$\mathcal{U}$-bootstrap percolation: critical probability, exponential decay and applications

Abstract: Bootstrap percolation is a wide class of monotone cellular automata with random initial state. In this work we develop tools for studying in full generality one of the three universality' classes of bootstrap percolation models in two dimensions, termed subcritical. We introduce the new notion ofcritical densities' serving the role of `difficulties' for critical models, but adapted to subcritical ones. We characterise the critical probability in terms of these quantities and successfully apply this link to prove new and old results for concrete models such as DTBP and Spiral as well as a general non-trivial upper bound. Our approach establishes and exploits a tight connection between subcritical bootstrap percolation and a suitable generalisation of classical oriented percolation, which will undoubtedly be the source of more results and could provide an entry point for general percolationists to bootstrap percolation. Furthermore, we prove that above a certain critical probability there is exponential decay of the probability of a one-arm event, while below it the event has positive probability and the expected infection time is infinite. We also identify this as the transition of the spectral gap and mean infection time of the corresponding kinetically constrained model. Finally, we essentially characterise the noise sensitivity properties at fixed density for the two natural one-arm events. In doing so we answer fully or partially most of the open questions asked by Balister, Bollob\'as, Przykucki and Smith (2016)---namely we are concerned with their Questions 11, 12, 13, 14 and 17.