Author Topic: How to understand the real convergence  (Read 4186 times)

0 Members and 1 Guest are viewing this topic.

Offline zhangguangping

  • QuantumATK Guru
  • ****
  • Posts: 193
  • Country: cn
  • Reputation: 2
    • View Profile
How to understand the real convergence
« on: November 15, 2010, 13:43 »
Hi all,
As we know if the convergence is below the criterion,we say the convergence is achived.However if you use the converged *.nc file as an initial guess for the same task,and perform it again,you may find the convergence is above the former criterion,i.e. it needs some scf for the new task.However this is the same task as before,and the initial guess is a converged one.
How to understand it?
And the so called converged *.nc file is reliable? Or we need to get it converged once again?
Even if it converged once again,if we repeat the steps used above,we will probably find the second converged *.nc file not converge to the criterion yet. I know if we repeat it time by time and at last we can find the converged *nc file will really converge for the next rerun.
But problem is :How can we do with the firt so called converged *nc file and the results calculted from it?

Any advice is welcome!

Offline zh

  • Supreme QuantumATK Wizard
  • *****
  • Posts: 1141
  • Reputation: 24
    • View Profile
Re: How to understand the real convergence
« Reply #1 on: November 16, 2010, 02:46 »
If you look closely into the iteration procedure of self-consistent calculation (SCF), your problem can be understood. Although a well-converged *.nc is used as an initial guess, in the new iteration procedure of a SCF two iteration steps at least should be done before checking the convergence. Meanwhile, the mixing technique will be used in the iteration procedure, the mixing during in the new SCF may bring the problem mentioned by you.

Offline zhangguangping

  • QuantumATK Guru
  • ****
  • Posts: 193
  • Country: cn
  • Reputation: 2
    • View Profile
Re: How to understand the real convergence
« Reply #2 on: November 16, 2010, 14:24 »
If you look closely into the iteration procedure of self-consistent calculation (SCF), your problem can be understood. Although a well-converged *.nc is used as an initial guess, in the new iteration procedure of a SCF two iteration steps at least should be done before checking the convergence. Meanwhile, the mixing technique will be used in the iteration procedure, the mixing during in the new SCF may bring the problem mentioned by you.
Thanks for your reply.What you said is really the key to this problem.After I post this question,I suddenly understood it.