mcodev31 / libmsym

molecular point group symmetry lib
MIT License
75 stars 32 forks source link

Trouble finding D5 / D5d / D5h #4

Closed ghutchis closed 8 years ago

ghutchis commented 8 years ago

Here are two ferrocenes that seem to give issues finding D5* groups.

https://www.dropbox.com/s/31566td477jk0h8/Fc-D5h.xyz?dl=0 https://www.dropbox.com/s/0iebvlz0bfjccua/Fc-D5v.xyz?dl=0

In general, it seems like the "loose" thresholds we discussed previously aren't definitely better than the "normal" ones.

mcodev31 commented 8 years ago

Well no, the way thresholds are defined does not really mean loose is better. Since this is not an exhaustive search algorithm what are good thresholds does depend a little on what type of molecule it is an generally the tighter the better.

The equivalence set detection algorithm is completely general for any molecule right now (since that was the topic of the project when I wrote it). It is basically a clustering algorithm for symmetry invariant properties, and so obviously a too loose threshold will make fewer clusters. It will be redesigned to handle these cases (and especially abelian groups) much better, but I’m a little pressed for time and the main focus has been to get SALC generation working (which it is!).

But I do agree that those loose thresholds are quite bad. Unfortunately some molecules will required those kind of thresholds, but these will probably give a better result in most cases.

msym_thresholds_t loose_thresholds = { .zero = 0.040, .geometry = 0.040, .angle = 0.016, .equivalence = 0.032, .permutation = 0.050, .eigfact = 1.0e-3, .orthogonalization = 0.1 };

The D5d molecule is quite symmetry broken and I’m not entirely sure how to detect those mirror planes, I’ll need to think about that.

/M

On 11 Nov 2015, at 23:35, Geoff Hutchison notifications@github.com wrote:

Here are two ferrocenes that seem to give issues finding D5* groups.

https://www.dropbox.com/s/31566td477jk0h8/Fc-D5h.xyz?dl=0 https://www.dropbox.com/s/0iebvlz0bfjccua/Fc-D5v.xyz?dl=0

In general, it seems like the "loose" thresholds we discussed previously aren't definitely better than the "normal" ones.

— Reply to this email directly or view it on GitHub.

mcodev31 commented 8 years ago

It turns out that the application of the angle threshold could have been more consistent for symmetric polyherda. The latest checkin should find these easier. I'll close this now.