Going forth and back in time: a fast and parsimonious algorithm for mixed initial/final-value problems

We present an efficient and parsimonious algorithm to solve mixed initial/final-value problems. The algorithm optimally limits the memory storage and the computational time requirements: with respect to a simple forward integration, the cost factor is only logarithmic in the number of time-steps. As an example, we discuss the solution of the final-value problem for a Fokker-Planck equation whose drift velocity solves a different initial-value problem -- a relevant issue in the context of turbulent scalar transport.