Tiny Molecules Cause Issues

[corinwagen]

I've seen some weird behavior with PRISM Pruner on very small systems. Here's my minimal script:

from prism_pruner.conformer_ensemble import ConformerEnsemble
from prism_pruner.pruner import prune

ensemble = ConformerEnsemble.from_xyz("h2.xyz")

pruned, mask = prune(
    ensemble.coords,
    ensemble.atoms,
    rot_corr_rmsd_pruning=True,
    debugfunction=print,
)

I made some trivial H2 and Ne ensembles for testing purposes:

H2 ensemble

h2.xyz

2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150
2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150
2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150
2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150
2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150
2
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
H     0.03753493   1.40368284  -0.00029150
H     0.65753493   1.40368284  -0.00029150

Ne ensemble

ne.xyz

1
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
Ne    0.03753493   1.40368284  -0.00029150
1
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
Ne    0.03753493   1.40368284  -0.00029150
1
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
Ne    0.03753493   1.40368284  -0.00029150
1
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
Ne    0.03753493   1.40368284  -0.00029150
1
charge: 0; multiplicity: 1; generated_by: Rowan; timestamp: 2026-03-17 13:34:38;
Ne    0.03753493   1.40368284  -0.00029150

When I run these, I get:

$ python h2.py
DEBUG: MOIPrunerConfig - Used cached data 0/5 times, 0.00% of total calls

prism_pruner/prism_pruner/pruner.py:513: RuntimeWarning: Mean of empty slice.
  centered = heavy - heavy.mean(axis=1, keepdims=True)  # (N, M, 3)
prism_pruner/.pixi/envs/dev/lib/python3.14/site-packages/numpy/_core/_methods.py:136: RuntimeWarning: invalid value encountered in divide
  ret = um.true_divide(
prism_pruner/prism_pruner/pruner.py:525: RuntimeWarning: invalid value encountered in divide
  rmsd_sq = (norm_sq[:, None] + norm_sq[None, :] - 2.0 * adjusted_sum_S) / M
DEBUG: RMSDPrunerConfig - rejected 0 (keeping 1/1), in 0.0 s

DEBUG: prune_by_rmsd_rot_corr - No subsymmetrical torsions found: skipping symmetry-corrected RMSD pruning

$ python ne.py
prism_pruner/pruner.py:199: RuntimeWarning: invalid value encountered in scalar divide
  if np.abs(im_1[j] - im_2[j]) / im_1[j] >= self.max_dev:
DEBUG: MOIPrunerConfig - k=1, rejected 4 (keeping 1/5), in 0.0 s
DEBUG: MOIPrunerConfig - keeping 1/5 (0.0 s)
DEBUG: MOIPrunerConfig - Used cached data 0/4 times, 0.00% of total calls

DEBUG: RMSDPrunerConfig - rejected 0 (keeping 1/1), in 0.0 s

DEBUG: prune_by_rmsd_rot_corr - No subsymmetrical torsions found: skipping symmetry-corrected RMSD pruning

It's not the end of the world, but causing a bit of messiness in the logs as we make PRISM Pruner part of scaled workflow computation here. Do you think it'd be worth adding some checks for cases like these?