This repository contains PET-MAD - a universal interatomic potential for advanced materials modeling across the periodic table. This model is based on the Point Edge Transformer (PET) model trained on the Massive Atomic Diversity (MAD) Dataset and is capable of predicting energies and forces in complex atomistic simulations.
In addition, it contains PET-MAD-DOS - a universal model for predicting the density of states (DOS) of materials, as well as their Fermi levels and bandgaps. PET-MAD-DOS is using a slightly modified PET architecture, and the same MAD dataset.
- Universality: PET-MAD models are generally-applicable, and can be used for predicting energies and forces, as well as the density of states, Fermi levels, and bandgaps for a wide range of materials and molecules.
- Accuracy: PET-MAD models achieve high accuracy in various types of atomistic simulations of organic and inorganic systems, comparable with system-specific models, while being fast and efficient.
- Efficiency: PET-MAD models are highly computationally efficient and have low memory usage, what makes them suitable for large-scale simulations.
- Infrastructure: Various MD engines are available for diverse research and application needs.
- HPC Compatibility: Efficient in HPC environments for extensive simulations.
- Installation
- Pre-trained Models
- Interfaces for Atomistic Simulations
- Usage
- Examples
- Fine-tuning
- Documentation
- Citing PET-MAD
You can install PET-MAD using pip:
pip install pet-madOr directly from the GitHub repository:
pip install git+https://github.com/lab-cosmo/pet-mad.gitAlternatively, you can install PET-MAD using conda package manager, which is
especially important for running PET-MAD with LAMMPS.
Warning
We strongly recommend installing PET-MAD using
Miniforge as a base conda
provider, because other conda providers (such as Anaconda) may yield
undesired behavior when resolving dependencies and are usually slower than
Miniforge. Smooth installation and use of PET-MAD is not guaranteed with
other conda providers.
To install Miniforge on unix-like systems, run:
wget "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"
bash Miniforge3-$(uname)-$(uname -m).shOnce Miniforge is installed, create a new conda environment and install PET-MAD with:
conda create -n pet-mad
conda activate pet-mad
conda install -c metatensor -c conda-forge pet-madCurrently, we provide the following pre-trained models:
v1.1.0: The dev version of the PET-MAD model with the non-conservative forces and stresses. This version has notably worse performance on molecular systems, and is not recommended for production use, as for now.v1.0.2: Stable PET-MAD model trained on the MAD dataset, which contains 95,595 structures, including 3D and 2D inorganic crystals, surfaces, molecular crystals, nanoclusters, and molecules. Use this version in the case you want to repoduce the results from the PET-MAD paper.
PET-MAD integrates with the following atomistic simulation engines:
- Atomic Simulation Environment (ASE)
- LAMMPS (including the KOKKOS support)
- i-PI
- OpenMM (coming soon)
- GROMACS (coming soon)
You can use the PET-MAD calculator, which is compatible with the Atomic Simulation Environment (ASE):
from pet_mad.calculator import PETMADCalculator
from ase.build import bulk
atoms = bulk("Si", cubic=True, a=5.43, crystalstructure="diamond")
pet_mad_calculator = PETMADCalculator(version="latest", device="cpu")
atoms.calc = pet_mad_calculator
energy = atoms.get_potential_energy()
forces = atoms.get_forces()These ASE methods are ideal for single-structure evaluations, but they are inefficient for the evaluation on a large number of pre-defined structures. To perform efficient evaluation in that case, read here.
PET-MAD also supports the direct prediction of forces and stresses. In that case, the forces and stresses are predicted as separate targets along with the energy target, i.e. not computed as derivatives of the energy using the PyTorch automatic differentiation. This approach typically leads to 2-3x speedup in the evaluation time, since backward pass is disabled. However, as discussed in this preprint it requires additional care to avoid instabilities during the molecular dynamics simulations.
To use the non-conservative forces and stresses, you need to set the non_conservative parameter to True when initializing the PETMADCalculator class.
from pet_mad.calculator import PETMADCalculator
from ase.build import bulk
atoms = bulk("Si", cubic=True, a=5.43, crystalstructure="diamond")
pet_mad_calculator = PETMADCalculator(version="latest", device="cpu", non_conservative=True)
atoms.calc = pet_mad_calculator
energy = atoms.get_potential_energy() # energy is computed as usual
forces = atoms.get_forces() # forces now are predicted as a separate target
stresses = atoms.get_stress() # stresses now are predicted as a separate targetMore details on how to make the direct forces MD simulations reliable are provided in the Atomistic Cookbook.
