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 43.2: Vortrag

Mittwoch, 18. März 2020, 17:30–17:45, BAR 205

High-temperature thermodynamics of Ni alloys with machine learning — •Nataliya Lopanitsyna and Michele Ceriotti — École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

Many thermodynamic properties of metals show a pronounced temperature dependence that is challenging to obtain fully from ab initio. However, to achieve agreement between theoretical considerations and experiment, temperature-dependent contributions also should be taken into account. One of the problems is the absence of an accurate and efficient technique to obtain the correction. It becomes even more evident for properties such as the excess free energy associated with a solid-liquid interface which requires a number of atoms and simulation times unaccessible with electronic structure calculations. Interatomic potentials could be used to overcome computational expenses encountered in relation to ab initio methods. In this paper, to achieve high accuracy in the description of the interatomic interaction, we trained a neural network to approximate the potential energy surface defined by a solution of the Kohn Sham equation of DFT and incorporated it into free energy sampling techniques to quantify anharmonic effects appearing at high temperatures. We present a wide range of properties computed at finite temperature including elastic properties, bulk modulus, melting temperature, formation energies of single point defects, surface tension for nickel chosen as a representative of metal used for high-temperature applications. Additionally, we show the importance of having an accurate underlying interatomic potential by comparing fitted NNp to potentials reported in the literature.