UC Davis

Gaussian accelerated molecular dynamics (GaMD) enhances sampling without predefined collective variables (CVs), but it scales poorly on GPUs. The weighted ensemble method (WE) is statistically unbiased and parallelizable, yet it requires CVs and can converge slowly from inadequate initial states. To mitigate these bottlenecks, we developed two hybrid methods: (1) GaMD–WE and (2) parallelizable GaMD (ParGaMD). GaMD–WE uses short GaMD simulations to generate diverse, well‐sampled starting configurations with reweighted probabilities for subsequent WE calculations, improving efficient estimation of both thermodynamic and kinetic properties. ParGaMD integrates GaMD dynamics within the WE framework: many short GaMD trajectories act as WE walkers in parallel, which improves computational scalability and accelerates convergence of the free energy landscape. We discuss both hybrids in detail and demonstrate their superiority in sampling compared to standalone GaMD and WE on complex biomolecular systems.