2008/4/15, Marcel Mohr <[EMAIL PROTECTED]>: > > > Then I get nearly linear behaviour PAOshift vs. Total energy, with 5 meV > PAO shift gives lowest energy (for my range 5 tp 100 meV).
Actually, as Dr. Postnikov correctly pointed out, I made a mistake here. The problem indeed is variational, but the optimal value is known in advance, it's 0.0 eV (i.e., non-localized basis) :) So, you have to use some other criteria (most probably, again, the balance of cost vs. performance). > Do you mean by that, the fact, that a larger basis gives larger lattice > constant? No, I meant that the overestimated lattice constant in the 'delocalized' limit is larger than the experimental value. And, as Dr. Postnikov pointed out, the more nonzero overlap matrix elements there are, the greater the variational freedom and hence, the lower the energy, which means that there's a bias 'force' favoring more compact structures that have more overlapping pairs of orbitals. Which explains the increase of l.c. with increasing localization radius (this effect becomes weaker and weaker).
