GDM Line Search Overstep

Questions
#1

I’m running the following job:

$rem
   JOBTYPE         sp
   METHOD          wB97X-D3
   BASIS           def2-tzvppd
   SYMMETRY        false
   SYM_IGNORE      true
   UNRESTRICTED    true
   SCF_MAX_CYCLES  1000
   SCF_ALGORITHM   diis_gdm
   SCF_CONVERGENCE 7
   MEM_TOTAL       64000
   CDFT            true
   CDFT_BECKE_POP  true
   SOLVENT_METHOD   SMD
   THRESH          10
$end

I’m using DIIS_GDM because plain DIIS fails to converge (and actually converges to a different state)

The first few cycles of optimization look normal:

---------------------------------------
  Cycle       Energy         DIIS Error
 ---------------------------------------
    1   -1514.3123093882      1.11E-02
    2   -1509.4561119885      1.75E-03 Done DIIS. Switching to GDM
    3   -1509.5921545528      6.00E-01 Descent step
    4   -1510.0674799206      2.91E-01 Descent step
    5   -1510.3269758906      1.93E-01 Descent step
    6   -1510.4240223605      6.48E-02 Normal BFGS step
    7   -1510.4446922618      1.30E-02 Normal BFGS step
    8   -1510.4454963533      2.07E-03 Normal BFGS step
    9   -1510.4455530461      2.62E-03 Normal BFGS step
   10   -1510.4455860852      6.71E-04 Normal BFGS step
   11   -1510.4455882307      3.99E-04 Normal BFGS step
   12   -1510.4455889806      8.67E-05 Normal BFGS step
   13   -1510.4455885945      1.80E-05 Line search: overstep
   14   -1510.4455890186      7.47E-05 Normal BFGS step
   15   -1510.4455894523      9.58E-06 Normal BFGS step
   16   -1510.4455895211      1.28E-05 Normal BFGS step
   17   -1510.4455900195      1.48E-05 Normal BFGS step
   18   -1510.4455897482      1.21E-05 Line search: overstep
   19   -1510.4455899958      1.57E-05 Line search: overstep
   20   -1510.4455893634      1.25E-05 Line search: overstep
   21   -1510.4455873121      1.74E-05 Line search: overstep
   22   -1510.4455841658      2.37E-05 Line search: overstep
   23   -1510.4455814043      3.01E-05 Line search: overstep
   24   -1510.4455800397      3.18E-05 Line search: overstep
   25   -1510.4455797063      3.13E-05 Line search: overstep
   26   -1510.4455797025      3.13E-05 Line search: overstep
   29   -1510.4455797035      3.13E-05 Line search: overstep
   30   -1510.4455894180      4.39E-05 Line search: overstep

The problem is that then I have 130 more lines which are identical to line 30:

  166   -1510.4455894180      4.39E-05 Line search: overstep
  167   -1510.4455894180      4.39E-05 Line search: overstep
  168   -1510.4455894180      4.39E-05 Line search: overstep
 Inaccurate integrated density:
   Number of electrons =   214
   Numerical integral  =     0.000000000000
   Relative error      =  -100.0000000000 %
Constraint 0 : 2.0000000000
 Lam  -0.2236920282351160
  169               -nan           NAN Convergence criterion met
 ---------------------------------------
 <S^2> = nan

Which ends with all energies being NaN and calculation terminating.

Is there anything I’m doing wrong? Anything I should be aware of?

The CDFT part constraints a charge transfer between solvent and a molecule, if that’s relevant.

#2

CDFT does sometimes require a tighter integration grid so you could try XC_GRID = 000099000590 for EML (99,590). However, in this case the grid warning may be a symptom of a bad GDM step rather than the cause. Suggest trying other choices for SCF_ALGORITHM (maybe ADIIS or RCA). Strip out the solvent model first and see if you can converge gas-phase. Try a smaller basis set, etc. Note there is no guarantee of finding a constrained DFT solution, it can fail depending on the system and nature of the constraint.

#3

RCA_DIIS performs better, thank you. some of calcs with SMD also have convergence issues, but i think it’s probably closer to your note about the existence\physicality of my constraint. thanks anyway!