🔢 mathematics

Cavity shape reconstruction with a homogeneous Robin condition via a constrained coupled complex boundary method with ADMM

This paper proposes a constrained coupled complex boundary method solved via the alternating direction method of multipliers (ADMM) to reconstruct unknown cavity shapes from Cauchy data by reformulating the inverse problem as a complex boundary value problem and minimizing a cost functional based on the imaginary part of the solution.

Original authors: Mustapha Essahraoui, El Mehdi Cherrat, Lekbir Afraites, Julius Fergy Tiongson Rabago

Published 2026-05-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mustapha Essahraoui, El Mehdi Cherrat, Lekbir Afraites, Julius Fergy Tiongson Rabago

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 have a solid block of cheese, but you can't see inside it. You know there's a hidden hole (a "cavity") somewhere inside, but you can't touch it directly. All you can do is poke the outside surface of the cheese, measure how the temperature changes, and listen to how the heat flows. Your goal is to figure out the exact shape of that hidden hole just by looking at the data on the outside.

This is the problem the paper tackles: finding a hidden shape inside an object using limited measurements.

Here is how the authors solve it, explained through simple analogies:

1. The Problem: The "One-Shot" Guessing Game

Usually, if you try to guess the shape of a hidden hole based on one set of measurements, you might get it wrong. It's like trying to guess the shape of a shadow cast by a complex 3D object using only one light source; many different shapes could cast the exact same shadow.

In the world of math, this is called an inverse problem. It's "ill-posed," meaning the answer isn't unique or stable. If your measurements have even a tiny bit of "noise" (like static on a radio), your guess for the shape could go wildly off track.

2. The Old Way: Trying to Fit a Puzzle Piece

Traditional methods try to fix this by guessing a shape, simulating what the measurements should look like, and then tweaking the shape to make the simulation match the real data.

The Analogy: Imagine you are trying to fit a key into a lock, but you can only see the keyhole from the outside. You keep guessing the shape of the key, testing it, and filing it down.
The Flaw: If the hidden hole has a "dent" or a concave curve (like a crescent moon), these old methods often struggle. They tend to smooth out the dents, making the hole look rounder than it really is.

3. The New Trick: The "Complex Color" Method (CCBM)

The authors introduce a clever new tool called the Coupled Complex Boundary Method (CCBM).

The Analogy: Instead of just looking at the "real" temperature data, they imagine the data has a "color" or a "shadow" attached to it (mathematically, they turn the problem into the complex number system, using real and imaginary parts).
How it works: They set up a rule where the "imaginary" part of the solution should be zero if the shape is correct.
The Goal: They treat the problem like a game of "Hot and Cold." They want to find the shape where the "imaginary" part of the data disappears completely. If the imaginary part is still there, they know they are in the wrong spot and adjust the shape.
Why it helps: This method acts like a better compass. It is much more sensitive to those tricky "dents" or concave curves that the old methods miss.

4. The Safety Net: The "Fence" (Inequality Constraints)

Even with the new "Complex Color" method, if the data is noisy or the starting guess is terrible, the algorithm might still wander off course.

To fix this, the authors add a constraint.

The Analogy: Imagine you are trying to find a hidden treasure in a field. You have a map, but it's a bit blurry. To stop yourself from wandering into a swamp, you build a fence around the area where you know the treasure must be. You tell your search robot: "You can move anywhere, but you cannot go outside this fence."
The Math: They use a technique called ADMM (Alternating Direction Method of Multipliers). This is like a smart project manager that splits the problem into two tasks:
1. Task A: Adjust the shape to match the data.
2. Task B: Make sure the shape stays within the "fence" (the physical limits of the data).
  The manager switches back and forth between these tasks until they agree.

5. The Results: Better Shapes, Even with Noise

The paper runs many computer simulations to test this new "Fenced Complex Color" method.

The Test: They tried to reconstruct shapes like ellipses, kites, squares, and L-shaped blocks (which have sharp corners and dents).
The Noise: They added "static" to the data to simulate real-world imperfections.
The Outcome:
- Without the fence: The method worked okay for simple shapes but struggled with dents and noise.
- With the fence (ADMM): The method became much more stable. Even when the starting guess was bad or the data was noisy, it still managed to find the correct shape, including those tricky concave dents.
- The "Lighting" Matters: They found that the way they "shined the light" on the object (the type of data they fed in) mattered a lot. Using a more complex, wavy pattern of data helped them see the dents much better than using a simple, flat pattern.

Summary

The paper presents a new way to find hidden holes inside objects.

They use a mathematical trick (complex numbers) to make the search more sensitive to dents.
They add a safety fence (constraints) to keep the search from going crazy when the data is noisy.
They use a smart manager (ADMM) to balance the search and the safety rules.

The result is a method that is more robust and accurate than previous techniques, especially when trying to find hidden shapes with "dents" or when the data isn't perfect.

Technical Summary: Cavity Shape Reconstruction with a Homogeneous Robin Condition via Constrained CCBM and ADMM

Problem Statement
This study addresses the inverse boundary identification problem of reconstructing an unknown interior boundary $\Gamma$ of a cavity within a conducting body $\Omega$ . The domain is doubly connected, bounded by an accessible exterior boundary $\Sigma$ and the inaccessible interior boundary $\Gamma$ . The state variable $u$ (representing electric potential or temperature) satisfies the Laplace equation $-\Delta u = 0$ in $\Omega$ . On the accessible boundary $\Sigma$ , Cauchy data $(f, g)$ is prescribed, where $u=f$ and $\partial_n u = g$ . On the unknown boundary $\Gamma$ , the function satisfies a homogeneous Robin condition: $\partial_n u + \alpha u = 0$ , where $\alpha$ is a strictly positive constant.

