Abstract: Calculating equilibrium properties of molecular systems is often hindered by high energy barriers that trap simulations in local minima. Traditional methods like Umbrella Sampling (US) typically provide results for a single temperature and require significant computational resources. This paper presents a nonequilibrium "quench" method that computes equilibrium densities of states by sampling at high temperatures and then cooling the system via dissipative dynamics. We implement this approach in LAMMPS and demonstrate that while quenching alone accurately captures low-energy basins, it struggles with high-energy transition regions. By combining quenching with umbrella sampling (Quench+US), we achieve a highly efficient framework that maps free energy surfaces (FES) across multiple temperatures simultaneously. For alanine dipeptide, this hybrid method reaches convergence in 60 ns, compared to 400 ns for standard US.
Introduction
The "sampling problem" remains a central challenge in molecular dynamics (MD). Rare events, such as protein folding or chemical reactions, occur on timescales far exceeding typical simulation capabilities because systems remain trapped in metastable states. While enhanced sampling techniques like Umbrella Sampling or Metadynamics can overcome these barriers, they often require pre-defined collective variables (CVs) and are typically limited to a single target temperature. If researchers need to understand behavior across a range of temperatures, they must often run multiple independent, expensive simulations.
The Quench Method
The authors propose a framework based on a nonequilibrium quench. The process involves three main steps:
- High-Temperature Sampling: Initial configurations are drawn from an equilibrium distribution at a high temperature ($T_0$), ensuring rapid mixing and barrier crossing.
- Dissipative Quench: These configurations are evolved using zero-temperature Langevin dynamics, effectively "draining" energy from the system.
- Reweighting Estimator: A mathematical formula, incorporating a Jacobian factor to account for phase-space volume changes, maps the nonequilibrium trajectories back to equilibrium properties at a target temperature ($T$).
A custom "fix" for the LAMMPS MD engine was developed to facilitate these dynamics, making the method accessible for standard molecular simulations.
Validation and Results
The method was first validated on a system of $N$-independent harmonic springs, where it proved to be exact. This allowed for the derivation of heuristic rules for quench duration: the unitless time ($\gamma \tau$) should exceed the log of the temperature ratio ($|\ln(T_0/T)|$).
When applied to alanine dipeptide in vacuum, the limitations of pure quenching became apparent. While the method accurately identifies the most stable configurations, it deviates from reference umbrella sampling in high-free-energy regions, particularly when quenching from very high temperatures.
To address this, the authors introduced a hybrid Quench+US approach. This combination leverages the high-temperature mixing of the quench with the targeted exploration of transition regions provided by US. This hybrid scheme provides the best of both worlds: total landscape coverage and rapid convergence.
The hybrid method achieved a converged FES in just 60 ns, a nearly seven-fold speedup over the 400 ns required for standard US. Furthermore, because the quench trajectories contain information across a temperature range, the FES at multiple temperatures can be reconstructed from a single simulation set at no additional cost.
Extending to Solvated Systems
For biological molecules in water, a "solute-quench" (s-quench) variant was developed. This approach quenches only the solute degrees of freedom while maintaining the solvent at a constant temperature. This prevents numerical instabilities and the "superheating" of water that would occur if the entire system were quenched from a high $T_0$. Tests on solvated alanine dipeptide showed that s-quench accurately reproduces the FES obtained via standard US, proving the method scales to realistic biological environments.
Conclusion
The nonequilibrium quench framework, especially when combined with umbrella sampling, offers a powerful tool for molecular simulation. It provides significant speedups, is robust across temperature ranges, and is CV-agnostic in its pure form. Future work may explore its application to phase equilibria and the volumes of energy basins in even more complex landscapes.
Summary updated