, since Ran T is closed, we
have that T is surjective. So T is bounded linear and bijective, and so by the Open
Mapping Theorem (Bounded Inverse Theorem) we have that T −1 is bounded,

kT −1 ykX ≤ ckykY .

Apply this inequality to y = T x to get the desired result.

7. Let L2 [0, 1] be the Lebesgue space of square integrable functions on the unit interval.
Similarly, let L1 [0, 1] be the space of integrable functions on the unit interval. Prove that
L2 [0, 1] is of the first category in L1 [0, 1].

Solution: Let En = {f ∈ L2 [0, 1] : |f |2 dx ≤ n}, and we have that L2 [0, 1] = ∞

R S
n=1 En .
We claim that En is closed in L1 [0, 1] but is nowwhere dense. This then proves the
desired result.

It is easy to see that En is closed. If we have a sequence {fk } ∈ En with fk → f in L1 ,

then we can find a subsequence fnk that converges pointwise almost everywhere. Then
as a Corollary to Fatou’s Lemma we have that

kf k2L2 [0,1] ≤ n.

Now observe that the interior of En is empty. Suppose that this isn’t true, then there
would exist an f0 ∈ En and some r > 0 such that D = {g : kf0 − gkL1 [0,1] < r} ⊂ En .
Let f ∈ L1 and consider the function g = f0 + 2r kf k f1 . We have that g ∈ D and so
L [0,1]

G ∈ En , and so g ∈ L2 [0, 1], rearrangement then gives that

2kf0 k1
f= (g − f0 ) ∈ L2 [0, 1].
r
 But f ∈ L1 [0, 1] was arbitrary, this implies that L2 [0, 1] = L1 [0, 1], which is false. So
we have that En has empty interior. This then gives that En is nowhere dense, and so
L2 [0, 1] is meager in L1 [0, 1].

8. Let S be a closed linear subspace of L1 [0, 1]. Suppose that for each f ∈ S there exists a
p > 1 such that f ∈ Lp [0, 1]. Show that there exists a q > 1 such that S ⊂ Lq [0, 1]

Solution: Since S ⊂ L1 [0, 1] and S is closed, it is complete since L1 [0, 1] is complete.

Let qn be a sequence decreasing to 1 and

En,m := {f ∈ S : f ∈ Lqn [0, 1], kf kLqn ≤ m}

where m ∈ N. Then we have that

[
S= En,m
n,m

since by hypothesis for each f ∈ S there is a p > 1 such that f ∈ Lp [0, 1] and if r > s
then Lr [0, 1] ⊂ Ls [0, 1] and choosing qn < p will place f ∈ En,m since the norm of f will
have to be smaller than some m.

Now each term in the union is closed, i.e. En,m = En,m . Since if we take a sequence
{fk } ⊂ En,m with fk → f , then f ∈ En,m . To see this, note that we are working in the
topology defined by the L1 norm, and so we can find a subsequence {fkj } that converges
to f almost everywhere on [0, 1]. Then as a Corollary of Fatou’s Lemma we have
Z 1
|f |qn dx ≤ m.
0

By Baire’s Category Theorem, there exists n0 , m0 with qn0 > 1 such that En0 ,m0 has
non-empty interior. So, there exists a ball of radius δ > 0 centered at some point f such
that
Bδ (f ) ⊂ En0 ,m0 .
For, the moment, lets assume that f = 0, the general case can be handled similarly and
by translation. Then we have that Bδ (0) ⊂ Lqn0 [0, 1].

Let 0 6= g ∈ S be any function, then the function g̃ = 2δ kgkg 1 ∈ Bδ (0), and so g̃ ∈

L
Lqn0 [0, 1]. But then since Lqn0 [0, 1] is a linear space, cg̃ ∈ Lqn0 [0, 1] and so g ∈ Lqn0 [0, 1],
which gives that S ⊂ Lqn0 [0, 1].
 Math 6338 : Real Analysis II
Mid-term Exam 2
09 April 2012
Instructions: Answer all of the problems.

1. Suppose that ϕ ∈ L∞ (R). Let Mϕ : L2 (R) → L2 (R) be the operator:

Mϕ (f ) = ϕf.
The essential range, Rϕ of a function ϕ ∈ L∞ (R) is the set of all w ∈ C such that
|{x : |ϕ(x) − w| < }| > 0
for all  > 0.
(a) Show that Mϕ is a bounded operator on L2 (R);
(b) Show that σ(Mϕ ) = Rϕ .

Solution: Part (a): Note that we have

Z
kMϕ f k2L2 (R) = |ϕ(x)f (x)|2 dx
R Z
≤ kϕkL∞ (R) |f (x)|2 dx
2
R
= kϕkL∞ (R) kf k2L2 (R) .
2

This gives that kMϕ k ≤ kϕk∞ . In fact, we have that kMϕ k = kϕk∞ .

Part (b): We will prove the two containments σ(Mϕ ) ⊂ Rϕ and Rϕ ⊂ σ(Mϕ ).

σ(Mϕ ) ⊂ Rϕ : If λ ∈
/ Rϕ , then there exists an  > 0 such that

|{x : |ϕ(x) − λ| < }| = 0 (1)

1
Consider the function g(x) = ϕ(x)−λ . Then we have that |g(x)| ≤ 1 almost everywhere
on R by (1). However, we have that Mϕ − λI = Mϕ−λ , and Mϕ−λ Mg = Mg Mϕ−λ = I,
and so Mϕ − λI is invertible with a bounded inverse. This then gives that λ ∈/ σ(Mϕ ).
Rephrasing, σ(Mϕ ) ⊂ Rϕ .

Rϕ ⊂ σ(Mϕ ): Now suppose that λ ∈ Rϕ . Then for each n, let

 
1
En = x : |ϕ(x) − λ| < .
n

Each of these sets have positive measure since λ ∈ Rϕ . Let fn = 1En and note that
kfn k2L2 (R) = |En |.
 Now, observe that for this value of λ we have the following
Z
2
k(Mϕ − λI)fn kL2 (R) = |ϕ(x) − λ|2 |fn (x)|2 dx
R
1 1
≤ 2
|En | = 2 kfn k2L2 (R) .
n n
This implies that Mϕ − λI is not bounded below and so is not invertible. This gives that
Rϕ ⊂ σ(Mϕ ).

2. Let C 2 = C 2 [0, 1] be the space of all complex valued functions on [0, 1] whose second
derivative is continuous. Choose a > 0 and b > 0 and define

kf ka,b = kf k∞ + akf 0 k∞ + bkf 00 k∞

where kgk∞ = supt∈[0,1] |g(t)|.

(a) Show that for any choice of a and b that we have (C 2 , k · ka,b ) is a Banach space;

(b) Show that (C 2 , k · ka,b ) is a Banach algebra if and only if 2b ≤ a2 .

Solution: Part (a): It is clear that kf ka,b satisfies all the axioms for a normed space. It
only remains to show completeness to obtain that it is a Banach space. However since
C 2 is a complete Banach space, and since
1
kf 00 k∞ ≤ kf ka,b
b
the completeness follows easily from this and the completeness of C 2 .

Part (b): By Part (a) we know that (C 2 , k·ka,b ) is a Banach space. Suppose that 2b ≤ a2 .
Now observe that we have

kf gka,b = kf gk∞ + ak(f g)0 k∞ + bk(f g)00 k∞

≤ kf k∞ kgk∞ + akf 0 k∞ kgk∞ + akf k∞ kg 0 k∞ + 2bkf 0 k∞ kg 0 k∞
+ bkf 00 k∞ kgk∞ + bkf k∞ kg 00 k∞ .

On the other hand, we have that

kf ka,b kgka,b = kf k∞ kgk∞ + akf 0 k∞ kgk∞ + akf k∞ kg 0 k∞ + a2 kf 0 k∞ kg 0 k∞

+ bkf k∞ kg 00 k∞ + bkgk∞ kf 00 k∞ + b2 kf 00 k∞ kg 00 k∞
+ abkf 0 k∞ kg 00 k∞ + abkg 0 k∞ kf 00 k∞ .

From these two expressions it is clear that since 2b ≤ a2 , we have that

kf gka,b ≤ kf gka,b .
 For the converse, suppose that k · ka,b is a Banach algebra norm. Then consider f (x) = x
and g(x) = x. Then we have

kf ka,b = kgka,b = kxka,b = kxk∞ + ak1k∞ + bk0k∞ = 1 + a.

While
kf gka,b = kx2 ka,b = kx2 k∞ + 2akxk∞ + 2bk1k∞ = 1 + 2a + 2b.
Then we have

1 + 2a + 2b = kf gka,b ≤ kf ka,b kgka,b = (1 + a)2 = 1 + 2a + a2 .

This then implies that 2b ≤ a2 .

3. Let T be a bounded linear operator on a Hilbert space H. Suppose that T ∗ T is compact.

Show that T is then compact.

Solution: Suppose that {xn } is a sequence in the unit ball of the Hilbert space H. It
suffices to show that {T xn } is a Cauchy sequence. Note that

kT xn − T xm k2H = hT xn − T xm , T xn − T xm iH
= hT ∗ (T xn − T xm ), xn − xm iH
≤ kT ∗ T xn − T ∗ T xm kH kxn − xm kH
≤ 2kT ∗ T xn − T ∗ T xm kH .

Since T ∗ T is compact, we have that {T ∗ T xn } is Cauchy in H, and so for any  > 0 there
exists a N such that if n, m > N then we have

2
kT ∗ T xn − T ∗ T xm kH < .
2
For this N , we then have that when n, m > N that

kT xn − T xm k2H < 2 .

4. For k(x, y) : R × R → C a R × R measurable function define the operator L2 (R) → L2 (R)

by Z
K(f )(x) := k(x, y)f (y) dy.
R
Suppose that
Z Z  21
2
M= |k(x, y)| dx dy <∞
R R
and show that
(a) kKkL2 (R)→L2 (R) ≤ M ;
 (b) K is a compact operator on L2 (R).

Solution: Part (a): We compute from the definition that

Z
2
kKf kL2 (R) = |Kf (x)|2 dx
R Z Z
2
≤ kf kL2 (R) |k(x, y)|2 dydx
R R
= M 2 kf k2L2 (R) .

This then gives that kKf kL2 (R) ≤ M kf kL2 (R) , which proves (a).

2
Pn (b): Suppose that {aj } and {bj } are sequences in L (R). For fixed n, set k̃n (x, y) =
Part
j=1,j=1 ai (x)bj (y). Define K̃n in the same manner as above. Then we have that K̃n is
finite dimensional, and so each K̃n is compact. Now choose the sequences {aj } and {bj }
so that
kK − K̃n kL2 (R)→L2 (R) → 0.
To see that such a sequence is possible, let {ej } be a orthonormal basis for L2 (R). It is
easy to see that {ej (x)ek (y)} is an orthonormal basis for L2 (R2 ). Then for some choice
of coefficients, we have that
X X
K(x, y) = cjk ej (x)ek (y) = ai (x)bj (y).
j,k i,j

Then by Part (a) above we have that

X
kK − K̃n k2L2 (R) ≤ kk − k̃n kL2 (R2 ) = |cij |2 → 0.
i,j>N

Then apply Problem 6a below to conclude that K is compact too.

