The ABINIT Users Mailing List ( CLOSED )

[Aaron Deskins <aarondesk@yahoo.com>]

Hello Marek,
   I had a question in regards to the following message that you posted a while ago on the abinit forum. The particular change you mentioned (making diag smaller) seems to only apply when using optcell not equal to 0. Would you have any suggestions what I should change in the code if I'm using optcell = 0?

thank you,

Aaron

 marek.veithen@ulg.ac.be wrote:

> Dear users of ABINIT
>
> I had recently some trouble with the structural optimization
> of Wurzite type II-VI semiconductors. The BFGS algorithm(ionmov 2 or 3)
> was not able to converge the forces on the atoms and the stresses
> on the unit cell to the imposed tolerance criterium.
>
> To overcome this problem, I modified the initialization of the inverse
> hessian that is done in the routine Src_9drive/brdmin.f.
>
> if(optcell/=0)then
> ! These values might lead to too large changes in some cases ...
> diag=30.0d0/ucvol
> if(optcell==1)diag=diag/3.0d0
> do idim=3*natom+1,ndim
> hessin(idim,idim)=diag
> enddo
> endif
> write(message, '(a)' )' Inverse hessian has been initialized.'
> call wrtout(06,message,'COLL')
> endif
>
> By imposing a smaller value of diag, I was able to converge the
> forces on the atoms and the stresses on the unit cell down to 5.0d-5 and5.0d-7 respectively usigng typically between 10 and 30 Broyden steps.
>
> In case other people have similar difficulties with structural optimizations
> it could be useful to know how they handle to overcome these problems in
> order to improve the efficiency of the BFGS algorithm.

---

Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now