Corrosion detection by identification of a nonlinear Robin boundary condition

Technical Summary: Corrosion Detection by Identification of a Nonlinear Robin Boundary Condition

Problem Formulation
This paper investigates an inverse boundary value problem motivated by corrosion detection. The physical model describes the electric potential $u$ within a bounded, connected domain $\Omega \subset \mathbb{R}^n$ ($n \geq 2$) with a $C^1$ boundary. The boundary is partitioned into an accessible region $\Gamma_A$, where current $f$ is prescribed and voltage is measured, and an inaccessible region $\Gamma_I$, where corrosion may occur.

The governing equation is the conductivity equation $-\text{div}(\gamma\nabla u) = 0$ in $\Omega$, where the conductivity $\gamma$ is symmetric and uniformly elliptic. The boundary conditions are:

1. Neumann condition on $\Gamma_A$: $\partial_{\nu,\gamma}u = f$.
2. Nonlinear Robin condition on $\Gamma_I$: $\partial_{\nu,\gamma}u + a(x, u) = 0$.

Here, $\partial_{\nu,\gamma}u = (\gamma\nabla u) \cdot \nu$ is the conormal derivative. The function $a(x, u)$ models the corrosion; if $a \equiv 0$, there is no corrosion, while a non-zero $a$ indicates its presence. The inverse problem seeks to determine the nonlinear term $a(x, u)$ uniquely from Cauchy data $(u|_{\Gamma_A}, \partial_{\nu,\gamma}u|_{\Gamma_A})$ measured on $\Gamma_A$.

The paper defines the parameter-to-measurement map via the global Cauchy data set $C_a$, consisting of all possible boundary measurements generated by solutions to the forward problem. The primary questions addressed are:

• IP1 (Global Identification): Does $C_{a_1} = C_{a_2}$ imply $a_1 = a_2$ on $\Gamma_I \times \mathbb{R}$?
• IP2 (Local Identification): Does the equality of local Cauchy data sets (measurements in a small neighborhood of a specific solution) imply local equality of the nonlinearities?

Methodology
The authors adapt a strategy originally developed for semilinear elliptic equations (specifically referencing [17] and [18]) to the nonlinear Robin setting. The core methodology relies on linearization and the parametrization of nonlinear solutions by solutions of the linearized equation.

1. Solvability and Linearization:
   The paper first establishes the solvability of the nonlinear equation (1.1) for a class of "admissible nonlinearities" $A_{ad}$. Unlike approaches requiring strict structural assumptions (e.g., positivity of the derivative), this work assumes only that the nonlinearity allows for at least one solution.
   A crucial step involves decomposing the solution space around a known solution $w$. Using the implicit function theorem and fixed-point arguments (Lemma 2.2 and 2.3), the authors show that solutions $u$ close to $w$ can be uniquely parametrized as $u = w + v + Q(v)$, where $v$ is a solution to the linearized equation:
   $$\begin{cases} -\text{div}(\gamma\nabla v) = 0 & \text{in } \Omega, \\ \partial_{\nu,\gamma}v = g & \text{on } \Gamma_A, \\ \partial_{\nu,\gamma}v + \partial_u a(x, w)v = 0 & \text{on } \Gamma_I. \end{cases}$$
   Here, $Q(v)$ represents a higher-order remainder term.

2. Unique Continuation and Runge Approximation:
   The identification proof utilizes two fundamental tools from the theory of linear elliptic equations:

   • Unique Continuation: If two solutions have identical Cauchy data on $\Gamma_A$, they are identical throughout $\Omega$.
   • Runge Approximation: The set of traces of solutions to the linearized equation on $\Gamma_I$ is dense in $C(\Gamma_I)$. This allows the construction of solutions $v$ that approximate arbitrary functions on the inaccessible boundary.

3. Local to Global Strategy:
   The method identifies the nonlinearity $a(x, z)$ on the "graph" of solutions. Specifically, if the local Cauchy data sets coincide, the nonlinearities are shown to be equal for arguments $z$ that correspond to the values of solutions $u(x)$ having Cauchy data in the local set. To achieve global identification, one must characterize the reachable set $E_a = \{(x, u(x)) \in \Gamma_I \times \mathbb{R} : u \text{ is a solution}\}$. If $E_a = \Gamma_I \times \mathbb{R}$, global identification follows.

Key Results

• Theorem 1.5 (Local Identification):
  The paper proves that if two admissible nonlinearities $a_1$ and $a_2$ generate identical local Cauchy data sets around solutions $w_1$ and $w_2$ (which themselves have equal Cauchy data), then $w_1 = w_2$ on $\Gamma_I$. Furthermore, there exists a $\lambda > 0$ such that:
  $$a_1(x, w_1(x) + \varepsilon) = a_2(x, w_1(x) + \varepsilon)$$
  for almost every $x \in \Gamma_I$ and $|\varepsilon| \leq \lambda$. This confirms that the nonlinear term is uniquely determined locally in the state variable $u$ around the solution $w_1$.

• Theorem 1.6 (Global Data and Reachable Sets):
  If the global Cauchy data sets are equal ($C_{a_1} = C_{a_2}$), then:

  1. The reachable sets are identical: $E_{a_1} = E_{a_2}$.
  2. The nonlinearities are equal on this reachable set: $a_1(x, z) = a_2(x, z)$ for almost every $(x, z) \in E_{a_1}$.
  3. The reachable set $E_a$ is an open subset of $\Gamma_I \times \mathbb{R}$.

Significance and Claims
The paper claims to provide a rigorous framework for identifying nonlinear corrosion terms from boundary measurements. Its primary contribution is the successful adaptation of linearization techniques from semilinear elliptic equations to the inverse nonlinear Robin problem.

The authors emphasize that their approach requires minimal assumptions on the nonlinearity (only that a solution exists), avoiding the need for the nonlinearity to be real-analytic or satisfy strict monotonicity conditions often required in higher-order linearization methods.

Regarding the global inverse problem (IP1), the paper is modest. It does not claim to have proven that $E_a = \Gamma_I \times \mathbb{R}$ for all admissible nonlinearities. Instead, it establishes that $E_a$ is an open set and that global identification is equivalent to characterizing this reachable set. The paper suggests that insights into the behavior of solutions (specifically, whether they can cover the entire state space) are the key to solving the global problem, framing the characterization of $E_a$ as a distinct, open problem (IP4) for future investigation.

The work also serves to clarify the definition of local Cauchy data in the nonlinear setting, utilizing boundary norms rather than interior norms to align with the physical reality of boundary measurements.