brings a factor X if the nearest-neighbour pair (i,j) consists of a d and an
e site, and a factor of 1 otherwise. X 2 1 is a new variational parameter which controls the degree of the d-e pair binding. Approximate analytic [110], and variational Monte-Carlo [457], studies with (10.39) suggest that the inclusion of intersite correlations brings essential improvements, especially in the large-U regime: for n = 1, the ground state energy density is of the order - t 2 / U . One might have hoped that once we have managed to get the energy scale - t 2 / U which is characteristic of Mott insulators, the trial ground state itself can be identified as an insulating state. As it turns out, this is not the case. Fig. 10.7 shows typical results for the band state occu- pation number of the one-dimensional Hubbard model. It has a number of features which indicate a substantial improvement over the piecewise constant behaviour15 shown Fig. 10.1. First, n(k) is a monotonically decreasing function of the band energy c(k). We are inclined to think this reasonable16: the occupation of higher-lying band states looks en- ergetically unfavourable. Second, the leading term in the approximate result [llol (10.41) is basically right, though the coefficient a is not the same as in the exact expansion (5.100). Note, however, the last term: the generalized Ansatz (10.39) describes a metal because n(k) has a finite Fermi step. It appears that by introducing nearest-neighbour d-e correlations, we managed to get a qualitatively correct behaviour to order t/U, but features from order (t/U)' upwards are still associated with a metallic character. We could try to remedy the situation by including farther-neighbour, or many-site, correlations which improve the behaviour in order ( t l ~ ! I ) ~ 16 That was obtained in the Gutzwiller approximation and is strictly correct only for D = 00. Accurate results for finite dimensions [271]reveal a puzzling feature of the Gutzwiller state: n ( k ) is found to be a piecewise increasing function. "Reasonable as it seems, it is not always true. Examples of a markedly non- monotonic behaviour have been found in the case of the U = 00 one-dimensional Hubbard model [310]. Remember that it is the sum of the band and interaction energies that has to be minimized and apparently, it may happen that the electrons can best avoid each other in real space if they preferentially occupy a higher-lying part of the band. All the same, we think that such cases are exceptional. For three- dimensional systems, perturbation theory predicts a monotonically falling n(k) [257].