🔢 mathematics

Interfaces and non-uniqueness in a cross-diffusion system with independent drifts

This paper demonstrates that a one-dimensional cross-diffusion system with independent drifts and segregated initial data admits multiple distinct solutions (including overlapping and segregated weak solutions), thereby proving the Cauchy problem is not well-posed in the class of weak solutions.

Original authors: Charles Elbar, Guy Parker

Published 2026-06-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Charles Elbar, Guy Parker

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 two crowds of people moving through a long hallway. Let's call them the "Red Team" and the "Blue Team."

In this mathematical story, the rules of the hallway are a bit unusual:

They push each other away: If a Red person and a Blue person get too close, they naturally want to spread out to avoid crowding. This is called "cross-diffusion."
They have different wind currents: The hallway has invisible winds (called "drifts"). The wind blowing on the Red Team is different from the wind blowing on the Blue Team.

The Setup: A Perfect Wall

At the start, the two teams are perfectly separated. The Red Team is on the left side of a specific door (the interface), and the Blue Team is on the right. They are not mixing at all.

The researchers asked a simple question: If the winds on both sides start pushing the teams toward that door, what happens?

Intuitively, you might think the teams will crash into each other and start mixing. Or, you might think they will just press against the door and stay separate, like two magnets pushing against a wall.

The Big Discovery: The Answer Depends on How You Look

The paper's main finding is shocking: The answer depends entirely on which "rulebook" you use to describe the movement.

The authors found that for the exact same starting situation, there are two completely different, mathematically valid outcomes:

1. The "Perfectly Segregated" Solution (The Wall Stays)

Imagine the teams are so stubborn that they refuse to mix. Even though the winds are pushing them together, they maintain a perfect, sharp wall between them. The Red Team stays on the left, the Blue Team stays on the right, and they just press harder against the invisible door.

The Math: This is a "weak solution." It's a valid answer if you define the rules loosely enough to allow for a sharp, unyielding boundary.

2. The "Vanishing Viscosity" Solution (The Wall Crumbles)

Now, imagine we add a tiny bit of "stickiness" or "friction" to the air (mathematicians call this "viscosity"). In the real world, nothing is perfectly sharp; there's always a tiny bit of fuzziness at the edge.

If you calculate the movement with this tiny bit of stickiness and then slowly remove the stickiness (letting it go to zero), the result is different.
The Outcome: The teams do mix. The sharp wall breaks down, and the Red and Blue people start overlapping in the middle. The "wind" pushing them together wins, and they blend into a purple zone.

The Analogy: The Traffic Jam

Think of it like two lanes of traffic merging into one.

Scenario A (Segregated): The cars in the left lane and the right lane are so disciplined that they never cross the yellow line, even if the lane merges. They just squeeze together in perfect order.
Scenario B (Mixing): If you look at how real cars actually behave (accounting for tiny adjustments and "friction" in driving), the cars naturally drift across the line and mix into a single stream.

The paper proves that both scenarios are mathematically correct for the same starting point. This means the "equation" describing the movement doesn't have a single, unique answer. It's like asking, "Where will the traffic be in 5 minutes?" and getting two different, equally valid maps depending on whether you assume drivers are robots or humans.

Why Does This Matter?

In the world of physics and math, we usually expect a system to have one unique future. If you know the starting position and the forces, you should be able to predict exactly what happens next.

This paper shows that for this specific type of crowd movement (cross-diffusion with different winds), the future is not unique.

If you choose the "robot" rulebook, the teams stay separate.
If you choose the "realistic/physical" rulebook (vanishing viscosity), they mix.

The authors used a clever tool called "Relative Entropy" (think of it as a "Mixing Meter") to prove this.

If the teams are perfectly separate, the meter reads 0.
If they mix, the meter reads positive.
They showed that the "physical" rulebook forces the meter to go up (mixing), while the "segregated" rulebook keeps it at zero. Since the meter can't be both 0 and positive at the same time, the two solutions must be different.

The Bottom Line

The paper demonstrates that when two groups are pushed together by different forces, the math allows for two realities: one where they stay perfectly apart, and one where they inevitably mix. Because the math allows for both, we cannot predict the outcome with certainty without deciding which version of reality (which definition of a "solution") we want to use. This breaks the standard expectation that nature has only one predictable path.

Technical Summary: Interfaces and Non-Uniqueness in a Cross-Diffusion System with Independent Drifts

Problem Statement
The paper investigates a one-dimensional cross-diffusion system describing the evolution of two population densities, $\rho(t, x)$ and $\mu(t, x)$ , subject to a common pressure law dependent on the total density $S = \rho + \mu$ and distinct external drift potentials $V$ and $W$ . The system is governed by:
$\begin{aligned} \partial_t\rho - \partial_x(\rho \partial_x f'(\rho + \mu)) - \partial_x(\rho \partial_x V) &= 0, \\ \partial_t\mu - \partial_x(\mu \partial_x f'(\rho + \mu)) - \partial_x(\mu \partial_x W) &= 0. \end{aligned}$
The central question addressed is the well-posedness of the Cauchy problem when the initial data is segregated (i.e., $\rho_0$ and $\mu_0$ have disjoint supports meeting at a single interface) and the drifts are distinct. Specifically, the authors examine the regime where the drifts push the two phases toward each other at the interface ( $\partial_x W(0) - \partial_x V(0) > 0$ ). While previous literature established that segregated solutions exist and remain segregated under certain conditions, this work questions whether such solutions are unique when compared to solutions obtained via the vanishing-viscosity limit.

