Dynamical mean-field theory for non-reciprocal spin-glasses

Abstract

In out-of-equilibrium systems, the lack of reciprocity in interactions is more the rule than the exception. Non-reciprocal interactions arise generically in out-of-equilibrium systems, such as metamaterials, neural networks, or ecosystems. In the context of glassy systems, it is known that they are crucial in the process of learning in neural networks but their role in glassy dynamics is still widely debated. In this work, we develop a generalization of a dynamical mean-field theory of spinglass models which includes non-reciprocal interactions among spins, with full analytical detail. Furthermore, we show how the dynamics of mean-field spin-glasses are quantitatively and qualitatively modified when considering non-reciprocal interactions, focusing on the high-temperature relaxational dynamics. Our theory predicts critical slowing down of the dynamics and glass melting when considering weakly non-reciprocal interactions, although we suspect that new physics can be further explored beyond that limit