Error in charges increases every OUTER_SCF cycle.

[John]

cp2k wizards,

It appears that the error in the electron density on grids increases every OUTER_SCF cycle in my recent builds of cp2k.  I am currently using the development and 2.3 branches of cp2k. 

These errors make the SCF very unstable, which is the opposite of the purpose of the OUTER_SCF.

I have noticed this happens in my intel and gnu builds of cp2k.  Almost all of the registry  tests pass in these builds.  Has anyone else noticed this issue with the OUTER_SCF?

I have attached sample output and input for a small case to demonstrate this issue with OUTER_SCF.

I would appreciate it, if anyone can explain why this happens.

Thanks.

John J. Low
Argonne National Laboratory

[shoutian sun]

Hi John,

I think you may miss to use the smooth method in your input.

Please try to add XC_GRID in the &XC part like this:

    &XC

...............

      &XC_GRID

         XC_SMOOTH_RHO NN10

         XC_DERIV SPLINE2_SMOOTH

      &END XC_GRID

..............

    &END XC

Regards,

---

发件人： John <jj...@yahoo.com>
收件人： cp...@googlegroups.com
发送日期： 2013年5月17日, 星期五, 2:19 上午
主题: [CP2K:4431] Error in charges increases every OUTER_SCF cycle.

--
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+uns...@googlegroups.com.
To post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Florian Schiffmann]

Hi John,

I doubt that it has to do with the smoothing (as in the posted previously, I am anyway not a big fan of that option). . I have not tested it yet but the usual reason for such a behaviour is a badly conditioned overlap matrix. Means, the inversion of the matrix goes wrong in the preconditioner and that causes a bad P.
There are various possibilities why that might happen. In your case it might be your box. 4A is very small. This causes some basis functions to go over the boundaries of your box. Furthermore you will use many periodic images (see total number of particle pairs). This might lead to a near linear dependence in S and so cause problems in the inversion.
In general, as Cp2K does not use k-points it is not a good idea to use such small unit cells. Using a 2x2x2 box won't be much more expensive and will give more reasonable results and probably fix your electron problem. Furthermore you have a very tight convergence criterium here you might even need a higher cutoff (just tried with 2x2x2, 800 works) to be able to reach that convergence level.

Flo