issues regarding PBE0 calculation

[Lorenzo Lagasco]

Good afternoon cp2k developers,
Regarding the setup of my input for the PDOS calculation (with the computation of inter-electronic couplings through diabatization), I encountered an unusual convergence issue. As shown in the attached output file, the total energy appears to "diverge" at each SCF iteration.

To address this problem, I decided to modify the restart wavefunction by using a file generated with the PBE+U method, instead of one obtained from a standard PBE calculation, as done in a previous test (whose input file is also attached). However, the convergence issue still persists in the following calculation, even after including the Hubbard correction.

    1 OT DIIS     0.80E-01 1116.7     0.00070560     -7523.8402746160 -7.52E+03
     2 OT DIIS     0.80E-01  845.8     0.84880918     -7399.9632959908  1.24E+02
     3 OT DIIS     0.80E-01  727.7     0.01083311     -7523.3761323117 -1.23E+02
     4 OT SD       0.80E-01  701.8     0.00252773     -7524.0396643421 -6.64E-01
     5 OT SD       0.80E-01  818.3     5.64458633     -8429.2536789938 -9.05E+02
     6 OT DIIS     0.80E-01  781.6    42.41696498    150357.5229304259  1.59E+05
     7 OT DIIS     0.80E-01  699.7     0.00163231     -7523.9647079113 -1.58E+05
     8 OT DIIS     0.80E-01  748.6     0.00216618     -7523.9831087634 -1.84E-02
     9 OT DIIS     0.80E-01  721.1     0.39080499     -7500.7599672907  2.32E+01
    10 OT DIIS     0.80E-01  747.7     0.28233406     -7501.4764303905 -7.16E-01
    11 OT DIIS     0.80E-01  877.7     0.55308111     -7512.0745363284 -1.06E+01
    12 OT DIIS     0.80E-01  691.1     1.21327090     -7412.3714625557  9.97E+01
    13 OT SD       0.80E-01  813.6     0.85821890     -7738.5230298085 -3.26E+02
    14 OT SD       0.80E-01  713.3    28.28919426     81675.8244234667  8.94E+04
    15 OT SD       0.80E-01  899.3    23.78977717     83513.7146849657  1.84E+03
    16 OT DIIS     0.80E-01  851.4    31.26519000     78050.0229091394 -5.46E+03
    17 OT SD       0.80E-01  807.5    11.92609899     27155.5632307179 -5.09E+04

In this email, I attach both input and output files of my hybrid functional calc (SLAB+DYE-old.inp and .out). 

As a second test, I performed a PBE+U+D2 calculation to generate a restart file and then used it as the starting point for a PBE0 calculation, modifying several input parameters such as the basis set, pseudopotentials, and SCF minimization scheme. In particular, I switched from DIIS to CG for the first few SCF iterations before returning to DIIS.

Unfortunately, the calculation did not complete even a single SCF step after one hour of runtime on 210 processors. The corresponding input file is attached as SLAB+DYE.inp.

Based on your experience, what additional tests or strategies would you recommend to identify the source of the problem and resolve it? Are there any specific convergence settings, parallelization options, or diagnostic calculations that could help pinpoint the bottleneck?  Thank you in advance for your patience and help.

Best regards,
Lorenzo Lagasco