Table of Contents
Fetching ...

Rodas6P and Tsit5DA - two new Rosenbrock-type methods for DAEs

Gerd Steinebach

TL;DR

The paper addresses efficient numerical solution of index-1 DAEs by expanding the Rosenbrock family with two new methods: Rodas6P, a sixth-order stiffly accurate Rosenbrock-type method with embedded fifth order, and Tsit5DA, a fifth-order method that blends an explicit differential solver with a linear-implicit algebraic component based on Tsit5. Rodas6P achieves high accuracy at the cost of many stages (s=19) but offers potential time savings for large, semi-discretized PDE-DAEs and remains L-stable; Tsit5DADA provides a significant improvement over RKF4DA for non-stiff problems by leveraging an explicit-implicit split and optimized coefficients. The authors validate theoretical orders via order tests and benchmarks, including Prothero-Robinson-type problems, semi-discretized parabolic and hyperbolic PDEs, and a multi-pendulum example, and supply extensive coefficient data and implementation details. Together, Rodas6P and Tsit5DA extend the Rosenbrock family with a high-order stiff method and a fifth-order explicit-implicit DAE solver, offering practitioners options for balancing accuracy, stiffness handling, and computational efficiency in large-scale DAEs.

Abstract

Two new Rosenbrock methods for solving index-1 differential algebraic equations are presented. Rodas6P is a sixth-order method based on the same design principles as the Rodas3P, Rodas4P, and Rodas5P methods. Tsit5DA is based on an explicit solution approach for the differential equations and a linear-implicit approach for the algebraic equations. Such a fourth-order method has already been presented in Rentrop, Roche & Steinebach, 1989. Tsit5DA now provides a significantly improved fifth-order method which is based on the well known Tsit5 method. The theoretical properties of the new methods are verified by some order tests and benchmarks.

Rodas6P and Tsit5DA - two new Rosenbrock-type methods for DAEs

TL;DR

The paper addresses efficient numerical solution of index-1 DAEs by expanding the Rosenbrock family with two new methods: Rodas6P, a sixth-order stiffly accurate Rosenbrock-type method with embedded fifth order, and Tsit5DA, a fifth-order method that blends an explicit differential solver with a linear-implicit algebraic component based on Tsit5. Rodas6P achieves high accuracy at the cost of many stages (s=19) but offers potential time savings for large, semi-discretized PDE-DAEs and remains L-stable; Tsit5DADA provides a significant improvement over RKF4DA for non-stiff problems by leveraging an explicit-implicit split and optimized coefficients. The authors validate theoretical orders via order tests and benchmarks, including Prothero-Robinson-type problems, semi-discretized parabolic and hyperbolic PDEs, and a multi-pendulum example, and supply extensive coefficient data and implementation details. Together, Rodas6P and Tsit5DA extend the Rosenbrock family with a high-order stiff method and a fifth-order explicit-implicit DAE solver, offering practitioners options for balancing accuracy, stiffness handling, and computational efficiency in large-scale DAEs.

Abstract

Two new Rosenbrock methods for solving index-1 differential algebraic equations are presented. Rodas6P is a sixth-order method based on the same design principles as the Rodas3P, Rodas4P, and Rodas5P methods. Tsit5DA is based on an explicit solution approach for the differential equations and a linear-implicit approach for the algebraic equations. Such a fourth-order method has already been presented in Rentrop, Roche & Steinebach, 1989. Tsit5DA now provides a significantly improved fifth-order method which is based on the well known Tsit5 method. The theoretical properties of the new methods are verified by some order tests and benchmarks.

Paper Structure

This paper contains 5 sections, 13 equations, 2 figures, 6 tables.

Figures (2)

  • Figure 1: Work-precision diagrams for parabolic and hyperbolic problems.
  • Figure 2: Trajectories of a pendulum with 5 masses.