Different values of start_mixing_after_step gave out distinctive results

[stclaireva]

When calculating the transmission spectrum of nanoribbon, I  obtained very different results with start_mixing_after_step = 0 (ref. 0.png) and =8 (ref. 8.png).
The bandstructure shows it is metallic. So start_mixing_after_step =0 probably gave the reasonable results. However, the calculation is not converged for the bias higher than 0.3 V. That's why I tried to use start_mixing_after_step =8, which indeed make the calculation converge fast.

Any idea what's going on here?

[Anders Blom]

The results from using 8 are simply wrong. It converges, but not to any sensible physical solution. If you look at the values of the transmission spectrum, they are all zero (or 1e-22, which is zero).

In my own experience, changing this keyword should only be necessary in extremely difficult convergence cases. If your calculation with 0 converges, there is no reason to change it, you will not converge faster by modifying it.

There is probably another reason why the finite bias calculation doesn't converge. Perhaps you have a perfect ribbon without scattering? That is difficult - and pointless - to treat at finite bias.

[stclaireva]

Thanks for reply.

Yeah, the system we are studing is the perfect nanoribbon and need to now the transmission properties ((I-V) curve). We already tried to change other possible parameters (i.e. mixing parameter, self_energy_calculator_real, initial_density_type, preconditioner, temperature) to solve the converge problem. But it is still not converging. The start_mixing_after_step=8 indeed help the calculation converge much faster for the bias situation. However, as you said it obviously gave out wrong results. Is this parameter similar to 'NELMDL' in vasp? It is supposed to give the right results or it just not fit for some cases?

[Anders Blom]

This is not the correct way to get an I-V curve for a perfectly periodic system. See http://quantumwise.com/publications/tutorials/mini-tutorials/167 instead. Basically, the ballistic, elastic I-V curve of a perfectly periodic system (esp. in 1D) can be inferred rather trivially from the band structure, and is in this respect rather uninteresting.

As a note on "start mixing after", I don't know how other codes do it, it may be a different method or just works differently with plane waves, but in ATK I have never been very successful using this parameters above 1 or 2. On the other hand, almost all calculations in ATK converge with the value 0 - again, the reason your calculation doesn't is not because of this (or any other) parameter, but because it's a rather ill-conditioned problem, to apply a finite voltage over a finite section of a perfectly periodic structure.