Explicit Multielement Extension of the Spectral Neighbor Analysis Potential for Chemically Complex Systems

Abstract

A natural extension of the descriptors used in the Spectral Neighbor Analysis Potential (SNAP) method is derived to treat atomic interactions in chemically complex systems. Atomic environment descriptors within SNAP are obtained from a basis function expansion of the weighted density of neighboring atoms. This new formulation instead partitions the neighbor density into partial densities for each chemical element, thus leading to explicit multielement descriptors. For N elem chemical elements, the number of descriptors increases as N ( ) elem 3 , while the computational cost of the force calculation as implemented in LAMMPS is limited to N ( ) elem 2 and the favorable linear scaling in the number of atoms is retained. We demonstrate these chemically aware descriptors by producing an interatomic potential for indium phosphide capable of capturing high-energy defects that result from radiation damage cascades. This new explicit multielement SNAP method reproduces the relaxed defect formation energies with substantially greater accuracy than weighted-density SNAP, while retaining accurate representation of the bulk indium phosphide properties.

Enhanced photodegradation of dimethoxybenzene isomers in/on ice compared to in aqueous solution

“…As a refinement to our former approach, we developed a universal ML model to predict the absorption spectra for all three DMOB isomers. To accomplish the transferability, we adopted a more sophisticated atomic descriptor -the Bispectrum Component (BC) (Bartok et al 2013, previously applied to ML interatomic potential development and material property predictions (Cusentino et al 2020, Legrain et al 2017. By using BC with the least absolute shrinkage and selection operator (LASSO) regression model (Tibshirani 2011), we attain a more precise estimate of the low-energy, long-wavelength tails of the spectra, which are important for calculating rates of photon absorption since the photon flux is increasing in this region.…”

UV-Visible Absorption Spectra of Solvated Molecules by Quantum Chemical Machine Learning

“…40 While originally BC was proposed as a descriptor to approximate the Born-Oppenheimer potential energy surface of single element systems, 40 it has been applied to develop ML potential of multi-component systems and predict material properties such as elastic constants, bulk modulus as well as vibrational free energies and entropies of solids. 42,61,62 Compared to other atomic environment descriptors such as smooth overlap of atomic positions (SOAP) 26,63,64 and atom-centered symmetry functions (ACSF), 26,65 BC is more keen at describing the nuances of atomic environments, as it is projected to a more complete set of basis functions with higher dimensions. 28,43 Additionally, in the development of spectral neighbor analysis potentials, BC was formulated to retain a linear relation with the target property.…”

“…28,43 Additionally, in the development of spectral neighbor analysis potentials, BC was formulated to retain a linear relation with the target property. 41,42 This development ensures the resulting ML models to achieve robust performance using only a moderate amount of training data. Such a trade-off between model complexity and size of the training data is the most critical factor for us to bridge BC with linear regression.…”

“…Here, we devise a framework that overcomes these limitations and allows one to compute spectral line shapes that are comparable to experimental measurements by coupling highlevel QC calculations with a ML model that can be applied to several different molecules. To this scope, we adopt the bispectrum component, an atomic environment descriptor commonly used to develop ML interatomic potentials, [40][41][42][43] as the input to a regularized regression model to predict the UV-visible absorption spectra of a set of ten aromatic molecules in aqueous solution. To pursue chemical accuracy we train the ML model on the EOM-CCSD excitation energies of configurations from first-principles molecular dynamics trajectories.…”

The MLIP package: Moment Tensor Potentials with MPI and Active Learning

“…Probably the third largest class of interatomic potentials is based on linear regression with a set of basis functions. This includes the spectral neighbor analysis potential (SNAP) [74,78] including the recent extension to multicomponent systems [13], and polynomial-based approaches [18,75] including moment tensor potentials (MTP) [67]-the main focus of the present work. Our extension of MTP to multicomponent systems [26,27] goes beyond the linear regression, however, there is an alternative formulation of multicomponent MTP that stays linear [40].…”