A stable gradient algorithm of adaptation using an output signal (Q4025868)

The author considers an error model \[ \begin{cases} \dot e= Ae+ bz^ T(t)[\theta(t)- \theta^*]\\ e_ 1= c^ T e\end{cases}\tag{1} \] where \(e\in\mathbb{R}^ n\), \(z(t)\) is a vector of known inputs, \(\theta^*\) is a vector of unknown parameters and \(\theta(t)\) the vector of adjustable parameters generated by an adaptation law. He proves that if \(A\) is Hurwitz, \((A,c)\) is observable, \(z(t)\) bounded and the adaptation law is defined by the so-called gradient algorithm of fast adaptation, then (1) is asymptotically stable. The gradient algorithm is given by \(\dot\theta= -k\text{ grad }J\), where \(J\) is a suitabble criterion to be minimized.