Li-Yau-Hamilton Inequality on the JKO Scheme for the Granular-Medium Equation

Basic Information

• Paper ID: 2510.09231
• Title: Li-Yau-Hamilton Inequality on the JKO Scheme for the Granular-Medium Equation
• Author: Fanch Coudreuse
• Classification: math.AP (Analysis of PDEs), math.OC (Optimization and Control)
• Publication Date: October 10, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.09231

Abstract

This paper establishes a version of the Li-Yau-Hamilton inequality for the granular-medium equation on the torus, covering both the PDE level and its time-discrete approximation given by the JKO scheme. Subsequently, this estimate is applied to derive further quantitative results for both continuous and discrete JKO flows, including Lipschitz and $L^\infty$ bounds, as well as quantitative Harnack inequalities. Finally, using the regularity provided by this estimate, the convergence of the JKO scheme for the Fokker-Planck equation in $L^2_{\mathrm{loc}}((0,+\infty); H^2(\mathbb{T}^d))$ is established.

Research Background and Motivation

Problem Background

Li-Yau type inequalities are fundamental estimates in heat equation theory and more broadly in the study of diffusion equations. Originally proved by P. Li and S. Yau in 1986, they demonstrate that on Riemannian manifolds with non-negative Ricci curvature, positive solutions to the heat equation satisfy the pointwise bound $\Delta \log \rho_t \geq -\frac{d}{2t}$. Later, Hamilton refined this inequality to a complete Hessian estimate under stricter geometric assumptions: $D^2 \log \rho_t \succeq -\frac{1}{2t}I$.

Research Motivation

1. Perfecting JKO Scheme Theory: Since the pioneering work of Jordan, Kinderlehrer, and Otto, the JKO scheme as a time-discrete implicit Euler scheme for gradient flows in Wasserstein space has been extensively studied. However, corresponding discrete versions of important estimates valid for continuous equations remain to be established.
2. Limitations of Existing Results: P.W.Y. Lee first established in 2018 an estimate of the form $D^2 \log \rho_t^\tau \succeq -\frac{C}{2t}$ for the JKO scheme on the torus, but with three major limitations:
   • Requires regularity assumptions on initial data
   • The constant $C \in (1/2, 1]$ is not optimal
   • Cannot handle more general equation types
3. Technical Challenges: For the Fokker-Planck equation, direct mimicry of classical maximum principle arguments encounters gradient terms that are difficult to control. The absence of non-constant convex functions on the torus renders standard methods ineffective.

Core Contributions

1. Established Li-Yau-Hamilton inequality for the granular-medium equation: First established a Li-Yau-Hamilton type estimate at the continuous level for the granular-medium equation on the torus.
2. Proved asymptotic Li-Yau-Hamilton estimate for the JKO scheme: Improved Lee's results with breakthroughs in four directions:
   • Removed dependence on regularity of initial data
   • Asymptotically recovered the optimal constant 1/2
   • Extended to more general equation types
   • Can be extended to time $t=0$ for regular initial data
3. Derived quantitative estimates: Utilized the Li-Yau-Hamilton inequality to derive Lipschitz bounds, $L^\infty$ bounds, and quantitative Harnack inequalities.
4. Proved strong convergence: For the Fokker-Planck equation, established convergence of the JKO scheme in $L^2_{\mathrm{loc}}((0,+\infty); H^2(\mathbb{T}^d))$.

Detailed Methodology

Problem Formulation

Study the granular-medium equation on the torus $\mathbb{T}^d$: $\partial_t \rho_t = \Delta \rho_t + \nabla \cdot (\rho_t \nabla V + \rho_t \nabla W * \rho_t)$ where $V$ is the potential energy and $W$ is the interaction potential.

