A fast neural network approach for direct covariant forces prediction in complex multi-element extended systems

Abstract

Neural network force field (NNFF) is a method for performing regression on atomic structureforce relationships, bypassing expensive quantum mechanics calculation which prevents the execution of long ab-initio quality molecular dynamics simulations. However, most NNFF methods for complex multielement atomic systems indirectly predict atomic force vectors by exploiting just atomic structure rotation-invariant features and the network-feature spatial derivatives which are computationally expensive. We develop a staggered NNFF architecture exploiting both rotation-invariant and covariant features separately to directly predict atomic force vectors without using spatial derivatives, thereby reducing expensive structural feature calculation by ~180-480×. This acceleration enables us to develop NNFF which directly predicts atomic forces in complex ternary and quaternary-element extended systems comprised of long polymer chains, amorphous oxide, and surface chemical reactions. The staggered rotation-invariant-covariant architecture described here can also directly predict complex covariant vector outputs from local physical structures in domains beyond computational material science.

Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

“…First, we evaluate GNNFF performance in predicting the forces of simple organic molecules in reference to SchNet which provides one of the best published benchmarks available for predicting forces of single-molecule MD 28,30 . Subsequently, GNNFF is evaluated on complex solid-state systems in reference to DCF, which has been trained and tested for predicting forces of complicated multielement solid-state systems 39 .…”

“…Although the computational efficiencies of these ML models are better than ab initio methods, the cost of using these models in practice is still significantly high due to the computational bottlenecks that come from (1) deriving the atomic fingerprints needed for PES predictions and (2) calculating the derivatives of the PES to obtain forces 33,39 . More recently developed models address one or the other of these computational bottlenecks, but not both.…”

“…While these models have shown excellent accuracies of atomic forces in single molecule systems, these models still rely on taking the derivatives of the PES to predict the atomic forces. In a different approach, the Direct Covariant Forces (DCF) model 39 proposes an ML framework that predicts atomic forces directly using atomic fingerprints without having to take the derivatives of the PES. However, because the model relies on human-engineered atomic fingerprints, its prediction accuracy is limited.…”

The impact of hydrogen valence on its bonding and transport in molten fluoride salts

“…The method follows that of Martinez, 45,56 which has previously been used to study reactions and transport in organic systems and surfaces. 44,[57][58][59] At each step in the trajectory, the system is represented by an atomic graph with nodes N and edges E. The nodes represent individual atoms, while the edges represent bonds between atoms, dening the graph G t as below:…”

Machine Learning at the Atomic Scale

“…Using a generic atom-centered symmetry-adapted representation has proven to be more effective 68 than frameworks that assume a rigid molecular frame to achieve covariance of the predicted properties, 69,70 and has made it possible to learn an atom-centered decomposition of a scalar field like the electron density. 23,71 Although learning schemes based on covariant features or kernels 55,[72][73][74] could in principle be used to machine-learn directly the inter-atomic forces rather than the underlying atomic potential, enforcing energy conservation has proven difficult. For this reason, most of the existing machine-learning interatomic potentials are built to predict the potential, although they can incorporate forces as an indirect learning target.…”