Neuroevolution machine learning potentials: Combining high accuracy and low cost in atomistic simulations and application to heat transport

We develop a neuroevolution-potential (NEP) framework for generating neural network based machine-learning potentials. They are trained using an evolutionary strategy performing large-scale molecular dynamics (MD) simulations. A descriptor of the atomic environment is constructed on Chebyshev and Legendre polynomials. The method implemented in graphic processing units within open-source GPUMD package, which can attain computational speed over $10^7$ atom-step per second one Nvidia Tesla V100. Furthermore, per-atom heat current available NEP, paves way efficient accurate MD simulations transport materials with strong phonon anharmonicity or spatial disorder, usually cannot be accurately treated either traditional empirical potentials perturbative methods.