Chemistry & ADMET · isayevlab
AIMNet2 ωB97M-D3
Neural network interatomic potential trained at the ωB97M-D3 level of theory for fast molecular simulations.
What AIMNet2 ωB97M-D3 is
AIMNet2 ωB97M-D3 is a neural network interatomic potential developed by the Isayev Lab for atomistic modelling and molecular simulations. It reproduces quantum chemical results at the ωB97M-D3 density functional theory (DFT) level, providing predictions of potential energy, interatomic forces, and partial atomic charges orders of magnitude faster than ab initio calculations.
Conventional DFT is computationally demanding, restricting long timescale molecular dynamics and high-throughput evaluations. By approximating the potential energy surface for neutral, charged, organic, and elemental-organic molecules, this model enables extended molecular simulations, structural optimisations, and property predictions with near-DFT accuracy.
How it works
The model architecture uses the AIMNet2 framework, combining Atomic Environment Vectors with message-passing multi-layer perceptrons over a 5.0 Å interaction cutoff radius. Electrostatics are treated using short-range embedded interactions alongside external long-range Coulomb solvers (DSF or Ewald), and dispersion interactions are handled through an external D3BJ dispersion correction.
The repository provides four ensemble members (ensemble_0 to ensemble_3) distributed in safetensors format. While ensemble member 0 serves as the default predictor, all four members can be evaluated together to estimate prediction uncertainty.
Key facts
- Architecture
- AIMNet2 (Atomic Environment Vectors with message-passing MLPs)
- Level of theory
- ωB97M-D3 with D3BJ dispersion
- Supported elements
- H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I (14 elements)
- Cutoff radius
- 5.0 Å
- Ensemble size
- 4 models
What you can use it for
- Molecular dynamics and geometry optimisation
- Evaluating potential energies, forces, Hessians, and stress tensors for organic and elemental-organic systems in gas-phase or periodic boundary conditions.
- Partial charge assignment
- Predicting per-atom partial charges for neutral and charged small molecules.
- Prediction uncertainty estimation
- Executing inference across all four ensemble members to quantify model variance and identify out-of-distribution structures.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
AIMNet2 ωB97M-D3
Strengths and limits
Good at
- Achieves near-DFT precision for energies, forces, and charges at orders of magnitude lower computational cost than standard DFT.
- Supports 14 elements spanning common organic, halogenated, and main-group elemental-organic chemistry.
- Calculates advanced physical properties including the Hessian matrix, stress tensor, and periodic boundary conditions via an ASE interface.
- Compatible with PyTorch compilation for up to fivefold GPU acceleration.
Watch out for
- Strictly limited to 14 elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, and I), throwing an error if unrecognised elements are present.
- Periodic system simulations require the Python API or ASE interface and cannot be run via the Gradio demonstration interface.
Inputs and outputs
You provide
Molecule (SMILES)
We generate and relax a 3D conformer before the model sees it.
You get back
Energy, maximum force and per-atom partial charges.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use AIMNet2 ωB97M-D3 commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run AIMNet2 ωB97M-D3 without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
Which chemical elements can the model handle?
The model supports 14 elements: hydrogen, boron, carbon, nitrogen, oxygen, fluorine, silicon, phosphorus, sulfur, chlorine, arsenic, selenium, bromine, and iodine.
What properties does the model predict?
It computes total energy in electronvolts, atomic forces in eV/Å, per-atom partial charges, Hessian matrices, and stress tensors.
Can the model handle periodic boundary conditions?
Yes, periodic boundary conditions are supported through the Python API and the Atomic Simulation Environment (ASE) calculator.
How is uncertainty calculated?
Uncertainty can be assessed by loading all four provided ensemble members (ensemble_0 to ensemble_3) and evaluating the variance across their predictions.
How to cite it
Anstine, Dylan and Zubatyuk, Roman and Isayev, Olexandr. AIMNet2: A Neural Network Potential to Meet your Neutral, Charged, Organic, and Elemental-Organic Needs. Chemical Science, 2025. DOI: 10.1039/D4SC08572H
Written from AIMNet2 ωB97M-D3's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.