next up previous contents
Next: 3 A worked example: Up: 2 Step-by-step Pseudopotential generation Previous: 2.3 Generating the pseudopotential   Contents


2.4 Checking for transferability

A simple way to check for correctness and to get a feeling for the transferability of a PP, with little effort, is to test the results of PP and AE atomic calculations on atomic configurations differing from the starting one. The error on total energy differences between PP and AE results gives a feeling on how good the PP is. Just to give an idea: an error $ \sim$ 0.001 Ry is very good, $ \sim$ 0.01 Ry may still be acceptable. The code ld1.x has a ``testing'' mode in which it does exactly the above operation. You provide the input PP file and a number of test configurations.

You are advised to perform also the test with a basis set of spherical Bessel functions jl(qr). In addition to revealing the presence of ``ghosts'', this test also gives an idea of the smoothness of the potential: the dependence of energy levels upon the cutoff in the kinetic energy is basically the same for the pseudo-atom in the basis of jl(qr)'s and for the same pseudo-atom in a solid-state calculation using PW's.

Another way to check for transferability is to compare AE and pseudo (PS) logarithmic derivatives, also calculated by ld1.x. Typically this comparison is done on the reference configuration, but not necessarily so. You should supply on input:

-
the radius rd at which logarithmic derivatives are calculated (rd should be of the order of the ionic or covalent radius, and larger than any of the rc's)
-
the energy range Emin, Emax and the number of points for the plot. The energy range should cover the typical valence one-electron energy range expected in the targeted application of the PP.
The files containing logarithmic derivatives can be easily read and plotted using for instance the plotting program gnuplot or xmgrace. Sizable discrepancies between AE and PS logarithmic derivatives are a sign of trouble (unless your energy range is too large or not centered around the range of pseudization energies, of course).

Note that the above checks, based on atomic calculations only, do not replace the usual checks (convergence tests, bond lengths, etc) one has to perform in at least some simple solid-state or molecular systems before starting a serious calculation.


next up previous contents
Next: 3 A worked example: Up: 2 Step-by-step Pseudopotential generation Previous: 2.3 Generating the pseudopotential   Contents
Filippo Spiga 2016-04-24