Author Topic: Convergence of simulation  (Read 3038 times)

0 Members and 1 Guest are viewing this topic.

Offline hsuya

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: in
  • Reputation: 0
    • View Profile
Convergence of simulation
« on: February 4, 2020, 06:47 »

1. Is there some way to understand whether running a particular script will lead to convergence without first running the geometry optimization.

2. How to do a convergence analysis to understand best which parameter set will be best suited for faster convergence.

3. Could you suggest some resource and ways to speed up a simulation (links to related articles will be helpful in understanding them in detail)

Offline Petr Khomyakov

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 1290
  • Country: dk
  • Reputation: 25
    • View Profile
Re: Convergence of simulation
« Reply #1 on: February 10, 2020, 11:04 »
1. Is there some way to understand whether running a particular script will lead to convergence without first running the geometry optimization.
You can do it for a static (e.g., initial) structure, performing a single SCF convergence test.

2. How to do a convergence analysis to understand best which parameter set will be best suited for faster convergence.
I think you just have to run a set of test calculations. I am not aware of any other for numerical simulations, in general.

3. Could you suggest some resource and ways to speed up a simulation (links to related articles will be helpful in understanding them in detail)
It is a very general question. One can always speed up calculations by doing more approximations, less accurate basis set, lower mesh cutoff and/or k-point sampling and so on.  One can also try using more computer resources, but one should also check if using more cpus would truly speed up particular calculations, because in same case it makes no sense, e.g., for relatively-small systems modeled within LCAO approach and few k-points. But one should always consider the specific case to say something about it.

Offline hsuya

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: in
  • Reputation: 0
    • View Profile
Re: Convergence of simulation
« Reply #2 on: February 18, 2020, 05:17 »
Okay.
Can you elaborate on the single SCF convergence test... I don't know how to do a single SCF convergence test. Usually what we do is we perform a geometry optimization that has many SCF tests so, should I stop that after one scf test is done?

Offline Petr Khomyakov

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 1290
  • Country: dk
  • Reputation: 25
    • View Profile
Re: Convergence of simulation
« Reply #3 on: February 19, 2020, 09:31 »
You just do not do optimization. QuantumATK will do a single SCF calculation for a fixed (initial) structure. You then run this calculation multiple times with different settings, e.g., gradually increasing k-point sampling; the same for other settings. In each run, you have to compute, e.g. total energy or any other physical quantity of your interest. That should be eventually converged within some tolerance (accuracy) that you set - it is your decision based on the needs of your particular study.   

Offline hsuya

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: in
  • Reputation: 0
    • View Profile
Re: Convergence of simulation
« Reply #4 on: February 19, 2020, 11:28 »
Okay so how do I do a single SCF calculation? I did not find any option for that. Do we have to modify the script or something?

Offline Petr Khomyakov

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 1290
  • Country: dk
  • Reputation: 25
    • View Profile
Re: Convergence of simulation
« Reply #5 on: February 19, 2020, 23:13 »
Adding Calculator and no Optimize Geometry means that ones does a single SCF calculation for a static, initial structure.

Offline hsuya

  • Heavy QuantumATK user
  • ***
  • Posts: 78
  • Country: in
  • Reputation: 0
    • View Profile
Re: Convergence of simulation
« Reply #6 on: February 20, 2020, 06:04 »
Thank you! I did not know about it before.