English

Group-averaged Markov chains: mixing improvement

Probability 2025-09-18 v2 Information Theory Group Theory math.IT Computation

Abstract

For Markov kernels PP on a general state space X\mathcal{X}, we introduce a new class of averaged Markov kernels Pda(G,ν)P_{da}(G,\nu) of PP induced by a group GG that acts on X\mathcal{X} and a probability measure ν\nu on G×GG \times G. Notable special cases are the group-orbit average P\overline{P}, left-average PlaP_{la}, right-average PraP_{ra} and the independent-double-average (Pla)ra(P_{la})_{ra}. For π\pi-stationary PP in which π\pi is invariant with respect to GG, we show that in general PdaP_{da} enjoys favorable convergence properties than PP based on metrics such as spectral gap or asymptotic variance, and within the family of PdaP_{da} the most preferable kernel is in general (Pla)ra(P_{la})_{ra}. We demonstrate that Pla,Pra,(Pla)raP_{la}, P_{ra}, (P_{la})_{ra} are comparable in terms of mixing times, which supports the use of Pla,PraP_{la}, P_{ra} in practice as computationally cheaper alternatives over (Pla)ra(P_{la})_{ra}. These averaged kernels also admit natural geometric interpretations: they emerge as unique projections of PP onto specific GG-invariant structures under the Kullback-Leibler divergence or the Hilbert-Schmidt norm and satisfy Pythagorean identities. On the other hand, in the general case if π\pi is not invariant with respect to GG, we propose and study a technique that we call state-dependent averaging of Markov kernels which generalizes the earlier results to this setting. As examples and applications, this averaging perspective not only allows us to recast state-of-the-art Markov chain samplers such as Hamiltonian Monte Carlo or piecewise-deterministic Markov processes as specific cases of PdaP_{da}, but also enables improvements to existing samplers such as Metropolis-Hastings, achieving rapid mixing in some toy models or when π\pi is the discrete uniform distribution.

Keywords

Cite

@article{arxiv.2509.02996,
  title  = {Group-averaged Markov chains: mixing improvement},
  author = {Michael C. H. Choi and Youjia Wang},
  journal= {arXiv preprint arXiv:2509.02996},
  year   = {2025}
}

Comments

68 pages

R2 v1 2026-07-01T05:18:41.694Z