Spectral test for the exponential stability (Q2670630)

Consider the \(n\)-th order differential equation \[ A_{0}X^{(n)} +A_{1}X^{(n-1)}+\dots+A_{n-1}\dot{X}+A_{n}X=0, \tag{1} \] where \(A_{0},A_{1},\dots,A_{n}\) are constant \(m\times m\) matrices, the matrix \(A_{0}\) is nonsingular and \(X^{(k)}= \frac{d^{k}X}{dt^{k}}\), \(0 \leq k \leq n\). Let \(S(A)\) denote the spectrum of a matrix \(A\). Consider the \(n\)-th order scalar algebraic equations \[ a_{0}\lambda^{n}+a_{1}\lambda^{n-1}+\dots+a_{n-1}\lambda+a_{n}=0, \tag{2} \] where \(a_{k}\in S(A_{k}), 0 \leq k \leq n\). Since \(A_{0}\) is assumed to be nonsingular, \(a_{0}\neq 0\). For each fixed choice of the coefficients \(a_{0},a_{1},\dots,a_{n}\), Eq. (2) has \(n\) roots. The characteristic equation for Eq. (1) is of order \(nm\). There are at most \(m^{n+1}\) such equations. Let \(\mathcal{R}\) be the set of all roots of (2). This is a nonempty finite set in \(\mathcal{C}\). They state the following: \textbf{Conjecture.} For each \(\epsilon > 0\), there exists a constant \(C(\epsilon)\) such that all solutions \(X(t)\) of the matrix differential equation (1) satisfy the estimate \[ \| X^{(j)}(t)\| \leq C(\epsilon)e^{t(\beta+\epsilon)}\max_{0 \leq k \leq n-1}\| X^{(k)}(0)\| \text{ with }0 \leq t <\infty \tag{3} \] for \(0 \leq j \leq n-1\), where the constant \(\beta\) is given by \(\beta = \max \{ \Re\lambda : \lambda \in \mathcal{R} \}\). If \(\beta < 0\), the matrix differential equation (1) is Lyapunov asymptotically stable and by the estimate (3), we obtain \[ \| X^{(j)}(t) \| \leq C(\epsilon)e^{t(\beta+\epsilon)}\max_{0 \leq k \leq n-1}\| X^{(k)}(0)\| \rightarrow 0 \] as \(t \rightarrow \infty\) for \(0 \leq j \leq n-1\), where the right-hand side tends to zero exponentially provided we take an \(\epsilon > 0\) such that \(\beta +\epsilon < 0\). Using the theory of commutative Banach algebras, the authors establish the estimate (3) for the solutions of the Eq. (1) under an additional assumption that the matrix coefficients are pairwise commuting, that is, \(A_{j}A_{k} = A_{k}A_{j}\) for \(j\neq k\) and \(0 \leq j,k \leq n\). As an example, they consider the matrix differential equation \[ a_{0}X^{(n)}+ a_{1}X^{(n-1)}+\dots+a_{n-1}\dot{X}+a_{n}X=0, \] that is, Eq. (1) with \(A_{k}=a_{k}E\), where \(E\) is the \(m\times m\) identity matrix and \(a_{0},a_{1},\dots,a_{n}\) are complex numbers with \(a_{0}\neq 0\).