Convergence-CP2K

Electronic convergence

Ideas to help electronic convergence problems with CP2K

Fix the Geometry

If the geometry is "bad", then convergence may be difficult. A bad geometry may be far from the optimized structure, or have an unrealistic structure (e.g. too short/long bonds). Change the geometry to a more reasonable guess then try again.

Increase number of SCF Steps

&SCF
  EPS_SCF 1.E-6
  SCF_GUESS RESTART
  MAX_SCF 700
  #MAX_SCF 400 
  &OT T
    PRECONDITIONER FULL_SINGLE_INVERSE
    MINIMIZER DIIS
    LINESEARCH 3PNT
  &END OT
&END SCF

Change XC_GRID

&XC_GRID
           XC_DERIV  SPLINE2
           XC_SMOOTH_RHO NN50
&END XC_GRID

Note: This is from the CP2K google group, and I haven't really tested it.

Change OT Minimizer

I usually use the DIIS minimizer, but changing to CG may work.

&OT T
 PRECONDITIONER FULL_SINGLE_INVERSE
 MINIMIZER CG
 LINESEARCH 3PNT
&END OT

Use Outer Loops

This basically resets the scf cycle rather than letting it continue to diverge or oscillate. DIIS may work also instead of CG.

&SCF
 EPS_SCF 1.E-6
 SCF_GUESS ATOMIC
 MAX_SCF 50
 &OUTER_SCF
  EPS_SCF 1.E-6
  MAX_SCF 20
 &END OUTER_SCF
 &OT T
  PRECONDITIONER FULL_SINGLE_INVERSE
  MINIMIZER CG
  LINESEARCH 3PNT
  &END OT
&END SCF