\[
\newcommand{\C}{\mathbb{C}}
\newcommand{\haar}{\mathsf{m}}
\newcommand{\B}{\mathscr{B}}
\newcommand{\P}{\mathcal{P}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\g}{>}
\newcommand{\l}{<}
\newcommand{\intd}{\,\mathsf{d}}
\newcommand{\Re}{\mathsf{Re}}
\newcommand{\area}{\mathop{\mathsf{Area}}}
\newcommand{\met}{\mathop{\mathsf{d}}}
\newcommand{\emptyset}{\varnothing}
\DeclareMathOperator{\borel}{\mathsf{Bor}}
\newcommand{\symdiff}{\mathop\triangle}
\DeclareMathOperator{\leb}{\mathsf{Leb}}
\DeclareMathOperator{\cont}{\mathsf{C}}
\DeclareMathOperator{\lpell}{\mathsf{L}}
\newcommand{\lp}[1]{\lpell^{\!\mathsf{#1}}}
\]
Week 5 Worksheet Solutions
-
- The periodic sequence
\[
x = \mathtt{0101010101010101010101010101\cdots}
\]
has a finite orbit because $T^2(x) = x$.
- The periodic sequence
\[
x = \mathtt{0101010101010101010101010101\cdots}
\]
has an orbit that is not dense, because if
\[
y = \mathtt{1111111111111111111111111111\cdots}
\]
then $\met(T^n(x),y) \ge \tfrac{1}{4}$ for all $n \in \N$.
- Let $n \mapsto w_n$ be an enumeration of all finite strings of zeroes and ones. Let $x$ be the sequence obtained by contatenating all the $w_n$. For any $y \in \{0,1\}^\N$ and any $\epsilon > 0$ there is $r \in \N$ so that $Tr(x)$ begins with $y(1) \cdots y(m)$ so that $\met(T^r(x),y) \l \epsilon$ if $m$ is chosen large enough.
- One calculates
\[
\begin{aligned}
T^{n+1}(x,y) &{} = T(T^n(x,y)) \\
&{} = T(x + n \alpha, y + nx + \tbinom{n}{2} \alpha ) \\
&{} = (x + n \alpha + \alpha, y + nx + \tbinom{n}{2} + x + n\alpha) \\
&{} = (x + (n+1)\alpha, y + (n+1)x + \tbinom{n+1}{2} \alpha)
\end{aligned}
\]
by induction.
- When $\alpha$ is irrational.
- Let $R \g 0$ be a bound on the sequence $n \mapsto x_n$. Fix $k \in \N$. We have
\[
\left| \sum_{n=1}^N x_n - \sum_{n=k}^{N+k} x_n \right| \le |x_1| + \cdots + |x_k| + |x_{N+1}| + \cdots + |x_{N+k}| \le 2kR
\]
which converges to zero upon division by $N$, giving the desired result.
- No, it is too lethargic. Indeed
\[
\log(2^{N+1}) - \log(2^N) = \log 2 \approx 0.30103
\]
so if we choose (as we may) a sequence $k(N)$ such that $\log(2^{k(N)}) \bmod 1$ converges, say to $\rho$, then the frequency
\[
\dfrac{|\{ 1 \le n \le 2^{k(N+1)} : \log(n) \notin [\rho,\rho + \log(2) )\}|}{2^{k(N+1)}}
\]
will not be larger than $\tfrac{1}{2}$ yet the set
\[
[0,1) \setminus [\rho,\rho + \log 2)
\]
has length strictly larger than $\tfrac{1}{2}$.
-
- Say that $x_n$ is uniformly distributed in $X$ if
\[
\lim_{N \to \infty} \dfrac{|\{ 1 \le n \le N : x_n \in [a,b) \times [c,d) \}|}{N} = (d-c)(b-a)
\]
for all $0 \le a \l b \le 1$ and all $0 \le c \l d \le 1$.
- Never, because the sequence never belongs to the set $[\tfrac{1}{8},\tfrac{2}{8}) \times [\tfrac{6}{8},\tfrac{7}{8})$.
-
- By passing to a subsequence one can arrange for
\[
N \mapsto \dfrac{1}{N} \sum_{n=1}^N f(T^n x)
\]
to converge for every continuous function $f : X \to \R$ by an application of the Stone-Weierstrass theorem.
The map $\phi : \mathsf{C}(X) \to \R$ defined by
\[
\phi(f) = \lim_{N \to \infty} \dfrac{1}{r(N)} \sum_{n=1}^{r(N)} f(T^n(x))
\]
is then a bounded linear functional on $\mathsf{C}(X)$ and there is therefore, by the Riesz representation theorem, a measure $\mu$ on $X$ with
\[
\lim_{N \to \infty} \dfrac{1}{r(N)} \sum_{n=1}^{r(N)} f(x_n) = \int f \intd \mu
\]
\for all $f \in \mathsf{C}(X)$.
- Yes, because both the sequences $r$, $s$ and the measure $\mu$,$\nu$ can be different. Explicitly, one can define a sequence $n \mapsto x_n$ by $x_n = 0$ whenever $2^{2k-1} \l n \le 2^{2k}$ and $x_n = \tfrac{1}{2}$ whenever $2^{2k} \l n \le 2^{2k+1}$.
- Since the orbit $\{n \alpha \bmod 1 : n \in \N \}$ is dense in $[0,1)$ we can find $n_1,\dots,n_6$ with
\[
\bigcup_{i=1}^6 (\tfrac{1}{4},\tfrac{3}{4}) - n_i \alpha = [0,1)
\]Say that $E \subset \N$ is syndetic if there is $r(1),\dots,r(k)$ with
\[
\N = (E - r(1)) \cup \cdots \cup (E - r(k))
\]
and taking
\[
E = \{ n \in \N : n \alpha \bmod 1 \in (\tfrac{1}{4},\tfrac{3}{4}) \}
\]
gives the desired result.