Bottlenecks in Hamiltonian-Adaptive Resolution Simulation Method for Modeling Interfaces

Abstract

The Hamiltonian-Adaptive Resolution Simulation (H-AdResS) method allows to combine atomistic and particle-based coarse-grained models in a single simulation box, which makes it very attractive to model systems containing interfaces or reactive regions surrounded by an interacting environment. In our previous work [J. Chem. Inf. Model. (2026) 66 (15): 9215-9225], we implemented H-AdResS in LAMMPS (2 Aug 2023 version) and extended its use to interfaces, focusing on MOF/CO$_2$ interfaces as an example. We found that, despite its advantages, using this method properly for this kind of systems is not trivial. In this work, an in-depth analysis of the impact of the choice of thermostatting schemes and long-range electrostatics models is presented. Even though its Hamiltonian formulation enables performing H-AdResS simulations within constant temperatures ensembles, not every thermostat is appropriate. We demonstrate that Langevin thermostat is a reliable choice for this method, while Nosé-Hoover results in artifacts. In addition, we show that using short-range models such as the Damped Shifted Force method for electrostatics, a popular choice for H-AdResS simulations, can lead to non-physical results when modeling interfaces. The need of capping strategies to deal with discontinuities in forces and energies arising from abrupt changes in resolution is also discussed. Finally, the impossibility of changing the definition of the H-AdResS Hamiltonian to include a gradual interpolation of the bonded degrees of freedom is discussed. We hope that this contribution helps the reader to appropriately set up H-AdResS simulations and to assess if this method can be used to accurately model their system of interest.