Numerical integration of a class of ordinary differential equations on the general linear group of matrices (Q1418844)

This paper is concerned with the numerical solution of initial value problems for matrix differential equations on \( { GL}(n)\) with the form \[ Y'(t) = Y^{-T} (t) F( Y(t) ), \quad t \geq t_0,\;Y(t_0)= Y_0, \] where \(f\) is assumed to be a continuous and globally Lipschitz matrix function. By using the theory of ordinary differential equations it can be seen that the solution \( Y(t)\) may have a escape for some finite \( \tau \) and then \( Y(t)\) approaches to a singular matrix when \( t \to \tau. \) The authors propose two alternative approaches to deal with this kind of problems when finite escapes are encountered. In the first one they use a low order explicit Runge-Kutta method on some modified scheme to avoid the unstable behaviour in the neighbourhood of the singularity. In the second one, they use the singular value decomposition of \( Y(t) = U(t) \Sigma (t) V(t)\) to integrate the components of \( U, \Sigma \) and \( V\). The paper ends with the results of a numerical experiment to test the behaviour of both approaches for some low order Runge-Kutta methods.