Multiple-Replica Strategies for Free-Energy Calculations in NAMD: Multiple-Walker Adaptive Biasing Force and Walker Selection Rules

From the most powerful supercomputers to multicore desktops and laptops, parallel computing architectures have been in the mainstream for some time. However, numerical schemes for calculating free energies in molecular systems that directly leverage this hardware paradigm, usually taking the form of multiple-replica strategies, are just now on the cusp of becoming standard practice. Here, we present a modification of the popular molecular dynamics program NAMD that is envisioned to facilitate the use of powerful multiple-replica strategies to improve ergodic sampling for a specific class of free-energy methods known as adaptive biasing force. We describe the software implementation in a so-called multiple-walker context, alongside the interface that makes the proposed approach accessible to the end users. We further evaluate the performance of the adaptive biasing force multiple-walker strategy for a model system, namely, the reversible folding of a short peptide, and show, in particular, in regions of the transition coordinate where convergence of the free-energy calculation is encumbered by hidden barriers, that the multiple-walker strategy can yield far more reliable results in appreciably less real time on parallel architectures, relative to standard, single-replica calculations.