Two problems of Chapter 2

Fourier Analysis

Problem 1

  One can construct Riemann integrable functions on [0,1][0,1] that have a dense set of discontinuities as follows.

 (a) Let f(x)=0f(x)=0, when x<0x<0, and f(x)=1f(x)=1 if x0x\geq0. Choose a countable dense sequence {rn}\{r_n\} in [0,1][0,1]. Then, show that function

F(x)=n=11n2f(xrn)F(x)=\sum_{n=1}^{\infty} \dfrac{1}{n^2} f(x-r_n)

is integrable and has discontinuities at all points of the sequence {rn}\{r_n\}

Proof:

 Step one: Note that F(x)F(x) is monotonic and bounded for the definition of ff,and F(x)F(x) \leq Euler’s series, so that it has countable discontinuities at most. Hence, F(x)F(x) is integrable on [0,1][0,1].

  Step two: We consider that F(rn)F(r_n). {rn}\{r_n\} is dense in [0,1][0,1] means ε>0,\forall \varepsilon >0, , nN\exists n \in \mathbb{N} such that xrn<ε|x-r_n|<\varepsilon. Or x1,x2[0,1]\forall x_1,x_2 \in [0,1], nN\exists n \in \mathbb{N} such that rnr_n is between x1x_1 and x2x_2. Fixed x2=rnx_2 = r_n, n=1,2,3,...n=1,2,3,... and consider that:

F(x)F(rn)|F(x)-F(r_n)|

δ>0\forall \delta > 0, when xrn<δ|x-r_n| < \delta, we can find rmr_m is between xx and rnr_n. Due to F(x)F(x) is monotonic, we have

Where Cn,mC_{n,m} is a constant(If we fix mm and nn). That is F(x)F(x) is discontinuous at all points x=rnx=r_n.

(#)(\#)

  (b)Consider next

F(x)=n=13ng(xrn)F(x)=\sum_{n=1}^{\infty} 3^{-n}g(x-r_n)

where g(x)=sin1xg(x)=\sin{\dfrac{1}{x}} when x0x\neq0, and g(0)=0g(0)=0. Then FF is integrable, discontinuous at each x=rnx=r_n, and fails to be monotonic in any subinterval of [0,1][0,1].

Proof:

  Note that g(x)g(x) is bounded, and the geometric series 3n\sum 3^{-n} is convergence. So F(x)F(x) is bounded.

Next, we discuss the integrability of F(x)F(x). Denote

U=j=1N[supxj1xxj{n=13ng(xrn)}](xjxj1)\mathcal{U}=\sum_{j=1}^{N}{\left[\sup_{x_{j-1} \leq x \leq x_j}\left\{\sum_{n=1}^{\infty} 3^{-n}g(x-r_n)\right\}\right](x_j-x_{j-1})}

L=j=1N[infxj1xxj{n=13ng(xrn)}](xjxj1)\mathcal{L}=\sum_{j=1}^{N}{\left[\inf_{x_{j-1} \leq x \leq x_j}\left\{\sum_{n=1}^{\infty} 3^{-n}g(x-r_n)\right\}\right](x_j-x_{j-1})}

Then,

UL=j=1Nωj(x)(xjxj1)\mathcal{U-L}=\sum_{j=1}^{N}{\omega_j(x)(x_j-x_{j-1})}

Where

So, we prove that there is at most countable ωj(x)>0\omega_j(x) > 0 as NN\to \infty.

 © The original example of Riemann is the function

F(x)=n=1(nx)n2F(x)=\sum_{n=1}^{\infty} \dfrac{(nx)}{n^2}

where (x)=x(x)=x for x(12,12]x \in (-\dfrac{1}{2},\dfrac{1}{2}], and (x)(x) is continued to R\mathbb{R} by periodicity, that is, (x+1)=(x)(x+1)=(x). It can be shown that FF is discontinuous whenever x=m2nx=\dfrac{m}{2n}, where m,nZm,n\in \mathbb{Z} with mm odd and n0n \neq 0.

Proof:

Problem 2

  Let DND_N denote the Dirichlet kernel

DN(θ)=k=NNeikθ=sin(N+12θ)sinθ2D_N(\theta)=\sum_{k=-N}^{N}e^{ik\theta}=\dfrac{\sin\left(N+\dfrac{1}{2} \theta\right)}{\sin\dfrac{\theta}{2}}

and define

LN=12πππDN(θ)dθL_N=\dfrac{1}{2\pi}\int_{-\pi}^{\pi}\left|D_N(\theta)\right|d\theta

  (a)Prove that

LNclogNL_N \geq c\log N

for some constant c>0c>0.

  Proof:

  We calculate LNL_N straitly:

The last inequality is used Euler’s formula

k=1n1k=logn+γn\sum_{k=1}^{n}{\dfrac{1}{k}}=\log n + \gamma_n

Where γn\gamma_n is Euler constant.

  A more careful estimate gives

LN4π2logN+O(1)L_N \ge \dfrac{4}{\pi^2} \log N +O(1)

  (b)Prove the following as consequence: for each n1n\geq 1, there exists a continuous function fnf_n such that fn<1\left|f_n \right|<1 and Sn(fn)(0)clogn\left| S_n(f_n)(0)\right| \geq c' \log n.