[CP2K-user] [CP2K:11276] Re: SCF convergence criterion options

[Matthew Okenyi]

Hi Thomas,

Thanks for your advice. I have been comparing the convergence of the 
orbital transformation and diagonalization (is this what you meant by 
'TD'?) methods, and found that for the same value of EPS_SCF=1.0e-7 the 
final change in energy at convergence is around 100 times smaller for OT 
than for TD (I have attached a plot of dE0 vs iteration). I think this 
suggests the opposite behaviour to what you've just described! Are there 
any circumstances where one might expect this? I used the CG minimizer with 
the FULL_KINETIC preconditioner and ENERGY_GAP parameter of 5.0e-4 a.u for 
OT and Broyden mixing with ALPHA = 0.4, NBUFFER = 8 for TD.

Best wishes,
Matthew

On Saturday, 16 February 2019 13:16:06 UTC, tkuehne wrote:
>
> Hi, 
>
> even though I agree with the comments made by Martin, I would like to add 
> that only for TD 
> EPS_SCF is the maximum norm of the difference between two subsequent 
> density matrices. 
> In case of OT, it is the mean deviation between the preconditioned 
> gradients, which is a much 
> weaker criterion! From this it follows first that in my opinion the change 
> of energy is not the best 
> choice of stopping criterion since it may be strongly dependent of the 
> employed preconditioner 
> and second that a calculations with TD and the default EPS_SCF value of 
> 1.0E-6 is typically 
> more accurate than the same calculation with EPS_SCF set to 1.0E-7. In 
> other words, the 
> choice of eigenvalue solver is much more important than the specific 
> stopping criterion!
>
> Best, 
> Thomas
>
> Am 16.02.2019 um 12:18 schrieb M. Brehm <brehm... at gmail.com 
> <javascript:>>:
>
> To the best of my knowledge, no other critera to signal SCF convergence 
> are implemented in CP2k.
>
> However, it is generally a bad idea to look at the SCF energy to determine 
> if the SCF has converged... There exist quite common cases where the energy 
> remains almost constant in some SCF iterations, but will heavily reduce 
> later on. Many years of practical experience have shown that the norm of 
> the difference between two subsequent density matrices is quite a good 
> criterion.
>
> Best regards,
> Martin
>
> -- 
> You received this message because you are subscribed to the Google Groups 
> "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to cp2k+... at googlegroups.com <javascript:>.
> To post to this group, send email to cp... at googlegroups.com <javascript:>.
> Visit this group at https://groups.google.com/group/cp2k.
> For more options, visit https://groups.google.com/d/optout.
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190216/fabd4eed/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dE0_comparison_OT_diag.png
Type: image/png
Size: 14899 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190216/fabd4eed/attachment.png>