Computational studies on beryllium, tungsten and their mixed systems

Selected materials that can be used as plasma facing materials (PFMs) in nuclear fusion devices will be studied by quantum chemical modelling. PFMs have to withstand particularly unfriendly conditions in terms of high temperatures and irradiation by neutrons and sputtering by ions and atoms from the plasma. Besides they should not contaminate the plasma by heavy atoms. In the ITER reactor beryllium will be used as a first wall material and tungsten for the divertor that suffers the highest load. These materials are sputtered by high energetic particles and can be re-deposited elsewhere, so that mixed materials such as Be-W alloys emerge. Furthermore, the surfaces can adsorb other atoms and molecules from the plasma and react with them. A computational study on the quantum chemical level of fundamental properties of these surfaces is needed to complement the scarce and often difficult experiments. It is also needed to assist and to provide input for more approximate methods of computational materials science that can work on larger systems, like classical molecular dynamics simulations and continuum models. We will perform density functional theory (DFT) calculations on surfaces of beryllium, tungsten and its mixed materials (Be2W and Be12W as prototypical alloys). Previously most of the DFT calculations on Be and W were restricted to bulk and rather small supercells and none exist for the surfaces of mixed materials. We will employ considerable computational resources to investigate larger structures and model directly surface stability and adsorption on surfaces. In the first part of the project we will disseminate results for surface binding energies and we will study the adsorption of atoms (Be, W, C, etc.) and small molecules (H2, N2, O2, etc.) on Be, W and mixed surfaces. Our results will provide decision criteria for material selection in fusion devices, since high adsorption energies lead to debris layer formation, whereas structures with low adsorption energies can readily be sputtered. They are also valuable for large scale erosion codes and for general material comparisons. The DFT results will also be used to construct a new classical force field for the ternary Be/W/H system employing the neural network approach in the second part of the project. Neural networks are in principle able to mimic any function to arbitrary accuracy and have been successfully employed for systems with as many as four different elements. The neural network potential can then be used in a subsequent project, or by other groups for highly accurate simulations of surface sputtering or H-retention in BeW alloys. In course of the project we will collaborate with experimentalists who are working on plasma sputtering and on scanning tunneling microscopy of such surfaces. Molecular dynamics simulations performed both in-house and by collaborating groups will also complement our work.

When a particle impacts on a surface with a certain velocity, a variety of things can happen, just like it does if a meteorite hits the earth: Surface material can fly away, melting processes and crater formation can occur. In case of an atom impacting, the removal of surface atoms is called sputtering and other processes are reflection or adsorption and many more. In all cases the surface can change, often dramatically, and is slowly destroyed. Especially surfaces at the walls of future fusion reactors will have to withstand high-energy impacts and experiments on them are difficult and expensive. We have developed and tested a computational method that is based on machine learning of all interactions taking place. These interactions are stored in a neural network that in turn allows for fast calculation of these plasma-wall interactions under different environmental conditions. In this way we also contribute to theoretical materials science, a field which plays a central role in development of high-tech products and in digitalization of industrial processes.