AIMNet2

AIMNet2 is a neural network potential that provides broad element coverage (14 elements) with ωB97M-D3BJ accuracy. It is a molecular, no-PBC backend that supports charge and multiplicity for charged or open-shell molecules.

Variants

Model	Keyword	Elements	Description
AIMNet2	`aimnet2`	H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I	Standard variant with ωB97M-D3BJ accuracy
AIMNet2-NSE	`aimnet2nse`	H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I	Open-shell variant using neural spin-charge equilibration for systems with arbitrary charge and spin multiplicity, including radicals.

Charge and Spin Support

AIMNet2 natively handles charged species and different spin multiplicities. Specify charge and multiplicity in the coordinate block:

#model=aimnet2
#opt(method=lbfgs)

-1 1
C    0.000000    0.000000    0.000000
O    1.250000    0.000000    0.000000
O   -1.250000    0.000000    0.000000

The first line specifies charge multiplicity (here: charge = −1, multiplicity = 1 for a formate anion).

Energy-only Implicit Solvation

AIMNet2 can use MAPLE's experimental implicit GB-polar/QEq correction for single-point energies. Add experimental=true and use #sp; solvent forces, optimizations, Hessians, and frequency workflows are not supported for this correction.

#model=aimnet2
#solv(method=gbsa, implicit=water, experimental=true)
#sp

-1 1
C    0.000000    0.000000    0.000000
O    1.250000    0.000000    0.000000
O   -1.250000    0.000000    0.000000

Calculator Capabilities

Hessian: analytic and numerical Hessians are available in gas phase. Implicit-solvent Hessians (analytic or numerical) are not supported.
PBC: AIMNet2 is a no-PBC wrapper in MAPLE. Periodic atoms fail fast, and coulomb_method=ewald is disabled until validated cell/PBC/MIC support exists.
HVP: AIMNet2 does not implement get_hvp(); use non-HVP transition-state workflows such as P-RFO, NEB, or String instead.
Model files: MAPLE can auto-download AIMNet2 checkpoints from the MAPLE model repository, or you can provide an explicit model_path. Supported Coulomb options are simple and dsf.

Usage Examples

Optimization with heavy elements (Br, I)

#model=aimnet2
#device=gpu0
#opt(method=lbfgs)

0 1
C    0.000000    0.000000    0.000000
Br   1.940000    0.000000    0.000000
H   -0.390000    0.000000    1.028000
H   -0.390000    0.890000   -0.514000
H   -0.390000   -0.890000   -0.514000

NSE variant using neural spin-charge equilibration

#model=aimnet2nse
#device=gpu0
#opt(method=lbfgs)

0 1
C    0.000000    0.000000    0.000000
I    2.140000    0.000000    0.000000
H   -0.390000    0.000000    1.028000
H   -0.390000    0.890000   -0.514000
H   -0.390000   -0.890000   -0.514000

Limitations

No D4 dispersion. The ωB97M-D3BJ training already includes dispersion corrections.
No periodic systems. AIMNet2 does not support periodic boundary conditions.
14 elements only. Systems with elements outside the supported set (e.g., transition metals) require uma instead.