5. For T ∈ B(X), where X is complex Banach space. The approximate point spectrum is
the set of λ ∈ C such that there exists a sequence {xn } ∈ X with kxn kX = 1 for all n and
k(T − λI)xn kX → 0. Prove the following are equivalent:

(a) λ ∈
/ σap (T );

(b) ker(T − λI) = {0} and ran(T − λI) is closed;

(c) There is a constant c > 0 such that k(T − λI)xkX ≥ ckxkX .

Hint: Attempt to show (a) ⇒ (c) ⇒ (b) ⇒ (a).

 Solution: Without loss we may take λ = 0.

(a) ⇒ (c): Suppose that (c) fails, then for every n there is a vector xn such that
kT xn kX ≤ n1 kxn kX . Set yn = kxxnnkX , then kyn kX = 1 and kT yn kX → 0, and so 0 ∈ σap (T ).

(c) ⇒ (b): Suppose that kT xkX ≥ ckxkX . Then clearly we have that ker T = {0}. If
we have that T xn → y, then we have kxn − xm kX ≤ c−1 kT xn − T xm kX . So {xn } is a
Cauchy sequence. Let x = lim xn and so we have that T x = y and ran T is closed.

(b) ⇒ (a): Let Y = ran T , so T : X → Y is a continuous bijection, and by the Inverse

Mapping Theorem there is a bounded operator S such that ST x = x for all x ∈ X. So
if kxkX = 1, then
1 = kxkX = kST xkX ≤ kSkkT xkX .
This gives that kT xkX ≥ ckxkX , when kxkX = 1 and so 0 ∈
/ σap (T ).

6. Let X and Y be complex Banach spaces. Suppose that T ∈ B(X, Y ) and Tn ∈ B(X, Y )
for all n ≥ 1.

(a) Suppose that Tn has finite dimensional range and that lim kT − Tn k = 0. Show that T
is compact.

(b) Assume that Y is a Hilbert space and prove the converse of (a): Every compact
operator T ∈ B(X, Y ) can be approximated in the operator norm by operators with
finite dimensional ranges. Hint: In a Hilbert space there are projections of norm 1 onto
any closed subspace.

Solution: Part (a): Since each {Tn } has finite dimensional range, it is a compact opera-
tor. Then since Tn → T uniformly, we have that T is compact by a Theorem from class.

Part (b): Let B denote the unit ball in the Hilbert space X. Then since T is compact,
T (B) has compact closure, and is totally bounded. So for any  > 0 there exists N =
N () < ∞ and {y1 , . . . , yN } ∈ Y such that
N ()
[
T (B) ⊂ {y : ky − yj kY < }. (2)
j=1

Set PN to be the orthogonal projection on the the span{y1 , . . . , yN }. Then consider the
operator TN = PN T . This operator is finite dimensional and hence compact. We then
have that
kT − TN k < .
This follows immediately from (2). Now simply choose a sequence n → 0. This then
give a collection {Tn } of compact operators such that kT − Tn k → 0.
7. If L2 = L2 (0, ∞) and if
Z s
1
T f (s) = f (t)dt 0 < s < ∞.
s 0

(a) Prove that T ∈ B(L2 );

(b) Prove that T is not compact.

Solution: Part (a): Assume that f ≥ 0 and is a continuous function with compact
support. Integration by parts gives
Z ∞ Z ∞
2
T f (s) ds = −2 T f (s)(T f )0 (s)sds.
0 0

Observe that s(T f )0 (s) = f − T (f ), and so we have

Z ∞ Z ∞ Z ∞
2
T f (s) ds = −2 f (s)T f (s)ds + 2 T (f )(s)2 ds.
0 0 0

Rearrangement gives
Z ∞ Z ∞
kT f k2L2 = 2
T f (s) ds = 2 f (s)T f (s)ds ≤ 2kf kL2 kT f kL2 .
0 0

This then gives that kT f kL2 ≤ 2kf kL2 .

√
Part (b): Set fj (x) = j1[0, 1 ] (x). Clearly this family is bounded uniformly in j since
j
kfj kL2 = 1. However, the sequence {T fj } is not Cauchy. Indeed, suppose that k ≥ j,
and consider
Z 1 Z s
√
2
k 1 p
2
kT fj − T fk kL2 ≥ 2
j1[0, 1 ] (t) − k1[0, 1 ] (t)dt ds
0 s
j k
0
Z 1 p √ √ !2
k √ 2 j− k
= j − k dt = √ .
0 k

So we have that when k ≥ j that

√
j
kT fj − T fk kL2 ≥ 1 − √ > 0.
k
So the operator T can not be compact.
 Math 6338 : Real Analysis II
Final Exam
Due: Friday May 4 2012 at 5:00pm
Instructions: Answer all of the problems. You may use course notes, but other sources are
not permitted.

1. Let H be a complex separable Hilbert space. A linear operator T : H → H is completely

continuous if any weakly convergent sequence {xn } is mapped to a strongly convergent
sequence {T xn }.

(a) Show that T is completely continuous if and only if T is compact.

(b) Suppose that T ∗ T is completely continuous. Show that T is completely continuous.

Solution: Part (a): If T is compact, then weakly convergent sequences are mapped to
strongly convergent sequences, as was proved in class. This gives that T is completely
continuous.

Suppose now that T is completely continuous. Since H is separable, the unit ball of
H is weakly compact. Let {xn } be any sequence in the unit ball of H, then there is a
x ∈ H and a subsequence {xnk } such that xnk → x weakly. But since T is completely
continuous we have that
kT xnk − T xkH → 0.
This implies that the image of the unit ball of H is sequentially compact, and so T is a
compact operator.

Part (b): Let {xn } be a weakly convergent sequence in H. Since it is weakly convergent,
it is in fact bounded and so kxn kH ≤ C for all n. Because T ∗ T is completely continuous
we have that {T ∗ T xn } is strongly convergent. We now claim that {T xn } is a Cauchy
sequence. Indeed, since

kT xn − T xm k2H = hT (xn − xm ), T (xn − xm )iH

= hT ∗ T (xn − xm ), xn − xm iH
≤ kT ∗ T (xn − xm )kH kxn − xm kH .
≤ 2C kT ∗ T xn − T ∗ T xm kH .

This last expression can be made small if n, m are large enough, and so we have that
{T xn } is Cauchy in H, and hence converges.

2. Let 1 < p < ∞ and p1 + 1q = 1. Suppose that K(x, y) : Rn × Rn → R is measurable and

K(x, y) is non-negative almost everywhere. If
Z Z
sup K(x, y) dy ≤ A and sup K(x, y) dx ≤ B
x∈Rn Rn y∈Rn Rn
 R
show that T f (x) = Rn
K(x, y)f (y) dy is a bounded operator on Lp (Rn ) and
1 1
kT f kLp (Rn ) ≤ A q B p kf kLp (Rn ) .

Solution: Note that we have

Z Z p
p
kT f kLp (Rn ) = K(x, y)f (y)dy dx
n n
ZR Z R Z p
≤ K(x, y) |f (y)| dy dx
Rn Rn Rn
Z Z Z p
1 1
= K(x, y) K(x, y) |f (y)| dy dx
q p

Rn Rn Rn
Z Z  pq Z 
p
≤ K(x, y)dy K(x, y) |f (y)| dx
Rn Rn Rn
Z Z
p
≤ A q K(x, y) |f (y)|p dydx
n n
ZR R Z 
p
p
= Aq |f (y)| K(x, y)dx dy
Rn Rn
p
≤ A B q kf kpLp (Rn ) .

Taking the p-th root gives the result.

3. Suppose that L is a linear functional on C[0, 1] such that

kLk = L(1) = 1.

Let f ∈ C[0, 1] with 0 ≤ f ≤ 1. Then show that 0 ≤ Lf ≤ 1.

Solution: Let f ∈ C[0, 1] with 0 ≤ f ≤ 1. Set Lf = α + iβ. We want to show that

β = 0 and 0 ≤ α ≤ 1. For all t ∈ R we have that
 
1 1
L f − + it = α − + i(β + t).
2 2
1
We have that f − 2 ∞
≤ 21 , we have that
 2   2 2
1 2 1 1 1
α− + (β + t) = L f − + it ≤ f− ≤ + t2 .
2 2 2 ∞ 4

Rearrangement gives that

α2 − α + β 2 + 2βt ≤ 0 ∀t ∈ R.

So β = 0, and hence α2 ≤ α, or rephrasing, 0 ≤ α ≤ 1 or 0 ≤ Lf ≤ 1.

4. Let X and Y be Banach spaces and let B(·, ·) be separately continuous bilinear mapping
X × Y to C. Namely, for fixed x, the mapping y → B(x, y) is a bounded linear transforma-
tion, and with a similar statement holding for fixed y. Show that B(·, ·) is jointly continuous.
Namely xn → 0 and yn → 0, then B(xn , yn ) → 0.

Solution: Set Tn (y) = B(xn , y). Since for fixed xn the mapping B(xn , ·) is continuous,
each Tn : Y → C is bounded. Because xn → 0 and B(·, y) is bounded, we have that
for all y ∈ Y that {kTn (y)k} is bounded for each fixed y. By the Uniform Boundedness
Principle, there is a constant C such that

kTn (y)k ≤ C kyk ∀n.

Then we have that

|B(xn , yy )| = kTn (yn )k ≤ C kyn k → 0.

5. Let f ∈ S(R).
(a) Show that kf k2L2 (R) ≤ 4π kxf kL2 (R) ξ fˆ .
L2 (R)
2
(b) Show that equality holds in this inequality if and only if f (x) = Ce−ax where a > 0
and C ∈ C.

Solution: Part (a): Note that by integration by parts and direct computation we have
Z Z Z 
2 2 d 2
kf kL2 (R) = |f (x)| dx = − x |f (x)| dx = −2Re f (x)f 0 (x)dx .
R R dx R

Then, applying Cauchy-Schwarz, we have

Z
2
kf kL2 (R) ≤ 2 |x| |f (x)| |f 0 (x)| dx
R
≤ 2 kxf kL2 (R) kf 0 kL2 (R) .

Now apply Plancherel’s Theorem to conclude that

kf 0 kL2 (R) = fˆ0 = 2π ξ fˆ ,

L2 (R) L2 (R)

where for the last equality we have used that fˆ0 (ξ) = 2πiξ fˆ(ξ).

Part (b): In the application of Cauchy-Schwarz we have an equality if and only if

−2axf (x) = f 0 (x).

2
Solving we have f (x) = Ce−ax for some C ∈ C and a ∈ R. The choice of a > 0 is
required for f ∈ L2 (R).
 Homework 1

1. Let u be a semi-inner product on H and put N = {x ∈ H : u(x, x) = 0}. Show that

(a) Show that N is a linear subspace of H;

(b) Show that if hx + N, y + N i ≡ u(x, y) for all x + N and y + N in the quotient space
H/N , then h·, ·i is a well-defined inner product on H/N .

Solution: Part (a): It is easy to see that if x ∈ N then λx ∈ N since u(λx, λx) =
|λ|2 u(x, x) = 0. To see that x, y ∈ N implies that x + y ∈ N , note that it suffices
to show that u(x, y) = 0. But, this follows from Cauchy-Schwarz (which is true for
semi-inner products). By Cauchy-Schwarz we have that
1 1
|u(x, y)| ≤ u(x, x) 2 u(y, y) 2 = 0

with the last equality holding by x, y ∈ N . So we have that u(x, y) = 0, and similarly
u(y, x) = 0. Then note that

u(x + y, x + y) = u(x, x) + u(x, y) + u(y, x) + u(y, y) = 0.

So we have that x + y ∈ N too. These two facts imply the N is a linear subspace of H.

