[Smeagol-discuss] Problems in charge and convergence

[Stefano Sanvito]

Hi,

OK, you probably can increase NEnergImCircle to 50-60 although probably it will
not help here. I believe the problem is with the fact you are trying to use 1D
leads. In these screening is really poor and you need huge unit cells to
converge. How many atoms of the leads you have in your scattering region? A
better strategy would be to use some more bulky leads, better if you use some
periodic boundary conditions. You usually don't need many k-points (2 or 4
should be enough for Al).

Stefano


Quoting Wei Fan <mail.fanwei at gmail.com>:

> Dear  Stefano,
>
> Thank you very much for your suggestions. In my calculations,  the
> parameters for charge integration are
>
> NEnergReal          857
> NEnergImCircle       20
> NEnergImLine         20
> NPoles               10
>
> I have tried to increase NEnergReal to 2000, but it helped a little. Then
> when I increase MaxScfSteps to 800,  the charges finally converged after
> really a lot of steps. Although it takes much, it works anyway.Is there a
> possible way to accelerate the convergence for higher bias?
>
> Thanks in advance!
> Wei
>
> On Sat, Mar 22, 2008 at 5:59 AM, Stefano Sanvito <sanvitos at tcd.ie> wrote:
>
> > Hi Wei,
> >
> > > 1) In my calculations, the zero bias charges converged to 194.02 when it
> > is
> > > 194.00 in the first scf step. I am not sure if it's acceptable. However,
> > the
> > > weird thing is that the converged charge gets bigger when bias is
> > larger,
> > > 194.03 when bias=0.1V, 194.037 when bias=0.2V,  194.044 when bias=
> > 0.3V......
> > > Does it make sense or is there something wrong?  Btw, The system
> > consists
> > > of  two wire Al leads along 111 with a 1,4-benzene-dithiolate in the
> > > center.  The system is recognized as a Chain in siesta.
> >
> > In general there is nothing particularly wrong with that. The
> > electrostatics is
> > different at different bias and it is not surprising that the
> > self-consistent
> > charge density in the scattering region changes with bias. Whether this is
> > correct for the specific system you are looking at, it is difficult to
> > say.
> > Usually problems arise when the charge density systematically decreases
> > with
> > bias (by many electrons). This is not the case here, so this looks
> > approximately OK.
> >
> > > 2) Generally, I can't get converged densities when bias is above 1 volt
> > with
> > > a 10-4 tolerance. The charge difference in those biases with zero bias
> > is
> > > more than 0.1.  Is there something responsible for this?
> >
> > There might be several reason for that. Did you try to change the number
> > of
> > energy points along the real axis? You might not have too many and the
> > charge
> > density is integrated on a too course grid. How many are you using?
> >
> > Regards,
> >
> > Stefano
> > --
> > Prof. Stefano Sanvito
> > School of Physics and CRANN,
> > Trinity College Dublin, IRELAND
> > Tel: +353-1-8963065
> > www.smeagol.tcd.ie
> >
> >
>
>
> --
> Wei Fan
> Room 332, Main Building
> Nano-Organic Photoelectronic Laboratory
> Technical Institute of Physics and Chemistry
> Beijing, China, 100101
> Tel: +86-13811270970
> +86-10-82680161
>


--
Prof. Stefano Sanvito
School of Physics and CRANN,
Trinity College Dublin, IRELAND
Tel: +353-1-8963065
www.smeagol.tcd.ie