📝 Journal of Applied Physics - Neuroevolution machine learning potential to study high temperature deformation of entropy-stabilized oxide MgNiCoCuZnO5 Paper
The project, led by Professor Keivan Esfarjani, with Bikash Timalsina as the graduate student, focuses on investigating the thermal properties of high-entropy alloys (HEAs) using molecular dynamics (MD) simulations on UVA Afton High-Performance Computing system.
- NEP Training: Conducting model generation by training a NEP potential model with extensive hyperparameter tuning to minimize energy and force loss results. The goal is to train a baseline model for HEO, then apply the same training process for CO2 and H2O data to get their model for molecular dynamics simulations.
- Molecular Dynamics Simulations: Focused on running MD simulations for Co₀.₂₅Ni₀.₇₅O and (Mg₀.₂Co₀.₂Ni₀.₂Cu₀.₂Zn₀.₂)O alloys using the J14 potential to investigate their thermal properties.
- Composition Tuning: Adjusted the atomic concentrations of Ni and Zn elements within the J14 potential to explore how thermal conductivity can be controlled. This involved studying the effects of varying concentrations on lattice distortion and analyzing the correlation between structural disorder and thermal conductivity.
- GPUMD for efficient GPU-based MD simulations.
- Python and MATLAB for data analysis and visualization, automating thermal conductivity and energy loss graphs.
- Linux commands and SLURM for managing simulation jobs on the UVA Afton supercomputer.
- Conducted over 100 thermal simulations for 4 HEAs (Mg₀.₅Ni₀.₅O, Co₀.₂₅Ni₀.₇₅O, J14, and J14 tuning).
- Optimized simulations with various runtime combinations and driving forces.
- Repeated simulations across different temperature ranges (100K-900K).
- Generated multiple graphs to visualize thermal conductivity and spectral heat currents at various temperatures.
- Automated analysis using Python and MATLAB, producing over 300 visual summaries of simulation results.