Efficient evaluation of PET-MAD on a desired dataset is also available from the
command line via metatrain, which
is installed as a dependency of PET-MAD. To evaluate the model, you first need
to fetch the PET-MAD model from the HuggingFace repository:
mtt export https://huggingface.co/lab-cosmo/pet-mad/resolve/v1.0.2/models/pet-mad-v1.0.2.ckpt -o pet-mad-latest.ptAlternatively, you can also download the model from Python:
import pet_mad
# Saving the latest version of PET-MAD to a TorchScript file
pet_mad.save_pet_mad(version="latest", output="pet-mad-latest.pt")
# you can also get a metatomic AtomisticModel for advance usage
model = pet_mad.get_pet_mad(version="latest")This command will download the model and convert it to TorchScript format. Then
you need to create the options.yaml file and specify the dataset you want to
evaluate the model on (where the dataset is stored in extxyz format):
systems: your-test-dataset.xyz
targets:
energy:
key: "energy"
unit: "eV"Then, you can use the mtt eval command to evaluate the model on a dataset:
mtt eval pet-mad-latest.pt options.yaml --batch-size=16 --extensions-dir=extensions --output=predictions.xyzThis will create a file called predictions.xyz with the predicted energies and
forces for each structure in the dataset. More details on how to use metatrain
can be found in the Metatrain documentation.
PET-MAD can also be used to calculate the uncertainty of the energy prediction. This feature is particularly important if you are interested in probing the model on the data that is far away from the training data. Another important use case is a propagation of the uncertainty of the energy prediction to other observables, like phase transition temperatures, diffusion coefficients, etc.
To activate the uncertainty quantification, you need to set the
calculate_uncertainty and / orcalculate_ensemble parameters to True when
initializing the PETMADCalculator class. The first feature will calculate the
uncertainty of the energy prediction, while the second one will calculate the
ensemble of the energy predictions based on the shallow ensemble of the last
layers of the model.
from pet_mad.calculator import PETMADCalculator
from ase.build import bulk
atoms = bulk("Si", cubic=True, a=5.43, crystalstructure="diamond")
pet_mad_calculator = PETMADCalculator(version="latest", device="cpu", calculate_uncertainty=True, calculate_ensemble=True)
atoms.calc = pet_mad_calculator
energy = atoms.get_potential_energy()
energy_uncertainty = atoms.calc.get_energy_uncertainty()
energy_ensemble = atoms.calc.get_energy_ensemble()More details on the uncertainty quantification and shallow ensemble method can be found in this and this papers.
To use PET-MAD with LAMMPS, you need to first install PET-MAD from conda (see
the installation instructions above). Then, follow the instructions
here to install lammps-metatomic. We recomend you also use conda to install lammps.
Fetch the PET-MAD checkpoint from the HuggingFace repository:
mtt export https://huggingface.co/lab-cosmo/pet-mad/resolve/v1.0.2/models/pet-mad-v1.0.2.ckpt -o pet-mad-latest.ptThis will download the model and convert it to TorchScript format compatible
with LAMMPS, using the metatomic and metatrain libraries, which PET-MAD is
based on.
Prepare a lammps input file using pair_style metatomic and defining the
mapping from LAMMPS types in the data file to elements PET-MAD can handle using
pair_coeff syntax. Here we indicate that lammps atom type 1 is Silicon (atomic
number 14).
units metal
atom_style atomic
read_data silicon.data
pair_style metatomic pet-mad-latest.pt device cpu # Change device to 'cuda' evaluate PET-MAD on GPU
pair_coeff * * 14
neighbor 2.0 bin
timestep 0.001
dump myDump all xyz 10 trajectory.xyz
dump_modify myDump element Si
thermo_style multi
thermo 1
velocity all create 300 87287 mom yes rot yes
fix 1 all nvt temp 300 300 0.10
run 100
Create the silicon.data data file for a silicon system.
# LAMMPS data file for Silicon unit cell
8 atoms
1 atom types
0.0 5.43 xlo xhi
0.0 5.43 ylo yhi
0.0 5.43 zlo zhi
Masses
1 28.084999992775295 # Si
Atoms # atomic
1 1 0 0 0
2 1 1.3575 1.3575 1.3575
3 1 0 2.715 2.715
4 1 1.3575 4.0725 4.0725
5 1 2.715 0 2.715
6 1 4.0725 1.3575 4.0725
7 1 2.715 2.715 0
8 1 4.0725 4.0725 1.3575
lmp -in lammps.in # For serial version
mpirun -np 1 lmp -in lammps.in # For MPI versionRunning LAMMPS with KOKKOS and GPU support is similar to the CPU version, but
you need to change the lammps.in slightly and run lmp binary with a few
additional flags.
The updated lammps.in file looks like this:
package kokkos newton on neigh half
units metal
atom_style atomic/kk
read_data silicon.data
pair_style metatensor/kk pet-mad-latest.pt # This will use the same device as the kokkos simulation
pair_coeff * * 14
neighbor 2.0 bin
timestep 0.001
dump myDump all xyz 10 trajectory.xyz
dump_modify myDump element Si
thermo_style multi
thermo 1
velocity all create 300 87287 mom yes rot yes
fix 1 all nvt temp 300 300 0.10
run_style verlet/kk
run 100
The silicon.data file remains the same.
To run the KOKKOS-enabled version of LAMMPS, you need to run
lmp -in lammps.in -k on g 1 -sf kk # For serial version
mpirun -np 1 lmp -in lammps.in -k on g 1 -sf kk # For MPI versionHere, the -k on g 1 -sf kk flags are used to activate the KOKKOS
subroutines. Specifically g 1 is used to specify, how many GPUs are the
simulation is parallelized over, so if running the large systems on two or more
GPUs, this number should be adjusted accordingly.
