An Adjoint Solver for Unsteady Navier-Stokes Flows and Application in Uncertainty Quantification

ORAL

Abstract

We present a parallel adjoint solver for unsteady incompressible Navier-Stokes equation. Backward time-stepping for the time-dependent adjoint equation is achieved using the dynamic checkpointing scheme. This adjoint solver is suitable for very long time integration without knowing the number of time steps a priori. The required computational time and memory is only three to five time the solution of the forward Navier-Stokes equation for tens of thousands of time steps. The adjoint solver is tested on a laminar cylinder vortex shedding calculation at Reynolds number 100. Application of the adjoint solver in propagating uncertainties with large number of random variables is also presented.

Authors

  • Qiqi Wang

    Massachusetts Institute of Technology

  • Frank Ham

    Stanford University

  • Gianluca Iaccarino

    Stanford University

  • Parviz Moin

    Stanford University, CTR, Stanford University