Author Topic: Is that ok to do MD relax then put the relaxed coordinate for DFT calculations?  (Read 4076 times)

0 Members and 1 Guest are viewing this topic.

Offline quantumtoday

  • Heavy QuantumATK user
  • ***
  • Posts: 29
  • Country: fr
  • Reputation: 0
    • View Profile
Dear all,

Relaxation with DFT is usually expensive, particularly for large structures and/or including defects. With vacancy inclusions, many times I can not get relax convergence with QE and Siesta.

So, my question is that: is that ok to do MD relaxation with VNL and then take the relaxed coordinates for other DFT calculations? Is that still safe?

Any ones with experience about this problem, please give comments and advices.

Thank you.

Offline ams_nanolab

  • Supreme QuantumATK Wizard
  • *****
  • Posts: 389
  • Country: in
  • Reputation: 11
    • View Profile
Don't think you will get the same level of accuracy with MD optimization as with DFT. Unless it's a Really GOOD classical potential which fits very well with DFT. So it's material and as a consequence parameter dependent.

Offline Julian Schneider

  • QuantumATK Staff
  • QuantumATK Guru
  • *****
  • Posts: 164
  • Country: dk
  • Reputation: 25
    • View Profile
Quote
Don't think you will get the same level of accuracy with MD optimization as with DFT. Unless it's a Really GOOD classical potential which fits very well with DFT. So it's material and as a consequence parameter dependen
Actually, running an MD simulation does not mean you are using classical potentials. You can run both MD and optimization with classical potentials OR DFT (or basically any method that gives you reasonable forces).

So, if the question was if you could use MD with DFT instead of Geometry Optimization using DFT, to speed up your optimization, I would say in most cases  running an actual geometry optimization is more efficient. You could run a few steps MD simulation to reduce initial large forces, which in some tricky cases may help to converge a subsequent geometry optimization. However, in MD you have no control over the max. forces, and therefore no control over how much your final configuration deviates from the true optimized configuration, so, unless you just want to have a very coarse guess of your structure, I would suggest running an optimization before you calculate any properties via DFT.

If your question was related to using classical potentials with to relax your configuration, then it can actually make sense to run a pre-relaxation with classical potentials, but, to be in the safe side, I would at least run a DFT force calculation before calculating any properties with DFT, to check if the residual DFT-forces are sufficiently small.
« Last Edit: February 12, 2017, 20:00 by Julian Schneider »

Offline ams_nanolab

  • Supreme QuantumATK Wizard
  • *****
  • Posts: 389
  • Country: in
  • Reputation: 11
    • View Profile
Quote
If your question was related to using classical potentials with to relax your configuration, then it can actually make sense to run a pre-relaxation with classical potentials, but, to be in the safe side, I would at least run a DFT force calculation before calculating any properties with DFT, to check if the residual DFT-forces are sufficiently small.

exactly. 
« Last Edit: February 13, 2017, 11:56 by ams_nanolab »

Offline quantumtoday

  • Heavy QuantumATK user
  • ***
  • Posts: 29
  • Country: fr
  • Reputation: 0
    • View Profile
Dear all,
Thank you very much for these fruitful discussions