How to solve SCF convergence problems 2

[uwe]

Convergence settings and hints for SCF convergence in Turbomole, continued...

• in cases when either very diffuse basis sets are used or electronic structures are near (or at) multi-reference cases, it can help to increase the accuracy of the integral screening. This will lead to probably much longer computational time, but can help at certain critical points.
  
  Note that in such cases it is highly recommended to check for correct input - in a multi-reference case the results are at least questionable, even if the calculation runs technically correct.
  
  Setting
  
  $scftol 1d-15
  
  should help, eventually 1d-16 or even smaller is necessary.
   
• check the output of eiger
  
  If the numbering scheme is not consistent with the orbital energies (i.e. you have a hole in the occupation, but the numbering has no holes), please force dscf or ridft to diagonalize the fock matrix at the very end of a SCF procedure to get rid of the (virtual) hole:
  
  $lastdiag
  
  Note that this option only works for Turbomole version 6.1 and newer.
  
  Again, this usually only happens when the HOMO-LUMO gap is very small and is a strong hint that you ran into a multi-reference case!