Abstract

Fluid mixing involves the interplay between advection and diffusion, which together cause any initial distribution of passive scalar to homogenize and ultimately reach a uniform state. However, this scenario only holds when the velocity field is non-divergent and has no normal component to the boundary. If either condition is unmet, such as for active particles in a bounded region, floating particles, or for filters, the ultimate state after a long time is not uniform, and may be time dependent. We show that in those cases of nonuniform mixing it is preferable to characterize the degree of mixing in terms of an f-divergence, which is a generalization of relative entropy. Unlike concentration variance (L2 norm), the f-divergence always decays monotonically, even for nonuniform mixing, which facilitates measuring the rate of mixing. We show by an example that flows that mix well for the nonuniform case can be drastically different from efficient uniformly mixing flows. We also discuss Some connections to the acceleration of convergence of Markov chains on graphs.