PCTS Special Physics Seminar

Abstract: I will discuss recent work on nonequilibrium systems that store information in the value of a local order parameter for thermodynamically long time scales, even in the presence of generic symmetry-breaking noise. Before this work, only one simple example of such a system was known, discovered by computer scientists in 1969. In this talk, I will present a general approach for constructing new examples, describe a simple machine-learning framework that enables many more to be discovered, and overview the novel nonequilibrium physics we have observed in these systems. Time permitting, I will also discuss ongoing work that uses these methods to reveal intriguing behavior in the physics of spatially-local diffusion models. This talk will be mostly based on arxiv:2509.20730 and arxiv:2508.15726. An interactive demo of some of the results is available at https://memorynca.github.io/2D/.