Title:Proposing new configurations in Monte Carlo simulations using truncated Markov Chains

Abstract:We propose a new Monte Carlo method in which trial configurations are generated by iterating a finite number of times an initial guess, sampled from a prior distribution, using Markov chains. Like in path-sampling Monte Carlo, we define an extended configurational space comprising all possible trajectories obtained by iterating the prior distribution by a given number of Markov updates. By enforcing detailed balance conditions between trajectories in the extended configurational space, we validate two prototypical schemes that sample desired target distributions while generating new configurations through truncated Markov Chains. As compared to existing path-sampling methods, trajectories are generated using non-local moves based on Bayesian sampling. Such a complication guarantees reasonable acceptances in the setting considered by this work. Using representative systems, we identify the important factors controlling the quality of the sampling of this type of algorithms. Considering the problem of sampling polymers with fixed endpoints, truncated Markov Chains can outperform an existing algorithm in overstretched conditions. Applications of the proposed methodology range from the design of new generative models to the improvement of the portability of Monte Carlo algorithms, like Configurational Bias schemes, relying on exact sampling of a subset of terms entering the Hamiltonian while generating new configurations.