Computing Nonequilibrium Responses with Score-shifted Stochastic Differential Equations

Abstract

Using equilibrium fluctuations to understand the response of a physical system to an externally imposed perturbation is the basis for linear response theory, which is widely used to interpret experiments and shed light on microscopic dynamics. For nonequilibrium systems, perturbations cannot be interpreted simply by monitoring fluctuations in a conjugate observable -- additional dynamical information is needed. The theory of linear response around nonequilibrium steady states relies on path ensemble averaging, which makes this theory inapplicable to perturbations that affect the diffusion constant or temperature in a stochastic system. Here, we show that a separate, ``effective'' physical process can be used to describe the perturbed dynamics and that this dynamics in turn allows us to accurately calculate the response to a change in the diffusion. Interestingly, the effective dynamics contains an additional drift that is proportional to the ``score'' of the instantaneous probability density of the system -- this object has also been studied extensively in recent years in the context of denoising diffusion models in the machine learning literature. Exploiting recently developed algorithms for learning the score, we show that we can carry out nonequilibrium response calculations on systems for which the exact score cannot be obtained.