Project README

Overview

This notebook compares MLIP-predicted energies and forces against DFT reference calculations. For each frame in a DFT trajectory, the MLIP takes the DFT-computed structure (atomic positions, cell) as input and performs a single-point evaluation, predicting E and F(r) for that fixed, pre-relaxed geometry.

This is used to find adsorption energy (E_ads = E_tot − E_slab − E_gas).

DFT Data

Download DFT runs from Box or DropBox and drop the unzipped folders into dft-data/. (A good starting point is COOH_on_YO4PdCu)
Note that DFT runs with no OUTCAR (FastRelax/OUTCAR doesn't count) and multiple OUTCARs (takes unreasonably long to run with this notebook) are invalid data here.

Conda Environment Setup

This project uses one primary Conda environment for the notebook and several model-specific environments for ML models with incompatible dependencies.

1. Create the primary environment

From the repository root:

conda env create -f model-scripts/conda_envs/mlip.yml

2. Create the model environments

From the repository root (for each model you want to run):

conda env create -f model-scripts/conda-envs/mlip-mace.yml
conda env create -f model-scripts/conda-envs/mlip-mattersim.yml
conda env create -f model-scripts/conda-envs/mlip-uma.yml
conda env create -f model-scripts/conda-envs/mlip-nep89.yml

Note on NEP89: To run this model, you must download nep89.txt (found here) and place the filepath to this in the model-scripts/run_nep89.py script.

3. Verify installation

conda env list

You should see:

mlip
mlip-mace
mlip-mattersim
mlip-uma
mlip-nep89

4. Run the notebook

Activate your primary notebook environment:

conda activate mlip

The notebook will automatically invoke model scripts using the appropriate Conda environment.

Troubleshooting Conda Env Setup

• Do not rename the Conda environments; notebook scripts expect the names defined in the YAML files.
• If environment creation fails, remove the environment and recreate it:

conda env remove -n <environment_name>
conda env create -f <yml_file>