# Probability-measure theory question

• Sep 11th 2010, 07:00 AM
artnoage
Probability-measure theory question
Hello http://www.mymathforum.com/images/sm...icon_smile.gif.

Does anyone know any way that you can characterize the set (or any nontrivial subset) of the probability measures in R^{2} with fixed marginals?

• Sep 11th 2010, 12:14 PM
Moo
If the marginals have the normal (gaussian) density wrt Lebesgue's measure, it's easy since we know the pdf of a bivariate gaussian distribution :D

I don't think it's easy to characterize, there is already such a wide range of possible marginal measures...
• Sep 11th 2010, 12:27 PM
artnoage
You cant imagine how much helpful that was. It would be great if i could have some informations in the more general case but still this is enough for what i am doing at the moment. Kolmogorov Bless u :).
• Sep 11th 2010, 12:32 PM
Moo
Huh ?
How did I help you exactly ? I just said that since we know the pdf (hence the probability measure) of (X,Y), a Gaussian vector (not just the joint distribution), we know the marginals of X and Y and can characterize the set of probability measures such that this stuff is verified.
But it is just a tiny point in an ocean of measures...

What exactly were you looking for ? I mean why did you ask such a question ?
• Sep 11th 2010, 12:44 PM
artnoage
It is quite complicated to explain what i am doing. But my first toy model has Gaussians as marginals, and i can test there if my assumption is true or not. (if my assumption doesnt even work with gaussians then it has no meaning working with something else). So i thought that what you said, is that when my marginals are Gaussian then always the joint measure is a bivariate Gaussian dist. But probably you meant the other way around :(.
• Sep 11th 2010, 12:54 PM
Moo
Yes, I'm afraid I was completely wrong in my first post... which I wrote partly joking because I was pretty sure it couldn't help you at all... I'm sincerely sorry :(
• Sep 11th 2010, 12:59 PM
artnoage
Np. You still gave me some ideas though. Thanks for replying anyway :)