FIND ME ON

GitHub

LinkedIn

Extension Theorem

🌱

Theorem
Probability

Extension Theorem

Let Ω\Omega be the sample space, J\mathcal{J} be a semialgebra of subsets. Suppose P:J[0,1]\mathbb{P}:\mathcal{J}\to[0,1] such that 1. P()=0\mathbb{P}(\emptyset)=0, P(Ω)=1\mathbb{P}(\Omega)=1 2. Finite superadditivity: For A1,,AnJA_{1},\dots,A_{n}\in\mathcal{J} disjoint and i=1nAiJ\bigcup_{i=1}^{n}A_{i}\in\mathcal{J} then P(i=1nAi)i=1nP(Ai)\mathbb{P}\left( \bigcup_{i=1}^{n}A_{i} \right)\ge \sum_{i=1}^{n}\mathbb{P}(A_{i}) 3. Countable Monotonicity: For A,A1,,AnJA,A_{1},\dots,A_{n}\in\mathcal{J} s.t. Ai1AiA\subseteq \bigcup_{i\ge 1}A_{i} then P(A)i1P(Ai)\mathbb{P}(A)\le \sum_{i\ge 1}\mathbb{P}(A_{i})

Then, we can find a σ-algebra MJ\mathcal{M}\supseteq\mathcal{J} and a σ\sigma-additive Probability Measure P\mathbb{P}^{*} on M\mathcal{M} s.t. P(A)=P(A)AJ\mathbb{P}^{*}(A)=\mathbb{P}(A)\quad\forall A\in\mathcal{J}

\begin{proof} We first begin with some definitions and lemmas:

For (Ω,J)(\Omega,\mathcal{J}) we say the outer measure P\mathbb{P}^{*} on AΩA\subseteq\Omega is defined as follow: P(A)=inf{iP(Ai):A1,A2,,J,AiAi}\mathbb{P}^{*}(A)=\inf\left\{ \sum_{i}\mathbb{P}(A_{i}):A_{1},A_{2},\dots ,\in\mathcal{J}, A\subseteq \bigcup_{i}A_{i} \right\}

For any (Bi)i1Ω(B_{i})_{i\ge 1}\subseteq\Omega, we have that the outer measure is countably subadditive: P(iBi)iP(Bi)\mathbb{P}^{*}\left( \bigcup_{i}B_{i} \right)\le \sum_{i}\mathbb{P}^{*}(B_{i})

AΩA\subseteq\Omega is Carathéodory measurable if P(E)=P(EA)+P(EAc)EΩ\mathbb{P}^{*}(E)=\mathbb{P}^{*}(E\cap A)+\mathbb{P}^{*}(E\cap A^{c})\quad\forall E\subseteq\Omegaand M={AΩ:A Caratheˊodory measurable}\mathcal{M}=\{ A\subseteq\Omega:A \text{ Carathéodory measurable} \}is the set of Carathéodory measurable sets.

For A1,A2,MA_{1},A_{2},\dots \in\mathcal{M} disjoint we have that the outer measure is countably additive: P(iAi)=iP(Ai)\mathbb{P}^{*}\left( \bigcup_{i}A_{i} \right)=\sum_{i}\mathbb{P}^{*}(A_{i})