Part (b):
Note that the computations above imply that for n ∈ N and h ∈ H that u(n, h) =
u(h, n) = 0. Suppose that x1 + N = x2 + N and y1 + N = y2 + N in the space H/N ,
i.e., we have that x1 − x2 ∈ N and y1 − y2 ∈ N . To show that the inner product is well
defined, we need to show that u(x1 , y1 ) = u(x2 , y2 ). Now observe

hx1 + N, y1 + N i = u(x1 , y1 ) = u(x2 + x1 − x2 , y1 )

= u(x2 , y1 ) + u(x1 − x2 , y1 )
= u(x2 , y2 + y1 − y2 ) + u(x1 − x2 , y1 )
= u(x2 , y2 ) + u(x2 , y1 − y2 ) + u(x1 − x2 , y1 )
= u(x2 , y2 ) = hx2 + N, y2 + N i .

Here, for the last equality, we used the fact just mentioned. This implies that the inner
product defined on the coset classes is well-defined.
It is easy to see that the pairing defined above is linear in each component. We also
have
hx + N, x + N i = u(x, x) ≥ 0
since u is a semi-inner product. Suppose that we have that,

hx + N, x + N i = u(x, x) = 0

then, by definition we have that x ∈ N and so x + N = N , and the final condition for
an inner product is satisfied.
2. Let H be a Hilbert space and suppose that f, g ∈ H are linearly independent with
kf k = kgk = 1. Show that for all 0 < t < 1 that ktf + (1 − t)gk < 1.

Solution: It is obvious that ktf + (1 − t)gk ≤ t kf k + (1 − t) kgk = 1, for all vectors

(regardless of linear independence). Suppose now that equality held for some 0 < t < 1.
Observe that

ktf + (1 − t)gk2 = t2 + (1 − t)2 + t(1 − t) hf, gi + t(1 − t) hg, f i

= 1 − 2t + 2t2 + t(1 − t) hf, gi + t(1 − t) hg, f i .

Then we would have that

0 = ktf + (1 − t)gk2 − 1 = 2t(1 − t) (Re hf, gi + 1) .

This implies that Re hf, gi = −1 since 0 < t < 1. This then gives that

1 = |Re hf, gi| ≤ |hf, gi| ≤ kf k kgk = 1.

So we have that |hf, gi| = 1 = kf k kgk, i.e., equality holds in the Cauchy-Schwarz
inequality. But, this then implies that f and g must be linearly dependent, which is a
contradiction to the hypothesis. Thus, we must have the strict inequality for all 0 < t < 1,
i.e.,
ktf + (1 − t)gk < 1.

3. Let K be a non-empty linear subspace of a Hilbert space H. Show that K is dense in H

if and only if K ⊥ = {0}.

Solution: Let x ∈ K ⊥ and assume that K is dense in H. Since K is dense, there exists
a sequence {xn } ∈ K such that x → x. We have that since x ∈ K ⊥ and xn ∈ K that

hxn , xi = 0.

By Problem 8 we have that x ⊥ x, but this implies that hx, xi2 = 0, or x = 0. This then
shows that K ⊥ = {0}.
Now suppose that K ⊥ = 0. If x ⊥ K, then we have that x ⊥ K and so x ∈ K ⊥ and
⊥
thus x = 0. This then implies that K = {0}. Now we can decompose the Hilbert space
⊥
H = K ⊕ K = K and so K is dense in H. (Here one should use the machinery we
developed on projections).

4. For the Hilbert space H = `2 (N ∪ {0}),

P∞ n
(a) If |λ| < 1, and Lλ h = n=0 λ hn for all h ∈ H, find the vector h̃ ∈ H such that
Lh = hh, h̃i for all h ∈ H.
(b) What is the norm of the linear functional defined in part (b)?
 Solution: It is easy to see that we have
1
|Lλ (h)| ≤ khk`2 q .
2
1 − |λ|

Then Lλ is a bounded linear functional on `2 (N ∪ {0}), so by the Riesz representation

theorem, there exists a vector h̃ so that
D E
Lλ (h) = h, h̃ .
`2
 2

From this equality, it is immediate that h̃ = 1, λ, λ , . . . , and clearly belongs to `2 with
norm given by
∞
2 X 1
h̃ = |λ|2k = .
`2
k=0
1 − |λ|2
Note that these computations also give that kLλ k = √ 1
.
1−|λ|2

5. Suppose that H is a Hilbert space over the real numbers R. Show that kxk = kyk implies
that
hx + y, x − yi = 0.

Solution: We expand the inner product to find

hx + y, x − yi = kxk2 − kyk2 − hx, yi + hy, xi = kxk2 − kyk2

since we are working in a real vector space. But, then the hypothesis that kxk = kyk
gives the result.

6. Show that in an inner product space H that

2
1 1
kz − xk2 + kz − yk2 = kx − yk2 + 2 z − (x + y) .
2 2
This is called Apollonius’ Identity.

Solution: This is a straight forward computation: Just expand both sides. Exact details
left to the student.

7. If in a Hilbert space we have xn → x and yn → y, show that

hxn , yn i → hx, yi .
 Solution: Now that we have

|hxn , yn i − hx, yi| = |hxn , yn i − hxn , yi + hxn , yi − hx, yi|

≤ |hxn , yn i − hxn , yi| + |hxn , yi − hx, yi|
≤ kxn k kyn − yk + kxn − xk kyk .

Now for the second term above, this clearly goes to zero since xn → x. For the first term,
note that since xn → x, we have that kxn k → kxk, so in particular, supn kxn k < +∞.
Then we can simply conclude that this term goes to zero since yn → y.

8. Show that y ⊥ xn and xn → x implies that x ⊥ y.

Solution: Observe that

hx, yi = hx, yi − hxn , yi = hx − xn , yi

An application of Cauchy-Schwarz then gives

|hx, yi| ≤ kx − xn k kyk .

But since xn → x, we must have that |hx, yi| = 0, which gives the result.

9. Show that for a sequence {xn } in a Hilbert space that the conditions kxn k → kxk and
hxn , xi → hx, xi implies that xn → x.

Solution:

kxn − xk2 = kxn k2 + kxk2 − hxn , xi − hx, xn i .

This then clearly gives the result, the hypotheses imply that the right hand side goes to
zero.

10. Let {ek } be any orthonormal sequence in a Hilbert space H. Show that for any x, y ∈ H
that ∞
X
|hx, ek i hy, ek i| ≤ kxk kyk
k=1

Solution: Note that by Bessel’s Inequality, we have that

∞
X
kxk2 ≥ |hx, ek i|2 .
k=1
 The result then follow easily from Cauchy-Schwarz,

∞ ∞
! 21 ∞
! 21
X X X
|hx, ek i hy, ek i| ≤ |hx, ek i|2 |hy, ek i|2
k=1 k=1 k=1
≤ kxk kyk .

11. Show that an element x of a Hilbert space can not have “too many” Fourier coefficients
hx, ek i that are “big”. Namely if {ek } is an orthonormal sequence, show that the number
N () of hx, ek i such that
|hx, ek i| > 
kxk2
must satisfy N () < 2
.

Solution: Note that by Bessel’s Inequality, we have that

∞
X
2
kxk ≥ |hx, ek i|2 .
k=1

Now, if we sum over only those elements such that |hx, ek i| > , then we clearly have

X ∞
X
2
N () ≤ 1≤ |hx, ek i|2 ≤ kxk2 .
k:|hx,ek i|> k=1

Rearrangement gives the result.

 Homework 2

1. Let X and Y be Hilbert spaces over C. Then a sesquilinear form h on X ×Y is a mapping

h : X × Y → C such that for all x1 , x2 , x ∈ X, y1 , y2 , y ∈ Y and all scalars α, β ∈ C we have

(a) h(x1 + x2 , y) = h(x1 , y) + h(x2 , y);

(b) h(x, y1 + y2 ) = h(x, y1 ) + h(x, y2 );

(c) h(αx, y) = αh(x, y);

(d) h(x, βy) = βh(x, y).

A sesquilinear form is bounded if |h(x, y)| ≤ C kxk kyk and the norm of the form h is given
by
|h(x, y)|
khkX×Y →C = sup = sup |h(x, y)| .
x6=0,y6=0 kxk kyk kxk=1,kyk=1

Show that if h is a bounded sesquilinear form on the Hilbert spaces X and Y , then h has
the representation
h(x, y) = hSx, yiY
where S : X → Y is a bounded linear operator. Moreover, S is uniquely determined by h
and has norm
kSkX→Y = khkX×Y →C .

Solution: The idea is to use the Riesz Representation to construct the linear operator
S. Fix x ∈ H and consider the functional on the space Y given by

L(y) = h(x, y).

By the properties of h being a sesquilinear form, we have that L is linear in y, and if h

is a bounded bilinear form, we have that L is a bounded linear functional on Y . By the
Riesz Representation Theorem, we have then that there exists a unique element z ∈ Y
such that
L(y) = hy, ziY ,
or in terms of h that
h(x, y) = hz, yiY .
While the point z is unique, it is depending on the point x that was fixed. We now define
the map S : X → Y by Sx = z. This map is clearly well-defined since the point z is well
defined given the x. Substituting in this representation we find,

h(x, y) = hSx, yiY

as desired. It remains to show that S is linear, bounded and unique.

 To see that S is linear, we simply use the linearity of h in the first variable. Let α, β ∈ C
and x1 , x2 ∈ X, then we have

hS(αx1 + βx2 ), yiY = h(αx1 + βx2 , y)

= αh(x1 , y) + βh(x2 , y)
= α hSx1 , yiY + β hSx2 , yiY
= hαSx1 , yiY + hβSx2 , yiY
= hαSx1 + βSx2 , yiY .

Since this holds for all y ∈ Y we have that

αSx1 + βSx2 = S(αx1 + βx2 )

and so S is linear.

To see that S is bounded, first note

kSxkY
kSkX→Y = sup
x6=0 kxkX
|hSx, SxiY |
= sup
x6=0,Sx6=0 kxkX kSxkY
|hSx, yiY |
≤ sup
x6=0,y6=0 kxkX kykY
|h(x, y)|
= sup = khkX×Y →C
x6=0,y6=0 kxkX kykY

In particular we have kSkX→Y ≤ khkX×Y →C , and so S is bounded. To see the other

inequality, we have

|h(x, y)| = |hSx, yiY | ≤ kSxkY kykY ≤ kSkX→Y kykY kxkX

which gives khkX×Y →C ≤ kSkX→Y .

For uniqueness, suppose that there are two linear operators S1 and S2 such that

h(x, y) = hS1 x, yiY = hS2 x, yiY .

This yields that S1 x = S2 x for all x ∈ X, and so S1 = S2 .

2. Suppose that H is Hilbert space that contains an orthonormal sequence {ek } which is
total in H. Show that H is separable (it has a countable dense subset).
 Solution: Let A denote the set of all linear combination of the form
X
γkj ekj
J

where J is a finite set, kj ∈ J, and γkj = akj + ibkj , where akj , bkj ∈ Q. It is clear that
A is countable. We now will show that A is the countable dense subset in H that we seek.

