On convergence conditions of waveform relaxation methods for linear differential-algebraic equations (Q629556): Difference between revisions

This paper is concerned with the numerical solution of initial value problems for linear constant-coefficient differential-algebraic equations \[ B\dot x(t) + A x(t) = f(t),\quad x(0)=x_0, \tag{1} \] where \(B,A\) are given square matrices, \(B\) is singular, and the pencil \(\lambda B + A\) is supposed to be regular. Using the Laplace transformation and the idea of dynamic iteration method by \textit{U. Miekkala} [J. Comput. Appl. Math. 25, No.~2, 133--151 (1989; Zbl 0669.65056)], the authors formulate a new waveform relaxation method for (1) and investigate its convergence. By suitable splittings \(B=M_B-N_B\), \(A=M_A-N_A\), the waveform relaxation method is given in the iteration form \[ x^{(k)}(t)={\mathcal K}(x^{(k-1)}(t))+\Phi(f(t)), \tag{2} \] where the operators \({\mathcal K}\) and \(\Phi\) are given in terms of functions of auxiliary matrices, the Laplace transform, and the inverse Laplace transform. By calculating the spectral radius of the iteration operator \({\mathcal K}\), new sufficient conditions are obtained for the convergence of the iteration (2). Here, the authors do not require the uniform boundedness of solutions on infinite interval. The latter fact improves the convergence result given by Miekkala [loc. cit.]. Finally, a numerical example with unbounded solution is given for illustrating the convergence of the numerical method.