Predictor-corrector strategies for single- and multi-term fractional differential equations


In E. A. Lipitakis (ed.): Proceedings of the 5th Hellenic-European Conference on Computer Mathematics and its Applications. LEA Press, Athens (2002), 117-122. [Zbl. Math. 1028.65081]


2000 Mathematics Subject Classification: Primary 65L05, 65L06; Secondary 26A33, 65R20.

Key words: Fractional differential equation, multi-term equation, predictor-corrector method


Abstract

We investigate strategies for the numerical solution of the initial value problem y(an)(x) = f(x, y(x), y(a1)(x), ..., y(an-1)(x)), y(k)(0) = y0(k) (k=0,1,..., b -1), where b is the smallest integer greater than or equal to an and 0 < a1 < a2 < ... < an. Here y(aj) denotes the derivative of order aj > 0 (aj not necessarily an integer) in the sense of Caputo. We begin our investigations with the simplest case, n=1. Here we propose a predictor-corrector (more precisely, PECE) method that can be interpreted in the spirit of the classical Adams-Bashforth-Moulton schemes for first-order equations. Specifically we analyse the discretization error of this approach under various assumptions on the given data. On the basis of these results we then look at the general problem for arbitrary n. Here we first generalize the classical approach for higher order differential equations and investigate the properties of the resulting scheme.

To overcome the disadvantages of this first approach, we alternatively suggest to construct an approximate solution in two steps. First we replace the given equation by a different n-term fractional differential equation without changing the initial conditions. This second differential equation is constructed in such a way that its solution z does not differ significantly from the solution y of the original problem, and that its structure allows a conversion into an equivalent system of one-term fractional differential equations of order a whose dimension is comparatively small. Then we solve the resulting system numerically with our predictor-corrector algorithm. In view of the properties of the system that are guaranteed by our construction, we find that the computational effort remains reasonably small without sacrificing too much precision.



List of Publications
Kai Diethelm's Homepage