Self-guided Langevin dynamics via generalized Langevin equation

Area of Science:

• Computational chemistry and molecular dynamics simulations.

Background:

• Self-guided Langevin dynamics (SGLD) accelerates molecular simulations by modifying system dynamics.
• Standard SGLD breaks detailed-balance, necessitating reweighting for equilibrium sampling.
• This limitation hinders accurate thermodynamic property calculations in molecular simulations.

Purpose of the Study:

• To develop a modified SGLD method that eliminates the need for reweighting.
• To achieve exact NVT and NPT ensemble sampling using a novel SGLD approach.
• To compare the enhanced conformational sampling capabilities of the new method against traditional Langevin dynamics (LD).

Main Methods:

• Introduced a generalized Langevin equation (GLE) with a modified random force to restore detailed-balance in SGLD.
• Developed the Self-guided Langevin dynamics with a generalized Langevin equation (SGLD-GLE) method.
• Applied SGLD-GLE to alanine dipeptide and argon liquid systems for validation.

Main Results:

• SGLD-GLE successfully samples NVT and NPT ensembles exactly, removing the need for reweighting.
• Demonstrated enhanced conformational sampling compared to regular Langevin dynamics (LD).
• Showcased comparable computational complexity to LD, with full size extensibility for large systems.

Conclusions:

• The SGLD-GLE method provides accurate equilibrium sampling without reweighting.
• This method offers superior conformational sampling efficiency over LD.
• SGLD-GLE is a scalable and appealing alternative for molecular dynamics simulations.