Use the machine Learning molecular dynamics (MLMD) calculations to generate the supercells with random displacements

[linwanhsing]

Dear Dr. Tadano,

To speed up the molecular dynamics (MD) calculations, we use machine Learning molecular dynamics (MLMD) calculations with the code of VASP.6.5.1, and 5000 steps have been simulated. The ab initio MD Iterations in the OUTCAR are about 300-400, and the Machine Learning MD Iterations in the OUTCAR are 5000. Furthermore, there are 5000 Direct configurations in the XDATCAR.

While we execute “python3 ${ALAMODE_ROOT}/tools/displace.py --VASP POSCAR_ref_supercell -md vasprun.xml -e 1001:5000:50 --random --mag 0.04 --prefix disp_aimd+random_”, it warns RuntimeError: The range specified by --every is larger than the loaded MD data.

Our questions are below,

1. How can we generate 50-80 supercells with random displacements using the MLMD with VASP.6.5.1 more efficiently?
2. Can we copy the Direct configurations in the XDATCAR as the supercells with random displacements? If we can, is there any script for generating 50-80 supercells with random displacements from the XDATCAR?

Best

Wanxing Lin
Lecturer of Physics
School of Materials Science and Engineering
Guangdong Ocean University (广东海洋大学)
No. 1, Luoqin Road, Yangjiang, 529500, Guangdong, P. R. China

[ttadano]

I think this issue occurred because displace.py script does not support a vasprun.xml file generated using the MLMD. I will investigate which part should be modified for that purpose.

[dengmou123]

Dear Dr. Tadano,
I would like to know if there is any corresponding strategy for the issue where displace.py fails to recognize vasprun.xml files generated by machine learning.
@ttadano