Absolute stability of nonlinear systems with non-stationary linear part (Q1902751)

The authors consider the system: \[ \dot x = A(t) x(t) + b(t) \xi (t) \tag{1} \] where \(x \in \mathbb{R}^n\), \(\xi \in \mathbb{R}^m\), \(A\) and \(b\) are bounded matrices. The input \(x\) and the output \(\xi\) satisfy the so- called integral square constraint: \(\exists t_0\), such that on an interval \([t_0, t_*]\) the following inequality is true: \[ \int^{t_k}_{t_0} F \bigl( t,x(t), \xi (t) \bigr) dt > - \gamma, \tag{2} \] for some \(\gamma > 0\), as \(t_k \to t_*\). Absolute stability is based on the inequality \[ \bigl |[x, \xi] \bigr |^2 \triangleq \int^\infty_0 \biggl( \bigl |x(t) \bigr |^2 + \bigl |\xi (t) \bigr |^2 \biggr) dt \leq c \biggl[ \bigl |x(0) \bigr |^2 + \gamma \biggr], \] with \(c > 0\), \(\gamma > 0\) and \(F(t,x(t),\;\xi (t)) = {1\over 2} [x^T Gx + 2x^Tg \xi + \xi^T K \xi]\), \(G = G^T\), \(K = K^T\), \(\exists \varepsilon > 0\) such that \(K < - \varepsilon I_m\) (unit \(m \times m\) matrix). Stability is investigated by considering a Lyapunov function \(V = x^T R_0 (t) x\), where \(R_0 (t)\) is bounded positive. (Also \(\exists \delta > 0\) such that \(V \leq - 2F - \delta (|x(t) |^2 + |\xi (t) |^2)\). A Pontryagin-type Hamiltonian \(H(t)\) is introduced, satisfying \(Jz = Hz\), where \[ J = \left[ \begin{matrix} 0 & - I_n \\ I_n & 0 \end{matrix} \right] H = \left[ \begin{matrix} G - g K^{-1} g^T, & A^T - g K^{-1} b^T \\ A^T - b K^{-1} b^T, & - b K^{-1} b^T \end{matrix} \right]. \] Non-oscillation of the system of \(n\)-independent solutions for an \(n\)- dimensional vector system is defined by non-vanishing of the determinant of the corresponding \(n \times n\) state matrix. It is \(t_0\) non- oscillating if this non-oscillation property is valid for all \(t > t_0\), and it is strictly \(t_0\) non-oscillating if this non-oscillation property persists under small deformation of the Hamiltonian with respect to the metric. The system described above is called minimally stable if there exists a piecewise continuous bounded matrix \(m\), such that the feedback vector \(\xi (t) = m(t) x(t)\) is in \(L_2 [0, \infty)\) and \(F(t,x(t)\), \(\xi (t)) \geq 0\) for all \(t, x\). The following theorems are proved: Let us assume that the system described above is minimally stable and the corresponding Hamiltonian system is 0-non-oscillating. Then (A) the system is absolutely stable, (B) there exists a Lyapunov function \(V\) as given above, and (C) there exists a matrix \(R(t)\) satisfying Riccati's equation: \[ \dot R = - (Rb - g) K^{-1} (Rb - g)^T - RA - A^T R + G \] such that all solutions of the linear system \(\dot x = (A(t) + b(t) r(t)^T)\) \(x(t)\) belong to \(L_2 [0, \infty)\), with \(r(t) = (Rb - g) K^{-1}\). The authors also prove that if the matrices \(A,G,K,g\) are all \(T\)- periodic, and the system is minimally stable, then the conditions \((A)\), \((B)\), \((C)\) are equivalent to 0-non-oscillation. The final remarks of this paper show that for high frequencies the averaging property becomes important. If the system is minimally stable, and nonlinear averaging is absolutely stable, then the system is absolutely stable for sufficiently large \(|\omega |\).