PBE0 geometric optimization not converging

[Lory Wang]

Dear CP2K experts,

I am currently experiencing difficulty in getting a geometric optimization test with PBE0 to converge. I am starting with a PBE relaxed structure, and the relaxed energy from PBE is approximately -212. From the output file (cp2k_full_all.out), it seems like the very first SCF step is in the correct energy range. However, the OT method then switches from DIIS to SD, and the energy diverges significantly.

     1 OT DIIS     0.15E+00  368.3     0.00789918      -212.0944035335 -2.12E+02
     2 OT SD       0.15E+00   55.4    54.87357862     -3785.7999788923 -3.57E+03
     3 OT SD       0.15E+00   55.1    78.83523027    -16117.9941684959 -1.23E+04
     4 OT DIIS     0.15E+00   54.8    36.24089744     -3042.2730297100  1.31E+04
     5 OT SD       0.15E+00   55.1    25.41660926      -813.4054260626  2.23E+03
     6 OT SD       0.15E+00   55.1    35.73628016    -14638.4364940160 -1.38E+04

I also tried different preconditioners and cutoff radii. I found that FULL_SINGLE_INVERSE and a cutoff radius of 6 work better (see cp2k_full_inverse.out), but I am still seeing energy spikes in the SCF cycles.
  ----------------------------------- OT ---------------------------------------

  Step     Update method      Time    Convergence         Total energy    Change
  ------------------------------------------------------------------------------
     1 OT DIIS     0.80E-01  113.1     0.00645985      -213.0167602266 -2.97E-01
     2 OT SD       0.80E-01   56.2     0.01323989      -213.9726833855 -9.56E-01
     3 OT SD       0.80E-01   56.2     0.05110050      -219.8677669985 -5.90E+00
     4 OT SD       0.80E-01   56.1     1.50838863      -526.6197195349 -3.07E+02
     5 OT SD       0.80E-01   55.9     6.44077824     -3560.7235910833 -3.03E+03
     6 OT DIIS     0.80E-01   55.9     6.93627366        -5.1310569943  3.56E+03
     7 OT SD       0.80E-01   56.3     0.00448314      -212.7274234258 -2.08E+02

I have attached the full input and output files for reference. Could someone help me identify what I might be missing?

Thank you very much for your time and assistance.

Best regards,

Lory

[Frederick Stein]

Dear Lory,

Regarding the convergence:

- you should replace your pseudopotential GTH-PBE-q4 with GTH-PBE0-q4 (there is one in the file POTENTIAL_UZH).

- are you sure about the low ENERGY_GAP parameter? Such a low value may decrease the convergence. the default (0.02 Eh should work well enough). Simply comment this line out.

- please double-check the convergence of the parameters EPS_DEFAULT, EPS_SCHWARZ, CUTOFF. Especially EPS_SCHWARZ has to be chosen tighter (lower) if ADMM is requested. Also check within ENERGY_FORCE (not GEO_OPT) runs whether forces and stress tensors are well converged. You can reuse the thereby obtained wavefunctions as an initial guess for your actual geometry optimization

Performance-wise:

- EPS_PGF_ORB need not be chosen that tight the default value is usually good enough. You rarely need values lower than 1.0E-8 . Simply comment it out.

- MOLOPT basis sets are usually not a good choice for Hartree-Fock calculations. Try the correlation-consistent basis sets in BASIS_ccGRB_UZH or the MOLOPT-PBE0-GTH-q4 basis sets in BASIS_MOLOPT_UZH.

HTH,

Frederick