Welcome to Randal Douc's wiki

A collaborative site on maths but not only!

User Tools

Site Tools


world:radon

$$ \newcommand{\arginf}{\mathrm{arginf}} \newcommand{\argmin}{\mathrm{argmin}} \newcommand{\argmax}{\mathrm{argmax}} \newcommand{\asconv}[1]{\stackrel{#1-a.s.}{\rightarrow}} \newcommand{\Aset}{\mathsf{A}} \newcommand{\b}[1]{{\mathbf{#1}}} \newcommand{\ball}[1]{\mathsf{B}(#1)} \newcommand{\bbQ}{{\mathbb Q}} \newcommand{\bproof}{\textbf{Proof :}\quad} \newcommand{\bmuf}[2]{b_{#1,#2}} \newcommand{\card}{\mathrm{card}} \newcommand{\chunk}[3]{{#1}_{#2:#3}} \newcommand{\condtrans}[3]{p_{#1}(#2|#3)} \newcommand{\convprob}[1]{\stackrel{#1-\text{prob}}{\rightarrow}} \newcommand{\Cov}{\mathbb{C}\mathrm{ov}} \newcommand{\cro}[1]{\langle #1 \rangle} \newcommand{\CPE}[2]{\PE\lr{#1| #2}} \renewcommand{\det}{\mathrm{det}} \newcommand{\dimlabel}{\mathsf{m}} \newcommand{\dimU}{\mathsf{q}} \newcommand{\dimX}{\mathsf{d}} \newcommand{\dimY}{\mathsf{p}} \newcommand{\dlim}{\Rightarrow} \newcommand{\e}[1]{{\left\lfloor #1 \right\rfloor}} \newcommand{\eproof}{\quad \Box} \newcommand{\eremark}{</WRAP>} \newcommand{\eqdef}{:=} \newcommand{\eqlaw}{\stackrel{\mathcal{L}}{=}} \newcommand{\eqsp}{\;} \newcommand{\Eset}{ {\mathsf E}} \newcommand{\esssup}{\mathrm{essup}} \newcommand{\fr}[1]{{\left\langle #1 \right\rangle}} \newcommand{\falph}{f} \renewcommand{\geq}{\geqslant} \newcommand{\hchi}{\hat \chi} \newcommand{\Hset}{\mathsf{H}} \newcommand{\Id}{\mathrm{Id}} \newcommand{\img}{\text{Im}} \newcommand{\indi}[1]{\mathbf{1}_{#1}} \newcommand{\indiacc}[1]{\mathbf{1}_{\{#1\}}} \newcommand{\indin}[1]{\mathbf{1}\{#1\}} \newcommand{\itemm}{\quad \quad \blacktriangleright \;} \newcommand{\jointtrans}[3]{p_{#1}(#2,#3)} \newcommand{\ker}{\text{Ker}} \newcommand{\klbck}[2]{\mathrm{K}\lr{#1||#2}} \newcommand{\law}{\mathcal{L}} \newcommand{\labelinit}{\pi} \newcommand{\labelkernel}{Q} \renewcommand{\leq}{\leqslant} \newcommand{\lone}{\mathsf{L}_1} \newcommand{\lrav}[1]{\left|#1 \right|} \newcommand{\lr}[1]{\left(#1 \right)} \newcommand{\lrb}[1]{\left[#1 \right]} \newcommand{\lrc}[1]{\left\{#1 \right\}} \newcommand{\lrcb}[1]{\left\{#1 \right\}} \newcommand{\ltwo}[1]{\PE^{1/2}\lrb{\lrcb{#1}^2}} \newcommand{\Ltwo}{\mathrm{L}^2} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mcbb}{\mathcal B} \newcommand{\mcf}{\mathcal{F}} \newcommand{\meas}[1]{\mathrm{M}_{#1}} \newcommand{\norm}[1]{\left\|#1\right\|} \newcommand{\normmat}[1]{{\left\vert\kern-0.25ex\left\vert\kern-0.25ex\left\vert #1 \right\vert\kern-0.25ex\right\vert\kern-0.25ex\right\vert}} \newcommand{\nset}{\mathbb N} \newcommand{\N}{\mathcal{N}} \newcommand{\one}{\mathsf{1}} \newcommand{\PE}{\mathbb E} \newcommand{\pminfty}{_{-\infty}^\infty} \newcommand{\PP}{\mathbb P} \newcommand{\projorth}[1]{\mathsf{P}^\perp_{#1}} \newcommand{\Psif}{\Psi_f} \newcommand{\pscal}[2]{\langle #1,#2\rangle} \newcommand{\pscal}[2]{\langle #1,#2\rangle} \newcommand{\psconv}{\stackrel{\PP-a.s.}{\rightarrow}} \newcommand{\qset}{\mathbb Q} \newcommand{\revcondtrans}[3]{q_{#1}(#2|#3)} \newcommand{\rmd}{\mathrm d} \newcommand{\rme}{\mathrm e} \newcommand{\rmi}{\mathrm i} \newcommand{\Rset}{\mathbb{R}} \newcommand{\rset}{\mathbb{R}} \newcommand{\rti}{\sigma} \newcommand{\section}[1]{==== #1 ====} \newcommand{\seq}[2]{\lrc{#1\eqsp: \eqsp #2}} \newcommand{\set}[2]{\lrc{#1\eqsp: \eqsp #2}} \newcommand{\sg}{\mathrm{sgn}} \newcommand{\supnorm}[1]{\left\|#1\right\|_{\infty}} \newcommand{\thv}{{\theta_\star}} \newcommand{\tmu}{ {\tilde{\mu}}} \newcommand{\Tset}{ {\mathsf{T}}} \newcommand{\Tsigma}{ {\mathcal{T}}} \newcommand{\ttheta}{{\tilde \theta}} \newcommand{\tv}[1]{\left\|#1\right\|_{\mathrm{TV}}} \newcommand{\unif}{\mathrm{Unif}} \newcommand{\weaklim}[1]{\stackrel{\mathcal{L}_{#1}}{\rightsquigarrow}} \newcommand{\Xset}{{\mathsf X}} \newcommand{\Xsigma}{\mathcal X} \newcommand{\Yset}{{\mathsf Y}} \newcommand{\Ysigma}{\mathcal Y} \newcommand{\Var}{\mathbb{V}\mathrm{ar}} \newcommand{\zset}{\mathbb{Z}} \newcommand{\Zset}{\mathsf{Z}} $$

2017/10/07 23:39 · douc

Introduction

In this post, I introduce the beautiful proof of Von-Neumann. To get the idea, we start with a very simple example that gathers all the different ingredients of this wonderful meal.

Assume that there exist two measures $\mu$ and $\nu$ on $(\Omega, \mcf)$ such that for all $A \in \Omega$, $\nu(A) \leq \mu(A)\leq \mu(\Omega)<\infty$. Then, $\nu(f) \leq \mu(f)$ for any nonnegative function $f$ and the Cauchy-Schwarz inequality yields $$ |\nu(f)| \leq [\nu(\Omega)]^{1/2} [\nu(f^2)]^{1/2} \leq [\nu(\Omega)]^{1/2} [\mu(f^2)]^{1/2} $$ Therefore, the linear mapping: $f \mapsto \nu(f)$ is continuous on $\Ltwo(\mu)$ and this allows to apply the Riez-Theorem: there exists $g \in \Ltwo(\mu)$ such that $\nu(f)=\mu(fg)$ for all $f\in \Ltwo(\mu)$. We can show that $\mu(g\notin[0,1])=0$ by combining the previous equality with $\nu(f) \leq \mu(f)$ and by taking the specific functions $f=\indi{g<0}$ or $f=\indi{g>1}$.

All the ingredients are here: the linear mapping $f \mapsto \nu(f)$ where $f$ in taken in a $\Ltwo$ space associated to another measure, the Radon-Nikodym theorem and some particular choices of $f$ to get properties on $g$. But since the assumption “for all $A \in \Omega$, $\nu(A) \leq \mu(A)\leq \mu(\Omega)<\infty$” is too restrictive, we will use $\nu(A) \leq \pi(A)$ where $\pi=\nu+\mu$, which is direct since $\mu$ is a (nonnegative) measure!!!!!

The Radon-Nikodym theorem

Theorem If $\mu$ and $\nu$ are two finite measures on $(\Omega,\mathcal{F})$ then there exist a nonnegative measurable function $g$ and a $\mu$-null set B such that $$ \nu(A)=\int_A g \, d\mu+ \nu(A \cap B) $$ for each $A \in \mathcal{F}$.

