⚡ electrical engineering

Safe Trajectory Tracking of the Stefan Problem with Second-Order Moving Boundary Dynamics

This paper proposes a globally exponentially stable, safe trajectory tracking controller for the Stefan problem with second-order moving boundary dynamics, utilizing a series expansion-based feedforward design and an energy-shaping feedback law to ensure constraint satisfaction.

Original authors: Shumon Koga, Miroslav Krstic

Published 2026-05-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Shumon Koga, Miroslav Krstic

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 trying to melt a block of ice into water, but you don't just want it to melt; you want the edge of the melting ice (the "moving boundary") to follow a very specific, wiggly path, like a snake slithering forward. This is the core challenge of the Stefan problem, a classic physics puzzle about how materials change phase (like ice turning to water).

This paper by Koga and Krstic solves a tricky version of this puzzle where the melting edge has "inertia." Think of it like pushing a heavy shopping cart: if you stop pushing, the cart doesn't stop instantly; it keeps rolling a bit. Similarly, the melting edge doesn't stop instantly when the heat changes; it has a "thermal inertia" that makes it hard to control.

Here is how the authors solved the problem of making this edge follow a path safely and accurately:

1. The "Recipe" (Feedforward Control)

First, the authors figured out a "recipe" for what the heat input should be if everything went perfectly. They used a mathematical trick called a series expansion.

The Analogy: Imagine you are trying to bake a cake that needs to rise to a specific height at a specific time. You can't just guess; you need a precise recipe. The authors calculated an infinite list of ingredients (mathematical terms) that, when combined, tell the heater exactly how much power to use to make the ice melt along the desired wiggly path. They proved that this infinite list actually adds up to a real, usable number.

2. The "Safety Guard" (Energy Shaping)

The biggest risk in melting ice is that if you get the heat wrong, the ice might refreeze (which breaks the physics of the problem) or the material might melt away completely. The paper introduces a safety guard based on "energy shaping."

The Analogy: Think of the system's energy like water in a bathtub. The authors designed a controller that acts like a smart faucet and drain. If the water level (energy) gets too low (risk of refreezing) or too high (risk of melting everything), the controller automatically adjusts the flow to keep it in the "safe zone."
The Magic: They proved mathematically that as long as you start with enough heat, the system will never accidentally refreeze, even if the desired path asks for a momentary pause in melting. It's like a self-correcting thermostat that guarantees you never run out of hot water.

3. The "Steering Wheel" (Tracking Stability)

Even with a perfect recipe and a safety guard, the ice might not follow the path exactly because real-world conditions aren't perfect. The authors added a "steering wheel" mechanism using a technique called backstepping.

The Analogy: Imagine driving a car on a winding road. The "recipe" tells you how to turn the wheel, but if you drift off course, the "steering wheel" (the feedback controller) gently nudges you back.
The Result: They proved that any mistake (drifting off the path) will shrink exponentially fast. It's like a rubber band: the further you get from the desired path, the harder the system pulls you back, ensuring you snap back to the correct line very quickly.

4. The Proof (Simulation)

To show this works, they ran a computer simulation using the properties of Zinc (a metal).

The Scenario: They asked the system to follow a path that wiggled back and forth (a sine wave) while slowly moving forward.
The Outcome: The simulation showed the melting edge hugging the wiggly path perfectly. Even when the "recipe" asked for a momentary pause (which would normally be dangerous), the safety guard kicked in, ensuring the heat never dropped below zero. The temperature stayed above the melting point the entire time, and the system settled into the correct path within an hour.

Summary

In short, this paper teaches a computer how to control a melting process with "heavy" inertia. They created a three-part system:

A Precise Recipe: To calculate the ideal heat input.
A Safety Net: To guarantee the material never refreezes or melts away.
A Correction Mechanism: To ensure any errors disappear quickly.

The result is a method that can make a melting boundary follow complex, wiggly paths safely and accurately, which is a big step forward for controlling phase-change materials in engineering.

Technical Summary: Safe Trajectory Tracking of the Stefan Problem with Second-Order Moving Boundary Dynamics

Problem Statement
This paper addresses the safe trajectory tracking control of the one-phase Stefan problem, a classical model for thermal phase change phenomena such as melting and solidification. The specific challenge tackled is the control of a system where the moving boundary (the interface between phases) is governed by second-order dynamics, representing thermal inertia, rather than the more common first-order dynamics. The system is modeled by a parabolic Partial Differential Equation (PDE) for the temperature profile defined on a time-varying spatial domain, coupled with a second-order Ordinary Differential Equation (ODE) for the boundary position.

