gridmlip 0.2.3

gridmlip

Contents

• About
• Installation
• How to use
• Notebooks
• How to cite

About

gridmlip is a library for the symmetry-aware grid-based sampling of the energy landscape of mobile species in solids using machine learning interatomic potentials (MLIPs).

(!) Note: The API and functionality may change. Errors are expected. The library is under development.

Installation

pip install gridmlip

or

git clone https://github.com/dembart/gridmlip
cd gridmlip
pip install .

How to use

Here we describe the pipeline in general. For a specific example, see Notebooks.

Percolation analysis

from gridmlip import Grid
from gridmlip.integrations.sevennet import evaluate_atoms_list # sevenn must be installed

file = './data/Li10Ge(PS6)2_mp.cif'
specie = 3

### Create grid
g = Grid.from_file(
    file,
    specie,
    r_min=0.8,
    r_cut=5.0,
    resolution=0.2,
    symprec=0.1,
    empty_framework=True,
    verbose=True
)

### Prepare inequivalent atomic configurations
atoms_list = g.construct_configurations(config_format='ase')

### Predict energies
energy_list, forces_list = evaluate_atoms_list(
    atoms_list,
    model_path='./data/checkpoint_sevennet_0.pth',
    device='cuda',
    compute_force=False,
    batch_size=256
)

### Load energies
g.load_energies(energy_list)

### Determine percolation barriers
barriers = g.percolation_barriers(n_jobs=-1)

### Save .grd file for visualization in VESTA 3.0
g.write_grd('test.grd')

Generate defected configurations for MLIP pre-training

from ase.io import read
from gridmlip._generate import DefectGenerator

GENERATOR_PARAMS = {
    'specie': 3,
    'resolution': 0.25,
    'r_min': 1.3,
    'r_cut': 7.5,
    'symprec': 0.1,
    'max_sep_dist': 'auto',
    'upper_bound': 6.0,
}

gen = DefectGenerator(atoms, **GENERATOR_PARAMS)

### Symmetrically distinct defect configurations
cfgs_int = gen.generate_dataset(defect_type='int') 
cfgs_vac= gen.generate_dataset(defect_type='vac')
cfgs_frk = gen.generate_dataset(defect_type='frenkel')

Notebooks

• Using SevenNet for calculating percolation barriers

• Using moment tensor potentials (as implemented in MLIP 2.0) for calculating percolation barriers

How to cite

If you use the gridmlip library, please, consider citing this repository.