Distributed Replica Sampling.

We present a simple and general scheme for efficient Boltzmann sampling of conformational space by computer simulation. Multiple replicas of the system differing in temperature T or reaction coordinate λ are simulated independently. In addition, occasional stochastic moves of individual replicas in T or λ space are considered one at a time on the basis of a generalized Hamiltonian containing an extra potential energy term or bias that depends on the distribution of all replicas. The algorithm is inherently suited for shared or heterogeneous computing platforms such as a distributed network.