Building machine learning force fields for nanoclusters

Abstract

We assess Gaussian process (GP) regression as a technique to model interatomic forces in metal nanoclusters by analyzing the performance of 2-body, 3-body, and many-body kernel functions on a set of 19-atom Ni cluster structures. We find that 2-body GP kernels fail to provide faithful force estimates, despite succeeding in bulk Ni systems. However, both 3- and many-body kernels predict forces within an ∼0.1 eV/Å average error even for small training datasets and achieve high accuracy even on out-of-sample, high temperature structures. While training and testing on the same structure always provide satisfactory accuracy, cross-testing on dissimilar structures leads to higher prediction errors, posing an extrapolation problem. This can be cured using heterogeneous training on databases that contain more than one structure, which results in a good trade-off between versatility and overall accuracy. Starting from a 3-body kernel trained this way, we build an efficient non-parametric 3-body force field that allows accurate prediction of structural properties at finite temperatures, following a newly developed scheme [A. Glielmo et al., Phys. Rev. B 95, 214302 (2017)]. We use this to assess the thermal stability of Ni nanoclusters at a fractional cost of full ab initio calculations.

From DFT to machine learning: recent approaches to materials science–a review

“…A recent review of applications of high-dimensional neural neural network potentials [430] summarized the notable number of molecular and materials systems studied, which ranges from simple semiconductors such as silicon [233,431,432] and ZnO [433], to more complex systems such as water and metallic clusters [434], molecules [435][436][437], surfaces [438,439], and liquid/solid interfaces [414,440]. Force fields for nanoclusters have been developed with 2-, 3-, and many-body descriptors [441], and the hydrogen adsorption on nanoclusters was described with structural descriptors such as SOAP [442].…”

Enabling QM-accurate simulation of dislocation motion in γ−Ni and α−Fe using a hybrid multiscale approach

“…However, since the scaling is only linear, the GP model remains orders of magnitude faster than computing forces by direct DFT calculations even for the largest database used in this work (N = 320 points). These results (as well as those of many recent works on the construction of ML force fields for a variety of materials [30,31,[56][57][58]) suggest that the production of data-driven force fields could allow for a drastic reduction of computational effort while maintaining QM accuracy. However, when simulating nonequilibrium chemomechanical phenomena, training a ML force field "once and for all" before the simulation is carried out will often not represent a sufficiently good strategy, since the system is likely to explore regions of configuration space not well represented in the initial training database.…”

Guest Editorial: Special Topic on Data-Enabled Theoretical Chemistry

“…Metallic nanoclusters are important in many areas of chemistry, but realistic simulations are limited by the computational cost of DFT-MD. Zeni et al 45 study such systems via classical n-body potentials derived from ML ("M-FFs") by constructing n-body kernels that can be exactly mapped to non-parametric classical potential forms such as 3D splines. This circumvents summing over training set entries for predictions, accelerating simulations by orders of magnitude.…”