Summary: We solve in this paper the problem of the antiferromagnetic transition for the -state Potts model (defined geometrically for generic using the loop/cluster expansion) on the square lattice. This solution is based on the detailed analysis of the Bethe ansatz equations (which involve staggered source terms of the type “real” and “anti-string”) and on extensive numerical diagonalization of transfer matrices. It involves subtle distinctions between the loop/cluster version of the model, and the associated RSOS and (twisted) vertex models. The essential result is that the twisted vertex model on the transition line has a continuum limit described by two bosons, one which is compact and twisted, and the other which is not, with a total central charge for . The non-compact boson contributes a continuum component to the spectrum of critical exponents. For generic, these properties are shared by the Potts model. For a Beraha number, i.e., with integer, and in particular integer, the continuum limit is given by a “truncation” of the two boson theory, and coincides essentially with the critical point of parafermions .
Moreover, the vertex model, and, for generic, the Potts model, exhibit a first-order critical point on the transition line – that is, the antiferromagnetic critical point is not only a point where correlations decay algebraically, but is also the locus of level crossings where the derivatives of the free energy are discontinuous. In that sense, the thermal exponent of the Potts model is generically equal to . Things are however profoundly different for a Beraha number. In this case, the antiferromagnetic transition is second order, with the thermal exponent determined by the dimension of the parafermion, . As one enters the adjacent “Berker-Kadanoff” phase, the model flows, for odd, to a minimal model of CFT with central charge , while for even it becomes massive. This provides a physical realization of a flow conjectured long ago by Fateev and Zamolodchikov in the context of integrable perturbations.
Finally, though the bulk of the paper concentrates on the square-lattice model, we present arguments and numerical evidence that the antiferromagnetic transition occurs as well on other two-dimensional lattices.