Step 1: prove For A1M,A2ΩA_{1}\in\mathcal{M},A_{2}\subseteq\Omega, such that A1,A2A_{1},A_{2} are disjoint P(A1A2)=P(A1(A1A2))+P(A1c(A1A2))=P(A1)+P(A2)\begin{align*} \mathbb{P}^{*}(A_{1}\cap A_{2})&= \mathbb{P}^{*}(A_{1}\cap(A_{1}\cup A_{2}))+\mathbb{P}^{*}(A_{1}^{c}\cap(A_{1}\cup A_{2}))\\ &= \mathbb{P}^{*}(A_{1})+\mathbb{P}^{*}(A_{2}) \end{align*}Where the first equality is done by applying the definition of and the second by disjoint property of both events. By induction, for A1,,AnMA_{1},\dots,A_{n}\in\mathcal{M} disjoint (by and finite superadditivity) P(A1An)=i=1nP(Ai)\mathbb{P}^{*}(A_{1}\cup\dots \cup A_{n})=\sum_{i=1}^{n}\mathbb{P}^{*}(A_{i})Thus P(i1Ai)P(i=1nAi)=i=1nP(Ai)n1\mathbb{P}^{*}\left( \bigcup_{i\ge 1}A_{i} \right)\ge \mathbb{P}^{*}\left( \bigcup_{i=1}^{n}A_{i} \right)=\sum_{i=1}^{n}\mathbb{P}^{*}(A_{i})\quad\forall n\ge 1     P(iAi)i1P(Ai)\implies \mathbb{P}^{*}(\cup_{i}A_{i})\ge \sum_{i\ge 1} \mathbb{P}^{*}(A_{i})But, by we have that P(iAi)=i1P(Ai)\mathbb{P}^{*}(\cup_{i}A_{i})=\sum_{i\ge 1}\mathbb{P}^{*}(A_{i}) Step 2: Show M\mathcal{M} is a σ-algebra where M:={AΩ:P(E)=P(EA)+P(EAc)EΩ}\mathcal{M}:=\{ A\subseteq\Omega:\mathbb{P}^{*}(E)=\mathbb{P}^{*}(E\cap A)+\mathbb{P}^{*}(E\cap A^{c})\quad\forall E\subseteq\Omega \} a) Show M\mathcal{M} is a algebra 1. ,ΩM\emptyset,\Omega \in\mathcal{M} ✅ 2. AM    AcMA\in\mathcal{M}\implies A^{c}\in\mathcal{M} ✅ 3. Let A,BMA,B\in\mathcal{M}, EΩE\subseteq\Omega, then P((AB)E)+P((AB)cE)=P(ABE)+P((AcBE)(ABcE)(AcBcE))P(ABE)+P(AcBE)+P(ABcE)+Pc(AcBcE)=P(BE)+P(BcE)=P(E)\begin{align*} &\mathbb{P}^{*}((A\cap B)\cap E)+\mathbb{P}^{*}((A\cap B)^{c}\cap E)\\ &= \mathbb{P}^{*}(A\cap B\cap E)+\mathbb{P}^{*}((A^{c}\cap B\cap E)\cup(A\cap B^{c}\cap E)\cup(A^{c}\cap B^{c}\cap E))\\ &\le \mathbb{P}^{*}(A\cap B\cap E)+\mathbb{P}^{*}(A^{c}\cap B\cap E)+\mathbb{P}^{*}(A\cap B^{c}\cap E)+\mathbb{P}^{c}(A^{c}\cap B^{c}\cap E)\\ &= \mathbb{P}^{*}(B\cap E)+\mathbb{P}^{*}(B^{c}\cap E)\\ &= \mathbb{P}^{*}(E) \end{align*} where the last two arguments are because A,BMA,B\in\mathcal{M}. Since this is the the one side of the inequality we wanted to prove (the other is trivial) we have that ABMA\cap B\in\mathcal{M}.

b) Lemmas… We now state some additional lemmas: >[!lemma|2.3.11] >Let A1,A2,MA_{1},A_{2},\dots \in\mathcal{M} be disjoint. For each nNn\in\mathbb{N}, let Bn=i=1nAiB_{n}=\bigcup_{i=1}^{n}A_{i}. Then nN,EΩ\forall n\in\mathbb{N},\forall E\subseteq\Omega we have P(EBn)=i=1nP(EAi)\mathbb{P}^{*}(E\cap B_{n})=\sum_{i=1}^{n}\mathbb{P}^{*}(E\cap A_{i})

Let A1,A2,MA_{1},A_{2},\dots \in\mathcal{M} be disjoint. Then nNAnM\bigcup_{n\in\mathbb{N}}A_{n}\in\mathcal{M}.

c) M\mathcal{M} is a σ-algebra: >[!lemma|2.3.14] >M\mathcal{M} is a σ-algebra

\begin{proof} Using we just use the fact that for any A1,A2,MA_{1},A_{2},\dots \in\mathcal{M} we have that iAi=A1(A2A1)(A3(A1A2))disjointM\bigcup_{i}A_{i}=\underbrace{ A_{1}\cup(A_{2}\setminus A_{1})\cup(A_{3}\setminus(A_{1}\cup A_{2}))\cup\dots }_{ \text{disjoint} }\in\mathcal{M} \end{proof}

Step 3: JM\mathcal{J}\subseteq \mathcal{M}:

JM\mathcal{J}\subseteq \mathcal{M}

\begin{proof} Let AJA\in \mathcal{J}. Since J\mathcal{J} is a semialgebra we can write Ac=J1JkA^{c}=J_{1}\sqcup\dots\sqcup J_{k} for some disjoint J1,,JkJJ_{1},\dots,J_{k}\in \mathcal{J}. Also, for any EΩE\subseteq\Omega and ϵ>0\epsilon>0 by the definition of : we can find A1,A2,JA_{1},A_{2},\dots \in \mathcal{J} with EnAnE\subseteq \bigcup_{n}A_{n} and nP(An)P(E)+ϵ\sum_{n}\mathbb{P}(A_{n})\le\mathbb{P}^{*}(E)+\epsilon. Then

