A fast and memoryless numerical method for solving fractional differential equations

TL;DR

This paper introduces a fast, memoryless numerical method for solving fractional differential equations by approximating the fractional kernel with exponential sums, transforming the problem into a system of ODEs, and efficiently solving the resulting stiff systems.

Contribution

The authors develop a novel approximation of the fractional kernel using exponential sums and demonstrate how to efficiently solve the resulting large, stiff linear systems with existing implicit solvers like Radau5.

Findings

The method achieves high accuracy in numerical experiments.

It significantly reduces memory usage compared to traditional methods.

The approach is computationally efficient for large-scale problems.

Abstract

The numerical solution of implicit and stiff differential equations by implicit numerical integrators has been largely investigated and there exist many excellent efficient codes available in the scientific community, as Radau5 (based on a Runge-Kutta collocation method at Radau points) and Dassl, based on backward differentiation formulas, among the others. When solving fractional ordinary differential equations (ODEs), the derivative operator is replaced by a non-local one and the fractional ODE is reformulated as a Volterra integral equation, to which these codes cannot be directly applied. This article is a follow-up of the article by the authors (Guglielmi and Hairer, SISC, 2025) for differential equations with distributed delays. The main idea is to approximate the fractional kernel $t^{\alpha -1}/ \Gamma (\alpha )$ ($\alpha >0$) by a sum of exponential functions or by a sum of exponential functions multiplied by a monomial, and then to transform the fractional integral (of convolution type) into a set of ordinary differential equations. The augmented system is typically stiff and thus requires the use of an implicit method. It can have a very large dimension and requires a special treatment of the arising linear systems. The present work presents an algorithm for the construction of an approximation of the fractional kernel by a sum of exponential functions, and it shows how the arising linear systems in a stiff time integrator can be solved efficiently. It is explained how the code Radau5 can be used for solving fractional differential equations. Numerical experiments illustrate the accuracy and the efficiency of the proposed method. Driver examples are publicly available from the homepages of the authors.