Universal $L^s$-rate-optimality of $L^r$-optimal quantizers by dilatation and contraction

Sagna, Abass

ESAIM: Probability and Statistics, Tome 13 (2009), pp. 218-246

Résumé

We investigate in this paper the properties of some dilatations or contractions of a sequence ${\left({\alpha }_n\right)}_{n\ge 1}$ of $L^r$-optimal quantizers of an ${ℝ}^d$-valued random vector $X\in L^r\left(\mathbb{P}\right)$ defined in the probability space $(\Omega ,𝒜,\mathbb{P})$ with distribution ${\mathbb{P}}_X=P$. To be precise, we investigate the $L^s$-quantization rate of sequences ${\alpha }_n^{\theta ,\mu }=\mu +\theta ({\alpha }_n-\mu )=\{\mu +\theta (a-\mu ),a\in {\alpha }_n\}$ when $\theta \in {ℝ}_{+}^{☆},\mu \in ℝ,s\in (0,r)$ or $s\in (r,+\infty )$ and $X\in L^s\left(\mathbb{P}\right)$. We show that for a wide family of distributions, one may always find parameters $(\theta ,\mu )$ such that ${\left({\alpha }_n^{\theta ,\mu }\right)}_{n\ge 1}$ is $L^s$-rate-optimal. For the gaussian and the exponential distributions we show the existence of a couple $({\theta }^{☆},{\mu }^{☆})$ such that ${\left({\alpha }^{{\theta }^{☆},{\mu }^{☆}}\right)}_{n\ge 1}$ also satisfies the so-called $L^s$-empirical measure theorem. Our conjecture, confirmed by numerical experiments, is that such sequences are asymptotically $L^s$-optimal. In both cases the sequence ${\left({\alpha }^{{\theta }^{☆},{\mu }^{☆}}\right)}_{n\ge 1}$ is incredibly close to $L^s$-optimality. However we show (see Rem. 5.4) that this last sequence is not $L^s$-optimal (e.g. when $s$ = 2, $r$ = 1) for the exponential distribution.

Détail

DOI : 10.1051/ps:2008008

Classification : 60G15, 60G35, 41A52
Keywords: rate-optimal quantizers, empirical measure theorem, dilatation, Lloyd algorithm