MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Hello all.

I have the following (perhaps basic) question: Let $X$ be a separable metric space. Does there necessarily exist a countable set $\mathcal{C}$ of Borel sets in $X$ such that any two probability measures which agree on $\mathcal{C}$ must agree on the whole of $\mathcal{B}(X)$?

(And slightly more generally: Let $(X,\Sigma)$ be a countably generated measurable space. Then does there necessarily exist a countable set $\mathcal{C}$ of $\Sigma$-measurable sets such that any two probability measures which agree on $\mathcal{C}$ must agree on the whole of $\Sigma$?)

Thanks, Julian.

share|cite|improve this question
up vote 1 down vote accepted

Yes. This can be proved using Dynkin's $\pi$-$\lambda$ theorem. The collection $\mathcal{L} := \{ B \in \Sigma : \mu(B) = \nu(B)\}$ is a $\lambda$-system. By Dynkin's theorem, if $\mathcal{L}$ contains a $\pi$-system which generates $\Sigma$ then $\mathcal{L} = \Sigma$, i.e. $\mu=\nu$. If $\mathcal{C}_0$ is countable and generates $\Sigma$, then the collection $\mathcal{C}$ of all finite intersections of sets from $\mathcal{C}_0$ is a countable $\pi$-system which generates $\Sigma$. So any two probability measures that agree on $\mathcal{C}$ must be equal.

share|cite|improve this answer
@Nate: Thank you very much. (The key step that I'd been missing was how to obtain a countable $\pi$-system that generates $\Sigma$.) When writing a paper, do you think I can assume (without proof or reference) the answer to my question? In other words, is the fact which I asked about sufficiently "standard" that I don't need to justify it? Would you want the paper to give acknowledgement of your help? – Julian Newman Apr 11 '13 at 20:23
@Julian: Well, I for one had to think about it for a couple of minutes. If it were my paper, I'd probably at least put in a few words to indicate to the reader how it goes. If you'd like to acknowledge me, my "publication name" is Nathaniel Eldredge. Glad to help! – Nate Eldredge Apr 12 '13 at 1:00
"Dynkin's theorem $\pi$-$\lambda$-theorem" is, possibly up to a minor variation, due Sierpi\'nski from about 1927. I don't have the precise reference(s) at hand, but see Bogachev's (2007) Measure Theory 1, p. 419, or Elstrodt's (2009) Ma\ss- und Integrationstheorie, 6. Auflage, p. 25. So, calling the result "Dynkin $\pi$-$\lambda$ theorem", honouring a popularizer of it, might be defensible, but writing here "Dynkin's" seems wrong to me. – Lutz Mattner Apr 28 '13 at 8:55

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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