Sensitivity analysis of chaotic systems using a frequency-domain shadowing approach

TL;DR

This paper introduces a frequency-domain shadowing method for sensitivity analysis of chaotic systems, overcoming limitations of traditional adjoint methods and enabling application to large, complex systems like turbulent flows.

Contribution

The paper reformulates the least-square shadowing method in Fourier space using harmonic balancing, reducing computational costs and enabling large-scale system analysis.

Findings

Method accurately computes sensitivities for the Kuramoto-Sivashinski system.

Cost of the new method is independent of positive Lyapunov exponents.

Proposed iterative approach reduces storage and computational requirements.

Abstract

We present a frequency-domain method for computing the sensitivities of time-averaged quantities of chaotic systems with respect to input parameters. Such sensitivities cannot be computed by conventional adjoint analysis tools, because the presence of positive Lyapunov exponents leads to exponential growth of the adjoint variables. The proposed method is based on the least-square shadowing (LSS) approach [1], that formulates the evaluation of sensitivities as an optimisation problem, thereby avoiding the exponential growth of the solution. However, all existing formulations of LSS (and its variants) are in the time domain and the computational cost scales with the number of positive Lyapunov exponents. In the present paper, we reformulate the LSS method in the Fourier space using harmonic balancing. The new method is tested on the Kuramoto-Sivashinski system and the results match with those obtained using the standard time-domain formulation. Although the cost of the direct solution is independent of the number of positive Lyapunov exponents, storage and computing requirements grow rapidly with the size of the system. To mitigate these requirements, we propose a resolvent-based iterative approach that needs much less storage. Application to the Kuramoto-Sivashinski system gave accurate results with very low computational cost. The method is applicable to large systems and paves the way for application of the resolvent-based shadowing approach to turbulent flows. Further work is needed to assess its performance and scalability.