In this paper a high order implicit algorithm is developed for solving instationary non-linear problems. This generic numerical method combines four mathematical techniques: a time discretization, a homotopy transformation, a perturbation technique and a space discretization. The time integration is performed by classical implicit schemes (Euler implicit for problems with a first order time derivative and Newmark for second order). The timediscretization leads to non-linear equations. In this paper a new technique is proposed to solve iteratively the latter equations. The key points in this approach are, first a high order solver based on perturbation techniques, second the possibility of choosing the iteration operator, which limits the number of matrices to be triangulated. To illustrate the performance of the proposed algorithm two examples are considered: the Korteweg-de Vries equation (KdV) and the non-linear oscillations of a 2D elastic pendulum.
IntroductionDuring the last few years, a considerable research amount has been done in order to elaborate efficient algorithms for solving non-linear evolution problems. A variety of numerical methods for these time dependent problems has been proposed. The time discretization of a non-linear evolution equation leads to a non-linear equation to define the solution at the next step. The most common methods to solve the latter equation are the classical iterative algorithms for evolution problems [10, 17, 18] and predictor-corrector ones for dynamic structural problems [2,3,7,12,19]. These algorithms permit ones to get the solution of the considered non-linear time dependent problems in a step wise manner and a full point by point description is then obtained. These algorithms work very well but they may be costly in term of computation time.In this paper, we propose to use high order iteration techniques to solve the non-linear problems deduced from time discretization. These methods are well established and efficient to solve path-following problems and they can be applied to a lot of physical models, see for instance [4,6,16]. It is also possible to solve in that way non-linear problems, that do not depend on a scalar parameter. The idea is to apply a homotopy transformation and to compute series with respect to the homotopy parameter, see [1,8]. We refer to recent papers [9, 15] for a general discussion and an optimisation of these high order iterative algorithms. A key point is the possibility to choose the operator that will be triangulated and to compute many time steps with a single matrix triangulation. An algorithm of this type has been proposed recently in [5,14]. A variant of the latter method will be presented and tested here, that leads to a strong reduction of the number of matrix inversions.To illustrate the proposed algorithm, let us consider the following non-linear evolution problem:where f ð:Þ is a spatial differential operator, u is the unknown and u 0 is the initial condition. Using, for instance, an Euler implicit scheme, the time discretizatio...