Assessment of Embedding Schemes in a Hybrid Machine Learning/Classical Potentials (ML/MM) Approach.

Machine learning (ML) methods have reached high accuracy levels for the prediction of in vacuo molecular properties. However, the simulation of large systems solely through ML methods (such as those based on neural network potentials) is still a challenge. In this context, one of the most promising frameworks for integrating ML schemes in the simulation of complex molecular systems are the so-called ML/MM methods. These multiscale approaches combine ML methods with classical force fields (MM), in the same spirit as the successful hybrid quantum mechanics-molecular mechanics methods (QM/MM). The key issue for such ML/MM methods is an adequate description of the coupling between the region of the system described by ML and the region described at the MM level. In the context of QM/MM schemes, the main ingredient of the interaction is electrostatic, and the state of the art is the so-called electrostatic-embedding. In this study, we analyze the quality of simpler mechanical embedding-based approaches, specifically focusing on their application within a ML/MM framework utilizing atomic partial charges derived in vacuo. Taking as reference electrostatic embedding calculations performed at a QM(DFT)/MM level, we explore different atomic charges schemes, as well as a polarization correction computed using atomic polarizabilites. Our benchmark data set comprises a set of about 80k small organic structures from the ANI-1x and ANI-2x databases, solvated in water. The results suggest that the minimal basis iterative stockholder (MBIS) atomic charges yield the best agreement with the reference coupling energy. Remarkable enhancements are achieved by including a simple polarization correction.

Grand Canonical Investigation of the Quasi Liquid Layer of Ice: Is It Liquid?

In this study, the solid-vapor equilibrium and the quasi liquid layer (QLL) of ice Ih exposing the basal and primary prismatic faces were explored by means of grand canonical molecular dynamics simulations with the monatomic mW potential. For this model, the solid-vapor equilibrium was found to follow the Clausius-Clapeyron relation in the range examined, from 250 to 270 K, with a Δ Hsub of 50 kJ/mol in excellent agreement with the experimental value. The phase diagram of the mW model was constructed for the low pressure region around the triple point. The analysis of the crystallization dynamics during condensation and evaporation revealed that, for the basal face, both processes are highly activated, and in particular cubic ice is formed during condensation, producing stacking-disordered ice. The basal and primary prismatic surfaces of ice Ih were investigated at different temperatures and at their corresponding equilibrium vapor pressures. Our results show that the region known as QLL can be interpreted as the outermost layers of the solid where a partial melting takes place. Solid islands in the nanometer length scale are surrounded by interconnected liquid areas, generating a bidimensional nanophase segregation that spans throughout the entire width of the outermost layer even at 250 K. Two approaches were adopted to quantify the QLL and discussed in light of their ability to reflect this nanophase segregation phenomena. Our results in the µVT ensemble were compared with NPT and NVT simulations for two system sizes. No significant differences were found between the results as a consequence of model system size or of the working ensemble. Nevertheless, certain advantages of performing µVT simulations in order to reproduce the experimental situation are highlighted. On the one hand, the QLL thickness measured out of equilibrium might be affected because of crystallization being slower than condensation. On the other, preliminary simulations of AFM indentation experiments show that the tip can induce capillary condensation over the ice surface, enlarging the apparent QLL.