247A Homework.

The two sources for notes are http://www.math.ubc.ca/~ilaba/wolff/

and http://www.its.caltech.edu/~schlag/notes_033002.pdf

1. Let us define Z 2
1


J0 (x) = 2 cos x sin() d
0
Show that Z
f (x) 7 F () = 2 J0 (2x)f (x)x dx


defines a unitary map from L2 [0, ), r dr to itself. Describe the relation to the
Fourier transform of radial functions in two dimensions.
2. Let d be a probability measure on R with x d(x) = 0 and x4 d(x) < .
R R

Prove the central limit theorem for the sum of independent random variables with
this distribution.
Specifically, if X1 , X2 , . . . are d-distributed, show that for any Schwartz function
f, Z
x2
E f X1 ++X 1


n1/2
n

2 2
exp{ 2 2 }f (x) dx

as n . Hint: first show convergence for f (x) = e2ix uniformly for in a

compact set.
3. Show that every continuous (group) homomorphism from T into C (the non-zero
complex numbers under multiplication) takes the form x 7 e2inx with n an integer.
What is the analogous statement for continuous homomorphisms R C .
4. Let us define a sequence functions on R by
hd in 2
n (x) = 2x ex
dx
where n = 0, 1, . . . . Show that n (x) form an orthogonal sequence of eigenfunctions
for the Fourier transform on L2 (R).
In fact they are a basis, but this is much harder to prove. One approach to
this latter problem is to realize that they are the eigenfunctions of the harmonic
oscillator:
hd ih d i d2 u
u(x) 7 2x 2x u(x) = 2 + (4 2 x2 2)u(x).
dx dx dx
we write
5. Let G be a finite cyclic group and H a subgroup. For G
X
f() = f (g)(g).

g

H if is constant on the cosets of H.

We say G
Prove the following analogue of the classical Poisson Summation formula:
1 X 1 X
f () = f (h).
|G| |H|
H
G hH

H = {e2inx : n
(The classical version has G = R and H = Z, which leads to G
Z}.)
1
 2

6. Suppose f L2 (R) is supported on [ 12 , 12 ] then we know that f can be recovered

from the values of f(n) for n Z (the characters form an orthonormal basis). Prove
the Shannon Sampling Theorem:
X sin[(n )]
f() = f(n)
n
(n )
(which includes proving convergence of this infinite sum).
Remark: The audible spectrum extends only to about 20kHz. Consequently, as
heard by a human, one may regard music as a function whose Fourier transform is
supported on a finite interval. The above theorem says that to faithfully reproduce
music, one need only sample the signal forty thousand times per second. This is
what happens in CD recording.
7. Given Rd , show that the following are equivalent:
(a) For m Zd , m = 0 implies m = 0.
(b) The curve t 7 t + Zd is dense in Rd /Zd .
(c) For any continuous function f on Rd /Zd ,
Z T Z 1 Z 1
1
lim f (t + Zd ) dt = f (x + Zd ) dx.
T 2T T 0 0

[Hint: prove (a)(c) and then (c)(b)(a).]

8. Let d be a finite complex measure on R.
(a) Show that
Z L
1  2
({x})2
X 
lim () d =
L 2L L
xR
(finiteness of the measure implies that only countably many terms in the sum are
non-zero).
(b) Suppose that d is purely atomic, that is, d is a (countable) linear combination
of delta measures. Show that is almost periodic.
A function on f on R is said to be almost periodic if for any  > 0, there exists
L > 0 so that any interval of length L contains an -almost period:
 
a R p [a, a + L] such that supf (x) f (x + p) < .
x

() = ei + e2i .
Hint: For part (b) begin by considering the case
9. The dyadic cubes in Rd are the sets of the form
Qn,k = [k1 2n , (k1 + 1)2n ) [kd 2n , (kd + 1)2n )
were n ranges over Z and k Zd .
(a) Given a collection of dyadic cubes whose diameters are bounded, show that one
may find a sub-collection which covers the same region of Rd but with all cubes
disjoint.
(b) Define the (uncentered) dyadic maximal function by
Z
1
[MD f ](x) = sup f (y) dy
Q3x |Q| Q

where the supremum is over all dyadic cubes that contain x. Show that this operator
is of weak type (1,1).
(c) Deduce boundedness of the Hardy-Littlewood maximal function from the above.
 3

