A new proof of convergence of MCMC via the ergodic theorem (Q634566): Difference between revisions

Markov chain Monte Carlo methods (MCMC) are considered. In this approach, the authors construct a Markov chain \(X=(X_n : n \geq 0)\) having a prescribed stationary distribution \(\pi\). By simulating a trajectory of \(X\) over \({0, 1, \ldots, n-1}\), the hope is that the time average \(n^{-1} \sum_{j=0}^{n-1} f(X_j)\) will converge to \(\int_S f(x) \pi(d x)\), where \(S\) is the state space. Such property is known for an irreducible discrete state space Markov chain, while the authors consider a general state space. A key result underlaying the theory of MCMC is that any \(\eta\)-irreducible Markov chain having a transition density with respect to \(\eta\) and possessing a stationary distribution \(\pi\) is automatically positive Harris current. The paper provides a short self-contained proof of this fact, using the ergodic theorem as the most advanced tool.