Available models All integrated pretrained potentials are registered in MLIPEnum. After installation you can inspect which models were loaded successfully:
from mlip_arena.models import MLIPEnum print ( list (MLIPEnum. __members__ .keys())) # e.g. ['MACE-MP(M)', 'CHGNet', 'M3GNet', 'MatterSim', 'ORBv2', 'SevenNet', # 'eqV2(OMat)', 'MACE-MPA', 'eSEN', 'MACE-OFF(M)', 'ANI2x', 'ALIGNN', # 'DeepMD', 'ORB', ...]
Run a single model The helper get_calculator constructs an ASE-compatible calculator for any entry in MLIPEnum, with optional D3 dispersion correction via torch_dftd.
from ase.build import bulk from ase import units from mlip_arena.models import MLIPEnum from mlip_arena.tasks import MD from mlip_arena.tasks.utils import get_calculator atoms = bulk( "Cu" , "fcc" , a = 3.6 ) * ( 5 , 5 , 5 ) result = MD( atoms = atoms, calculator = get_calculator( MLIPEnum[ "MACE-MP(M)" ], calculator_kwargs = {}, dispersion = True , dispersion_kwargs = { "damping" : "bj" , "xc" : "pbe" , "cutoff" : 40.0 * units.Bohr, }, ), ensemble = "nvt" , # "nve", "nvt", or "npt" dynamics = "langevin" , # any ASE Dynamics class name total_time = 1e3 , # ps — 1 ps = 1e3 fs time_step = 2 , # fs )
get_calculator also accepts a raw ASE Calculator instance or a class, so you are not limited to models registered in MLIPEnum.
Loop over all models Iterate over MLIPEnum to run the same task against every installed model:
from mlip_arena.models import MLIPEnum from mlip_arena.tasks import MD from mlip_arena.tasks.utils import get_calculator from ase.build import bulk atoms = bulk( "Cu" , "fcc" , a = 3.6 ) * ( 5 , 5 , 5 ) results = [] for model in MLIPEnum: result = MD( atoms = atoms, calculator = get_calculator(model, calculator_kwargs = {}), ensemble = "nvt" , dynamics = "langevin" , total_time = 1e3 , time_step = 2 , ) results.append(result)
Parallelize with Prefect All tasks in MLIP Arena are decorated with @prefect.task. Call .submit() instead of a direct invocation and wrap everything in a @flow to dispatch tasks concurrently to Prefect workers.
from prefect import flow from ase.build import bulk from mlip_arena.models import MLIPEnum from mlip_arena.tasks import MD from mlip_arena.tasks.utils import get_calculator atoms = bulk( "Cu" , "fcc" , a = 3.6 ) * ( 5 , 5 , 5 ) @flow def benchmark_all_models (): futures = [] for model in MLIPEnum: future = MD .submit( atoms = atoms, calculator = get_calculator(model, calculator_kwargs = {}), ensemble = "nvt" , dynamics = "langevin" , total_time = 1e3 , time_step = 2 , ) futures.append(future) return [f.result( raise_on_failure = False ) for f in futures] if __name__ == "__main__" : results = benchmark_all_models()
Available tasks All tasks live under mlip_arena.tasks and are importable directly:
from mlip_arena.tasks import OPT , EOS , MD , PHONON , NEB , NEB_FROM_ENDPOINTS , ELASTICITY
PHONON requires phonopy to be installed. If phonopy is missing the wildcard import from mlip_arena.tasks import * will log a warning but the other tasks will still load.OPT — Structure optimization Relax atomic positions and/or unit cell using any ASE Optimizer and Filter:
from ase.build import bulk from mlip_arena.tasks import OPT from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum atoms = bulk( "Fe" , "bcc" , a = 2.87 ) result = OPT( atoms = atoms, calculator = get_calculator(MLIPEnum[ "MACE-MP(M)" ], calculator_kwargs = {}), optimizer = "BFGSLineSearch" , # any ASE Optimizer name filter = "FrechetCell" , # relax cell shape and volume criterion = { "fmax" : 0.01 }, # convergence threshold in eV/Å ) # result["atoms"] — relaxed Atoms; result["converged"] — bool; result["steps"] — int
EOS — Equation of state Compute the energy–volume curve and fit a Birch–Murnaghan EOS to extract bulk modulus and equilibrium volume:
from ase.build import bulk from mlip_arena.tasks import EOS from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum atoms = bulk( "Si" , "diamond" , a = 5.43 ) result = EOS( atoms = atoms, calculator = get_calculator(MLIPEnum[ "CHGNet" ], calculator_kwargs = {}), max_abs_strain = 0.1 , # ±10 % volume range npoints = 11 , # number of EV sample points ) # result["K"] — bulk modulus (GPa); result["v0"] — eq. volume (ų); result["e0"] — eq. energy (eV)
MD — Molecular dynamics Flexible NVE / NVT / NPT simulation with temperature and pressure scheduling:
from ase.build import bulk from mlip_arena.tasks import MD from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum atoms = bulk( "Cu" , "fcc" , a = 3.6 ) * ( 4 , 4 , 4 ) result = MD( atoms = atoms, calculator = get_calculator(MLIPEnum[ "MACE-MP(M)" ], calculator_kwargs = {}), ensemble = "nvt" , dynamics = "langevin" , total_time = 10_000 , # fs — 10 ps time_step = 2 , # fs temperature = 600 , # K (scalar or array for annealing) traj_file = "cu_nvt.traj" , traj_interval = 10 , ) # result["atoms"], result["n_steps"], result["runtime"]
PHONON — Phonon calculation Finite-displacement phonon calculation powered by phonopy :
from ase.build import bulk from mlip_arena.tasks import PHONON from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum atoms = bulk( "Al" , "fcc" , a = 4.05 ) result = PHONON( atoms = atoms, calculator = get_calculator(MLIPEnum[ "M3GNet" ], calculator_kwargs = {}), supercell_matrix = [ 2 , 2 , 2 ], t_min = 0 , t_max = 1000 , t_step = 10 , ) # result["phonon"] — a phonopy.Phonopy object
NEB — Nudged elastic band Minimum energy path calculation from a list of pre-built images:
from ase.build import bulk, make_supercell from mlip_arena.tasks import NEB from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum # Provide start, intermediate (empty), and end images images = [initial_atoms] + [initial_atoms.copy() for _ in range ( 5 )] + [final_atoms] result = NEB( images = images, calculator = get_calculator(MLIPEnum[ "SevenNet" ], calculator_kwargs = {}), optimizer = "MDMin" , climb = True , # climbing-image NEB interpolation = "idpp" , # or "linear" criterion = { "fmax" : 0.05 }, ) # result["barrier"], result["images"], result["forcefit"]
NEB_FROM_ENDPOINTS — NEB with automatic interpolation Convenience wrapper that handles image interpolation from just two endpoint structures:
from mlip_arena.tasks import NEB_FROM_ENDPOINTS from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum result = NEB_FROM_ENDPOINTS( start = initial_atoms, end = final_atoms, n_images = 7 , calculator = get_calculator(MLIPEnum[ "SevenNet" ], calculator_kwargs = {}), relax_end_points = True , # relax start/end before NEB interpolation = "idpp" , climb = True , criterion = { "fmax" : 0.05 }, )
ELASTICITY — Elastic tensor Compute the full elastic tensor by applying symmetry-adapted normal and shear strains:
from ase.build import bulk from mlip_arena.tasks import ELASTICITY from mlip_arena.tasks.utils import get_calculator from mlip_arena.models import MLIPEnum import numpy as np atoms = bulk( "W" , "bcc" , a = 3.16 ) result = ELASTICITY( atoms = atoms, calculator = get_calculator(MLIPEnum[ "MACE-MP(M)" ], calculator_kwargs = {}), filter = "FrechetCell" , normal_strains = np.linspace( - 0.01 , 0.01 , 4 ), shear_strains = np.linspace( - 0.06 , 0.06 , 4 ), ) # result contains the elastic tensor and derived moduli
Next steps
Benchmarks View live benchmark results on the MLIP Arena leaderboard.
Contributing models Add your own pretrained potential as an external ASE calculator or a HuggingFace model.
Prefect docs Learn how to write Prefect tasks and flows to orchestrate large-scale benchmarks.
GitHub Source code, issue tracker, and contribution guidelines.