Remarks: Part (a) provides a replacement for the Vitali Covering Lemma. I propose
you address (c) geometrically; draw some pictures in the planar (d = 2) case.
10. (a) Evaluate
N
X
DN (x) = e2inx
n=N
and show that it is not an approximate identity on T.
(b) Show that 2N1+1 |DN (x)|2 is an approximate identity and derive its relation to
the Fejer kernel.
(c) Calculate
X
r|n| e2inx
nZ
for 0 < r < 1 and show that for r 1 it gives rise to an approximate identity.
(d) Suppose n is an approximate identity and d, a finite complex measure on T.
Show that n d converges weak- to d.
R Note: dRn converges weak- to d iff for every bounded continuous function, f ,
f dn f d.
11. (a) Given f Lp (R), 1 p < , show that t 7 f (x + t) defines a continuous map
of R into Lp (R, dx).
(b) Show that it is not equi-continuous as f varies over the set of f with kf kLp 1.
(That is,  cannot be chosen from independently of f .)
(c) Show that part (a) is false for L and M (R).
12. (From Wolff 4.) Find a sequence of Schwartz functions n such that (a) kn kLp
and kn kLp0 are constant. The supports of n are disjoint and those of n are
PN
almost disjoint. Use n=1 n to show that if kfkLp0 . kf kLp then p 2.
PN PN
By almost disjoint we mean k n=1 n kpLp 100 99
p
n=1 kn kLp . Notice that if
the supports were actually disjoint, then 100/99 could be replaced by 1.
Hint: Take a single Cc function and modify it by translation and multiplication
by characters.
13. Prove the Rising Sun Lemma: Given a non-negative f L1 (R), define
1 t
Z
[MR f ](x) = sup f (x + s) ds.
t>0 t 0

If S = {x : MR f > } then |S| = 1 S f (x) dx. [Hint: S is open.]

R

14. Prove the following theorem of Milicer-Gruzewska: Let d be a complex measure

(n) 0 as n ( is called a Rajchman measure).
on T with the property that
If f L1 (d||) and d = f d then (n) 0. [Hint: mimic the proof of the
RiemannLebesgue Lemma from Schlags notes.]
15. Let R(k) be the smallest number such that in any colouring of the edges of the
complete graph on R(k) vertices by two colours, one can find a monochromatic
complete graph on k vertices. These are known as Ramsey numbers; it is not
difficult to show that R(k) 22k . The problem here is to prove that 2k/2 R(k),
which is due to Erd os.
(a) Determine the expected number of monochromatic complete graphs on k ver-
tices contained within a random colouring of the complete graph on n vertices.
(b) Show that this is less that one when n = 2k/2 and so complete the problem.
 4

16. Let f C with < 1, and let Kn denote the Fejer kernel.
(a) Show that
kf Kn f kC 0 . n kf kC .
(b)[Optional] Show that kf Kn f kC 0 may fail. However, it is true if one
restricts to those f with
(1) sup |f (x) f (y)| = o( ).
|xy|<

(c)[Optional] Show that the set of f C that obey (1) is exactly the closure of
C in C .
17. Let f be a continuous function on T. Suppose that for each n > 0 there is a
trigonometric polynomial pn of degree n (or less) such that
kf pn kC 0 . n
where < 1. Show that f is Holder continuous. Hint: write

