Machine learning coarse grained models for water

Abstract

An accurate and computationally efficient molecular level description of mesoscopic behavior of ice-water systems remains a major challenge. Here, we introduce a set of machine-learned coarse-grained (CG) models (ML-BOP, ML-BOPdih, and ML-mW) that accurately describe the structure and thermodynamic anomalies of both water and ice at mesoscopic scales, all at two orders of magnitude cheaper computational cost than existing atomistic models. In a significant departure from conventional force-field fitting, we use a multilevel evolutionary strategy that trains CG models against not just energetics from first-principles and experiments but also temperature-dependent properties inferred from on-the-fly molecular dynamics (~ 10’s of milliseconds of overall trajectories). Our ML BOP models predict both the correct experimental melting point of ice and the temperature of maximum density of liquid water that remained elusive to-date. Our ML workflow navigates efficiently through the high-dimensional parameter space to even improve upon existing high-quality CG models (e.g. mW model).

Ice Ic without stacking disorder by evacuating hydrogen from hydrogen hydrate

“…The stacking-disordered ice has more dislocations, which promote the phase transformation from "ice Ic" to ice Ih by reducing the activation energy required to change the stacking sequence 13 . This is also supported by a recent mesoscopic-size calculation 24 . The diffraction pattern observed at 250 K looks a mixture of bulk ice Ic and Ih, rather than stacking-disordered ice with many stacking faults, judging from "stackogram" reported in the literature 8,18 .…”

Machine learning for multi-fidelity scale bridging and dynamical simulations of materials

“…This model is able to successfully capture the various thermodynamic anomalies of water and outperforms most other existing atomistic and coarsegrained models in their predictive power [41]. Multi-objective optimization such as Pareto optimization and the recent success of HOGA (hierarchical objective genetic algorithm) [41] to find optimal solutions that represent a compromise between the various desired objectives is a step in the right direction. One needs to exercise caution in defining the objective appropriately otherwise even the best global optimization strategy may not yield the best results.…”

“…A major bottleneck in the use of GA (or another optimization scheme) for FF parameterization is the evaluation of the cost function (step 2) for each candidate parameter set, which involves performing several expensive MD computations (to estimate correct phase energetics, high-temperature high-pressure stability, etc.). To overcome this problem, Chan and co-workers used a hierarchical objective function scheme, termed HOGA, presented in Figure 2b [41,42,43]. In this approach, the desired set of property objectives are segregated into different hierarchical classes based on their computational cost.…”

“…The potential was fit to not only first principles data, but the temperature dependent stability of WSe 2 structures was also among the fitting criteria. The idea is to run on-the-fly MD simulations during the potential fitting process to assess the stability of various structures at different target temperatures [41,42].…”

“…ML strategies can allow for development of coarsegrained surrogates that address the timescale challenges without sacrificing accuracy. ML CG water models developed by Chan et al [41] that allow for on-the-fly sampling of temperature dependent properties have succeeded in capturing the various thermodynamic properties and anomalies of water at fraction of the computational cost of its atomistic counterpart. Such coarser surrogate model development augurs well for capturing hitherto unexplored dynamical phenomena in the mesoscopic regime.…”