学术文献库

Time-lapse microscopy imaging provides direct access to the dynamics of soft and living systems. At mesoscopic scales, such microscopy experiments reveal intrinsic fluctuations, which may have both thermal and non-equilibrium origins. These intrinsic fluctuations, together with measurement noise, pose a major challenge for the analysis of the dynamics of these "Brownian movies". Traditionally, methods to analyze such experimental data rely on tracking embedded or endogenous probes. However, it is in general unclear how to select appropriate tracers; it is not evident, especially in complex many-body systems, which degrees of freedom are the most informative about their non-equilibrium nature. Here, we introduce an alternative, tracking-free approach that overcomes these difficulties via an unsupervised analysis of the Brownian movie. We develop a dimensional reduction scheme that selects a basis of modes based on dissipation, and we subsequently learn the non-equilibrium dynamics in this basis and estimate the entropy production rate. In addition, we infer time-resolved force maps in the system and show that this approach is scalable to large systems, thus providing a potential alternative to microscopic force-probes. After benchmarking our method against a minimal two-beads model, we illustrate its broader applicability with an example inspired by active biopolymer gels.