Quantitative stability for quasilinear parabolic equations

This paper establishes explicit quantitative convergence rates for viscosity solutions of quasilinear parabolic equations under perturbations, including variations in the exponent $p$ and regularized approximations, by adapting standard comparison arguments to cover both singular and degenerate cases like the normalized and variational $p$-parabolic equations.

The Gist

Imagine you are trying to predict the weather. You have a complex computer model that simulates how wind, temperature, and pressure interact. But your model isn't perfect; it relies on certain "knobs" or settings (like how sensitive the wind is to temperature changes) that we can't measure with infinite precision.

This paper is essentially about how much your weather forecast changes when you tweak those knobs slightly.

In the world of mathematics, these "weather models" are called Partial Differential Equations (PDEs). Specifically, the authors are looking at a class of equations that describe how things spread out or evolve over time, like heat moving through a metal rod or a drop of ink spreading in water. These are called parabolic equations.

Here is a breakdown of what the authors did, using simple analogies:

1. The Problem: The "Fuzzy" Knobs

The equations in this paper describe things that can get very tricky. Sometimes, the math "breaks" or becomes singular when the slope of the solution is zero (think of a perfectly flat spot on a hill).

• The Scenario: Imagine you have a specific equation (let's call it the "Perfect Model") that describes a physical process. But in reality, we often use a slightly different version of the equation because the "Perfect Model" is too hard to solve, or because we don't know the exact value of a parameter (like the exponent $p$ in a $p$-Laplace equation).
• The Question: If we change the parameter slightly (turn the knob a tiny bit), does the solution (the forecast) change wildly, or does it stay close to the original? And if it stays close, how close is it?

2. The Goal: Measuring the "Drift"

Previous research knew that if you turn the knob a tiny bit, the answer doesn't change much (this is called stability). But they didn't know exactly how fast the answer converges back to the truth.

• The Analogy: Imagine you are walking toward a target. You know that if you aim slightly off, you will still hit near the target. But this paper asks: "If I am off by 1 degree, am I 1 meter away? 10 meters? Or 0.1 meters?"
• The Achievement: The authors calculated the exact speed of this convergence. They provided a formula that tells you: "If you change the parameter by an amount $\epsilon$, your answer will be off by roughly $\epsilon^\nu$."

3. The Method: The "Double Vision" Technique

To prove this, the authors used a mathematical trick called the "Doubling of Variables" method.

• The Metaphor: Imagine you have two versions of a movie playing on two screens. One screen shows the "Perfect Model" ($u_0$), and the other shows the "Approximate Model" ($u_\epsilon$).
• The authors create a giant, invisible "rubber band" that connects the two movies. They stretch this rubber band to see where the two movies differ the most.
• They then use a tool called the Crandall-Ishii Lemma (think of it as a super-precise microscope) to zoom in on the exact moment and place where the two movies diverge the most.
• By analyzing the tension in the rubber band at that specific point, they can mathematically prove how much the two movies can possibly differ based on how much the "knob" was turned.

4. The Results: It Depends on the "Smoothness"

The paper finds that the accuracy of the prediction depends on how "smooth" the solution is.

• Smooth Solutions (Lipschitz): If the solution is very smooth (like a gentle hill), the error is very small. If you turn the knob by 0.01, the error might be 0.01.
• Rough Solutions (Hölder): If the solution is a bit jagged or rough (like a rocky mountain), the error is slightly larger. If you turn the knob by 0.01, the error might be $0.01^{0.5}$ (which is 0.1).
• The Singularity Issue: The authors also dealt with equations that have "singularities" (mathematical black holes where the slope is zero). They showed that even in these tricky cases, the stability holds, provided the solution doesn't get too jagged.

5. Real-World Applications

Why does this matter?

• Image Processing: When computers smooth out images or remove noise, they use these types of equations. Knowing the convergence rate helps engineers know how much they can simplify the math without ruining the image quality.
• Game Theory: The paper mentions "Tug-of-War" games. In these games, two players pull a token, and the outcome is determined by a specific type of math equation. This research helps predict how the game's outcome changes if the rules are tweaked slightly.
• Material Science: When modeling how materials deform or how heat flows through irregular shapes, these equations are used. Knowing the stability ensures that small measurement errors in the lab don't lead to massive errors in the engineering design.

Summary

In short, Kurkinen and Liu took a complex, messy class of mathematical equations that describe how things change over time, and they built a quantitative ruler to measure how sensitive these equations are to small changes.

They proved that even when the math gets weird and singular (like a flat spot on a hill), the solutions remain stable, and they gave us the exact formula to calculate how much the solution will wiggle when we tweak the input. It's like giving engineers a precise manual that says, "If you change this setting by 1%, your result will change by no more than X%."

Technical Summary

Here is a detailed technical summary of the paper "Quantitative Stability for Quasilinear Parabolic Equations" by Tapio Kurkinen and Qing Liu.

1. Problem Statement

The paper addresses the quantitative stability of viscosity solutions for a broad class of quasilinear parabolic partial differential equations (PDEs) under perturbations of their parameters. Specifically, the authors investigate how the solutions $u_\varepsilon$ of a perturbed equation converge to the solution $u_0$ of a limiting equation as the perturbation parameter $\varepsilon \to 0$.

The general form of the equations studied is:
$$\partial_t u - \text{tr}(A(\nabla u)\nabla^2 u) + H(x, t, \nabla u) = 0$$
Key challenges addressed include:

• Singularities and Degeneracy: The elliptic operator may exhibit singularities when the gradient $\nabla u$ vanishes (e.g., in the $p$-Laplacian for $1 < p < 2$) or degeneracy when$p > 2$.
• Specific Applications: The framework covers:
  • Normalized $p$-Laplace equations: $\partial_t u - \Delta_p^N u = 0$.
  • Variational $p$-Laplace equations: $\partial_t u - \Delta_p u = 0$.
  • Regularized approximations: Replacing singular operators with non-singular ones (e.g., adding $\varepsilon^2$ to the gradient term).
  • Hamilton-Jacobi equations: Vanishing viscosity limits.
• Goal: Unlike previous works that established qualitative convergence (uniform convergence), this paper aims to derive explicit convergence rates (e.g., $O(\varepsilon^\nu)$) and determine the optimality of these rates.

