A decoupled direct method for the solution of ordinary boundary value problems (Q1312684)

The problem of numerical resolution of a general nonlinear two point boundary value problem for ordinary differential equations is considered. The author proposes to apply the so-called shooting method, coupled with Newton's algorithm for resolution of the resulting nonlinear, nondifferential system of equations. The shooting method when applied the Newton's process consists in an iterative resolution of a sequence of initial value problems for an extended system composed by the original equation and an auxiliary system -- its derivative. This auxiliary system is needed in order to compute the Jacobi matrix of the resulting nonlinear, nondifferential system of equations (to be solved by Newton's process). The number of equations in the auxiliary system is in general much greater than this of the original system. The author observes that a great precision while computing the Jacobi matrix for the Newton process is not so important as in computing the solution of the original system: it may influence the order of the iterative process and not the precision of the final result. This observation leads to the following proposition concerning the implementation of the algorithm: -- apply the semiimplicit three-level Runge-Kutta method (strongly \(A\)- stable!) for integration of the original equation; -- apply the two level implicit Runge-Kutta method (less expensive!) for integration of the auxiliary system. The paper contains numerical examples and a discussion comparing the proposed algorithm with the finite difference method.