On matrix functions which commute with their derivative (Q1064382)

Matrix functions which commute with their derivatives play an important role in the solution of systems of linear differential equations with time-varying coefficients. The present paper generalizes some results on matrix functions published during 1950-1982 in the sense that the independent variable t is not only defined on an open interval of \({\mathbb{R}}\) but constitutes an open subset of a Banach space on \({\mathbb{R}}\) or \({\mathbb{C}}\). With the aid of a condition for a finite-dimensional vector subspace of a normed space not to depend on t it is shown that the spectral subspaces of a matrix function commuting with its derivative are constant. Such a matrix function turns out to be decomposable, on a neighborhood of any point \(t_ 0\), into matrix functions of the same class but having only one eigenvalue at \(t_ 0\). As a consequence pointwise diagonalizable matrix functions commuting with their derivatives, whose eigenvalues do not bifurcate, are globally diagonalizable. Further, it is proved that any cyclic matrix function commuting with its derivative is commutative of the polynomial form. A sufficient condition states that under certain hypotheses a matrix function commuting with its derivative is a polynomial in a constant matrix. The matrix function then has only eigensubspaces of extremal dimension. If a matrix function A is globally diagonalizable into submatrix functions and if each of them is a polynomial in a constant matrix., then A itself is a polynomial in a constant matrix. As a last result it is shown that if a matrix function has only eigensubspaces of extremal dimensions, satisfies some conservative hypotheses, and commutes with its derivative, then it is commutative of the polynomial form.