a distinguished role to quarter-filling. Though the reason here is almost
trivial, it seems to have a more general validity: even higher-dimensional lattices built of triangular blocks seem to possess a particularly stable pocket of ferromagnetism around n = 1/4. We can actually get an idea of the €0-dependenceof the stabilization energy of the ferromagnetic state from (8.72) in the solution to Problem 8.2. The lesson is that the perturbative result (8.6) goes over smoothly into an expression of the order of the hopping energy. Since for medium or small eo the electrons are no longer confined to the lower chain (Fig. 8.6, left), or to sites 1 and 2 in terms of the toy model, we cannot use an effective Heisenberg model, However, it should be still possible to write down an effective t-J model, where the qualitative behaviour of J is something like (8.72). The “least decorated” version of the triangular strip (which is not covered by the arguments of [392] or [323]) was introduced by Miiller- Hartmann [297]. It is obtained from the model shown in Fig. 8.6 (left) by setting t 2 = ti, €0 = 0, and U = U‘. Interpreting the model as the tl-tz- U Hubbard chain, we find another example of how the no-go theorem by Lieb and Mattis [245] can be circumvented. The point is that switching on the second-neighbour hopping, we have short loops which means an essential deviation from strict one-dimensionality. At the same time, the lattice ceases to be bipartite. A Nagaoka argument [265] shows that the single-hole ground state is fully polarized if t2 is positive17. Let us remember that we met the tight-binding term in Problem 4.3.b where we found that the dispersion has two degenerate minima if t2 > -tr/4. Though the proven existence of a Nagaoka state shows that the model can become ferromagnetic also for 0 < tZ < -t1/4, the possibility of a double-well dispersion seems to help18 the FM phase to extend over an especially large region of the parameter space: it appears that (for sufficiently large U ) the model can become ferromagnetic at all fillings n < 1/2, for a broad range of t 2 > 0. In particular, one finds
“We can gauge t l to be either positive or negative. Here we fix t l < 0.
“Though any naive application of the Stoner criterion should be avoided, it is certainly suggestive that the DOS diverges not only at the band edges but also somewhere inside the band (Fig. 4.13, right). Later, we will realize that the strongly asymmetrical shape of the DOS is more important than the presence of a divergency.