Towards a MATLAB Toolbox to compute backstepping kernels using the power series method

Xin Lin, Rafael Vázquez, Miroslav Krstić

Abstract

In this paper, we extend our previous work on the power series method for computing backstepping kernels. Our first contribution is the development of initial steps towards a MATLAB toolbox dedicated to backstepping kernel computation. This toolbox would exploit MATLAB's linear algebra and sparse matrix manipulation features for enhanced efficiency; our initial findings show considerable improvements in computational speed with respect to the use of symbolical software without loss of precision at high orders. Additionally, we tackle limitations observed in our earlier work, such as slow convergence (due to oscillatory behaviors) and non-converging series (due to loss of analiticity at some singular points). To overcome these challenges, we introduce a technique that mitigates this behaviour by computing the expansion at different points, denoted as localized power series. This approach effectively navigates around singularities, and can also accelerates convergence by using more local approximations. Basic examples are provided to demonstrate these enhancements. Although this research is still ongoing, the significant potential and simplicity of the method already establish the power series approach as a viable and versatile solution for solving backstepping kernel equations, benefiting both novel and experienced practitioners in the field. We anticipate that these developments will be particularly beneficial in training the recently introduced neural operators that approximate backstepping kernels and gains.

Introduction

Backstepping is a powerful technique for stabilizing systems governed by Partial Differential Equations (PDEs). However, its implementation relies on solving "kernel equations"—a set of PDEs defined on a triangular domain. Traditionally, these are solved using symbolic software or finite difference methods, both of which have significant drawbacks in terms of speed or handling specific boundary conditions. This work introduces a high-performance MATLAB framework that leverages power series expansions to solve these equations efficiently.

Methodology: From Symbols to Sparse Matrices

The core innovation lies in transitioning the power series method from a symbolic exercise into a numerical linear algebra problem. By representing power series coefficients as vectors and differential operators as sparse matrices, the kernel computation is reduced to solving a linear system $M\kappa = b$.

Sparse Matrix Optimization

By utilizing MATLAB’s optimized sparse matrix routines, the researchers achieved a speedup of approximately 18,000x compared to symbolic solvers like Mathematica. For a 50th-order approximation, the MATLAB solver took less than 0.01 seconds, whereas the symbolic counterpart required nearly three minutes. This efficiency allows for high-order approximations (up to $N=100$) that were previously computationally prohibitive.

Convergence of the kernel approximation $K(L, \xi)$ for increasing series order $N$.

Localized Power Series

Standard power series expansions around the origin often fail when encountering singularities or non-analytic coefficients. To address this, the authors introduced localized power series. By expanding the series around strategically chosen points $(x_0, \xi_0)$, the solver can "dodge" singularities and maintain convergence in regions where global expansions would diverge. This is particularly useful for systems with space-varying parameters or discontinuous kernels.

Complex multi-kernel solutions for coupled hyperbolic systems, demonstrating the toolbox's versatility.

Enabling Neural Operators

A primary motivation for this toolbox is the generation of large-scale datasets. Modern "neural operators" (like DeepONets) can learn to approximate backstepping kernels for real-time control but require thousands of high-fidelity training examples. The speed and robustness of this power series approach make it an ideal candidate for such data generation tasks.

Conclusion and Future Work

The power series method, when implemented via sparse matrix algebra and localized expansions, provides a fast and robust solution for backstepping kernel equations. Future work will focus on "patched power series," using multiple overlapping expansions to handle even more complex, highly oscillatory, or singular domains.

Summary updated

Abstract

In this paper, we extend our previous work on the power series method for computing backstepping kernels. Our first contribution is the development of initial steps towards a MATLAB toolbox dedicated to backstepping kernel computation. This toolbox would exploit MATLAB's linear algebra and sparse matrix manipulation features for enhanced efficiency; our initial findings show considerable improvements in computational speed with respect to the use of symbolical software without loss of precision at high orders. Additionally, we tackle limitations observed in our earlier work, such as slow convergence (due to oscillatory behaviors) and non-converging series (due to loss of analiticity at some singular points). To overcome these challenges, we introduce a technique that mitigates this behaviour by computing the expansion at different points, denoted as localized power series. This approach effectively navigates around singularities, and can also accelerates convergence by using more local approximations. Basic examples are provided to demonstrate these enhancements. Although this research is still ongoing, the significant potential and simplicity of the method already establish the power series approach as a viable and versatile solution for solving backstepping kernel equations, benefiting both novel and experienced practitioners in the field. We anticipate that these developments will be particularly beneficial in training the recently introduced neural operators that approximate backstepping kernels and gains.
BibTeX
@misc{lin2024towards,
  author = {Xin Lin and Rafael Vázquez and Miroslav Krstić},
  title = {{Towards a MATLAB Toolbox to compute backstepping kernels using the power series method}},
  year = {2024},
  eprint = {2403.16070},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2403.16070},
}
Figure