[Pw_forum] Help with atomic relaxation (Issue with convergence)

giuseppe.mattioli at mlib.ism.cnr.it giuseppe.mattioli at mlib.ism.cnr.it
Mon Jan 23 17:25:51 CET 2012


Dear Vikas
Hard to say without looking at your input file...

> So, I wanted to ask how I can make is equilibrate faster? Would a lower
> value of conv_thr (1e-4 oe 1e-5)

Maybe, but you are going to calculate poorly converged forces on
ions... Dangerous!

I suppose that you are running a nspin=2 calculation; open shell
systems are sometimes very nasty. You may try to:

lower your mixing_beta parameter, maybe down to 0.05 or 0.02;
perform a (slower) diagonalization="cg" calculation;
increase the smearing value (but check the results at the end of the
calculation with the lowest value which allows you to converge)

HTH but it may turn out not to be enough...

Giuseppe

Giuseppe Mattioli
ISM-CNR
ITALY

Quoting Vikas Varshney <vv0210 at gmail.com>:

> Dear Users,
>
> I want to relax (not vc-relax) the combined system of Al2O3 slab with 2 or
> 3 layers of Ni-atoms on the top of it. Before this combined study, I have
> successfully relax the Al2O3 slab and Ni layers seperately. When I combine
> them together, I dont know (in advance) the seperation at which I should
> keep the Ni layers on the top of Al2O3 slab. So, I keep it ~2 Angs apart
> and was hoping that with relax, I should get a relaxed geometry.
>
> The issue I am having is that the electronic relaxation is not converging
> (1e-8) even after 500 steps (electron_maxstep = 500).
>  I tried 1000 steps also. It ran through few iterations of atom relaxations
> but then got crashed after few iterative steps (at a certain step).
>
> So, I wanted to ask how I can make is equilibrate faster? Would a lower
> value of conv_thr (1e-4 oe 1e-5)  at the initial stages help (to relax
> atomic coordinates) for frequently. Once, I know the approximate good
> positions for Ni, atoms, I could increase the conv_thr then.
>
> Please guide, any other suggestions are also highly welcomed
>
> Best Regards,
> Vikas
>
> Computational Reseacher,
> Wright Patterson Air Force Base,
> Dayton, OH.
>





----- End forwarded message -----






More information about the users mailing list