🔢 mathematics

Identification of space-dependent coefficients in two competing terms of a nonlinear subdiffusion equation

This paper proposes and analyzes a fixed-point scheme for uniquely identifying space-dependent coefficients in a nonlinear subdiffusion equation using interior observations under specific excitation scenarios, with theoretical guarantees of convergence and local uniqueness supported by numerical experiments.

Original authors: Barbara Kaltenbacher, William Rundell

Published 2026-02-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Barbara Kaltenbacher, William Rundell

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 a detective trying to solve a mystery inside a sealed room. The room is filled with a mysterious, slow-moving fog (this represents the subdiffusion equation). This fog doesn't just drift randomly; it reacts to two invisible forces inside the room:

The "Growth" Force ( $q$ ): A hidden heater that tries to make the fog expand or multiply.
The "Limit" Force ( $p$ ): A hidden sponge that tries to absorb the fog or stop it from growing too fast.

The paper is about figuring out exactly where these heaters and sponges are located and how strong they are, just by looking at the fog at specific moments in time.

The Mystery: The "Fog" Equation

In the real world, this "fog" could be anything from bacteria growing in a petri dish, a chemical reaction, or even the spread of a rumor. The math used to describe this fog is complex because:

It moves slowly and erratically (like a drunkard's walk), not smoothly like water.
The growth and absorption forces change depending on where you are in the room (they are "space-dependent").
The fog interacts with itself in a non-linear way (the more fog there is, the more it might react).

The authors want to work backward. They know the rules of the game (the equation), but they don't know the specific settings of the heater ( $q$ ) and the sponge ( $p$ ). They only have a few snapshots of the fog.

The Detective's Toolkit: The "Fixed Point" Scheme

To solve this, the authors invented a clever guessing game called a Fixed Point Scheme. Think of it like tuning a radio to find a clear station:

The Guess: You start with a wild guess about where the heaters and sponges are.
The Simulation: You run a computer simulation using your guess to see what the fog should look like.
The Comparison: You compare your simulated fog with the actual fog you measured in the room.
The Correction: The math calculates exactly how much you were wrong. It doesn't just say "you're off"; it gives you a specific formula to adjust your guess for the heater and the sponge.
The Loop: You repeat this process. With every loop, your guess gets closer and closer to the truth, until the simulation matches the real fog perfectly.

The Two Ways to Catch the Culprit

The paper proves that this detective work works, but you need at least two pieces of evidence (measurements) to solve the case. The authors tested two scenarios:

Scenario A: The "Two-Source" Test (The "Two-Run" Case)
Imagine you can turn on two different "fog machines" (inputs) one after the other.

Run 1: Turn on Fog Machine A, wait until time $T$ , and take a photo.
Run 2: Turn on Fog Machine B, wait until time $T$ , and take a photo.
Result: This is like having two different suspects. Because the fog reacts differently to each machine, it's very easy to separate the effects of the heater from the sponge. The paper shows this method works very well.

Scenario B: The "Two-Time" Test (The "Single-Run" Case)
Imagine you can only turn on one Fog Machine, but you can take photos at two different times.

Photo 1: Take a picture at time $T_1$ (early in the process).
Photo 2: Take a picture at time $T_2$ (later in the process).
Result: This is harder. If you take the photos too close together in time, the fog hasn't had enough time to show the difference between the heater and the sponge. The paper finds that you need a big gap between $T_1$ and $T_2$ (one very early, one very late) to get a good solution. If the gap is small, the detective work fails.

The "Magic" of the Math

The authors proved two big things:

Uniqueness: There is only one correct answer for the heater and sponge locations. You won't get tricked into thinking there are two different solutions that look the same.
Convergence: The guessing game (the algorithm) is guaranteed to stop eventually. It won't spin in circles forever; it will lock onto the correct answer, provided you wait long enough for the fog to settle.

What the Experiments Showed

The authors ran computer experiments to see how well their detective worked:

Stronger Reactions Help: If the fog reacts strongly to the sponge (a specific mathematical term called $f(u)$ ), it's easier to find the sponge.
Time Matters: Waiting longer (letting the fog settle) makes the solution more accurate.
Noise Tolerance: Even if the photos are a little blurry (noisy data), the method still works, as long as you run enough iterations (guesses) to smooth out the errors.

The Bottom Line

This paper provides a mathematical "recipe" for reverse-engineering hidden properties inside a complex, slow-moving system. By using a smart, iterative guessing game and taking measurements at the right times (or using two different inputs), we can accurately map out invisible forces that control how things spread and react in the real world.

Technical Summary: Identification of Space-Dependent Coefficients in a Nonlinear Subdiffusion Equation

Problem Statement
This paper addresses the inverse problem of identifying two spatially dependent coefficients, $p(x)$ and $q(x)$ , in a nonlinear subdiffusion equation of the form:
$\partial_t^\alpha u - D\Delta u = q(x)u - p(x)f(u) + r(t,x)$
where $\partial_t^\alpha$ denotes the Djirbashian-Caputo fractional derivative of order $\alpha \in (0,1]$ , and $f(u)$ is a known smooth nonlinearity (e.g., $u^2$ for Fisher-KPP, $u^3$ for Allen-Cahn). The goal is to reconstruct the unknown functions $p(x)$ and $q(x)$ from interior observations of the solution $u(x,t)$ .

