Machine Learning Potentials

These examples demonstrate kALDo workflows using machine learning interatomic potentials (MLIPs). MLIPs combine near-DFT accuracy with the computational efficiency of classical potentials, enabling accurate thermal transport predictions at a fraction of the cost.

Setup Instructions

Requirements

Before running these examples, install the following packages:

  • MatterSim — Installation guide available here

  • Orb - Installation guide available here

  • TDEP - Installation guide available here

  • pyACE - Installation guide available here

  • calorine - Instllation guide available here

  • upet - Installation guide available here


List and content of examples folder

For each example, more detailed information is provided by the README.md file contained in the corresponding directory.

  • cesium_lead_bromide_NEP_TDEP: This example illustrates how to perform thermal transport simulation for a cubic cesium lead bromide system (5 atom per cell) with TDEP and GPUMD packages as force calculator.

  • gallium_arsenide_orb-v3_ORBCalculator: This example illustrates how to perform thermal transport simulation for a gallium arsenide (2 atoms per cell) system using Orb package as force calculator.

  • magnesium_oxide_MatterSim-v1-1M_mattersimcalculator: This example illustrates how to model thermal expansion coefficients for a magnesium oxide (2 atoms per cell) system using MatterSim package as force calculator.

  • silicon_MatterSim-v1-1M_mattersimcalculator: This example illustrates how model thermal expansion coefficients for a silicon diamond (2 atoms per cell) system using MatterSim package as force calculator.

  • silicon_NEP89_calorine: This example illustrates how to perform thermal transport simulation for a silicon diamond system (2 atom per cell) with calorine packages as force calculator.

  • silicon_carbide_MatterSim-v1-1M_mattersimcalculator: This example illustrates how to perform thermal transport simulation for a silicon carbide system (2 atom per cell) with MatterSim packages as force calculator.

  • wurtzite_aluminum_nitride_ACE_PyACE: This example illustrates how to perform thermal transport simulation for an aluminum nitride system (4 atom per cell) with pyACE packages as force calculator.

  • sodium_chloride_UPETCalculator: This example illustrates how to perform thermal transport simulation for a sodium chloride system (2 atom per cell) with UPET packages as force calculator.


GPU/CPU Configuration

For TensorFlow-based calculations, you can specify GPU or CPU usage following these instructions.


Git Large File Storage (LFS)

This repository uses Git LFS to handle large files. Ensure Git LFS is installed on your system by following the instructions on the Git LFS website.

Once installed, clone the repository as usual with git clone — large files will be downloaded automatically. If you’ve already cloned without Git LFS, retrieve the large files with:

git lfs pull