We need to show that for every x ∈ H and  > 0 there is a v ∈ A such that

kx − vkH < 

Since the sequence {ek } is total in H, there exists an integer n such that Yn =
span{e1 , . . . , en } contains aPpoint y whose distance in x is less than 2 . In particular, by
Parseval, we can take y = nk=1 hx, ek iH ek and have
n
X 
x− hx, ek iH ek < .
k=1
2
H

Now for each coefficient hx, ek iH , we can find an element γk ∈ C whose real and
imaginary parts are rational such that
n
X 
(hx, ek iH − γk ) ek <
k=1
2
H
Pn
Then define v = k=1 γk ek ∈ A, and note that
n
X
kx − vkH = x− γk ek
k=1 H
Xn n
X
≤ x− hx, ek iH ek + (hx, ek iH − γk ) ek
k=1 H k=1 H
< .

So we have that A is dense in H and A is countable, so H is separable.

3. If p is a sublinear functional on a real vector space X, show that there exists a linear
functional f˜ on X such that −p(−x) ≤ f˜(x) ≤ p(x).

Solution: Note that by Hahn-Banach we have that for any linear functional f with
f (x) ≤ p(x) we have a corresponding extension f˜(x) that satisfies f˜(x) ≤ p(x).
Now simply note that if we evaluate this expression at −x then we have
−f˜(x) = f˜(−x) ≤ p(−x)
and so we have f˜(x) ≥ −p(−x). Combining these inequalities gives the result.
4. If x in a normed space X such that |f (x)| ≤ c for all f ∈ X ∗ of norm at most 1. Show
that kxk ≤ c.

Solution: We have the following fact at our disposal:

|f (x)|
kxkX = sup
f ∈X ∗ ,f 6=0 kf kX→C

. It is easy to show that

|f (x)|
sup = sup |f (x)|
f ∈X ∗ ,f 6=0 kf kX→C f ∈X ∗ ,kf kX→C =1

The result then follows easily, since

kxkX = sup |f (x)| ≤ c

f ∈X ∗ ,kf kX→C =1

5. Show that for any sphere centered at the origin with radius r, Sr (0), in a normed space
X and any point y ∈ Sr (0) there is a hyperplane Hy 3 y such that Br (0) lies entirely in one
of the two half spaces determined by the hyperplane Hy .

Solution: Let y ∈ Sr (0), and let Hy be the hyperplane containing y. As X is a normed

space, it is possible to write X = Hy ⊕ Ry (here we view things as a real vector space, the
complex case is not any different). The space Ry is the span of the vector y. Note that
the functional f : X → R given by fy (h+ry) = r has kernel Hy . This algebraic reasoning
can be turned around. From this decomposition of the space X, we can then see that a
hyperplane is given by the level sets of a linear functional, i.e., Hy = fy−1 (0) = ker fy .
Now consider the bounded linear functional f : Ry → R given by f (ry) = r. Then,
we can extend this linear functional to all of X by Hahn-Banach. Then we have that
Hy = f −1 (0), and it is easy to see that the set Br (0) must lie in one of the half-spaces.
It isn’t too much more work to show that the same conclusion in fact holds when one
is given a non-empty convex subset C of X. The interested student should attempt to
work this out.

6. Of what category is the set of all rational numbers Q in R?

Solution: The rationals Q are meagre in R. Note that the rationals Q are countable.
Let q1 , q2 , . . . be an enumeration of Q, i.e., we can write Q = ∞
S
k=1 k Then it is clear
q .
that each set {qk } is nowhere dense. Thus, we have written Q as a countable union of
nowhere dense sets in R.
7. Show that the complement M c of a meager subset M of a complete metric space X is
non-meager.

Solution: Let M be a meager subset of a complete metric space X. Suppose, for a

contradiction, that M c is also meager. Then we can write
∞
[ ∞
[
c
M = Ak M= Bk
k=1 k=1

where each Ak and Bk is nowhere S dense inSX.

But we have that X = M M = ∞ c
S∞ S∞
k=1 Ak ∪ k=1 Bk = k=1 Ck . So we have
written X as a countable union of nowhere dense sets, and so X would be of the first
category. But since X is complete, it must be of the second category, and so we have a
contradiction. This then implies that M c is non-meager.

8. Let X and Y be Banach spaces. Suppose that Tn ∈ B(X, Y ) is such that supn kTn k = +∞.
Show that there is a point x0 ∈ X such that supn kTn x0 kY = +∞. The point x0 is called a
point of resonance.

Solution: Suppose that there does not exist a point x0 ∈ X with the desired conclusion.
Then for all x ∈ X we have that

sup kTn xkY < +∞

n

But, then the operators Tn satisfy the hypotheses of the Uniform Boundedness Principle,
so we have that the sequence of norms kTn k is finite, i.e.,

sup kTn k < ∞.

n

However, this clearly contradicts the hypotheses of the operators Tn , and so we must
have that there exists x0 ∈ X such that

sup kTn x0 kY = +∞
n

9. If X and Y are Banach spaces and Tn ∈ B(X, Y ), show that the following are equivalent:

(a) {kTn k} is bounded;

(b) {kTn xk} is bounded for all x ∈ X;

(c) {|g(Tn x)|} is bounded for all x ∈ X and for all g ∈ Y ∗ .

 Solution: It is immediate that (a) implies (b) and (b) implies (c). So it suffices to prove
that (c) implies (a). But, two applications of the Uniform Boundedness Principle gives
this result. The first application of UBP implies that the sequence {Tn x} is bounded for
all x. This is problem 10 below. Then we can simply apply the standard version of the
UBP to get that {kTn k} is bounded.

10. If {xn } is a sequence in a Banach space is such that for all f ∈ X ∗ the sequence {f (xn )}
is bounded, show that {kxn k} is bounded.

Solution: Define an operator Tn : X ∗ → C by Tn (f ) = f (xn ), then we have that Tn is

a linear operator on the space X ∗ . By the hypothesis, we have that

sup |Tn (f )| = sup |f (xn )|

n n

is bounded for all f ∈ X ∗ . Then, by the Uniform Boundedness Principle, we have there
there exists some universal constant C such that

kTn kX ∗ →C ≤ C.

This implies that for all f ∈ X ∗ of norm at most 1 that

sup |f (xn )| = sup |Tn (f )| ≤ C

f ∈X ∗ ,kf kX ∗ ≤1 f ∈X ∗ ,kf kX ∗ ≤1

But, by the Theorem in class, we have that

kxn kX = sup |f (xn )| ≤ C

f ∈X ∗ ,kf kX ∗ ≤1

which gives that {kxn kX } is bounded.

 Homework 3

1. If the inverse T −1 of a closed linear operator exists, show that T −1 is a closed linear
operator.

Solution: Assuming that the inverse of T were defined, then we will have to have that
D(T −1 ) = Ran(T ). Suppose that {un } ∈ D(T −1 ) is a sequence such that un → u and
T −1 un → x. We need so that that u ∈ D(T −1 ) and T −1 u = x. Then by the simple
criteria, we would have that T −1 is closed.

Since un ∈ D(T −1 ) we have that un = T xn , where xn ∈ D(T ). Also, note we have that
T −1 un = xn → x. So, we have a sequence of vectors {xn } ∈ D(T ) with xn → x and
T xn → u. Since T is closed, we have that x ∈ D(T ) and T x = u. But, this last equality
implies that u ∈ D(T −1 ) and that x = T −1 u, which is what we wanted.

2. Let T be a closed linear operator. If two sequences {xn } and {x̃n } in the domain of T
both converge to the same limit x, and if {T xn } and {T x̃n } both converge, show that {T xn }
and {T x̃n } both have the same limit.

Solution: Suppose that {xn }, {x̃n } ∈ D(T ) are such that xn → x and x̃n → x. Let
T xn → y and T x̃n → ỹ. We need to show that y = ỹ.

Since T is closed, then we know that for xn → x and T xn → y implies that x ∈ D(T )
and T x = y. Similarly, we have that x̃n → x and T x̃n → ỹ implies that x ∈ D(T ) and
T x = ỹ. But, this then gives y = T x = ỹ, and we they have the same limit as claimed.

3. Let X and Y be Banach spaces and T : X → Y an injective bounded linear operator.

Show that T −1 : Ran T → X is bounded if and only if Ran T is closed in Y .

Solution: First suppose that Ran T is closed in Y . By problem 1 we know that T −1 is

closed and D(T −1 ) = Ran(T ). Then by the Closed Graph Theorem, we have that T −1
is a bounded linear operator from Ran T → X.

Now, suppose that T −1 : Ran T → X is bounded, i.e., kT −1 ykX ≤ c kykY . The goal
is to show that y ∈ Ran T . Let y ∈ Ran T , and then there exists xn ∈ X such that
T xn := un → y. We then have that the sequence {xn } is Cauchy since

kxn − xm kX = T −1 T xn − T −1 T xm X
= T −1 (un − um ) X
≤ c kun − um kY

and since un → y this gives that the claim. Since X is complete, let x denote the limit
of the sequence {xn }. Then we will show that y = T x ∈ Ran T , and so the range of T is
 closed. But, note that

ky − T xkY = ky − un + un − T xkY
≤ ky − un kY + kun − T xkY
≤ ky − un kY + kT k kxn − xkX .

Here we have used that T is bounded. But, the expression on the right hand side can be
made arbitrarily small, and so y = T x as desired.

4. Let T : X → Y be a bounded linear operator, where X and Y are Banach spaces. If T is

bijective, show that there are positive real numbers a and b such that

a kxkX ≤ kT xkY ≤ b kxkX ∀x ∈ X.

Solution: Since T is a bounded linear operator, then we immediately have that

kT xkY ≤ kT k kxkX ∀x ∈ X.

However, since T is bijective and bounded we have that T −1 is bounded as well. Note
that T −1 : Y → X, so for any y ∈ Y we have that

T −1 y X
≤ T −1 kykY .

Apply this when y = T x, and we then have

kxkX ≤ T −1 kT xkY .

Combining inequalities we have that

a kxkX ≤ kT xkY ≤ b kxkX ∀x ∈ X

−1
with a = kT −1 k and b = kT k.

5. Let X1 = (X, k·k1 ) and X2 = (X, k·k2 ) be Banach spaces. If there is a constant c such
that kxk1 ≤ c kxk2 for all x ∈ X, show that there is a constant C such that kxk2 ≤ C kxk1
for all x ∈ X.

Solution: Consider the map I : X2 → X1 given by Ix = x. Then I is clearly linear

from X2 → X1 , and by the hypotheses, we have

kIxk1 ≤ c kxk2 ∀x ∈ X

so I is bounded. But, we also have that I is bijective, and so by the Bounded Inverse
Theorem, we have that I −1 : X1 → X2 is a bounded linear operator as well, i.e., there
 exists a constant C such that

I −1 x 2
≤ C kxk1 ∀x ∈ X.

However, I −1 x = x and so
kxk2 ≤ C kxk1 ∀x ∈ X.

6. Let X be the normed space whose points are sequences of complex numbers x = {ξj }
with only finitely many non-zero terms and norm defined by kxk = supj |ξj |. Let T : X → X
be defined by    
1 1 1
T x = ξ1 , ξ2 , ξ3 , . . . = ξj .
2 3 j
Show that T is linear and bounded, but T −1 is unbounded. Does this contradict the Bounded
Inverse Theorem?

Solution: It is pretty obvious that T is linear. Since x = {ξn } and y = {ηn }, then we
have that      
1 1 1
T (x + y) = (ξj + ηj ) = ξj + ηj = T x + T y.
j j j
The case of constants λ, is just as easy to verify, T λx = λT x. It is also very easy to see
that T is bounded, since we have for any x ∈ X that

|ξj | 1
kT xkX = sup ≤ kxkX sup ≤ kxkX .
j≥1 j j≥1 j

We now show that T −1 is not bounded. First, note that T −1 is given by

T −1 x = {jξj } .

