Area per lipid calculation fails because of `Duplicate` bug in Voronoi [BUG]

[Aleksandr-biochem]

I'm analysing 1 microsecond long simulations of solvated POPC membrane in Martini2.2 and 3.
LiPyPhilic is incorporated into a script called from terminal. Leaflet identification runs successfully, but APL calculation consistently fails for different replicas as in the following example:

Assigning leaflets...
100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10001/10001 [53:50<00:00,  3.10it/s]
Calculating area per lipid...
 54%|██████████████████████████████████████████████████████████████████████████████████████████▌                                                                              | 5362/10001 [39:15<33:14,  2.33it/s]Duplicate: 707 (0.0194327,144.786,0) matches 141 (0.0194397,144.786,0)

Interestingly, calculation runs smoothly for analogous POPC simulations with SIRAH and CHARMM36m force-fileds.
Visual inspection shows no obvious abnormalities in Martini2/3 trajectories.

This seems to be a problem with Freud dependency as there is a related issue: glotzerlab/freud#1008
(which makes sense, as a resolution and numerical precision issue) It might be worth making LiPyPhilic resistant to this issue.

Expected behaviour
Completion of APL calculation in each frame.

Actual behaviour
APL run fails at certain frames of Martini2 and 3

Additional context

• Which version of lipyphilic are you using? 0.11.0 (freud 3.1.0)
• Which version of Python are you using? 3.10.12
• Which OS are you using? Ubuntu 22.04.5 LTS

[p-j-smith]

Thanks for reporting this @Aleksandr-biochem. I think I ran into the same error before and added a workaround to handle overlapping atoms. This was several years ago though so perhaps Freud has changed since then and it's a different issue.

Is there a particular error message you get? Would you be able to share a frame of your trajectory that causes this issue?