0
$\begingroup$

Consider a probability space $(\Omega, \mathcal{F}, P)$, as well as two sub-$\sigma$-fields $\mathcal{A}$ and $\mathcal{B}$. The Hausdorff pseudo-distance between $\mathcal{A}$ and $\mathcal{B}$ is defined as follows: $$D(\mathcal{A},\mathcal{B}):=\max\{\sup_{A\in \mathcal{A}}\inf_{B\in\mathcal{B}}P(A\Delta B) ; \sup_{B\in \mathcal{B}}\inf_{A\in\mathcal{A}}P(A\Delta B)\}, $$ where $\Delta$ stands for the usual symmetric difference between sets, i.e.: $A\Delta B = (A-B\cap A)\cup (B-B\cap A)$. Now suppose that $F, N$ are two independent random vectors of some common dimension $d\geq 1$ (we may even assume that $N$ is Gaussian and that $F$ has a density), and also that $\mathcal{A} = \sigma(F)$ and $\mathcal{B} = \mathcal{B}_{\epsilon} = \sigma(F+\epsilon N)$. Is there some known method in order to obtain estimates on the quantity $D(\mathcal{A},\mathcal{B}_\epsilon)$, as $\epsilon $ converges to zero? I am expecting that such a quantity converges to zero, and I would like to have an upper bound on this convergence (the faster the better!).

Relevant references on the topic (among many others) are a paper by Rogge

<Link>

and one by van Zandt

<Link>

$\endgroup$
3
  • $\begingroup$ If $F$ degenerate and $N$ nondegenerate there is no convergence, since $\sigma(F)$ is trivial and $\sigma(F + \epsilon N) = \sigma(N)$ is not. $\endgroup$
    – Dan
    Feb 19, 2013 at 13:25
  • $\begingroup$ Dan: this is a pertinent remark (even if I specified that $F$ has a density)! Anyway, I have really a poor understanding of the Hausdorff distance, whenever it is applied to arbitrary $\sigma$-fields; moreover, it seems to me (hopefully, I am wrong) that there are few examples in the literature with some explicit estimates. It seems to me that the most explicit results are in a martingale framework. I wonder whether one can build a counterexample (that is, an example where one has not convergence to zero) involving a random variable $F$ which is not trivial, say with a density. $\endgroup$
    – user31542
    Feb 20, 2013 at 10:17
  • $\begingroup$ This paper, for instance, projecteuclid.org/DPubS/Repository/1.0/… has a lot of explicit computations $\endgroup$
    – user31542
    Feb 20, 2013 at 10:18

0

Your Answer

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

Browse other questions tagged or ask your own question.