A necessary condition for boundary sensitivity of attractive non-linear stochastic cellular automata in Z2

This paper tackles the question of the environmental robustness of a particular class of two-dimensional finite threshold Boolean cellular automata when they are subjected to distinct fixed boundary instances. More precisely, focusing on a non-linear stochastic version of the classical threshold function governing the evolution of formal neural networks, we show the existence of a necessary condition under which attractive cellular automata of this form become boundary sensitive, i.e., we highlight a condition without which a cellular automaton hits the same asymptotic dynamical behaviour whatever its boundary conditions are. To go further, we give an explicit formula for this necessary condition.