Page 1 of 1
Hybrid functionals, recommended practice for defects calculations
Posted: Wed Aug 28, 2013 5:26 pm
by Neutrino
In using hybrid functionals which of the following is recommended to obtain reliable energies and electronic structure for defects in semiconductors and insulators and in the same time the computational cost is kept reasonable.
1- Performing one ionic step using PBE to get WAVECAR and then performing the full structure relaxation using the hybrid functional.
Or,
2- Performing the full structure relaxation using PBE and then use the hybrid functional to perform one ionic step just to get the energy.
Or
3- Performing the full structure relaxation using PBE and then use the hybrid functional to continue the full relaxation process.
Thanks in advance for your help
Hybrid functionals, recommended practice for defects calculations
Posted: Fri Aug 30, 2013 1:17 am
by Neutrino
Any suggestions?
Hybrid functionals, recommended practice for defects calculations
Posted: Sun Jan 05, 2014 7:19 pm
by WolverBean
Neutrino,
The answer to your question depends on what tradeoff you're willing to accept between cost and accuracy. The optimal geometry in a hybrid functional will differ from that in PBE. So the real question is, how sensitive is the electronic structure to small errors in geometry?
Here's a suggestion:
Fully optimize your structure at the PBE level. Then run an energy/DOS calculation at the hybrid level. Now, take the same relaxed geometry, and scale the unit cell by 99% and by 101%. Run energy/DOS calculations on the slightly smaller and slightly bigger unit cells, and see how different the electronic structures are from that of the relaxed PBE geometry. If the electronic structures don't differ much, then you don't need to do any geometry relaxation at the hybrid level to get a qualitatively correct answer. This approach requires only 3 single-point calculations at the hybrid level and only 1 geometry optimization at the PBE level, so it should be an inexpensive way to test for the importance of geometric relaxation.
Of course, if the electronic structures do differ by a lot, then you will need to re-optimize the geometry at the hybrid level. That will be very expensive, but at least now you know that it's necessary. It's not necessarily obvious whether you should start from the literature crystal structure or from the PBE-relaxed geometry when doing relaxations at the hybrid level. I'd be tempted to start from the literature structure, but your mileage may vary.