Relaxation of product Markov chains on product spaces (Q1273731)

Let us suppose that \(d\) finite sets \(X_1,\dots,X_d\) and corresponding distributions \(\pi_1,\dots,\pi_d\) are given. The prototype of this setup is provided by \(d\)-dimensional grids on a given domain in \(R^d\) with possibly direction dependent mesh size (adapted to a function on the domain). The relaxation time of product-type Markov chains on \({\mathbf X}:=\prod^d_{j=1}X_j\) which asymptotically approach \(\Pi: =\prod^d_{j=1} \pi_j\) is studied. Also, it is supposed that homogeneous Markov chains on the component sets \(X_1, \dots, X_d\) are given, driven by transition matrices \(P_j\), respectively. Firstly the author canonically extends each of the Markov chains \(P_j\) to the product by letting for \(x=(\xi_1, \dots, \xi_d)\) and \(y=(\eta_1, \dots,\eta_d)\) the extended chain be \[ \widetilde P_j (x,y):= \begin{cases} P_j(\xi_j, \eta_j), \quad & \text{if } \xi_l= \eta_l,\;l=1, \dots,d,\;l\neq j,\\ 0,\quad & \text{otherwise}. \end{cases} \] Hence, the Markov chains \(\widetilde P_j\) accept transitions in the components \(X_j\) only, whereas they remain unchanged during steps in different components. Particularly, \(\widetilde P_i \widetilde P_j= \widetilde P_j \widetilde P_i\). In Sections 3 and 4 the author studies the following problems: 1. What can be inferred about the overall mixing time of a product-type Markov chain when there are known the mixing times of the components? 2. Can we speed up mixing by a properly chosen visiting scheme? Finally an application is given: Metropolis sampling with a separable energy function. It is a good paper that contains a rigorous study of a very interesting subject.