🔢 mathematics

Recovery of a Measure-valued Source in the Heat Equation from Sparse Boundary Measurements

This paper establishes a theoretical framework and numerical methodology for uniquely recovering general measure-valued sources in the heat equation using sparse boundary flux measurements, thereby extending existing results from point or $L^2$ sources to a broader class of Radon measures.

Original authors: Ulysse Dalmasso, Siyu Cen, Yavar Kian

Published 2026-07-20

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Ulysse Dalmasso, Siyu Cen, Yavar Kian

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 in a large, empty room, and someone has hidden a secret object inside. You can't see the object, and you can't walk inside the room to find it. However, you have two tiny microphones placed on the walls. If the hidden object starts "screaming" (releasing heat or energy), the sound of that scream travels through the air, hits the walls, and is picked up by your microphones. The big question is: Can you listen to just those two microphones and figure out exactly where the object is, what shape it is, or even if it's a single point or a long, thin line? This is the heart of a field called "inverse problems," where scientists try to work backward from the effects they can measure to find the hidden cause. Usually, these puzzles are incredibly hard because the "scream" gets muffled and mixed up as it travels, making it difficult to tell if the source was a single dot or a long line just from a few listening spots.

This paper tackles a specific version of that puzzle involving heat. The authors are trying to find a "heat source" hidden inside a circular room (a unit disc) by only listening to the heat flow at two specific points on the wall. What makes this paper special is that they are looking for a very messy, complicated kind of source. Most previous studies only looked for simple, neat sources like a single glowing dot or a smooth cloud of heat. But in the real world, pollution often comes from long, thin lines (like a busy road) or strange, jagged shapes. The authors wanted to know: Can we still find these messy, "measure-valued" sources if we only have two microphones? They proved that, mathematically, the answer is yes, provided the two microphones are placed at a very specific, non-repeating angle relative to each other. They also ran computer simulations to show that their method works in practice, successfully reconstructing both single dots and curved lines, even when the microphone data was a bit noisy.

The Mystery of the Two Microphones

Let's dive into the story of how the authors solved this. Imagine the room is a perfect circle, and the heat source is something hidden inside. The heat spreads out over time, like ink dropping into water, following the rules of the "heat equation." The authors set up a scenario where the source is a mix of a time-varying signal (like a siren that gets louder and softer) and a spatial shape (the hidden object itself).

The challenge is that the source isn't necessarily a nice, smooth blob. It could be a "Radon measure," which is a fancy math way of saying it could be a single point, a collection of points, or even a source spread out along a thin curve (like a line of pollution). Previous math tools were great for smooth blobs or single points, but they broke down when the source got too "spiky" or irregular. The authors' main goal was to extend the math to handle these messy, real-world shapes using only two tiny sensors on the boundary.

The Magic of the "Two-Point" Rule

The paper's big breakthrough is a mathematical proof showing that if you place your two sensors at the right spots, you can uniquely identify the hidden source. The authors found a specific rule for where to put these sensors: the angle between them cannot be a simple fraction of a circle (like 90 degrees or 180 degrees). They need to be at an "irrational" angle relative to each other. Think of it like this: if you and a friend are standing on a clock face, and you are at 12 o'clock, your friend shouldn't be at 3, 6, or 9. They need to be at a weird spot, like 12:07 and 12:13, where the gap between you doesn't repeat in a simple pattern. If this condition is met, the math guarantees that no two different hidden sources will ever produce the exact same signal at those two points.

To prove this, the authors used a clever trick involving "time analyticity." Imagine the heat signal as a song. The authors showed that if you know the song for a short while, you can mathematically predict the rest of the song perfectly, even if you can't hear it yet. Because the signal at the wall is "analytic" (smooth and predictable in time), knowing it at two points for a short time is enough to figure out the entire hidden source. They combined this with a detailed breakdown of the heat equation using "eigenfunctions," which are like the natural notes a drum can play. By listening to how the source "sings" these natural notes, they could reverse-engineer the source's shape.

