Setting correct cutoff/convergence criteria

[Matt W]

>
>
> Thanks for your suggestions. After playing around with the parameters, I 
> figured the Extrapolation issue and cut-off issue (so fixed that). 
> I already see improvement in time after some trials. I am wondering though 
> which method would be best for extrapolation: ASPC etc.
>

I'd stick with ASPC. PS is maybe a bit faster for geo_opt, but ASPC should 
be good and robust.
 

> I haven't switched to GAPW though. I will try that. 
>

Only needed really when you have some elements that need stupidly large 
cutoff (F, Na, Mg) with normal pseudos. It does tend to make forces a bit 
more accurate, so can speed up final convergence a lot too.
 

> Can you please elaborate on "Mixing and broyden optimizer will be 
> moderately inefficient for insulators"
> and a possible alternative.
>
> The Orbital Transform (OT) method is CP2Ks flagship optimizer for 
insulators. For a smallish system

    &SCF
      &OUTER_SCF
         EPS_SCF 5.0E-7
         MAX_SCF 1
      &END OUTER_SCF
      SCF_GUESS RESTART
      EPS_SCF 5.0E-7
      MAX_SCF  15
      &OT
        MINIMIZER DIIS
        PRECONDITIONER FULL_ALL
        ENERGY_GAP 0.001
      &END
    &END SCF

would be my default setup. 

        PRECONDITIONER FULL_SINGLE_INVERSE
        ENERGY_GAP 0.1

for large systems (>500 atoms or so)

Matt

 

> Thanks
>
>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170710/400460a2/attachment.htm>