Finding magnetic sites with (Sj) # 0 means that we have to do with a symmetry-breaking ground state of the Hubbard Hamiltonian. This should be understood as follows: For finite quantum-mechanical systems, it is expected that the ground state, as well a,a finite-T equilibrium states, show the full symmetry of the Hamiltonian. Consequently, for a spin-rotationally invariant Hamiltonian, such as the Hubbard model, expectation values like (7.1) should vanish. In finite systems, we cannot really speak about magnetic (or any other kind of) order4. This may sound as if the understanding of spontaneous ordering were denied t o us because strictly speaking, all systems are finite. However, macroscopic systems are so large that it can be argued that their behaviour is essentially the same 89 it would be for an infinitely large system. Technically speaking, one must take the thermodynamic limit: say, for a system with a given density, we let the particle number N -+ 00, and the volume V -+ 00, with N / V kept fixed. For lattice models, as in our present discussion, the filling has to be kept fixed: we let N -+ 00, and the number of lattice sites L -+ 00, with n = N/L fixed. The important consequence of this step is that we gain a principal understanding of ordering transitions: in the thermodynamic limit, symmetry breaking becomes possible. The way it comes about is this: For a finite system, the free energy is an analytic function of the temperature T for all finite T,so the behaviour of the system must be perfectly continuous, there can be no phase transition. In contrast, in the thermodynamic limit the free energy can show non-analytic behaviour at some finite transition temperature Tt,. This has the effect that the behaviour at T < T,, cannot be understood by smooth continuation from T > Ttr; the ordered state necessitates a new method of description. Approximate theories (such as the mean field theories we are going to discuss) are usually not sensitive to the difference between finite and infinite systems, and may predict magnetic long-range order even for small systems5. We understand that such a result must be spurious, and that the
4Moreprecisely: the ground.state belongs to one of the irreducible representations
of the symmetry group of the Hamiltonian. If the ground state is non-degenerate, we expect this to be the identity representation, so the ground state is fully symmetrical; this is the case relevant for systems with a tendency to antiferromagnetic ordering. Systems inclined to become ferromagnetic can have a high-spin ground state, i.e., several degenerate ground states which can be rotated into each other. Then you might say that any of the individual ground states breaks spin-rotational invariance even for a finite system; recall the case of the 0 2 molecule. However, the thermal equilibrium state of the system (includingT + 0 where we average over all degenerate ground states) remains fully symmetrical. 'We will see an example of this in Problem 7.6 (see p. 385 and p. 413).