From Theory to the Computer Lab

Proving it on paper is one thing, but does it work in the real world? The authors didn't stop at the math; they built a computer simulation to test their theory. They created a virtual circular room and tried to find hidden sources using their two-point method.

First, they tested "point sources." They hid one, two, three, and even four glowing dots in the room. Their computer algorithm, which uses a method called "Gauss-Newton," tried to guess where the dots were. Even when they added "noise" to the microphone data (simulating static or errors, up to 10% noise), the algorithm successfully found the locations. The more dots they hid, the harder the puzzle became, and the algorithm needed a bit more help (a technique called "regularization") to stay on track, but it still worked.

Next, they tackled the "line source." This is where the source isn't a dot but a wiggly curve, like a snake made of heat. They modeled these curves using a mix of waves (sines and cosines). Again, they used a computer algorithm (this time "Levenberg-Marquardt") to try and reconstruct the shape of the curve from the two microphone readings. The results were impressive: even with noisy data, the computer could redraw the hidden curves with surprising accuracy.

What This Means

The authors didn't just say "it might work"; they provided a rigorous mathematical proof that the solution is unique under their specific conditions. They also showed through simulations that the method is robust enough to handle real-world imperfections like noise. However, they were careful to note that their proof relies on the source being "constant" for the last part of the time interval and that the sensors must be in that specific irrational arrangement. They didn't claim to solve every possible version of this problem, but they successfully opened the door to identifying much messier, more realistic sources than ever before.

In short, this paper takes a difficult math puzzle about finding hidden heat sources and proves that even if the source is a messy line or a cluster of points, two cleverly placed sensors are enough to find it. It's a reminder that sometimes, you don't need a whole army of sensors to solve a mystery; you just need the right two, and the right math to listen to what they're saying.

Technical Summary: Recovery of a Measure-valued Source in the Heat Equation from Sparse Boundary Measurements

Problem Formulation
This work addresses the inverse source problem (ISP) for the heat equation on the unit disc $\Omega \subset \mathbb{R}^2$ over a time interval $(0, T)$ . The forward problem is defined by the initial-boundary value problem:
$\begin{cases} \partial_t u - \Delta u = F(t, x) & \text{in } (0, T) \times \Omega, \\ u = 0 & \text{on } (0, T) \times \partial\Omega, \\ u(0, x) = 0 & \text{on } \Omega, \end{cases}$
where the source term is assumed to have separated variables, $F(t, x) = \sigma(t)\mu(x)$ . The primary objective is to uniquely determine the spatial component $\mu$ , which belongs to the class of finite signed Radon measures ( $\mu \in (C(\overline{\Omega}))'$ ), from sparse boundary measurements. Specifically, the data consists of the normal flux $\partial_\nu u(t, x_\ell)$ observed at two distinct boundary points $x_1, x_2 \in \partial\Omega$ over the time interval $(0, T)$ .

The paper distinguishes itself from existing literature by extending the scope of recoverable sources beyond point sources (Dirac measures) or $L^2$ functions to the general class of Radon measures. This extension is motivated by applications in environmental modeling where pollution sources may be supported on lower-dimensional subsets (e.g., line sources) that cannot be adequately modeled by standard $L^2$ functions.

Methodology and Analytical Framework
The theoretical analysis relies on a combination of functional analysis, spectral theory, and complex analysis. The methodology proceeds through the following steps:

