Convergence with the optB88vdW functional
Posted: Sun Jun 21, 2015 9:50 pm
I am doing calculations with the van der Waals corrected optB88vdW functional. I am trying to reproduce the graphene - Ni(111) interaction from Phys. Rev. B 84, 201401(R) (http://journals.aps.org/prb/abstract/10 ... .84.201401). It seems to be working, however the electronic SCF cycle does not seem to be able to converge to the required precision. I have in my INCAR an EDIFF = 1.e-5 eV requirement for a 32 C graphene interacting with a 64 Ni slab (4 layers). VASP basically manages to converge to perhaps one order of magnitude less accurate energies and then stays there until the NELM value is reached (which I have increased to 100).
I have a couple of technical questions:
*Does it make sense to require this precision with this functional and why/why not?
*Why is it so difficult to achieve self consistency down to a small energy tolerance?
And a more physics-related one:
*Would it make sense to run a SC PBE calculation updating the ionic positions and then run a static (NSW = 0) calculation with optB88vdW to get the correct energy? Are the fully SC optB88vdW ionic positions expected to differ substantially from the PBE ones?
I have a couple of technical questions:
*Does it make sense to require this precision with this functional and why/why not?
*Why is it so difficult to achieve self consistency down to a small energy tolerance?
And a more physics-related one:
*Would it make sense to run a SC PBE calculation updating the ionic positions and then run a static (NSW = 0) calculation with optB88vdW to get the correct energy? Are the fully SC optB88vdW ionic positions expected to differ substantially from the PBE ones?