Advanced Methods in Applied Analysis

Assignment 3

Alberto Castro - Jose Melo - Alejandro Vallejo

November 14, 2024

As it will be very useful during the following problems, we will first enounce the Parseval Theorem.

Theorem 1 (Parseval). If f (x) is a square-integrable function on the interval [−π, π], then the sum
of the squares of its complex Fourier coefficients cn is equal to the L2 norm of the function:
Z π ∞
1 X
|f (x)|2 dx = |cn |2 ,
2π −π n=−∞

where cn are the Fourier coefficients of f (x) given by:

Z π
1
cn = f (x)e−inx dx.
2π −π

Problem 1 Let f (x) = x for x ∈ [−π, π]. Find the Fourier coefficients of f and use the Parseval
inequality to bound the sum of the squares of these coeffients.

Using the Parseval theorem (Th.1), we get a bound for the sum of the squares of the Fourier
coefficients,
∞ Z π
X 1
|cn |2 ⩽ |f (x)|2 dx.
n=−∞
2π −π

Doing the integral for f (x) = x,

π
2π 3
Z
x2 dx =
−π 3
∞
X 1 2π 3 π2
|cn |2 ⩽ = .
n=−∞
2π 3 3

For the Fourier coefficients of f (x) = x we need to use the second formula from Th.1,
Z π
1
Cn = xe−inx dx .
2π −π

It is obvious that for n = 0 the result of the integral is 0, so we will only do the calculations for n ̸= 0.

1
 We integrate by parts, we recall the integration by parts formula,
Z b Z b
udv = [uv]ba − vdu .
a a

We use the following change of variables,

−1 −inx
u = x, du = dx, dv = e−inx dx, v= e ,
in
π !
 π  
−x −inx 1 −π
Z
1 −inx 1
Cn = e − e dx = (2 cos(nπ)) + (−2 i sin(nπ)) =
2π in −π −π 2π in in
i i(−1)n
cos(nπ) = .
n n

For every n ̸= 0, as mentioned before.

P∞Problem 2
2. Prove that for the function f (x) = sin(x) on [−π, π], Parseval identity gives that
1
n=1 |cn | = 4 , where cn are the Fourier coefficients of f .

Looking at Th.1 we see that there is a difference in the range of the summatory compared with
the one that is presented in this exercise. Thus, we need to calculate the |cn | in order to see what
coefficients can we keep and if we can or cannot apply the theorem.

For that, we calculate the generic formula for the complex Fourier coeffients,
Z π Z π
1 −ixn 1
cn = f (x)e dx = sin(x)e−inx dx =
2π −π 2π −π
Z π ix Z π Z π
e − e−ix −inx

1 1
e dx = eix(1−n) dx − e−ix(1+n) dx .
2π −π 2i 4πi −π −π
We can simplify this expression making the following changes,

1−n=k 1 + n = k′ , (1)

resulting in, Z π Z π 
1 ikx −ik′ x
e dx − e dx .
4πi −π −π

It is easy to realise that these integrals only have solution for k = 0 and k ′ = 0 (same solution 2π),
elsewhere the will both be zero.

So, looking at eq.1, we see that the only cn that do not vanish are c±1 . Obtaining these results for
each of them,
1 1
c1 = · 2π = ,
4πi 2i
1 1
c−1 = · (−2)π = − .
4πi 2i

The summatory that we are being asked in the exercise has a range from 1 to ∞, so it would only
be taking into account c1 , resulting in,
∞ 2
X 1 1
|cn |2 = |c1 |2 = = .
n=1
2i 4

2
P∞Problem2
3. Let f (x) = 1 on [−π, π]. Show that the Parseval inequality gives the bound
−∞ |cn | ⩽ 1.

Th.1 gives us a bound for the sum of the coefficients, so we can also express it as an inequality,
∞ Z π
X 1
|cn |2 ⩽ |f (x)|2 dx.
−∞
2π −π

Now, solving the integral for f (x) = 1, we get

∞ Z π Z π π
X 1 1 1 2π
|cn |2 ⩽ |f (x)|2 dx = 1dx = ·x = = 1.
−∞
2π −π 2π −π 2π −π 2π

Problem 4. For the function f (x) = eix on [−π, π], find the sum of the squares of the Fourier
coefficients.

