Stochastic sensitivity: a computable Lagrangian uncertainty measure for unsteady flows

Abstract

Uncertainties in velocity data are often ignored when computing Lagrangian particle trajectories of fluids. Modeling these as noise in the velocity field leads to a random deviation from each trajectory. This deviation is examined within the context of small (multiplicative) stochasticity applying to a two-dimensional unsteady flow operating over a finite time. These assumptions are motivated precisely by standard availability expectations of realistic velocity data. Explicit expressions for the deviation's expected size and anisotropy are obtained using an Itô calculus approach, thereby characterizing the uncertainty in the Lagrangian trajectory's final location with respect to lengthscale and direction. These provide a practical methodology for ascribing spatially nonuniform uncertainties to predictions of flows, and also provide new tools for extracting fluid regions that remain robust under velocity fluctuations.