X

f = p1 + p2k p2k1 .
k=1

18. Let be a simply-connected open domain bounded by a Jordan curve. By a

theorem of Caratheodory, any conformal map f of D onto can be extended to a
homeomorphism of D onto .
We say that a curve : S 1 C is rectifiable if there exists a constant L so that
for any 0 0 < 1 < < n < 2,
n
X
|(eik ) (eik+1 )| L
k=0

where n+1 = 0 .
Prove the following theorem of F. and
P M. i Riesz: f 0 H 1 if and only if is
rectifiable. [Hint: the function z 7 |f (ze k ) f (zeik+1 )| is continuous and
sub-harmonic on D.]
19. Prove the following result of Privalov: For 0 < < 1, f C implies f C .
20. (a) Suppose T is a rotation invariant operator on L2 (R/Z), that is, Ry T = T Ry
for any rotation [Ry f ](x) = f (x y). Show that e2inx , n Z, are eigenfunctions
of T .
(b) Let T be a bounded operator on L2 (Rn ) such that there is a function K obeying
|K(x, y)| . |x y|n so that whenever f and g have disjoint supports,
Z Z
hg, T f i = g(x)K(x, y)f (y) dy dx.

Show that if T is translation invariant, then K(x, y) = F (x y), which means that
T is a convolution operator. [Hint: Treat (a) and (b) independently.]
21. (a) Let I R be an interval and let z C+ = {z : Im z > 0}. Show that the
harmonic measure of I C+ with respect to z is equal to the angle subtended
by I at z divided by . Deduce that the the harmonic measure of I is constant on
arcs of circles.
d
(b) Calculate the conjugate function of [0,a] () L2 (S 1 ; 2 ).
 5

22. (a) Suppose f : D C+ = {z : Im z > 0} is analytic. Show that there exists a

finite positive measure d and a real constant a so that
Z 2 i
e +z
f (z) = a + i d().
0 ei z
This result is due to Herglotz. [Hint: First look at Im(f ).]
(b) Deduce that any holomorphic mapping of C+ into itself admits the representa-
tion Z
1 + tz
f (z) = a + bz + d(t).
R tz
where a R, b 0, and d is a positive measure.
23. Prove the following theorem of Kolmogorov: suppose 0 < p < 1 then
Z 2 Z
d()
f (z) = i z
implies sup |f (rei )|p d <
0 e 0<r<1

for any finite complex measure d.

24. Let `p denote the weighted `p space
X
kckp = (|n| + 1)2 |cn |p .
Let n , n Z, be an orthonormal basis for L2 (R/Z) which obeys kn kL . 1 and
define T : L2 `2 by


[T f ](n) = (|n| + 1) n (x), f (x) .
(a) Show that T extends to a bounded map of Lp into `p for all 1 < p 2. This
result is due to Hardy and Littlewood. [Hint: Prove a weak-type bound and use
Marcinkiewicz.]
(b) Given a sequence cj indexed by j Z, define the rearrangement cj as follows:
For j 0, cj is the (j +1)th largest element of the set {|c0 |, |c1 |, . . . } while for j < 0,
it is the |j|th largest element of {|c1 |, |c2 |, . . . }. Derive the following inequality
of Payley: X
(1 + |j|)p2 |cj |p . kf kpLp
where cj = f(j).
(c) By splitting the sum dyadically, show that this implies the usual Hausdorff
Young inequality for 1 < p 2.
25. Suppose f L1 (R/Z) and
R let M f denote its (uncentred) dyadic maximal function.
(a) Show that for > |f |,
Z
1
|f (x)| dx . |{x : M f > }|.
|f |>
[Hint: Do a CalderonZygmund style decomposition.]
(b) Deduce that if M f L1 , then |f | log[1 + |f |] L1 . This result is due to Stein.
(c) Use the fact that M : L L and L1 L1w to show
Z
1
|{x : M f > }| . |f (x)| dx.
|f |>c
for some small constant c.
(d) Deduce that if |f | log[1 + |f |] L1 then M f L1 .
 6

26. For 1 p < , let Lpw (R) denote the set of measurable functions on R for which
1/p
kf kp = sup p |{x : |f | > }|

>0

is finite. The is to warn that this isnt a norm; however,

(a) For 1 < p < , the following defines a norm on Lpw (R):
Z
1
kf kp,w = sup (p1)/p
|f (x)|.
E |E| E

