Lemma 25.6.4 (Existence of Simple Approximations of the Identity).label Let $X$ be a separable metric space, $\mathcal{A}: X \to 2^{X}$ be a Borel measurable admissible approximant function, and $\seq{x_n}\subset X$ be a dense subset with $x_{1} \in \bigcap_{x \in X}\mathcal{A}(x)$, then there exists $\seq{I_n}\subset X^{X}$ such that:

  1. (1)

    $\seq{I_n}$ is a Borel measurable, $\mathcal{A}$-admissible approximation of the identity.

  2. (2)

    For each $N \in \natp$, $I_{N}(X) \subset \bracsn{x_n|1 \le n \le N}$.

  3. (3)

    For each $N \in \natp$ and $x \in X$,

    \[d(x, I_{N}(x)) = \min\bracs{d(x, x_n)| 1 \le n \le N, x_n \in \mathcal{A}(x)}\]

Proof. By removing duplicate elements from the sequence, assume without loss of generality that for each $m, n \in \natp$ with $m \ne n$, $x_{m} \ne x_{n}$.

Let $N \in \natp$. For each $x \in X$, let

\[C_{N}(x) = \bracs{1 \le n \le N| x_n \in \mathcal{A}(x)}\]

Since $x_{1} \in \bigcap_{y \in X}\mathcal{A}(y)$, $1 \in C_{N}(x)$ and $C_{N}(x) \ne \emptyset$. Now, let

\[k_{N}(x) = \min\bracs{n \in C_N(x) \bigg | d(x, x_n) = \min_{m \in C_N(x)}d(x, x_m)}\]

be the minimum $n \in C_{N}(x)$ on which the minimal distance from $x$ to $\bracs{x_m|m \in C_N(x)}$ is achieved. Define

\[I_{N}: X \to X \quad x \mapsto x_{k_N(x)}\]

(2): For each $x \in X$, $k_{N}(x) \in [N]$, so $I_{N}(x) \in \bracsn{x_n|1 \le n \le N}$.

(3): Let $x \in X$, then by definition of $k_{N}$ and $C_{N}$,

\begin{align*}d(x, I_{N}(x))&= d(x, x_{k_N(x)}) = \min_{n \in C_N(x)}d(x, x_{n}) \\&= \min\bracsn{d(x, x_n)|1 \le n \le N, x_n \in \mathcal{A}(x)}\end{align*}

(1, Borel Measurable): Fix $N \in \natp$, then for each $n \in \natp$,

\begin{align*}\bracs{k_N \le n}&= \bigcup_{j = 1}^{n} \bracs{x \in X \bigg | j \in C_N(x), d(x, x_j) = \min_{m \in C_N(x)}d(x, x_m)}\\&= \bigcup_{j = 1}^{n} \bracs{j \in C_N}\cap \bracs{x \in X \bigg | d(x, x_j) = \min_{m \in C_N(x)}d(x, x_m)}\\&= \bigcup_{j = 1}^{n}\bigcup_{J \subset [N]}\bracs{j \in C_N, J = C_N}\cap \bracs{x \in X \bigg | d(x, x_j) = \min_{m \in J}d(x, x_m)}\end{align*}

Given that $\mathcal{A}$ is Borel measurable, $\bracs{n \in C_N}= \bracs{x_n \in \mathcal{A}(x)}$ is a Borel set for each $1 \le n \le N$. As a result, $\bracs{J = C_N}$ is Borel for each $J \subset [N]$. Thus $\bracs{j \in C_N, J = C_N}$ is Borel for each $1 \le j \le n$ and $J \subset [N]$.

On the other hand, for each $1 \le n \le N$, the function $x \mapsto d(x, x_{n})$ is continuous and hence Borel measurable. Similarly, for each $J \subset [N]$, the mapping $\real^{J} \to \real$ with $\alpha \mapsto \min_{j \in J}\alpha_{j}$ is also Borel measurable.

The above facts combined show that $\bracs{k_N \le n}$ is a Borel set, and $k_{N}: X \to [N]$ is a Borel measurable function. Now, let

\[I_{N}: X \to \bracsn{x_n|1 \le n \le N}\quad x \mapsto x_{k_N(x)}\]

By assumption that $\seq{x_n}$ are distinct, $\bracs{I_N = x_n}= \bracs{k_N = n}$ is a Borel set for each $1 \le n \le N$. Therefore $I_{N}$ is Borel measurable.

(1, $\mathcal{A}$-Admissible): Let $N \in \natp$ and $x \in X$, then

\[I_{N}(x) = x_{k_N(x)}\in \bracs{x_n|n \in C_N(x)}\subset \mathcal{A}(x)\]

(1, Approximation): Let $x \in X$ and $\eps > 0$. Since $x \in \ol{\mathcal{A}(x)^o}$ and $\seq{x_n}$ is dense in $X$, there exists $N_{0} \in \natp$ such that $x_{N_0}\in \mathcal{A}(x)$ and $d(x, x_{N_0}) < \eps$. By (3),

\[\limv{N}d(x, I_{N}(x)) = \limv{N}\min\bracs{d(x, x_n)| 1 \le n \le N, x_n \in \mathcal{A}(x)}= 0\]

$\square$

Post a Comment

Name:Email:
Please enter the tag of the current page (13U) to post the comment.
Tag: