[Pw_forum] Problems with integrated charge for a large supercell relaxation

[Gabriele Sclauzero]

JR Schmidt wrote:
> I am having problems with a calculation repeatedly aborting during a 
> supercell relaxation.  After MANY successful (force / displacement) 
> iterations, and when the force on the atoms is quite small (i.e. near 
> convergence of the optimization), the calculation aborts with the error:
> 
> WARNING: integrated charge=   405.28031513, expected=   408.00000000
> 
>  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
>      from electrons : error #         1
>      charge is wrong
>  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> 
> I checked the archives, and this problem has come up before for 
> supercell calculations.  I realize that the most obvious explanations are:
> 
> (1)  Bad geometry
> (2)  Bad cutoffs

Maybe I'm wrong, but this could happen also if your charge density and/or wfc files are 
not written correctly in the outdir. Is your problem reproducible, or did it happen just 
only once?

If it is reproducible, please specify version of QE, input file and exact point of crash 
in the program (i.e. first diagonalization step or before diagonalization or etc.).

GS


> 
> With regards to (1), I have essentially ruled out by:
>  * Checking the geometry in xcrysden.
>  * Noting that if I restart the relaxation from the final geometry of 
> the failed relaxation, the resulting SCF converges just fine, and the 
> optimization continues (although the approximate Hessian is lost, so the 
> optimization does not proceed very well).
> 
> With regards to (2), I am using an ultrasoft PP.  For ectuwfc I am using 
> 30 Ry, which works fine for all the SCF calculations.  I tried 
> increasing ecutrho from 200 Ry to 300 Ry (which I think is quite 
> conservative) to no avail.
> 
> Is there something happening in the course of the optimization (say, for 
> example, an atom moving from one side of the unit cell to another across 
> the boundary?) that would render the previous charge density a horrible 
> guess, and cause the optimization to bomb out?
> 
> I don't think the optimization itself is leading to unreasonable 
> geometries, as I looked at the final result (right before the crash); it 
> looks fine.  Also, as I mentioned, I can successfully restart 
> ('from_scratch', but with the final geometry) the optimization from that 
> point.
> 
> Thanks for any guidance.
> 

-- 

La legge n. 133 uccide l'Università e la Ricerca italiana!

Save Italian Brains! (http://saveitalianbrains.wordpress.com/)

o ------------------------------------------------ o
| Gabriele Sclauzero, PhD Student                  |
| c/o:   SISSA & CNR-INFM Democritos,              |
|        via Beirut 2-4, 34014 Trieste (Italy)     |
| email: sclauzer at sissa.it                         |
| phone: +39 040 3787 511                          |
| skype: gurlonotturno                             |
o ------------------------------------------------ o