(588d) Free Energy Calculation with Non-Uniform Windows for Improved Computational Efficiency: A Proof of Concept

Umbrella sampling combined with WHAM (weighted histogram analysis method) is widely used for computing free energy profiles along specified reaction coordinates. Success of the approach relies on the adequate sampling in each sampling window. For larger molecules with long relaxation times, it becomes costly to achieve this criterion. Using oligomer adsorption on a crystalline cellulose surface as a case study, we show that the actual sampling time required to reach a robust free energy profile, in the conventional approach with uniform windows, is much larger than that seen in typical literature. Lesser sampling time can result in significant errors. Inspection of sampling quality in each window shows that it is often a small number of windows being insufficiently sampled that limits the overall performance. By allowing non-uniform window sizes and biasing potentials, optimization can be applied to obtain numerical parameters that enable most efficient sampling. We show that accurate free energy profiles can be obtained with much less overall simulation time if computational resources are redistributed and optimized between different windows.