CAAM/Stat 31460 Applied Fourier Analysis, Winter 2024

Homework 1

Instructions: Please write your solution to each problem on a separate page, and please include the full
problem statement at the top of the page. All solutions must be written in legible handwriting or typed (in each
case, the text should be of a reasonable size). Your solutions to all problems should be written in complete
sentences, with proper grammatical structure. Reminder: In all cases where there is collaboration or external
resources are consulted or used, proper citation must be given.
Due date: Tuesday, Jan. 16, 8pm, by online Gradescope submission.
Reading Assignment:
• Sections 2.1–2.2 of Folland (click on “Library Reserves” on the left-hand side of our Canvas page to
access the library’s electronic copy of the textbook).

Written Assignment:

(1) (a) Show that for all m, n ∈ Z, m 6= n implies

Z π 
i(m−n)x 0, if m 6= n,
e dx =
−π 2π, if m = n.

(b) Let n be an integer with n ≥ 1. Show that

2 π −4/(n2 π),
Z 
if n is odd,
x cos(nx) =
π 0 0, if n is even.
(This is the computation we used in Lecture 1 when we computed the coefficients of the Fourier
series of f : R → R defined by f (x) = |x| for x ∈ [−π, π] and extended periodically.)

(c) Compute the Fourier series of the function k : R → R defined by

k(x) = | sin(θ)|
for x ∈ R (see, e.g. Item 8 from Table 1 in Chapter 2 of our Folland text.)

(2) Let f : R → C be a 2π-periodic function which is Riemann integrable on [−π, π], and suppose that
N
X
sup f (x) − cn einx → 0
x∈[−π,π] n=−N

as N → ∞. Show that for all n ∈ Z, we have

Z π
1
cn = f (x)e−inx dx.
2π −π

(3) (Plots of partial sums) The Python code in the file plot fourier.py (posted on Canvas in the
Assignment page and in the Files HW section) can be used to plot partial sums of the Fourier
series for our two “sawtooth” examples. Similar plots can be made using Octave or Matlab, using
similar syntax. Generate similar plots for items 4, 6, 7, and 8 from Table 1 in Chapter 2 of Folland.
Write a brief report containing your code, images of the plots you generate, and discussion of any
aspects of the plots you find notable observations.

1
sin((N + )x)
2
(4) (a) For N ≥ 1, define DN : R → R by DN (x) = 2π sin(x/2) for x ∈ R (we will see in Lecture 2
that this function, which is referred to as the “Dirichlet kernel,” is closely related to results on
 convergence of Fourier series). Show that
N
1 X
+ cos(nx) = πDN (x).
2 n=1
(Hint: For n ≥ 1, write cos(nx) = (einx + e−inx )/2. Now, note that 1 +
P
cos(nx) =
P−1 inx
PN inx
1 + n=−N e + n=1 e can be written as a truncated geometric series. Now, recall
a(1−r n+1 )
that for a, r ∈ C with r 6= 1 we have a + r + ar2 + · · · + arn = 1−r .)

(b) Show that for all a ∈ R, | sin(ax) sin((a + 21 )x)| ≥ | sin2 (ax) cos(x/2)| − | sin(x/2)|.

2 Qn 2
(5) Set λ0 = 1, and, for each n ≥ 1, set λn = 2(n!) λn−1 (so that λn = k=1 2(k!) ). Define g : (0, π) → R
by setting, for x ∈ (0, π),
 1
√ sin(λn x),
n
π
if x ∈ ( λπn , λn−1 ], n ≥ 1,
g(x) =
0, if x = 0.
Now, let f˜ : [−π, π) → R be the even extension of g (defined by setting f (x) = g(x) for x ∈ [0, π)
and f (x) = g(−x) for x ∈ (−π, 0)), and let f : R → R be the (2π-)periodic extension of f˜.
(a) Show that f is continuous on R.

(b) Recall that the Fourier coefficients of f are given by

2 π
Z
an = f (x) cos(nx)dx (n ≥ 0), bn = 0 (n > 0).
π 0
Show that for all k ≥ 1, the truncation of the evaluation at x = 0 of the Fourier series of f to
the first λk non-constant terms is given by
kλ
a0 X
+ an = Ak + Bk + Ck ,
2 n=1
where
Z π/λk Z π/λk−1
Ak := 2 f (x)Dλk (x)dx, Bk := 2 f (x)Dλk (x)dx,
0 π/λk

and Z π
Ck := 2 f (x)Dλk (x)dx,
π/λk−1
where Dλk is as in (b) above (with N = λk ).
(Hint: Use the formulas for a0 and an , 1 ≤ n ≤ λk , in the left-hand side of the desired identity,
and make use of 5(a) above. Then split up the resulting integral into a sum of integrals over
appropriate intervals.)

