IV curve calculation did not converge after 0.4v bias

[alokrs]

I am trying to calculate IV curve fore MTJ .  The calculation is not converging after bias of 0.4V. I have used DZP basis set for all the elements and have kept damping factor to 0.05.

[Petr Khomyakov]

Please take a look at the following 2 guides:
https://docs.quantumwise.com/technicalnotes/negf_convergence_guide/negf_convergence_guide.html
https://docs.quantumwise.com/technicalnotes/spin_polarized_negf_convergence_guide/spin_polarized_negf_convergence_guide.html.

In addition to that, since you have decreased a damping factor value, you should also consider reducing the history steps to something like 12. I would also suggest increasing the k-point sampling, e.g., to na=18, nb=10, nc=300 or whatever. You may also need to do calculations with smaller voltage step, not 0.1 V as it is now, but something like 0.05 V.

You may also use IVCharacteristics study object (note this is only available in the recent QATK versions) for doing IVCharacteristics calculations: docs.quantumatk.com/manual/Types/IVCharacteristics/
docs.quantumatk.com/tutorials/ivcharacteristics/ivcharacteristics.html

This study object has many convenient restart options when you do device calculations.

[alokrs]

As suggested by you, I tried K-point= 18, 10 and 300, damping factor 0.05, history steps 12 and equivalent bulk method. The calculation for zero bias converged in 115 steps, 0.05 V in 36 steps, but for bias 0.1V it didn't converge to the requested tolerance of 1E-05 in 200 steps.

[Petr Khomyakov]

It would be helpful to see the log files for this calculation. Note that if k-point sampling is not the cause for slow convergence, you may now reduce it; I would however still keep nc at 300. 

[alokrs]

I guess it's the num_histoty_steps that is creating problem in convergence, I reduced the k-point to na=13, nb=7, and nc=141 and num_history_steps=12. The zero bias calculation converged in 147 steps, but 0.05V, 0.3V, 0.35V, 0.4V bias doesn't converge to the requested tolerance of 1E-06 in 200 steps. The detailed log file is attached



../ivcurve_0.10000V.log:| Calculation Converged in 165 steps                                           |
../ivcurve_0.15000V.log:| Calculation Converged in 116 steps                                           |
../ivcurve_0.20000V.log:| Calculation Converged in 123 steps                                           |
../ivcurve_0.25000V.log:| Calculation Converged in 91 steps                                            |
../trans_195612.log:| Calculation Converged in 21 steps                                            |
../trans_195612.log:| Calculation Converged in 45 steps                                            |
../trans_195612.log:| Calculation Converged in 147 steps


ivcurve_0.05000V.log:# Warning: The calculation did not converge to the requested tolerance!        #
ivcurve_0.30000V.log:# Warning: The calculation did not converge to the requested tolerance!        #
ivcurve_0.35000V.log:# Warning: The calculation did not converge to the requested tolerance!        #
ivcurve_0.40000V.log:# Warning: The calculation did not converge to the requested tolerance!        #

[Petr Khomyakov]

Did you try other tricks (different pseudopotential/basis set, different Poisson solver, e.g., PCG) suggested in these guides?
https://docs.quantumwise.com/technicalnotes/negf_convergence_guide/negf_convergence_guide.html
https://docs.quantumwise.com/technicalnotes/spin_polarized_negf_convergence_guide/spin_polarized_negf_convergence_guide.html.

Actually, in the latter guide, the example system is Fe-based MTJ. What about trying to adopt similar computational settings to see if that would work for your Fe-based MTJ?