The control objective is to design a boundary heat flux input, $q_c(t)$ , that drives the system's moving boundary and temperature profile to track a desired reference trajectory, $s_r(t)$ and $T^r(x,t)$ . Crucially, the controller must ensure safety, meaning the physical constraints of the model must be satisfied at all times: the temperature must remain above the melting point ( $T(x,t) \ge T_m$ ) to prevent the liquid phase from freezing at the heat inlet, and the boundary position must remain within the physical domain ( $0 < s(t) < L$ ).

Methodology
The proposed control strategy combines a feedforward component with a feedback controller designed via energy shaping and PDE backstepping.

Feedforward Control (Inverse Stefan Problem):
To generate the reference trajectory, the authors solve the inverse Stefan problem for a given reference boundary motion $s_r(t)$ . They employ a series expansion approach, representing the reference temperature profile as a power series in the spatial coordinate relative to the moving boundary. The coefficients of this series are determined recursively. The paper proves the convergence of this infinite series under the assumption that the reference trajectory is of Gevrey order, ensuring the existence of a well-defined reference temperature profile and the corresponding feedforward heat flux $q_c^r(t)$ .
Feedback Control (Energy Shaping and Safety):
A trajectory tracking controller is derived using an energy-shaping approach. The total energy of the system is defined to include both the thermal energy and the kinetic/potential energy of the moving boundary. The control law is designed as:
$q_c(t) = q_c^r(t) - c(E(t) - E_r(t))$
where $E(t)$ and $E_r(t)$ are the actual and reference energies, respectively, and $c$ is a positive gain.
To guarantee safety (specifically, that the heat flux remains non-negative and the temperature stays above the melting point), the authors utilize the Maximum Principle. They impose specific assumptions on the initial energy and the reference trajectory to ensure that the closed-loop system satisfies the physical constraints. This approach effectively acts as a safety filter, ensuring that even if the reference trajectory demands cooling (negative heat flux), the actual control input remains non-negative, preventing the violation of the phase-change condition.
Stability Analysis:
The global exponential stability of the tracking error is established using Lyapunov analysis combined with a PDE backstepping transformation. The tracking error system is transformed into a target system with desirable stability properties. By constructing a Lyapunov function that accounts for the energy of the error states and the moving boundary dynamics, the authors prove that the tracking error converges exponentially to zero. The analysis carefully handles the time-varying domain and the coupling between the PDE and the second-order ODE.

Key Contributions

Second-Order Dynamics: The paper extends the control of the Stefan problem to include second-order moving boundary dynamics, which better captures thermal inertia in phase-change processes.
Rigorous Safety Proof: Unlike previous works that often assumed safety or static references, this paper provides a rigorous proof that the proposed control law guarantees the satisfaction of physical constraints (non-freezing and domain bounds) for a dynamic reference trajectory.
Convergence of Series Solution: The authors prove the convergence of the series expansion solution for the inverse Stefan problem under second-order boundary dynamics, a necessary step for generating the feedforward control.
Global Exponential Stability: The paper establishes global exponential stability for the closed-loop tracking error using a novel combination of energy shaping, backstepping, and Lyapunov analysis, without relying on linearization or simplified assumptions about the error dynamics.

Results
Numerical simulations were performed using parameters for Zinc melting. The reference trajectory was designed to be a non-trivial sinusoidal function that decays over time, ensuring the boundary moves monotonically but with varying speed.

Tracking Performance: The closed-loop system successfully tracked the reference boundary position and temperature profile, with the tracking error converging to zero within the simulation time (approx. 60 minutes).
Safety Verification: The simulation confirmed that while the reference heat flux occasionally became negative (indicating a theoretical need for cooling), the actual control input remained non-negative, satisfying the safety constraint. The temperature profile remained strictly above the melting point throughout the simulation.
Transient Behavior: The results showed that the controller initially applied a large heat flux to compensate for the lower initial energy of the system compared to the reference, leading to a brief overshoot before settling into accurate tracking.

Significance
The authors claim this work as the first to provide a rigorous proof of both safety and tracking stability for the Stefan problem with a trajectory tracking control. By addressing the geometric nonlinearity and the coupling between the PDE and the second-order ODE, the paper offers a systematic control design that ensures the physical validity of the solution (safety) while achieving precise tracking. The methodology bridges the gap between theoretical control design for PDEs and the practical necessity of satisfying hard physical constraints in phase-change applications. The code for the proposed method is made available to facilitate further research and validation.