Now let x = {ξj } where ξj = 0 if j 6= n and 1 if n = j. Then we have that kxkX = 1

and T −1 x = {jξj } = (0, . . . , 0, n, 0, . . . , ) where the value 1 occurs in the nth coordinate.
This then implies that kT −1 xkX = n. But, this is enough to show that the operator
is not bounded. This doesn’t contradict the Bounded Inverse Theorem since X is not
complete.

7. Give an example to show that boundedness need not imply closedness.

Solution: Let T : D(T ) → D(T ) ⊂ X be the identity operator on D(T ), where D(T ) is
a proper dense subspace of the normed space X. Then it is immediate that T is linear
and bounded. However, we have that T is not closed. To see this, take x ∈ X \ D(T )
and a sequence {xn } ∈ D(T ) that converges to x, and then use the simple test for an
operator to be closed that we gave in class.
8. Show that the null space of a closed linear operator T : X → Y is a closed subspace of X.

Solution: Recall that the null space of a linear operator is given by Null(T ) := {x ∈
X : T x = 0}. Let x ∈ Null(T ) and suppose that xn ∈ Null(T ) is a sequence such that
xn → x. We need to show that x ∈ Null(T ).

If we have that T xn → y, then since each xn ∈ Null(T ) we have that y = 0. But

since T is closed, then we have that T x = y = 0 and so x ∈ Null(T ).

9. Let H be a Hilbert space and suppose that A : H → H is everywhere defined and

hAx, yiH = hx, AyiH ∀x, y ∈ H.

Show that A is a bounded linear operator.

Solution: It suffices to show that the graph of A is closed. So suppose that we have
(xn , Axn ) → (x, y). Then we need to show that y = Ax. To accomplish this last equality,
it is enough to show that for all z ∈ H that

hz, yiH = hz, AxiH .

But, with this observation the problem becomes rather easy:

hz, yiH = lim hz, Axn iH

n
= lim hAz, xn iH
n
= hAz, xiH
= hz, AxiH .

10. Let {Tn } be a sequence of bounded linear transformations from X → Y , where X and
Y are Banach spaces. Suppose that kTn k ≤ M < ∞ and there is a dense set E ⊂ X such
that {Tn x} converges for all x ∈ E. Show that {Tn x} converges on all of X.

Solution: Let  > 0 be given. Pick any x ∈ X. Now choose x0 ∈ E such that

kx − x0 kX < 3M , which will be possible since E is dense in X. For this x0 we have
that Tn x0 → y. So there exists an integer N such that if n, m > N then

kTn x0 − Tm x0 kY < .
3
Now, we claim that {Tn x} is Cauchy in Y , and since Y is a Banach space it is complete
and so the result follows. To see that {Tn x} is Cauchy, when n, m > N we have

kTn x − Tm xkY = kTn x − Tn x0 + Tn x0 − Tm x0 + Tm x0 − Tm xkY

≤ kTn x − Tn x0 kY + kTn x0 − Tm x0 kY + kTm x − Tm x0 kY

< kTn k kx − x0 kX + kTm k kx − x0 kX +
3
 
< 2M + = .
3M 3
 Homework 4

1. Let X and Y be normed spaces, T ∈ B(X, Y ) and {xn } a sequence in X. If xn → x

weakly, show that T xn → T x weakly.

Solution: We need to show that

g(Tn x) → g(T x) ∀g ∈ Y ∗ .

It suffices to do this when kgkY ∗ = 1. Observe that

|g(T xn ) − g(T x)| = |g(T xn − T x)|

≤ kgkY ∗ kT xn − T xkY
≤ kgkY ∗ kT k kxn − xkX
= kT k kxn − xkX .

Note that then we have

|f (xn − x)|
kxn − xkX = sup
06=f ∈X ∗ kf kX ∗

So, let  > 0 be given, and select f ∈ X ∗ with kf k∗X = 1 such that

kxn − xkX < |f (xn − x)| + .
2 kT k

Since xn → x weakly, there exists an integer N such that if n > N we have


|f (xn ) − f (x)| = |f (xn − x)| < .
2 kT k

This then gives,

 

|g(T xn ) − g(T x)| ≤ kT k kxn − nkX ≤ kT k |f (xn − x)| + < .
2 kT k

2. If {xn } is a weakly convergent sequence in a real normed space X with xn → x0 . Show

that there is a sequence {ym } of linear combinations of elements of {xn } which converge
strongly to x0 .

Hint: Prove that the weak closure of a convex set is the same as the strong closure.

Solution: Note that if suffices to prove that the following statement. If E ⊂ X is a

convex subset of a normed space, then the weak closure of E, denoted by E w is the same
as the strong closure of E, denoted by E. Once we have this statement, we can apply it to
 E = span{xn } in the following way. We have that x0 ∈ E w by hypothesis. But, E w = E
by the claim. So we have that x ∈ E, which when we unpack the definitions gives the
existence of {ym } a sequence of linear combinations of {xn } that converges strongly to x0 .

So it suffices to prove the claim. Since strong convergence implies weak convergence,
we have that E ⊂ E w . Suppose that x0 6∈ E. By your exam, there exists a functional
f ∈ X ∗ such that
f (x0 ) < δ ≤ f (x) ∀x ∈ E.
Consider the set {x : f (x) < δ}, this is a weak open set of x0 that does not intersect E,
/ E w . Which gives E w ⊂ E.
and so x0 ∈

3. Let Tn ∈ B(X, Y ). Show that Tn → T uniformly if and only if for every  > 0 there is an
N , depending only on  such that for all n > N and all x ∈ X of norm 1 we have
kTn x − T xkY < .

Solution: Suppose first that Tn → T uniformly. Then we have that for any  > 0 there
exists N such that
kTn − T kX→Y < .
Then we clearly have that

kTn x − T xkY ≤ kTn − T kX→Y kxkX = kTn − T kX→Y < 

when n > N and x ∈ X with kxkX = 1.

For the other direction, suppose that  > 0 is given, then we need to show that

kTn − T kX→Y < 

when n > N (). Note that

kTn x − T xkY
kTn − T kX→Y = sup
06=x∈X kxkX

But, by the hypothesis, we have that

kTn x − T xkY <  kxkX

when n > N (). This then clearly implies that

kTn x − T xkY
kTn − T kX→Y = sup ≤
06=x∈X kxkX

which proves the result.

4. If a normed space X is reflexive, show that X ∗ is reflexive.

 Solution: Let h ∈ X ∗∗∗ , then for every g ∈ X ∗∗ there is an x ∈ X such that g = Cx
since X is reflexive. Here C is the canonical map identifying X and X ∗∗ . Hence,

h(g) = h(Cx) := f (x)

defines a bounded linear functional f on X, and C̃f = h, where C̃ : X ∗ → X ∗∗∗ is the

canonical mapping. Thus, C̃ is surjective and so X ∗ is reflexive.

5. Let X be a normed space and {fn } a sequence of linear functionals on X. Show

(a) If fn → f weakly, then fn → f weak* too;
(b) Show that if X is reflexive, then the converse holds.

Solution: Part (a) is immediate since we have that X ⊂ X ∗∗ where we have used
the canonical mapping. Namely, each x ∈ X defines a functional x̂ : X ∗ → C by
x̂(f ) := f (x). So, if we have weak convergence of the sequence of functionals, then we
have weak-* convergence as well.

If X is reflexive, then we have that X ∗∗ = X, and so every linear functional on X ∗ can

be identified with a functional of the form x̂.

6. A weak Cauchy sequence in a normed space is a sequence {xn } in X such that for every
f ∈ X ∗ the sequence {f (xn )} is Cauchy. Show that a weak Cauchy sequence is bounded.

Solution: Define x̂n : X ∗ → C by x̂n (f ) = f (xn ). Since {xn } is a weak Cauchy sequence,
then we have that {f (xn )} is a Cauchy sequence in C and hence bounded. Thus, we
have that {x̂n (f )} is a bounded sequence, and since X ∗ is a Banach space we can apply
the Uniform Boundedness Principle to conclude that

sup kx̂n k < ∞

n

But, this then gives that supn kxn kX < ∞, which is what we wanted.

7. A normed space X is said to be weakly complete if each weak Cauchy sequence in X

converges weakly in X. If X is reflexive, show that X is weakly complete.

Solution: Let {xn } be a weak Cauchy sequence in X. Again, we will identify xn with x̂n
via the canonical mapping. Note that by Problem 6 above, we have that supn kxn kX < ∞.
Since {f (xn )} is Cauchy in C, it has a limit, and we can define

ϕ(f ) = lim f (xn ) = lim x̂n (f )

n→∞ n→∞
 which exists for each f ∈ X ∗ . This defines a map ϕ : X ∗ → C. This map is clearly linear
since each x̂n is linear. Also, it is bounded since

|x̂n (f )| = |f (xn )| ≤ kf kX ∗ kxn kX ≤ sup kxn k kf kX ∗ .

n

Passing to the limit we have

|ϕ(f )| ≤ sup kxn k kf kX ∗ .

n

So we have that ϕ ∈ X ∗ ∗. But, since X is reflexive, there exists x ∈ X such that x̂ = ϕ.

But, then we have

f (x) = x̂(f ) = ϕ(f ) = lim x̂n (f ) = lim f (xn )

n→∞ n→∞

so that xn → x weakly since f (xn ) → f (x) for all f ∈ X ∗ .

8. Put fn (t) = eint , t ∈ [−π, π]. Let Lp = Lp (−π, π) with respect to Lebesgue measure. If
1 ≤ p < ∞, show that fn → 0 weakly in Lp , but not strongly.

Solution: Note that if p is a polynomial on [−π, π] then we clearly have that

hp, fn iL2 → 0.

The polynomials are dense in Lp when 1 ≤ p < ∞, so for any f ∈ Lp , select a polynomial
p such that kf − pkLp < . Then we have that

hf, fn iL2 = hf − p, fn iL2 + hp, fn iL2 .

The second term goes to zero as n → ∞, in particular if n > N we have |hp, fn iL2 | <  .
For the first term, note that we have

|hf − p, fn iL2 | ≤ kf − pkp kfn k q = kf − pkp < 

1 1
where p
+ q
= 1. Thus for n > N we have that

|hf, fn iL2 | < 2.

It is immediate that fn does not converge strongly to 0 in Lp when 1 ≤ p < ∞ since each
element has norm 1.

9. A subset of a Banach space X is weakly bounded if for all f ∈ X ∗ , supx∈S |f (x)| < ∞.
Prove that S is weakly bounded if and only if it is strongly bounded, i.e., supx∈S kxkX < ∞.
 Solution: It is clear that if the set S is strongly bounded, then it must be weakly
bounded since

sup |f (x)| ≤ sup kf kX ∗ kxkX = kf kX ∗ sup kxkX < ∞.

x∈S x∈S x∈S

Suppose now that S is weakly bounded. Then we have that

sup |f (x)| < ∞

x∈S

for all f ∈ X ∗ . Define x̂ : X ∗ → C by x̂(f ) = f (x). Then, we have that

sup |x̂(f )| = sup |f (x)| < ∞.

x∈S x∈S

So, we have that for all x ∈ S that x̂ is a uniformly bounded family of operators from
X ∗ → C. Hence by the Uniform Boundedness Principle, we have that

sup kx̂kX ∗∗ < ∞

x∈S

But kxkX = kx̂kX ∗∗ , and so we have the result.

