Sampling from Gaussian Markov random fields conditioned on linear constraints

Abstract

Gaussian Markov random fields (GMRFs) are important modeling tools in statistics. They are often utilised to model spatially structured uncertainty, seasonal variation and other trends in the data. These last two examples of GMRFs are part of a larger class of GMRFs conditioned on linear constraints. Performing Monte Carlo Markov Chain inference on these models requires a large number of samples from GMRFs conditioned on linear constraints. Therefore it is vital to have fast and efficient methods for performing these samples. This article presents three Krylov subspace methods for sampling from a GMRF conditioned on linear constraints based on solving a Karush--Kuhn--Tucker, or saddle point, system. References

• H. Rue and L. Held. Gaussian Markov Random Fields: Theory and Applications. Chapman and Hall/CRC, 2005. USA.
• H. Rue. {Fast Sampling of {G}aussian {M}arkov Random Fields}. J. R. Statist. Soc. B, 63:325--338, 2001. doi:10.1111/1467-9868.00288
• N. Cressie. Reply to {Wahba}. The American Statistician, 44(3):256--258, 1990.
• M. Benzi, G. Golub, and J. Liesen. Numerical solutions of saddle point problems. Acta Numerica, 14:1--137, 2005. doi:10.1017/S0962492904000212
• D. P. Simpson, I. W. Turner, and A. N. Pettitt. Fast sampling from Gaussian Markov random field using Krylov subspace approaches, Scandinavian Journal of Statistics, Submitted, 2008.
• Y. Saad. Iterative Methods for Sparse Linear Systems. PWS Publishing Company, Boston, 1993.
• C. C. Paige and M. A. Saunders. Solution of sparse indefinite systems of linear equations. SIAM J. Numerical Analysis, 12:617--629, 1975. doi:10.1137/0712047
• K. Burrage and J. Erhel. On the performance of various adaptive preconditioned {GMRES} strategies. Numerical Linear Algebra with Applications, 5(2):101--121, 1998. doi:0.1002/(SICI)1099-1506(199803/04)5:2<101::AID-NLA127>3.0.CO;2-1
• A. N. Pettitt, I. S. Weir, and A. G. Hart. A conditional autoregressive {Gaussian} process for irregularly spaced multivariate data with application to modelling large sets of binary data. Statistics and Computing, 12, 2002. doi:10.1023/A:1020792130229