Again, using Th.1 for the sum of the squares of the Fourier coefficients,
∞ Z π Z π
X
2 1 ix 2 1 2π
|cn | = |e | dx = 1dx = = 1.
−∞
2π −π 2π −π 2π

Where we have used p

|eix | = cos2 (x) + sin2 (x),
|eix |2 = cos2 (x) + sin2 (x) = 1 .

Problem 5. Let f (x) = ex on [0, 1]. Give at least two sequences of polynomials {pn } and {qn }
such that f can be approximated uniformly on [0, 1] by these sequences.

For the first sequence {pn } we will do a Taylor expansion for the function f around x = 0. But
first, lets recall the Taylor formula around a certain point a,

f ′ (a) f ′′ (a) f (n) (a)

f (x) = f (a) + (x − a) + (x − a)2 + ... + (x − a)n ,
1! 2! n!
the bigger the n, the better the approximation.

So, for f (x) = ex around x = 0, we get

n
X xk
pn = ,
k!
k=0

3
which converges uniformly to f as n → ∞.

For the second sequence, {qn }. We will use the Bernstein polynomial approximation, lets also recall
it first.
Pn
Theorem 2 (B1). Let f : [0, 1] → R be a bounded function, and Bn (f ; x) = k=0 f nk nk xk (1 −

x)n−k then
lim Bn (f ; x) = f (x)
n→−∞
for each x ∈ [0, 1] where f is continuous. Furthermore, if f ∈ C[0, 1] then Bn (f ; x) converges uniformly
to f .

So, for f (x) = ex we get

n  
X k n k
qn (x) = en x (1 − x)n−k ,
k
k=0

which converges uniformly to f as n → ∞.

Problem 6. Let f, w ∈ C[a, b] and w(x) ⩾ 0. Prove that

s s
Z b Z b Z b
|f (x)|w(x)dx ⩽ w(x)dx |f (x)|2 w(x)dx
a a a

Hint: Use the Cauchy-Schwarz inequality

First of all, the Cauchy-Schwarz inequality for integrals is the following,

!2
Z b Z b Z b
u(x)v(x) dx ⩽ u(x)2 dx · v(x)2 dx. (2)
a a a

Taking the square root of both sides of the expression,

s s
Z b Z b Z b
u(x)v(x) dx ⩽ u(x)2 dx · v(x)2 dx.
a a a

In order to apply this formula to our demonstration, the following ”change of variables” is done
p p
u(x) = w(x) , v(x) = |f (x)| · w(x).

Replacing on the last equation we get,

s s
Z b Z b p Z b p
|f (x)|w(x) ⩽ ( w(x))2 dx · |f (x)|2 ( w(x))2 dx
a a a
s s
Z b Z b Z b
|f (x)|w(x) ⩽ w(x)dx · |f (x)|2 w(x)dx.
a a a

4
 Problem P 7. Prove that if f is a continuous 2π-periodic function whose Fourier coefficients have
the property (|an | + |bn |) < ∞, then the Fourier serier converges uniformly to f .

We know that if f˜ converges uniformly to f , then ∀ϵ > 0, ∃N with n ≥ N such that we have,

|f˜n − f | ≤ ϵ .

We know that if f is a 2π-periodic function, it has a Fourier series of the form:

∞
a0 X
f (x) ∼ f˜ = + (ak cos(kx) + bk sin(kx)) ,
2
k=1

where an and bn are the Fourier coefficients. Then we need to find an ϵ that does not depend on x.
Lets consider the partial sum of f˜n ,

n
a0 X
f˜n = + (ak cos(kx) + bk sin(kx)) .
2
k=1

Then |f˜n − f | becomes,

∞
X
|f˜n − f | = | (ak cos(kx) + bk sin(kx))| .
k=n+1

Notice that cos(nx) ≤ 1 and sin(nx) ≤ 1. Then we have,

∞
X ∞
X
| (ak cos(kx) + bk sin(kx))| ≤ |ak | + |bk | .
k=n+1 k=n+1

P∞ P∞
We know that k=1 |ak | + |bk | < ∞, then the remaining sum k=n+1 |ak | + |bk | gets smaller as n
increases and it does not depend on x, so it works as our bound:

∞
X
|f˜n − f | ≤ |ak | + |bk | = ϵ ,
k=n+1

So fn converges uniformly to f .