10. If {xn } ⊂ `1 show that ∞

P
j=1 y(j)xn (j) → 0 for every y ∈ c0 if and only if supn kxn k`1 <
∞ and xn (j) → 0 for all j ≥ 1.

Solution: Note that we have c∗0 = `1 . For notational simplicity, let hxn , yi denote the
infinite sum ∞
X
xn (j)y(j).
j=1

The easy direction then follows fromPtesting on an obvious element in c0 and the Uniform
Boundedness Principle. If we have ∞ j=1 y(j)xn (j) → 0 for every y ∈ c0 , then taking

y = (0, . . . , 0, 1, 0, . . .)
with the one occurring in the jth position, we have that
∞
X
xn (j) = y(j)xn (j) → 0.
n=0

Since xn ∈ `1 , then it is a bounded linear functional on so we have that |hxn , yi| ≤

kxn k`1 kyk`∞ . The Uniform Boundedness Principle gives that
sup kxn k`1 < ∞.
n

Conversely,
P∞ if we have supn kxn k`1 < ∞ and xn (j) → 0 for all j ≥ 1, we need to show
j=1 y(j)xn (j) → 0. Fix y ∈ c0 and note that for any  > 0 there exists J such that

|y(j)| < .
Now we have that
∞
X J−1
X ∞
X
y(j)xn (j) = y(j)xn (j) + y(j)xn (j).
j=1 j=1 j=J

For the second term, we have that

∞
X ∞
X
y(j)xn (j) ≤ |y(j)| |xn (j)| ≤  kxn k`1 ≤  sup kxn k`1 .
n
j=J j=J

This can be made as small as we wish by choosing  appropriately. Now for the first
term, we have that xn (j) → 0 for j = 1, . . . , J − 1. Select N so that |xn (j)| ≤ J . Then
we have that
J−1
X J−1
X
y(j)xn (j) ≤ kyk`∞ |xn (j)| ≤ kyk`∞ .
j=1 j=1

Combining these two estimates gives that hxn , yi → 0 as n → ∞.

 Homework 5

1. Let H be the Hilbert space `2 . Let Sn : `2 → `2 be defined by:

Sn (x) = (0, . . . , 0, x1 , . . . , xn , . . .)

where we have shifted the sequence x to the right by n zeros. Show that Sn∗ → 0 strongly.

Solution: Note that

∞
X
hSn x, yi`2 = xj y j+n .
j=1

This then implies that for any x ∈ `2 that we have

Sn∗ x = (xn , xn+1 , . . .).

This representation then shows that

∞
X
k(Sn∗ − 0)xk2`2 = kSn∗ xk2`2 = |xj |2 → 0
j=n

since it is the tail of a convergent series. This gives the desired claim.

2. Let X and Y be Banach spaces, and suppose that T ∈ B(X, Y ) has a bounded inverse.
Show that (T −1 )∗ = (T ∗ )−1 .

Solution: We have that T T −1 = I and so I = I ∗ = (T −1 )∗ T ∗ . Similarly, T −1 T = I and

so I = I ∗ = T ∗ (T −1 )∗ , Uniqueness of inverses gives that

(T −1 )∗ = (T ∗ )−1 .

3. Show that T → T ∗ is continuous in the weak operator topology.

Solution: We need to show that if {Tn } is a sequence of operators with Tn → T then

hTn∗ x, yi → hT ∗ x, yi .

(Here we are using the ‘pairing’ notation to make sense of the action on linear functionals).
But, we then have
hTn∗ x, yi = hx, Tn yi → hx, T yi = hT ∗ x, yi .
4. Let S, T ∈ B(X) where X is a Banach space. Suppose that T is invertible in B(X) with
kT −1 k = α1 and S ∈ B(X) with kSk = β < α.
(a) Show that (S + T )−1 ∈ B(X) and find an estimate on the norm of (S + T )−1 .
(b) Show that the set of all invertible operators on B(X) is open and that the map T → T −1
is a homeomorphism

Solution: Note that we have

S + T = T (T −1 S + I) and (S + T )−1 = T −1 (I + (T −1 S))−1 .

So (S + T )−1 will be invertible if (I − (T −1 S))−1 exists. Let A = −T −1 S, and note that

kAk ≤ kT −1 k kSk = αβ by the hypotheses. Since β < α we have that αβ < 1, and so
kAk < 1. Thus we have that (I − T −1 S) is invertible, and
∞
X βj α
(I − T −1 S)−1 ≤ = .
j=0
αj α−β

We also then have that

1
(S + T )−1 ≤ T −1 (I − T −1 S)−1 ≤ .
β−α

For part (b), note that part (a) implies that the set of invertibles is an open set. Since
−1
if T is invertible, then for all operators S such that kSk < kT −1 k we have that T − S
−1
is invertible, i.e. the ball of radius kT −1 k about the operator T contains invertible
elements. Finally note that the map T → T −1 is its own inverse and maps the set of
invertible elements onto itself. It suffices to show that T → T −1 is continuous. Using the
computations from above one can see that
∞
X
−1 −1 −1 −1 −1 −1
T − (T + S) =T (I − (I + ST )) =T (−1)j+1 (ST −1 )j .
j=1

The computations above then show that if kSk is small enough (depending on kT −1 k)
then the right hand side can be made sufficiently small, and so the mapping T → T −1 is
continuous.

5. For the identity operator on a normed space X, find the eigenvalues and eigenspaces as
well as σ(I) and Rλ (I).

Solution: Clearly we have that

Rλ (I) = (I − λI)−1 = (1 − λ)−1 I.

So, Rλ (I) exists, defines a bounded operator, and has dense domain when λ 6= 1. This
clearly gives σ(I) = {1}.
 Also, it is easy to see that σp (I) = {1}, and so the eigenvalue of I is 1 with corresponding
eigenspace X.

6. Show that for a given linear operator T the sets ρ(T ), σp (T ), σc (T ) and σr (T ) are mutu-
ally disjoint and their union is C.

Solution: It is obvious that we have ρ(T )∩σ(T ) = ∅ and ρ(T )∪σ(T ) = C. So it suffices
to show that
σ(T ) = σp (T ) ∪ σc (T ) ∪ σr (T )
and each of these sets are disjoint.

Note that we have, using the definition of the point spectrum,

σ(T ) = σp (T ) ∪ (σ(T ) \ σp (T )) .

This is true since λ ∈ σ(T ) then Rλ either exists or doesn’t, but σp (T ) is the set for
which it doesn’t exist. The remaining set is the collection of λ ∈ σ(T ) such that Rλ (T )
exists. Clearly these sets are disjoint.
Now if λ ∈ σ(T ) \ σp (T ), then we can partition this set into the collection of λ for which
Rλ (T ) is defined on a set which is dense in X, and those for which the property fails.
But, this gives, upon referring to the definition of the residual and continuous spectrum,
that
σ(T ) \ σp (T ) = σr (T ) ∪ σc (T )
and clearly these sets are then disjoint.

7. Let X = C[0, 1] and fix v ∈ X define T : X → X by T x = vx. Find σ(T ).

Solution: Consider the operator Tλ x = T x − λx = (v − λ)x. So if λ ∈ / Ran v, then we

have that Rλ exists. So σ(T ) = Ran v = v([0, 1]). Note that since v ∈ X, we have that v
is continuous, and since [0, 1] is compact, it must assume a maximum and minimum, so
v([0, 1]) = [a, b].

8. Find a linear operator T : C[0, 1] → C[0, 1] whose spectrum is a given interval [a, b].

Solution: To solve this problem, we use Problem 7. For the operator T : C[0, 1] →
C[0, 1] given by T x = vx, we know that

σ(T ) = v([0, 1]).

As we observed at the end of Problem 7, v([0, 1]) is a closed interval. Set

v(x) = a + (b − a)x ∈ C[0, 1]

 and v([0, 1]) = [a, b].

9. Let T ∈ B(X). Show that kRλ (T )kB(X) → 0 as |λ| → ∞.

Solution: Let |λ| > kT k. Suppose that y = T x − λx. Then we have

kyk = kλx − T xk ≥ |λ| kxk − kT xk ≥ (|λ| − kT k) kxk .

This then gives that

1
kRλ k ≤ .
|λ| − kT k
Since the right hand side above clearly goes to zero when |λ| → ∞ we have the result.

10. For 1 ≤ p ≤ ∞ let `p denote the set of p-summable sequences with the standard norm.
Let T ((ξ1 , ξ2 , . . .)) = (ξ2 , ξ3 , . . .), i.e., the left shift.
(a) Consider T : `∞ → `∞ . If |λ| > 1 show that λ ∈ ρ(T ) and if |λ| ≤ 1 that λ is an
eigenvalue and find the corresponding eigenspace.
(b) Consider T : `p → `p when 1 ≤ p < ∞. If |λ| = 1 is λ an eigenvalue of T ?

Solution: Part (a): Note that we have

kT k`∞ →`∞ ≤ 1.
In fact we have equality by testing on the element (0, 1, 0, . . .). Since we know that for
any λ ∈ σ(T ) we must have that |λ| ≤ kT k = 1, and so for λ ∈ ρ(T ) we must have that
|λ| > 1. Suppose that λ is such that |λ| ≤ 1, then we want to solve
T x = λx (x2 , x3 . . . , ) = λ(x1 , x2 , . . .)
So we have that xj+1 = λxj for all j. Note that this implies
xj+1 = λxj = λ2 xj−1 = · · · = λj x1 .
So corresponding to any λ with |λ| < 1, the sequence
(λ, λ2 , λ3 , . . .) ∈ `∞
is an eigenvector (and eigenspace the space of this vector) with eigenvalue λ.

Part (b): The computations above imply that if we solve the equation
T x = λx
where x ∈ `p , then we have to have that x is a scalar multiple of the vector,
xλ = (λ, λ2 , λ3 , . . .).
However, if |λ| = 1, then xλ ∈/ `p when 1 ≤ p < ∞. So there is no vector x ∈ `p such
that T x = λx, and so if |λ| = 1, then this is not an eigenvalue of T .
 Homework 6
1. Let S, T ∈ B(X) show that for any λ ∈ ρ(T ) ∩ ρ(S) that
Rλ (T ) − Rλ (S) = Rλ (T )(S − T )Rλ (S).

Solution: Note that

Tλ Rλ (T ) = I Rλ (S)Sλ = I.
Then we have

Rλ (S)(T − S)Rλ (T ) = Rλ (S)(Tλ − Sλ )Rλ (T )

= (Rλ (S)Tλ − I)Rλ (T )
= Rλ (S) − Rλ (T )

2. Let X be a complex Banach space, with T ∈ B(X) and p a polynomial. Show that the
equation
p(T )x = y
has a unique solution x for every y ∈ X if and only if p(λ) 6= 0 for all λ ∈ σ(T ).

Solution: Note that if p(λ) 6= 0 for λ ∈ σ(T ) then the function g = p1 is holomorphic
on s set that contains σ(T ). But we have that pg = 1, and by the Spectral Theorem
p(T )g(T ) = I, and so p(T ) is invertible, and thus has a unique solution x for all y ∈ X.

Conversely, if there is a unique solution then p(T ) is invertible. Suppose that p(λ) = 0
for some λ ∈ σ(T ), then we have p(z) = (z − λ)h(z), which gives

(T − λI)h(T ) = p(T ) = h(T )(T − λI).