P(EA)+P(EAc)P((nAn)A)+P((nAn)Ac)monotonicity=P(n(AnA))+P(ni=1k(AnJi))definition of semialgebranP(AnA)+ni=1kP(AnJi)subadditivity=nP(AnA)+ni=1kP(AnJi)since PJ=P=n(P(AnA)+i=1kP(AnJi))nP(An)superadditivity then CaratheˊodoryP(E)+ϵby assumption.\begin{align*} &\mathbb{P}^{*}(E\cap A)+\mathbb{P}^{*}(E\cap A^{c})\\ &\le \mathbb{P}^{*}\left( \left( \bigcup_{n}A_{n} \right)\cap A \right)+\mathbb{P}^{*}\left( \left( \bigcup_{n}A_{n} \right)\cap A^{c} \right)&\text{monotonicity}\\ &= \mathbb{P}^{*}\left( \bigcup_{n}(A_{n}\cap A) \right)+\mathbb{P}^{*}\left( \bigcup_{n}\bigcup_{i=1}^{k}(A_{n}\cap J_{i}) \right)&\text{definition of semialgebra}\\ &\le \sum_{n}\mathbb{P}^{*}(A_{n}\cap A) + \sum_{n}\sum_{i=1}^{k}\mathbb{P}^{*}(A_{n}\cap J_{i})&\text{subadditivity}\\ &= \sum_{n}\mathbb{P}(A_{n}\cap A)+\sum_{n}\sum_{i=1}^{k}\mathbb{P}(A_{n}\cap J_{i})&\text{since }\mathbb{P}^{*}|_{\mathcal{J}}=\mathbb{P}\\ &= \sum_{n} \left(\mathbb{P}(A_{n}\cap A)+\sum_{i=1}^{k}\mathbb{P}(A_{n}\cap J_{i})\right)\\ &\le \sum_{n}\mathbb{P}(A_{n})&\text{superadditivity then Carathéodory}\\ &\le \mathbb{P}^{*}(E)+\epsilon&\text{by assumption}. \end{align*} This is true ϵ>0\forall\epsilon>0, hence since ϵ\epsilon is arbitrary, P(EA)+P(EAc)P(E),EΩ\mathbb{P}^{*}(E\cap A)+\mathbb{P}^{*}(E\cap A^{c})\le\mathbb{P}^{*}(E),\forall E\subseteq\Omega. Since the other direction is trivial we have that the condition holds giving us AMA\in \mathcal{M}, since this is for any AJA\in \mathcal{J} we have JM\mathcal{J}\subseteq \mathcal{M}. \end{proof} All these lemmas prove what we needed to prove ✅

\end{proof} # Extensions of the Extension Theorem

Let J\mathcal{J} be a semialgebra of subsets of Ω\Omega. Let P:J[0,1]\mathbb{P}:\mathcal{J}\to[0,1] such that: 1. P(Ω)=1\mathbb{P}(\Omega)=1, 2. : P(nIn)=nP(In) for I1,I2,J disjoint with nInJ.\mathbb{P}\left( \bigcup_{n}I_{n} \right)=\sum_{n}\mathbb{P}(I_{n})\text{ for }I_{1},I_{2},\dots \in \mathcal{J}\text{ disjoint with }\bigsqcup_{n}I_{n}\in \mathcal{J}.

Then there is a σ-algebra MJ\mathcal{M}\supseteq\mathcal{J}, and a countably additive probability measure P\mathbb{P}^{*} on M\mathcal{M}, such that: (Ω,M,P)(\Omega,\mathcal{M},\mathbb{P}^{*}) is a probability space and P(A)=P(A),AJ\mathbb{P}^{*}(A)=\mathbb{P}(A),\quad\forall A\in \mathcal{J}

Whereas requires finite superadditivity and countable monotonicity under disjoint sets this one requires countable additivity. The only things we need to prove in this case are Probability Measure and under non-disjoint sets.

Let Ω\Omega be the sample space, J\mathcal{J} a semialgebra of subsets of Ω\Omega, and P:J[0,1]\mathbb{P}:\mathcal{J}\to[0,1] such that (Ω,M,P)(\Omega,\mathcal{M},\mathbb{P}^{*}) is a probability space. If (Ω,F,Q)(\Omega,\mathcal{F},\mathbb{Q}) is a prob. space, P=Q\mathbb{P}=\mathbb{Q} on J\mathcal{J}, and JFM\mathcal{J}\subseteq \mathcal{F}\subseteq \mathcal{M} then Q(A)=P(A),AF\mathbb{Q}(A)=\mathbb{P}^{*}(A),\quad\forall A\in \mathcal{F}

Let P\mathbb{P} and Q\mathbb{Q} be two probability measures defined on the collection B\mathcal{B} of Borel subsets of R\mathbb{R}. Suppose P((,x])=Q((,x]),xR.\mathbb{P}((-\infty,x])=\mathbb{Q}((-\infty,x]),\quad\forall x\in \mathbb{R}.Then P(A)=Q(A),AB\mathbb{P}(A)=\mathbb{Q}(A),\quad\forall A\in \mathcal{B}

Linked from