Methodology
The authors employ a relative entropy approach to distinguish between different notions of solutions. The core of their methodology involves:

Relative Entropy Functional: They define a functional $\Phi[\rho, \mu] = \int_{\mathbb{R}} h(\rho, \mu) dx$ , where $h(u, v) = (u+v)\log(u+v) - u\log u - v\log v$ . This functional is non-negative and vanishes if and only if the densities are segregated ( $\rho\mu = 0$ ).
Viscous Approximation: They analyze the system with added viscosity terms ( $\varepsilon \partial_{xx}$ ) and study the limit as $\varepsilon \to 0$ . They derive an entropy dissipation identity at the viscous level and pass to the limit to obtain a lower bound for the relative entropy of the vanishing-viscosity solution.
Distributional Analysis at the Interface: A critical technical step involves characterizing the distributional derivative of the quantity $q = \partial_x \rho - \frac{\rho}{S}\partial_x S$ at the interface. For a segregated solution with a sharp interface, this term behaves like a Dirac mass.
Contradiction via Mixing: The authors prove that if the drifts push the phases together, the lower bound on the relative entropy for a vanishing-viscosity solution is strictly positive, implying mixing. Conversely, a segregated solution maintains $\Phi \equiv 0$ . This contradiction establishes non-uniqueness.
Handling General Data: For general initial data, the authors must ensure the interface does not vanish (creating a vacuum) or jump away from the region where drifts induce mixing. They utilize:
- Regularity results for segregated solutions from prior work ([56]).
- Comparison principles and Harnack inequalities for decoupled drift-diffusion equations to show that if a vacuum were to form at the interface, the species would instantaneously mix, contradicting the segregation assumption.
- Construction of pressure sub-solutions (inspired by travelling waves) for the slow-diffusion regime ( $\alpha > 1$ ) to prove interface persistence.

Key Contributions and Results

Explicit Non-Uniqueness for Stationary States:
The authors construct an explicit stationary segregated solution $(\bar{\rho}, \bar{\mu})$ where the pressure gradient balances the drifts on each side of the interface. They prove that this solution is a valid weak solution but cannot be a vanishing-viscosity solution. Under the condition that drifts push phases together, the vanishing-viscosity limit necessarily produces an overlap (mixing), whereas the explicit stationary solution remains segregated.
Non-Uniqueness for General Initial Data:
Extending the result beyond stationary states, the paper proves that for a broad class of pressure laws (including power laws $f'(s) \sim s^{\alpha-1}$ with $\alpha > 1/3$ ) and general segregated initial data, the Cauchy problem admits at least two distinct weak solutions:
- A segregated solution (constructed via minimizing movement schemes as in [56]).
- A vanishing-viscosity solution (constructed via compactness arguments as in [58]).
  These solutions are distinct because the vanishing-viscosity solution develops a mixing zone (positive relative entropy), while the segregated solution does not.
Entropy Dissipation Inequality:
A central technical result is the derivation of the inequality:
$\Phi[\rho(t), \mu(t)] \geq \int_0^t \langle q(s), \partial_x V - \partial_x W \rangle ds$
for vanishing-viscosity solutions. This inequality quantifies the mixing induced by the drift difference and serves as the primary tool for distinguishing solution classes.

Significance and Claims
The paper claims to provide the first demonstration of non-uniqueness for this specific class of cross-diffusion systems with independent drifts. The authors highlight that:

Physical vs. Mathematical Solutions: The phenomenon illustrates that the "physical" solution selected by vanishing viscosity (which allows mixing) differs from the mathematically valid segregated solution. This challenges the well-posedness of the system within the class of weak solutions.
Analogy to Muskat Problem: The behavior bears a striking resemblance to the Muskat problem, where an unstable interface between fluids of different densities spontaneously yields to a mixing zone.
Limitations of Existing Theory: The work underscores that standard existence theories for cross-diffusion systems (which often rely on compactness of smooth approximations) may select solutions that do not preserve segregation, even when segregated solutions exist.
Scope: The authors explicitly state that these results are specific to cross-diffusion systems where species do not exhibit self-diffusion. They note that systems with self-diffusive terms possess higher regularity that prevents sharp interfaces, and thus, this non-uniqueness mechanism is not expected to apply to single-species advection-diffusion models or self-diffusive cross-diffusion systems.

The paper concludes that for cross-diffusion systems with independent drifts pushing phases together, the Cauchy problem is ill-posed in the class of weak solutions without a more restrictive selection criterion (such as the vanishing-viscosity limit) to distinguish between the segregated and mixed states.