But since T − λI would not be invertible, we would have that p(T ) is not invertible, and
this is a contradiction and so p(λ) 6= 0 for all λ ∈ σ(T ).

3. Let T be a bounded operator on a Banach space. Suppose that T 2 = T , but T 6= 0 and

T 6= I. Compute σ(T ).

Solution: Note that T 2 − T = 0, and so p(T ) = 0 where p(z) = z 2 − z. By the Spectral

Theorem we have
{0} = σ(p(T )) = p(σ(T )).
So points λ ∈ σ(T ) must satisfy the equation
p(λ) = 0 = λ2 − λ = λ(λ − 1).
Thus σ(T ) = {0, 1}.
4. Let X 6= {0} be a complex Banach space. A bounded linear operator T is nilpotent if
there is a positive integer m such that T m = 0. Suppose that T : X → X, compute σ(T ).

Solution: By the Spectral Theorem we have that

σ(T m ) = (σ(T ))m

But since T m = 0, then σ(T m ) = {0} and so σ(T ) = {0}.

5. Let T ∈ B(X). Show that exp T is a well defined bounded linear operator on X with

kexp T k ≤ exp kT k

Solution: Note that ∞

X 1 j
exp T = T .
j=0
j!

Taking norms gives

∞ ∞
X 1 j X 1
kexp T k = T ≤ kT kj = exp kT k .
j=0
j! j=0
j!

6. Show that the spectral radius depends upper semincontinuously in the norm topology.
Namely if Tn → T then
lim rσ (Tn ) ≤ rσ (T ).

Solution: Suppose that λ ∈ ρ(T ). Then Tλ = T − λI is invertible, has a bounded

inverse and is densely defined. Since the set of invertible operators is open, and since
Tn → T is norm we have that λ ∈ ρ(Tn ). This implies that ρ(T ) ⊂ ρ(Tn ), or equivalently
σ(Tn ) ⊂ σ(T ). Which gives
rσ (Tn ) ⊂ rσ (T )
and
lim rσ (Tn ) ≤ rσ (T ).

7. If X is a complex Banach space, S, T ∈ B(X) and ST = T S, show that

rσ (ST ) ≤ rσ (S)rσ (T )
 Solution: We have that

(ST )2 = (ST )(ST ) = (T S)(ST ) = (T S)(T S) = T (ST )S = T (T S)S = T 2 S 2 .

Induction then proves

(ST )k = T k S k
and so
(ST )k ≤ T k Sk .
This inequality then clearly implies the result.

8. Show the commutativity can not be dropped from the problem above.

Solution: One can construct lots of different examples. Using 2x2 matrices is good
place to start looking for an example. Let
   
1 2 1 0
S= and T = .
0 1 1 1

Then rσ (S) = rσ (T ) = 1. Now compute that

   
3 2 1 2
ST = and TS = .
1 1 1 3

√ 6= T S. A simple computation then

So ST √ gives that the eigenvalues of ST are given by
2 ± 3. So we have that rσ (ST ) = 2 + 3 > rσ (S)rσ (T ) = 1.

9. Let H be a Hilbert space over C. Show

(a) If U is a bounded unitary operator and λ ∈ σ(U ), then |λ| = 1;

(b) If S is a bounded self-adjoint and λ ∈ σ(S), then λ ∈ R.

Solution: For part (a) first note that U ∗ U = U U ∗ = I and so kU k = 1. If |λ| > 1 then
we have Uλ < 1 and thus λI −U = λ(I −λ−1 U ) is invertible. So we have that λ ∈ / σ(U ).
∗ ∗
If |λ| < 1, then we have kλU k < 1 and then λI − U = U (λU − I) is invertible, and so
λ∈ / σ(U ). Thus we have that σ(U ) ⊂ {λ : |λ| = 1}.

For part (b) note that if λ is an eigenvalue with eigenvector x, then we have

λ hx, xi = hAx, xi = hx, A∗ xi = hx, Axi = λ hx, xi .

This gives that λ = λ for eigenvalues, and so must be real. Now observe that λ ∈ σ(T )
if and only if λ ∈ σ(T ∗ ). But since T = T ∗ , then we have that λ ∈ σ(T ) if and only if
 λ ∈ σ(T ). We just need to rule out the case of the residual spectrum. So if λ = a + ib
then we have
k(T − λI)xk2 = k(T − aI)xk2 + b2 kxk2 ≥ b2 kxk2 .
This estimate implies that T −λI is one-to-one and has closed range. If Ran(T −λI) 6= H,
then λ belongs to the residual spectrum. But, then we must have that λ = a − bi must
be an eigenvalue of T , which is not real. This contradicts what we had from before.

10. Let X = C[0, 1] with the standard norm. Define the operator
Z t
V (f )(t) = f (x) dx
0

Show that σ(V ) = {0} by two different methods:

(a) Compute the spectral radius of V ;

(b) Explicitly compute the inverse of the operator V − λI when λ 6= 0.

Solution: For part (a), note that

Z t Z s 
2
V (f )(t) = f (u)du ds
0 0

and so Z t Z s
t2

2
V (f )(t) ≤ kf kX du ds = kf kX
0 0 2
2 1
So we have that kV k ≤ 2
. A similar computation shows that

1
Vk ≤ .
k!
In fact one can easily show that
Z s
k 1
V f (s) = (s − t)k−1 f (t) dt.
(k − 1)! 0

These computations easily imply that

1
rσ (V ) = lim V k k
= 0.
k→∞

And so we have that σ(T ) = {0}.

For part (b), we set up an ODE that we can solve. For the moment, assume that
y ∈ C[0, 1] is actually differentiable. We want to solve the equation

V x(t) − λx(t) = y(t) t ∈ [0, 1].

This is the same as the ODE:
1 1
x0 (t) − x(t) = − y 0 (t).
λ λ
1
Set µ(t) = e− λ t , and then note that we have
Z t Z t Z t 
1 0 1 0 0
x(t) = − y (s)µ(s) ds = − (y(s)µ(s)) ds − y(s)µ (s) ds
λµ(t) 0 λµ(t) 0 0
1 t −1s
 Z 
1
= − y(t)µ(t) − y(0)µ(0) + e λ y(s) ds
λµ(t) λ 0
Z t
1 1 1t 1 1
= − y(t) + e y(0) − 2
λ y(s)e λ s ds.
λ λ λ 0

So we have that given y ∈ C[0, 1] that its solution is given by

Z t
1 1 1t 1 1
x(t) = − y(t) + e y(0) − 2
λ y(s)e λ s ds.
λ λ λ 0
It is clear from this representation that we didn’t need to apriori assume that y was
differentiable.
 Homework 7

1. If X and Y are normed spaces, and T1 and T2 are compact linear operators, show that
T1 + T2 is a compact linear operator. Show that the compact linear operators from X into
Y constitute a subspace K(X, Y ) of B(X, Y ).

Solution: There are two ways to prove that T1 + T2 are compact if T1 and T2 are com-
pact. One can resort to the characterization of compact operators by sequences, and
simply note that if {xn } is a bounded sequence then {T1 xn } and {T2 xn } have convergent
subsequences. Then use linearity and select a common subsequence so that {(T1 +T2 )xn }
has a convergent subsequence. The other way to prove that T1 +T2 is compact is to recall
that the sum of two compact sets is again compact.

That K(X, Y ) is a subspace is then immediate. We know that K(X, Y ) ⊂ B(X, Y )

since every compact linear operator is bounded. It is immediate that if λ ∈ C and
T ∈ K(X, Y ) then λT ∈ K(X, Y ) too. The above gives that T1 + λS ∈ K(X, Y ) when
T, S ∈ K(X, Y ) and λ ∈ C. So K(X, Y ) is a subspace.

2. If X is a normed space and Y is a Banach space, show that K(X, Y ) is a Banach space.

Solution: If X is a normed space and Y is a Banach space, then B(X, Y ) is a Banach

space. By Problem 1, we have that K(X, Y ) is a subspace of B(X, Y ). We just need to
show that if {Tn } is Cauchy in K(X, Y ) then the limit is a compact operator. Suppose
that {Tn } is Cauchy in K(X, Y ), then since B(X, Y ) is complete, we have that there is
a limit operator T ∈ B(X, Y ). However, since kTn − T k → 0, then we must have that T
is compact (an 3 -argument from class). Thus, we have that {Tn } is convergent and so
K(X, Y ) is a Banach space.

3. Show that a linear operator T : X → X is compact if and only if for every sequence {xn }
of vectors of norm not exceeding 1 the sequence {T xn } has a convergent subsequence.

Solution: To prove this we will use the alternate (equivalent) definition of compactness,
for any bounded sequence {xn } the sequence {T xn } has a convergent subsequence.

First suppose that T is compact and that the the sequence of vectors {xn } is bounded
in norm not exceeding 1. Then it is a bounded sequence and so {T xn } has a convergent
subsequence.

Now suppose that every sequence {xn } of vectors of norm not exceeding 1 the sequence
{T xn } has a convergent subsequence. Let {xn } be a given bounded sequence. Then
there exists a number R > 0 such that kxn k ≤ R for all n. So the sequence yn = xRn

 is bounded in norm by 1. Thus,we have that {T yn } has a convergent subsequence. But
since T is linear, we have that R1 T xn has a convergent subsequence. This then gives
that T is compact.

4. If z ∈ X and f ∈ X ∗ , show that T : X → X defined by T x = f (x)z is compact.

Solution: The range of the map is at most one dimensional. Compactness follows easily
from this.

xj
5. Show that T : `p → `p , 1 ≤ p < ∞, with T x = y defined by yj = (T x)j = j
is compact

Solution: Define Tn : `p → `p by
 x xn 
2
Tn x = x1 , , . . . , , . . . .
2 n
Then we clearly have that Tn is compact since it has finite dimensional range. Note that
∞
X |xj |p 1
kTn x − T xkp`p = p
≤ p
kxkp`p .
j=n+1
j (n + 1)

So we have that Tn → T uniformly. Then, we have that T is compact by the Theorem

in class.

6. Suppose that X is a Banach space. Let T ∈ B(X) and let S ∈ K(X). Show that
T S, ST ∈ K(X). Note that this implies that K(X) is an ideal in B(X).

Solution: Let B ⊂ X be any bounded set. Since T is bounded, we have that T (B) is
a bounded set too. Thus, ST (B) = S(T (B)) is relatively compact, and so ST is compact.

Let {xn } be any bounded sequence in X. Then since S is compact the sequence {Sxn }
has a convergent subsequence, denoted by {Sxnk }. Since T is bounded, we have that
{T Sxnk } is a convergent sequence. This gives that T S is compact.

7. Define the shift operator on `2 := `2 (N ∪ {0}) by


0 : n=0
(Sx)n =
xn−1 : n ≥ 1
1
Similarly, define the multiplication operator M : `2 → `2 by (M x)n = x .
n+1 n
Set T = M S.
1
Show that T is compact. Compute kT n k and limn kT n k n
Solution: We switch notation to avoid double subscripts. A sequence will be denoted
by x(n) instead of xn . First observe that

0 : n=0
T x(n) = x(n−1)
n+1
: n≥1
For k ∈ N define the operator

 0 : n=0
x(n−1)
Tk x(n) = n+1
: 1≤n≤k
0 : n > k.


Note that Tk is linear and compact. A simple computation then gives that
2
X |x(n − 1)|2
kT x − Tk xk`2 =
n≥k
(n + 1)2
1
≤ kxk2`2 .
(k + 1)2
1
This gives that kTk − T k ≤ k+1
and so Tk → T uniformly, and so T is compact.

