The ABINIT Users Mailing List ( CLOSED )

[Igor Lukacevic <ilukacevic@fizika.unios.hr>]

Hi!

Convergence study for GaAs is not much harder than that of H molecule. The
acell convergence is analogous to that of H bond length: one doesn't change 
the
positions of atoms but the cell dimension.

You could check an ecut convergence by watching a total energy or the acell
values. Since, if one assumes that you are interested in ambient pressure 
phase
of GaAs with zinc blende structure (in which the positions of atoms are fixed
by symmetry), equilibrium acell only correspondes to the energy minimum, a 
good
energy convergence ensures you a good acell convergence. So, you could make a
simple total energy calculation for a series of ecuts and see when your energy
is converged under wanted tolerance. This will give you the ecut convergence.
Don't forget that the converged energy is for the very large ecut, so you
should check the energy for the very large ecut also and compare all other
total energies with that one - that is your tolerance. These calculations you
should make at a reasonable ngkpt. For properties other than structural, you
should check the ecut convergence again!

Acell convergence could be made by calculating the equilibrium acell (only one
value for zinc blende) for many ecuts until the wanted tolerance is achieved.
Or compare its value with an experimental one (or other theoretical results).
But have in mind at which pressure is the experiment undertaken. You can
calculate equilibrium acell in two ways, which give you the same result:
a) calculate total energy for many acells and the equilibrium acell is the one
for the energy minimum
b) use BFGS minimization algorithm for example, with optcell=2 for zinc blende
structure, or optcell=0 first and then optcell=2 for high pressure structures
of GaAs like Cmcm.
Of course, you should have all the parameters converged before doing 
structural
optimization: ecut, ngkpt, tsmear (for high pressure metallic structures 
only),
tolvrs and tolmxf (for b)) or toldfe (for a)).

On the other hand, check the PHYSICAL REVIEW B 66, 045209 (2002) (for bond
lengths) and references therein (McMahon&Nelmes and Mujica&Needs for example).

Cheers,

Igor Lukacevic