Materials Design from the Melt and the Benefit of Neutron Scattering Experiments at high-flux Reactor Sources
by
Institute for advanced studies (IAS)
Recent advances in machine learning interatomic potentials allow the study of very large systems or at significantly reduced computing time by means of classical molecular dynamics simulation at a quasi-similar accuracy from DFT calculations. This is achieved via a training procedure with a high dimensional neural network with an optimal number of input variables from a data set generated from ab-initio simulations. It has e.g. been shown that this accuracy is relevant for the study of nucleation and vitrification phenomena. The long-term objective is now to establish a training procedure that is gauged by experimental static and dynamic structure factors from neutron scattering in order to derive reliable force-fields for MD simulation to model liquid properties and solidification in realistic systems. Beside macroscopic properties like density and viscosity, mean square displacements and radial distribution functions evolve as crucial input parameters in such a novel training procedure.
We measure accurate thermophysical property data on liquid alloys with various levitation techniques on ground and in space. These are also applied in combination with neutron time-of-flight spectroscopy and diffraction to study liquid structures and dynamics in multicomponent metallic melts. In special cases partial static and even dynamic structure factors are obtained via isotope substitution resulting in a comprehensive data set that further enables the development of realistic potentials. Results from MD simulation with ML trained potential than allow to elucidate the role of topological and chemical ordering fluctuations and their contributions to long-range transport, crystal growth, or vitrification. In this presentation a special emphasis is also given on the role that ILL and FRM-II will play to achieve a materials design from the melt with their high-flux, large source size, tunable repetition rate, and unprecedented intensity in a single diffraction peak.
Dr. Nicolas Walte
Dr. Debasish Saha