Computing bounds for entropy of stationary \(\mathbb{Z}^d\) Markov random fields (Q2870521)

The paper studies stationary \(\mathbb Z^d\) Markov random fields (MRF) satisfying a strong spatial mixing condition (SSM) (with an exponential decay rate \(\alpha\)). Given a stationary measure \(\mu\) which is nearest-neighbor \(\mathbb Z^d\) MRF, the authors provide lower and upper bounds for its entropy which depend on \(\alpha\). These bounds provide an algorithm to explicitly approximate the entropy which is accurate to within \(\epsilon\) in time \(\exp^{O((\log(1/\epsilon))^{(d-1)^2})}\). For \(d=2\) it is accurate to within \(O(1/n)\) in polynomial time (in \(n\)). The algorithms are entirely deterministic and rigorous and do not rely on randomized approaches such as, for example, a Monte Carlo method. The authors discuss the particular case of a stationary Gibbs measure, giving explicitly the SSM condition for such a measure. Finally, they indicate briefly how to extend their approach to study pressure which provides an algorithm with analogous precision.