Modeling Electronic Response Properties with an Explicit-Electron Machine Learning Potential

Abstract

Explicit-electron force fields introduce electrons or electron pairs as semiclassical particles in force fields or empirical potentials, which are suitable for molecular dynamics simulations. Even though semiclassical electrons are a drastic simplification compared to a quantum-mechanical electronic wave function, they still retain a relatively detailed electronic model compared to conventional polarizable and reactive force fields. The ability of explicit-electron models to describe chemical reactions and electronic response properties has already been demonstrated, yet the description of short-range interactions for a broad range of chemical systems remains challenging. In this work, we present the electron machine learning potential (eMLP), a new explicit electron force field in which the short-range interactions are modeled with machine learning. The electron pair particles will be located at well-defined positions, derived from localized molecular orbitals or Wannier centers, naturally imposing the correct dielectric and piezoelectric behavior of the system. The eMLP is benchmarked on two newly constructed data sets: eQM7, an extension of the QM7 data set for small molecules, and a data set for the crystalline β-glycine. It is shown that the eMLP can predict dipole moments, polarizabilities, and IR-spectra of unseen molecules with high precision. Furthermore, a variety of response properties, for example, stiffness or piezoelectric constants, can be accurately reproduced.

A Carbon Is a Carbon Is a Carbon: Orbital-Free Simulations of Hydrocarbon Chemistry without Resort to Atom Types

“…The effectiveness of pairwise interactions has been demonstrated previously with closed form and machine learned potentials. In the present case, purely classical interactions, with no dispersion, have sufficed for kernel–kernel interactions, including at the intermolecular range.…”

“…The central challenges are (i) that the potentials need to be continuously differentiable functions so that molecular dynamics simulations will be energy conserving and (ii) that since chemically significant energies are small differences between very large attractive energies and similarly large kinetic and repulsive energies, extreme care is required in devising and tuning each of the potential functions. Recently, machine learning has been applied to avoid the need to devise closed-form potentials . This very powerful approach has demonstrated efficient replication and extension of calculations of molecular properties by density functional wave mechanics.…”

“…Recently, machine learning has been applied to avoid the need to devise closed-form potentials. 2 This very powerful approach has demonstrated efficient replication and extension of calculations of molecular properties by density functional wave mechanics. However, it provides relatively little physical insight or grounds for simulating reaction pathways.…”

Spectroscopy from Machine Learning by Accurately Representing the Atomic Polar Tensor

“…A complementary approach is to train partial charges such that molecular dipole moments are reproduced correctly , or to train the positions of Wannier centers. , Molecular dipole moments are generally measurable and can thus be validated against experiments. In the case of gas-phase systems (without periodic boundary conditions), the total dipole moment vector has recently been trained as a whole to predict IR spectra of protonated water clusters as well as of an ethanol and an aspirin molecule .…”

Spectroscopy from Machine Learning by Accurately Representing the Atomic Polar Tensor

“…A complementary approach is to train partial charges such, that molecular dipole moments are reproduced correctly [34,31] or to train the positions of Wannier centers [35,36]. Molecular dipole moment are generally measurable and can thus be validated against experiments.…”