[gpaw-users] Mn2O3 lattice parameter using StrainFilter

[Jens Jørgen Mortensen]

Den 03-10-2013 13:02, Teemu Parviainen skrev:
> Dear GPAW users
>
> I'm trying to resolve the lattice parameter of cubic Mn2O3 bulk 
> structure with StrainFilter class (GPAW version 0.9.1.rev10611 and ASE 
> 3.7.1). However, the method does not yield any good results as the 
> lattice immediately starts to shrink and deform. Basically here are 
> the steps I try to do:
>
> 1. I generate the structure using Crystal-package and relax it. The 
> energy of the final structure is -636.18 eV with total force of 0.0262
> 2. Load up the output file to the Strainfilter-script and try to 
> optimize the lattice parameter
>
> However step 2 fails. The calculation converges fine but the total 
> force of the structure is ridiculously large which I find very weird. 
> The .log file looks basically like this:
>
>     -636.179147    6725.2886
>     -630.665535    6649.4279
>     -618.996444    6554.3936
>     -599.934434    6439.1574
>
> Calculator parameters in both scripts are the same:
>
>  calc = GPAW(
>                 mode=PW(900),
>                 nbands=-40,
>                 xc='PBE',
>                 maxiter=240,
>                 width=0.2,
>                 spinpol= True,
>                 mixer=MixerSum(0.02, 2, 50),
>                 kpts=(1,1,1),
>                 txt=name + '.txt',
>                 setups={'Mn': 'paw:d,3.9'}  #Hubbard U for increasing 
> band gap
>         )
>
> The same method works fine for example for Cu2O and CuO bulk 
> structures, which we have calculated recently.
>
> Other things I have tried:
>     -Perform the calculation without using spins
>     -Generate the structure within StrainFilter-script and not relax 
> it first
>     -Use different values of plane-wave cut-off energy
>     -Use different values of initial lattice constant
>
> All these have yielded same kind of results.
>
> Attached are the relevant scripts and output files. Any help would be 
> kindly accepted.

The StrinFilter multiplies the 6 components of the stress by the volume 
and then converts those numbers to forces acting on two atoms so that we 
can use our normal optimization algorithms which can only handle atoms.

Since your volume is quite high, the "atomic" forces will be high too.  
Maybe that is confusing the algorithm - I'll take a look.

Jens Jørgen

> -Teemu Parviainen
> PhD student
> University of Jyväskylä, Nanoscience Center, Finland

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20131007/25dd6d3d/attachment.html>