Adjoint-based Sensitivity Analysis of Chaotic Flows using Stabilized March with Entropy-Stable Schemes

  • Thakur, Pranshul (McGill University)
  • Nadarajah, Siva (McGill University)

Please login to view abstract download link

Sensitivity analysis is of significant importance for optimization, uncertainty quantification and mesh adaptation. While the conventional method of linearization and sensitivity analysis is well established for non-chaotic flows, it yields unbounded sensitivities for chaotic flows due to the presence of positive Lyapunov exponents [1]. Since the high-fidelity simulation of turbulent flows is inherently chaotic, it is of interest to obtain accurate sensitivities for such cases. In the past decade, shadowing-based approaches have been developed to compute accurate sensitivities for chaotic dynamical systems. Notable examples include the least-squares shadowing [2], non-intrusive least-squares shadowing [3], stabilized march [4], etc. The stabilized march approach computes the shadow trajectory by specifying suitable boundary conditions for the adjoint differential equation, and is provably convergent to the correct sensitivity with large integration times [4]. The current work uses stabilized march to compute the sensitivity for a three-dimensional test case involving turbulent flow. Comparisons against finite differences verify the accuracy of the approach. In addition, the convective term of the governing Navier-Stokes equation is discretized using the entropy-stable scheme, which discretely satisfies the entropy inequality [5]. An investigation is conducted on the impact of entropy-stable schemes, which are known to lack the local energy stability property [6], on the sensitivity of chaotic flows.