5.6 Lebesgue-Stieltjes Measure

Definition 5.6.1 (Stieltjes Function).label Let $F:\R\to\R$, then $F$ is a Stieltjes function if $F$ is right-continuous and non-decreasing. Denote by $\fraks{S}=\fraks{S}(\R,\R)=\curl{F:\R\to\R:F\text{ is Stieltjes}}$ the space of Stieltjes functions and fix $x\in\R$. Define $\fraks{S}_{x}=\curl{F\in\fraks{S}:F(a)=0}$ the space of normalized Stieltjes functions. Let $\sim$ be the equivalence relation on $\fraks{S}$ given by

\begin{align*}\sim:\fraks{S}\ti\fraks{S}\to\fraks{S}&&F\sim G\iff F-G\text{ is constant}\end{align*}

Let $\leq$ be the partial order on $\fraks{S}$ given by

\begin{align*}F\leq G\iff G-F\text{ is non-decreasing}\end{align*}

Then there is an ordered convex cone isomorphism $\fraks{S}_{x}\iso \fraks{S}/\sim$.

Proof. $\frak{S}$ has an action of $(\R,+)$ by translation $c*F=F+c$, for $F\in\fraks{S},x\in\R$. For any $F\in\fraks{S},x\in\R$ we have $\til{F}\define F-F(x)\in\fraks{S}_{x}$ so $\R F\cap\fraks{S}_{x}\neq \emp$. Let $G,H\in \R F\cap\fraks{S}_{x}$ then $G=H+c$ for some $a\in\R$ so

\begin{align*}0=G(x)=H(x)+c=0+c\end{align*}

so $G=H$. By Lemma 5.6.2 $\fraks{S}_{x}\iso \fraks{S}/\sim$ as sets. Since $G-F+c$ is non-decreasing if and only if $G-F$ is non-decreasing for any $c\in\R$ we have by Lemma 5.6.2(1) $\fraks{S}_{x}\iso \fraks{S}/\sim$ as posets.

Let $F,G\in\fraks{S}$, $\al,\be\geq 0$ then

\begin{align*}\al F+\be G\in \fraks{S}\end{align*}

so $\fraks{S}$ is convex cone. Similarly so is $\fraks{S}_{x}$[1]. By Lemma 5.6.2(2) we have an ordered convex cone isomorphism.$\square$

Lemma 5.6.2.label Let $G$ be a group acting on a set $X$. Suppose $Y\suf X$ satisfies $\abs{[x]\cap Y}=1$ for each $x\in X$ then $X/G\iso Y$. Addtionally

  1. (1)

    If $X$ has a partial order $\leq_{X}$ such that for any $x_{1}\sim x_{1}',x_{2}\sim x_{2}'\in X$ we have

    \begin{align*}x_{1}\leq x_{2}\iff x_{1}'\leq x_{2}'\end{align*}

    then $X/G\iso Y$ as posets.

  2. (2)

    If $X,Y$ is a convex cone and for any $x_{1}\sim x_{1}',x_{2}\sim x_{2}'\in X,\lam\geq0$ we have

    \begin{align*}x_{1}+x_{2}\sim x_{1}'+x_{2}'&&\lam x_{1}\sim \lam x_{1}'\end{align*}

    then $X/G\iso Y$ as convex cones.

Proof. Consider the map $\p:Y\to X/G$ defined by $y\mto [y]$. Let $y_{1},y_{2}\in Y$ then $\p(y_{1})=\p(y_{2})\iff [y_{1}]=[y_{2}]\iff y_{1}=y_{2}$ since $\curl{y_1,y_2}\suf [y_{1}]\cap Y$. Let $E\in X/G$ then $E=[x]$ for some $x\in X$. Since $\abs{[x]\cap Y}=1$ there exists $y\in Y$ such that $y\in [x]$ so $[y]=[x]=\p(x)$.

  1. (1)

    By hypothesis $[x_{1}]\leq_{X/G}[x_{2}]\iff x_{1}\leq_{X} x_{2}$ is a well-defined partial order on $X/G$. The restriction of $\leq_{Y}\define \leq_{X}|_{Y}$ to $Y$ is a partial order on $Y$. Suppose $y_{1},y_{2}\in Y$ then

    \begin{align*}y_{1}\leq_{Y} y_{2}\iff y_{1}\leq_{X} y_{2}\iff \braks{y_1}\leq_{X/G}\braks{y_2}\iff\p(y_{1})\leq_{X/G}\p(y_{2})\end{align*}

  2. (2)

    By hypothesis $X/G$ is a convex cone with

    \begin{align*}\braks{x_1}+\braks{x_2}=\braks{x_1+x_2}&&\lam\braks{x}=\braks{\lam x_1}\end{align*}

    Suppose $y_{1},y_{2}\in Y,\lam\geq 0$ then

    \begin{align*}\p(y_{1}+y_{2})=\braks{y_1+y_2}=\braks{y_1}+\braks{y_2}=\p(y_{1})+\p(y_{2})\\ \p(\lam y_{1})=\braks{\lam y_1}=\lam \braks{y_1}=\lam\p(y_{1})\end{align*}

$\square$

Definition 5.6.3 (h-interval).label A h-interval is an interval of the form $(a,b]\suf\R$ where $-\infty<a\leq b<\infty$.

