🔢 mathematics

Sobolev estimates for the Keller-Segel system and applications to the JKO scheme

This paper establishes $L^{\infty}_{t}W^{1,p}_{x}$ Sobolev estimates for the linear diffusion Keller-Segel system using a Brezis-Gallouët-Wainger-inspired functional inequality and demonstrates their validity in the discrete JKO scheme to prove $L^2_t H^{2}_{x}$ convergence, thereby extending recent Fokker-Planck results to the Keller-Segel context.

Original authors: Charles Elbar

Published 2026-01-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Charles Elbar

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

The Big Picture: A Crowd of Cells and a Digital Simulation

Imagine a crowded room full of people (cells) who are trying to find their way to a party (chemical signals). Some people are naturally drawn to the music (positive chemotaxis), while others just wander around randomly (diffusion).

The Keller-Segel system is a set of mathematical rules that predicts how this crowd moves over time.

The Problem: If the music is too loud and the crowd is too dense, everyone rushes to the same spot at once. In math terms, the crowd "blows up" (clumps into an infinitely dense point) in a finite amount of time. This is bad because it breaks the simulation.
The Goal: The author, Charles Elbar, wants to prove that if the crowd starts out small enough (a "subcritical" case), they will never crash into a singularity. Instead, they will move smoothly forever, and we can predict their behavior with high precision.

The Tool: The JKO Scheme (The "Step-by-Step" Simulator)

To solve these complex equations, mathematicians often use a method called the JKO scheme (named after Jordan, Kinderlehrer, and Otto).

Think of the JKO scheme like a stop-motion animation or a video game with a slow frame rate.

Instead of watching the crowd move in a smooth, continuous flow, the computer takes a snapshot, calculates the best move for the next second, takes another snapshot, and repeats.
The paper proves that if you make these time steps ( $\tau$ ) smaller and smaller (making the animation smoother), the result converges to the true, real-life movement of the crowd.

The Three Big Hurdles

The author identifies three specific difficulties in making this work for the Keller-Segel system, especially in 3D space (which is harder than 2D):

The Existence Problem: In higher dimensions, the math gets messy. The "energy" of the system can go negative in ways that make it hard to prove a solution even exists. The author uses a "penalty trick" (adding a temporary rule to the game) to force the math to behave, then proves this rule isn't actually needed in the end.
The "No-Zero" Problem: To make the math work, the crowd density ( $\rho$ ) must never hit zero. If the crowd disappears completely in one spot, the equations break. The author proves that if you start with a crowd, the density stays strictly positive everywhere—it never vanishes.
The Smoothness Problem: The author needs to prove the crowd doesn't just move, but moves smoothly. They need to show that the "curvature" of the crowd's movement is controlled. This is the hardest part.

The Secret Weapon: A New "Ruler" (Functional Inequality)

To solve the Smoothness Problem, the author invents a new mathematical tool, a Functional Inequality.

The Analogy: Imagine you are trying to measure the roughness of a mountain range. Usually, you measure the height (average) and the jaggedness (peaks).
The Old Way: Standard math tools say that if you know the average height, you can guess the peaks. But this fails if the mountain has a sharp, jagged spike.
The New Way (Brezis-Gallouët-Wainger style): The author creates a special "ruler" that says: "If the mountain is mostly smooth, but has a few jagged bits, the worst peak you can have is only slightly bigger than the average, plus a tiny bit of logarithmic growth."
Why it matters: This "logarithmic" control is very gentle. It allows the author to prove that the crowd's movement stays smooth enough to prevent the "blow-up" (the singularity). It's like having a safety net that catches the crowd before they fall off a cliff.

The Main Results

Using this new ruler and the step-by-step simulator, the paper claims three main things:

Global Existence: If the initial crowd isn't too big, they will never crash into a singularity. They will exist for all time.
Smoothness: The crowd's movement is not just "okay"; it is highly smooth (mathematically, they have bounds in $L^\infty$ and Sobolev spaces). This means the density doesn't have sudden, jagged spikes.
Convergence: The "stop-motion" simulation (JKO scheme) doesn't just look like the real thing; it mathematically becomes the real thing as the time steps get smaller. Specifically, the simulation converges strongly in a high-precision space ( $L^2_t H^2_x$ ), meaning the shape of the crowd in the simulation matches the real crowd perfectly in terms of position and curvature.

Summary in One Sentence

The author proves that a mathematical model of cells moving toward a signal will never collapse into a singularity if started small enough, and that a specific step-by-step computer simulation of this process is not just an approximation, but a mathematically rigorous way to find the exact, smooth solution.

Technical Summary: Sobolev estimates for the Keller-Segel system and applications to the JKO scheme

Problem Statement
The paper addresses the analytical properties of the Keller-Segel (KS) system with linear diffusion in a smooth, convex, bounded domain $\Omega \subset \mathbb{R}^d$ . The system is given by:
$\partial_t\rho - \Delta\rho + \text{div}(\rho\nabla u[\rho]) = 0, \quad -\Delta u[\rho] = \rho$
subject to no-flux and Dirichlet boundary conditions. The primary challenges in this context are:

Finite-time blow-up: Solutions can blow up if the initial mass exceeds a critical threshold (e.g., $8\pi$ in 2D). The paper focuses on a "small-data subcritical regime" where the initial $L^{d/2}$ norm is sufficiently small to prevent blow-up.
Existence and Regularity: While weak solutions exist, obtaining $L^\infty$ bounds and higher-order Sobolev estimates (specifically $L^\infty_t W^{1,p}_x$ for all $1 \le p < \infty$ ) in dimensions $d \ge 3$ is non-trivial due to the lack of control over the negative part of the free energy and the singularity of the interaction kernel.
Convergence of the JKO Scheme: The Jordan-Kinderlehrer-Otto (JKO) scheme, a variational time-discretization of gradient flows in the Wasserstein space, has been established for the KS system. However, extending the strong convergence results (specifically in $L^2_t H^2_x$ ) known for the Fokker-Planck equation to the KS system requires overcoming specific difficulties related to the time-dependent, non-smooth potential $u[\rho]$ .

