Skip to content

SCM AMS (Amsterdam Modeling Suite)

Status: supported upstream.

The Amsterdam Modeling Suite ships first-class support for AIMNet2 through the MLPotential engine, introduced in AMS2024.1. Two AIMNet2 variants are exposed as named models:

  • AIMNet2-wB97MD3
  • AIMNet2-B973c

Per the SCM documentation, "these are currently the only ML potential models that support charged systems (ions), and that predict atomic charges and dipole moments and that give IR intensities when calculating normal modes" -- a meaningful capability gap over the other backends in the same engine (TorchANI, M3GNet, MACE, NequIP, FAIRChem).

Element coverage: H, B, C, N, O, F, Si, P, S, Cl, As, Se, Br, I. The AMS integration exposes AIMNet2 for aperiodic systems only -- this is a property of AMS's MLPotential wrapper, not of AIMNet2 itself, which supports PBC via the in-tree AIMNet2Calculator.

Model coverage in this engine: AIMNet2-wB97MD3 and AIMNet2-B973c only. The NSE (open-shell) and rxn (reactive) AIMNet2 model families are not currently exposed by AMS.

Minimal AMS input

Task SinglePoint

System
  Atoms
    O 0.0 0.0 0.0
    H 0.0 0.7 0.6
    H 0.0 -0.7 0.6
  End
End

Engine MLPotential
  Model AIMNet2-wB97MD3
EndEngine

The Model keyword is sufficient -- the AIMNet2 backend is activated implicitly when an AIMNet2-prefixed model is selected. Other backends in the same engine (FAIRChem, M3GNet, MACE, NequIP, TorchANI) are selected via Backend directly.

Workflows

Engine MLPotential works with the standard AMS driver tasks: single-point, geometry optimization, transition-state search, vibrational analysis, molecular dynamics. AIMNet2's atomic-charge and dipole predictions feed AMS's normal-mode IR intensity reporting.

See also