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.
Introduction
In the field of numerical simulation for physical systems, variational integrators (VIs) have emerged as a gold standard for long-term stability. Unlike general-purpose integrators, VIs are designed to be symplectic, meaning they naturally preserve the geometric structure of a physical system. This property leads to excellent energy conservation over long durations, making them indispensable for celestial mechanics and molecular dynamics.
However, constructing high-order variational integrators remains a significant challenge. A common strategy involves "shooting-based" VIs, which utilize a standard one-step method (such as Runge-Kutta) to approximate the trajectory between two boundary points. The primary bottleneck in this approach is the computational cost: achieving high-order accuracy requires evaluating the system at multiple quadrature points along the path, often necessitating numerous expensive sub-steps. This paper introduces Taylor Variational Integrators (TVIs) as a solution to this efficiency bottleneck.
Methodology
The core innovation of the Taylor Variational Integrator lies in its use of Taylor's method as the underlying engine for trajectory approximation. This choice provides several distinct advantages over traditional one-step methods:
- Polynomial Evaluation: A Taylor method produces a polynomial approximation of the trajectory. Consequently, evaluating the solution at any required quadrature point becomes a simple matter of polynomial evaluation. This is significantly more computationally efficient than taking additional integration steps or performing complex interpolations.
- Automatic Differentiation (AD): To make high-order Taylor methods viable, the authors utilize Automatic Differentiation techniques. AD allows for the efficient computation of high-order derivatives, which are necessary for the Taylor coefficients.
- Symmetry: The authors propose a symmetric version of the Lagrangian TVI. Symmetry is a highly desirable property in geometric integration, as it typically results in superior long-term error behavior and enhanced stability.
- Hamiltonian Extensions: The framework is extended to include discrete Hamiltonian formulations (Type II and III generating functions). This extension is critical for simulating systems where the Lagrangian may be degenerate or where a Hamiltonian perspective is more natural.
Theoretical Insights
One of the most compelling theoretical findings is the "Order + 1" Bonus. The authors demonstrate that a Taylor VI of order $r$ 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 gain is a direct result of how the Taylor method exploits boundary data within the variational framework.
Furthermore, the method benefits from a unique Predictor-Corrector Synergy. While TVIs are implicit and require solving a nonlinear system (typically via Newton's method), the standard Taylor method provides an exceptionally accurate initial guess. This allows the nonlinear solver to converge rapidly, often requiring only one or two iterations to reach the desired tolerance.
Numerical Results
The efficacy of TVIs was evaluated against several classic benchmarks, including the simple pendulum, Kepler’s planar 2-body problem, the Henon-Heiles model, and the Fermi-Pasta-Ulam (FPU) model. Key findings include:
- Efficiency: The 4th-order symmetric Taylor variational integrator (SV4) consistently outperformed other methods in terms of the trade-off between computational time and global energy error.
- Robustness: In the Kepler problem, traditional Taylor methods failed to capture correct orbital behavior at larger step sizes (e.g., $h=0.25$), whereas TVIs maintained the correct qualitative behavior.
- Energy Conservation: The research highlighted that symplecticity is closely tied to the accuracy of the nonlinear solver. By setting solver tolerances slightly tighter than the integrator's order, the authors achieved near-perfect energy conservation without secular drift.
- Large-Scale Simulation: In a simulation of the outer solar system over 200,000 days, 6th-order TVIs provided a qualitatively accurate representation where lower-order methods failed to maintain stability.
Limitations and Future Work
Despite their advantages, TVIs face certain limitations. The computational cost of evaluating higher-order derivatives grows at a rate of $2^n$, which may limit the practical order of the integrator for extremely complex systems. Additionally, like most VIs, these methods are currently optimized for fixed stepsizes; extending them to variable stepsizes while maintaining symplecticity is a subject for future research. The authors also suggest that combining these integrators with Poincaré transformations could further improve performance for degenerate Hamiltonian systems.
Summary updated