Energy conserving nonholonomic integrators

Jorge Cortés

This paper, presented at the 2002 Conference on Dynamical Systems and Differential Equations, tackles a classic "holy grail" problem in computational physics: how to build a numerical simulation that perfectly respects the laws of physics—specifically energy and momentum—when dealing with complex, constrained systems.

TL;DR

The paper introduces a new class of numerical integrators for nonholonomic Lagrangian systems (like a wheel rolling without slipping) that preserve energy, momentum, and symplectic structure. By treating time as a dynamic variable rather than a fixed step, the author bypasses theoretical "no-go" theorems that previously prevented simultaneous conservation of these properties.

What Problem Does This Paper Address?

In the field of Geometric Integration, researchers aim to create simulation algorithms that preserve the underlying geometric structures of physical systems. For a standard "holonomic" system (where constraints only depend on position), we have great tools like variational integrators. However, nonholonomic systems—where constraints depend on velocity in a way that can't be simplified to position (e.g., a knife edge sliding on ice or a rolling ball)—are much harder.

The core challenge is a theoretical obstruction known as the Ge-Marsden theorem. It states that for a fixed time step, a numerical integrator cannot simultaneously preserve energy, momentum, and the symplectic form (the "area" in phase space) unless it is the exact solution. Most existing integrators for nonholonomic systems fail to preserve energy, leading to "numerical drift" where the system artificially gains or loses energy over long simulations.

Methods

The author, Jorge Cortés, proposes a solution based on Extended Discrete Mechanics.

  1. Time as a Coordinate: Instead of using a fixed time step ($h$), the author treats time ($t$) as a coordinate in the configuration space, just like position ($q$). This creates an "extended" state space.
  2. Discrete Lagrange-d’Alembert (DLA) Principle: Standard variational integrators work by finding the "path of least action." For nonholonomic systems, you can't just minimize action; you must use the DLA principle, which is essentially a discrete version of the principle of virtual work.
  3. The EDLA Algorithm: By applying the DLA principle to the extended state space, the author derives the Extended Discrete Lagrange-d’Alembert (EDLA) equations. Because time is now a variable, the algorithm calculates the necessary time step at each iteration to ensure that the discrete energy remains constant.

Datasets

Not discussed in the paper. This is a theoretical and mathematical derivation paper. It focuses on proving the existence and properties of the integrator rather than providing numerical benchmarks or experimental data.

Key Insights and Ideas

  • Bypassing the "No-Go": The most significant insight is that by allowing the time step to vary dynamically as part of the system's evolution, the theoretical barrier to energy conservation is removed.
  • Energy as Momentum: In this extended framework, energy conservation is mathematically treated as the "momentum" associated with the symmetry of time translation. If the system is autonomous (doesn't change its rules over time), the algorithm naturally preserves this "momentum" (energy).
  • Nonholonomic Momentum: The paper proves that these integrators also satisfy a discrete version of the nonholonomic momentum equation. If the system has "horizontal symmetries" (like a rolling sphere being symmetric regardless of its orientation), the corresponding momentum is preserved exactly.
  • Chaplygin Systems: The author shows how these complex integrators can be "reduced" for a specific class of systems called Chaplygin systems, leading to a simplified set of equations (REDLA) that are easier to implement.

Key Results

  • Proposition 4 (Energy Conservation): The primary result is a formal proof that if a discrete nonholonomic system is autonomous, the EDLA algorithm preserves its associated discrete energy exactly.
  • Symplecticity: The author demonstrates that the integrator is symplectic in an extended sense, meaning it preserves the geometric "volume" of the state space over time, which is crucial for the long-term stability of simulations.
  • Momentum Preservation: The paper provides a rigorous derivation showing that the discrete nonholonomic momentum map evolves correctly, matching the behavior of the continuous physical system.

Limitations and Future Work

  • Well-posedness: The author notes that the EDLA equations might not always have a solution. Specifically, if the discrete energy is near zero or if the chosen Lagrangian is "pathological," the algorithm might fail to find a valid next time step.
  • Numerical Implementation: While the math is derived, the paper does not provide a specific implementation or a study of the computational cost compared to standard fixed-step methods.
  • Error Analysis: The author explicitly identifies the need for a formal numerical error analysis (using "backward error" techniques) as a vital next step to understand the accuracy of these integrators over long durations.

Summary updated

Abstract

We address the problem of constructing numerical integrators for nonholonomic Lagrangian systems that enjoy appropriate discrete versions of the geometric properties of the continuous flow, including the preservation of energy. Building on previous work on time-dependent discrete mechanics, our approach is based on a discrete version of the Lagrange-d’Alembert principle for nonautonomous systems.
BibTeX
@inproceedings{cortes2003energy,
  author = {Jorge Cortés},
  title = {{Energy conserving nonholonomic integrators}},
  booktitle = {Conference},
  year = {2003},
  doi = {10.3934/proc.2003.2003.189},
}

Loading related content...

Figure