Convergece problem in tw probe calculation

[jjhskang]

Hi

I am trying to calculate I-V characteristics for a two-probe system of an organometallic molecule between Au111 wires. I used PBE XC functional. I was able get the curren I between -1.5 < V < 1.6 Volt. I want t extend the caculation at least up to +2.5V and down to -2.5V.

(1) For this, I have tried to calculate the current gradually lowering temperature from a high (say, > 1000K) T to 300K at several steps. That does not work and scf calculton diverged when V >= 1.6 or V <= -1.5. It even diverges at a very high Temp !

(2) I also tried to adjust mixing parameter (from 0.1 to 0.05) and history steps, which do not solve the problem.

Can you give me a suggestion to this probelm?

[jjhskang]

Hi!

For above question, I attach the python script file for the current calculation at high voltage. Please refer to it. All parameters are default values except the mixing parameter.

Thanks.

[Anders Blom]

It is not easy to go to very high bias. It is typically necessary to initiate the calculation at bias V from the converged results at V-dV (the previous point on the curve), but even then it might not converge.

Also, regarding the temperature, it would seem that we might have been a bit too aggressive in promoting a higher temperature as the miracle cure for all problems. Sometimes a lower temperature (even as low as 10 K) works much better, esp. if there are several degenerate states close to the Fermi level, like in a gold nanowire.

Still, and in general, it is not expected that the present version of ATK can achieve convergence at very high bias. There is a however a method by which the available bias range can be increased, know as double contour integration, and this method will, in fact, be introduced in the next version of ATK, of which you can already download a preview version (2009.11) today!
necess

[jjhskang]

Hi

Thanks for your reply.
It seems that the new version doesnot work on CENTOS. When do you expect that I can get a working verson on that OS?

[Anders Blom]

Soon. Can't say a date, but I would expect within 3-4 weeks.