Feedback control methodologies for nonlinear systems (Q5925731)

In the first part of the paper the authors recall in a succinct manner five distinct numerical methods in the literature for computing the optimal feedback control \(u^*(.):\Omega \to R^k\) of an infinite-horizon autonomous problem that consists in the minimization of the functionals: \[ J(x_0;u(.)):=\int_0^\infty [(x(t))^TQx(t)+(u(t))^TRu(t)]dt, \;x_0\in \Omega \;\subset R^n, \] subject to: \[ x'(t)=f(x(t))+Bu(t), \;u(t)\in R^k \text{ a.e. }([0,\infty)), \;x(0)=x_0. \] In the second part, one uses tables and figures to report computer numerical results of each of the five methods on four examples taken from the same bibliographical sources; moreover, comparing the performance of the methods regarding computational time, overall cost and ``flops'', the authors reach some conclusions and make suitable recommendations for choosing the ``best'' method for each type of problem, depending on certain characteristics such as ``the nature of nonlinearities'', the ``size'' of the system, etc.. One may note that, as in most recent papers on numerical methods in optimal control, very little of the abundant ``optimal control theory'' [e.g., \textit{M. Bardi} and \textit{I. Capuzzo-Dolcetta}, ``Optimal control and viscosity solutions of Hamilton-Jacobi-Bellman equations'' (1997; Zbl 0890.49011)]) is actually used; for instance, in the first three methods it is (implicitly) assumed that the value function: \[ V(x_0):= \inf_{u(.)}J(x_0;u(.)), \quad x_0\in \Omega, \] is smooth (even analytical) and the optimal feedback control is given by the (Carathéodory's) formula: \[ u^*(x):=-(1/2)R^{-1}B^TDV(x), \quad x\in \Omega, \] while in the last two methods it is assumed (also implicitly) that any ``normal extremal'' satisfying the transversality condition is optimal. Similar results and examples may be found in [\textit{W. Prager}, Calcolo 37, No. 2, 97-123 (2000; Zbl 0956.65050)] while a rather complex but rigorous theory of \(\varepsilon\)-optimal feedback control may be found in [\textit{H. Ishii} and \textit{S. Koike}, Ann. Inst. Henri Poincaré, Anal. Non Linéaire 17, No. 4, 473-502 (2000; Zbl 0969.49019)].