Learning scheme to predict atomic forces and accelerate materials simulations

By V. Botu, Rampi Ramprasad. 2015

183 citations recorded by the source corpus.

The behavior of an atom in a molecule, liquid, or solid is governed by the force it experiences. If the dependence of this vectorial force on the atomic chemical environment can be learned efficiently with high fidelity from benchmark reference results---using ``big-data'' techniques, i.e., without resorting to actual functional forms---then this capability can be harnessed to enormously speed up in silico materials simulations. The present contribution provides several examples of how such a force field for Al can be used to go far beyond the length-scale and time-scale regimes presently accessible using quantum-mechanical methods. It is argued that pathways are available to systematically and continuously improve the predictive capability of such a learned force field in an adaptive manner, and that this concept can be generalized to include multiple elements.

Subfields: Atomic and Molecular Physics, and Optics, Computational Theory and Mathematics, Materials Chemistry

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atomic and Molecular Physics, and Optics

In Computational Theory and Mathematics

In Materials Chemistry

Citations in the corpus

Physically informed artificial neural networks for atomistic modeling of materials (2018, methodology): “The energies predicted by the PINN (a,c,e) and NN (b,d,f) potentials are compared with DFT calculations from [20, 21].”
Machine Learning Force Fields: Construction, Validation, and Outlook (2016, background): “16,17 The force ﬁ eld is henceforth dubbed AGNI.”
Deep Learning for Deep Chemistry: Optimizing the Prediction of Chemical Patterns (2019, unknown): “…2016; Rupp et al., 2012, 2015; Snyder et al., 2012; Hansen et al., 2013, 2015; Montavon et al., 2013; Schütt et al., 2014; Alipanahi et al., 2015; Botu and Ramprasad, 2015b; De et al., 2016; Faber et al., 2016; Sadowski et al., 2016; Wei et al., 2016; Brockherde et al., 2017; Chmiela et al.,…”
Machine-Learning Interatomic Potentials for Materials Science (2021, methodology): “The most common choices are the Gaussian process regression [80, 95, 97–101] (underlying the GAP potentials), the kernel ridge regression [29, 102, 103], the SNAP model [75, 96, 104], the MTP potentials [90], and the artificial neural network (NN) regression [21, 40, 68, 69, 105–113].”
An Atomistic Fingerprint Algorithm for Learning Ab Initio Molecular Force Fields (2017, methodology): “owever, triggers the need for methods to decompose the total energy among the atoms, which is a property of the entire system rather than individual atoms [21]. Another approach proposed by Li et al. [13, 22] is to project vector quantities onto a potentially overcomplete set of non-orthogonal basis”
Accurate interatomic force fields via machine learning with covariant kernels (2016, background): “We note that noncovariant kernels can be used and avoid these difficulties, and some have been successfully implemented [28,36].”
Active learning of linearly parametrized interatomic potentials (2016, unknown): “Three closely related works, [28, 29, 30], use Gaussian process regression to predict the forces on atoms directly, without predicting the energy and taking its gradient.”
The MLIP package: moment tensor potentials with MPI and active learning (2020, background): “These are symbolic-regression potentials [29], non-conservative force fields [8,38], force-fields for a fixed-size molecular system (typically applied to one or two organic molecules) [9, 42], on-lattice potentials [66], and cheminformatics-type models [10, 14,19,21,28,30,62,64,71], see also a recen”