Perspective: Machine learning potentials for atomistic simulations

Abstract

Nowadays, computer simulations have become a standard tool in essentially all fields of chemistry, condensed matter physics, and materials science. In order to keep up with state-of-the-art experiments and the ever growing complexity of the investigated problems, there is a constantly increasing need for simulations of more realistic, i.e., larger, model systems with improved accuracy. In many cases, the availability of sufficiently efficient interatomic potentials providing reliable energies and forces has become a serious bottleneck for performing these simulations. To address this problem, currently a paradigm change is taking place in the development of interatomic potentials. Since the early days of computer simulations simplified potentials have been derived using physical approximations whenever the direct application of electronic structure methods has been too demanding. Recent advances in machine learning (ML) now offer an alternative approach for the representation of potential-energy surfaces by fitting large data sets from electronic structure calculations. In this perspective, the central ideas underlying these ML potentials, solved problems and remaining challenges are reviewed along with a discussion of their current applicability and limitations. Published by AIP Publishing. [http://dx

Non-covalent interactions across organic and biological subsets of chemical space: Physics-based potentials parametrized from machine learning

“…In most existing approaches, the damping functions aim at modeling the outer Slater-type orbitals of atoms-e.g., note the presence of exponential functions in Eq. (9). Unfortunately, penetration effects due to the higher moments are not presently corrected.…”

“…A radically different strategy consists in predicting the potential energy surface of a system from machine learning (ML). [7][8][9] ML encompasses a number of statistical models that improve their accuracy with data. Recent studies have reported unprecedented accuracies in reproducing reference energies from electronic-structure calculations, effectively offering a novel framework for accurate intramolecular interactions freed from molecular-mechanics-type approximations (e.g., harmonic potential).…”

A universal strategy for the creation of machine learning-based atomistic force fields

“…[5][6][7][8]13,15,19,20 Typically, a fingerprint must be invariant with respect to translations and rotations of the whole system, and to permutations of like atoms. To be useful for a ML force field, it must also be directionally resolved and continuous, i.e., proportionately change with small changes of the atomic arrangement.…”

Perturbed path integrals in imaginary time: Efficiently modeling nuclear quantum effects in molecules and materials

“…When compared to the largest system sizes accessible with the most efficient DFT implementations, accurately designed force fields such as the q-TIP4P/F water model 28 allow us to consider systems containing orders of magnitude more atoms. Furthermore, machine-learned potentials 29 can become indispensable tools for cases where suitably reliable force fields are not available, for instance, the recently designed gradient-domain machine learning approach, 30 which uses only 1000 conformational geometries for training, is able to reproduce global potential energy surfaces of smalland intermediate-sized molecules with an accuracy of 0.3 kcal mol 1 for energies and 1 kcal mol 1 Å 1 for atomic forces. Successes such as these are making it more evident that the combination of machine-learned potentials with highly accurate quantum chemistry 31 or quantum Monte Carlo 32 approaches will pave the way toward a new era of electronic structure calculations.…”