QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: Sabyasachi Sen on July 12, 2014, 11:45

Title: calculation did not converged-even-after chaning the tolerance-two-probe-job
Post by: Sabyasachi Sen on July 12, 2014, 11:45

I am performing two-probe run on a system having 7x7 Au eletrodes with SZ basis and molecule consisting of C, H, N, Mn; with DZP basis. Large molecule needed large electrode for better connectivity.
However, in the outpt, an error appers just befor transmission spectrum analysis, the error is repoted as

+------------------------------------------------------------------------------+
|  99 E = -313.274 dE =  7.651696e-04 dH =  3.574468e-02                       |
+------------------------------------------------------------------------------+
################################################################################
#                                                                              #
# Warning: The calculation did not converge to the requested tolerance!        #


Looking at this I have given rerun of the job with higher tolerance given as,

left_electrode_iteration_control_parameters = IterationControlParameters(
    tolerance=0.1,
    )

right_electrode_iteration_control_parameters = IterationControlParameters(
    tolerance=0.1,
    )

device_iteration_control_parameters = IterationControlParameters(
    tolerance=0.1,
    )

However, the output shows similar covergence problem.

Kindly, suggest me how to overcome this problem.

Title: Re: calculation did not converged-even-after chaning the tolerance-two-probe-job
Post by: Umberto Martinez on July 14, 2014, 10:00

Changing the accuracy so much is probably not a good idea. Your results will not be.... accurate!

Seems that your job is not doing so bad. A small push may solve your issue.
I would rather change other parameters like damping_factor (try in increase that to 0.5/0.8, although we found this to improve mainly bulk calculations).
Sometimes you can improve convergence by increasing k-points or mesh cut-off.
If you see some improvement but 100 steps are not enough consider increasing max_steps a little bit.

Title: Re: calculation did not converged-even-after chaning the tolerance-two-probe-job
Post by: tara on February 9, 2015, 19:34

For slater koster calculations, how much can we increase density mesh cut off for better convergence at higher bias??