\[
\newcommand{\C}{\mathbb{C}}
\newcommand{\haar}{\mathsf{m}}
\newcommand{\P}{\mathcal{P}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\d}{\mathsf{d}}
\newcommand{\g}{>}
\newcommand{\l}{<}
\newcommand{\intd}{\,\mathsf{d}}
\newcommand{\Re}{\mathsf{Re}}
\newcommand{\area}{\mathop{\mathsf{Area}}}
\newcommand{\met}{\mathop{\mathsf{d}}}
\newcommand{\emptyset}{\varnothing}
\DeclareMathOperator{\borel}{\mathsf{Bor}}
\]
Week 10 Tutorial Solutions
- $R(x) = \mathtt{0010011010}\cdots$
- $R(x) = \mathtt{1010100101}\cdots$
- The rule does not make sense at the constant sequence
\[
x = \mathtt{11111111111111} \cdots
\]
because no $x(n)$ is equal to 0. Define $R$ at this point to take the value
\[
\mathtt{000000000000} \cdots
\]
and think of the odometer resetting.
- $R^{-1}([1]) = [0]$ because if $x(1) = 0$ then $(R(x))(1) = 1$ and if $(R(x))(1) = 1$ it must have been the case that $\gamma(x) = 1$ and therefore $x(1) = 0$.
- Let $\mu$ be the $(1-p,p)$ coin measure. Thus $\mu([1]) = p$ and $\mu([0]) = 1-p$. If $\mu$ is $R$ invariant then
\[
p = \mu([1]) = \mu(R^{-1}([1])) = \mu([0]) = 1-p
\]
which is impossible unless $p = 0.5$.
- Since $R^{-1}([0]) = [1]$ and $R^{-1}([1]) = [0]$ we have
\[
f \circ R = 1_{R^{-1} [0]} - 1_{R^{-1}[1]} = 1_{[1]} - 1_{[0]}
\]
and from
\[
1_{[1]} - 1_{[0]} = - (1_{[0]} - 1_{[1]})
\]
we see it is an eigenfunction of eigenvalue $-1$.
- Take
\[
\begin{align*}
f_1 &{} = 1_{[00]} +1\cdot 1_{[10]} + 1\cdot 1_{[01]} +1\cdot 1_{[11]} \\
f_2 &{} = 1_{[00]} +i\cdot 1_{[10]} - 1\cdot1_{[01]} -i\cdot 1_{[11]} \\
f_3 &{} = 1_{[00]} -1\cdot 1_{[10]} + 1\cdot1_{[01]} -1\cdot 1_{[11]} \\
f_4 &{} = 1_{[00]} -i\cdot 1_{[10]} -1\cdot 1_{[01]} + i\cdot 1_{[11]}
\end{align*}
\]
which are eigenfunctions of $R$ with eigenvalues $1$, $i$, $-1$ and $-i$ respectively. As their eigenvalues are distinct they are linearly independent.
- From
\[
1_{[0]} - 1_{[1]} = 1_{[00]} + 1_{[01]} - 1_{[10]} - 1_{[11]}
\]
we see that $1_{[0]} - 1_{[1]}$ and $f_3$ span the same subspace.
- We can write
\[
\begin{align*}
1_{[00]} &{} = \tfrac{1}{4} ( f_1 + f_2 + f_3 + f_4) \\
1_{[10]} &{} = \tfrac{1}{4} ( f_1 - i f_2 - f_3 + i f_4) \\
1_{[01]} &{} = \tfrac{1}{4} ( f_1 - f_2 + f_3 - f_4) \\
1_{[11]} &{} = \tfrac{1}{4} ( f_1 + i f_2 - f_3 - i f_4)
\end{align*}
\]
by some linear algebra.
- The collection $\mathcal{D}$ of cylinders (including the empty set) is a π-system so it suffices to check that $\mu(R^{-1}(C)) = \mu(C)$ for every cylinder $C$. But for every cylinder $[\epsilon(1) \cdots \epsilon(r)]$ its inverse image under $R$ is another cylinder of the same length, defined by the symbols $\delta(1)\cdots\delta(r)$ that odometrically precede $\epsilon(1) \cdots \epsilon(r)$. With respect to the fair coin measure, all cylinders of length $r$ have the same measure. Thus $\mu$ is an invariant measure for $R$.
- Mimicking the answer to 7 with longer cylinders, we can find for every $\eta \in \C$ with $|\eta| = 1$ and $\eta^{2^n} = 1$ for some $n \in \N$ an eigenfunction of $R$ with eigenvalue $\eta$. As, for every cylinder set $C$, we can write $1_C$ as a linear combination of such eigenfunctions, we can conclude that the eigenfunctions span $\mathsf{L}^{\!\mathsf{2}}(X,\mathscr{B},\mu)$. There are therefore no other eigenvalues of $R$ and $R$ has discrete spectrum.