[QE-users] convergence criterion?

[Matic]

Hi Matthew,
> I've often heard it said that 1meV/atom is a good criterion for 
> convergence tests to set ecutwfc, ecutrho and K-point sampling.  I 
> wonder if that makes
> sense in if the structure has very light and heavy atoms.  One I'm 
> working on now contains Ca, Si, O and H.  Since each H contributes 
> only one electron, should
> it count equally in the tally of 104 atoms?  Should the criterion be a 
> certain energy/total electron?  Energy/valence electron (defined in 
> the pseudo)?

Convergence of ecutwfc/ecutrho is a property of the pseudpotential used, 
so you should converge ecutwfc/ecutrho for the each new atom type 
(pseudopotential) in your system and then used the maximum value 
obtained for all calculations. A hint for appropriate values is 
typically listed in the pseudopotential file. The easiest way to do this 
is just to make a small box containing the atom and increase ecutwfc 
(using lets say ecutrho=8*ecutwfc_max) until the energy stays within 
approx ~1 mRy. Once you obtain the converged ecutwfc then you can 
decrease ecutrho to a more appropriate value.

Here is a hands-on presentation from a workshop where this is described: 
http://indico.ictp.it/event/7921/session/320/contribution/1261/material/0/

Best Regards,

Matic

-- 
Matic Poberznik
J. Stefan Institute, Jamova 39, Ljubljana, Slovenia