🔢 mathematics

Existence of weak solutions for nonlinear drift-diffusion equations with measure data

This paper establishes the existence of nonnegative weak solutions with gradient estimates for nonlinear drift-diffusion equations with measure data under specific drift integrability conditions, demonstrating that divergence-free drifts allow for relaxed requirements and enlarged diffusion exponents, while also providing counterexamples to prove the sharpness of these results and applying them to equations coupled with incompressible Navier-Stokes systems.

Original authors: Sukjung Hwang, Kyungkeun Kang, Hwa Kil Kim, Jung-Tae Park

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sukjung Hwang, Kyungkeun Kang, Hwa Kil Kim, Jung-Tae Park

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 trying to predict how a crowd of people (or a cloud of gas, or a drop of ink in water) moves through a room over time. This movement is governed by two main forces:

Diffusion (The "Spreading"): People naturally spread out from crowded areas to empty ones. In the paper, this is the "nonlinear diffusion" part. If the crowd is very dense, they spread differently than if they are sparse.
Drift (The "Push"): There is a wind blowing through the room, or a crowd leader shouting directions, pushing people in a specific direction. This is the "drift" term.

The paper tackles a very messy version of this problem. Usually, mathematicians like to start with a smooth, well-behaved crowd. But here, the authors are dealing with "Measure Data."

The "Measure Data" Problem: The Instant Crowd

Imagine that at the very start ( $t=0$ ), instead of a smooth crowd, you have a single, infinitely dense point of people appearing out of nowhere (like a Dirac delta function, or a "point source"). Or perhaps people are being injected into the room in sudden, unpredictable bursts.

This is the "Measure Data." It's mathematically ugly because the density is infinite at a point. Standard tools for predicting movement break down when you have these infinite spikes.

The Two Main Challenges

The paper asks: Can we still predict the movement if we have these messy "infinite spikes" AND a "wind" (drift) pushing them around?

The authors found that the answer depends heavily on the shape of the wind and the nature of the crowd.

1. The "General Wind" (The Hard Way)

If the wind is just blowing randomly (it has no special structure), the math gets very difficult.

The Analogy: Imagine a chaotic wind that swirls and pushes people in ways that fight against the natural spreading.
The Result: To prove a solution exists, the "crowd" (the diffusion exponent $m$ ) must be "thick" enough. If the crowd is too "thin" (fast diffusion), the chaotic wind can completely destroy the mathematical structure needed to make a prediction.
The Limit: The authors proved that for a general wind, the crowd must be thick enough ( $m > 1 - 1/d$ ). If it's thinner than that, the wind can cancel out the spreading so effectively that the math breaks down. They even built a "counter-example" (a specific scenario) to show that if you try to go below this limit, the prediction fails.

2. The "Organized Wind" (The Easy Way)

Now, imagine the wind is divergence-free.

The Analogy: Think of a river flowing. Water flows in, but it doesn't pile up or disappear; it just moves. The amount of wind entering any small area equals the amount leaving. It's a "conservative" flow.
The Magic: Because this wind is organized, it has a "cancellation property." When you do the math, the messy parts of the wind cancel each other out.
The Result: Because of this cancellation, the authors could relax the rules. They could handle much "thinner" crowds (down to $m > 1 - 2/d$ ) and still get a valid prediction. The organized wind doesn't destroy the math like the chaotic wind does.

The "Energy" of the System

The authors developed a new way to measure the "energy" of this system.

In normal physics, energy is conserved. Here, because of the "infinite spikes" (measure data), standard energy formulas don't work.
They created a "Truncated Energy Estimate." Think of this as putting a "cap" on the height of the crowd. Instead of trying to measure the infinite spike directly, they measure the crowd up to a certain height, prove the math works there, and then show that as they remove the cap, the solution remains stable.
This new estimate is the key tool that allowed them to prove solutions exist for both the "General Wind" and the "Organized Wind" scenarios.

The Real-World Application: The "Fluid-Crowd" System

Finally, the paper applies this theory to a specific, complex system: The Keller-Segel-Fluid system.

The Setup: Imagine bacteria (the crowd) swimming in a fluid (like water). The bacteria diffuse, but they are also pushed by the current of the water. The water itself moves according to the Navier-Stokes equations (the laws of fluid dynamics), and the bacteria push back on the water.
The Connection: In this system, the fluid velocity (the "wind") is divergence-free (water is incompressible; it doesn't pile up).
The Outcome: Because the fluid is "organized" (divergence-free), the authors could use their "Organized Wind" theorem. They successfully proved that even if you start with a messy, infinite spike of bacteria, a valid solution exists where the bacteria and the fluid move together in a predictable way.

Summary

The Problem: Predicting movement of a substance with "infinite spikes" when pushed by a wind.
The Discovery:
- If the wind is chaotic, the substance must be "thick" enough to resist being destroyed by the wind.
- If the wind is organized (divergence-free), the substance can be "thinner," because the wind's chaos cancels itself out.
The Proof: They built a new mathematical "ruler" (energy estimate) to measure these messy systems and proved that solutions exist, provided the wind and the substance meet specific criteria.
The Application: They used this to solve a problem involving bacteria swimming in fluid, showing that even with messy starting conditions, the system behaves predictably.

