Rigorous confidence bounds for MCMC under a geometric drift condition

We assume a drift condition towards a small set and bound the mean square error of estimators obtained by taking averages along a single trajectory of a Markov chain Monte Carlo algorithm. We use these bounds to construct fixed-width nonasymptotic confidence intervals. For a possibly unbounded function f:X→Rf:X→R, let I(f)=∫Xf(x)π(dx) be the value of interest and Iˆt,n(f)=(1/n)∑i=tt+n−1f(Xi) its MCMC estimate. Precisely, we derive lower bounds for the length of the trajectory nn and burn-in time tt which ensure that P(∣Iˆt,n(f)−I(f)∣≤ε)≥1−α. The bounds depend only and explicitly on drift parameters, on the VV-norm of ff, where VV is the drift function and on precision and confidence parameters ε,α. Next we analyze an MCMC estimator based on the median of multiple shorter runs that allows for sharper bounds for the required total simulation cost. In particular the methodology can be applied for computing posterior quantities in practically relevant models. We illustrate our bounds numerically in a simple example.