Constrained DFT-based magnetic machine-learning potentials for magnetic alloys: a case study of Fe-Al.

We propose a machine-learning interatomic potential for multi-component magnetic materials. In this potential we consider magnetic moments as degrees of freedom (features) along with atomic positions, atomic types, and lattice vectors. We create a training set with constrained DFT (cDFT) that allows us to calculate energies of configurations with non-equilibrium (excited) magnetic moments and, thus, it is possible to construct the training set in a wide configuration space with great variety of non-equilibrium atomic positions, magnetic moments, and lattice vectors. Such a training set makes possible to fit reliable potentials that will allow us to predict properties of configurations in the excited states (including the ones with non-equilibrium magnetic moments). We verify the trained potentials on the system of bcc Fe-Al with different concentrations of Al and Fe and different ways Al and Fe atoms occupy the supercell sites. Here, we show that the formation energies, the equilibrium lattice parameters, and the total magnetic moments of the unit cell for different Fe-Al structures calculated with machine-learning potentials are in good correspondence with the ones obtained with DFT. We also demonstrate that the theoretical calculations conducted in this study qualitatively reproduce the experimentally-observed anomalous volume-composition dependence in the Fe-Al system.

Ring polymer molecular dynamics and active learning of moment tensor potential for gas-phase barrierless reactions: Application to S + H2.

Ring polymer molecular dynamics (RPMD) has proven to be an accurate approach for calculating thermal rate coefficients of various chemical reactions. For wider application of this methodology, efficient ways to generate the underlying full-dimensional potential energy surfaces (PESs) and the corresponding energy gradients are required. Recently, we have proposed a fully automated procedure based on combining the original RPMDrate code with active learning for PES on-the-fly using moment tensor potential and successfully applied it to two representative thermally activated chemical reactions [I. S. Novikov et al., Phys. Chem. Chem. Phys. 20, 29503-29512 (2018)]. In this work, using a prototype insertion chemical reaction S + H2, we show that this procedure works equally well for another class of chemical reactions. We find that the corresponding PES can be generated by fitting to less than 1500 automatically generated structures, while the RPMD rate coefficients show deviation from the reference values within the typical convergence error of the RPMDrate. We note that more structures are accumulated during the real-time propagation of the dynamic factor (the recrossing factor) as opposed to the previous study. We also observe that a relatively flat free energy profile along the reaction coordinate before entering the complex-formation well can cause issues with locating the maximum of the free energy surface for less converged PESs. However, the final RPMD rate coefficient is independent of the position of the dividing surface that makes it invulnerable to this problem, keeping the total number of necessary structures within a few thousand. Our work concludes that, in the future, the proposed methodology can be applied to realistic complex chemical reactions with various energy profiles.