Lagrangian and Hamiltonian Taylor variational integrators

Jérémy Schmitt, Tatiana Shingel, Melvin Leok

TL;DR

This paper introduces a new class of high-order variational integrators that leverage Taylor methods to solve the Euler–Lagrange boundary-value problem. By exploiting the polynomial nature of Taylor approximations, these integrators achieve higher accuracy and better computational efficiency than traditional shooting-based methods while strictly preserving the symplectic structure of physical systems.

What Problem Does This Paper Address?

In the world of numerical simulation, variational integrators (VIs) are the gold standard for long-term stability because they are "symplectic"—they naturally preserve the geometric structure of a physical system, leading to excellent energy conservation.

However, building high-order VIs is traditionally difficult. One common approach is "shooting-based" VIs, which use a standard one-step method (like Runge-Kutta) to approximate the path between two points. The bottleneck here is that to get high-order accuracy, you need to evaluate the system at multiple "quadrature points" along that path. For most methods, this requires taking many tiny sub-steps, which is computationally expensive. The authors aim to break this bottleneck by using Taylor methods as the underlying engine.

Methods

The authors propose Taylor Variational Integrators (TVIs). The core innovation lies in how they handle the trajectory approximation:

  1. Taylor Engine: Instead of a generic one-step method, they use Taylor's method to approximate the flow of the system.
  2. Polynomial Evaluation: Because a Taylor method produces a polynomial approximation of the trajectory, evaluating the solution at any quadrature point becomes a simple matter of polynomial evaluation. This is significantly "cheaper" than taking additional integration steps.
  3. Automatic Differentiation (AD): To make high-order Taylor methods practical, the authors leverage AD techniques to compute high-order derivatives efficiently.
  4. Symmetry: They introduce a symmetric version of the Lagrangian Taylor VI. Symmetry is a prized property in geometric integration as it often leads to even better long-term error behavior.
  5. Hamiltonian Extensions: The paper extends the framework beyond the Lagrangian side to include discrete Hamiltonian formulations (Type II and III generating functions), which are essential for dealing with systems where the Lagrangian might be degenerate.

Datasets

The paper evaluates the proposed methods on several classic "benchmark" problems in dynamical systems and celestial mechanics:

  • Simple Pendulum: A standard test for energy preservation.
  • Kepler’s Planar 2-Body Problem: Testing orbits and long-term stability.
  • Henon-Heiles Model: A system that exhibits chaotic motion at high energy levels.
  • Fermi-Pasta-Ulam (FPU) Model: A highly oscillatory chain of mass points.
  • Outer Solar System: A simulation of the five outer planets (including Pluto) relative to the sun over 200,000 days.

Key Insights and Ideas

  • The "Order + 1" Bonus: A Taylor VI of order $r$ actually results in a shooting method that is one order higher ($r+1$) than other shooting methods based on a one-step method of the same order. This is due to how the Taylor method exploits the boundary data.
  • Efficiency through Polynomials: The "aha moment" is realizing that once you have the Taylor coefficients at the start of a step, you have a continuous approximation of the path. You don't need to "re-integrate" to find values at quadrature points; you just plug numbers into a polynomial.
  • Predictor-Corrector Synergy: While TVIs require solving a nonlinear system (usually via Newton's method), a standard Taylor method provides an excellent "initial guess," allowing the solver to converge extremely quickly—often in just one or two iterations.

Key Results

  • Efficiency: The 4th-order symmetric Taylor variational integrator (SV4) emerged as the "clear winner" in terms of efficiency (computational time vs. global energy error) across multiple tests.
  • Robustness at Large Steps: In the Kepler problem, Taylor methods failed to capture the correct behavior at a step size of $h=0.25$, while the Taylor Variational Integrators maintained close to the correct orbital behavior (see Figure 6).
  • Energy Conservation: The authors demonstrated that the accuracy of the nonlinear solver directly impacts symplecticity. By setting the solver tolerance slightly tighter than the integrator's order, they achieved near-perfect energy conservation without drift (see Figure 1).
  • Celestial Mechanics: In the Outer Solar System test, 6th-order TVIs provided a qualitatively accurate representation of the system over massive timescales where lower-order methods failed.

Limitations and Future Work

  • Derivative Cost: The computational cost of evaluating higher-order derivatives grows at a rate of $2^n$ (where $n$ is the order). While efficient for orders under 10, it may become a bottleneck for extremely high-order requirements.
  • Fixed Stepsizes: Like most variational integrators, these methods are primarily designed for fixed stepsizes. Extending them to variable stepsizes while maintaining symplecticity remains a challenging area for future research.
  • Non-Separable Hamiltonians: While the symmetric version (SVHd) performs better than the standard Störmer–Verlet for non-separable problems, it is implicit and thus more expensive.
  • Future Direction: The authors suggest combining these integrators with the Poincaré transformation to better handle degenerate Hamiltonians.

Summary updated

Abstract

In this paper, we present a variational integrator that is based on an approximation of the Euler--Lagrange boundary-value problem via Taylor's method. This can viewed as a special case of the shooting-based variational integrator. The Taylor variational integrator exploits the structure of the Taylor method, which results in a shooting method that is one order higher compared to other shooting methods based on a one-step method of the same order. In addition, this method can generate quadrature nodal evaluations at the cost of a polynomial evaluation, which may increase its efficiency relative to other shooting-based variational integrators. A symmetric version of the method is proposed, and numerical experiments are conducted to exhibit the efficacy and efficiency of the method.
BibTeX
@inproceedings{schmitt2017lagrangian,
  author = {Jérémy Schmitt and Tatiana Shingel and Melvin Leok},
  title = {{Lagrangian and Hamiltonian Taylor variational integrators}},
  booktitle = {BIT},
  year = {2017},
  doi = {10.1007/s10543-017-0690-9},
}

Loading related content...

Figure