2. Methodology

The authors employ a viscosity solution framework adapted for singular operators, known as F-solutions (introduced by Ohnuma and Sato). The core methodology relies on:

• Doubling of Variables: A standard technique in viscosity solution theory where a test function involving two space-time points $(x, t)$ and $(y, s)$ is maximized to compare two solutions $u_\varepsilon$ and $u_0$.
• Crandall-Ishii Lemma: The parabolic version of this lemma is used to obtain second-order sub/super-jets $(\tau, \xi, X)$ and $(\sigma, \eta, Y)$ at the maximum point, allowing the comparison of the PDE operators.
• Handling Singularities:
  • The authors define a class of admissible test functions ($C^2_A$) compatible with the singularity structure of the operator $A$.
  • They utilize a generalized notion of viscosity solutions (F-solutions) that handles points where $\nabla u = 0$ by imposing specific conditions on the test functions (using functions like $f(r) = r^k$).
• Regularity Assumptions: The proofs crucially rely on the equi-Hölder continuity of the family of solutions $\{u_\varepsilon\}$. Specifically, they assume:
  $$|u_\varepsilon(x_1, t_1) - u_\varepsilon(x_2, t_2)| \leq L(|x_1 - x_2|^\theta + |t_1 - t_2|^\theta)$$
  for some $\theta \in (0, 1]$.
• Optimization: The convergence rate is derived by optimizing the error terms arising from the doubling variables (specifically the distance between the maximizers) with respect to the perturbation parameter $\varepsilon$.

3. Key Contributions and Main Results

A. General Quantitative Stability Theorem (Theorem 1.1)

The paper establishes a general convergence rate for the difference between the perturbed solution $u_\varepsilon$ and the limit solution $u_0$:
$$\sup_{\Omega_T} |u_\varepsilon - u_0| \leq \sup_{\partial_p \Omega_T} |g_\varepsilon - g_0| + C \varepsilon^\nu + c_H T \varepsilon^\gamma$$
where the exponent $\nu$ depends on the Hölder exponent $\theta$, the convergence rate of the operator coefficients ($\alpha$), and the singularity strength ($\beta$):
$$\nu = \frac{\alpha \theta}{1 + (1-\theta)\max\{\beta, 0\}}$$

• Significance: This formula explicitly quantifies how the regularity of the solution ($\theta$) and the nature of the operator's singularity ($\beta$) affect the convergence speed.
• Observation: If solutions are Lipschitz ($\theta=1$), the rate simplifies to $O(\varepsilon^\alpha)$, and the singularity strength $\beta$ becomes irrelevant.

B. Applications to $p$-Laplace Equations

The authors apply the general theorem to specific cases, deriving concrete rates:

1. Normalized $p$-Laplacian: For $p \to q$, if solutions are equi-Hölder continuous with exponent $\theta$, the convergence rate is $O(|p-q|^\theta)$.
2. Variational $p$-Laplacian:
   • For $p < 2$ (singular case), the rate is $O(|p-q|^\theta)$.
   • For $p > 2$ (degenerate case), the rate is $O(|p-q|^\nu)$ where $\nu < \frac{2\theta}{(1-\theta)q + 2\theta}$.
3. Regularization of Generalized $p$-Laplacian: For equations regularized by $\varepsilon$, the convergence rate $O(\varepsilon^\nu)$ depends on the regularization parameter $p'$:
   • If $p'=2$ (mean curvature flow type), $\nu < \theta/2$.
   • If $p' > 4$, $\nu < \frac{\theta}{1 + (1-\theta)(p'-4)}$.

C. Elliptic Counterparts (Theorem 3.6)

The methodology is extended to elliptic equations. A key distinction is the requirement of a monotonicity condition on the lower-order term $H$ with respect to $u$ (i.e., $H(x, u, \xi) - \lambda u$ is non-decreasing) to ensure uniqueness and stability in the elliptic setting.

D. Optimality and Examples

The paper provides examples (e.g., Barenblatt solutions, 1D heat diffusion) to demonstrate that the derived rates are sharp for Lipschitz solutions. It also highlights a limitation: for non-Lipschitz initial data, the regularizing effect of the parabolic flow might improve the rate over time, a nuance not captured by the global-in-time estimate which assumes uniform Hölder continuity.

4. Significance

• Bridging Qualitative and Quantitative: While qualitative stability (convergence) for $p$-Laplace equations was known, this work provides the first explicit, quantitative rates for a general class of singular/degenerate parabolic equations using purely PDE methods.
• Unified Framework: The approach unifies the treatment of normalized and variational $p$-Laplacians, as well as various regularization schemes, under a single theoretical umbrella.
• Handling Singularities: The rigorous treatment of singularities at vanishing gradients via F-solutions and specific test function classes allows for the analysis of the critical range $1 < p < 2$, which is often difficult due to the singularity.
• Practical Implications: The explicit rates are valuable for numerical analysis (error estimation in finite difference/element methods) and for understanding the sensitivity of physical models (e.g., image processing, game theory models like tug-of-war) to parameter variations.

In summary, Kurkinen and Liu successfully extend the comparison principle techniques to derive precise error bounds for perturbed quasilinear parabolic equations, offering a robust tool for analyzing the stability of solutions in the presence of singularities and degeneracies.