Variational principles for Hamiltonian systems

Brian K. Tran, Melvin Leok

TL;DR

This paper introduces a new "Type II" variational framework for Hamiltonian systems that works intrinsically on curved manifolds and handles "degenerate" systems where traditional methods fail. By using a virtual work principle to enforce boundary conditions, the authors provide a robust mathematical foundation for building next-generation numerical integrators, optimization algorithms, and control systems.

What Problem Does This Paper Address?

In classical mechanics, we usually derive equations of motion using Hamilton’s Principle, which involves finding a path that minimizes "action" given fixed starting and ending positions (Type I boundary conditions). While this works well for standard systems, it hits a wall in two specific scenarios:

  1. Degenerate Systems: Many important systems (like those in adjoint sensitivity analysis or certain constrained dynamics) have "degenerate" Hamiltonians. In these cases, fixing both the start and end positions is mathematically inconsistent—the system is simultaneously overdetermined and underdetermined.
  2. Manifolds: Defining variational principles on curved spaces (manifolds) is tricky. In a "Type II" approach, you fix the initial position $q(0)$ and the final momentum $p(T)$. However, on a manifold, a momentum vector is tied to a specific base point. If the final position $q(T)$ is unknown/variable, it is difficult to define the final momentum $p(T)$ in a way that doesn't depend on the specific coordinate chart you are using.

The authors aim to bridge this gap by creating a coordinate-independent (intrinsic) variational principle that handles these boundary conditions naturally.

Methods

The authors move away from the standard "stationary action" approach and instead adopt a Type II d’Alembert Variational Principle based on virtual work.

  • Virtual Work Principle: Instead of requiring the action functional itself to be stationary, they require the variation of the action to equal the "virtual work" done by the boundary momentum. This allows the boundary conditions to emerge as "natural" consequences of the variation rather than "essential" constraints imposed at the start.
  • Parallelizable Manifolds: They first show how to define this on manifolds that can be "flattened" globally (like Lie groups). They derive the Hamel equations, which are the non-canonical equivalents of Hamilton's equations.
  • Free Boundary Problems: For general manifolds that aren't parallelizable, they treat the system as a "free boundary problem." Here, the final momentum is defined as a section of the cotangent bundle (a field of possible momenta), allowing the final position to remain variable while still maintaining a well-defined variational principle.
  • Infinite-Dimensional Extension: The framework is extended to reflexive Banach spaces, providing a path to apply these ideas to Hamiltonian Partial Differential Equations (PDEs).

Datasets

Not discussed in the paper. This is a theoretical mathematical physics paper focused on deriving new formalisms and proofs rather than empirical benchmarking.

Key Insights and Ideas

  • Degeneracy Demands Type II: The authors prove that for a model degenerate Hamiltonian, Type I and Type IV boundary conditions are fundamentally incompatible with the physics, whereas Type II and III are complete. This justifies why we need to move beyond standard Lagrangian-based variational integrators.
  • Natural vs. Essential Boundaries: By treating the final momentum condition as a "natural" boundary condition (emerging from the virtual work), the authors bypass the need to fix the final position. This is the "aha moment" that allows the principle to work on curved manifolds.
  • The Adjoint Connection: The paper highlights that "maximally degenerate" Hamiltonians are exactly the forms used in adjoint sensitivity analysis. This means their new variational principle provides the "natural" geometric setting for understanding how sensitive a system's output is to its initial conditions.

Key Results

  • Unified Framework: The authors successfully derived a Type II variational principle that transitions seamlessly from simple vector spaces to parallelizable manifolds, general manifolds, and finally to infinite-dimensional systems.
  • Hamel Equations Derivation: They showed that their intrinsic variational principle correctly recovers the Hamel equations on parallelizable manifolds, which are crucial for simulating systems like rigid bodies or fluids.
  • Stationarity for Cost Functions: They demonstrated that if the boundary momentum is chosen as the derivative of a cost function ($p_1 = dC$), the free boundary variational principle becomes a standard stationarity condition ($\delta \mathcal{J} = 0$), linking classical mechanics directly to optimal control theory.

Limitations and Future Work

  • Existence and Uniqueness: The authors acknowledge that the analysis of "free boundary problems" is generally complicated and problem-dependent. They do not provide a universal proof of existence/uniqueness for all possible Hamiltonians on general manifolds.
  • Numerical Implementation: While the paper provides the theoretical blueprint for "Type II Variational Integrators" on manifolds, it does not implement them. The authors suggest that the next logical step is to use this framework to build discrete-time algorithms (integrators) that preserve the symplectic structure of Hamiltonian systems on curved spaces.
  • Stochastic Systems: While touched upon, the extension of these intrinsic manifold-based principles to stochastic Hamiltonian systems remains a fertile ground for future research.

Summary updated

Abstract

Motivated by recent developments in Hamiltonian variational principles, Hamiltonian variational integrators, and their applications such as to optimization and control, we present a new Type II variational approach for Hamiltonian systems, based on a virtual work principle that enforces the Type II boundary conditions through a combination of essential and natural boundary conditions; particularly, this approach allows us to define this variational principle intrinsically on manifolds. We first develop this variational principle on vector spaces and subsequently extend it to parallelizable manifolds, general manifolds, as well as to the infinite-dimensional setting. Furthermore, we provide a review of variational principles for Hamiltonian systems in various settings as well as their applications.
BibTeX
@inproceedings{tran2025variational,
  author = {Brian K. Tran and Melvin Leok},
  title = {{Variational principles for Hamiltonian systems}},
  booktitle = {Geometric},
  year = {2025},
  doi = {10.1142/s2972458925500042},
}

Loading related content...

Figure