Well-posedness and Regularity: The authors define solutions in the "transposition sense" to handle the low regularity of measure-valued sources. They establish the unique existence of solutions in $L^2(Q)$ and prove that away from the support of the source, the solution possesses higher regularity ( $C^1$ in space near the boundary).
Spectral Decomposition: Using the eigen-system $\{(\lambda_n, \phi_n)\}$ of the Laplacian on the unit disc, the solution $u$ is decomposed into a Fourier series. This leads to a representation of the boundary flux $\partial_\nu u$ involving the solution $G$ of an associated elliptic problem ( $-\Delta G = \mu$ ) and a time-dependent convolution term involving $\sigma'$ .
Time Analyticity: A crucial step involves the assumption that the time-dependent coefficient $\sigma(t)$ is constant on a sub-interval $(T_1, T)$ and satisfies $\sigma(0)=0$ . Under these conditions, the authors demonstrate that the boundary flux admits an analytic extension in time. This allows the use of the unique continuation principle for analytic functions to extend the equality of fluxes from the observation interval $(0, T)$ to the entire half-line $(0, +\infty)$ .
Laplace Transform and Uniqueness: By applying the Laplace transform to the extended flux equality, the problem is reduced to an algebraic identity involving the Laplace transform of $\sigma$ and the spectral coefficients of the measure $\mu$ . Leveraging the fact that $\sigma$ is not uniformly vanishing, the authors derive a fundamental identity requiring the spectral coefficients of the difference measure to vanish.
Geometric Conditions: The uniqueness proof for the general Radon measure relies on the specific geometry of the unit disc and the explicit form of the eigenfunctions (Bessel functions). The proof requires the observation points $x_1 = e^{i\theta_1}$ and $x_2 = e^{i\theta_2}$ to satisfy the generic condition $\theta_1 - \theta_2 \notin \pi\mathbb{Q}$ . This condition ensures that the resulting linear system for the Fourier coefficients of the measure has a trivial solution, implying the measure is zero.

Key Contributions

Extension to Radon Measures: The paper provides the first positive uniqueness result for the inverse source problem with sparse boundary measurements where the source $\mu$ is a general finite signed Radon measure, rather than being restricted to point sources or $L^2$ functions.
Relaxed Time-Dependence Assumptions: Unlike previous works that required $\sigma$ to be piecewise constant, this study allows $\sigma$ to be a general function in $H^1(0, T)$ , provided it is constant on $(T_1, T)$ and vanishes at $t=0$ .
Numerical Feasibility: The authors complement the theoretical analysis with numerical experiments demonstrating the reconstruction of:
- Point sources (sums of Dirac measures) using a regularized Gauss-Newton method.
- Line sources (measures supported on curves) using the Levenberg-Marquardt method.
  The numerical results indicate that the proposed methods can recover source locations and shapes even with noisy data (up to 10% noise), though the ill-posedness increases with the number of unknowns.

Results and Claims
The main theoretical result (Theorem 1.1) states that if two measures $\mu_1$ and $\mu_2$ generate identical fluxes at two boundary points $x_1, x_2$ over $(0, T)$ , and if the time coefficient $\sigma$ satisfies the stated regularity and non-vanishing conditions, then $\mu_1 = \mu_2$ , provided the angular separation of the observation points is not a rational multiple of $\pi$ .

The authors explicitly note the limitations of their current approach:

The assumption that $\sigma$ is constant on $(T_1, T)$ is a technical requirement imposed in all known works on this specific sparse measurement formulation, and it is unclear if it can be relaxed.
The analysis relies on the explicit properties of Laplacian eigenfunctions on the unit disc, limiting the immediate application to other domains without further development.
The condition $\sigma(0)=0$ is required for the general Radon measure case, though it may be removable for more regular sources (e.g., $L^2$ ).

Significance
The paper claims significance by bridging a gap between theoretical inverse problems and practical applications involving singular sources (such as line pollution). By establishing uniqueness for Radon measures from only two boundary points, the work provides a rigorous mathematical foundation for scenarios where dense boundary data is unavailable or impractical. The numerical section further validates the theoretical findings, showing that the recovery of such singular sources is computationally feasible despite the sparsity of the measurements.

The Mystery of the Two Microphones

The Magic of the "Two-Point" Rule

From Theory to the Computer Lab

What This Means

More like this