2022 | 2021 | 2020 | 2019 | 2018 | 2017 | 2016 | 2015 | 2014 | 2013 | 2012 | 2011 | 2010 | 2009 | 2008 | 2007 | 2006 | 2005 | 2004 | 2003 | 2002 | 2001 | 2000 | 1999 | 1997

Neural network force field model of MoxW1-xS2 alloys for mechanical and thermal properties

Authors: Martin Keller

Ref.: Master thesis, Martin-Luther University of Halle-Wittenberg (2021)

Abstract: A deep neural network was trained to predict the atomic forces of MoxW1-xS2 alloys. For this a dataset of about 2000 DFT calculations of 3x3 to 5x5 MoxW1-xS2 supercells with atomic displacements was created. The neural network was used to compute the phonon dispersion relations of large supercells (up to 16x16) in the second harmonic approximation. Using a model based on the polarizabilities of the individual bonds in a cell Raman spectra of the alloys were predicted from the phonon eigenmodes at the Γ point. The calculations were performed on so-called special quasi-random supercells. Some analysis of the behaviour of the A1 mode during the change of composition from MoS2 to WS2 is conducted. A splitting of the mode at 50% concentration into 4 subpeaks is observed in agreement with experiments. Correlation between the subpeaks and the different environments of the S atoms is shown. Further the discrepancy between the structures usually used for alloy calculations and real systems is investigated. Additionally the neural network was used to compute the thermal conductivity of the alloys using a linearized phonon Boltzmann equation and the relaxation time approximation. The results are only qualitative information due to the small size of the cells used (5x5).