[gpaw-users] Calculations in LCAO mode

[Ulrik Grønbjerg]

Hi Ask

Thanks for your help, those changes improved convergence. However, one
system still fails to converge in 120 iterations, even at linear mixing
parameter of 0.02. I am therefore trying with 150 iterations as maximum.

Would you have other tips to improve convergence?

Best wishes
Ulrik

2012/2/23 Ask Hjorth Larsen <askhl at fysik.dtu.dk>

> Hi
>
> On Thu, 23 Feb 2012, Ulrik Grønbjerg wrote:
>
>  Hi all
>>
>> I have attached the txt-file. In this case it managed to converge the
>> first
>> SCF-cycle in 113 steps, but the second one failed to converge.
>>
>> Best wishes
>> Ulrik
>>
>
> A couple of comments.
>
>  * Reduce linear mixing parameter to 0.03.  This should fix the
>   convergence problem.  Larger systems may require lower values.
>
>  * The Poisson solver is not the problem, so you can use the default
>   convergence criterion.
>
>  * Use gpts=h2gpts(h=0.18, atoms.get_cell(), idiv=8) to get a nicely
>   divisible-by-large-power-of-2 grid (as opposed to the present one
>   which is only divisible by 4).
>
>  * Enable ScaLAPACK
>
>  * (You can use more nodes and it should still be efficent due to
>    k-points.)
>
> Regards
> Ask
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20120228/1d251967/attachment.html