Higher order implicit multistep methods for matrix differential equations (Q679261)

This paper deals with the numerical solution of initial value problems for first order matrix differential equations by means of linear multistep methods. Assuming that the matrix differential equation is given by \(Y'(t)= f(t,Y(t))\), where \(Y\) and \(f\) are complex \(p \times q\) matrices with some smoothness assumptions, and denoting by \(Y_j\) the approximation to \(Y(t_j)\), the \(k\) step linear multistep methods under consideration have the form \(\sum^k_{j=0} (A_jY_{n+j} -hB_jY_{n+j}'-h^2C_jY_{n+j}'' =0\) where \(A_j\), \(B_j\), \(C_j\) are constant complex \(p\times p\) matrices that define the method, \(h\) the constant stepsize and \(Y_k'= f(t_k,Y_k)\), \(Y_k'' =(f_t+ f_Yf) (t_k,Y_k)\). In this setting, the authors establish a theory of stability and convergence for the above methods which is similar to the well-known theory of Dahlquist for scalar linear multistep methods given by \textit{P. Henrici} [Discrete variable methods in ordinary differential equations (1962; Zbl 0112.34901)].