[CP2K-user] OT for Aluminum Oxide?

[Fabian Ducry]

Hi,

I can confirm that OT works well with alumina. Settings that I typically 
use (in MD) are
        &SCF
            SCF_GUESS RESTART
            EPS_SCF 1.0E-6
            MAX_SCF 30
            &OT
                MINIMIZER DIIS
                ALGORITHM IRAC
                ENERGY_GAP 0.5
                PRECONDITIONER FULL_SINGLE_INVERSE
            &END OT
            &OUTER_SCF
                MAX_SCF 20
                EPS_SCF 1.0E-6
            &END OUTER_SCF
        &END SCF

IRAC is much faster than CG, but apparently not always reliable. I haven't 
noticed any problems so far though.
If you have more than a few dozen atoms, I recommend FULL_SINGLE_INVERSE 
over FULL_ALL with a relatiely large ENERGY_GAP (~0.5).

OT is not necessarily faster than diagonalization, it just scales better 
with number of atoms. Without knowing more about your systems it's 
difficult to tell what's happening.

Best,
Fabian

On Saturday, 15 February 2020 14:58:56 UTC+1, MD Simulation wrote:
>
> Does anyone know if the OT method could be used for aluminum oxide 
> (Al2O3)?  It seemed a bit slow when I tried it compared to using SMEAR.
>
> Here's my two SCF solving routines, the OT method is slow, ~9s per SCF 
> iteration, while the SMEAR method is ~3s per SCF iteration.  Thanks for the 
> help.
>
> #### OT
>         &SCF
>             SCF_GUESS ATOMIC
>             EPS_SCF 1.0E-8
>             MAX_SCF 200
>             &OT ON
>                 MINIMIZER DIIS
>                 PRECONDITIONER FULL_SINGLE_INVERSE
>             &END OT
>             &OUTER_SCF
>                 MAX_SCF 10
>                 EPS_SCF 1.0E-8
>             &END OUTER_SCF
>         &END SCF
> ####
>
> #### Fast SMEAR
>         &SCF
>             SCF_GUESS ATOMIC
>             EPS_SCF 1.0E-8
>             MAX_SCF 200
>             ADDED_MOS 10
>             &DIAGONALIZATION ON
>                 ALGORITHM STANDARD
>             &END DIAGONALIZATION
>             &MIXING  .TRUE.
>                 METHOD BROYDEN_MIXING
>                 ALPHA 0.4
>                 NBROYDEN 8
>             &END MIXING
>             &SMEAR ON
>                 METHOD FERMI_DIRAC
>                 ELECTRONIC_TEMPERATURE [K] 300.0
>             &END SMEAR
>         &END SCF
> ####
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200217/287fea2f/attachment.htm>