Why more difficult to converge by using ATK2011.2.2?

[hejun8035]

Hello, everyone

    For the same system and the same parameters, it can be converged  with 39 setps by using ATK2008.10 while did not converge to the requested tolerance more than 100 steps by using new version ATK2011.2.2? (The py file except for configuration post below )

   Moreover, it can't  set verbosity_level in ATK2011, the output file is too large and it is diffcult for user read.

[nori]

Poor convergence in ATK 11.2.2 might come from bound states within bias window.
How about using single contour instead of double contour?

[hejun8035]

Thanks for your reply!
 
Whether only change DoubleContourIntegralParameters to SingleContourIntegralParameters?

[nori]

Whether only change DoubleContourIntegralParameters to SingleContourIntegralParameters?

Yes, and if you want to include bound states within bias window in density matrix, you should set as "SingleContourIntegralParameters(contour_type=Right)".
(In the case of that a left voltage is higher than a right voltage)

[hejun8035]

Thank you very much!

In other words, we will use (contour_type=Left) at the negative bias voltage?

However, what is the DoubleContourIntegralParameters used for ?(It is Poor convergence)

[nori]

In other words, we will use (contour_type=Left) at the negative bias voltage?

Exactly

However, what is the DoubleContourIntegralParameters used for ?(It is Poor convergence)

Are you saying why SCF convergence with double contour gets worse if there are bound states within bias window?
If so, I'm sorry that I don't have clear answer about it.
I've not understood the property of the formalism used for weighting scheme perfectly yet
http://www.quantumwise.com/documents/manuals/latest/ReferenceManual/index.html/ref.doublecontourintegralparameters.html

So please note that even if there are bound states within bias window and double contour is used, SCF would converge well in some cases.
But there've also been the case that SCF can't converge due to it in my experience.
In such case, I could overcome the issue by using single contour.
And at the same time I did the two type of calculation (contour_type=Left, contour_type=Right), consequently I obtained the different electronic structure.
This is the direct evidence that there are indeed bound states within bias window.
 

[hejun8035]

Thanks!

I mainly understand you!

I have read the Reference Manual and Ref.[23]. However, still have one question.

Compare to DoubleContour, whether the SingleContour have large effect on result

and whether the bound state unconsidered in all the older versions (such as ATK-2008).

[nori]

Compare to DoubleContour, whether the SingleContour have large effect on result

Yes, it could be especially when SingleContour DOES NOT include bound states within bias window.
Let me show you one example of the influence of Bound states with attached.
You will clearly understand including bound states would make the result difference so much.

I guess that the density matrix includes bound states is more accurate.
But please note that bound states are added to density matrix as an equilibrium state even if the states are within bias window when you use SingleContour integration.
So I also feel that SingleContour does not treat bound states within bias window perfectly.

and whether the bound state unconsidered in all the older versions (such as ATK-2008).

Yes.
More precisely, old ATK only includes bound states in density matrix blow the lower electrode fermi level.

[hejun8035]

Thank your patient reply!

[hejun8035]

Dear nori ！

   It is indeed benifit for converge when we use single contour. However, it is still

more difficult to converge than ATK-2008.10 and there are some bias voltage still

can not converge to the requested tolerance more than 100 steps.

   More importantly, the reluts are very different from ATK-2008.10? why?

   What should I do ?

[Anders Blom]

It is possible that the system is not set up correctly geometrically. Sometimes the "bulk to device" conversion isn't perfect, and then you have to adjust it a bit. Without seeing the coordinates, however, it's impossible to say.

I'm not sure I agree with the discussion of double vs. single contour. The idea of the double contour is to make it possible to converge even difficult cases where a bound state is present in the bias window. In practice this typically occurs at high bias, and so 11.2 is potentially able to handle much larger bias values than 2008.10, where you usually had to struggle to get about 1 V bias. We have successfully tried up to 5 V bias with 11.2.

So my guess, again without having seen the system, is that the problem at hand is more a matter of geometry than contour integration.

[hejun8035]

Dear Anders Blom:

       Firstly, thank your reply!

       However, the transmission spectrums at zero bias are consistently very well !
      

[nori]

   More importantly, the reluts are very different from ATK-2008.10? why?

Perhaps it comes from the difference of the algorithm.
Finite bias calculation of ATK 11.2 is more reliable than that of ATK 2008.10.
http://www.quantumwise.com/documents/tutorials/latest/UpgradeGuide/index.html/chap.upgrade.html#sect1.upgrade.twoprobe

About convergence, It seems that more fine tuning for SCF parameters is needed.
But I agree with Anders that we can't give you further advice without the geometry information.