High-order implicit time integration scheme based on Pad\'e expansions

TL;DR

This paper introduces a high-order implicit time integration scheme based on Padé expansions that achieves high accuracy and efficiency for large-scale transient and wave propagation problems without requiring mass matrix inversion.

Contribution

The paper develops a novel high-order implicit time integrator using Padé expansions that is computationally efficient and avoids mass matrix inversion, extending the accuracy of existing methods.

Findings

Achieves high accuracy with large degrees of freedom

Comparable complexity to standard methods like Newmark

Demonstrates superior efficiency in numerical examples

Abstract

A single-step high-order implicit time integration scheme for the solution of transient and wave propagation problems is presented. It is constructed from the Pad\'e expansions of the matrix exponential solution of a system of first-order ordinary differential equations formulated in the state-space. A computationally efficient scheme is developed exploiting the techniques of polynomial factorization and partial fractions of rational functions, and by decoupling the solution for the displacement and velocity vectors. An important feature of the novel algorithm is that no direct inversion of the mass matrix is required. From the diagonal Pad\'e expansion of order $M$ a time-stepping scheme of order $2M$ is developed. Here, each elevation of the accuracy by two orders results in an additional system of real or complex sparse equations to be solved. These systems are comparable in complexity to the standard Newmark method, i.e., the effective system matrix is a linear combination of the static stiffness, damping, and mass matrices. It is shown that the second-order scheme is equivalent to Newmark's constant average acceleration method, often also referred to as trapezoidal rule. The proposed time integrator has been implemented in MATLAB using the built-in direct linear equation solvers. In this article, numerical examples featuring nearly one million degrees of freedom are presented. High-accuracy and efficiency in comparison with common second-order time integration schemes are observed. The MATLAB-implementation is available from the authors upon request or from the GitHub repository (to be added).