Processing math: 100%

Thursday, 30 October 2014

Borel Sets of Infinite Binary Sequences


Note: The content of this blog is an expansion of some results in Chapter 2 of Parthasarathy's Book on Probability measures on Metric Spaces.

M=\{0,1\}^\infty is a compact metric space as shown previously. Let X,Y be two metric spaces. Borel sets B_1 \subset X, B_2 \subset Y are said to be isomorphic if there exists a bijective map, \phi : B_1 \to B_2 such that both \phi, \phi^{-1} are measurable. This is denoted as B_1 \sim B_2.

Here the sigma algebra on B_1 is B_1 \cap \mathcal{B}_X and similarly on B_2. Measurability of \phi,\phi^{-1} should be thought of in this sense. (Note that B_1 \cap \mathcal{B}_X = \mathcal{B}_{B_1}). Throughout, \mathcal{B}_X represents the \sigma-algebra generated by the open sets in X.
There exists a Borel set E \subset M such that E \sim [0,1]
Define \tau : M \to [0,1] by \tau(\{x_1,x_2,\cdots\}) = \sum_{i=1}^\infty \frac{x_i}{2^i}. Then \tau is continuous from the uniform limit theorem. \tau is also onto since every number in [0,1] has a binary expansion. Let E be set of all binary sequences which contain infinitely many zeros. Then, M - E is countable. As each singleton in a metric space is a closed set, M-E is a Borel set. Hence, E is also Borel.

The restriction of \tau to E is a bijective map as every number in [0,1] has exactly one binary expansion which contains infinitely many zeros. As \tau is continous, it is measurable map from M \to [0,1] and hence \tau : E \to [0,1] is also measurable.

It needs to be shown that \tau^{-1} is also measurable. \tau^{-1} : [0,1] \to E and it needs to be shown that \tau(A) is Borel in [0,1] for all A Borel in E i.e. A \in \mathcal{B}_E. As \mathcal{B}_E = E \cap \mathcal{B}_M and the fact that \mathcal{B}_M is generated by the sets F_j = \{ \{x_1,x_2\cdots\} | x_j = 1\} it is enough to show that \tau(A) is Borel for all A of the form E\cap F_j.

To understand what \tau(E\cap F_j) is, it helps to fix the first j-1 entries and let the entries from j+1 to \infty be arbitrary. Each such set gives rise to finite union of intervals whose end points are dyadic rationals. Hence \tau(E\cap F_j) is Borel.
M is homeomorphic to M^\infty
Let x be an infinite binary sequence. From this we construct a sequence of infinite binary sequences as follows. From the given sequence x choose elements alternatingly to obtain the first sequence. After removing this subsequence from the original sequence, the procedure is repeated ad infinitum. This can be written mathematically as \{x_1,x_2,x_3,\cdots,\} \to \{x^1,x^2,x^3,\cdots\} where each x^j \in M and ith element of the jth sequence x^j_i = x_{2^{j-1}2(i-1)}. Clearly, this map is a bijection. \begin{array} dd(\{x^1,x^2,x^3,\cdots\}, \{y^1,y^2,y^3,\cdots\}) &= \sum_{j=1}^\infty \frac{1}{2^j} \sum_{i=1}^\infty \frac{1}{2^i} \frac{d_M(x^j_i,y^j_i)}{1+d_M(x^j_i,y^j_i)} \\ &= \sum_{j=1}^\infty \frac{1}{2^j} \sum_{i=1}^\infty \frac{2^{2^{j-1} + (2i-1)}}{2^i} \frac{1}{2^{2^{j-1} + (2i-1)}} \frac{d_M(x_{2^{j-1} + (2i-1)},y_{2^{j-1} + (2i-1)})}{1+d_M(x_{2^{j-1} + (2i-1)},y_{2^{j-1} + (2i-1)})} \\ &\geq d_M(x,y) \end{array} The other direction is not too hard to show.
There exists a Borel set E_1 \subset M such that E_1 \sim I^\infty
Let \tau,E be the same as above. Define \tau^\prime : M^\infty \to I^\infty by \tau^\prime(x^1,x^2,\cdots) = (\tau(x^1),\tau(x^2),\cdots) where x^i = \{x^i_1,x^i_2,\cdots\} \in M. We show that \tau^\prime is continuous.

\begin{array} dd_{I^\infty}(\tau^\prime(x),\tau^\prime(y)) &= \sum_{i=1}^\infty \frac{1}{2^i}\frac{|\tau(x^i) - \tau(y^i)|}{1+|\tau(x^i) - \tau(y^i)|} \\ &= \sum_{i=1}^\infty \frac{1}{2^i}\frac{|\sum_{j=1}^\infty \frac{1}{2^j}(x^i_j - y^i_j)|}{1+|\sum_{j=1}^\infty \frac{1}{2^j}(x^i_j - y^i_j)|} \\ &\leq \sum_{i=1}^\infty \frac{1}{2^i} \sum_{j=1}^\infty \frac{1}{2^j} {|x^i_j - y^i_j|} \\ &\leq d_{M^\infty}^0(x,y) \end{array} where d_{M^\infty}^0 is equivalent to d_{M^\infty}(x,y) = \sum_{i=1}^\infty \frac{1}{2^i}\frac{d_M(x^i,y^i)}{1+d_M(x^i,y^i)}. Hence \tau^\prime is continuous and therefore measurable. Now, it will be shown that \tau^\prime : E^\infty \to I^\infty is an isomorphism.
  • \tau^\prime is bijective as \tau : E \to I is bijective.
  • E^\infty is a measurable set in \mathcal{B}_{M^\infty}
  • (\tau^\prime)^{-1} : I^\infty \to E^\infty is measurable. \mathcal{B}_{E^\infty} = E^\infty \bigcap \mathcal{B}_{M^\infty} = \sigma(\{E^\infty \cap \rho_n^{-1}(U_{ni}), U_n \in \mathcal{D}\}) where \mathcal{D}_i is a denumerable base for M. However, each U_n can in-turn be written as a countable union of finite intersection of sets of the form V_{m,j}=\{\gamma_m^{-1}(\{j\}), j \in \{0,1\}\} (where \gamma_m : M \to \{0,1\} is the co-ordinate projection). Hence, \mathcal{B}_{E^\infty} = \sigma(\{E^\infty \cap \rho_n^{-1}(V_{mj})\}). \tau^\prime(E^\infty \cap \rho_n^{-1}(V_{mj})) = \tau^\prime(E \times E \times \cdots \times E\cap V_{mj} \times E \times E \cdots) =\tau(E) \times \tau(E) \cdots \times \tau(E \cap V_{mj}) \times \tau(E) \times \tau(E) \cdots. Therefore, (\tau^\prime)^{-1} is also measurable.
As M and M^\infty are homeomorphic, they are also isomorphic (in the sense defined initially) and it follows that E^\infty is isomorphic to some E_1 \subset M. Therefore, we get that E_1 \sim I^\infty.

No comments:

Post a Comment