P\'eclet-number dependence of optimal mixing strategies identified using multiscale norms

Abstract

The optimization of the mixing of a passive scalar at finite P\'eclet number $Pe=Uh/\kappa$ (where $U,h$ are characteristic velocity and length scales and $\kappa$ is the scalar diffusivity) is relevant to many significant flow challenges across science and engineering. While much work has focused on identifying flow structures conducive to mixing for flows with various values of $Pe$, there has been relatively little attention paid to how the underlying structure of initial scalar distribution affects the mixing achieved. In this study we focus on two problems of interest investigating this issue. Our methods employ a nonlinear direct-adjoint looping (DAL) method to compute fluid velocity fields which optimize a multiscale norm (representing the `mixedness' of our scalar) at a finite target time. First, we investigate how the structure of optimal initial velocity perturbations and the subsequent mixing changes between initially rectilinear `stripes' of scalar and disc-like `drops'. We find that the ensuing stirring of the initial velocity perturbations varies considerably depending on the geometry of the initial scalar distribution. Secondly, we examine the case of lattices of multiple initial `drops' of scalar and investigate how the structure of optimal perturbations varies with appropriately scaled P\'eclet number defined in terms of the drop scale rather than the domain scale. We find that the characteristic structure of the optimal initial velocity perturbation we observe for a single drop is upheld as the number of drops and $Pe$ increase. However, the characteristic vortex structure {\color{red} and associated mixing exhibits some nonlocal variability,} suggesting that rescaling to a local $Pe$ {\color{red} will} not capture all the significant flow dynamics.