Core Technical Framework

1. Pressure Variable Method

Define the pressure function: $u[\rho] := \log \rho + V + W * \rho$

This function satisfies a nonlocal Hamilton-Jacobi equation: $\partial_t u_t = \Delta u_t + |\nabla u_t|^2 - \nabla V \cdot \nabla u_t - R(\nabla W, \nabla u_t, \rho_t)$

2. Key Lemma: Gradient Estimate for Semi-Convex Periodic Functions

Lemma 1.1: Let $u: \mathbb{R}^d \to \mathbb{R}$ satisfy $D^2 u \succeq -\lambda I_d$ in the weak sense (i.e., $u$ is $-\lambda$-convex). Then for any $i = 1,\ldots,d$, we have $\|\nabla u(x)\|_\infty \leq \frac{1}{2\lambda}$.

This lemma is the key tool for handling gradient terms on the torus.

3. Proof Strategy for the Continuous Case

For the smooth case, employ a maximum principle argument:

1. Set $-\lambda_t := \min_{x \in \mathbb{T}^d, \nu \in S^d} D^2 u[\rho_t](x)(\nu,\nu)$
2. At the optimal point $(x_t, \nu_t)$, utilize optimality conditions and semi-convexity bounds
3. Through calculation, obtain the differential inequality $-\dot{\lambda}_t \geq 2\lambda_t^2 - \Lambda\lambda_t$
4. Apply Grönwall's lemma for the logistic equation

4. Discrete Analysis of the JKO Scheme

One-Step Improvement Theorem: Let $D^2 u[\eta] \succeq -\lambda_0$ and $\rho \in \text{Prox}_\tau[\eta]$. Then: $G[\tau\lambda_1, \tau] \leq \tau\lambda_0$ where $G[E,\tau] := \frac{E}{(1-E)^2}(1-\tau(2\lambda^* + L^*) + \tau(\lambda^* + L^*)E)$.

Discrete Comparison Principle: Define the comparison sequence $(E_k^\tau)_{k \geq 0}$ satisfying:

• $E_0^\tau = \tau\lambda_0$
• $G[E_{k+1}^\tau, \tau] = E_k^\tau$

Then for any JKO flow, we have $D^2 u[\rho_k^\tau] \succeq -\frac{1}{\tau}E_k^\tau$.

Technical Innovations

1. Semi-Convexity Gradient Control: Cleverly controlled gradient terms on the torus using Lemma 1.1, circumventing difficulties with convexity assumptions.
2. Finite Difference Approximation: In non-regular cases, used finite differences in place of second derivatives, combined with limiting arguments to establish the maximum principle.
3. Asymptotic Analysis: Through linearization of the function $G$ and comparison with the corresponding ODE solution, obtained precise asymptotic estimates.

Experimental Setup

Theoretical Verification Framework

This is primarily a theoretical work verified through rigorous mathematical proofs. Main verification contents include:

1. Exact Estimates for the Heat Equation: When $V = W = 0$, obtain exact asymptotic behavior $E_k \sim \frac{1}{2k}$.
2. Asymptotic Estimates for General Cases: Proved that when $\Lambda > 0$, the asymptotic behavior is consistent with the continuous case.

Comparison Benchmarks

• P.W.Y. Lee (2018) results: Requires regularity assumptions, non-optimal constants
• Classical Li-Yau-Hamilton inequality: Standard results for the continuous case
• Aronson-Bénilan estimates: Similar results for porous medium equations

Experimental Results

Main Theoretical Results

Theorem 1.2 (Asymptotic Li-Yau-Hamilton Estimate)

For the JKO flow $(\rho_t^\tau)_{t \geq 0}$, when $t_0 > 0$ and $t \geq t_0$: $D^2(\log \rho_t^\tau + V + W * \rho_t^\tau) \succeq \begin{cases} -\frac{1+o(\tau)}{2t} & \text{if } \Lambda = 0 \\ -\frac{(1+o(\tau))\Lambda}{2(1-e^{-\Lambda t})} & \text{otherwise} \end{cases}$

Theorem 1.3 (Continuous Li-Yau-Hamilton Estimate)

For gradient flow solutions $\rho$ of the granular-medium equation, when $t > 0$: $D^2(\log \rho_t + V + W * \rho_t) \succeq \begin{cases} -\frac{1}{2t} & \text{if } \Lambda = 0 \\ -\frac{\Lambda}{2(1-e^{-\Lambda t})} & \text{otherwise} \end{cases}$

