Re: Convergence of NEGF for defective GNRs ( No.1 ) |
- Date: 2016/10/27 18:27
- Name: Artem Pulkin
- Following are my suggestions:
1. Having 110 k-points does not really matter. You may decrease it down to 10-1 k-point for sure.
2. I guess you have a charge sloshing problem. To check it, look if the occupation numbers go bananas (unphysically low or high) at 121st iteration and after. Set Kerker factor to something large to avoid it. I set
scf.Kerker.factor 35
for a 6-7 nm long scattering region. It does not look like you have that long system so you may try above value. Generally, larger - better (though there are drawbacks, ofc).
3. Try delaying the start of Pulay mixing by a few iterations. For example, I use the following combination:
NEGF.scf.Iter.Band 40 scf.Mixing.History 40 scf.Mixing.StartPulay 90
Here, I waive converged charged density for periodic boundary conditions: looking at your NormRD values it does not really help (same was in my case). Instead, I get a better history to start the Pulay mixing.
4. Finally, you may have to increase your scattering region. But I really encourage you to try cooking a proper mixing scheme because the computation times for a larger system will explode.
|
Re: Convergence of NEGF for defective GNRs ( No.2 ) |
- Date: 2016/10/28 15:37
- Name: T. Ozaki
- Hi,
While something happens at SCF=86,
SCF= 85 NormRD= 0.000000266408 Uele= 0.000000000000 SCF= 86 NormRD= 10.093133746743 Uele= 0.000000000000
the value of NormRD at SCF=85 is already enough to consider the calculation is convergent. NormRD of 10^{-6} might be enough for transport calculations in most cases. Regards,
TO
|
Re: Convergence of NEGF for defective GNRs ( No.3 ) |
- Date: 2016/10/28 18:07
- Name: Artem Pulkin
- At SCF=86 the periodic boundary conditions were replaced by the fixed ones.
NEGF.scf.Iter.Band 85
To the best of my knowledge, in older versions of OpenMX the charge density obtained at this step is not physical (because off-diagonal coupling terms in the Hamiltonian were neglected for simplicity, I guess). I believe one cannot use this Hamiltonian for transport calculations. It is easy to check it on an ideal nanowire model.
Artem
|
Re: Convergence of NEGF for defective GNRs ( No.4 ) |
- Date: 2016/11/07 19:58
- Name: Vipin Kumar
- Dear all,
This problem frequently occurs when we run the NEGF calculation. In NEGF calculation, the NormRD, after the completion of NEGF.SCF.Iter.Band steps, increases suddenly in most of the cases. Similar problem, I am also facing. I want to ask that, is there any simple way or a proper route to avoid this problem. We can play with Mixing parameters, no doubt, but is there any way to choose proper mixing scheme to avoid it in less time and less effort. How to pick the proper mixing scheme. Any help would be highly appreciated.
Thanks.
|
Re: Convergence of NEGF for defective GNRs ( No.5 ) |
- Date: 2016/11/08 23:02
- Name: Artem Pulkin
- "I want to ask that, is there any simple way or a proper route to avoid this problem."
If you think it is a problem, just set
NEGF.SCF.Iter.Band 0
|
Re: Convergence of NEGF for defective GNRs ( No.6 ) |
- Date: 2016/11/10 20:07
- Name: S Dechamps <samuel.dechamps@student.uclouvain.be>
- Dear all,
setting :
NEGF.scf.Iter.Band 40 scf.Mixing.History 40 scf.Mixing.StartPulay 90
(note that the Kerker factor was set automatically to 20.5)
worked just fine for me.
Thanks a lot for the help, Best
|