SHARP ENTRYWISE PERTURBATION BOUNDS FOR MARKOV CHAINS

被引:12
作者
Thiede, Erik [1 ]
Van Koten, Brian [2 ]
Weare, Jonathan [2 ,3 ]
机构
[1] Univ Chicago, Dept Chem, Gordon Ctr Integrat Sci, Chicago, IL 60637 USA
[2] Univ Chicago, Dept Stat, Chicago, IL 60637 USA
[3] Univ Chicago, James Franck Inst, Chicago, IL 60637 USA
关键词
Markov chains; stochastic matrices; perturbation bounds; condition numbers; sensitivity analysis; STATIONARY DISTRIBUTION; SENSITIVITY; PROBABILITIES; 1ST;
D O I
10.1137/140987900
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
For many Markov chains of practical interest, the invariant distribution is extremely sensitive to perturbations of some entries of the transition matrix, but insensitive to others; we give an example of such a chain, motivated by a problem in computational statistical physics. We have derived perturbation bounds on the relative error of the invariant distribution that reveal these variations in sensitivity. Our bounds are sharp, we do not impose any structural assumptions on the transition matrix or on the perturbation, and computing the bounds has the same complexity as computing the invariant distribution or computing other bounds in the literature. Moreover, our bounds have a simple interpretation in terms of hitting times, which can be used to draw intuitive but rigorous conclusions about the sensitivity of a chain to various types of perturbations.
引用
收藏
页码:917 / 941
页数:25
相关论文
共 21 条
[1]  
Berman A., 1994, NONNEGATIVE MATRICES, DOI [DOI 10.1137/1.9781611971262, 10.1137/1.9781611971262.fm]
[2]   Markov chain sensitivity measured by mean first passage times [J].
Cho, GE ;
Meyer, CD .
LINEAR ALGEBRA AND ITS APPLICATIONS, 2000, 316 (1-3) :21-28
[3]   Comparison of perturbation bounds for the stationary distribution of a Markov chain [J].
Cho, GE ;
Meyer, CD .
LINEAR ALGEBRA AND ITS APPLICATIONS, 2001, 335 :137-150
[4]   ON THE 1ST AND 2ND ORDER DERIVATIVES OF THE PERRON VECTOR [J].
DEUTSCH, E ;
NEUMANN, M .
LINEAR ALGEBRA AND ITS APPLICATIONS, 1985, 71 (NOV) :57-76
[5]   SENSITIVITY OF THE STATIONARY DISTRIBUTION VECTOR FOR AN ERGODIC MARKOV-CHAIN [J].
FUNDERLIC, RE ;
MEYER, CD .
LINEAR ALGEBRA AND ITS APPLICATIONS, 1986, 76 :1-17
[6]   USING THE QR FACTORIZATION AND GROUP INVERSION TO COMPUTE, DIFFERENTIATE, AND ESTIMATE THE SENSITIVITY OF STATIONARY PROBABILITIES FOR MARKOV-CHAINS [J].
GOLUB, GH ;
MEYER, CD .
SIAM JOURNAL ON ALGEBRAIC AND DISCRETE METHODS, 1986, 7 (02) :273-281
[7]   PERTURBATION BOUNDS FOR THE STATIONARY PROBABILITIES OF A FINITE MARKOV-CHAIN [J].
HAVIV, M ;
VANDERHEYDEN, L .
ADVANCES IN APPLIED PROBABILITY, 1984, 16 (04) :804-818
[8]   Stationary distributions and mean first passage times of perturbed Markov chains [J].
Hunter, JJ .
LINEAR ALGEBRA AND ITS APPLICATIONS, 2005, 410 :217-243
[9]   UNIFORM STABILITY OF MARKOV-CHAINS [J].
IPSEN, ICF ;
MEYER, CD .
SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 1994, 15 (04) :1061-1074
[10]  
KATO T., 1966, Perturbation Theory for Linear Operators