mat-defect-energy

Point-Defect Formation Energy (MLIP)

Goal

To calculate the formation energy ($E_f$) of neutral point defects (vacancies, substitutions, and interstitials) using Machine Learning Interatomic Potentials (MLIPs). Formation energy is defined as:

$$E_f = E_\mathrm{defect} - \frac{n_\mathrm{defect}}{n_\mathrm{bulk}} E_\mathrm{bulk} + \sum_i \Delta n_i \mu_i$$

where $E_\mathrm{defect}$ and $E_\mathrm{bulk}$ are the total energies of the defective and pristine supercells, $n$ is the number of atoms, $\Delta n_i$ is the change in number of species $i$, and $\mu_i$ is the chemical potential of species $i$.

Instructions

1. Select Level of Theory

Choose an MLIP model. See ml-foundation-potentials for guidance.

• Recommended: r2SCAN-level potentials for inorganic defects (e.g., MACE-MH-1 with matpes_r2scan head).
• Use the same model for bulk and defect calculations.

2. Obtain Bulk Structure

Start with a relaxed bulk primitive cell. You can retrieve one from Materials Project:

mcp_base_search_materials_project_by_formula(formula="MgO", save_to_file="MgO.cif")

3. Relax Bulk Structure

Relax the bulk unit cell to get the reference energy:

mcp_mace_load_model(model_name="MACE-MH-1", task_name="matpes_r2scan")
mcp_mace_relax_structure(
    structure_data="MgO.cif",
    relax_cell=True,
    fmax=0.01,
    output_dir="bulk_relaxation/"
)

Record the final energy per atom from the output.

4. Generate Defect Supercells

Use the defect generation script with pymatgen-analysis-defects:

# Env: base-agent
python .agents/skills/mat-defect-energy/scripts/generate_defects.py \
    --bulk bulk_relaxation/relaxed_structure.cif \
    --supercell_size 2 2 2 \
    --defect_type vacancy \
    --output defect_structures/

Options for --defect_type:

• vacancy — removes each symmetry-unique atom
• substitution — replaces atoms with --substitute_element at each unique site
• interstitial — inserts --interstitial_element at Voronoi interstitial sites
• all — generates all vacancy types