English

On Low-Dimensional Projections of High-Dimensional Distributions

Statistics Theory 2013-11-26 v3 Statistics Theory

Abstract

Let PP be a probability distribution on qq-dimensional space. The so-called Diaconis-Freedman effect means that for a fixed dimension d<<qd << q, most dd-dimensional projections of PP look like a scale mixture of spherically symmetric Gaussian distributions. The present paper provides necessary and sufficient conditions for this phenomenon in a suitable asymptotic framework with increasing dimension qq. It turns out, that the conditions formulated by Diaconis and Freedman (1984) are not only sufficient but necessary as well. Moreover, letting P^\hat{P} be the empirical distribution of nn independent random vectors with distribution PP, we investigate the behavior of the empirical process n(P^P)\sqrt{n}(\hat{P} - P) under random projections, conditional on P^\hat{P}.

Keywords

Cite

@article{arxiv.1107.0417,
  title  = {On Low-Dimensional Projections of High-Dimensional Distributions},
  author = {Lutz Duembgen and Perla Zerial},
  journal= {arXiv preprint arXiv:1107.0417},
  year   = {2013}
}