Weakly chiral networks and two-dimensional delocalized states in a weak magnetic field

We study numerically the localization properties of two-dimensional electrons in a weak perpendicular magnetic field. For this purpose we construct weakly chiral network models on the square and triangular lattices. The prime idea is to separate in space the regions with phase action of magnetic field, where it affects interference in course of multiple disorder scattering, and the regions with orbital action of magnetic field, where it bends electron trajectories. In our models, the disorder mixes counterpropagating channels on the links, while scattering matrices at the nodes describe exclusively the bending of electron trajectories. By artificially introducing a strong spread in the scattering strengths on the links (but keeping the average strength constant), we eliminate the interference and reduce the electron propagation over a network to a classical percolation problem. In this limit we establish the form of the disorder-magnetic field phase diagram. This diagram contains the regions with and without edge states, i.e., the regions with zero and quantized Hall conductivities. Taking into account that, for a given disorder, the scattering strength scales as inverse electron energy, we find agreement of our phase diagram with levitation scenario: energy separating the Anderson and quantum-Hall insulating phases floats up to infinity upon decreasing magnetic field. From numerical study, based on the analysis of quantum transmission of the network with random phases on the links, we conclude that the positions of the weak-field quantum-Hall transitions on the phase diagram are very close to our classical-percolation results. We checked that, in accord with the Pruisken theory, presence or absence of time-reversal symmetry on the links has no effect on the line of delocalization transitions. We also find that floating up of delocalized states in energy is accompanied by doubling of the critical exponent of the localization radius. We establish the origin of this doubling within classical-percolation analysis.