Application Results

1. Quantitative Lipschitz and $L^\infty$ Bounds

$|\nabla u[\rho_t]|_\infty \leq \frac{1}{2}E_t^{\lambda_0}$$e^{-\frac{d\sqrt{d}}{2}L_t^{\lambda_0}} \leq \rho \leq e^{\frac{d\sqrt{d}}{2}L_t^{\lambda_0}}$

2. Quantitative Harnack Inequality

$\rho_t(x) \leq \rho_{t+h}(y)\left(\frac{e^{\Gamma(t+h)}-1}{e^{\Gamma t}-1}\right)^d \exp(D_\rho(x,y;t,h))$

3. Strong Convergence Results

Theorem 1.4: For $F[\rho_0] < +\infty$:

1. $(\rho_t^\tau)_{t \geq 0}$ converges in $L^p_{\text{loc}}((0,+\infty); C^{0,\alpha}(\mathbb{T}^d))$
2. When $W = 0$, converges in $L^2_{\text{loc}}((0,+\infty); H^2(\mathbb{T}^d))$

Related Work

Development of Li-Yau Type Inequalities

1. Li-Yau (1986): Original results for heat equations
2. Hamilton (1993): Complete Hessian estimates
3. Bakry-Ledoux, Bakry-Bolley-Gentil: Extensions to Markov diffusion operators

Estimate Theory for JKO Schemes

1. Jordan-Kinderlehrer-Otto (1998): Establishment of JKO schemes
2. Lee (2018): First Li-Yau type estimate for JKO schemes
3. Ferrari-Santambrogio, Caillet-Santambrogio: Extensions of Lipschitz estimates

Granular-Medium Equations

1. Extensive research on aggregation-diffusion equations
2. Connections with McKean-Vlasov SDEs
3. Applications of optimal transport theory

Conclusions and Discussion

Main Conclusions

1. Successfully established the Li-Yau-Hamilton inequality for the granular-medium equation on the torus
2. Proved that the JKO scheme preserves an asymptotic version of this estimate
3. Derived multiple important quantitative estimates and convergence results

Limitations

1. Torus Restriction: Methods depend on compactness and periodicity of the torus
2. Regularity Assumptions: Some results require $C^{2,1}$ regularity of potential functions
3. Interaction Terms: Handling the case $W \neq 0$ involves more complex techniques

Future Directions

1. Extension to more general manifolds and domains
2. Treatment of more general nonlinear diffusion terms
3. Investigation of other types of gradient flow estimates

In-Depth Evaluation

Strengths

1. Theoretical Innovation: First establishment of Li-Yau-Hamilton inequality for the granular-medium equation, filling a theoretical gap
2. Technical Breakthrough: Cleverly solved the technical difficulty of controlling gradient terms on the torus
3. Complete Results: Forms a complete theoretical framework from one-step estimates to asymptotic analysis
4. Rich Applications: Derives multiple important quantitative results and convergence theorems

Weaknesses

1. Limited Scope: Primarily restricted to the torus setting; extension to general domains requires new techniques
2. Computational Complexity: Some proofs are highly technical; practical computation may be complex
3. Constant Optimization: While asymptotically recovering optimal constants, finite-time constants still have room for improvement

Impact

1. Theoretical Contribution: Provides important analytical tools for JKO scheme theory
2. Methodological Value: Semi-convexity gradient control techniques may be applicable to other problems
3. Practical Value: Quantitative estimates provide guidance for numerical analysis and algorithm design

Applicable Scenarios

1. Numerical Analysis: Convergence analysis and error estimation for JKO schemes
2. Optimal Transport: Theoretical research on Wasserstein gradient flows
3. Partial Differential Equations: Qualitative analysis of aggregation-diffusion equations

References

The paper cites 33 important references, covering classical and cutting-edge work in optimal transport, gradient flows, Li-Yau inequalities, and other fields, reflecting the depth and breadth of the research.

Overall Assessment: This is a high-quality theoretical mathematics paper that makes important contributions at the intersection of Li-Yau-Hamilton inequalities and JKO schemes. It demonstrates strong technical innovation, complete theoretical results, and provides a solid foundation for further research in related fields.