[CP2K:1948] Re: OUTER_SCF

Fawzi Mohamed fa... at gmx.ch
Fri Apr 3 09:48:39 UTC 2009

On 3-apr-09, at 01:36, Jörg Saßmannshausen wrote:

> Dear Marc,
> as far as I understood it from the Tutorial, it works like that:
> Imagine 2 loops, the inner and the outer SCF cycle. First you run  
> the inner
> cycle. If it converges, you only have one outer cycle done (starting  
> the
> inner cycle).
> If the inner cycle does not converge, rather then going down 'the  
> wrong path',

Not down the wrong path, but simply converging slowly

> you go back (into the outer cycle), make some corrections,

This "correction", if you use the outer cycle just to improve the  
convergence is simply recalculating the preconditioner of OT.

The preconditioner might depend on the actual C (this is not true for  
example for the kinetic one), and in this case a better C means a  
better preconditoner, which means better convergence.

As building the preconditioner costs time (especially for the better  
ones like full_all) you do not want to rebuild it each step, but with  
the outer SCF you can say that if you don't converge in MAX_SCF  
(inner) steps, then the preconditioner should be rebuilt, and you  
should try again (hopefully with better convergence), repeating this  
procedure at most (outer) MAX_SCF times.

As consequence when used in this way you should have the same EPS_SCF  
in both cycles.


> and go into the
> inner cycle again (so second outer circle run). By doing so, the  
> chances of
> convergence are increased.
> I am sure somebody an give a less pictorial and more mathematical
> interpretation of that, but that was my understanding at the time.
> I am aware that this methode is used by other programs as well but  
> in a
> different manner.
> Right, I am tired, braindead and prolly that is full of typos.
> All the best
> Jörg
> On Donnerstag 02 April 2009 marc wrote:
>> Dear all,
>> I'm trying to get a better understanding of the different options  
>> that
>> are available in cp2k by reading some papers and also to get a more
>> hands-on experience with DFT-calculations I'm comparing the results  
>> of
>> different functionals.  To get some help I was reading the cp2k user
>> self-support(part 3) and there I came across the OUTER_SCF option
>> (FORCE_EVAL%DFT%SCF%OUTER_SCF).  So my question
>> is what does an outer_scf cycle do exactly?  I didn't find any
>> reference to the literature or something similar...  Apparently it is
>> important for the convergence but I don't really see how.  And what  
>> is
>> the relation to the 'inner scf' cycle?  There should be an important
>> interaction between the two because the total number of iteration
>> steps is the product of the maximum internal and maximum external
>> steps...
>> Maybe this explained somewhere in a paper?
>> Thanks in advance,
>> marc
> -- 
> *************************************************************
> Jörg Saßmannshausen
> Research Fellow
> University of Strathclyde
> Department of Pure and Applied Chemistry
> 295 Cathedral St.
> Glasgow
> G1 1XL
> email: jorg.sassm... at strath.ac.uk
> web: http://sassy.formativ.net
> Please avoid sending me Word or PowerPoint attachments.
> See http://www.gnu.org/philosophy/no-word-attachments.html

More information about the CP2K-user mailing list