Diffusion approximation in turbulent two-particle dispersion

We solve an inverse problem for fluid particle pair statistics: we show that a time sequence of probability density functions (PDFs) of separations can be exactly reproduced by solving the diffusion equation with a suitable time-dependent diffusivity. The diffusivity tensor is given by a time integral of a conditional Lagrangian velocity structure function, weighted by a ratio of PDFs. Physical hypotheses for hydrodynamic turbulence (sweeping, short memory, mean-field) yield simpler integral formulas, including one of Kraichnan and Lundgren (K-L). We evaluate the latter using a space-time database from a numerical Navier-Stokes solution for driven turbulence. The K-L formula reproduces PDFs well at root-mean-square separations, but growth rate of mean-square dispersion is overpredicted due to neglect of memory effects. More general applications of our approach are sketched.