(c) Show that there exists C > 0 and c > 0 such that for all k ≥ 1 we have
|Ak | ≤ C,
c  λ 
k
|Bk | ≥ √ log − C,
k λk−1
|Ck | ≤ C log(λk−1 ).
(Hint:
R To estimate |AkR| and |Ck |, one strategy is to note that |f (x)| ≤ 1 on [0, π] implies
|2 I f (x)Dλk (x)dx| ≤ 2 I |Dλk (x)|dx for any interval I. To estimate the right-hand side of this
bound, you may find it useful to recall that | sin(x)/x| ≤ 1 for x ∈ R (and thus | sin(kx)/x| ≤ k
for all x) and to observe that |x/ sin(x)| ≤ 1 for x ∈ [0, π/2] (and thus |(x/2)/ sin(x/2)| ≤ 1 for
x ∈ [0, π]). To estimate |Bk |, use 5(b) above. You may also find it useful to use the estimate
| cos(x/2)| ≥ 1 for x ∈ [0, π] (together with estimates as before for sin(x)/x). The calculation
 concludes by noting that the integral of sin2 (ax)/x can be evaluated.)

(d) Use the triangle inequality to combine parts (b) and (c) to show that for all M ≥ 1 there exists
K ≥ 1 such that for all k ≥ K, we have 1
k λ
a0 X
+ an ≥ M.
2 n=1

(Hint: You may find it useful to observe that log(λk /λk−1 ) = (k!)2 log(2) and log(λk−1 ) =
Pk−1
log(2) n=1 (n!)2 ≤ (k − 1)[(k − 1)!]2 log(2). It follows that if K0 is chosen large enough to
ensure that k ≥ K0 implies ck 3/2 − C(k − 1) > 1, then for all such k we have |Bk | − |Ck | −
|Ak | ≥ log(2)[(k − 1)!]2 − 2C. Now, for each M , look for K ≥ K0 such that k ≥ K implies
log(2)[(k − 1)!]2 − 2C ≥ M .)

Suggested Additional Problems (not collected or graded):

(1) Read Sections 1.2 and 1.3 of Folland on separation of variables. Complete the following exercises:
(a) pg. 11, exercise 5(a).
(b) pg. 17, exercise 5.
(2) Let f be as in Problem (3) above. Show that if f is even (in the sense that f (x) = f (−x) for all x)
then for all n ≥ 1 we have
1 π
Z
cn = c−n = f (x) cos(nx)dx.
π 0
Similarly, show that if f is odd (in the sense that f (x) = −f (−x) for all x) then for all n ≥ 1 we
have
i π
Z
cn = −c−n = − f (x) sin(nx)dx.
π 0
(3) (Riemann-Lebesgue lemma) In this exercise we outline a proof of the Riemann-Lebesgue lemma: for
any f ∈ L1 (T),
lim cn = 0,
n→±∞
where cn denotes the nth Fourier coefficient of f .

Let f ∈ L1 (T) be given, and let (cn ) denote the sequence of Fourier coefficients of f .

(a) Use the change of variable x 7→ x + π/n to rewrite the formula for cn as
Z Z
1 1
cn = f (x + π/n)e−in(x+π/n) dx = − f (x + π/n)e−inx dx.
2π T 2π T
(b) Add the formula in (a) to the usual formula for cn to find
Z  
4πcn = f (x) − f (x + π/n) e−inx dx.
T
(c) Use the above to show that if f is continuous then cn → 0 as n → ±∞.
1Part (e) of this exercise in particular shows that the Fourier series of f ,
∞
a0 X
+ an cos(nx),
2 n=1
diverges at x = 0. This example is essentially that of du Bois-Reymond (1873) (there are also some related arguments
due to Lebesgue and Fejér; see also an influential and useful “soft analysis”/functional analysis argument using the uniform
boundedness principle – a textbook reference is Prop. 1.5.1 and Prop. 1.6.4 in M. Pinsky, Introduction to Fourier Analysis
and Wavelets); the result that a continuous function can have a Fourier series which does not converge everywhere was quite
striking and played a key role in the development of modern analysis. Further developments are due to Kolmogorov (1923) and
Kahane-Katznelson (1966). We’ll see a variety of hypotheses in class (stronger than continuity) that can be imposed to ensure
that various notions of convergence do hold. Moreover, as we’ll briefly discuss (but not prove) in class, a remarkable theorem
of Carleson (1966) shows that if f belongs to L2 (T), then its Fourier series converges to f almost everywhere (i.e. except for a
set of measure zero); this was extended by Hunt (1968) to functions in Lp , p > 1.
(d) We now extend part (c) to general f ∈ L1 (T). Fix  > 0. Standard properties of the Lebesgue
integral allow us to find a continuous function f˜ on T such that
Z
|f (x) − f˜(x)|dx < .

Set g := f − f˜, and let (dn ) denote the Fourier coefficients of g, defined via
Z
1
dn := (f (x) − f˜(x))e−inx dx.
2π T
Note that Z
1 
|dn | ≤ |f (x) − f˜(x)|dx < .
2π T 2π
(e) Let (c̃n ) denote the sequence of Fourier coefficients of f˜. Show that
cn = c̃n + dn
for n ∈ Z, and conclude that

|cn | ≤ |c̃n | + |dn | ≤ |c̃n | + .
2π
(f) Combine (c), (d), and (e) to obtain

lim sup |cn | ≤ .
n→±∞ 2π
Since  > 0 was arbitrary, this gives the claim.