ZZ PWSCF论坛：
Dear Eduardo,
in this order, I would consider
1) wfc cutoff. For a reasonable ultrasoft psp should be between
25 and 30 Ry, and unlikely to be more than 40 Ry.
2) dual (i.e. ecutrho). 6 to 12 times the wfc cutoff.
3) k-points - you might want to use m-p meshes containing gamma,
since the phonon q-mesh will also need to contain gamma (last
I checked, the q-to-r and r-to-q interpolators require that
you have gamma in your q mesh). Usually shifted k-point meshes
are slightly better for convergence, and I would use those if
only a total energy and force calculation was needed.
4) degauss. I tend to use ~0.5eV for the cold smearing (note that
in the code degauss is given in Ry). Something slightly larger
(up to 0.7eV) can be used for certain systems (the larger the
smearing, the fewer the k-points needed to achieve convergence).
Occasionally you might need to go down to ~0.3eV. It's system
dependent (how complex the fermi energy surface is, how parabolic
the dos is...). Note that this smearings are good for integrated
quantities (like the forces); if you needed to study a propery dependent
only on the fermi surface, you would need to be more carefull.
5) convergence in scf cycle. As accurately ocnverged as the code
would go....
6) convergence in the nscf/phonon. You might want to try a few tests here.
In addition, you need to make sure that all your atomic positions
are accuarately relaxed. Volume is often the theoretical volume,
but there are cases in which it might be wise to use the
experimental parameters (e.g. the c/a in graphite - we have
a recent preprint on arxiv/cond-mat on the matter).
At the end, you want a convergence in the frequencies with not more
than 1% error due to all the parameters above, so you can roughly
estimate (from sqrt(k/m), where k is the force constant, and
you know the experimental frequencies) the error on the forces that
you can tolerate.
Anyone on the list with additional suggestions is welcome to add them...
All the best,
阅读(2133) | 评论(0) | 转发(0) |
