Diffusion bounds for non-autonomous degenerate parabolic equations

1. Problem Statement

The paper addresses the problem of establishing diffusive upper bounds (specifically, off-diagonal decay estimates) for the propagators of parabolic partial differential equations (PDEs).

• Equation Class: The authors consider the evolution equation $\partial_t u = Lu$ where $L$ is a non-autonomous, divergence-form operator:
  $$Lu = \text{div}(a \nabla u) + \langle b, \nabla u \rangle + cu$$
  defined on a domain $\Omega$ (either Euclidean space $\mathbb{R}^n$ or a smooth Riemannian manifold).
• Key Challenges:
  1. Degeneracy: The diffusion matrix $a(x,t)$ is only required to be non-negative semi-definite ($a \ge 0$), meaning it can vanish on arbitrary open subsets of the domain. This breaks standard uniform ellipticity and hypoellipticity assumptions.
  2. Non-Autonomy: Coefficients depend explicitly on time $t$.
  3. Non-Linearity: The framework extends to nonlinear equations where coefficients depend on the solution $u$ and its gradient.
  4. Drift: The presence of a drift term $b$ introduces ballistic transport, complicating pure diffusive estimates.

The goal is to prove Davies-Gaffney type bounds (integrated Nash-Aronson estimates) in the $L^p$-sense ($1 \le p \le \infty$). These bounds quantify how an initially localized solution spreads over time, specifically showing that the solution decays exponentially with the square of the distance divided by time ($d^2/t$) outside a "ballistic wave front."

2. Methodology

The authors develop a unified geometric exponential deformation technique that avoids reliance on classical tools like Li-Yau gradient estimates, the maximum principle, or Sobolev inequalities, which typically require strict ellipticity.

• Exponential Tilting (The Core Argument):
  The proof relies on a "conjugation" or "tilting" of the propagator $P_{t,s}$ by an exponential weight. Let $\phi$ be a signed cutoff function adapted to two disjoint sets $X$ and $Y$. The authors define the operator $T = e^\phi$.
  The key identity used is:
  $$\chi_X P_{t,s} \chi_Y = \chi_X T^{-1} (T P_{t,s} T^{-1}) T \chi_Y$$
  By Hölder's inequality, the norm is bounded by:
  $$\|\chi_X P_{t,s} \chi_Y\| \le \| \chi_X T^{-1} \|_\infty \| T P_{t,s} T^{-1} \|_{L^1 \to L^1} \| T \chi_Y \|_\infty$$

• Estimating the Deformed Propagator:
  The term $\| T P_{t,s} T^{-1} \|$ is estimated by analyzing the generator of the deformed semigroup. If $L$ generates the original semigroup, the deformed generator is $L_\phi = e^\phi L e^{-\phi}$.
  The authors show that the growth of the $L^1$ norm of the deformed solution is controlled by the supremum of a specific quantity $A$ derived from the coefficients and the gradient of $\phi$:
  $$A \approx \sup \left( \text{div}(a\nabla \phi) + \langle a\nabla \phi, \nabla \phi \rangle - \langle b, \nabla \phi \rangle \right)_+$$
  Crucially, this estimate holds even if $a$ vanishes, provided $a \ge 0$.

• Construction of Cutoff Functions:
  The method requires constructing a smooth function $\phi$ that transitions between values on sets $X$ and $Y$ with controlled derivatives.

  • In $\mathbb{R}^n$, this is achieved via Whitney's extension theorem.
  • On Riemannian manifolds, the authors introduce the $\gamma$-cutoff property, ensuring the existence of functions with bounded Laplacian and gradient relative to the distance $d_{XY}$.

• Optimization:
  The decay rate is optimized by choosing the amplitude of the cutoff function $\phi$ (denoted as $\mu$) to maximize the exponential decay exponent, leading to the Gaussian-like bound.

