0
$\begingroup$

I have a distance on the space of probability measures on $[0,2]$. It is defined as such for two probability measures $\mu_1$ and $\mu_2$ :

$d_p(\mu_1,\mu_2) := \sum_{k=0}^p ( \mathbb{E}[X_1 ^k]- \mathbb{E}[X_1 ^k] ) ^2$ where $X_1$ and $X_2$ are random variables whose laws are $\mu_1$ and $\mu_2$ respectively.

Let $d_{TV}$ be the total variation distance on the same space. I'm considering distributions lying in a set $\Sigma$ with :

$\Sigma = \{ \alpha U(s,t) + (1-\alpha)W_1(R) \lvert (s,t,R,\alpha) \in [0,\epsilon_1]^2 \times [0,\epsilon_2] \times [0,1] \} $

where :

$ \epsilon_1, \epsilon_2 >0 $, $U(s,t)$ is the uniform distribution on $[s,t]$ and $W_1(R)$ is the semi-circular law of parameter $R>0$ and centered on 1.

I want to know if there exists a constant $C>0$ (possibly for $p$ big enough) such that

$\forall (\mu_1,\mu_2) \in \Sigma^2, \ d_{TV}(\mu_1,\mu_2) \leqslant C d_p(\mu_1,\mu_2) $.

Is is possible to possible to find such a constant when $\mu_1$ and $\mu_2$ are both uniform distributions for instance but I do not know if it's true in the general case. Do you have an idea or reference for this problem? Any suggestion is welcome!

Thanks.

$\endgroup$
  • 1
    $\begingroup$ This is hopeless. For any $p$, One can find pairs of probability measures where any the first $p$ moments agree. $\endgroup$ – Anthony Quas Jan 29 at 21:59
  • $\begingroup$ Yes, for general probability measures, this is impossible but here I am studying the possibility of such a control on some subset $\Sigma$ of the probability space. $\endgroup$ – YZ22 Jan 30 at 0:03
  • $\begingroup$ Still hopeless unless your subset is basically finite dimensional. $\endgroup$ – Anthony Quas Jan 30 at 3:20
  • $\begingroup$ I don't understand Anthony's objection. Doesn't the question state that the subset is finite dimensional? $\endgroup$ – Steve Jan 30 at 8:57
  • $\begingroup$ Yes indeed Steve, the subset is not finite dimensional per se but is parametrized by a small number of parameters. I'm going to update its definition in order to be clearer. $\endgroup$ – YZ22 Jan 30 at 9:39

Your Answer

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

Browse other questions tagged or ask your own question.