MLatom is an open-source package for atomistic simulations with machine learning and quantum chemical methods — DFT, wavefunction-based, and semi-empirical approximations. Use it as a Python library, through input files, or from the command line — run it locally, or online with no installation.
Website · Documentation · Protomia · Aitomistic Hub · Aitomistic Lab@XMU · Releases
Run MLatom in your browser on either online platform — both powered by Protomia, with an AI assistant for autonomous atomistic simulations:
- Aitomistic Hub — registration-free.
- Aitomistic Lab@XMU — free for academic users (registration with an academic email).
The easiest way to run MLatom is online — no installation. To install locally:
python3 -m pip install -U mlatomThat pulls in the required dependencies automatically (including the PyTorch/TorchANI and geometry-optimization backends the AIQM2 quick start uses). AIQM2 additionally needs the DFT-D4 program, installed via conda:
conda install -c conda-forge dftd4
export dftd4bin=$(which dftd4) # point MLatom at the dftd4 executableSee the installation guide for the full dependency list and other methods.
Optimize the geometry of a water molecule with AIQM2 — an AI-enhanced quantum-mechanical method (native to MLatom, CHNO elements) that reaches beyond-DFT accuracy at semi-empirical cost:
import mlatom as ml
mol = ml.data.molecule.from_xyz_string('''3
O 0.00000 0.00000 0.11779
H 0.00000 0.75545 -0.47116
H 0.00000 -0.75545 -0.47116
''')
aiqm2 = ml.methods(method='AIQM2')
opt = ml.optimize_geometry(model=aiqm2, initial_molecule=mol).optimized_molecule
print(opt.energy) # optimized energy in hartree (≈ -76.3838)The same calculation as an input file (geomopt.inp, with init.xyz holding the geometry):
AIQM2 # method
geomopt # task: geometry optimization
xyzfile=init.xyz # input geometry
optxyz=opt.xyz # output geometry
mlatom geomopt.inpPrefer zero setup? Run these online on the Aitomistic Hub or Aitomistic Lab@XMU (both powered by Protomia) — no installation needed.
Full manuals and tutorials are at mlatom.com/docs — begin with installation and get started.
- Methods — universal ML potentials (ANI, AIMNet2); AI-enhanced QM methods (UAIQM, AIQM1, AIQM2) approaching coupled-cluster accuracy at semi-empirical cost; and DFT, semi-empirical, and wavefunction methods via interfaces (PySCF, Gaussian, ORCA, xtb, MNDO, Turbomole, DFTB+, Sparrow, Columbus).
- ML models — train and use ML potentials (KREG, GAP-SOAP, ANI, MACE, and more) with active learning, Δ-learning, transfer learning, and self-correction.
- Simulations — geometry optimization, transition-state search, IRC, frequencies & thermochemistry, molecular dynamics, nonadiabatic dynamics, IR/Raman/UV–vis spectra, and periodic boundary conditions.
- Workflows — compose methods and tasks into complex pipelines from Python, input files, or the command line.
If you use MLatom in scientific work, please cite:
Pavlo O. Dral et al. MLatom 3: A Platform for Machine Learning-Enhanced Computational Chemistry Simulations and Workflows. J. Chem. Theory Comput. 2024, 20, 1193–1213. DOI: 10.1021/acs.jctc.3c01203
Feature-specific references appear in the program output and in CITATION.cff. The full list with BibTeX is on the License and citations page.
- MLatom Skills — agent skills for MLatom (new and growing rapidly).
- Aitomic add-ons — cutting-edge methods (e.g. AIQM3), free for academic, non-commercial use.
- Aitomia — an AI assistant for autonomous atomistic simulations with MLatom.
Full release notes · CHANGELOG.md
- 3.23 — AIQM3 public add-on (
pip install aitomic-addons); direct Gaussian workflows; version/commit/build-date banner with a daily update check. - 3.22 — OMNI-P2x model; faster nonadiabatic dynamics.
- 3.21 — refactored ORCA interface supporting many more excited-state methods.
- 3.18 — FSSH; MDtrajNet-1 (directly predicting MD trajectories); KRR in Julia; ECTS diffusion model for transition states.
Contributions are welcome — see CONTRIBUTING.md and the Code of Conduct. Report bugs and request features via GitHub issues. You may also create your own private derivatives by following the license requirements.
Don't miss MLatom updates — subscribe to the Aitomistic email newsletter and social channels.
MLatom was founded by Pavlo O. Dral on 10 September 2013, who continues to lead its development. It is open-source under the Apache License 2.0 and supported by Aitomistic.