Design sensitivity analysis and optimization of molecular dynamic simulation with Langevin dynamics

Abstract

In molecular dynamics the system is highly nonlinear and often develops unpredictable chaotic behavior so the design sensitivity diverges and reliable values cannot be obtained. In this study, we addressed this challenge by employing the stochastic Langevin thermostat, which suppressed the divergence of sensitivities and enabled stable evaluation over long-time simulations. To efficiently compute sensitivities in systems with a large number of design variables, the adjoint variable method was introduced. An optimal LJ potential for liquid argon was successfully designed based on the developed methodology. The objective function was defined as the difference between the simulated and experimental radial distribution functions, which allowed us to obtain an accurate LJ potential parameter set. Furthermore, it was demonstrated that tabulated potentials can also be constructed through the proposed optimization framework.