FIND ME ON

GitHub

LinkedIn

Showing F_m is closed

🌱

Since vv is Continuous:v(x)μ(dx):μv(x)μ(dx)\int v(x) \mu(dx): \mu \mapsto \int v(x) \mu(dx) is lower semicontinuous.

This is because we can write vv as the pointwise limit of nondecreasing family (vn)nN(v_{n})_{n\in \mathbb{N}} of continuous real-valued functions by

Let L(X)L(X) be the family of all functions on XX that are lower semicontinuous and bounded below. Then vL(X)    {vn}Cb(X):vnv.v\in L(X)\iff \exists \{ v_{n} \}\subset C_{b}(X):v_{n}\uparrow v.

Then using this we can prove the following lemma

Let X\mathcal{X} be polish spaces and v:XR{+}v:\mathcal{X}\to \mathbb{R}\cup \{ +\infty \} a nonnegative, lower semicontinuous function. Let (μn)nN(\mu_{n})_{n\in \mathbb{N}} be a sequence of probability measures weakly converging to some μP(X)\mu \in \mathcal{P}(\mathcal{X}). Then vdμlim infkvdμk\int\limits v \, d\mu \le \liminf_{ k \to \infty } \int\limits v \, d\mu_{k} i.e. μvdμ\mu\mapsto \int\limits v \, d\mu is lower semicontinuous.

https://math.stackexchange.com/questions/3859243/characterization-of-weak-convergence-with-lower-semicontinuity \begin{proof} We first know that by the definition of lower semicontinuity that for l.s.c. ff, the set {xX:f(x)>α},αR\{ x\in X:f(x)>\alpha \},\,\alpha \in \mathbb{R}is Open. Hence, we can then apply Portmanteau’s Theorem to get μ({xX:f(x)>α})lim infkμk({xX:f(x)>α}).\mu(\{ x\in X:f(x)>\alpha \})\le\liminf_{ k \to \infty }\mu_{k}(\{ x\in X:f(x) >\alpha \}) .Now, integrating over the positive reals and applying Fatou’s Lemma we get 0μ({xX:f(x)>α})dα0lim infkμk({xX:f(x)>α})dαlim infk0μk({xX:f(x)>α})dα\begin{align*} \int\limits _{0}^{\infty}\mu(\{ x\in X:f(x)>\alpha \}) \, d\alpha &\le \int\limits _{0}^{\infty}\liminf_{ k \to \infty } \mu_{k}(\{ x\in X:f(x)>\alpha \}) \, d\alpha\\ &\le \liminf_{ k \to \infty } \int\limits _{0}^{\infty}\mu_{k}(\{ x\in X:f(x)>\alpha \}) \, d\alpha \end{align*} where the first inequality is simply by monotonicity of the integral and the second is by Fatou’s. Hence, Xfdμlim infnfdμn\int\limits _{\mathbb{X}}f \, d\mu \le\liminf_{ n \to \infty }\int\limits f \, d\mu_{n} \end{proof}

\begin{proof} By Monotone Convergence Theorem we have v(x)μ(dx)=limnvn(x)μ(dx)(MCT)=limn(limkvn(x)μk(dx))(Portmanteau)limn(lim infkvn(d)μk(dx))(w.c of μ)lim infkv(x)μk(dx)(MCT)\begin{align*} \int\limits v(x) \, \mu(dx) &= \lim_{ n \to \infty } \int\limits v_{n}(x) \, \mu(dx) &\text{(MCT)}\\ &= \lim_{ n \to \infty } \left( \lim_{ k \to \infty } \int\limits v_{n}(x) \, \mu_{k}(dx) \right)&\text{(Portmanteau)}\\ &\le \lim_{ n \to \infty } \left( \liminf_{ k \to \infty } \int\limits v_{n}(d) \, \mu_{k}(dx) \right) &\text{(w.c of }\mu)\\ &\le \liminf_{ k \to \infty } \int\limits v(x) \, \mu_{k}(dx) &\text{(MCT)} \end{align*}so we first apply MCT to get an integral in terms of a continuous and bounded function. This allows us to apply Portmanteau’s Theorem to then write the integral as a limit of the measure, we then get the inequality from the fact that vnCb(X)v_{n}\in C_{b}(\mathbb{X}) meaning the integral itself is continuous and bounded by the Feller Property property of μk\mu_{k} and hence is lower semicontinuous. Then we apply MCT again to take the limit of the function.

\end{proof}

Therefore, if μn\mu_n is a sequence in FmF_m converging to μ\mu; we have thatmlim infnμn(dx)v(x)μ(dx)v(x)m \geq \liminf_{n \to \infty} \int \mu_n(dx) v(x) \geq \int \mu(dx) v(x) and therefore  μ(dx)v(x)m\int \mu(dx) v(x) \leq m. This shows that μFm\mu \in F_{m}, and therefore the set FmF_m is Closed.