Methodology
The author employs a combination of functional analysis, optimal transport theory, and discrete gradient flow techniques.

Functional Inequality (Brezis-Gallouët-Wainger type):
A central tool is a new functional inequality (Theorem 1.1) inspired by the Brezis-Gallouët-Wainger inequality. It provides an $L^\infty$ bound on the Hessian $D^2 u$ in terms of the $L^\infty$ and $C^{0,\alpha}$ norms of the Laplacian $\Delta u$ :
$\|D^2 u\|_{L^\infty} \le C \| \Delta u \|_{L^\infty} \left( 1 + \log \left( \frac{\|\Delta u\|_{C^{0,\alpha}}}{\|\Delta u\|_{L^\infty}} \right) \right)$
This logarithmic control is crucial for closing estimates in the presence of the nonlinearity.
Dissipation of the Functional $F_p$ :
The paper analyzes the dissipation of the functional $F_p[\rho] = \int_\Omega \rho |\nabla(\log \rho - u[\rho])|^p dx$ . By computing the time derivative (in the continuous case) or the discrete difference (in the JKO scheme), the author derives a differential inequality of the form:
$\frac{d}{dt} F_p \lesssim F_p (\log(F_p + 1) + 1)$
The logarithmic growth on the right-hand side, derived via the functional inequality and the Calderón-Zygmund estimates, allows for global-in-time bounds via Gronwall's lemma, preventing the blow-up of Sobolev norms.
The Penalized JKO Scheme:
To handle existence in high dimensions ( $d \ge 3$ ) where the standard free energy is unbounded from below, the author introduces a penalized functional $\tilde{J}$ . This includes an artificial term controlling the $L^{d/2}$ norm. The paper proves that for sufficiently small initial data, this penalization term remains inactive, and the scheme coincides with the standard JKO scheme.
Discrete Estimates and Five-Gradients Inequality:
To prove convergence, the author adapts the "five-gradients inequality" (originally for BV estimates in optimal transport) to the discrete JKO setting. This allows for the control of the dissipation of the Fisher information functional $F_2$ at the discrete level. Key steps include:
- Establishing a uniform $L^\infty$ bound on the discrete solutions using an Alikakos iteration method adapted to the JKO scheme.
- Proving a strict lower bound (positivity) for the discrete solutions, which is essential for controlling the Kantorovich potentials and their Hessians.
- Showing that the Kantorovich potentials converge to zero with a specific rate, bridging the gap between discrete dissipation and continuous limits.

Key Contributions and Results

Theorem 1.1 (Functional Inequality): Proves a new inequality bounding $\|D^2 u\|_{L^\infty}$ using logarithmic interpolation between $L^\infty$ and $C^{0,\alpha}$ norms of $\Delta u$ . This is a refinement of classical Sobolev embeddings suitable for the specific structure of the KS system.
Theorem 1.5 (Sobolev Estimates): Establishes that for small initial data ( $L^{d/2}$ norm), the weak solution to the KS system exists globally and is bounded in $L^\infty((0,T) \times \Omega) \cap L^\infty(0, T; W^{1,p}(\Omega))$ for all $1 \le p < \infty$ . The proof relies on the dissipation of $F_p$ and the derived logarithmic differential inequality.
Theorem 1.7 (JKO Scheme Convergence):
- Existence: Proves the existence of the JKO scheme sequence in high dimensions using the penalized functional.
- Regularity: Shows the discrete curve $\rho_\tau$ is uniformly bounded in $L^\infty((0,T) \times \Omega)$ and $L^\infty(0, T; W^{1,p}(\Omega))$ .
- Strong Convergence: Proves that the discrete curve $\rho_\tau$ converges strongly in $L^2(0, T; H^2(\Omega))$ to the weak solution of the KS system as the time step $\tau \to 0$ .

Significance and Claims
The paper claims to extend recent results by Santambrogio and Toshpulatov (who proved $L^2_t H^2_x$ convergence for the Fokker-Planck equation) to the aggregation-diffusion context of the Keller-Segel system. The author highlights three specific difficulties overcome:

Existence in High Dimensions: Overcoming the lack of free energy control in $d \ge 3$ via the penalization trick.
Lower Bounds: Establishing a uniform lower bound for the discrete density $\rho_\tau$ without assuming a priori bounds on $D^2 u$ , which is non-trivial since the potential $u$ depends on $\rho$ .
Regularity of Potentials: Proving uniform $C^{0,\alpha}$ estimates for the JKO curves to control the convergence of the Kantorovich potentials, a step that previously relied on $L^\infty_t W^{1,p}_x$ estimates which were only known for $p < 2$ in the KS context.

The author notes that while the $L^\infty$ bound for the continuous PDE was considered an open problem in some literature (e.g., [23]), this work provides a constructive proof via the JKO scheme. The results are presented as a rigorous justification of the JKO scheme's convergence in high Sobolev norms for the Keller-Segel system, relying on the interplay between the functional inequality and the gradient flow structure.