Chemistry & ADMET · isayevlab
AIMNet2-NSE
A neural network interatomic potential for open-shell chemistry, radicals and non-singlet molecular states.
What AIMNet2-NSE is
AIMNet2-NSE is a spin-polarised neural network interatomic potential developed by the Isayev Lab for simulating open-shell molecular systems. Standard interatomic potentials typically target closed-shell, singlet-state molecules, making them unsuitable for systems with unpaired electrons. AIMNet2-NSE accepts total charge and spin multiplicity as explicit inputs, enabling simulations of radicals, triplet states, excited states and transition states.
The model achieves near-density functional theory (DFT) accuracy for energies, atomic forces and partial charges, whilst running orders of magnitude faster than conventional quantum chemical methods. It also supports calculating properties such as the molecular Hessian, stress tensor and simulations under periodic boundary conditions across 14 chemical elements.
How it works
AIMNet2-NSE is based on the AIMNet2 architecture, which combines atomic environment vectors with message-passing multilayer perceptrons using a 5.0 Å local cutoff radius. Electrostatic interactions are captured using embedded short-range Coulomb interactions paired with external long-range DSF or Ewald summations, while dispersion corrections rely on an externalised D3BJ scheme handled via the calculator interface.
The model evaluates molecular geometries based on atomic coordinates, nuclear charges, molecular charge and spin multiplicity against reference calculations at the wB97M DFT functional level. An ensemble of four distinct parameter sets is provided, allowing predictions to be averaged to estimate model uncertainty.
Key facts
- Architecture
- AIMNet2 (Atomic Environment Vectors with message-passing MLPs)
- Supported elements
- 14 elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I)
- Reference DFT functional
- wB97M with D3BJ dispersion correction
- Local interaction cutoff
- 5.0 Å
- Ensemble size
- 4 members (ensemble_0 to ensemble_3)
- Supported modalities
- Small molecules, periodic systems, radicals
What you can use it for
- Radical and open-shell simulation
- Compute energies, forces and partial charges for neutral or charged radicals and open-shell species by specifying the spin multiplicity.
- Molecular dynamics of reactive intermediates
- Drive atomic simulations of transient, non-singlet intermediates and transition states at computational costs far below density functional theory.
- Periodic material and condensed-phase modelling
- Model extended or crystal systems containing the 14 supported elements using periodic boundary conditions via the Python or ASE interfaces.
- Uncertainty quantification in potential energy surfaces
- Evaluate variance across all four ensemble members to quantify predictive uncertainty in energies and atomic forces.
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-NSE
Strengths and limits
Good at
- Delivers near-DFT accuracy at speeds orders of magnitude faster than conventional quantum mechanics calculations.
- Provides native support for open-shell systems such as radicals and triplet states via explicit spin multiplicity inputs.
- Supplies an ensemble of four model checkpoints for direct uncertainty estimation.
- Supports performance acceleration through torch.compile, providing roughly a fivefold speedup on GPUs.
- Computes full analytical forces, partial charges, Hessians and stress tensors.
Watch out for
- Restricted to 14 elements (H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I), raising an error if unparameterised elements are present.
- Periodic system calculations are accessible only via the Python API and ASE interface, not through the Gradio 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-NSE 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-NSE 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.
What inputs does AIMNet2-NSE require?
Calculations require atomic Cartesian coordinates, atomic numbers, total molecular charge and spin multiplicity (mult). SMILES strings can also be supplied through supported pipeline runners.
Which elements are supported by the potential?
The potential supports 14 elements: hydrogen, boron, carbon, nitrogen, oxygen, fluorine, silicon, phosphorus, sulphur, chlorine, arsenic, selenium, bromine and iodine.
How does the model handle open-shell configurations?
It is a spin-polarised model that takes the spin multiplicity parameter alongside the total charge, allowing it to distinguish between closed-shell configurations and open-shell species such as doublets and triplets.
How can uncertainty be estimated?
The model distribution includes four separate ensemble parameter files (ensemble_0 through ensemble_3). Researchers can run all four members and evaluate the variance across predictions.
What units are used for the primary outputs?
Potential energy is output in electronvolts (eV), atomic forces in eV/Å and atomic partial charges in elementary charge units (e).
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-NSE'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.