You are on page 1of 1

164 Ch.

4 Mott 2lansition and Hubbard Model

the limit of extremely strong interaction with U / t = 00.

4.4.1 The Band Limit


Most of the literature about the Hubbard model uses the simple form
(4.13) of the band term, on a D-dimensional hypercubic lattice. There
is a single band with the dispersion relation
D
c(k) = -2t C ~09Icja (4.27)
j=1

where a is the lattice constant. We assume familiarity with the essen-


tial features of the one-, and three-dimensional cases but we will pay
particular attention to D = 2 (the square lattice) and D + 00.
It is conventional to assume that in (4.13), t > 0. Does the sign
of t really matter? We can convince ourselves that it does not if the
lattice is bipartite, i.e., it can be divided into sublattices A and B in
such a manner that the hopping connects an A-site only to B-sites, and
vice versa. Namely, we can redefine the Wannier functions on sublattice
B by incorporating a minus sign: 4(r - Rj) -+ -4(r - Rj) if j E B.
This change of definition'* cannot change the physical properties of the
system but it changes the sign o f t . A glance at (4.27) shows that the
spectrum of the tight binding models with t and -t is the same. Then
for any number of electrons, we find the same ground state energy and
excitation energies, and thus the same free energy. We should remember,
though, that the argument does not work for non-bipartite lattices like
the triangular lattice, and for those the t > 0 and t < 0 models have
different physics.
In the grand-canonical version

(4.28)

Changing n from 0 to 2, p is increasing in a smooth fashion from -2tD


to 2tD. For a non-interacting metal, the charge compressibility (or
l4 We could have argued more formally by performing the canonical transformation
i
& + -cigfor j E B,but no change (ciu + &) for 1 Er A. This leads to %band +
-xbtband.

You might also like