we have done better by introducing a more complicated trial state? Can
we get a meaningful description of the insulating state? And what about magnetic long-range order? We do not raise here the even more fundamental question whether the (non-degenerate) Hubbard model is good enough. In the previ- ous chapters, we had ample opportunity to convince ourselves that or- bital degeneracy, and the extra terms included in the extended Hubbard model, can be important. However, the simplest Hubbard model should suffice for the discussion of the metal-insulator transition.
10.6.1 Gutzwiller Method: Exact Treatment
For decades, working with the Gutzwiller Ansatz involved approxima- tions of unknown quality (such as the Gutzwiller approximation), or numerical evaluation for finite systems, or series expansions of limited range of validity. Since one of the main objectives was to describe such a subtle phenomenon as the metal-insulator transition, which (if it hap- pens at all) is certain to be associated with non-analytic behaviour at some intermediate value of U ,the lack of secure knowledge was most worrying. The matter was finally settled when Vollhardt and cowork- ers [271]devised a method for carrying out the Gutzwiller variational procedure exactly, for arbitrary interaction strength, and lattice dimen- sionality. Unfortunately, the solution is far too lengthy and complicated to be included here; we have to content ourselves with quoting some of the essential results. Though the scheme works for any dimension, closed-form results can be obtained only for D = 1, and D = 00. It turns out, however, that the character of the results is the same for all finite D. Thus it is enough to cite the D = 1 results to get an impression of the behaviour in finite dimensions. The first step is to express the expectation values in terms of the variational parameter q; the second is to determine the optimum value of q. It is interesting that at half-filling (n= 1) the discontinuity of the band state occupation number nk,, is