-
For CPU calculations, use a single MPI task unless simulating large systems (30+ Å box size). Multi-threading can be enabled via:
export OMP_NUM_THREADS=4 -
For GPU calculations, use one MPI task per GPU.
You can combine the PET-MAD calculator with the torch based implementation of
the D3 dispersion correction of pfnet-research - torch-dftd:
Within the PET-MAD environment you can install torch-dftd via:
pip install torch-dftdThen you can use the D3Calculator class to combine the two calculators:
from torch_dftd.torch_dftd3_calculator import TorchDFTD3Calculator
from pet_mad.calculator import PETMADCalculator
from ase.calculators.mixing import SumCalculator
device = "cuda" if torch.cuda.is_available() else "cpu"
calc_MAD = PETMADCalculator(version="latest", device=device)
dft_d3 = TorchDFTD3Calculator(device=device, xc="pbesol", damping="bj")
combined_calc = SumCalculator([calc_MAD, dft_d3])
# assign the calculator to the atoms object
atoms.calc = combined_calcPET-MAD packages also allows the use of the PET-MAD-DOS model to predict electronic density of states of materials, as well as their Fermi levels and bandgaps. Similarly to the PET-MAD model, the PET-MAD-DOS model is also available in the ASE interface.
from pet_mad.calculator import PETMADDOSCalculator
atoms = bulk("Si", cubic=True, a=5.43, crystalstructure="diamond")
pet_mad_dos_calculator = PETMADDOSCalculator(version="latest", device="cpu")
energies, dos = pet_mad_dos_calculator.calculate_dos(atoms)Predicting the densities of states for every atom in the crystal, or a list of atoms, is also possible:
# Calculating the DOS for every atom in the crystal
energies, dos_per_atom = pet_mad_dos_calculator.calculate_dos(atoms, per_atom=True)
# Calculating the DOS for a list of atoms
atoms_1 = bulk("Si", cubic=True, a=5.43, crystalstructure="diamond")
atoms_2 = bulk("C", cubic=True, a=3.55, crystalstructure="diamond")
energies, dos = pet_mad_dos_calculator.calculate_dos([atoms_1, atoms_2], per_atom=False)Finally, you can use the calculate_bandgap and calculate_efermi methods to
predict the bandgap and Fermi level for the crystal:
bandgap = pet_mad_dos_calculator.calculate_bandgap(atoms)
fermi_level = pet_mad_dos_calculator.calculate_efermi(atoms)You can also re-use the DOS calculated earlier:
bandgap = pet_mad_dos_calculator.calculate_bandgap(atoms, dos=dos)
fermi_level = pet_mad_dos_calculator.calculate_efermi(atoms, dos=dos)This option is also available for a list of ase.Atoms objects:
bandgaps = pet_mad_dos_calculator.calculate_bandgap([atoms_1, atoms_2], dos=dos)
fermi_levels = pet_mad_dos_calculator.calculate_efermi([atoms_1, atoms_2], dos=dos)You can use PET-MAD last-layer features together with a pre-trained sketch-map dimensionality reduction to obtain 2D and 3D representations of a dataset, e.g. to identify structural or chemical motifs. This can be used as a stand-alone feature builder, or combined with the chemiscope viewer to generate an interactive visualization.
import ase.io
import chemiscope
from pet_mad.explore import PETMADFeaturizer
featurizer = PETMADFeaturizer(version="latest")
# Load structures
frames = ase.io.read("dataset.xyz", ":")
# You can just compute features
features = featurizer(frames, None)
# Or create an interactive visualization in a Jupyter notebook
chemiscope.explore(
frames,
featurize=featurizer
)More examples for ASE, i-PI, and LAMMPS are available in the Atomistic Cookbook.
PET-MAD can be fine-tuned using the Metatrain library.
Additional documentation can be found in the metatensor, metatomic and metatrain repositories.
If you use any of the PET-MAD models in your research, please cite the corresponding articles:
@misc{PET-MAD-2025,
title={PET-MAD, a universal interatomic potential for advanced materials modeling},
author={Arslan Mazitov and Filippo Bigi and Matthias Kellner and Paolo Pegolo and Davide Tisi and Guillaume Fraux and Sergey Pozdnyakov and Philip Loche and Michele Ceriotti},
year={2025},
eprint={2503.14118},
archivePrefix={arXiv},
primaryClass={cond-mat.mtrl-sci},
url={https://arxiv.org/abs/2503.14118}
}
@misc{PET-MAD-DOS-2025,
title={A universal machine learning model for the electronic density of states},
author={Wei Bin How and Pol Febrer and Sanggyu Chong and Arslan Mazitov and Filippo Bigi and Matthias Kellner and Sergey Pozdnyakov and Michele Ceriotti},
year={2025},
eprint={2508.17418},
archivePrefix={arXiv},
primaryClass={physics.chem-ph},
url={https://arxiv.org/abs/2508.17418},
}