Chemistry & ADMET · isayevlab
AIMNet2 (B97-3c, 2025)
A neural network interatomic potential calculating molecular energies, forces, and charges at the B97-3c level.
What AIMNet2 (B97-3c, 2025) is
AIMNet2 (B97-3c, 2025) is a neural network interatomic potential developed by the Isayev Lab for fast molecular simulations. It predicts quantum chemical properties—including total potential energies, atomic forces, and partial charges—at speeds orders of magnitude faster than conventional density functional theory (DFT). This 2025 release updates and supersedes the earlier aimnet2-b973c model, offering improved treatment of intermolecular interactions and broader parameterisation.
The model is suited for chemical systems where non-covalent and intermolecular forces dominate, such as molecular clusters, solution-phase solvation environments, and organic crystal packing. It supports calculations for 14 chemical elements across neutral and charged organic or elemental-organic molecules, and includes support for periodic boundary conditions.
How it works
The model combines Atomic Environment Vectors (representations of local atomic neighbourhoods) with message-passing multilayer perceptrons using a 5.0 Å radial cutoff. It models electrostatic effects via an embedded short-range Coulomb formulation paired with an externalised long-range treatment using damped shifted force (DSF) or Ewald summation. Dispersion corrections are calculated via an externalised D3BJ term handled directly by the simulation calculator.
Trained against quantum mechanical reference data at the B97-3c level of theory, the model comprises four separate ensemble members. These ensemble weights can be queried individually or evaluated together to average predictions and derive uncertainty estimates for computed energies and atomic forces.
Key facts
- Model architecture
- AIMNet2 (Atomic Environment Vectors + message-passing MLPs)
- Level of theory
- B97-3c
- Supported elements
- H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I (14 elements)
- Radial cutoff
- 5.0 Å
- Ensemble size
- 4 members
- Dispersion correction
- D3BJ (externalised)
What you can use it for
- Solvation and Solution Modelling
- Simulating solvent-solute interactions and liquid mixtures where intermolecular hydrogen bonding and dispersion forces govern behaviour.
- Molecular Cluster and Crystal Packing Studies
- Evaluating potential energies, forces, stress tensors, and lattice arrangements in periodic crystal structures and non-covalent aggregates.
- Molecular Dynamics and Geometry Optimisation
- Driving atomic trajectories and structural minimisations within the Atomic Simulation Environment (ASE) with near-DFT accuracy.
- Uncertainty Quantification
- Evaluating predictions across all four ensemble members to quantify model variance on unfamiliar chemical conformers.
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 (B97-3c, 2025)
Strengths and limits
Good at
- Performs orders of magnitude faster than conventional density functional theory, with up to roughly 5x additional acceleration when compiled using torch.compile on a GPU.
- Provides a broad set of physical outputs including energies, atomic forces, partial charges, Hessians, and stress tensors.
- Includes four pre-trained ensemble members that allow straightforward calculation of model uncertainty.
- Supports periodic boundary conditions for condensed-phase systems via the Python and ASE interfaces.
Watch out for
- Restricted strictly to 14 elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I); any other element will trigger an error.
- Periodic system calculations are only available through the Python API and ASE interface rather than the interactive demo.
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 (B97-3c, 2025) 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 (B97-3c, 2025) 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 elements does this model support?
The model supports 14 elements: hydrogen, boron, carbon, nitrogen, oxygen, fluorine, silicon, phosphorus, sulfur, chlorine, arsenic, selenium, bromine, and iodine.
How are dispersion and electrostatics handled?
Dispersion is addressed using an externalised D3BJ correction managed by the calculator. Electrostatics combine an embedded short-range Coulomb network with external long-range DSF or Ewald summation.
How can uncertainty be estimated from predictions?
The repository includes four ensemble members (ensemble_0 through ensemble_3). Researchers can load all four checkpoints and compute the variance or standard deviation across their predictions.
Can this potential be used in ASE?
Yes, the package includes an AIMNet2ASE calculator that directly interfaces with the Atomic Simulation Environment for molecular dynamics, relaxations, and periodic boundary calculations.
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 (B97-3c, 2025)'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.