Moreover, kf kp . kf kp,w . kf kp . [Remark: with this norm, Lpw (R) is actually a

Banach space.]
(b) Show that there is no norm on L1w (R) comparable to kf k1 by considering the
following family of functions
N
X 1
|x k|
k=0
as N .
27. (a) Let cn denote the surface area of S n1 Rn . Show that for n 3,
1
G(x) =
(n 2)cn |x|n2
is the Green function for the Laplace equation in Rn : if f S, then (Gf ) = f .
(b) For any f, g S,
Z 2 ZZ
 f (x)g(x) dx kf k2 2
 
  L g(x)G(x y)g(y) dx dy.

(c) Deduce the following Sobolev inequality:

f S, kf kLq . kf kL2 where q = 2n/(n 2)
by choosing g appropriately.
(d) Show that on R, one does not have
kf k2L . kf 0 k2L2
however it is true that
kf k2L . kf 0 k2L2 + kf k2L2 .
[Remark: In this regard, R2 is like R; there is no estimate without adding kf kL2 .
However, one has only
kf k2Lq . kf k2L2 + kf k2L2
for all 2 q < .]
28. Suppose a : R2 R obeys
n+m
a(x, ) L
xn m
for all n, m 0. We then define an operator on L2 (R) by
Z
[T f ](x) = a(x, )e2ix f() d.

(This is the pseudo-differential operator with symbol, a, which belongs to the exotic
0
symbol class S0,0 .) Show that it is bounded. [Hint: let j be a partition of unity
 7

adapted to the partition of R by [j, j +1), then apply the Cotlar-Stein Lemma using
the operators with symbols ai,j (x, ) = (x i)a(x, )( j).]
29. (a) Prove that Z  Z
f (x)g (x) dx
 

 f (x)g(x) dx
Rn Rn
(b) Suppose f 7 f K is a bounded operator on L2 (Rn ) and K(x) . |x|n . Show
that there exists C so that
Z
K(x) dx C
<|x|<N

for all 0 <  < N < .

30. Given a measurable function t : R (0, ), let us define
Z
1 2
[Tt f ](x) = p exp{ (xy)
2t(x) }f (y) dy.
2t(x)
(a) Determine the adjoint of the operator Tt ; write it as an integral operator.
(b) Consider Tt Tt and show that for f 0,
[Tt Tt f ](x) . [T2t f ](x) + [T2t

f ](x).
(c) Deduce that maximal operator
Z
1 2
[M f ](x) = sup exp{ (xy)
2t }f (y) dy.
t>0 2t
is bounded on L2 (R). [Remark: There is nothing special about the Gaussian, it
was just chosen for concreteness.]
31. Given n Z3 , let us write |n| for the `1 norm: |n| = |n1 | + |n2 | + |n3 |. Consider
the following operator on `2 (Z3 ):
X
[Hu](n) = u(m).
|nm|=1

Schurs test (or part (b)) shows that this is a bounded operator.
(a) Given n Zd , let us write n for the function k 7 k,n . Show that hm |H N n i
is equal to the number of paths of length N from n to m in the Z3 lattice.
(b) As H is translation invariant, we know that we can write it as a Fourier multi-
plier. Find the Fourier multiplier.
(c) Determine the leading term in the t asymptotics of
h0 |etH 0 i.
(d) [Optional] Use the BorelCantelli Lemma to deduce that in three dimensions,
a random walker starting at the origin will return to the origin only finitely many
times (with probability one).
32. Let denote a hyperplane in Rd and let d denote the induced Lebesgue measure.
For s 0, H s denotes the Sobolev space of functions f L2 for which
Z
2
f s = |f|2 (1 + ||2 )s d
H

is finite.
Show that for  > 1/2, f 7 f | defines a continuous map from H s (Rd ) to
s
H (). Also show that for  1/2, it does not.
 8

33. Let denote the cone |0 |2 = |1 |2 + + |d |2 in Rd+1 and let d denote the
induced surface measure.
(a) If f is a smooth function supported in a compact subset of Rd+1 \ {0}, show
that Fourier transform of f d has a natural interpretation as a solution of the wave
equation:
d2 u X d2 u
= .
dt2 j
dx2j

(b) Calculate the leading term asymptotics of fd d as || in a fixed direction.

