Optimal Poisson quantisation (Q2643385): Difference between revisions

Quantisation of a probability measure \(P\) in \(\mathbb{R}^{d}\) means that the measure \(P\) is represented using a discrete measure supported by a fixed number \(n\) of points \(\mathcal{X} \subset \mathbb{R}^{d}\). The quantisation error is given as an expected distance from the nearest representing point. The authors study the case when the representative set \(\mathcal{X}\) is a Poisson point process with the intensity \(\mu\), \(\mu(\mathbb{R}^{d})=n\). In such case the quantisation error becomes a random variable. It is quite natural to study the infimum of the expected quantisation errors over all possible intensity measures. This is discussed in Section 2 of the paper. The limit behaviour of the rescaled quantisation error as \(n \to \infty\) and when the probability measure \(P\) is uniform on a unit cube may be found in Section 3. An extension of the asymptotic results to more general absolutely continuous probability measures with Riemann integrable density \(h\) of bounded support is given in Section 4.