Sampling large hyperplane-truncated multivariate normal distributions

Generating multivariate normal distributions is widely used in various fields, including engineering, statistics, finance and machine learning. In this paper, simulating large truncated on the intersection of a set hyperplanes investigated. Specifically, proposed methodology focuses cases where prior extracted from stationary Gaussian process (GP). It based combining both Karhunen-Loève expansions (KLE) Matheron’s update rules (MUR). The KLE requires computation decomposition covariance matrix random variables, which can become expensive when vector too large. To address issue, input domain split smallest subdomains eigendecomposition be computed. Due to property, only first subdomain required. Through strategy, computational complexity drastically reduced. mean-square truncation block errors have been calculated. efficiency approach has demonstrated through synthetic real data studies.