Constructing particular sequence of finitely supported measures converging to a given measure

17 Views Asked by At

Let $(\Omega, \mathcal{B}, \mu)$ be a measure space where $\Omega \subset \mathbb{R}^K$ is the unit sphere in the $K$-dimensional space, and $\mu$ is a finite, positive measure (not necessarily a probability measure). Let $\mu(\Omega)=m \in (0,\infty)$. The CDF of $\mu$ -- let us call it $F$ -- admits a continuous density $f$ which is strictly positive on $[0,1]$. I want a construct a sequence of finitely supported measures $\mu_n$, with supports contained in $\Omega$, converging (weakly) to $\mu$, ensuring the following property: If $x,x'$ are in the support of both $\mu_n$ and $\mu$ then we must have $\mu_n({x})/\mu_n({x'})=f({x})/f({x'})$.

I have constructed the distribution sequence as follows. Define $\Omega_n = \{x \in [0,1]: x = k/n\;\text{for some}\;k \in \mathbb{Z}\}$. Define:

$$f_n(x)=\begin{cases} &\frac{mf(x)}{\sum\limits_{x'\in \Omega_n}f(x')}\;\text{if}\;x \in \Omega_n,\\ &0,\;\text{otherwise}. \end{cases}$$

Fix $x\in \Omega$.

\begin{align*} \therefore\;F_n(x)&=\frac{m\sum\limits_{x'\in \Omega_n, x' \leq x}f(x')}{\sum\limits_{x'\in \Omega_n}f(x')}\\ &=\frac{m\sum\limits_{x'\in \Omega_n, x' \leq x}f(x')\times\frac{1}{n^K}}{\sum\limits_{x'\in \Omega_n}f(x')\times\frac{1}{n^K}} \end{align*}

where $x' \leq x$ indicates componentwise inequality.

By the definition of the Reimann integral, the limit of the numerator and the denominator as $n \rightarrow \infty$ are $m\int\limits_{x' \in \Omega, x'\leq x}f(x')dx'$ and $\int\limits_{x' \in \Omega}f(x')dx'$ respectively, which gives us pointwise convergence of the CDF's.

I have two questions:

  1. Is the argument for showing pointwise convergence of the sequence of CDF's correct?
  2. Is pointwise convergence of the sequence of CDF's sufficient for weak convergence of the corresponding sequence of measures, like it is in case of probability measures?

Any help is most appreciated.