For simplicity, just treat the case d = 2 with f supported in the region {0 > 0}.
Warning: the cone does not have non-vanishing Gaussian curvature!
34. (a) Given 0 with 0 Cc (R), write the solution of the free Schrodinger equation
2
i = 2 , (x, t = 0) = 0 (x)
t x
as an integral involving 0 .
(b) Study the asymptotics in the regime t with x = vt and v R fixed.
Specifically, prove that
 2

1
(x, t) 4t ei/4 eix /4t 0 ( 4t
x 
) . (t2 + x2 )3/4


for t sufficiently large.

(c) Let us call the map of 0 into the leading asymptotic behaviour V (t). That is,
the LHS of the equation above is |(x, t) V (t)0 |. Check that this determines a
unitary map and that (t) V (t)0 converges to zero in L2 .
(d) Use the fact that for t fixed, 0 7 (t) is also a unitary map to deduce that
the above asymptotic holds in L2 sense for any initial data 0 L2 .
35. Prove the van der Corput Lemma: (a) Suppose is real-valued and smooth in (a, b)
and that for some k 1, (k) (x) 1 on [a, b]. Show that
Z 
 b 
ei(x)
dx 3k 1/k
 

 a 
for k 2 and also for k = 1 if we assume that 0 is monotone. [Hints: Proceed
by induction. For k = 1, integrate by parts wisely. For the step from k to k + 1,
treat any interval with |(k) (x)| () separately from those where it is bigger
than ().]
(b) Deduce that
Z 
 b   Z b 
i(x) 1/k 0
e (x) dx . |(b)| + | (x)| dx .
 

 a  a

36. (a) Prove Debyes asymptotics for Bessel functions: given (0, ),
e[tanh()] h i
1 + O( 1 )


J sech() = p
2 tanh()
as .
(b) Prove that
X
cos[z sin()] = J0 (z) + 2 J2k (z) cos(2k)
k=1
 9

for all R. Why does the series converge?

37. Let E be a compact subset of Rn of non-zero -capacity. In class we proved the
existence of a probability measure d so that
1
= inf I () = I ().
C (E) supp()E
Recall that Z
d(y)
V (x) =
|x y|
denotes the potential generated by d.
(a) Show that V (x) 1/C (E) for p.p. x E. (Recall that p.p. means except
for a set of zero capacity.)
(b) Show that for any positive measure , V (x) is upper semi-continuous, that is,
for every a R, the set {x : V (x) > a} is open. Equivalently,
lim inf V (xn ) V (x).
xn x

(c) From part (a), it follows that C (E)V (x) 1 for -almost all x. Explain.
From this and part (b), show that C (E)V (x) 1 for all x supp().
(d) Show that
C (E) = sup{kk : supp() E and x supp(), V (x) 1}.

38. Let E Rn , be a compact set of non-zero -capacity (0 < < n). Let us define
X
2
Dn = inf n(n1) |xi xj |
1i<jn

and X
Mn = sup inf 1
|x xi | .
xE n
1in
The infimum in the definition of Dn and the supremum in the definition of Mn are
over {x1 , . . . , xn } E.
Show that Dn+1 Mn , that Mn 1/C (E), and that lim inf Dn 1/C (E).
Conclude that C (E) = lim Dn = lim Mn . [You may use the results of Question 3.]
39. Functions defined by

X an
f (s) = ,
n=1
ns
are known as Dirichlet series. The most famous example is the Riemann zeta
function, where an 1. By writing s = + it we have ns = e log(n) eit log(n)
which shows the connection to Fourier integrals.
bn ns , show that f (s)g(s) can also be
P
(a) Given f (s) as above and g(s) =
written as a Dirichlet series and find the formula for the coefficients. In this way,
interpret the coefficients of (s)2 . (This operation is the multiplicative analogue of
convolution.)
(b) Let f and g be Dirichlet series absolutely convergent for Re(s) > 0 . Show that
Z T
1 X an bn
lim f ( + it)g( it) =
T 2T T n+
for Re() and Re() larger than 0 . This is the analogue of the Plancherel Theorem.
 10

(c) Prove the following simple Abelian theorem: Given < 1,

lim log (n)an = A = lim u1 f (1 + u) = C A
n u0

and determine the value of C . What if = 1?

40. (a) Let d(n) = #{d > 0 : d|n}. Prove that if n = pai i then
Q
 
d(n) Y ai + 1 n 1/ o
2
= < exp log(2)
n pa
i
i

[Hint: be wasteful, a+1

pa
1 + paa 1 + log(2)
1
.]
(b) Refine the above argument to show that
(1 + ) log(2) log(n)
log[d(n)]
log log(n)
for n sufficiently large (depending on ). P
(c) By the prime number theorem, (x) = px log(p) obeys (x)/x 1. Use
this to show that
(1 ) log(2) log(n)
log[d(n)] .
log log(n)
infinitely often.
(d) By counting lattice points under the hyperbola xy = n, show that
d(1) + d(2) + d(3) + + d(n) = n log(n) + O(n).
While part (c) shows that d(n) can be enormous, this result shows that it is typically
much smaller.
41. Let = e2i/3 .
(a) Show that Z[] is a Euclidean domain using the norm N (a + b) = |a + b|2 =
a2 ab + b2 .
(b) Determine the units (there are six).
(c) Show that the following is a complete list of the primes in Z[] (without repe-
tition):
(i) 1 and its associates,
(ii) the rational primes of the form 3n + 2 and their associates, and
(iii) the (non-trivial) factors a + b of rational primes of the form 3n + 1.
(d) Deduce that the number of solutions (n, m) Z2 to N = n2 + 3m2 is bounded
by C N  . (You may use results from the previous problem.)
42. (a) Use the previous problem to prove the following result of Bourgain1:
2

X 2 X
2i[nx+n t] 

an e .N |an |2
|n|N 6
L
6 6 2 2
where L denotes L (R /Z ; dx dt).
(b) Gauss proved that if a, b are integers and q is an odd prime with a, b [1, q 1],
then
 q1 2i[an2 +bn]/q 
X 

 e  = q.

n=0

1Fourier transform restriction phenomena for certain lattice subsets and applications to non-
linear evolution equations, I. Geom. Funct. Analysis, 3 (1993) 107156.
 11

(Such sums are known as Gauss sums.) From this we may deduce
 N 2i[an2 +bn]/q 
X 
N
 e &

n=0
 q
2
when N q, for example. By studying small regions around x = b/q and t = a/q
with q [3, N ] show that N  cannot be replaced by an N -independent constant
in part (a).
43. Prove the following two Strichartz estimates due to Bourgain2:
2 2
X X
an e2i(nx+n t) . |an |2


L4
where L4 denotes L4 (R2 /Z2 ; dx dt) and
2

X 2 2 X
2i[nx+my+(n +m )t] 
|a(n, m)|2


a(n, m)e .N
n2 +m2 N 2 4
L
where L4 denotes L4 (R3 /Z3 ; dx dy dt). [Hint: Use the method from the proof of
Zygmunds restriction theorem.]
44. Let d and dn be probability measures on [0, ).
(a) Show that if dn converges weak- to d, then
lim sup n (K) (K)
n
for any closed set K. Also show that for any open set, O,
lim inf n (O) (O).
n
(b) Give examples that show that the inequalities in part (a) can fail to be equalities.
(c) If we do not assume that d is a probability measure, half of (a) can fail. Which
half and why?
(c) Show that if Z Z
lim emx dn (x) = emx d(x)
n
for all m {0, 1, 2, . . . } then dn converges weak- to d.

2Fourier transform restriction phenomena for certain lattice subsets and applications to non-
linear evolution equations, I. Geom. Funct. Analysis, 3 (1993) 107156.