Comment on "Antilocalization in a 2D Electron Gas in a Random Magnetic Field"

In a recent Letter, Taras-Semchuk and Efetov reconsider the problem of electron localization in a random magnetic field in two dimensions. They claim that due to the long-range nature of the vector potential correlations an additional term appears in the effective field theory ($\sigma$-model) of the problem, leading to delocalization at the one-loop level. This calls into question the results of earlier analytical studies, where the random magnetic field problem was mapped onto the conventional unitary-class $\sigma$-model, implying that the leading quantum correction is of two-loop order and of a localizing nature. We show in this Comment, however, that the new term in fact does not exist and was erroneously obtained by Taras-Semchuk and Efetov because of an inconsistent treatment violating gauge invariance.