I have been struggling quite a bit with reconciling my intuitive understanding of probability distributions with the weird properties that almost all topologies on probability distributions possess.

For example, consider a mixture random variable Xn: pick a Gaussian centered at 0 with variance 1, and with probability 1n, add n to the result. A sequence of such random variables would converge (weakly and in total variation) to a Gaussian centered at 0 with variance 1, but the mean of the Xn is always 1 and the variances converge to +∞. I really don’t like saying that this sequence converges because of that.

I took me quite some time to remember everything I’ve forgotten about topologies, but I finally figured out what was so unsatisfying to me about such examples: the limit of the sequence is not a conventional distribution. In the example above, the limit is a weird “Gaussian of mean 1 and of infinite variance”. In topological terms, the set of probability distributions isn’t complete under the weak (and TV, and all the other topologies I’ve looked at).

I then face the following question:

does there exist a topology such that the ensemble of probability distributions is complete ?

If no, does that absence reflect an interesting property of the ensemble of probability distributions ? Or is it just boring ?

Note: I have phrased my question about “probability distributions”. These can’t be closed because they can converge to Diracs and stuff like that which don’t have a pdf. But measures still aren’t closed under the weak topology so my question remains

crossposted to mathoverflowhttps://mathoverflow.net/questions/226339/topologies-for-which-the-ensemble-of-probability-measures-is-complete?noredirect=1#comment558738_226339

**Answer**

Looking at the question from a more narrow statistical angle (the general mathematical topological issue is valid), the fact that the sequence of moments may not converge to the moments of the limiting distribution is a well-known phenomenon. This in principle, does not automatically set in doubt the existence of a well behaved limiting distribution of the sequence.

The limiting distribution of the above sequence {Xn+nBern(1/n)} is a well-behaved N(0,1) distribution with finite moments. It is the sequence of the moments that does not converge. But *this is a different sequenc*e, a sequence comprised of *functions* of our random variables (integrals, densities and such), not the sequence of the random variables themselves whose limiting distribution we are interested at.

**Attribution***Source : Link , Question Author : Guillaume Dehaene , Answer Author : Alecos Papadopoulos*