Towards Polynomial Immersion of Port-Hamiltonian Systems

This paper proposes a method to immerse non-polynomial Port-Hamiltonian systems into higher-dimensional polynomial representations that preserve key structural and energy properties, thereby enabling the design of stabilizing feedback laws through sum-of-squares optimization and passivity-based control.

The Gist

Here is an explanation of the paper "Towards Polynomial Immersion of Port-Hamiltonian Systems," translated into everyday language with creative analogies.

The Big Picture: Translating a Foreign Language into a Universal One

Imagine you are an engineer trying to control a complex machine, like a self-balancing robot or a chemical plant. This machine is governed by the laws of physics, specifically energy. It stores energy, loses energy (friction/heat), and moves energy around. In the world of control theory, we call these Port-Hamiltonian (pH) systems. They are beautiful because they respect the laws of thermodynamics.

However, there's a problem. The math describing these machines often involves "weird" functions like exponentials ($e^x$), sines, or logarithms. These are non-polynomial.

Think of polynomials as the "Lego bricks" of mathematics. They are simple, predictable, and easy to snap together. We have powerful, automated tools (like Sum-of-Squares optimization) that can instantly solve problems if the math is built entirely out of these Lego bricks. But if your machine's math is built out of "alien blocks" (exponentials and sines), those tools can't touch it. You have to solve the equations by hand, which is slow, hard, and prone to error.

The Goal of this Paper:
The authors want to take that complex machine with "alien blocks" and translate it into a higher-dimensional version made entirely of "Lego bricks" (polynomials), without breaking the machine's physics. They want to keep the energy balance, the friction, and the interconnections exactly the same, just in a new, easier-to-solve language.

---

The Core Concept: "Lifted Immersion"

The paper proposes a technique called Lifted Immersion. Let's break that down with an analogy.

1. The Problem: The "Alien" Equation

Imagine you are trying to predict the path of a ball rolling down a hill with a weird, bumpy surface. The equation for the bump involves $e^x$. Your computer software for designing controllers (the "Lego tools") gets confused by the $e^x$ and refuses to work.

2. The Solution: The "Shadow Puppet" Trick

Instead of trying to force the software to understand $e^x$, the authors say: "Let's add a new variable to our system."

Imagine you have a shadow puppet show. The puppet (the real system) is complex and moves in strange ways. But if you shine a light from a specific angle, the shadow cast on the wall is a simple, perfect circle.

• The Real System: The complex puppet with $e^x$ and sines.
• The Shadow (The Immersion): A new, higher-dimensional system where the math is purely polynomial (just $x^2$, $x^3$, etc.).
• The "Lifted" part: We don't just look at the shadow; we keep the original puppet inside the new system. We "lift" the original state into a bigger room where we can also track the "shadow" variables.

By adding these extra variables (like adding $e^x$ as a new variable called $z$), the complex relationship becomes a simple multiplication ($z \cdot z$). Suddenly, the "alien" math becomes "Lego" math.

What They Proved: Keeping the Soul of the Machine

The scary part of this trick is usually that when you translate a system, you might lose its essential properties. If you translate a car into a video game, does it still have brakes? Does it still burn fuel?

The authors proved that their specific translation method is structure-preserving.

• Energy is Safe: The total energy of the new polynomial system is exactly the same as the old one.
• Friction is Safe: The way the system loses energy (dissipation) is preserved.
• The "Ports" are Safe: The way the system talks to the outside world (inputs and outputs) remains identical.

They showed that if you start the new system in the right "shadow" position (consistent initialization), it will behave exactly like the original system forever. It's like a perfect twin that speaks a simpler language.

The Payoff: Building Better Controllers

Why do we care? Because once the system is translated into "Lego bricks" (polynomials), we can use powerful, automated tools to design controllers.

In Section 5, the authors show an example. They took a system with exponential terms (hard to control) and translated it into a polynomial system. Then, they used a method called Sum-of-Squares (SOS) optimization.

• Analogy: Imagine you want to build a bridge.
  • Old Way: You have to calculate every stress point by hand, guessing if it will hold. It takes weeks.
  • New Way (This Paper): You translate the bridge design into a format that a super-computer can instantly check. The computer says, "Yes, this bridge is stable," and gives you the exact blueprint for the support beams.

The result? They designed a controller that stabilized the system, proving that the "Lego" version works just as well as the "alien" version.

Summary for the Non-Mathematician

1. The Problem: Real-world machines often have math that is too complex for our best automated design tools.
2. The Trick: We add extra variables to the system to turn complex math (like $e^x$) into simple math (like $x \cdot x$).
3. The Guarantee: We proved that this translation doesn't break the physics. The new system still obeys the laws of energy and friction.
4. The Benefit: Now we can use fast, automated computer tools to design controllers for these complex machines, making them safer and more efficient.

In short, the authors found a way to translate the "foreign language" of complex physics into the "universal language" of polynomials, allowing us to use powerful new tools to control the world around us.

Technical Summary

Here is a detailed technical summary of the paper "Towards Polynomial Immersion of Port-Hamiltonian Systems."

1. Problem Statement

Port-Hamiltonian (pH) systems provide a powerful, energy-based framework for modeling physical systems, capturing interconnection geometry, energy storage, and dissipation. However, many physical systems exhibit non-polynomial nonlinearities (e.g., trigonometric, exponential, or logarithmic functions).

