[Paper Review] Splitting methods for Levitron Problems

Jürgen Geiser, Karl Lueskow|arXiv (Cornell University)|Jan 9, 2012

Numerical methods for differential equations11 references3 citations

TL;DR

This paper proposes novel splitting methods based on the Verlet integrator to solve the non-separable Hamiltonian system of the Levitron problem, enhancing numerical stability and accuracy through iterative and extrapolation techniques. The approach achieves high-precision simulations of magnetically levitated spinning tops by decoupling the system into geometric integrators, significantly reducing computational time while preserving symplectic structure.

ABSTRACT

In this paper we describe splitting methods for solving Levitron, which is motivated to simulate magnetostatic traps of neutral atoms or ion traps. The idea is to levitate a magnetic spinning top in the air repelled by a base magnet. The main problem is the stability of the reduced Hamiltonian, while it is not defined at the relative equilibrium. Here it is important to derive stable numerical schemes with high accuracy. For the numerical studies, we propose novel splitting schemes and analyze their behavior. We deal with a Verlet integrator and improve its accuracy with iterative and extrapolation ideas. Such a Hamiltonian splitting method, can be seen as geometric integrator and saves computational time while decoupling the full equation system. Experiments based on the Levitron model are discussed.

Motivation & Objective

• To develop stable, high-accuracy numerical schemes for simulating the Levitron, a nonlinear magnetostatic system where a spinning top levitates via magnetic repulsion.
• To address the challenge of instability in the reduced Hamiltonian system, particularly at relative equilibrium points where the Hamiltonian is undefined.
• To improve the standard Verlet integrator through iterative and extrapolation techniques for better accuracy in non-separable Hamiltonian systems.
• To implement geometric integrators via Hamiltonian splitting that preserve symplectic structure and reduce computational cost.

Proposed method

• The authors employ a Hamiltonian splitting method decomposing the system into two parts: A (kinetic-like) and B (potential-like), represented as Lie operators A and B.
• They apply the symmetric second-order splitting scheme T2,VV(h) = exp(h/2 B) exp(h A) exp(h/2 B), corresponding to the velocity Verlet algorithm.
• An iterative Verlet scheme is introduced, updating position and momentum iteratively using updated force and velocity estimates at each step.
• Extrapolation techniques are applied to the Verlet kernel to achieve higher-order accuracy in time integration.
• The method is tested using a reduced Hamiltonian model from Gans and an extended magnetic field model from Dullin for a disk-shaped magnet.
• Numerical experiments compare computational time and error metrics across different schemes, including 4th-order MPE with Verlet kernel.

Experimental results

Research questions

• RQ1 How can standard Verlet integrators be improved to achieve higher accuracy in non-separable Hamiltonian systems like the Levitron?
• RQ2 What is the impact of iterative and extrapolation techniques on the stability and convergence of time integrators for the Levitron problem?
• RQ3 Can splitting methods preserve the geometric structure of the Hamiltonian system while reducing computational cost?
• RQ4 How do different splitting schemes compare in terms of error and computational efficiency for long-term simulation of the Levitron?

Key findings

• The iterative Verlet scheme achieves consistent mean and maximal errors around 0.007 and 0.023, respectively, across multiple test cases.
• The 4th-order MPE scheme with Verlet kernel reduced mean error to 0.0068 and maximal error to 0.0188, demonstrating improved accuracy.
• Computational time was significantly reduced using splitting methods: 14 minutes for the fastest scheme versus 272 minutes for the slowest, indicating substantial efficiency gains.
• The proposed splitting schemes preserve the symplectic structure of the Hamiltonian system, ensuring long-term energy conservation and stability.
• The iterative and extrapolation extensions to the Verlet integrator successfully improved convergence and accuracy without increasing computational complexity unduly.
• The method effectively handles the singularity at relative equilibrium by decoupling the system and applying stable, structure-preserving integrators.