to finite temperatures, which yields fairly good estimates of the critical
temperature, and allows even the derivation of a Curie-Weiss-like sus- ceptibility in the paramagnetic state of certain band ferromagnets like bcc Fe [374]. The relativistic version of LSDA gives promising results for magnetic anisotropy. Since LSDA seems to be a theory of almost everything, why do not we simply regard it as the theory of magnetism? The answer is twofold. First, LSDA (at least, in its conventional form) tends to break down for strongly correlated systems: it predicts that Mott insulators like COO or LazCuO4 should be metals [306], and it needs a phenomenological input parameter to describe the large effective mass of heavy fermion compounds. Since our main interest is in Mott phenomena, we would not wish to rely on density functional theory alone35. However, this is only the lesser of our reasons to stick to the Hub- bard model and its likes. The rapid development of density functional theory may soon make our first objection unfounded. Still, we can argue that even when we will have a complete, ab initio description of, say V203, giving its structure, lattice constants, orbital and spin order all coming out from a single calculation, theories of simple Hubbard-like models will not become obsolete. The point is the following: in LSDA theory, the primary quantity is the electron density, and orbitals play a merely auxiliary role. It turns out that for a sufficiently accurate expansion of the ground state density no(r), one may need dozens of orbitals per lattice site, well beyond the number which is “reasonably expected” in a tight binding scheme. Indeed, the local basis one uses is not the tight binding basis at all, but a set of functions chosen for quick convergence of the calculation; the local basis functions have no imme- diate physical meaning. Correspondingly, if we wanted to represent the Coulomb interaction in the local basis, there would be a great many in- trasite matrix elements, and we still would have the long-range part to describe! Reformulated in terms of localized orbitals, LSDA would be a many-parameter theory36, with a large local basis. Apparently, “the
35We do not attempt to do justice to later developments, such as the LDA+U
approach, which are supposed to cover the Mott-Hubbard gap and other strong correlation features. 3eRemember that those many parameters would not be arbitrarily chosen but