Gaussian approximation potentials for body-centered-cubic transition metals

Abstract

We develop a set of machine-learning interatomic potentials for elemental V, Nb, Mo, Ta, and W using the Gaussian approximation potential framework. The potentials show good accuracy and transferability for elastic, thermal, liquid, defect, and surface properties. All potentials are augmented with accurate repulsive potentials, making them applicable to radiation damage simulations involving high-energy collisions. We study melting and liquid properties in detail and use the potentials to provide melting curves up to 400 GPa for all five elements.

Computational study of crystal defect formation in Mo by a machine learning molecular dynamics potential

“…In addition, the correct structure of the liquid phase and re-crystallization process should be well described, to accurately emulate atomic mixing together with defect creation and annihilation during the collision cascade. In this work, we use the ML interatomic potential for molybdenum that was recently developed [15] within the Gaussian approximation potential (GAP) framework [13,19]. Here, the total energy of a system of N atoms is expressed as…”

“…In the computation of the ML potential the descriptors for two bodies, 2b, are utilized to take into account most of the interatomic bond energies, while the many-body, mb, contributions are treated by the SOAP descriptor. More details about the computation of the ML potentials for Mo can be found in reference [15].…”

“…For each PKA energy value, the projectile travels on ten different crystal orientation: 001 , 110 , 111 , and 7 cases for r 1 r 2 r 3 , where r i are random numbers uniformly distributed in an interval of [0, 1]. The velocity Verlet integration ) obtained by GAP [15] and EAM [10], as reported in the literature [15], and their comparison to experimental measurements [25]. algorithm is utilized to model the collision cascade, which is performed for 6 ps, followed by an additional relaxation time of 4 ps.…”

“…These have the advantage that they can be systematically improved toward the accuracy of the DFT training data set. The goal of the present work is to numerically model the damage in crystalline Mo samples due to irradiation in a fusion reactor by utilizing the recently developed ML interatomic potential by Byggmästar et al [15]. Hence, we perform MD simulations to emulate neutron bombardment at intermediate primary knock-on atom (PKA) energies, providing an understanding about the modeling of the re-crystallization process after the collision cascade, which has been an issue for numerical simulations based on fixed functional forms [11,12].…”

“…Our paper is organized as follows: in section 2 we briefly discuss the theory to develop the machine learned (ML) potential [15] for Mo, and the software workflow for fingerprinting and visualizing defects in damaged crystal structures (FaVAD) [16,17] that is applied to quantify and classify the damage in Mo samples [18]. Our results for the total number of points defects, Frenkel pairs and atomic displacement are presented in section 3.…”