The convergence of the solution of a matrix Riccati equation to the maximal stationary solution in the critical case (Q1874844)

Consider the optimal control problem for the linear system \[ x=Ax + Bu,\quad y = Lx,\quad x(0) = x_0 \tag{1} \] (\(x\in\mathbb R^n\) is the state vector, \(y\in\mathbb R^m\), \(m\leq n\), is the observed output, \(u\in\mathbb R^r\) is the control, and \(x_0\) is the initial condition) with the quadratic objective functional \[ J(u) = y^{\text{T}}(T)Gy(T) + \int_0^T\left[y^{\text{T}}(t)Vy(t) + u^{\text{T}}(t)S u(t)\right]dt, \tag{2} \] where \(S =S^{\text{T}}> 0\), \(G = G^{\text{T}}\geq 0\), and \(V = V^{\text{T}}\geq0\). It is known that the minimum of the functional (2) on trajectories of system (1) is provided by the control \linebreak \(u =-S^{-1}B^{\text{T}} R(t)x (t)\), where \(R(t)\) is the solution of the Cauchy problem for the matrix Riccati differential equation \[ R = -A^{\text{T}} R -RA + RDR - C,\quad R(T) = Q;\tag{3} \] here \(D = BS^{-1}B^{\text{T}}\), \(Q = L^{\text{T}} GL\), and \(C = L^{\text{T}} VL\). The present paper deals with the analysis of the equation (3) under the assumption that \(A\), \(C\), \(D\), and \(Q\) are arbitrary constant \(n\times n\) matrices; moreover, \(C\), \(D\), and \(Q\) are symmetric. Under some known conditions, the set \(\mathcal H\) of symmetric stationary solutions of (3) contains the so-called minimal and maximal stationary solutions \(R^-\) and \(R^+\) such that \(R^-\leq R\leq R^+\) for any \(R\in\mathcal H\) and all eigenvalues of the matrices \(A-DR^+\) and \(A-DR^-\) have nonpositive and nonnegative real parts, respectively. The main result of the paper is a sufficient condition at which \(R(t)\to R^+\) as \(t\to -\infty\).