[CP2K-user] Convergence of OT with transition metals

[Nicholas Winner]

Thanks very much Fabian. Why would IRAC and ENERGY_GAP 0.5 lead to reduced 
accuracy? You mean only in the forces, not in the energy, it seems, but if 
we arrive at a density corresponding to very fine SCF convergence, then why 
would the forces calculated from that be any worse?

On Thursday, March 19, 2020 at 1:20:18 AM UTC-7, Fabian Ducry wrote:
>
> Dear Nick,
>
> I typically use something like this (for HfO2, TiO2 etc.)
>
>       EPS_SCF 1.0E-6
>       MAX_SCF 30
>       &OT
>         PRECONDITIONER FULL_SINGLE_INVERSE
>         ALGORITHM IRAC
>         ENERGY_GAP 0.5
>         MINIMIZER DIIS
>       &END OT
>       &OUTER_SCF
>          EPS_SCF 1.0E-6
>          MAX_SCF 20
>       &END OUTER_SCF
>
> In my experience this is a good compromise between accuracy and speed, 
> e.g. for MD. If you need more accurate forces, e.g. CELL_OPT comment the 
> following two lines out
> !        ALGORITHM IRAC
> !        ENERGY_GAP 0.5
>
> You should compare this to GC to make sure this works for your system.
>       EPS_SCF 1.0E-6
>       MAX_SCF 30
>       &OT
>         PRECONDITIONER FULL_SINGLE_INVERSE
>         MINIMIZER GC
>         LINESEARCH 3PNT
>       &END OT
>       &OUTER_SCF
>          EPS_SCF 1.0E-6
>          MAX_SCF 20
>       &END OUTER_SCF
>
> I hope this helps,
> Fabian
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200319/640e6055/attachment.htm>