[Wien] which optimization is more important, E vs volume or E vs c/a?

[Stefaan Cottenier]

>         For the calculation of impurity doped 2*2*2 Ni3Al (L12 structure)
supercell, which optimization is more important, E vs volume or E vs c/a?

Hard to say, it depends on the property. If the total energy is involved
(for cohesive energy calculations, for instance), I expect E vs volume to be
more important. For an EFG, c/a matter more. But experience shows that often
you are very close to the global minimum if you optimize first the volume,
and then c/a at your minimal volume.

In your case with an impurity, I would follow the following strategy:

1) optimize the volume of pure Ni3Al (not of the supercell).
2) fix the minimal volume, and optimize c/a
3) (if you have time, repeat 1) and 2) starting from the new c/a, until you
find you are in the global minimum)
4) make the supercell, insert the impurity
5) relax the atoms that feel forces due to the impurity (often very
important! -- you didn't mention it in your question)

That is the fastest and quite safe procedure. If you would relax volume and
c/a of the supercell containing the impurity already, it would in principle
be a little bit more accurate: the a=b,c lattice constants that you find in
this way will be somewhat different from the ones of pure Ni3Al, but they
will mimic to some extent the displacements of the atoms that are far away
from the impurity, outside your unit cell. (and here, in principle you have
to relax neighbours at any volume or c/a ratio, very time consuming).

Stefaan