Iterative operator splitting methods for differential equations: proof techniques and applications (Q2905993)

The paper focuses on a novel iterative operator-splitting method for bounded operators that can be applied as a splitting method to ordinary differential equations (ODEs) and partial differential equations (PDEs). The paper concentrates on an approximate solution of the linear evolution equation given by \(\partial_tu \equiv Lu=(A+B)u\), \(u(0)=u_0\), where \(L\), \(A\) and \(B\) are bounded operators, using a two-stage iterative splitting scheme.NEWLINENEWLINEFeatures of iterative splitting methods are identified in Section 1 and an introduction to operator-splitting methods is presented in Section 2, where consistency of the iterative splitting methods is proved for unbounded generators of a strongly continuous semigroup. In Section 3, following relevant definitions relating to consistency, the consistency of the novel iterative operator splitting method are proved via a sequence of propositions.NEWLINENEWLINESection 4 relates to the compuation of the iterative splitting method using \(\phi\)-functions. Section 4.1 discusses the exact computation of the integrals and includes a comment by the author that ``the main disadvantage of computing the iterative scheme exactly are the time-consuming inverse matrices''. The new method involves only matrix multiplications of the given exponential functions and avoids the integral formulation of the exponential functions. Computation of the operators using the trapezium rule, Simpson's rule and the Bode rule are included.NEWLINENEWLINENumerical examples are presented in Section 5 to enable a comparison of the iterative splitting method with standard splitting methods. The examples include the application of the method to a large system of ODEs to demonstrate the benefit to ODEs, to PDEs using the diffusion equation and a heat conduction equation with initial and boundary conditions and to the 2-dimensional advection-diffusion equation with periodic boundary conditions. CPU times, and the number of iterations involved, are compared to demonstrate the benefit of the iterative schemes. A comparison of errors for the heat conduction equation illustrates that more accurate results, in shorter CPU time, are achieved using the iterative splitting methods.NEWLINENEWLINESome problems with AB splitting are identified in Section 5.4.1, and in Section 5.4.2 a table is presented which shows which methods are practicable for each kind of splitting scheme.