Lemma 5.6.4.label Let $\cali{I}=\curl{(a,b]:-\infty<a\leq b<\infty}$ then

  1. (1)

    $\cali{I}$ is an elementary family.

  2. (2)

    The collection

    \begin{align*}\cali{A}=\curl{\dcups{n}{j=1}I_j:\curl{I_j}_1^n\suf\cali{I},n\in\N^+}\end{align*}

is a ring.

Proof. (1): Let $(a,b],(c,d]\in\cali{I}$ and assume without loss of generality that $a<b$ and $c<d$ then

\begin{align*}(a,b]\cap(c,d]&=(\max\curl{a,c},\min\curl{b,d}]\in\cali{I}\\ (a,b]\del(c,d]&=(a,\min\curl{b,c}]\sqcup(\max\curl{a,d},b]\end{align*}

(2): By Proposition 4.4.2, $\cali{A}$ is a ring.$\square$

Definition 5.6.5 (Lebesgue-Stieltjes Measure).label Let $\fraks{S}_{x}=\fraks{S}_{x}(\R,\R)$ and $\cali{M}_{lf}=\cali{M}_{lf}(\R)$ be as in Definition 5.6.1 and Definition 5.1.6 respectively. Let $\cali{A}$ be the ring from Lemma 5.6.4 then $\cali{B}_{\R}=\s(\cali{A})$. There exists a unique Borel measure $\mu_{F}\in\cali{M}_{lf}$ satisfying

\begin{align*}\mu_{F}((a,b])=F(b)-F(a)&&\fall-\infty<a<b<\infty\end{align*}

Moreover for any fixed $x\in\R$, this gives an ordered convex cone isomorphism

\begin{align*}\p:\fraks{S}_{x}&\bij \cali{M}_{lf}\\ F&\mto\mu_{F}\end{align*}

whose inverse is

\begin{align*}\mu\mto F_{\mu}&&F_{\mu}(t)=\begin{cases}\mu((x,t])&t\geq x\\ -\mu((t,x])&t\leq x\end{cases}\end{align*}

The measure $\mu_{F}$ is called the Lebesgue-Stieltjes measure associated with $F$.

Proof. By Definition 5.6.1 it is equivalent to consider $\fraks{S}_{x}$ over $\fraks{S}/\sim$. By Proposition 4.3.4, $\cali{B}_{\R}=\s(\cali{I})=\s(\cali{A})$. We define

\begin{align*}\mu_{0}:\cali{A}\to [0,\infty)&&\dcups{n}{j=1}(a_{j},b_{j}]\mto \sums{n}{j=1}\braks{F(b_j)-F(a_j)}\end{align*}

Let $(a,b]\in\cali{I}$ with $(a,b]=\dcups{n}{j=1}(a_{j},b_{j}]$. Assume without loss of generality that $\curl{a_j}_{1}^{n}$ is non-decreasing, then $b_{j}=a_{j+1}$ for each $1\leq j\leq n-1,b=b_{n}$ and $a=a_{1}$. Thus

\begin{align*}F(b)-F(a)=F(b_{n})+\sums{n-1}{j=1}\braks{F(b_j)-F(a_{j+1})}-F(a_{1})=\sums{n}{j=1}\braks{F(b_j)-F(a_j)}\end{align*}

so $\mu_{0}$ is well-defined and finitely-additive on $\cali{A}$. Suppose $(a,b]=\dcups{}{n\in\N}(a_{n},b_{n}]\in\cali{A}$ for $\curl{(a_n,b_n]}_{n\in\N}\suf\cali{I}$. By finite additivity, for each $N\in\N$

\begin{align*}\sums{}{n\leq N}\mu_{0}((a_{n},b_{n}])\leq\mu_{0}((a,b])\end{align*}

thus

\begin{align*}\sums{}{n\in\N}\mu_{0}((a_{n},b_{n}])\leq\mu_{0}((a,b])\end{align*}

Let $\e>0$. By right-continuity of $F$, pick $\de,\de_{n}>0$ for each $n\in\N$ such that

\begin{align*}F(a+\de)-F(a)<\frac{\e}{2}&&F(b_{n}+\de_{n})-F(b_{n})<\frac{\e}{2^{n+1}}\end{align*}

