🔢 mathematics

Recovery of coefficients for a convection-diffusion equation from partial data

This paper establishes the unique determination of time- and space-dependent convection and potential coefficients in a parabolic convection-diffusion equation from partial boundary measurements by developing special solutions and new nonlinear-weight Carleman estimates to extend elliptic inverse problem techniques to the parabolic setting.

Original authors: Liam Buisson

Published 2026-07-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Liam Buisson

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, invisible room. You can't see inside, and you can't walk through the walls. The only way to learn about what's happening in the room is to shout at the walls (sending in a signal) and listen to how the walls echo back (measuring the response). In the world of physics and mathematics, this is called an "inverse problem." Instead of predicting what happens when you know the rules, you are trying to figure out the hidden rules just by watching the results.

The specific rules we are looking at here describe how things move and spread out, like heat warming up a metal plate, smoke drifting through a room, or a virus spreading through a crowd. This movement is governed by a set of equations known as "convection-diffusion" equations. Think of "diffusion" as the natural tendency of a drop of ink to spread out in water until it's evenly mixed. Think of "convection" as a current in the water pushing that ink in a specific direction. In real life, these currents (wind, blood flow, ocean tides) and the rate of spreading often change depending on where you are and when you look.

The big challenge is that in the real world, we rarely have a perfect view of the whole room. We might only be able to shout at one side of the wall and listen to the echo from a small patch on the opposite side. This is called "partial data." For a long time, mathematicians could solve these puzzles if the room was perfectly still (like a static building), but they struggled when the rules changed over time, like a stormy day where the wind shifts every minute. This paper tackles that exact difficulty: figuring out the hidden, time-changing currents and spreading rates inside a room, using only a limited view of the echoes.

The Detective's New Tool

In this paper, the author, Liam Buisson, acts as a mathematical detective who has developed a new, super-sharp tool to solve this specific mystery. The goal is to uniquely identify two hidden ingredients in the equation: the "convection" term (the invisible wind or current pushing things around) and the "potential" term (a local force that might be creating or destroying the substance, like a heater turning on or a chemical eating away at the material).

Usually, when you try to find these hidden ingredients, you run into a "gauge invariance" problem. Imagine you are trying to describe the wind in a room, but you can also change your definition of "zero wind" by shifting your perspective. If you shift your perspective just right, the wind looks different, but the actual echoes off the wall stay exactly the same. This means that without extra information, you can't tell the difference between the real wind and a "fake" wind created by a shift in perspective. The paper proves that while you can't always find the exact wind and potential, you can find their "gauge class"—essentially, you can find the wind up to that specific, harmless shift in perspective.

How the Magic Trick Works

To crack the case, Buisson uses a clever trick involving "geometric optics solutions." Imagine shining a very bright, focused beam of light into the dark room. In this mathematical world, these beams are not just light; they are specially crafted waves that travel along specific paths, bouncing off the invisible currents. The author constructs these waves to be extremely sensitive to the hidden coefficients.

The real breakthrough here is the use of a new type of mathematical "flashlight" called a Carleman estimate with nonlinear weights. In simpler terms, previous methods used flashlights with a steady, linear beam that got brighter in a straight line. But the hidden currents in this problem are tricky and change in complex, curved ways. The author invented a new kind of flashlight where the beam's intensity bends and curves (nonlinear) to perfectly match the shape of the hidden currents. This allows the math to "see" the hidden details that the old, straight-beam flashlights missed.

The Findings

Using these curved beams and the new flashlight, the paper proves two main things:

Uniqueness up to a Shift: If you measure the echoes from a specific "illuminated" part of the boundary (the side of the room lit up by a point source outside), you can uniquely determine the hidden wind and potential, except for that specific gauge shift mentioned earlier. If two different sets of hidden rules produce the exact same echoes, they must be related by that specific mathematical shift.
Breaking the Lock: The paper also shows that if you already know one of the ingredients (for example, if you know the "potential" or the local creation/destruction rate), you can then uniquely determine the other one (the wind/current) with no ambiguity. Similarly, if the wind field has a specific property (like being "divergence-free," meaning it doesn't create or destroy matter, just moves it), you can uniquely find the wind.

Why It Matters

This work is a significant step forward because it extends a powerful method that was previously only used for static, unchanging rooms to dynamic, time-changing environments. It proves that even with limited data—only listening to a small part of the wall—you can still reconstruct the complex, time-varying forces driving physical processes like heat transfer, pollutant dispersion, or even financial models. The author doesn't just say it's possible; they provide the rigorous mathematical proof and the specific construction of the tools needed to do it, opening the door for more accurate models in science and engineering where time and space are constantly shifting.

Technical Summary: Recovery of Coefficients for a Convection-Diffusion Equation from Partial Data

Problem Statement
The paper addresses the inverse problem of uniquely determining the time-dependent zeroth-order coefficient (potential, $q$ ) and the first-order coefficient (convection field, $A$ ) in a parabolic convection-diffusion equation from partial boundary measurements. The governing equation is:
$\partial_t u - \Delta u + A(t, x) \cdot \nabla u + q(t, x)u = 0$
defined on a space-time cylinder $Q = (0, T) \times \Omega$ , where $\Omega \subset \mathbb{R}^n$ ( $n \geq 2$ ) is a bounded domain with $C^2$ boundary. The data available is the partial Dirichlet-to-Neumann (DN) map, $\Lambda(A, q)$ , which maps Dirichlet boundary data $g$ to the Neumann flux $\partial_\nu u$ restricted to a subset $U \subset \Sigma = (0, T) \times \partial\Omega$ . Specifically, $U$ is a neighborhood of the "illuminated" portion of the boundary, $\Gamma_-(x_0)$ , as viewed from a point $x_0$ located outside the closure of $\Omega$ .

