AIMNet2 Backend
===============

AIMNet2 is a machine-learning-potential backend available in PyAR. It is useful
when you want fast molecular energy and force evaluations without calling an
external quantum-chemistry executable.

When to use AIMNet2
-------------------

Use ``aimnet_2`` when you want:

* fast optimisation of molecular structures
* a machine-learning-potential route for aggregation, solvation, or reaction
  exploration
* AFIR/geomeTRIC reaction searches without relying on Turbomole coordinate
  updates

Example commands
----------------

Molecular cluster search:

.. code-block:: bash

   pyar-cli -s water.xyz water.xyz --software aimnet_2 -ss 10 -N 16 -c 0 0 -m 1 1

Reaction search:

.. code-block:: bash

   pyar-cli react A.xyz B.xyz -N 8 -gmin 100 -gmax 1000 --software aimnet_2

Practical notes
---------------

* PyAR bundles AIMNet2 model assets for its AIMNet2 interfaces.
* AIMNet2 is a third-party project from the Isayev Lab and is MIT licensed
  upstream. See ``THIRD_PARTY_LICENSES/AIMNet2-LICENSE`` and
  ``THIRD_PARTY_LICENSES/AIMNet2-PROVENANCE.md``.
* As with all ML potentials, use AIMNet2 aggressively for exploration but
  validate important structures and energetics with an appropriate higher-level
  method.

AFIR/geomeTRIC status
---------------------

``aimnet_2`` is registered as an energy-gradient provider for the
geomeTRIC-backed AFIR reaction route.