Sputtering of the beryllium tungsten alloy Be₂W by deuterium atoms: molecular dynamics simulations using machine learned forces

Abstract

Material erosion and fuel retention will limit the life and the performance of thermonuclear fusion reactors. In this work, sputtering, reflection and retention processes are atomistically modeled by simulating the non-cumulative sputtering by deuterium projectiles on a beryllium–tungsten alloy surface. The forces for the molecular dynamics trajectories were machine learned from density functional theory with a neural network architecture. Our data confirms and supplements previous results for simulated sputtering rates. In the non-cumulative scenario we simulate, we did not observe reaction mechanisms leading to swift chemical sputtering. Thus, our sputtering rates at low impact energies are smaller than in comparable non-cumulative studies. The sputtering yields of the Be2W alloy are generally lower than those of pure beryllium. We found a strong dependence of the sputtering yield on the incident angle with an increase by about a factor of 3 for larger incident angles at 100 eV impact energy. In the pristine surface, a large majority of the impacting hydrogen projectiles at perpendicular impact remains in the surface.

Theory and molecular simulations of plasma sputtering, transport and deposition processes

“…Both the quality of the potential and the computer time required when using complex many-body and targeted phenomena drive the choice of the force fields. Moreover, very recently machine learning based force fields emerged in the field of sputtering studies [209].…”

2022 Review of Data-Driven Plasma Science

“…Figure 37 shows results from sputtering simulations of a Be 2 W surface. The trajectories of MD runs with different angles of the incoming deuterium atoms are analysed to obtain density distributions (histograms) of the angles with which Be atoms are sputtered away [622]. Similar studies have been performed also for other surfaces as well [623].…”

“…FIG. 37: Effect of the angles (0°, 20°,45°, and 60°to the surface normal) of deuterium atoms incoming with 100 eV on the angular distributions with which Be atoms are sputtered [622].…”

Modelling the impact of argon atoms on a tungsten surface

“…We model the non-cumulative sputtering from tungsten surfaces by Ar projectiles, as well as retention events. Based on former works [29][30][31], we continue to develop neural network potential energy functions using the Behler-Parrinello approach [28] as implemented in the n 2 p 2 code [32,33]. With neural networks representing the potential energy surfaces, sputtering, reflection, and retention probabilities for various incident angles α (0°, 20°, 40°, and 60°) on bcc-W(110) are calculated for projectiles with 100-800 eV kinetic energy and surface temperatures of 300, 1500, and 2500 K.…”