3. Key Contributions and Results

A. Main Theorems

1. Theorem 2.1 (Euclidean Case): Establishes the bound for $\Omega = \mathbb{R}^n$.
   $$\|\chi_X P_{t,s} \chi_Y\|_{L^p \to L^p} \le \exp\left( -\frac{d_{XY}^2}{4k^2 \alpha (t-s)} \right)$$
   Valid under the condition $k(\frac{\alpha}{d_{XY}} + \beta)(t-s) \le d_{XY}$.

   • $\alpha = \sup \|a\|_\infty$, $\beta = \sup (\|b\|_\infty + \|\text{div } a\|_\infty)$.
   • $k$ is a constant depending on dimension.

2. Theorem 2.4 (Riemannian Manifolds): Extends the result to manifolds $(M, g)$ provided the sets $X$ and $Y$ satisfy the $\gamma$-cutoff property. This property is verified for manifolds with non-negative Ricci curvature (Proposition 2.5) and Cartan-Hadamard manifolds.

3. Theorem 2.7 (Nonlinear Equations): Generalizes the bounds to nonlinear equations of the form $\partial_t u = L(u)u$ (e.g., porous medium, McKean-Vlasov). The constants $\alpha$ and $\beta$ become dependent on the solution $u$ and its derivatives, providing a priori estimates.

4. Theorem 2.9 (Sharp Constants): For operators where $a$ is independent of $x$ (and $b=0$), the authors prove that the constant $k$ can be taken as 1, matching the sharp Gaussian heat kernel bounds known for the Laplacian on non-negatively curved manifolds.

B. The "Ballistic Validity Interval"

A critical insight is the explicit inclusion of a validity interval condition:
$$k \left( \frac{\alpha}{d_{XY}} + \beta \right) t \le d_{XY}$$

• Significance: This condition accounts for the "ballistic" transport caused by the drift $b$ or spatial variations in $a$ (i.e., $\nabla a$).
• Interpretation: The diffusive decay ($e^{-d^2/t}$) only applies outside a wave front moving at speed proportional to $\beta$. Inside this front, the solution may have transported ballistically. Previous literature often hid this in the prefactor; this paper makes it explicit, sharpening the estimate.

C. Degeneracy Handling

The method works for strongly degenerate operators where $a$ vanishes on open sets (not just measure zero sets).

• Example: The paper notes that if $a \equiv 0$ and $b$ is constant, the equation is a pure transport equation (no diffusion). The derived bound correctly reflects this by requiring the distance to be larger than the ballistic distance $\beta t$ for any diffusive decay to appear.

4. Significance and Impact

1. Unification: The paper provides a single, robust framework covering linear, nonlinear, autonomous, non-autonomous, elliptic, and degenerate parabolic equations.
2. Relaxation of Assumptions: It removes the need for hypoellipticity or uniform ellipticity. This is a major advancement, as many physical models (e.g., porous media, kinetic equations) involve degenerate diffusion.
3. Geometric Flexibility: By introducing the $\gamma$-cutoff property, the results apply to a wide class of Riemannian manifolds, including those with boundaries and non-trivial topology, without requiring global curvature bounds for the general case.
4. Sharpness: The method recovers the optimal constant ($k=1$) for standard Laplacians, proving that the technique is not just a qualitative tool but quantitatively sharp.
5. Applications: The framework is directly applied to:
   • Porous Medium Equation: Showing sub-diffusive spreading ($t^\gamma$ with $\gamma < 1/2$) and deriving explicit bounds.
   • McKean-Vlasov Equations: Analyzing diffusion in velocity space for interacting particle systems, showing that velocity spreads diffusively ($t^{1/2}$) despite the presence of drift.

Conclusion

This paper fundamentally advances the theory of parabolic PDEs by proving that diffusive decay estimates hold even in the absence of ellipticity, provided one accounts for the ballistic transport induced by drift and coefficient gradients. The "exponential deformation" method offers a powerful, elementary alternative to the heavy machinery of gradient estimates and functional inequalities traditionally used in this field.