LDIAG=False has not effect for ionic relaxation

[martin.schlipf]

I read the whole thread again and I think I understand now that you issue is with the PROCAR file. The 124th band loses its p character.

Code: Select all

grep -A3 'band   124' step*/PROCAR
step1/PROCAR:band   124 # energy    7.93852113 # occ.  1.00000000
step1/PROCAR-
step1/PROCAR-ion      s     py     pz     px    dxy    dyz    dz2    dxz  x2-y2    tot
step1/PROCAR-    1  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000
--
step1/PROCAR:band   124 # energy    8.46042572 # occ.  1.00000000
step1/PROCAR-
step1/PROCAR-ion      s     py     pz     px    dxy    dyz    dz2    dxz  x2-y2    tot
step1/PROCAR-    1  0.000  0.007  0.033  0.071  0.000  0.000  0.000  0.000  0.000  0.112
--
step2/PROCAR:band   124 # energy    8.35380209 # occ.  1.00000000
step2/PROCAR-
step2/PROCAR-ion      s     py     pz     px    dxy    dyz    dz2    dxz  x2-y2    tot
step2/PROCAR-    1  0.000  0.050  0.000  0.051  0.000  0.000  0.000  0.000  0.000  0.100
--
step2/PROCAR:band   124 # energy    9.73363336 # occ.  0.00000000
step2/PROCAR-
step2/PROCAR-ion      s     py     pz     px    dxy    dyz    dz2    dxz  x2-y2    tot
step2/PROCAR-    1  0.006  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.000  0.006

I tested this again by running just one iteration in step 2 and found that the orbital character is still the same. So something goes wrong when proceeding to the next step in the ionic update.

[martin.schlipf]

Another update: The root of this issue seems an if statement labeled pre_subrot in main.F (line 4719 in VASP.6.4). You will see that this calls the diagonalization (EDDIAG). We will need to test what the proper path to transverse this code path is. You can explore whether skipping this part or the diagonalization works for you if you do not want to wait for an official solution.

[yihuang_xiong]

Sorry for the late response. Thanks a lot! I will test this block in main.F.
Looking forward to the official solution. Thank you!