QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: sayantanu on May 13, 2015, 11:10

Title: convergence problem
Post by: sayantanu on May 13, 2015, 11:10

Dear atk users,
I was doing transmission spectrum analysis of a two probe system, having a molecule of 4 cobalt atoms. But it is not converging. When I am increasing step numbers it is doing only 99 steps. I have tried to converge by increasing basis, but the job suddenly stopped. Please help.

Title: Re: convergence problem
Post by: Jess Wellendorff on May 19, 2015, 13:47

How did you increase the maximum number of SCF steps? Did you increase "Maximum steps" in "Iteration Control" using the Script Generator?

Title: Re: convergence problem
Post by: sayantanu on May 21, 2015, 07:39

Yes, I also increased history steps and decreased the tolerance limit. Here is the required part :

left_electrode_iteration_control_parameters = IterationControlParameters(
    damping_factor=0.5,
    max_steps=120,
    tolerance=0.001,
    number_of_history_steps=30

Title: Re: convergence problem
Post by: Jess Wellendorff on May 21, 2015, 10:02

ok, so you have increased the maximum number of SCF steps to 120. That's not really a big increase from 100.

I suggest you try lower the damping factor and increase the max_steps:

left_electrode_iteration_control_parameters = IterationControlParameters(
    damping_factor=0.05,
    max_steps=500,
    )

Title: Re: convergence problem
Post by: sayantanu on May 26, 2015, 09:39

I have done what u have said. But the job still stopping  at 99th step.
Please help.

Title: Re: convergence problem
Post by: Umberto Martinez on May 26, 2015, 10:40

Please, attach your full script.

You probably increase max step of the electrodes only. You have to do it for your device calculator as well.
Use the Script Generator to do it for your.

There might be other things to try before increasing that much the number of SCF steps.
Again, if you add your script we may come up with more precise suggestions.

Title: Re: convergence problem
Post by: sayantanu on May 27, 2015, 09:07

Here is the script.

Title: Re: convergence problem
Post by: Umberto Martinez on May 27, 2015, 14:32

In my opinion:
- your central region is too short. You actually do not have gold atoms that are not part of the electrode extension.
- increase the number of k-point in the C direction. Even to about 200.
- include polarization in your basis set. SZP for gold and DZP for the other elements.
- keep Iteration Control Settings as default
- perform a zero bias calculation first.

- When you converge your calculation, run the transmission spectrum separately (higher parallelization, if you can afford it) with a much higher k-point sampling.

Yes, it is a demanding calculation but that is how your system is.

Title: Re: convergence problem
Post by: sayantanu on June 3, 2015, 07:36

zero bias calculation is converging well, but when I gave some bias the job stopped after some time without showing any error.

Title: Re: convergence problem
Post by: Quhe on June 5, 2017, 06:30

Quote from: sayantanu on June  3, 2015, 07:36

zero bias calculation is converging well, but when I gave some bias the job stopped after some time without showing any error.

meet the same problem..