Domain of attraction. Analysis and control via SOS programming. (Q637980): Difference between revisions

The manuscript deals with the important problem of estimating the domain of attraction of an equilibrium point (say, the origin) for solutions of smooth Ordinary Differential Equations (ODEs) of the form \(\dot{x}(t)=f(x(t))\) where \(x(t) \in {\mathbb R}^n\) is a function of time \(t \in \mathbb R\), the dot denotes time derivative, and \(f \in {\mathbb R}^n \to {\mathbb R}^n\) is a given analytic vector function. The author approaches the problem with Lyapunov and convex optimisation techniques: polynomial Lyapunov functions whose sublevel sets approximate the domain of attraction are numerically computed via SemiDefinite Programming (SDP), a broad extension of linear programming to linear matrix inequalities. Instrumental to these computations is the use of polynomial Sum-Of-Squares (SOS) to deal constructively with polynomial positivity conditions related to definiteness of the Lyapunov function and its time-derivative along trajectories. The book is structured into two parts. The first part is a tutorial on (convex) polynomial SOS programming, a branch of SDP (Chapter 1) and its applications to (nonconvex) polynomial optimization (Chapter 2). The second part deals with the estimation of the domain of attraction (defined in Chapter 3), first for polynomial ODEs (Chapter 4), then for uncertain (parametric) polynomial ODEs (Chapter 5) and finally for non-polynomial but analytic ODEs (Chapter 6). The material of the first part can already be found in various other references, see e.g. [\textit{J. B. Lasserre}, Moments, positive polynomials and their applications, Imperial College Press, London, UK (2009)] or [\textit{M. Laurent}, ``Sums of squares, moment matrices and optimization over polynomials'', New York, NY: Springer. The IMA Volumes in Mathematics and its Applications 149, 157--270 (2009; Zbl 1163.13021)]. The second part is the essential contribution of the book, the core development being located in Chapters 4 and 5. The book is a timely and useful contribution to the systems control literature. It contains many explicitly worked examples, with nice and informative visual illustrations, and also pointers to public-domain Matlab codes written by the author. One may however regrets that the author sticks to his own (and sometimes awkward, to this reviewer's taste) notations, and does not follow the mainstream literature on the topic, especially in the opening Chapters 1 and 2. Besides this, the author's approach to the problem of estimating the domain of attraction of polynomial ODEs is essentially dual, in the sense that a Lyapunov certificate of local stability is sought and the corresponding polynomial sublevel set is used as an approximation of the domain of attraction. As a result, optimization problems related to the maximisation of the volume of the approximation (or to the design of a control law to improve the approximation) are formulated as nonconvex bilinear matrix inequalities, with all their inherent numerical difficulties. It is expected that the use of measures and moments, jointly with SDP, as explained in the above references, could pave the way for a more direct, genuinely primal approach to the problem, so that convexity of the related optimization and design problems could be preserved.