The convergence analysis of an accelerated iteration for solving algebraic Riccati equations (Q2068332)

The author considers the discrete-time algebraic Riccati equation written as \(X=R_{d}(X):=H+A^{\ast }X(I+GX)^{-1}A\), where \(A\in \mathbb{C}^{n\times n}\), \(G\) and \(H\) are positive and semidefinite matrices of size \(n\times n\), \(X\) is the unknown \(n\times n\) and Hermitian matrix, and \(A^{\ast }\) stands for the complex conjugate transpose of \(A\). For the resolution of this problem, iterative algorithms are proposed, among which the Newton and structure-preserving doubling ones. A Hermitian solution \(X\) of the above equation is called stabilizing (respectively, almost stabilizing) if all the eigenvalues of the matrix \(T_{X}=(I+GX)^{-1}A\) are in the open (respectively, closed) unit disk. The purpose of the paper is to introduce conditions leading to an accelerated fixed-point iteration convergence (AFPI) result. The main result shows that, under certain conditions, the quadratic convergence of AFPI still holds in the critical case and even if \(\rho (T_{X})>1\), where \(X_{+}\) is an almost stabilizing solution to the above Riccati equation. The author recalls the notions of \(R\)-linear and \(R\)-superlinear convergence of a sequence \( \{X_{k}\}\subset \mathbb{C}^{n\times n}\) to its limit, that of order-preserving property for a matrix operator on the set \(\mathbb{P}_{n}\) of positive and semidefinite matrices. He also revisits convergence properties of the sequence \(\{X_{k}\}\subset \mathbb{C}^{n\times n}\) defined through the iteration scheme \(X_{k+1}=F(X_{k})\) for a matrix operator \(F\) which is order-preserving in the set \(\mathbb{H}_{n}\) of Hermitian matrices. The first main result proves that if \(H\geq 0\) and \(G\geq 0\), any positive definite solution \(X\) of the above equation is an almost stabilizing solution. Furthermore, \(X\) is a stabilizing solution if \(H>0\). The second main result proves that if \(\mathbb{R}_{\geq }:=\{X\in \mathbb{N}_{n}\mid X\geq R_{d}(X)\}\neq \varnothing \) and \(H\neq 0\), the sequence \( X_{k+1}=R_{d}(X_{k})\) starting from the initial matrix \(X_{1}=H\) converges \(R \)-linearly to the minimal positive and semidefinite solution \(X\) of the above equation. Moreover, the author gives further details on the convergence rate of this fixed-point algorithm. The author finally present numerical results and examples concerning the resolution of this discrete-time algebraic Riccati equation.