aimnet2 updates; additional loss options; multi cutoff neighborlists

[chrisiacovella]

Pull Request Summary

The PR is primarily focused on getting all the odds and ends together for proper utilization of aimnet2.

The changes are outlined below:

Key changes

Notable points that this PR has either accomplished or will accomplish.

• Aimnet2 returns predicted partial charges
• Add in post processing operation to allow summation of different energy contributions
• Adds partial charges into the available properties
• Partial charges can now be computed in loss functions
• Multi- cutoff neighbor lists have been reimplemented, now compatible with torch script
• Documentation for how to properly use AimNet2
• Add in DFTD3 as a post processing operation

Associated Issue(s)

• per-atom-charge and per-atom-spin loss #359 - Note per_atom_spin predictions/losses haven't been added in this; those will be explored separately
• Aimnet2 implementation #360

Pull Request Checklist

• Issue(s) raised/addressed and linked
• Includes appropriate unit test(s)
• Appropriate docstring(s) added/updated
• Appropriate .rst doc file(s) added/updated
• PR is ready for review

[chrisiacovella]

I'll note this is still partially a WIP. I still need to implement some CI tests for the multi-cutoff neighbor lists and add in the DFTD3 post-processing routine (although I might end up skipping that for this PR and do that in a separate one).

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.98606% with 23 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.98%. Comparing base (6d3272c) to head (fb2a8da).
⚠️ Report is 271 commits behind head on main.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull Request Overview

This PR integrates per-atom partial charge support, extends loss functions to include charge errors, and refactors neighborlist routines to support multiple cutoffs alongside adding DFTD3-based dispersion and energy‐summation postprocessing.

Key changes:

• Added partial_charge flow through PropertyNames, Metadata, BatchData, training, and Loss.
• Refactored NeighborListFor* to return PairlistOutputs with local, vdw, and electrostatic cutoffs.
• Introduced DispersionPotential, SumPerSystemEnergies, and updated PostProcessing for new per-system energy terms.

Reviewed Changes

Copilot reviewed 22 out of 23 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
modelforge/utils/prop.py	Added partial_charge to PropertyNames and Metadata.
modelforge/train/training.py	Updated CalculateProperties to handle per-atom and total charges; renamed helper methods.
modelforge/train/losses.py	Added PerAtomChargeError and updated Loss to include per_atom_charge.
modelforge/potential/processing.py	Added SumPerSystemEnergies, DispersionPotential, and multi-cutoff postprocessing; contains a key‐naming bug.
modelforge/potential/neighbors.py	Refactored neighborlist classes to return PairlistOutputs; updated internal logic for multiple cutoffs.
modelforge/tests	Updated tests for .local_cutoff API and added test_per_atom_charge_loss.

Comments suppressed due to low confidence (2)

modelforge/potential/processing.py:435

• The key electrostatic_r_ij] contains an extra bracket and won't match downstream usage (electrostatic_r_ij). Remove the trailing ] inside the string.

        Returns

modelforge/potential/processing.py:336

• The forward method in SumPerSystemEnergies is defined inside __init__, making it a local function rather than a class method. Move def forward(...) out of __init__ and indent it at the class level.

        def forward(self, data: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]: