🔢 mathematics

Discrete-time generalized canonical transformations for non-autonomous systems

This paper proposes a geometric discretization method for non-autonomous Hamiltonian systems using generalized canonical transformations to construct a structure-preserving discrete flow that maintains key invariants such as the volume form, Poisson bracket, and symplectic structure on each time fiber.

Original authors: Leonardo Colombo, David Martin de Diego, Riccardo Muradore, Damiano Rigo, Nicola Sansonetto

Published 2026-07-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Leonardo Colombo, David Martin de Diego, Riccardo Muradore, Damiano Rigo, Nicola Sansonetto

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are watching a movie of a swinging pendulum or a planet orbiting a star. In the real world, these systems follow strict, invisible rules that keep their energy and shape consistent over time. Mathematicians call these rules "geometric structures." Now, imagine trying to recreate that movie using a computer. The computer doesn't see a smooth, flowing film; it sees a series of still frames, or "discrete steps."

The problem is that when you use standard, old-school computer methods (like the ones taught in basic physics classes) to take these steps, the movie starts to glitch. The pendulum might slowly gain energy and swing higher and higher until it flies off the screen, or it might lose energy and stop too soon. This is because the standard methods break the invisible geometric rules that keep the system stable. They are like a clumsy photographer who takes pictures of a spinning dancer but accidentally makes her look like she's growing taller with every frame.

This paper introduces a new, smarter way to take those digital snapshots. The authors, a team of mathematicians, propose a method called "discrete-time generalized canonical transformations." Think of this as a special camera lens that doesn't just take a picture of the dancer; it understands the rules of her spin and ensures that every single frame respects those rules, even when time is moving forward and the system is changing (which they call "non-autonomous").

The Big Idea: The Extended Stage
To understand how they fix the glitch, you have to imagine the stage differently. Usually, we think of a system as having a position (where it is) and a momentum (how fast it's going). But because the system changes over time, the authors suggest adding "time" as a character on the stage, just like the position and momentum. They build a giant, extended playground called the "extended phase space."

In this playground, they use a special trick: they treat the time-dependent system as if it were a standard, time-independent system living in this bigger world. They then use a "discretization map"—think of it as a precise ruler that measures the distance between two frames in a way that perfectly preserves the geometry. When they project this perfect, big-world movement back down to our normal, time-changing world, the result is a digital simulation that never loses its shape.

What They Prove and What They Reject
The authors explicitly argue against using standard methods like the "explicit Euler scheme" for these tricky, time-changing systems. They show through a simple example of a damped harmonic oscillator (a spring that loses energy) that the standard method fails immediately. If you use the standard method, the computer calculates that the system's volume (a measure of how much space the system occupies in its state) grows larger than it should. It's like the dancer suddenly expanding into a giant balloon. The authors prove that their new method, however, keeps that volume exactly the same, just like the real physics demands.

They also show that their method preserves something called the "Poisson bracket," which is a fancy way of saying the relationship between position and momentum stays perfectly locked in sync. In their simulations, while the standard method (Runge-Kutta) starts to drift and diverge over long periods, their new integrator keeps the system dancing in a tight, stable rhythm.

The Proof is in the Simulations
The authors didn't just guess this would work; they ran the numbers. They tested their method on two specific scenarios:

A Damped Harmonic Oscillator: A spring with friction, where the energy decays over time. They used a damping constant of $\gamma = 0.2$ and ran the simulation for $50$ seconds. They also ran a longer test with a very small damping constant ( $\gamma = 2 \cdot 10^{-4}$ ) over a massive time horizon of $5 \cdot 10^5$ seconds. In these simulations, their method kept the "volume form" (the system's shape) constant, while the standard method let it drift apart.
A Harmonic Oscillator with a Time-Dependent Potential: A spring where the stiffness changes over time like a sine wave. They used parameters $\epsilon = 0.0035$ and $\alpha = 0.0008$ , running for $50$ seconds and then a long horizon of $2 \cdot 10^5$ seconds. Again, their method kept the geometric structures intact, while the standard method showed a slight bias and divergence.

They also applied this to a charged particle moving in an electromagnetic field. Here, they showed that their method respects "gauge transformations." Imagine you can describe the same magnetic field using different mathematical maps (gauges). The authors proved that if you switch from one map to another, their digital simulation switches perfectly with it, keeping the physical reality (the particle's actual path) exactly the same, whereas the standard method might get confused by the switch.

The Bottom Line
The paper suggests that by using this geometric approach, we can build computer simulations of time-changing systems that are far more reliable over long periods. They don't claim to have solved every problem in the universe, but they have built a robust tool that prevents the "energy drift" and "shape distortion" that plague standard methods.

Looking ahead, the authors suggest this framework could be very useful for "geometric filters" on Lie groups (which are complex shapes used in robotics and navigation). They imagine using this method to help robots estimate their position in real-time without losing their geometric "balance." But for now, they have successfully shown that by treating time as a special coordinate and using these generalized transformations, we can keep the digital dance of the universe in perfect step with the real one.

Technical Summary: Discrete-time Generalized Canonical Transformations for Non-Autonomous Systems

Problem Statement
The discretization of non-autonomous Hamiltonian systems presents a significant challenge in geometric numerical integration. While symplectic integrators are well-established for autonomous systems, preserving the geometric structures of time-dependent systems remains difficult. Standard integrators, such as explicit Euler schemes, typically fail to conserve the cosymplectic volume or the underlying Poisson structure when applied to time-dependent Hamiltonians. Specifically, while the continuous flow preserves the canonical Poisson bracket and symplecticity on each time fiber $T^*Q \times \{t\}$ , standard numerical methods often introduce spurious energy drift and fail to maintain "slice-symplecticity." Existing approaches, such as those by Marthinsen and Owren, characterize discrete canonical transformations as symplectomorphisms on the extended phase space $T^*(Q \times \mathbb{R})$ but do not explicitly describe how these transformations act on the projected, time-dependent space $T^*Q \times \mathbb{R}$ to preserve cosymplectic invariants.

Methodology
The authors propose a geometric method for discretizing non-autonomous Hamiltonian systems based on generalized canonical transformations. The core of the methodology involves:

Extended Phase Space Formulation: The system is lifted from the time-dependent space $T^*Q \times \mathbb{R}$ (endowed with a cosymplectic structure $(\omega_0, \eta)$ ) to the extended phase space $T^*(Q \times \mathbb{R})$ (endowed with a canonical symplectic structure $\Omega_0 = dq \wedge dp + dt \wedge dp_t$ ). An extended Hamiltonian $\tilde{H} = H \circ \mu + p_t$ is defined, where $\mu$ is the projection.
Discretization Maps: The approach utilizes the concept of discretization maps (retractions) defined on the tangent bundle, which are then lifted to the cotangent bundle. A discretization map $R_d: TQ \to Q \times Q$ is generalized to the extended setting $M = Q \times \mathbb{R}$ .
Cotangent Lifts and Symplectomorphisms: The construction employs a sequence of symplectomorphisms and anti-symplectomorphisms to lift the discretization map from the configuration space to the cotangent bundle. This includes:
- The cotangent lift of the diffeomorphism $R_d$ .
- The canonical symplectomorphism $\alpha_Q$ between $TT^*Q$ and $T^*TQ$ .
- The symplectomorphism $\Phi$ between $T^*(Q \times Q)$ and $T^*Q \times T^*Q$ .
- The canonical anti-symplectomorphism $I_{TM}$ relating $T^*T^*M$ and $T^*TM$ .
Lagrangian Submanifolds: The discrete flow is implicitly defined by a Lagrangian submanifold in $T^*T^*(Q \times \mathbb{R})$ derived from the differential of the extended Hamiltonian and the discretization map. This submanifold defines a map $\Psi_h$ on the extended phase space.
Projection: The map $\Psi_h$ is shown to be $\mu$ -projectable, meaning it induces a map $\phi_h$ on the projected space $T^*Q \times \mathbb{R}$ .

Key Contributions

Generalized Canonical Transformations: The paper defines a pair of diffeomorphisms $(\psi, \phi)$ where $\psi$ is a symplectomorphism on the extended space $T^*(Q \times \mathbb{R})$ and $\phi$ is its projection onto $T^*Q \times \mathbb{R}$ .
Structure Preservation Proofs: The authors prove that if the discrete map on the extended space is a canonical transformation satisfying specific fiber-commuting conditions, its projection $\phi$ $ϕ$ preserves:
1. The cosymplectic volume form $\text{vol}_0 = \omega_0^n \wedge \eta$ .
2. The Poisson bracket $\{\cdot, \cdot\}_0$ associated with the cosymplectic structure.
3. The symplectic form $\omega_Q$ on each time fiber $T^*Q \times \{t\}$ , ensuring the induced map is a symplectomorphism for fixed time steps.
Global, Coordinate-Free Framework: The work extends local matrix conditions (previously established by Marthinsen and Owren) to a global, coordinate-free framework that connects symplectic and cosymplectic discretizations via the projection $\mu$ .
Gauge Invariance Application: The framework is applied to gauge-invariant systems (specifically a charged particle in an electromagnetic field). The authors demonstrate that gauge transformations induce canonical transformations on the extended phase space and that the discrete evolution maps in different gauges are related by conjugation, preserving the geometric structure across gauge choices.

Results
Numerical simulations were conducted on two benchmark systems:

Damped Harmonic Oscillator: A system with explicit time dependence due to an exponential damping factor.
Harmonic Oscillator with Time-Dependent Potential: A system where the potential energy varies with time.

In both cases, the proposed geometric integrator (based on the implicit midpoint rule derived from the discretization map) was compared against a standard fourth-order Runge-Kutta (RK4) scheme.

Symplectic and Volume Preservation: The proposed integrator maintained the symplectic form and volume form with bounded oscillations around a constant value, whereas the RK4 method exhibited divergence.
Hamiltonian Behavior: While the symplectic integrator did not preserve the Hamiltonian value exactly, it produced oscillations around the correct average value, avoiding the systematic energy drift observed in the RK4 method.
Gauge Consistency: In the electromagnetic field application, the simulations confirmed that physical observables (position and mechanical momentum) remained consistent across different gauge choices (Coulomb, symmetric, Landau, Lorenz), while canonical momenta varied as expected. The discrete trajectories in different gauges were shown to be exactly related by the discrete gauge transformation.

Significance
The paper claims to provide the first geometric proof of structure preservation for discrete canonical transformations in the non-autonomous case. By bridging the discretization-map formalism with canonical transformation theory, the authors offer a faithful discrete analogue of continuous non-autonomous Hamiltonian dynamics. The method ensures that numerical simulations do not suffer from artificial energy drift and correctly reproduce the geometric features of the continuous flow, such as volume conservation and symplecticity on time slices. This framework is presented as a robust alternative to traditional discretization schemes, particularly for long-term integration and applications requiring geometric consistency, such as filtering and control on Lie groups.

Technical Summary: Discrete-time Generalized Canonical Transformations for Non-Autonomous Systems

More like this