Dresden 2020 – wissenschaftliches Programm

Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MA: Fachverband Magnetismus

MA 1: Computational Magnetism I

MA 1.1: Vortrag

Montag, 16. März 2020, 09:30–09:45, HSZ 04

Accelerated evaluation of thermal conductivity via machine learning: A case study of two-dimensional (2D) BN — •Yixuan Zhang, Chen Shen, and Hongbin Zhang — Institute of Materials Science, TU Darmstadt, 64287 Darmstadt, Germany

Accurate density functional theory (DFT) calculations to evaluate the anharmonic effect are demanding, as accurate force constants from DFT up to the third or even higher orders are needed. In this work, using the recently developed machine learning technique, we obtained accurate force constants by learning over a limited number of configurations and demonstrated that the thermal conductivity can be evaluated accurately. The interatomic potential is developed using the GAP model, and the resulting forces are fed into Alamode to evaluate the thermal conductivity. The configurations for training were automatically selected using the active learning method, which enables future on-the-fly calculations. For 2D BN sheets, it is demonstrated that the final training set can be reduced to 123 out of total 867 geometries, and the resulting thermal conductivity is only slightly deviated from the values obtained via explicit DFT calculations. It is suspected that the method can be applied to other 2D/3D compounds where the computational effort required can be reduced by one order of magnitude.