13
$\begingroup$

Until recently, all my knowledge of measure theory and Lebesgue integration are from Rudin's book, which focuses solely on the Lebesgue measure, its construction and nothing else. I have just put my hands on a nice book "Measure and Integration Theory" by Heinz Bauer and I'm currently enjoying it. I have encountered the definition of a Dynkin system $\mathcal D$, which is a family of subsets of a set $\Omega$ satisfying

1.) $\Omega\in\mathcal D$.

2.) If $A\in\mathcal D$, then $A^c\in\mathcal D$.

3.) For $n\in\Bbb N$, if $A_n\in\mathcal D$ are pairwise disjoint then $\bigcup_{n=1}^{\infty}A_n\in\mathcal D$.

I have some idea about what a $\sigma$-algebra is, but not about a Dynkin system. I would really appreciate if someone could give me an intuition about Dynkin systems or what they're supposed to represent. What is the characteristics of a Dynkin system that let you recognize it once you see it?

I know the $\pi$-$\lambda$ theorem and facts like a Dynkin system $\mathcal D$ is a $\sigma$-algebra if it is closed under intersection, it would be also nice if anyone could explain to me why should we expect such a result. Thank you in advance.

$\endgroup$
  • $\begingroup$ I'd suggest looking at a probability theory book for details. Theres a bit on the topic in Durrett's probability text, for example. $\endgroup$ – Batman May 3 '16 at 2:42
20
$\begingroup$

A motivation for Dynkin systems (and specially, the $\pi$-$\lambda$ Theorem) is the following problem:

How many sets should I check to be sure that two probability measures $\mu$ and $\nu$ are the same?

For the sake of definiteness, both $\mu$ and $\nu$ are defined on a measurable space $(\Omega,\Sigma)$. A trivial answer to the previous question is “If they coincide over $\Sigma$, they are the same”. True, and useless. So let's think about this for a moment. If I have checked that $\mu(A) =\nu(A)$ for some $A\in\Sigma$, it is not necessary to check for the complement, since $$ \mu(A^c) =1-\mu(A) = 1-\nu(A) = \nu(A^c). $$ It is also immediate by $\sigma$-additivity that if they coincide on a sequence of pairwise disjoint sets $A_n$, they must coincide on their union. Hence we conclude

If two probability measures coincide on a family $\mathcal{A}\subseteq\Sigma$, then they coincide on the Dynkin system generated by $\mathcal{A}$ (i.e., the smallest $\lambda$-system containing it).

The above arguments can't be generalized to intersections; we can't compute $\mu(A\cap B)$ from $\mu(A)$ and $\mu(B)$. So, it would be desirable that our initial data (sets checked for coincidence) is a family of sets closed under binary intersections. Here, the $\pi$-$\lambda$ confirms this intuition: If $\Sigma$ is generated by a family $\mathcal{A}$, then it equals the smallest Dynkin system including the closure of $\mathcal{A}$ under intersections (i.e., the $\pi$-system generated by it).

Therefore, to check that $\mu$ and $\nu$ are equal, it is enough to take a family $\mathcal{A}$ such that of $\Sigma =\sigma(\mathcal{A})$ and check that the measures coincide on every finite intersection of members of $\mathcal{A}$.

$\endgroup$

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.