Geometry-resolved avalanche criterion in disordered spin systems: transverse connectivity corrections and asymptotic recovery of the flat-front law
Hikaru Wakaura, Taiki Tanimae
Abstract
Abstract Isolated quantum many-body systems can fail to thermalize in the presence of strong disorder, a phenomenon known as many-body localization (MBL). Whether this breakdown of statistical mechanics survives in the thermodynamic limit is believed to be decided by quantum avalanches: rare thermal inclusions that grow by consuming their localized surroundings, a mechanism that renders MBL unstable in more than one dimension. Quantitative tests of the avalanche criterion have so far been confined to strictly one-dimensional chains, leaving the geometry dependence at the heart of the higher-dimensional argument unmeasured. Here we show, using a bath-coupled relaxation-rate diagnostic validated against exact Lindblad dynamics, that the avalanche stability criterion acquires systematic and lawful geometry corrections as a disordered XXZ system is widened from a chain to two-, three-, and four-leg strips and to quasi-three-dimensional tubes. In contrast to the flat-front expectation, in which geometry enters only through the cross-sectional area, we find that the measured decay slope of the slowest relaxation rate decreases with transverse connectivity: the slope deficit is organized, to leading approximation, by the number of transverse bonds per column, and geometries containing closed transverse loops retain their deficits up to the strongest disorder accessible. At the same time, all deficits vanish as disorder grows, so that the area-only avalanche criterion is recovered asymptotically. These results promote the avalanche instability argument from a scaling estimate to a measured, geometry-resolved law, explain why apparent critical disorder grows with system width more slowly than analytic predictions, and provide the quantitative framework needed to interpret the system-size drift of the MBL crossover observed in two-dimensional quantum simulators.