$\bproof$ Define $\pi:=\mu+\nu$ and consider the linear operator $T(f):=\nu(f)$. Note that $T$ is continuous on $L^2(\pi)$. Indeed, by the Cauchy-Schwarz inequality, for all $f \in L^2(\pi)$, $$ |T(f)| \leq [\nu(\Omega)]^{1/2} \|f\|_{L^2(\nu)}\leq [\nu(\Omega)]^{1/2} \|f\|_{L^2(\pi)} $$ According to the Riesz Theorem, there exists a $h \in L^2(\pi)$ such that for all $f\in \Ltwo(\pi)$, $T(f)=\pi(f h)$ so that \begin{equation} \label{eq:fond} \forall f \in \Ltwo(\pi),\quad \nu(f)=\mu(f h)+\nu( f h)\eqsp. \end{equation} From this equation, we would like to deduce that $\nu(f(1-h))=\mu(fh)$ and then $\nu(f)=\mu(fh/(1-h))$ but we have to be rigorous since we don't know yet if $h$ takes values only between $0$ and $1$. Consider the following sets $$ N \eqdef \{h <0\}, \quad M\eqdef \{0 \leq h < 1\}, \quad B\eqdef \{h \geq 1\}. $$ Let $A \in \mcf$. Define $M_n:=\{0 \leq h \leq 1-1/n\}$ and note that $f\eqdef\frac{\indi{A} \indi{M_n}}{1-h} \in \Ltwo(\pi)$ since this function is bounded by $n$. Since \eqref{eq:fond} can be rewritten as $ \nu(f(1-h))=\mu(fh)$, the monotone convergence then yields $$ \nu(A \cap M)=\lim_{n \to \infty} \nu\lr{\indi{A \cap M_n}}=\lim_{n \to \infty} \nu\lr{\frac{\indi{A\cap M_n}}{1-h} (1-h)}=\lim_{n \to \infty} \mu\lr{\frac{\indi{A\cap M_n}}{1-h} h}=\mu\lr{\indi{A} \underbrace{\frac{\indi{M}h}{1-h}}_{g}}=\mu(\indi{A}g) $$ Plugging $f=\indi{N}$ into \eqref{eq:fond} implies $0\leq \nu(N)=\mu(\indi{N} h)+\nu(\indi{N} h) \leq 0$ so that $\nu(N)=0$ and $$ \nu(A)=\underbrace{\nu(A \cap N)}_{0} + \nu(A \cap M) + \nu(A \cap B) = \int_A g \, d\mu + \nu(A \cap B) $$ To complete the proof, it remains to show that $B$ is a $\mu$-null set, that is $\mu(B)=0$. But plugging $f=\indi{B}$ into \eqref{eq:fond}, we get $0 \geq \int_B (1-h) d\nu=\mu(\indi{B} h)\geq \mu(\indi{B}) \geq 0$ so that $\mu(B)=0$. $\eproof$

From this theorem, we can show that if $\mu$ and $\nu$ are $\sigma$-finite measures on $(\Omega,\mcf)$, then there exist a nonnegative measurable function $g$ and a $\mu$-null set B such that $$ \nu(A)=\int_A g \, d\mu+ \nu(A \cap B) $$ for each $A \in \mathcal{F}$. As a direct consequence, if $\mu$ is $\sigma$-finite and if $\nu \preceq \mu$ in the sense that $\mu$-null sets are $\nu$-null sets, then there exists a nonnegative, measurable function $g$ such that $\rmd \nu= g \ \rmd \mu$ (which is an “infinitesimal” notation for saying that $\nu= g \cdot \mu$).

An easy exercise on Radon Nikodym derivatives

Statement

Let $\mu,\nu$ be two measures on $(\Xset,\Xsigma)$. Assume that $\nu \preceq \mu$ and that $\frac{\rmd \nu}{\rmd \mu}(x)>0$, for $\mu$-almost all $x\in \Xset$. Show that $\mu \preceq \nu$ and $\frac{\rmd \mu}{\rmd \nu}(x)=\lr{\frac{\rmd \nu}{\rmd \mu}(x)}^{-1}$ , for $\nu$-almost all $x\in \Xset$.

Proof

By assumption, for any non-negative measurable function $f$ on $\Xset$,

\begin{equation} \label{eq:rel} \nu(f)=\int_\Xset \mu(\rmd x) \frac{\rmd \nu}{\rmd \mu}(x) f(x) \end{equation}

Taking $f=\indi{\Xset}$, we get $\mu(\frac{\rmd \nu}{\rmd \mu})=\nu(\Xset)=1<\infty$ and therefore $\mu(\frac{\rmd \nu}{\rmd \mu}<\infty)=1$. Since by assumption $\mu(\frac{\rmd \nu}{\rmd \mu}=0)=1$, we obtain $\mu(A)=1$ where $$ A=\set{x \in \Xset}{\frac{\rmd \nu}{\rmd \mu}(x)\in (0,\infty)} $$ This, combined with \eqref{eq:rel}, in turn implies $$ \nu(A)=\int_\Xset \mu(\rmd x) \frac{\rmd \nu}{\rmd \mu}(x) \indi{A}(x)=\int_\Xset \mu(\rmd x) \frac{\rmd \nu}{\rmd \mu}(x)= \nu(\Xset)=1 $$ For all $B\in \Tsigma$, using successively $\mu(A)=1$, Eq. \eqref{eq:rel} and $\nu(A)=1$, \begin{align*} \mu(B)&=\mu(B \cap A)=\int_\Xset \mu(\rmd x)\lrb{ \frac{\rmd \nu}{\rmd \mu}(x) \times \lr{\frac{\rmd \nu}{\rmd \mu}(x)}^{-1}}\indi{B}(x)\indi{A}(x)\\ &=\int_\Xset \nu(\rmd x) \lr{\frac{\rmd \nu}{\rmd \mu}(x)}^{-1} \indi{B}(x)\indi{A}(x)= \int_\Xset \nu(\rmd x) \lr{\frac{\rmd \nu}{\rmd \mu}(x)}^{-1} \indi{B}(x) \end{align*} This completes the proof.

world/radon.txt · Last modified: 2022/03/16 07:40 (external edit)