Fix Wannier localization when using LOW SPIN ROKS #3108
Conversation
- Added the ability to calculate Wannier Сenters for the ROKS method (only LOCALIZE NONE and JACOBI). - Added the error message when ROKS and LOCALIZE were used incorrectly (instead of Segmentation Fault). - Replaced the output of two spins in ROKS by the output of spin 1 only, since the densities of spin_1/2 are the same. I have previously raised this issue in the Google Group discussion https://groups.google.com/g/cp2k/c/yJPJS2vUkBI This Wannier centres analysis is added in analogy to the implementation in CPMD code. See https://github.com/CPMD-code/CPMD/blob/main/src/ddipo_utils.mod.F90, The fragment “ELSEIF (lspin2%tlse) THEN…” means that for singly filled orbitals the Jacobi rotation is disabled. In this case, the number of orbitals to which the Jacobi rotation can be applied has been reduced by 2 since SOMO-1 and SOMO-2 orbitals should not rotate.
src/qs_localization_methods.F
Outdated
| @@ -365,6 +370,11 @@ SUBROUTINE jacobi_rotations_serial(weights, zij, vectors, max_iter, eps_localiza | |||
| CALL cp_fm_get_info(rmat, nrow_global=nstate) | |||
| tolerance = 1.0e10_dp | |||
|
|
|||
| IF (restricted > 0) THEN | |||
| WRITE (6, *) "JACOBI: The last ", restricted, " orbitals do not rotate" | |||
There was a problem hiding this comment.
Note CP2K's coding convention #c012 concerning hardwired unit numbers.
There was a problem hiding this comment.
Sorry, I missed it. Replaced it in the code with cp_logger_get_default_unit_nr().
src/qs_localization_methods.F
Outdated
| @@ -2208,6 +2220,13 @@ SUBROUTINE jacobi_rot_para(weights, zij, vectors, para_env, max_iter, eps_locali | |||
|
|
|||
| CALL cp_fm_get_info(rmat, nrow_global=nstate) | |||
|
|
|||
| IF (restricted > 0) THEN | |||
| IF (output_unit > 0) THEN | |||
| WRITE (6, *) "JACOBI: The last ", restricted, " orbitals do not rotate" | |||
There was a problem hiding this comment.
Sorry, I missed it. Replaced it in the code with cp_logger_get_default_io_unit().
|
Thanks for your contribution. Could you also add a regression test input for that new feature? |
|
I added two tests to QS/regtest-lsroks The first test checks the output of localized Wannier Centers for ROKS. It would be more correct to check the centers themselves, but as far as I understand, the regression tests do not support this and I added a check for “Total Spread (Berry)”. The second test checks the MO_CUBES output for ROKS. It would be more correct to check that it outputs cube files only for spin-1, but as I understand it, this is also not supported, so I added a check that the program worked through "ENERGY| Total FORCE_EVAL" and a comment to the input. |
|
I'm sorry, could you explain what's wrong with Regtest sdbg, it says that all tests were successful... |
Just one slow test was detected. That can happen sometimes without a serious reason. I requested to re-run the sdbg test. |
Thank you. |
I have previously raised this issue in the Google Group discussion https://groups.google.com/g/cp2k/c/yJPJS2vUkBI
This Wannier centres analysis is added in analogy to the implementation in CPMD code. See https://github.com/CPMD-code/CPMD/blob/main/src/ddipo_utils.mod.F90, The fragment “ELSEIF (lspin2%tlse) THEN…” means that for singly occupied orbitals the Jacobi rotation is disabled. In this case, the number of orbitals to which the Jacobi rotation can be applied has been reduced by 2 since SOMO-1 and SOMO-2 orbitals should not rotate.