The Kohn Sham matrix is not 100% occupied

[mguidon]

Hi Claudio.

There are several issues that could cause troubles in a periodic HFX
calculation:

1. Always use the truncated Coulomb operator, otherwise your results
might converge to an unphysical wavefunction (i.e. total energies a
far lower than what you would expect)
2. The sparsity-patterns of the Fock- and Kohn-Sham matrice are not
the same. This issue can be solved by putting EPS_PGF_ORB to 1.0E-32
or similar. (But usually, a 90 % occupied Kohn-Sham matrix is good
enough, especially when you apply the truncated potential)
3. By putting a threshold on all the integrals (EPS_SCHWARZ) you
introduce some random errors of the same order. As a rule of thumb,
eps_schwarz should be roughly equal to the inverse square of the
condition number of the overlap matrix. For example: If the condition
number of the overlap matrix is 1.0E4, then you should put eps_schwarz
to 1.0E-8 (but this is a worst case estimate!) Keep in mind, that the
amount of work roughly doubles by decreasing eps_schwarz by one order
of magnitude. It might be better to decrease the condition number of
the overlap matrix by choosing a less diffuse basis set.
4. Always add the keyword EPS_STORAGE_SCALING 1.0E-1 in order to avoid
suppress noise in the compression/decompression routines.

For more details see

J. Chem. Theory Comput., 2009, 5 (11), pp 3010-3021

Hope that helps

Manuel

On Nov 16, 1:56 pm, Claudio <marg... at gmail.com> wrote:
> Dear all, when I try to run a periodic HF calculation or a mixed HF
> DFT run (b3lyp) for example. I always get
>
>  *** 06:49:15 WARNING in
> hfx_energy_potential:integrate_four_center         ***
>  *** err=-300  The Kohn Sham matrix is not 100% occupied. This may
> result   ***
>  *** in uncorrect Hartree-Fock results. Try to decrease EPS_PGF_ORB in
> the  ***
>  *** QS
> section.
> ***
>
> Same system converges without problem without HF using for example
> PBE. I tried decreasing EPS_PGF_ORB and EPS_DEFAULT without much
> success. I also tried decreasing basis set size. I am not using a
> truncated potential.
>
> Is this warning relevant? I am having a very hard time converging the
> b3lyp calculation.
>
> Thanks,
> Claudio