Now we first compute T 2 . A straightforward computation gives that


2 0 : n=0
T x(n) = T (T x(n)) = T x(n−1)
n+1
: n≥1

 0 : n=0
= 0 : n=1
 x(n−2)
(n+1)n
: n ≥ 2.
More generally, we have that
(
0 : 0≤n≤k−1
T k x(n) = x(n−k)
Qk−1 : n ≥ k.
j=0 (n+1−j)

From this representation we conclude that

∞
2
X 1 1
T k x `2 = |x(n)|2 Qk−1 ≤ Qk−1 kxk2`2 .
2 2
n=0 j=0 (n + k + 1 − j) j=0 (k + 1 − j)

And choosing x = (1, 0, 0, . . .) gives that

1
Tk `2 →`2
= Qk−1 .
j=0 (k + 1 − j)
1
From this it is easy to deduce that limn kT n k n = 0. Indeed,
k−1
1 1X
log T k = − (k + 1 − j)
k k j=0
 
1 k(k + 1) k+3
= − (k + 1)k − =− .
k 2 2
This last expression converges to −∞ as k → ∞ then exponentiation gives the claim.
 Homework 8
1. Prove that if f ∈ L1 (Rn ) then fˆ is uniformly continuous on Rn .

Solution: Note that we have

Z
fˆ(y + h) − fˆ(y) = f (x) e−2πix·(y+h) − e−2πix·y dx


n
ZR
= f (x)e−2πix·y (e−2πix·h − 1)dx
Rn

Thus we have that

Z
fˆ(y + h) − fˆ(y) ≤ |f (x)| e−2πix·h − 1 dx
Rn

It suffices to show that this last expression can be made arbitrarily small as we let
h → 0, independently of y. The idea is to apply the Dominated Convergence Theorem.
Set gh (x) = f (x)e−2πix·(y+h) . Then we have that |gh | ≤ |f |, and gh → g0 = f (x)e−2πix·y
almost everywhere in Rn . So we then have that
Z Z
fˆ(y + h) = gh (x)dx → g0 (x)dx = fˆ(y)
Rn Rn

as h → 0 by Dominated Convergence.

2. Give f ∈ L2 (Rn ), prove that

Z
ξ→ f (x)e−2πix·ξ dx
|x|≤N

converges to fˆ in L2 (Rn ) as N → ∞.

Solution: Let Z
gN (ξ) = f (x)e−2πix·ξ dx.
|x|≤N

Observe that
Z
fˆ(ξ) − gN (ξ) = f (x)1{x:|x|>N } e−2πix·ξ dx = f 1\
{x:|x|>N } (ξ).
Rn

Then we have that

fˆ − gN = f 1\
{x:|x|>N }
L2 (Rn ) L2 (Rn )

= f 1{x:|x|>N } L2 (Rn )

But, for N large enough, we have that this last expression can be made smaller than any
given  > 0 since f ∈ L2 (Rn ).
3. If fk , f ∈ S(Rn ) and fk → f in S(Rn ), then fˆk → fˆ and fˇk → fˇ in S(Rn ).

Solution: Recall that fk → f ∈ S(Rn ) if for all multi-indices α and β we have

ρα,β (fk − f ) = sup xα (∂ β (fk − f )) → 0

x∈Rn

as k → ∞. We then have that

sup xα ∂ β (fˆk − fˆ) = C(α, β) sup ∂ α (x\

β (f − f )) ≤ ∂ α (xβ (f − f ))
k k L1 (Rn )
x∈Rn x∈Rn

Recall that if fk → f in S(Rn ), then we have that fk → f in Lp (Rn ) for any 0 < p ≤ ∞.
Moreover, we have X
∂ β g Lp (Rn ) ≤ C(p, n) ρα,β (g).
|α|=b n+1
p
c+1

Apply this estimate with g = fk − f and p = 1 to conclude that fˆk → fˆ. Similar
computations prove the statement for fˇk → fˇ.

4. Find the set of eigenvalues of the Fourier transform, namely the λ such that

fˆ = λf.

Hint: Apply the Fourier transform to the above identity, and consider functions of the form
2 2 2
xe−πx , (a + bx2 )e−πx and (cx + dx3 )e−πx for good choices of a, b, c, d.

Solution: Note that we have

ˆ
fˆ(x) = f (−x)
ˆˆ
ˆ
and so fˆ(x) = f (x). If f is an eigenfunction then we see that the corresponding eigenvalue
must satisfy
λ4 − 1 = 0.
From this we see that the eigenvalues of the Fourier transform are 1, −1, i, −i. Using
the remaining part of the hint, one can deduce the corresponding eigenfunctions to be
Hermite polynomials.

2
5. If 0 < c < ∞, define fc (x) = e−cx
(a) Compute fˆc in the following way: Let ϕ = fˆc and show that 4π 2 tϕ(t) + 2cϕ0 (t) = 0
and then solve the resulting differential equation;

(b) Show that there is one (and only one) value of c for which fc = fˆc ;

(c) Show that fa ∗ fb = γfc where γ = γ(a, b) and c = c(a, b).

Solution: Part (a): Set
Z
2
ϕ(t) = fˆc (t) = e−cx e−2πixt dx.
R

Then we have
Z
0 2
ϕ (t) = 2πi xe−cx e−2πixt dx
ZR
π d  −cx2  −2πixt
= − i e e dx
c R dx
Z
π 2 d
= − i e−cx e−2πixt dx


c R dx
2
π
= −2 tϕ(t).
c
Rearrangement gives the resulting differential equation. Interchange of the derivative
2
with respect to t, and switching the derivative with respect to x is justified since e−cx is
a Schwarz class function. Solving the resulting differential equation gives that
π 2 2
fˆc (t) = ke− c t

where k is some constant. Note that we have that fˆc (0) = k =

2
e−cx dx =
R pπ
R c
. Thus
we have that r
ˆ π − π2 t
fc (t) = e c .
c
Part (b): Suppose that we have a value of c such that
fc (t) = fˆc (t).
Then for all t ∈ R we have that
r
π t2 π2
“ ”
−c
e c
= 1.
c
In particular, it must be true when t = 0, and we obtain that c = π is the only value
that works.

Part (c): Taking the Fourier Transform, we have that

π − π2 t2 − π2 t2 π 2 1 1 2 π
e a e b = √ e−π ( a + b )t = √ e−π ab t .
2 a+b 2
ˆ ˆ
a ∗ fb = fa fb = √
f\
ab ab ab
ab π
p
Set c = c(a, b) = a+b and γ = γ(a, b) = a+b , then we have that
r
π −π2 a+b t2 π − π2 t2
√ e ab =γ e c = γ fˆc (t).
ab c
ab
p π
So we have that when c = c(a, b) = a+b and γ = γ(a, b) = a+b that
fa ∗ fb = γfc .
6. Suppose that f ∈ L1 (Rn ) and f > 0. Show that fˆ(y) < fˆ(0) for y 6= 0.

Solution: Note that for any y ∈ Rn we have

Z
fˆ(y) ≤ kf kL1 = f (x)dx = fˆ(0).
Rn

Suppose that for some y 6= 0 we have that

fˆ(y) = fˆ(0).

This then leads to a contradiction. Indeed, we have that

Z Z Z
ˆ
f (x)dx = f (0) = f (x)e−2πix·y
dx = η f (x)e−2πix·y dx
Rn Rn Rn

where η is a constant complex number with |η| = 1. Re-arrangement of this inequality

gives that Z
f (x)(1 − ηe−2πix·y )dx = 0.
Rn

This implies that f (x)(1 − ηe−2πix·y ) = 0 almost everywhere on Rn . But, since y 6= 0, we

have that (1 − ηe−2πix·y ) 6= 0 almost everywhere on Rn , and so f = 0 almost everywhere
on Rn . This is a contradiction to the conditions on f , and so there can not be a y 6= 0
with equality holding. Thus we must have for y 6= 0 that

fˆ(y) < fˆ(0).

7. Compute the Fourier transform of g(x) = e−2π|x| using the following steps:

(a) Let f ∈ L1 (R) and show that

Z Z  
1
f (x)dx = f x− dx.
R R x

2
(b) Use part (a) with f (x) = e−tx and t > 0 to obtain the following identity:
Z ∞
1 t2 dy
−2t
e =√ e−y− y √ ;
π 0 y

(c) Set t = π |x| and integrate with respect to e−2πiξ·x dx to obtain that

Γ( n+1
2
) 1
gb(ξ) = n+1 n+1 .
π 2 (1 + |ξ|2 ) 2
Solution: Part (a): For one proof, one can check the identity when f (x) = 1[0,1] (x), and
then use that simple functions are dense in L1 (R). Here is another proof that one can
give. Note that we have
Z Z ∞ Z 0
f (x)dx = f (x)dx + f (x)dx.
R 0 −∞

We work with each of these integrands separately. For the first one, we have that
Z ∞ Z ∞   
1 1
f (x)dx = f x− 1 + 2 dx.
0 1 x x
While for the second one, we have
Z 0 Z −1   
1 1
f (x)dx = f x− 1 + 2 dx.
−∞ −∞ x x
Now, note that we have
Z   Z −1   Z 1   Z ∞  
1 1 1 1
f x− dx = f x− dx + f x− dx + f x− dx
R x −∞ x −1 x 1 x
Using these identities from above, it is easy to see that
Z Z   Z   Z 1  
1 1 dx 1
f (x)dx = f x− dx + f x− − f x− dx.
R R x |x|>1 x x2 −1 x
To conclude the computation, we are left with showing that
Z   Z 1  
1 dx 1
f x− = f x− dx.
|x|>1 x x2 −1 x
To prove this last identity one shows that
Z 1   Z −1  
1 1 dx
f x− = f x−
0 x −∞ x x2
and  Z ∞ 
Z 0  
1 1 dx
f x− = f x−
−1 x 1 x x2
via a standard change of variables. Part (b): We use part (a) applied to the function
2
f (x) = e−tx . Now observe that
Z r
−tx2 π
e dx = .
R t
But, we also have that
Z 2
Z
−t(x− x1 ) 2 −t 1
e dx = e 2t
e−tx x2 dx
R R
e2t
Z 2
−y 2 − t 2
= √ e y dy
t R
e2t ∞ −u− t2 du
Z
= √ e u √ .
t 0 u
Using Part (a), we have that
∞
r
e2t
Z
π t2 du
=√ e−u− u √ .
t 2t 0 u

Rearrangement gives the result.

Part (c): Now set t = π |x| in Part (b) and obtain,

Z ∞
1 π 2 |x|2 dy
e −2π|x|
=√ e−y− y √ .
π 0 y

Then integrate this expression with respect to e−2πix·ξ dx to obtain that

Z
ĝ(ξ) = e−2π|x| e−2πix·ξ dx
n
ZR  Z ∞ 
1 −y− y
π 2 |x|2 dy
= √ e √ e−2πix·ξ dx
R n π 0 y
Z ∞ Z 
1 −y
2
− πy |x|2 −2πix·ξ dy
= √ e e e dx √
π 0 Rn y
Z ∞
1 2 n−1
= n+1 e−y(1+|ξ| ) y 2 dy
π 2 0 Z ∞
1 1 n−1
= n+1 2 n+1
y 2 e−y dy
π 2 (1 + |ξ| ) 2 0
Γ( n+1
2
) 1
= n+1 n+1 .
π 2 (1 + |ξ|2 ) 2