The calculation did not converge to the requested tolerance!

[Megan]

Hi!
     I constructed a graphyne-graphene heterojunction,and I calculated the I-V curve. There is a WARNING "The calculation did not converge to the requested tolerance"!
    The parameter settings "  exchange_correlation = LSDA.PZ, max_steps=300 , electrode_voltages= 0.6*Volt, 0.0*Volt , infinitesimal=0.001*eV"
    In addition, I also used spin .
    could you please tell me that how can i fix this problem?   Thanks a lot!

[Umberto Martinez]

did you managed to converge the zero bias state?

[Megan]

Thanks for your reply! in  the zero bias state,it converged! but when I increased the bias,until 0.6,it cant converge.

[Jess Wellendorff]

Do you increase the bias in small steps (e.g. 0.1 eV steps) from zero bias?

[Megan]

I first calculated transmisson spectum, it did did not converge.

[Jess Wellendorff]

I think we need to see the script you use in order to help you here.

[Megan]

This is my script

[Jess Wellendorff]

Your k-point sampling is all wrong: You need much more than 1 k-point along the C-direction for device calculations. The default is 100 k-points along C, so I suggest you try a (1,1,100) grid. Note: This does not mean the calculation will become very slow! The 100 k-points are used only in the electrodes, which are computed in the beginning of the device calculation.

[Megan]

sorry,I give you a wrong script.  here is the scrip.    the k-point sampling is (1,1,100).   Would you please see it again?
  Thank you  so much !

[Jess Wellendorff]

I suggest that instead of doing the 0.6 eV bias calculation from scratch, you do an IVCurve analysis. This will do the zero-bias calculation and then a range of non-zero bias calculations where the previous self-consistent state is used as initial guess.

[Megan]

  I followed your suggestion, and calculated the IV curve.
 At first ,the left electrode  converged in 21steps,and the right electrode calculation converged in 19steps.But the device DFT calculation did  not converge to the requested tolerance until 99 steps. Could you please have a look at my script ?Thank you very much. I wonder whether my settings is right. 

[Jess Wellendorff]

Dear Megan.

I have run your script and I do actually find that the device SCF converges at the initial 0 Volt calculations (in 89 steps). However, I agree that the convergence is rather slow. You should probably increase the max number of SCF steps, and you could perhaps also try a slightly more aggressive density mixing (In "Iteration control parameters", try "Maximum steps" = 200 and "Damping factor" = 0.25).

There is one detail I do not think you do correctly: You want to do an SGGA calculation with an initial guess for the spins. But you set the initial spins only after you have done the first SCF calculation. Therefore you need yet another SCF after setting the initial spins. That's double work. You only need to do a single SCF before doing the IVCurve analysis.

[Megan]

Thanks for your reply . I followed your suggestion,and added the two parameter settings. But the calculation results still had 5WARNINGs. Then I  changed the bias to( -0.5,+0.5),instead of (0,1).  This time ,the results had 2WARNINGs. What should I do?
 By the way, I also see that I have two SCF. But I dont know how to edit my script specifically .Could you help me to reduce one ?
 Thank you, Jess. Looking forward to your reply.
This is my script.

[Megan]

I have a new try,and it has no WARNING!  I changed the energy to (-1,1),instead of (-2,2)

[Jess Wellendorff]

Sounds great! It is usually such that SCF convergence gets easier when the bias window is decreased.

About the double SCF in script C16.py: All you need to dop is delete lines 386-389, which set the calculator on the configuration and executes SCF. That is not needed.