0
$\begingroup$

Let $\{X_j\}_{j=1}^{n}$ be a independent identically distributed random variables taking values in $\mathbb{R}^d$. We write $\mu$ for the distributions. We assume moreover that $\mu$ is absolutely continuous to the $d$-dim Lebesgue measure.

For $x \in \mathbb{R}^d$ and $r>0$, we denote by $B_x(r) \subset \mathbb{R}^d$ the closed ball centered at $x$ with radius $r>0$. Then For $x \in \mathbb{R}^d$ and $A \in \mathcal{F}$, we can consider $ P\left[A \mid X_1=x \right] $, the regular conditional probability given $X_1=x$:

My question. In a published paper, the authors conclude that \begin{align*} &P\left[\bigcap_{j=2}^{n-1}\{X_{j} \notin B(X_n,|x-X_n|)\} \mid X_1=x \right]=P\left[\bigcap_{j=2}^{n-1}X_{j} \notin B(X_n,|x-X_n|) \right]. \end{align*}

I do not know the reason why this identity is valid... Why is this true? The left-hand side should be interpreted as $P[A_x \mid X_1=x]$. Here, $A_x=\bigcap_{j=2}^{n-1}\{X_{j} \notin B(X_n,|x-X_n|)\} \in \mathcal{F}$ is a measurable set. From the definition of the regular conditional probability, can we obtain the following identity?: for any measurable $C$, \begin{align*} &\int_{C}P\left[\bigcap_{j=2}^{n-1}X_{j} \notin B(X_n,|x-X_n|) \mid X_1=x \right]\,d\mu(x) =E\left[\bigcap_{j=2}^{n-1}X_{j} \notin B(X_n,|X_1-X_n|) ,\,X_1 \in C \right]. \end{align*} I feel like it's impossible because the event $A_x$ is dependent on $x$.

$\endgroup$
0

2 Answers 2

1
$\begingroup$

$\newcommand\ov\overline\newcommand\R{\mathbb R}$This is just an application of Tonelli's theorem. Indeed, let $X:=X_1$, $Y:=(Y_2,\dots,Y_n)$, $Y_i:=X_i$ for $i\in\ov{2,n}$, where $\ov{k,l}:=[k,l]\cap\mathbb Z$. Let \begin{align*} A&:=\{(x,y):=(x,y_2,\dots,y_n)\in(\R^d)^n \colon\\ &\qquad\qquad\qquad\forall j\in\ov{2,n-1}\ y_j\notin B_{y_n}(|x-y_n|)\} \\ &:=\{(x,y)\in(\R^d)^n \colon\forall j\in\ov{2,n-1}\ |y_j-y_n|>|x-y_n|\}, \end{align*} so that $A$ is a Borel set. Let
\begin{equation*} f:=1_A. \end{equation*} Then for any Borel $C\subseteq\R^d$ \begin{align*} P(\{(X,Y)\in A\}\cap\{X\in C\})&=Ef(X,Y)1(X\in C\} \\ &=\int_{\R^d} P(X\in dx)\int_{(\R^d)^{n-1}} P(Y\in dy)f(x,y)1(x\in C) \\ &=\int_{\R^d} P(X\in dx)1(x\in C)\,Ef(x,Y) \\ &=\int_C P(X\in dx)\,P(Y\in A_x), \end{align*} where $A_x:=\{y\in(\R^d)^{n-1}\colon(x,y)\in A\}$; the second equality in the above display follows by Tonelli's theorem and the independence of $X$ and $Y$.

So, the regular conditional probability of the event $\{(X,Y)\in A\}$ given $X=x$ is $P(Y\in A_x)$, as desired.

$\endgroup$
1
$\begingroup$

More general: Let $(\Omega,\mathcal{A},\mathbb{P})$ a probability space and $X \colon \Omega \to E$ and $Y \colon \Omega \to F$ be independent random variables with values in Polish spaces $E$ and $F$ and $\mathcal{B}(E)$ be the Borel space of $E$, similarly for $F$. Then $\mathcal{B}(E \times F) = \mathcal{B}(E) \otimes \mathcal{B}(F)$. Let $A \in \mathcal{B}(E) \otimes \mathcal{B}(F)$ be arbitrary. Then for all $B \in \mathcal{B}(E)$ first by the theorem of Fubini and second by the definition of conditional expectation

$$\mathbb{P}((X,Y) \in A \cap B \times F) = \mathbb{P}^X \otimes \mathbb{P}^Y(A \cap B \times F) = \int_B \mathbb{P}^X(dx) \mathbb{P}^Y(A_x) = \int \mathbb{P}_B^X(dx) \mathbb{P}(A_x|X=x).$$ It follows that $\mathbb{P}^Y(A_x) = \mathbb{P}(A_x|X=x)$ for $\mathbb{P}^X$-a.e. $x \in E$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.