The collection $\curl{(a_n,b_n+\de_n)}_{n\in\N}$ forms an open cover for $[a+\de,b]$ so by compactness tehre exists $N\in\N$ such that

\begin{align*}[a+\de,b]\suf \cups{}{n\leq N}(a_{n},b_{n}+\de)\suf \cups{}{n\leq N}(a_{n},b_{n}+\de_{n}]\end{align*}

Thus

\begin{align*}\mu_{0}((a,b])&\leq F(a+\de)-F(a)+\mu_{0}([a+\de,b])\\&\leq \frac{\e}{2}+\sums{}{n\leq N}\mu_{0}((a_{n},b_{n}+\de_{n}])\\&\leq \frac{\e}{2}+\sums{}{n\leq N}\mu_{0}((a_{n},b_{n}])+\sums{}{n\leq N}\braks{F(b_n+\de_n)-F(b_n)}\\&\leq \e+\sums{}{n\leq N}\mu_{0}((a_{n},b_{n}])\end{align*}

Since $\e$ was arbitrary

\begin{align*}\mu_{0}((a,b])\leq \sums{}{n\leq N}\mu_{0}((a_{n},b_{n}])\leq \sums{}{n\in\N}\mu_{0}((a_{n},b_{n}])\end{align*}

Thus $\mu_{0}$ is a premeasure onn $\cali{A}$. Since $\R=\cups{}{k\in\Z}(k,k+1]$ and $\mu_{0}((k,k+1])=F(k+1)-F(k)<\infty$ the premeasure is $\s$-finite. By Carathéodory Extension Theorem, there exists a unique Borel measure $\mu_{F}$ on $\s(\cali{A})=\cali{B}_{\R}$ extending $\mu_{0}$. Let $K\suf\R$ be compact then $K\suf [a,b]$ for some $-\infty<a<b<\infty$ then by monotonicity of measure $\mu_{F}(K)<\infty$.

If $F\sim G$ then for any $-\infty<a<b<\infty$

\begin{align*}\mu_{F}((a,b])=F(b)-F(a)=G(b)-G(a)=\mu_{G}((a,b])\end{align*}

Since $\mu_{F}$ and $\mu_{G}$ are Borel measures and $\cali{B}_{\R}=\s(\cali{A})$, we have $\mu_{F}=\mu_{G}$ so the map is well-defined. Suppose $\mu_{F}=\mu_{G}$ then for all $a<b\in\R$

\begin{align*}F(b)-F(a)=G(b)-G(a)\end{align*}

so $F-G$ is constant. Since $F(x)=G(x)=0$, the constant is $0$ so $F=G$. Therefore the map is injective. Let $\mu\in\cali{M}_{lf}$ then $F_{\mu}\in\fraks{S}$ by monotonicity and continuity from above. Observe $F_{\mu}(x)=0$ so $F_{\mu}\in\fraks{S}_{x}$. For $-\infty<a<b<\infty$

\begin{align*}\mu((a,b])=\begin{cases}\mu((0,b])-\mu((0,a])&a,b\geq0\\ \mu((a,0])-\mu((0,b])&a\leq 0,b\geq0\\ \mu((a,0])-\mu((b,0])&a,b\leq0\end{cases}\end{align*}

In all three cases, $\mu((a,b])=F(b)-F(a)$. By definition of $\mu_{F_\mu}$, $\mu_{F_\mu}|_{\cali{A}}=\mu|_{\cali{A}}$ so $\mu_{F_\mu}=\mu$. Therefore the map is surjective.

Suppose $F,G\in\fraks{S}_{x},\lam\geq 0$ then

\begin{align*}\p(\lam F+G)((a,b])&=\mu_{\lam F+G}((a,b])\\&=(\lam F+G)(b)-(\lam F+G)(a)\\&=\lam (F(b)-F(a))+ (G(b)-G(a))\\&=\lam\mu_{F}((a,b])+\mu_{G}((a,b])\\&=\lam \p(F)((a,b])+\p(G)((a,b])\end{align*}

Suppose $F\leq G$ then $G-F$ is non-decreasing so $G(a)-F(a)\leq G(b)-F(b)$ thus

\begin{align*}\p(F)((a,b])=\mu_{F}((a,b])=F(b)-F(a)\leq G(b)-G(a)=\mu_{G}((a,b])=\p(G)((a,b])\end{align*}

$\square$

  1. The quotient inherits a convex cone structure because $\sim$ is quotient by constant functions which is a linear subspacekeyboard_return

Post a Comment

Name:Email:
Please enter the tag of the current page (E9) to post the comment.
Tag: