viewpoint of applicability of the one-parameter scaling. It is shown that the distribution
becomes nonuniversal and the scaling description fails in d= 2 dimensions and in the critical
region in d> 2 dimensions. It is argued that on approaching the region of strong localization
(d= 2) or Anderson transition (d> 2) a crossover should exist from a close-to-gaussian
distribution with logarithmically normal tails to a completely logarithmically normal …