Abstract:This paper considers the speed of convergence (mixing) of a finite Markov kernel $P$ with respect to the Kullback-Leibler divergence (entropy). Given a Markov kernel one defines either a discrete-time Markov chain (with the $n$-step transition kernel given by the matrix power $P^n$) or a continuous-time Markov process (with the time-$t$ transition kernel given by $e^{t(P-\mathrm{Id})}$). The contraction of entropy for $n=1$ or $t=0+$ are characterized by the famous functional inequalities, the strong data processing inequality (SDPI) and the modified log-Sobolev inequality (MLSI), respectively. When $P=KK^*$ is written as the product of a kernel and its adjoint, one could also consider the ``half-step'' contraction, which is the SDPI for $K$, while the ``full-step'' contraction refers to the SDPI for $P$. The work [DMLM03] claimed that these contraction coefficients (half-step, full-step, and continuous-time) are generally within a constant factor of each other. We disprove this and related conjectures by working out a number of different counterexamples. In particular, we construct (a) a continuous-time Markov process that contracts arbitrarily faster than its discrete-time counterpart; and (b) a kernel $P$ such that $P^{m+1}$ contracts arbitrarily better than $P^m$. Hence, our main conclusion is that the four standard inequalities comparing five common notions of entropy and variance contraction are generally not improvable. In the process of analyzing the counterexamples, we survey and sharpen the tools for bounding the contraction coefficients and characterize properties of extremizers of the respective functional inequalities. As our examples range from Bernoulli-Laplace model, random walks on graphs, to birth-death chains, the paper is also intended as a tutorial on computing MLSI, SDPI and other constants for these types of commonly occurring Markov chains.

Explosion, implosion, and moments of passage times for continuous-time Markov chains: a semimartingale approach

Markov chains revisited

Statistical Methods in Markov Chains

Circulations of Markov-Chains with Continuous-Time and the Probability Interpretation of Some Determinants

Markov Chains Approximations of jump-Diffusion Quantum Trajectories

Entropy Contractions in Markov Chains: Half-Step, Full-Step and Continuous-Time

Stationary measures of continuous time Markov chains with applications to stochastic reaction networks

An introduction to the Markov chain Monte Carlo method

On obtaining sharp bounds of the rate of convergence for a class of continuous-time Markov chains

Compressibility and Stochastic Stability of Monotone Markov Chain

Irreversible local Markov chains with rapid convergence towards equilibrium

Metastability of Exponentially Perturbed Markov Chains

On Three Classical Problems for Markov Chains with Continuous Time Parameters

Long time behavior of semi-Markov modulated perpetuity and some related processes

Scaling features of two special Markov chains involving total disasters

On the nonexplosion and explosion for nonhomogeneous Markov pure jump processes

Markov chains: Structure and applications

Asymptotics of Markov Kernels and the Tail Chain

On cover times of Markov chains

Full $Γ$-expansion of reversible Markov chains level two large deviations rate functionals

Algebraic Convergence of Markov Chains