Analysis of Circulant Embedding Methods for Sampling Stationary Random Fields

Abstract

A standard problem in uncertainty quantification and in computational statistics is the sampling of stationary Gaussian random fields with given covariance in a d-dimensional (physical) domain. In many applications it is sufficient to perform the sampling on a regular grid on a cube enclosing the physical domain, in which case the corresponding covariance matrix is nested block Toeplitz. After extension to a nested block circulant matrix, this can be diagonalised by FFT -the "circulant embedding method". Provided the circulant matrix is positive definite, this provides a finite expansion of the field in terms of a deterministic basis, with coefficients given by i.i.d. standard normals. In this paper we prove, under mild conditions, that the positive definiteness of the circulant matrix is always guaranteed, provided the enclosing cube is sufficiently large. We examine in detail the case of the Matérn covariance, and prove (for fixed correlation length) that, as h 0 → 0, positive definiteness is guaranteed when the random field is sampled on a cube of size order (1 + ν 1/2 log h −1 0 ) times larger than the size of the physical domain. (Here h 0 is the mesh spacing of the regular grid and ν the Matérn smoothness parameter.) We show that the sampling cube can become smaller as the correlation length decreases when h 0 and ν are fixed. Our results * March 21, 2018. are confirmed by numerical experiments. We prove several results about the decay of the eigenvalues of the circulant matrix. These lead to the conjecture, verified by numerical experiment, that they decay with the same rate as the Karhunen-Loève eigenvalues of the covariance operator. The method analysed here complements the numerical experiments for uncertainty quantification in porous media problems in an earlier paper by the same authors in

Recycling Samples in the Multigrid Multilevel (Quasi-)Monte Carlo Method

“…Note also that there is an end effect for the smallest eigenvalues, where the asymptotic decay rate is not preserved. Similar numerical results in case of a circulant embedding method and corresponding analysis can be found in [15].…”

Multilevel quasi-Monte Carlo integration with product weights for elliptic PDEs with lognormal coefficients

“…The SPDE (1.1) can be numerically solved by Fourier methods. We refer to [16,27] for details on this. Unlike the product weights for QMC integration which were derived in the present work, the appearance of QMC weights with POD structure in [26,36] implies that the construction cost for these QMC integration methods scales as O(s 2 N + sN log(N )) [41].…”

Application of Quasi-Monte Carlo Methods to PDEs with Random Coefficients – An Overview and Tutorial

“…The resulting large matrix factorization problem could potentially be handled by circulant embedding and FFT, if the covariance function is stationary and the grid is regular, see [10]. In fact, this was the approach taken in the first QMC paper for PDEs with random coefficients [20], and the corresponding analysis is being considered in [21,22].…”