results

[ferroteam]

Hello,
I performed a crystal search using RS method (1000 structures). There are 5 stages in the process. For some structures the search is completed (i.e., all 5 stages are done) and when I look into cryspy_rslt_energy_asc file I see the following:

 Spg_num     Spg_sym  Spg_num_opt Spg_sym_opt  E_eV_atom  Magmom      Opt

668 92 P4_12_12 92 P4_12_12 -3.187351 NaN not_yet
646 62 Pnma 62 Pnma -3.187223 NaN not_yet
586 19 P2_12_12_1 19 P2_12_12_1 -3.185611 NaN not_yet
390 92 P4_12_12 92 P4_12_12 -3.184457 NaN not_yet
682 56 Pccn 56 Pccn -3.183607 NaN not_yet
809 110 I4_1cd 110 I4_1cd -3.181608 NaN not_yet
661 229 Im-3m 221 Pm-3m -3.176966 NaN not_yet
941 229 Im-3m 221 Pm-3m -3.176839 NaN not_yet
381 173 P6_3 173 P6_3 -3.175062 NaN not_yet
841 64 Cmce 64 Cmce -3.174583 NaN not_yet
481 106 P4_2bc 106 P4_2bc -3.173147 NaN not_yet
200 161 R3c 161 R3c -3.173107 NaN not_yet
777 106 P4_2bc 106 P4_2bc -3.172855 NaN not_yet
......

I'm confused by the last column which says "not_yet" for Opt status. I remember to see in CrysPY_examples that it should read "done" if it is over. So, does it mean my calculations still not completed?

Thank you.

[Tomoki-YAMASHITA]

When you use the VASP code, CrySPY looks for "reached required accuracy" from OUTCAR in the final stage.
If it is found, CrySPY shows 'done' in cryspy_rslt. This is just a rough guide. If you do not perform structure optimization in the final stage, "reached required accuracy" does not appear. So it depends on INCAR_5.

[USFmatscilab]

Thank you, Tomoki!
Yes, I use VASP code and last stage is just SCF calculation.

I was wondering, what is acceptable number of trial structures for RS method? Is 1000 enough? Or should I stick to EA or BO or LAQA, where only 1 stage is required?

Thanks!

[Tomoki-YAMASHITA]

Acceptable number of trial structures depends on the system; number of elements and atoms, potential energy surface, and so on.
For example, simple ionic crystals such as Na8Cl8 is easy, but Si16 (covalent) is more difficult.

If you cannot find stable structure with 1000 or 2000 trials in RS, better to use EA.
In EA and BO, we usually 2 or 3 stages like RS. LAQA is a slightly special method, so the stage is fixed at 1.