General Techniques for Constructing Variational Integrators

Melvin Leok, Tatiana Shingel

Abstract

The numerical analysis of variational integrators relies on variational error analysis, which relates the order of accuracy of a variational integrator with the order of approximation of the exact discrete Lagrangian by a computable discrete Lagrangian. The exact discrete Lagrangian can either be characterized variationally, or in terms of Jacobi's solution of the Hamilton-Jacobi equation. These two characterizations lead to the Galerkin and shooting-based constructions for discrete Lagrangians, which depend on a choice of a numerical quadrature formula, together with either a finite-dimensional function space or a one-step method. We prove that the properties of the quadrature formula, finite-dimensional function space, and underlying one-step method determine the order of accuracy and momentum-conservation properties of the associated variational integrators. We also illustrate these systematic methods for constructing variational integrators with numerical examples.

Introduction

In the simulation of physical systems, preserving the underlying geometric structure is often more critical than minimizing local truncation error. Classical mechanical systems are governed by the Principle of Least Action, where the path taken by a system is a stationary point of the action functional. While standard numerical methods, such as the ubiquitous Runge-Kutta schemes, are highly accurate over short durations, they often fail to preserve physical invariants like energy and momentum over long-term simulations. This failure manifests as "numerical drift," where the system artificially gains or loses energy.

Variational integrators offer a robust alternative by discretizing the action principle itself. By ensuring the discrete motion satisfies a discrete version of Hamilton’s principle, these integrators naturally inherit symplectic properties and preserve momentum maps. This paper provides a systematic framework for constructing high-order variational integrators using two primary techniques: Galerkin and shooting-based constructions.

Systematic Construction of Discrete Lagrangians

The core of a variational integrator is the discrete Lagrangian, $L_d(q_0, q_1, h)$, which approximates the action integral along a path segment of duration $h$. The authors identify two systematic ways to build these approximations.

1. Galerkin-Based Construction

The Galerkin approach replaces the infinite-dimensional space of possible trajectories with a finite-dimensional subspace, such as the space of polynomials of degree $k$. The action is then approximated using a numerical quadrature rule.

  • Function Spaces: By choosing specific basis functions, one can control the accuracy of the resulting integrator.
  • Lie Group Extensions: A significant contribution of this work is the extension of Galerkin methods to Lie groups (e.g., $SO(3)$ for rotations). By using the exponential map and "natural charts," the authors ensure that the discrete trajectory remains on the manifold, preserving properties like the orthogonality of rotation matrices to machine precision.

2. Shooting-Based Construction

The shooting-based approach is a "plug-and-play" framework that leverages existing numerical expertise. It uses a standard, non-variational one-step method (like a 4th-order Runge-Kutta) to approximate the trajectory between two points. This trajectory is then used within a quadrature rule to compute the discrete Lagrangian.

  • Inherited Accuracy: The paper proves that the accuracy of the resulting variational integrator is determined by the minimum of the order of the underlying solver ($p$) and the quadrature rule ($q$).
  • Symmetry: If both the underlying solver and the quadrature rule are symmetric (self-adjoint), the resulting variational integrator is also symmetric, which typically leads to an even order of accuracy and enhanced long-term stability.

Numerical Analysis and Performance

The authors evaluate these techniques using classic benchmarks: the free rigid body, the planar pendulum, and the simple harmonic oscillator.

Geometric and Energy Stability

In tests of the free rigid body, the Lie Group Variational Integrator (LGVI) demonstrated exceptional geometric fidelity. While standard methods exhibited drift in the orthogonality of rotation matrices, the LGVI maintained these constraints to the level of machine precision ($10^{-15}$).

Furthermore, while variational integrators do not preserve energy exactly, they are known to preserve a "modified Hamiltonian." The numerical experiments confirm that for the pendulum and oscillator, energy error remains bounded and oscillates near zero over long periods. In contrast, non-variational methods like the explicit midpoint rule show a clear trend of energy error growth over time.

Computational Efficiency

A key insight is the efficiency of the LGVI. By carefully constructing the discrete Lagrangian, the authors developed a method that requires only one force evaluation per time-step. This allows it to outperform more complex methods like Crouch-Grossman or Implicit Midpoint schemes in terms of CPU time versus error benchmarks.

Discussion and Limitations

Despite their advantages, these systematic constructions introduce certain complexities. The shooting-based approach requires solving a small boundary value problem at each step to determine the initial velocity that connects two points. While this can be handled efficiently using iterative methods like Newton’s method with Cayley transforms, it remains more computationally intensive than a simple explicit step. Additionally, implementing higher-order derivatives for specific quadrature rules (such as Euler-Maclaurin) can be mathematically demanding.

Conclusion

This work bridges the gap between abstract variational mechanics and practical numerical analysis. By providing "recipes" for constructing high-order variational integrators from familiar components—polynomials, quadrature rules, and standard one-step solvers—the authors make these powerful tools more accessible for complex physical simulations. Future work may extend these techniques to stochastic systems, multiphysics environments, and systems involving non-smooth collisions.

Summary updated

Abstract

The numerical analysis of variational integrators relies on variational error analysis, which relates the order of accuracy of a variational integrator with the order of approximation of the exact discrete Lagrangian by a computable discrete Lagrangian. The exact discrete Lagrangian can either be characterized variationally, or in terms of Jacobi's solution of the Hamilton-Jacobi equation. These two characterizations lead to the Galerkin and shooting-based constructions for discrete Lagrangians, which depend on a choice of a numerical quadrature formula, together with either a finite-dimensional function space or a one-step method. We prove that the properties of the quadrature formula, finite-dimensional function space, and underlying one-step method determine the order of accuracy and momentum-conservation properties of the associated variational integrators. We also illustrate these systematic methods for constructing variational integrators with numerical examples.
BibTeX
@inproceedings{leok2012general,
  author = {Melvin Leok and Tatiana Shingel},
  title = {{General Techniques for Constructing Variational Integrators}},
  booktitle = {Frontiers of Mathematics in China},
  year = {2012},
  doi = {10.1007/s11464-012-0190-9},
}
Figure