A fundamental obstruction to uniqueness is identified: the problem possesses a gauge invariance. The DN map remains unchanged under the transformation $(A, q) \mapsto (A + 2\nabla \phi, q - (\partial_t \phi - \Delta \phi + |\nabla \phi|^2 + A \cdot \nabla \phi))$ for any function $\phi$ vanishing with its normal derivative on the boundary. Consequently, the primary goal is to recover the gauge class of $(A, q)$ , and under additional assumptions, the coefficients themselves.

Methodology
The authors develop a framework that extends partial data results from elliptic equations to the parabolic setting with time-dependent coefficients. The methodology relies on three main pillars:

Geometric Optics (GO) Solutions: The construction of special solutions to the parabolic equation and its adjoint in the form of geometric optics waves:
$u_\tau(t, x) = e^{\theta_\tau(t, x)} [c_+(t, x) + R_{+, \tau}(t, x)]$
$v_\tau(t, x) = e^{-\theta_\tau(t, x)} [c_-(t, x) + R_{-, \tau}(t, x)]$
where $\tau$ is a large parameter. Unlike previous works on parabolic equations that utilized linear weights, this paper employs nonlinear weights in the phase function $\theta_\tau(t, x) = \tau^2 t + \tau \psi(x)$ , where $\psi(x) = |x - y|$ for a point $y$ outside $\Omega$ . This choice allows the weight to satisfy the eikonal equation $|\nabla \psi|^2 = 1$ and facilitates the use of partial data.
Nonlinear Carleman Estimates: A significant technical hurdle is that the nonlinear phase of the GO solutions requires Carleman estimates with nonlinear weights. The authors derive a new family of Carleman estimates for parabolic operators with variable first-order coefficients. These estimates are established for the weight $\theta_\tau$ and its negative, serving as a parabolic analogue to the "limiting Carleman weights" known in the elliptic setting. These estimates control the remainder terms $R_{\pm, \tau}$ , ensuring they decay sufficiently as $\tau \to \infty$ .
Integral Identities and X-ray Transform: By inserting the GO solutions into an integral identity derived from the difference of two DN maps, the authors isolate terms involving the unknown coefficients. Taking the limit $\tau \to \infty$ eliminates the remainder terms, leaving integral identities involving the principal parts $c_\pm$ . These identities are shown to imply that the X-ray transform of the difference of the coefficients (specifically the solenoidal part of $A$ and the potential $q$ ) vanishes on lines intersecting the illuminated region. The uniqueness of the X-ray transform for compactly supported distributions is then invoked to recover the coefficients.

Key Results

Theorem 2.1 (Gauge Class Uniqueness): If two sets of coefficients $(A_1, q_1)$ and $(A_2, q_2)$ generate the same partial DN map and agree on the lateral boundary $\Sigma$ , then they belong to the same gauge class. That is, there exists a gauge function $\phi$ such that $(A_2, q_2) = S_\phi(A_1, q_1)$ . This result holds for time-dependent coefficients and measurements restricted to the illuminated boundary, extending previous elliptic results to the parabolic case.
Theorem 2.2 (Recovery of Potential): If the convection field $A$ is known and fixed, the potential $q$ is uniquely determined by the partial DN map. This is proven by showing the X-ray transform of the difference $q_1 - q_2$ vanishes.
Corollary 2.3 (Recovery of Convection Field): If the potential $q$ is known, the convection field $A$ is uniquely determined (up to the boundary condition $A|_\Sigma$ ).
Corollary 2.4 (Fokker-Planck Equation): For the Fokker-Planck equation (where the zeroth-order term is the divergence of the convection field), the vector field $A$ is uniquely determined from the partial DN map, provided $A$ is sufficiently smooth.

Significance and Claims
The paper claims to provide the first positive answer to the simultaneous recovery of time-dependent convection and potential coefficients in a parabolic equation using partial boundary data restricted to the illuminated portion of the boundary, a configuration previously studied only for elliptic equations (e.g., in [San+07; KSU07]).

The authors highlight that their approach overcomes the difficulty of time-dependence, which prevents the reduction of the problem to elliptic or spectral problems used in earlier literature. A major contribution is the development of Carleman estimates with nonlinear weights for parabolic equations. The authors note that while Carleman estimates with linear weights have been used for parabolic inverse problems (e.g., [CK18b]), the extension to nonlinear weights is novel. They assert that this new class of estimates is of independent interest and may enable further applications, such as the simultaneous restriction of both the observation set and the excitation set.

The work also improves upon existing literature by covering the two-dimensional case (which was not addressed in [KLL24]) and by relaxing regularity assumptions on the coefficients through a novel approximation technique involving extensions of the coefficients to the whole space. The results are presented as a rigorous extension of the partial data framework from elliptic to parabolic settings, establishing uniqueness up to a natural gauge transformation.

More like this