The problem is inherently ill-posed. While local uniqueness may hold for specific domains, a single Cauchy pair generally corresponds to infinitely many admissible domains, making the reconstruction of $\Gamma$ indeterminate without additional constraints or regularization. Furthermore, classical shape optimization methods often struggle to accurately recover concave or non-convex features of the boundary, particularly in single-measurement settings.

Methodology
The authors propose a novel application of the Coupled Complex Boundary Method (CCBM) combined with an Alternating Direction Method of Multipliers (ADMM) framework to address these challenges.

CCBM Formulation:
The original overdetermined real-valued problem is reformulated as a complex-valued boundary value problem. The Cauchy data on $\Sigma$ are coupled via a complex Robin condition: $\partial_n u + i\rho u = g + i\rho f$ on $\Sigma$ , while the physical Robin condition $\partial_n u + \alpha u = 0$ remains on $\Gamma$ . Here, $u = u_1 + iu_2$ is a complex-valued solution.
The inverse problem is transformed into a shape optimization problem: finding the domain $\Omega$ such that the imaginary part of the solution, $u_2$ , vanishes in $\Omega$ . The cost functional to be minimized is $J(\Omega) = \frac{1}{2} \int_\Omega |u_2|^2 dx$ .
Shape Sensitivity Analysis:
The paper derives the shape derivative of the cost functional using shape calculus. The gradient is expressed as a boundary integral over $\Gamma$ involving the state solution $u$ and an adjoint state $p$ . The adjoint system is a complex PDE with specific boundary conditions derived from the Lagrangian of the optimization problem.
Constrained Optimization via ADMM:
To improve stability against noise and poor initialization, particularly for concave geometries, the authors introduce an inequality constraint on the real part of the state variable: $a \leq \Re\{u\} \leq b$ almost everywhere in $\Omega$ .
This leads to a constrained shape optimization problem (Problem 3). To solve this, the authors employ an ADMM strategy:
- Variable Splitting: An auxiliary variable $v$ is introduced to decouple the inequality constraint from the state equation.
- Augmented Lagrangian: The problem is solved by minimizing an augmented Lagrangian functional.
- Iterative Scheme: The algorithm alternates between:
  - Shape Update (SP1): Minimizing the augmented Lagrangian with respect to the shape $\omega$ (using the derived shape gradient and a Sobolev gradient method for mesh deformation).
  - Variable Update (SP2): Updating the auxiliary variable $v$ via a projection onto the admissible convex set $K = \{v \in L^2(\Omega) : a \leq v \leq b\}$ .
  - Multiplier Update: Updating the Lagrange multiplier $\lambda$ .

Key Contributions

Novel Application of CCBM: This work extends the CCBM framework, previously used for inverse source and conductivity problems, to the identification of Robin boundaries.
Constrained Formulation: The integration of inequality constraints into the CCBM framework via ADMM is a primary contribution. This approach enforces physical bounds on the state variable without introducing additional geometric regularization terms (like perimeter penalization), thereby preserving the original CCBM structure while enhancing stability.
Shape Derivative Derivation: The paper provides a rigorous derivation of the shape gradient for the complex state system coupled with the ADMM splitting, including multiple alternative expressions for the gradient based on different adjoint formulations.
Numerical Implementation: The method is implemented using the Finite Element Method (FEM) in FreeFem++, utilizing an $H^1$ -Riesz representation of the shape gradient to ensure smooth mesh deformations.

Numerical Results
The proposed method was validated through various numerical experiments involving different cavity shapes (ellipse, kite, square, L-block) and noise levels.

Concavity Detection: The unconstrained CCBM method showed improved performance over classical approaches in detecting concave regions but still exhibited limitations in resolving the depth of concavities, especially with small Robin coefficients ( $\alpha$ ) or noisy data.
Impact of Constraints: The ADMM-constrained approach demonstrated superior robustness. It effectively suppressed spurious oscillations induced by noise and provided more stable reconstructions across different initial guesses.
Parameter Sensitivity: The weight parameter $\rho$ in the complex coupling was found to influence sensitivity to geometric features; larger values improved the resolution of fine details but required careful tuning to avoid noise amplification.
Data Influence: The quality of reconstruction was highly dependent on the prescribed Dirichlet data $f$ . Spatially varying excitations (e.g., trigonometric functions) yielded significantly better recovery of concave features compared to constant data.
Noise Robustness: Under low noise levels ( $\delta = 0.005, 0.01$ ), the constrained ADMM method maintained geometric consistency, whereas the unconstrained method showed greater sensitivity to initialization and noise.

Significance and Claims
The paper claims that the proposed ADMM-CCBM framework offers a more stable and robust approach for inverse shape reconstruction problems involving homogeneous Robin conditions compared to the standard CCBM formulation. The significance lies in the ability to handle the severe ill-posedness of single-measurement inverse problems, particularly in recovering non-convex geometries.

The authors modestly note that while the method improves stability and reduces oscillations, the accurate recovery of the depth and sharpness of concavities remains challenging due to the inherent ill-posedness of the problem. The improvement is attributed to the enforcement of physically motivated state constraints via ADMM, which guides the optimization process more effectively than unconstrained least-squares or energy-gap functionals. The work does not claim to solve the identifiability issue (uniqueness) but rather provides a numerical strategy that yields "reasonable reconstructions" and "improved qualitative recovery" where classical methods often fail or produce artifacts. Future work is suggested to include multiple measurements and adaptive parameter selection.