10
$\begingroup$

Let $(S,\mathcal F)$ be a measurable space, and let $\nu \in\mathcal P(S,\mathcal F)$ be a probability measure on $(S,\mathcal F)$. Fix some $x\in S$ and consider Dirac measure $\delta_x$. Would like to prove

If $\mu \in \mathcal P(S×S,\mathcal F\otimes \mathcal F)$ and has marginals $ν$ and $δ_x$ $then$ $μ=ν×δ_x$

So we should show $μ(A×B)=ν(A)δ_x(B)$ for $∀A,B∈\mathcal F$.

If $x∉B$ then right-hand side is $0$ but so is the left-hand side since $μ(A×B)≤μ(S×B)=δ_x(B)=0$.

How to deal with the case $x∈B$ ?

$\endgroup$

2 Answers 2

8
$\begingroup$

If $x\in B$, then for each $A\in\mathcal F$, $$\nu(A)=\mu(A\times S)\geqslant \mu(A\times B)\geqslant \mu(A\times \{x\}).$$ Since all the involved measures are probability measures, applying the previous inequality with $S\setminus A$ gives that $$ 1-\nu\left(A\right)\geqslant \mu\left(S\times \{x\}\right)-\mu\left(A\times \{x\}\right), $$ and since the marginal in the second coordinate is $\delta_x$, we got that $$ -\nu\left(A\right)\geqslant-\mu\left(A\times \{x\}\right) $$ hence $\nu\left(A\right)\leqslant \mu\left(A\times \{x\}\right)\leqslant \mu\left(A\times B\right)$ which proves that $\mu\left(A\times B\right)=\nu\left(A\right)\delta_x\left(B\right)$. This identity is also true for $x\notin B$, and it can be extended to disjoint finite unions of cartesian products of elements of $\mathcal F$, which is sufficient to ensure the equality of two measures.

$\endgroup$
4
  • $\begingroup$ Could you explain the last conclusion in more detail. Im not sure how it works. You claim it follows if we apply the inequality with A = S\A ? $\endgroup$
    – White
    Jan 16, 2019 at 14:08
  • 1
    $\begingroup$ @White I have edited in order to include more details. $\endgroup$ Jan 16, 2019 at 16:07
  • $\begingroup$ Thank you. I think you don't need that last sentence. The product measure is the unique measure satisfying $\nu \otimes \delta_x(A,B) = \nu(A) \delta_x(B)$ and this was proved for $\mu.$ $\endgroup$
    – White
    Jan 16, 2019 at 16:35
  • $\begingroup$ @White Indeed, it seems that the opening poster knows that this is a property of product measure. $\endgroup$ Jan 16, 2019 at 16:36
0
$\begingroup$

For any measurable set $B\subset S$, $\mu(S\times B)=\delta_x(B)=\mathbb{1}_B(x)$. In particular, $\mu(S\times\{x\})=1$, and if $x\notin B$, $$\mu(A\times B)=0=\nu(A)\delta_x(B),\qquad A\in\mathcal{F}$$ for $\mu(A\times B)\leq\mu(A\times(S\setminus\{x\})=0$.

Suppose now that $x\in B$. Then, $\delta_x(B)=1$, and $\mu(A\times(S\setminus B))=0$. Consequently, $$\begin{align} \nu(A)\delta_x(B)=\nu(A)&=\mu(A\times S)=\mu((A\times B) \cup (A\times(S\setminus B))\\ &=\mu(A\times B)+\mu(A\times(S\setminus B))=\mu(A\times B) \end{align}$$

Putting things together, we have that $$\mu(A\times B)=\nu(A)\times\delta_x(B),\qquad A,B\in\mathcal{F}$$ From this (why?) it follows that $\mu=\nu\otimes\delta_x$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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