ferromagnetic) kinetic exchange in Sec. 5.1, even though we know that it
would be very difficult to calculate the superexchange coupling strength for a real compound. Subsequent work on the Hubbard model and its cousins made us more cautious. Now we would say that there are a few important phe- nomena which we can understand qualitatively by studying the one- band Hubbard model; there are some which demand two- or multi-band models or an extended Hubbard model; and finally, there are phenom- ena for which the canonical model has not been found yet. Foremost, let us reassert that the insight gained from deriving the antiferromagnetic Heisenberg coupling as the leading interaction term in the large-U effective Hamiltonian (5.36) is correct: the Hubbard model gives a solid basis for understanding antiferromagnet ism, especially the antiferromagnetism of Mott insulators. Quite probably, we can also de- scribe metallic states with an incommensurate magnetic order which arises by a gradual distortion from a “nearby” two-sublattice antifer- romagnetic state. Thus, choosing the right parameter regime, we can in principle understand insulating antiferromagnets like NiO, and per- haps also a spin density wave metal like Cr. These phases are easily found in mean field calculations, and their existence and robustness is corroborated by a variety of more sophisticated techniques. Then what about the ferromagnetism of iron? The T = 0 mean field phase diagram of the cubic lattice Hubbard model (schematically repro- duced in Fig. 4.11) shows an extended ferromagnetic domain around quarter-filling, starting from intermediate coupling strengths; it looks like a good imitation of itinerant ferromagnetism, and it does not look any more fragile than the antiferromagnetic phase. Of course, one has to keep in mind that a mean field solution is obtained by neglecting fluctuations and that in an exact solution, the phase boundaries would be shifted from their Hartree-Fock positions. Still, conventional wisdom would have held that for a three-dimensional system, such shifts would not alter the overall appearance of the phase diagram and, in particu- lar, if there is a ferromagnetic phase in the mean field approximation, then you would expect that it also exists in the exact solution. Well, it is still possible that the cubic lattice Hubbard model is ferromagnetic somewhere, but even if it is, then probably only at much larger values of