Problem 8. Prove that if all the Fourier coefficients of a continuous 2π-periodic function f vanish,
then f = 0. Hint:use Fejér theorem
Theorem 3. (Fejér)

Let f (x) be a continuous and periodic function on the interval [−π, π]. Let Sn (f )(x) denote the
n-th partial sum of the Fourier series of f (x):
n
X
Sn (f )(x) = ck eikx ,
k=−n

5
where ck are the Fourier coefficients of f .

The Fejér sum σn (f )(x) is defined as the average of the first n partial sums of the Fourier series
of f (x):
n
1 X
σn (f )(x) = Sk (f )(x).
n+1
k=0

If f (x) is continuous and periodic on [−0, 2π], then the Fejér sums σn (f )(x) converge uniformly to
f (x) on [0, 2π].

If the Fourier coefficients vanish then,

Sn (f )(x) = 0 ,

so,
σn (f )(x) = 0

As our f ∈ C̃[0, 2π], the Fejér sum (σn , which is constantly 0) converges uniformly to f , conse-
quently f = 0.

Problem 9. Starting with the Rformula f ≈ Bn f , where Bn is the Bernstein polynomial operator,
1 1
Pn k
we obtain the quadrature formula 0 f ≈ n+1 k=0 f ( n ). Prove this, and discuss the convergence.

The Bernstein polynomial operator Bn for a function f ∈ C[0, 1] is defined in Th.2 as

n   
X k n k
Bn f (x) = f x (1 − x)n−k .
n k
k=0

To prove the quadrature formula, consider the integral

Z 1
f (x) dx.
0

Using the Bernstein approximation, we have

Z 1 Z 1
f (x) dx ≈ Bn f (x) dx.
0 0

k


Now, since Bn f (x) is a sum involving the function values f n , we can write
Z 1 Z n
1X   
k n k
Bn f (x) dx = f x (1 − x)n−k dx.
0 0 k=0 n k

By interchanging the summation and integration, we get

Z 1 n   Z 1
X k n
Bn f (x) dx = f xk (1 − x)n−k dx.
0 n k 0
k=0

After this, the problem now is to prove that

 Z 1
n 1
xk (1 − x)n−k dx = .
k 0 n+1

6
With the help of the Gamma function and its properties,
Z 1
Γ(x)Γ(y)
tx−1 ty−1 dt = , Γ(n) = (n − 1)! ,
0 Γ(x + y)

we can perform the previous integral with a change a simple change of variables and obtain
Z 1
k!(n − k)!
xk (1 − x)n−k dx = .
0 (n + 1)!

n n!


Since k = k!(n−k)! , this simplifies to
Z 1 n  
1 X k
Bn f (x) dx = f .
0 n+1 n
k=0

Therefore, we have shown that

Z 1 n  
1 X k
f (x) dx ≈ f .
0 n+1 n
k=0

Now we are going to discuss the convergence of the quadrature. For this, we are going to use the
following theorem:
Theorem 4. If fn (x) is a sequence of functions that converges uniformly to f (x), both defined on
[a, b] and are Riemann integrable, then
Z 1 Z 1 Z 1
lim fn (x) dx = lim fn (x) dx = f (x) dx .
n→∞ 0 0 n→∞ 0

Recalling Th.2, as n → ∞, the Bernstein operator Bn f converges uniformly to f for all f ∈ C[0, 1].
Therefore, the quadrature formula
Z 1 n  
1 X k
f (x) dx ≈ f
0 n+1 n
k=0

R1
converges to 0 f (x) dx as n → ∞ because of Th.4. This convergence is guaranteed for continuous
functions, but since this condition in Th.2 cannot be dropped, we cannot say that it still holds for
bounded functions defined on [0, 1]. All we would have to do to prove it is to find a counter-example.