Technical Summary: Existence of Weak Solutions for Nonlinear Drift-Diffusion Equations with Measure Data

Problem Statement
This paper investigates the existence of nonnegative weak solutions to nonlinear drift-diffusion equations of the form:
$u_t - \Delta u^m + \nabla \cdot (uV) = \mu$
defined on a bounded domain $\Omega_T = \Omega \times (0, T)$ with $d \ge 2$ . The equation encompasses both porous medium equations ( $m > 1$ ) and fast diffusion equations ( $0 < m < 1$ ). The forcing term $\mu$ is a nonnegative finite Radon measure, representing low-regularity data, and $V$ is a drift vector field. The primary challenge lies in the simultaneous presence of measure-valued data and a drift term, which alters the compactness mechanisms essential for existence proofs compared to cases where either the drift or the measure data is absent.

Methodology
The authors employ a direct compactness argument applied to a sequence of regularized problems. The methodology involves:

Regularization: Approximating the measure data $\mu$ and the drift $V$ with smooth functions to construct regular solutions.
A Priori Estimates: Deriving energy-type estimates using truncated test functions. A critical distinction is made between the general drift case and the divergence-free case ( $\nabla \cdot V = 0$ $\nabla \cdot V = 0$ ).
- In the general drift case, the drift term $uV$ introduces additional terms in local truncation estimates that cannot be controlled solely by measure-data compactness methods. The authors derive new energy estimates involving the product $uV$.
- In the divergence-free case, the drift term possesses a cancellation structure in the local truncation estimates. This allows the troublesome drift contribution to be reduced to a lower-order cutoff term, enabling the adaptation of compactness methods from previous works (specifically [2, 4]) to the current setting.
Compactness: Utilizing the Aubin–Lions lemma and parabolic embedding theorems to establish strong convergence of the approximate solutions. The choice of the exponent $\alpha$ for gradient estimates is crucial for ensuring sufficient Sobolev compactness.
Sharpness Analysis: The paper constructs specific counterexamples in dimension $d=2$ to demonstrate the necessity of the derived conditions. These examples show that without the divergence-free condition, the gradient estimates fail even when the drift satisfies the same integrability scales as in the divergence-free theory.

Key Contributions and Results

Existence for General Drifts: For general (non-divergence-free) drifts, the authors establish the existence of weak solutions provided the drift $V$ belongs to a sub-scaling class relevant to the $L^1$ space. This requires the diffusion exponent to satisfy $m > 1 - \frac{1}{d}$ .
- Theorem 1.4: Establishes existence for $1 - \frac{1}{d} < m \le 2$ under specific integrability conditions on $V$ (classes $S^{(q_1, q_2)}_m$ ). The upper bound $m \le 2$ is linked to the singularity of the factor $u^{1-\frac{m}{2}}$ near the zero set when $m > 2$ , which obstructs the direct compactness argument.
- Gradient Estimates: The solutions satisfy gradient estimates of the form $\int_{\Omega_T} |\nabla u^{m/2}|^\alpha dxdt < \infty$ for $\alpha \in (0, 2)$ .
Existence for Divergence-Free Drifts: When $\nabla \cdot V = 0$ , the requirements on both the drift and the diffusion exponent are relaxed.
- Theorem 1.8: The lower bound on $m$ is improved to the intrinsic measure-data threshold $m > (1 - \frac{2}{d})_+$ . The drift $V$ can belong to a class that is supercritical with respect to the $L^1$ -scaling (classes $S^{(q_1, q_2)}_{m, \sigma}$ ).
- The divergence-free condition restores the optimal measure-data range by preventing the drift from destroying the gradient estimates via cancellation.
Sharpness of Conditions:
- Lower Threshold: Section 4.1 constructs a counterexample showing that for $m \le 1 - \frac{1}{d}$ , the gradient estimate fails for general drifts even if they satisfy the integrability conditions required for the divergence-free case. This confirms that the threshold $m > 1 - \frac{1}{d}$ is sharp for non-divergence-free drifts.
- Upper Threshold: Section 4.2 addresses the range $m > 2$ . It demonstrates that without the divergence-free structure, a general drift can cancel the diffusion term entirely, destroying the gradient estimate for $u^{m/2}$ even if the drift is integrable. This explains the natural appearance of the $m \le 2$ restriction in the general drift case.
Application to Fluid Coupling: As an application, the authors construct weak solutions for a simplified Keller–Segel–fluid system where the density equation (with measure data and nonlinear diffusion) is coupled to the incompressible Navier–Stokes equations. Since the fluid velocity is divergence-free, the divergence-free existence theorem (Theorem 1.8) applies directly, allowing for the construction of solutions satisfying a global energy inequality.

Significance
The paper claims that the simultaneous presence of measure data and drift fundamentally changes the compactness mechanism. The primary contribution is the identification of the divergence-free condition as the precise structural assumption that allows the problem to bypass the stricter lower bound ( $m > 1 - \frac{1}{d}$ ) required for general drifts, thereby recovering the optimal range ( $m > (1 - \frac{2}{d})_+$ ) known for drift-free measure data problems. Furthermore, the work clarifies the limitations of general drifts in highly degenerate regimes ( $m > 2$ ), showing that the cancellation structure is essential not just for the lower bound but also for preserving gradient estimates in the upper range. The results are presented as sharp, with counterexamples demonstrating that the integrability scales cannot be improved without the divergence-free cancellation.