The Aronson-Bénilan estimate for a Lagrangian particle discretization of the Porous Medium Equation

Technical Summary: The Aronson-Bénilan Estimate for a Lagrangian Particle Discretization of the Porous Medium Equation

Problem Statement
The paper addresses the numerical approximation of the one-dimensional porous medium equation (PME), given by $\partial_t \rho = \partial_{xx}(\rho^m)$ with $m > 1$. While the PME is a classical model for gas filtration and nonlinear diffusion, its numerical treatment via Lagrangian particle methods presents specific challenges. Unlike "blob methods" that convolve the velocity field to handle discontinuities (thereby turning the local diffusion operator into a nonlocal transport operator), this work focuses on a nearest-neighbor Lagrangian particle discretization.

The primary mathematical difficulty lies in the fact that, while this specific particle scheme (first considered in [20] and [27]) successfully reproduces the variational structure of the PME in the Wasserstein gradient flow setting, it had not previously been proven to satisfy the fundamental regularizing estimates characteristic of the continuum equation. Specifically, the scheme lacked a discrete analog of the celebrated Aronson-Bénilan estimate, which is crucial for controlling the speed of support propagation and ensuring the $L^\infty$ decay of solutions. Without this control, rigorous convergence proofs for general $L^1$ initial data (including those with compact support or singularities) were previously unavailable.

Methodology
The authors propose and analyze a deterministic particle approximation (DPA) where $N+1$ ordered particles $x_0(t) < \dots < x_N(t)$ evolve according to a system of ODEs derived from the Lagrangian formulation of the PME. The velocity of each particle is determined by the difference in the $m$-th power of the local densities of the adjacent intervals:
$$\dot{x}_i(t) = -N \left( (R_{i+1}(t))^m - (R_i(t))^m \right)$$
where $R_i(t) = \frac{1}{N(x_i(t) - x_{i-1}(t))}$ represents the piecewise constant density reconstruction on the interval $(x_{i-1}, x_i)$.

The core methodological innovation is the derivation and rigorous proof of a discrete Aronson-Bénilan estimate. The authors define a discrete quantity $Z_k(t)$ analogous to the continuum quantity $\partial_{xx}(\frac{m\rho^{m-1}}{m-1})$. By analyzing the time evolution of the minimum of these discrete quantities using a discrete maximum principle and summation-by-parts techniques, they establish a lower bound for $Z_k(t)$ that mirrors the continuum behavior.

Key Contributions and Results

1. Discrete Aronson-Bénilan Estimate (Theorem 1):
   The paper proves that the discrete quantity $Z_k(t)$ satisfies the inequality:
   $$Z_k(t) \geq -\frac{1}{|Z^{(N)}|^{-1} + (m+1)t}$$
   where $|Z^{(N)}|$ is a constant depending on the initial data. This is the first rigorous proof that this specific nearest-neighbor scheme satisfies a discrete version of the fundamental regularizing estimate known for the continuum PME.

2. Uniform Support Growth Estimate (Theorem 2):
   Leveraging the discrete Aronson-Bénilan estimate, the authors derive a uniform bound on the growth of the support of the approximate density $\rho^{(N)}$. They show that the length of the support $L_N[X^{(N)}(t)]$ grows at most as $t^{1/(m+1)}$, with constants independent of the number of particles $N$. This resolves the previous difficulty in controlling the propagation speed of the support uniformly with respect to discretization.

3. Uniform $L^\infty$ Decay (Theorem 3):
   The authors establish a uniform $L^\infty$ decay estimate for the discrete densities:
   $$\max_k R_k^{(N)}(t) \leq C t^{-1/(m+1)}$$
   While the constant $C$ derived is not the optimal sharp constant found in the continuum self-similar solutions, it is universal and independent of the discretization parameter $N$.

4. Convergence to Weak Solutions (Theorem 4):
   Combining the uniform estimates (support growth and $L^\infty$ decay) with compactness arguments (Aubin-Lions theorem), the paper proves that the sequence of piecewise constant reconstructions $\rho^{(N)}$ converges in $L^1([0, T] \times \mathbb{R})$ to the unique distributional solution of the PME. Crucially, this convergence holds for general $L^1$ initial data, including cases where the solution is not strictly positive or smooth, a regime where previous convergence results for this scheme were limited.

Significance and Claims
The paper claims that the discrete Aronson-Bénilan estimate is the "key tool" necessary to bridge the gap between the discrete particle dynamics and the continuum weak solution theory. The authors assert that:

• The discrete version of the estimate alone is sufficient to prove the convergence of the scheme to the weak solution in the full generality of $L^1$ initial data.
• The scheme successfully replicates the qualitative properties of the PME (finite speed of propagation and smoothing effects) at the discrete level without resorting to nonlocal regularization (blob methods).
• The results provide a rigorous theoretical foundation for using this specific nearest-neighbor Lagrangian discretization for nonlinear diffusion problems, validating its use in contexts requiring the preservation of local interaction structures and support properties.

The work does not propose new applications or experimental setups but rather solidifies the mathematical justification for an existing numerical approach, ensuring its reliability for a broad class of initial conditions previously unaddressed in the literature for this specific scheme.