We raise and discuss the following question. Why does the spectrum for
the three-band model of Hybertson, Stechel, Schluter, and Jennison, c
laimed not to be approachable by perturbation theory because of rather
large hopping integrals compared to site energy differences, follow p
recisely what would be expected by low-order perturbation theory? The
latter is, for the insulating case, that the low-lying levels are desc
ribable by a Heisenberg Hamiltonian with nearest-neighbor interactions
plus much smaller next-nearest-neighbor interactions and n-spin terms
, n greater than or equal to 4. We first check whether perturbation th
eory actually does not converge, treating the hopping and p-d exchange
terms as perturbations. For the crystal, we find that the first three
terms contributing to the nearest-neighbor exchange coupling J (which
are of third, fourth, and fifth order) increase in magnitude, and are
not of the same sign, i.e., there is no sign of convergence to this o
rder. We also consider the small cluster, Cu2O7, for which we have car
ried out the perturbation series to 14th order; there is still no sign
of convergence. Thus the nonconvergence of this straightforward pertu
rbation theory is convincingly established. Yet the apparent perturbat
ive nature of the spectrum suggests the existence of some perturbation
theory that does converge. The possibility of a particular transforma
tion of the Hamiltonian leading to a convergent perturbation series, t
hereby answering the above question, is discussed. (C) 1996 American I
nstitute of Physics.