Track invalid spacegroups for structure creation

[scott-materials]

Describe the desired feature

Hye Sol and I are doing structure-prediction.toolkit.fixed-composition calculations on Co12 O17. As expected, many space groups do not work for atom placement.

We find that the same space groups are retested, even though they fail:

grep "spacegroup 52" -A 2

                    INFO     Creating structure with spacegroup 52
                    INFO     Generating possible wyckoff combinations for
                             spacegroup 52
                    INFO     Done generating combinations.
                    WARNING  Spacegroup 52 is invalid for Co12 O17
--
                    INFO     Creating structure with spacegroup 52
                    INFO     Generating possible wyckoff combinations for
                             spacegroup 52
                    INFO     Done generating combinations.
                    WARNING  Spacegroup 52 is invalid for Co12 O17
--
                    INFO     Creating structure with spacegroup 52
                    INFO     Generating possible wyckoff combinations for
                             spacegroup 52
                    INFO     Done generating combinations.
                    WARNING  Spacegroup 52 is invalid for Co12 O17

It seems like we could get a big speedup in structure creation if we could remove invalid spacegroups once they've been tested.

Additional context

No response

To-do items

No response

[jacksund]

This was on my to-do list, so thank you for writing an issue for it! Just a heads up though -- this request will be lower priority because I don't have much time to work on the evo search at the moment

As another side note -- so far, I've only tested the simmate evo search up to 20 atoms (in Mg4Si4O12 without any seeds/prototyping). So for >20 atom systems (and especially a highly disordered system like the 12-17 stoich in your case), there's no guarantee that the search will be efficient or even work successfully. As of v0.14.0, I'd stick with this rule of thumb: use simmate for <20atoms and then uspex for >20atoms

[scott-materials]

No worries -- not too urgent.

I do remember the 20 atom limit. I will say that it looks like this is working OK, since this isn't slowing us down too much for what we're doing.

At this size of 29 atoms, structure creation time ranges between 5 seconds and 5 minutes. Since the staged relaxation takes 3 hrs. (on average) per candidate, even 5 minutes is OK.

Prior to trying the 29 atom system, we tried something with about 50 atoms (again using a non-divisible atom count). No structure was successfully created within 10 minutes, so we gave up on that one.

[jacksund]

I'm guessing we'll have to switch to @sionab's "random-walk" creation method for >20 atom systems -- or alternatively switch to the creator I built for xtalopt's method. Either of these will give much more of a speedup than tracking invalid spacegroups