Dear Marcel, just by chance, I optimized the same system quite recently. The trend you report is indeed (generally?) present, but apparently to somehow smaller extent: I've got a=4.212, c=7.016 Ang for EnergyShift = 5 mRy; a=4.255, c=6.969 Ang for EnergyShift = 200 mRy (default) The basis I used is the standard PAO, 2ZP for s and p and 2Z (or 2ZP or TZ, it doesn't change much) for Cd4d and Se3d. The phonon frequencies depend somehow on the extension of basis, but not much. In principle, once your basis is large (complete) enough, the effect can be discarded. As usual, everything depends on which properties you want to get, and with which accuracy.
> Hello > I am calculating wurtzite CdSe with the LDA pseudos from the SIESTA page. > Depending on the PAO.energyshift, 5 to 100 meV, I obtain results for the > lattice constant between 7.04 ang and 6.94 ang, respectively. > This is close to the exp. value of 6.99 ang. > The smaller energyshift, should in principle be better, and results in > larger basis size. This is normal; more extended basis provides better variational freedom at larger interatomic distances (or, over larger unteratomic distances) > Which energyshift would you recomment in this case? This is a compromise between precision and performance... > And can anyone explain to me, why the larger basis gives > an overestimation of the lattice constant? This is a good question; my understanding is the following. The optimization of lattice parameter occurs in the course of variational search over overlaps (atom displacements) and occupations of orbitals (the SCF loop in each step). If the orbitals overlap by their maxima, the variational freedom (the chance to reduce the energy by varying the occupations) can be used more efficiently, that is, it improves (a bit) chances to get the equilibrium lattice constant near to where the basis orbitals do better manifest their non-trivial spatial structure. In other words: you should better provide more variational freedom at the place where the chemistry may better make use of it; at another orbitrary place your variational freedom will be spent in vain. This is a general limitation of all atom-centered basis schemes. Out of this, there are two possible issues: i) to use possibly large, flexible, unbiased etc. bases; ii) to tune our limited basis to what we know is a correct result. Even as in principle opposite, the two strategies are in practice often used in combination... Best regards Andrei Postnikov
