You are on page 1of 1

10.

6 Discussion and Outlook 557

that the ground state is insulating for any U > 0. A metal-insulator


transition occurs at U = 0 but we would not like to call this a Mott
transition because it is not a strong correlation phenomenon: it is an
SDW instability. We can, to some extent, get around this difficulty by
constraining the ground state to be paramagnetic but it would be nicer
to work with a model which really has a Mott transition at finite U.
One possibility would be to make our previous models more compli-
cated by including some second-neighbour hopping: this would destroy
the perfect nesting. In any case, such amendments would be necessary
to make the band term more realistic-looking. Another possibility is to
study the nearest-neighbour hopping model on non-bipartite lattices,
such as the triangular lattice. However, there is little secure knowledge
about the behaviour of these models. It is clearly desirable to have at
least one model, however artificial it may be, where the existence of a
Mott transition at finite U is rigorously known. This could be used its
a testing ground to decide whether, and in which sense, the variational
approach is capable of detecting the Mott transition. Such a model
was constructed by 'Gebhard and Ruckenstein [126]. Consider L sites
arranged equidistantly on the perimeter of a circle with radius L. The
hopping is long-range: electrons can hop between any two sites

j,m=l Q .i
j#m

with the distance-dependent hopping


it (- 1)-
tjm = = tkj (10.45)
djm
being inversely proportional to the chord distance between j and m
L .(j -m)
djm = ; sin
L .
(10.46)

Rigorous arguments show that the Mott transition occurs when U reaches
the bandwidth: U,, = W = 2.t [126].
Let us discuss the ground state of the half-filled ( n = 1) model. The
Gutzwiller trial state is the familiar (10.42), where ISu=o>,the ground

You might also like