Standard control design techniques, particularly those based on Sum-of-Squares (SOS) optimization, are highly effective for polynomial systems but cannot be directly applied to systems with non-polynomial dynamics. While "lifting" techniques exist to convert non-polynomial systems into polynomial ones, they often:

1. Destroy the intrinsic pH structure (interconnection and dissipation matrices).
2. Obscure passivity and energy balance properties.
3. Introduce algebraic constraints that complicate control synthesis.

The Core Question: Is it possible to immerse a non-polynomial pH system into a higher-dimensional polynomial system while strictly preserving its pH structure (interconnection geometry, energy balance, and passivity)?

2. Methodology

The authors propose a novel technique called Lifted Immersion, which combines concepts from differential algebra and system immersion to achieve structure-preserving polynomial representations.

Key Concepts & Steps:

1. Differential Algebraic (DA) Functions: The paper assumes system functions are real-analytic and satisfy algebraic differential equations. This class includes polynomials, rational functions, exponentials, and trigonometric functions.
2. Lifted Immersion:
   • Instead of a standard immersion (mapping state $x$ to $\tilde{x}$), the authors define a lifted immersion $\Psi(x) = [x^\top, \psi_1(x), \dots, \psi_r(x)]^\top$.
   • This ensures the original state $x$ is explicitly contained in the new state vector $\bar{x}$, allowing the Hamiltonian and original interconnection structure to be preserved directly.
3. Construction of the Polynomial System:
   • Step 1 (Rational Immersion): Using the theory of differential fields, the non-polynomial terms are replaced by auxiliary variables (e.g., $e^{x_1} \to \bar{x}_3$). This yields a rational system where dynamics are ratios of polynomials.
   • Step 2 (Polynomialization): The rational system is further immersed into a polynomial system by introducing new coordinates for the denominators (e.g., $1/\bar{x}_3$).
4. Structure Preservation:
   • The authors construct the new interconnection matrix $\bar{J}(\bar{x})$ and dissipation matrix $\bar{R}(\bar{x})$ such that they retain the skew-symmetry and positive semi-definiteness properties of the original system on the invariant manifold defined by the immersion.
   • The Hamiltonian $H(\bar{x})$ is defined to be identical to the original $H(x)$ when restricted to the manifold.

3. Key Contributions

• Structure-Preserving Polynomial Immersion: The paper proves that any pH system with a rational Hamiltonian and differential algebraic vector fields can be immersed into a higher-dimensional polynomial pH system.
• Preservation of Physical Properties:
  • Interconnection Geometry: The skew-symmetric structure of the interconnection matrix is preserved (modulo a trivial lifting).
  • Energy Balance & Passivity: The paper proves that the dissipation inequality holds along the trajectories of the lifted system. Specifically, the power balance equation $\dot{H} = y^\top u - d$ is preserved, ensuring the polynomial system remains passive with respect to the original Hamiltonian.
  • Invariant Manifold: The image of the immersion $\Psi(D)$ forms an invariant embedded submanifold. If the system is initialized consistently on this manifold, it remains there, and the input-output behavior is identical to the original system.
• Control Design Framework (SOS-IDA-PBC): The authors demonstrate how the resulting polynomial structure enables the use of Interconnection and Damping Assignment Passivity-Based Control (IDA-PBC) combined with Sum-of-Squares (SOS) optimization. This allows for the systematic design of stabilizing controllers for systems that were previously intractable with SOS methods due to non-polynomial terms.

4. Results and Illustrative Examples

The paper validates the theory through two examples and a control design case study:

1. Exponential System Example:

   • A 2D system with exponential dissipation terms ($e^{x_1}, e^{x_2}$) is lifted to a 4D polynomial system.
   • The resulting system preserves the skew-symmetric $J$ and the specific dissipation properties on the manifold.

2. Rolling Coin Example:

   • A classic non-holonomic system involving trigonometric functions ($\sin x_4, \cos x_4$) is modeled as a pH system.
   • The authors successfully lift this to an 8-dimensional polynomial system, preserving the Hamiltonian and interconnection structure.
   • Simulation results show that the state and output trajectories of the polynomial system match the original system exactly when initialized on the manifold.

3. Stabilization via SOS-IDA-PBC:

   • The authors apply the method to a system with exponential terms to design a stabilizing controller.
   • They formulate the matching equations for IDA-PBC as a polynomial problem.
   • Using SOS optimization, they compute a desired Hamiltonian $H_d$ and a feedback law that ensures global asymptotic stability (locally constrained to a ball to ensure feasibility).
   • Simulations confirm the convergence of the Hamiltonian to zero and the stabilization of the state trajectories.

5. Significance and Conclusion

• Bridging the Gap: This work bridges the gap between the rich physical modeling capabilities of Port-Hamiltonian systems and the computational tractability of polynomial control synthesis (SOS).
• No Structural Loss: Unlike previous lifting methods that might obscure the energy-based structure, this approach guarantees that the fundamental physical properties (passivity, energy storage, dissipation) are mathematically preserved on the invariant manifold.
• Enabling Advanced Control: By converting non-polynomial pH systems into polynomial ones, the paper opens the door for applying powerful convex optimization tools (like SOS) to a much broader class of physical systems, including those with complex nonlinearities like friction, aerodynamics, or chemical kinetics.

Future Outlook: The authors suggest extending this framework to data-driven modeling and further exploiting the polynomial structure for more complex control design problems.