This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Submitted to SIAM Journal on Scientific Computing, March 2009 ALGEBRAIC MULTIGRID FOR MARKOV CHAINS H. DE STERCK , T.A. MANTEUFFEL , S.F. MCCORMICK , K. MILLER , J. RUGE bardbl , AND G. SANDERS Abstract. An algebraic multigrid (AMG) method is presented for the calculation of the sta tionary probability vector of an irreducible Markov chain. The method is based on standard AMG for nonsingular linear systems, but in a multiplicative, adaptive setting. A modified AMG interpo lation formula is proposed that produces a nonnegative interpolation operator with unit row sums. It is shown how the adoption of a previously described lumping technique maintains the irreducible singular Mmatrix character of the coarselevel operators on all levels. Together, these properties are sufficient to guarantee the wellposedness of the algorithm. Numerical results show how it leads to nearly optimal multigrid efficiency for a representative set of test problems. Key words. multilevel method, Markov chain, stationary probability vector, algebraic multigrid AMS subject classifications. 65C40 Computational Markov chains, 60J22 Computational methods in Markov chains, 65F10 Iterative methods for linear systems, 65F15 Eigenvalues, eigenvec tors 1. Introduction. This paper describes an algebraic multigrid (AMG) method for computing the stationary probability vector of large, sparse, irreducible Markov transition matrices. While multigrid methods of aggregation type have been considered before for Markov chains [13, 10, 9], our present approach is based on standard AMG for non singular linear systems, but in a multiplicative, adaptive setting. The current method is, in fact, an extension to nonvariational coarsening of the variational adaptive AMG scheme originally developed in the early stages of the AMG project by A. Brandt, S. McCormick, and J. Ruge [3] (described earlier in [18]). One of the features of the earlier approach is that it constructed interpolation to exactly match the minimal eigenvector of the matrix. A closely related technique called the Exact Interpola tion Scheme (EIS) was proposed by Brandt and Ron [4]. The EIS has been applied to eigenvalue problems, for example, as a multigrid solver for onedimensional Helmholtz eigenvalue problems [14]. Moreover, the current method also incorporates some as pects of early work on aggregation multigrid for Markov chains. In particular, it uses a multiplicative correction form of the coarsegrid correction process that is similar to the twolevel aggregated equations proposed in [21], and its framework is similar to the twolevel iterative aggregation/disaggregation method for Markov chains pio neered in [25] and since used and analyzed extensively (see [22] and [9] for references)....
View
Full
Document
 Fall '09
 Chong

Click to edit the document details