A stabilized march approach to adjoint-based sensitivity analysis of chaotic flows

TL;DR

This paper introduces a stabilized march method for adjoint-based sensitivity analysis of chaotic flows, providing a more efficient and convergent approach compared to traditional shadowing techniques.

Contribution

It proposes a novel stabilized march approach that computes the adjoint shadowing trajectory without solving a minimization problem, improving efficiency and convergence.

Findings

The method converges to true sensitivities for large integration times.

Sensitivity errors are of the order of the local truncation error.

Numerical verification on Lorentz 63 and Kuramoto-Sivashinsky equations confirms effectiveness.

Abstract

Adjoint-based sensitivity analysis is of interest in computational science due to its ability to compute sensitivities at a lower cost with respect to several design parameters. However, conventional sensitivity analysis methods fail in the presence of chaotic flows. Popular approaches to chaotic sensitivity analysis of flows involve the use of the shadowing trajectory. The state-of-the-art approach computes the shadowing trajectory by solving a least squares minimization problem, resulting in a space-time linear system of equations. The current paper computes the adjoint shadowing trajectory using the stabilized march, by specifying the adjoint boundary conditions instead of solving a minimization problem. This approach results in a space-time linear system that can be solved through a single backward substitution of order $\mathcal{O}(n_u^2)$ with $n_u$ being the dimension of the unstable subspace. It is proven to compute sensitivities that converge to the true sensitivity for large integration times and that the error in the sensitivity due to the discretization is of the order of the local truncation error of the scheme. The approach is numerically verified on the Lorentz 63 and Kuramoto-Sivasinsky equations.