The authors consider two distinct observation scenarios requiring at least two measurements to resolve the two unknowns:

"Two-run" scenario: Two different source inputs $r_1, r_2$ are applied, and measurements are taken at a single final time $T$ .
"Two-time" scenario: A single source input is applied, and measurements are taken at two distinct time instances, $T_1$ and $T_2$ .

Methodology
The core methodology involves the construction of a fixed-point scheme for the simultaneous reconstruction of $p(x)$ and $q(x)$ .

Fixed-Point Formulation: By evaluating the governing PDE at the observation points and utilizing the known values of the solution $g_i(x) = u(T_i, x)$ , the authors derive a system of linear equations for $p$ and $q$ at each spatial point. Specifically, they eliminate the time derivative term $\partial_t^\alpha u$ by combining the equations from the two measurements. This yields an explicit pointwise update formula:
$\begin{pmatrix} p \\ q \end{pmatrix} = \mathcal{T}(p, q)$
where the operator $\mathcal{T}$ depends on the residuals between the model prediction and the observed data. Crucially, this scheme does not require a basis representation (e.g., finite elements or spectral methods) for the unknown coefficients, allowing for pointwise updates.
Convergence Analysis: The convergence of the fixed-point iteration is established via a contraction argument. The analysis relies on the assumption that the system evolves toward a steady state $u^\infty$ as $t \to \infty$ . The authors prove that for sufficiently large $T$ (in the two-run case) or sufficiently large $T_2$ with $T_1$ close to zero (in the two-time case), the determinant of the linear system (involving $u$ and $f(u)$ ) remains bounded away from zero.
- The proof utilizes energy methods and decay estimates for the difference between the solution with true coefficients and the solution with iterated coefficients.
- Key conditions include the positivity of the linearized potential at the steady state and specific growth conditions on the nonlinearity $f(u)$ .
- The authors demonstrate that the fixed-point operator is a contraction in a neighborhood of the true coefficients, implying local uniqueness and convergence of the scheme.
Handling Noise and Regularity: The paper notes that since the reconstruction scheme involves spatial derivatives of the data (specifically $\Delta g$ ), noisy measurements must be smoothed to lie in $H^2(\Omega)$ . The authors show that the inverse problem is well-posed in $H^2(\Omega)$ , meaning that with sufficient iterations (scaling as $O(|\log \delta|)$ for noise level $\delta$ ), the reconstruction error is $O(\delta)$ without the need for early stopping.

Key Results

Theoretical Guarantees: The paper proves that the fixed-point scheme converges to the unique solution $(p, q)$ within a local ball around the true coefficients, provided the observation time is sufficiently large and the coefficients satisfy specific regularity and positivity conditions.
Numerical Performance: Numerical experiments illustrate the scheme's performance across various nonlinearities ( $u^2, u^3, 4u^2, \frac{1}{4}u^3$ $u^{2}, u^{3}, 4 u^{2}, \frac{1}{4} u^{3}$ ) and fractional orders $\alpha$ $α$ .
- Two-run vs. Two-time: The "two-run" scenario (two sources, one time) generally yields superior reconstructions compared to the "two-time" scenario (one source, two times).
- Time Gap: In the two-time scenario, the quality of reconstruction is highly sensitive to the time gap between measurements. A larger gap (where $T_1$ is near 0 and $T_2$ is large) produces significantly better results than a small gap between $T_1$ and $T_2$ .
- Nonlinearity Impact: The magnitude of the nonlinearity $f(u)$ influences the sensitivity of the reconstruction. Larger values of $f(u)$ (e.g., $u^3$ ) tend to improve the reconstruction of both $p$ and $q$ by emphasizing the difference between the terms.
- Robustness: The scheme demonstrates robustness to noise, with relative errors decreasing linearly with the noise level $\delta$ after a sufficient number of iterations.
- Initial Guess: The method works effectively even with crude initial guesses (e.g., constant zero values), converging rapidly.

Significance and Claims
The authors position this work as a contribution to the mathematical theory of inverse problems for subdiffusion equations with multiple unknown coefficients.

Minimal Data Requirement: The paper emphasizes that it addresses the mathematical question of what can be achieved with the minimum amount of information sufficient to guarantee uniqueness (two measurements for two unknowns), rather than relying on oversampling.
Algorithmic Efficiency: The proposed fixed-point scheme is highlighted for its simplicity and lack of need for complex basis representations of the unknowns, distinguishing it from Newton-type methods.
Generalization: The framework extends existing results on reaction-diffusion equations to the time-fractional (subdiffusion) setting and handles general nonlinearities $f(u)$ , not just specific polynomial forms.
Modesty: The authors acknowledge that while the theoretical framework assumes specific regularity and decay conditions, the numerical experiments confirm the practical viability of the approach. They note that the fractional order $\alpha$ plays a minor role in the convergence speed for the time ranges considered, though it may be more significant for very large times.

The paper concludes by suggesting that while the primary focus is on the two-measurement case, the framework could be extended to scenarios with $N$ sampling points or alternative experimental setups (e.g., varying initial conditions or boundary conditions) if the determinant condition remains non-singular.