Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 43: Computational Materials Modelling - Alloys II
MM 43.3: Vortrag
Mittwoch, 18. März 2020, 17:45–18:00, BAR 205
Machine trained interatomic potentials for multicomponent alloys — •Konstantin Gubaev1, Yuji Ikeda2, Fritz Körmann1,2, Evgeny Podryabinkin3, and Alexander Shapeev3 — 1Materials Science and Engineering, Delft University of Technology, 2628 CD, Delft, The Netherlands — 2Computational Materials Design, Max-Planck-Institut fu*r Eisenforschung GmbH, 40237 Düsseldorf, Germany — 3Skolkovo Institute of Science and Technology, Skolkovo Innovation Center, Nobel St. 3, Moscow, 143026, Russia
Accurate density-functional theory (DFT) calculations have become an indispensable tool in computational exploration and design of multicomponent alloys alas high entropy alloys (HEAs). The vastly unknown compositional and structural phase space of HEAs as well as the methodological obstacles related to the description of finite-temperature properties challenges the computational limits of standard DFT approaches and reveals the need for more efficient techniques.
Here we present DFT trained machine-learning potentials (MLIPs) as efficient force-fields for investigating finite-temperature properties of multicomponent alloys. The MLIP construction for a given alloy is performed within an iterative framework comprising the phase space sampling and subsequent MLIP training in an automated manner. As a result, MLIPs reconstructing the DFT atomic forces and total energies with unprecedented speed/accuracy combination are obtained. We demonstrate the approach by performing finite-temperature molecular-dynamics simulations to assess elastic properties of a set of bcc TiZrHfTax HEAs.