improvements of periodic GW #1776
Conversation
JWilhelm
commented
Dec 5, 2021
JWilhelm
commented
- regularized RI for periodic GW, enables use of large RI basis sets together with the minimum image convention (could be also interesting for periodic RI-Hartree-Fock)
- diagonalization + inversion instead of Cholesky decomposition + inversion
- still to do: k-points for self-energy for computing band structures
Cholesky decomposition
| usage="PERIODIC", & | ||
| CALL keyword_create(keyword, __LOCATION__, name="PERIODIC_CORRECTION", & | ||
| description="If true, the periodic correction scheme is used employing k-points. "// & | ||
| "Method is not recommended to use, use instead PERIODIC_LOW_SCALING which much "// & |
There was a problem hiding this comment.
which is much more accurate
src/input_cp2k_mp2.F
Outdated
| CALL keyword_create( & | ||
| keyword, __LOCATION__, & | ||
| name="MAKE_CHI_POS_DEFINITE", & | ||
| description="If true, makes and eigenvalue decomposition of chi(iw,k) and removes negative "// & |
There was a problem hiding this comment.
Maybe: makes eigenvalue decomposition of chi(iw,k) positive semi-definite by removing negative values.
src/rpa_util.F
Outdated
| ! ELSE | ||
| ! ! we have only one kpoint per group | ||
| ! num_proc_per_kp = para_env%num_pe/nkp | ||
| ! END IF |
There was a problem hiding this comment.
I would rather remove it for the production code.
src/rpa_main.F
Outdated
| ! "GENERAL", virtual, para_env%num_pe, & | ||
| ! kpgeneral=kpgeneral, with_xc_terms=.FALSE.) | ||
| ! | ||
| ! DEALLOCATE(kpgeneral) |
There was a problem hiding this comment.
It is an info for you but not useful for production code.
| IF (.FALSE.) THEN | ||
| IF (ASSOCIATED(kpoints_Sigma)) CALL kpoint_release(kpoints_Sigma) | ||
| IF (ASSOCIATED(kpoints_Sigma_no_xc)) CALL kpoint_release(kpoints_Sigma_no_xc) | ||
| END IF |
|
Maybe, I have overseen it, but have you added a regtest for the regularized RI? |
|
Thank you for reviewing! The description of keywords is now improved. The regularized RI is implicitly tested using the test "regtest-gw-cubic/G0W0_kpoints_from_Gamma.inp" using the default regularization parameter 1E-2. I agree that it could be a good idea to have a separate, explicit test, that is now "regtest-gw-cubic/G0W0_kpoints_from_Gamma_RI_regularization_1E-3.inp". Concerning comments in the code: I understand that for production code, such comments should be avoided. As most of the comments will be anyway gone in the future (after completing band structure calculations from GW), might it be acceptable to leave the three comments (12 lines in total)? |
|
I am in favor of removing these lines because a new CP2K release is planned for the near future. |
|
I hope that it is ok now... I will be happy if you could merge the pull request to see whether the other regtester are ok with the code changes. Thank you! |
3 participants
