Entropy-constrained functional quantization of Gaussian processes
The paper is devoted to the quantization problem of probability distributions. The history of the theory and practice of quantization dates to 1948, although similar ideas had appeared in the literature as long ago as 1898. The first systematic mathematical treatment of this problem one can find in [the authors, ``Foundations of quantization for probability distributions (2000; Zbl 0951.60003)]. The main result of the paper is the sharp asymptotics for the entropy-constrained \(L^2\)-quantization error for a large class of Gaussian measures on a Hilbert space. Let \(B\) be a real separable Hilbert space and let \(X\) be a centered Gaussian \(B\)-valued random vector. Moreover, let \(\mathcal F\) denote the set of quantizers \(f\) of \(X\) (i.e. \(f:B\to B\) is Borel measurable and \(\operatorname{card} f(B)<\infty\)). For \(H\geq0\) the \(L^2\)-quantization problem under the entropy-constraint \(H\) consists in minimizing \(E\| X-f(X)\|^2\) over all \(f\in\mathcal F\) with \(H_X(f)\leq H\), where \(H_X(f)\) is the entropy of \(f\). The minimal \(H\)-quantization error of \(X\) is defined by the equality \[ e_H(X)= \inf\{(E\| X-f(X)\|^2)^{1/2}: f\in{\mathcal F}, \;H_X(f)\leq H\}. \] Let \(K_X\subset B\) be the reproducing kernel Hilbert space and let \(\lambda_1\geq\lambda_2\geq\ldots>0\) be ordered nonzero eigenvalues of the covariance operator of \(X\). The following theorem, which is the main result of the paper, is devoted to the case \(\dim K_X=\infty\) and regularly varying eigenvalues, and, therefore, it includes a wide class of Gaussian processes (recall that \(a_n\sim b_n\) means \(a_n/b_n\to1\)). Theorem 1.1. Let \(X\) be a centered Gausian \(B\)-valued random vector with \(\dim K_X=\infty\). Assume \(\lambda_j\sim\varphi(j)\) as \(j\to\infty\), where \(\varphi:(s,\infty)\to (0,\infty)\) is a decreasing, regularly varying function at infinity of index \(-b<-1\) for some \(s\geq0\). Then \[ e_H(X)\sim((b/2)^{b-1} b/(b-1))^{1/2} (H\varphi(H))^{1/2} \quad {\text{as}} \quad H\to\infty. \]
- Sharp asymptotics of the functional quantization problem for Gaussian processes.
- Functional quantization and small ball probabilities for Gaussian processes
- Functional quantization of Gaussian processes
- Asymptotically optimal quantization schemes for Gaussian processes on Hilbert spaces
- Quantization of Gaussian measures with Rényi--entropy constraints
- On the entropy of a function
- An operator theoretical characterization of -entropy in Gaussian processes
- Small ball probabilities around random centers of Gaussian measures and applications to quantization
- Functional quantization of Gaussian processes
- Sharp asymptotics of the functional quantization problem for Gaussian processes.
- Sharp asymptotics of the Kolmogorov entropy for Gaussian measures
- Constructive quadratic functional quantization and critical dimension
- High-resolution product quantization for Gaussian processes under sup-norm distortion
- Asymptotically optimal quantization schemes for Gaussian processes on Hilbert spaces
- Quantization of stationary Gaussian random processes and their generalizations
- Uniform and non-uniform quantization of Gaussian processes
- Error bounds for high-resolution quantization with Rényi-\(\alpha\)-entropy constraints
- Quantization of Gaussian measures with Rényi--entropy constraints
- Higher-order terms in asymptotic expansion for information loss in quantized random processes
- Small deviations for two classes of Gaussian stationary processes and \(L^p\)-functionals, \(0<p\leq\infty\)
This page was built for publication: Entropy-constrained functional quantization of Gaussian processes
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5461360)
