The 8th International Conference on Matter and Radiation at Extremes (ICMRE2025)

A wide P-T range machine learning inter-atomic potential of alumina and its property calculations
Qiannian Zhu¹,Qiyu Zeng¹,Bo Chen¹,Dongdong Kang¹ and Jiayu Dai^1,*
1Department of Physics, National University of Defense Technology, Hunan Provincial Key Laboratory of Extreme Matter and Applications, Hunan Research Center of the Basic Discipline for Physical States, Changsha 410073, China
²National Key Laboratory of Plasma Physics, Laser Fusion Research Center, Chinese Academy of Engineering Physics, Mianyang,621900, China
Abstract
Alumina is a common metal oxide with excellent insulating properties and corrosion resistance, making it crucial for industrial applications. Alumina is also an important component in mantle , which is widely used in high pressure experiments. Alumina has different kinds of phases, in high pressure, there is four main phases: α-Al₂O₃, Rh₂O₃(Ⅱ)-type Al₂O₃, CaIrO₃-type Al₂O₃ and U₂S₃-type Al₂O₃[1,2,3]. We use DeepMD approach[4,5], training a wide P-T range machine learning inter-atomic potential (MLIP) of alumina, which contains liquid phase and four phases mentioned above. This MLIP, valid from 0-600GPa and 300-13300K, is used for calculating some basic properties of alumina, such as radius distribution function (RDF), phonon spectrum and meting curve of different phases. The configuration distribution in P-T space of MLIP, the dptest of data and the comparison of DFT vs DPMD for phonon spectrum of Corundum is shown in FIG. 1. The computed properties demonstrate that this MLIP achieves high fidelity in describing the properties of alumina in extreme condition.


FIG. 1. (a)configuration distribution in P-T space of MLIP (b)phonon spectrum(c)dptest of energy per-atom(d)dptest of force(e)dptest of virial per-atom
References
[1] Mashimo T, Tsumoto K, Nakamura K, et al. High-pressure phase transformation of corundum (α-Al2O3) observed under shock compression[J]. Geophysical research letters, 2000, 27(14): 2021-2024.
[2] Tsuchiya J, Tsuchiya T, Wentzcovitch R M. Transition from the Rh2O3(II)- to-CaIrO3 structure and the high-pressure-temperature phase diagram of alumina[J]. Physical Review B—Condensed Matter and Materials Physics, 2005, 72(2): 020103.
[3] Umemoto K, Wentzcovitch R M. Prediction of an U2S3-type polymorph of Al2O3 at 3.7 Mbar[J]. Proceedings of the National Academy of Sciences, 2008, 105(18): 6526-6530.
[4] Zhang L, Han J, Wang H, et al. Deep potential molecular dynamics: a scalable model with the accuracy of quantum mechanics[J]. Physical review letters, 2018, 120(14): 143001.
[5] Wang H, Zhang L, Han J, et al. DeePMD-kit: A deep learning package for many-body potential energy representation and molecular dynamics[J]. Computer Physics Communications, 2018, 228: 178-184.





Corresponding author: J. Dai, jydai@nudt.edu.cn.
 

extreme condition,machine learning interatomic potentiail