A single-step third-order temporal discretization with Jacobian-free and Hessian-free formulations for finite difference methods

TL;DR

This paper introduces a novel third-order temporal discretization method for finite difference PDE solvers that simplifies the computation of Jacobian and Hessian terms, enhancing efficiency in high-order time integration.

Contribution

The paper presents an extension of the Picard integration formulation enabling single-step, third-order accurate temporal discretization with Jacobian-free and Hessian-free formulations.

Findings

Achieves third-order temporal accuracy.

Simplifies Jacobian and Hessian computations.

Enhances efficiency of high-order PDE solvers.

Abstract

Discrete updates of numerical partial differential equations (PDEs) rely on two branches of temporal integration. The first branch is the widely-adopted, traditionally popular approach of the method-of-lines (MOL) formulation, in which multi-stage Runge-Kutta (RK) methods have shown great success in solving ordinary differential equations (ODEs) at high-order accuracy. The clear separation between the temporal and the spatial discretizations of the governing PDEs makes the RK methods highly adaptable. In contrast, the second branch of formulation using the so-called Lax-Wendroff procedure escalates the use of tight couplings between the spatial and temporal derivatives to construct high-order approximations of temporal advancements in the Taylor series expansions. In the last two decades, modern numerical methods have explored the second route extensively and have proposed a set of computationally efficient single-stage, single-step high-order accurate algorithms. In this paper, we present an algorithmic extension of the method called the Picard integration formulation (PIF) that belongs to the second branch of the temporal updates. The extension presented in this paper furnishes ease of calculating the Jacobian and Hessian terms necessary for third-order accuracy in time.