Lusin-type approximation of Sobolev by Lipschitz functions, in Gaussian and $RCD(K,\infty)$ spaces

Luigi Ambrosio, Elia Brué, Dario Trevisan

Abstract

We establish new approximation results, in the sense of Lusin, of Sobolev functions by Lipschitz ones, in some classes of non-doubling metric measure structures. Our proof technique relies upon estimates for heat semigroups and applies to Gaussian and $RCD(K, \infty)$ spaces. As a consequence, we obtain quantitative stability for regular Lagrangian flows in Gaussian settings.

Introduction

In classical functional analysis, the relationship between Sobolev functions and Lipschitz functions is fundamental. A Sobolev function, while not necessarily smooth, possesses derivatives that are integrable. A cornerstone result in Euclidean space is the Lusin-type approximation: for any Sobolev function $f$ and any $\epsilon > 0$, there exists a Lipschitz function $f_\epsilon$ such that $f = f_\epsilon$ except on a set of measure $\epsilon$. Essentially, Sobolev functions are "almost" Lipschitz.

However, traditional proofs of this property rely heavily on the doubling condition—the geometric assumption that the volume of a ball grows by at most a constant factor when its radius is doubled. Many critical mathematical environments, including Gaussian spaces (essential to probability) and infinite-dimensional Wiener spaces (central to stochastic analysis), are non-doubling. This paper bridges this gap, extending Lusin-type approximations to these complex, non-doubling metric measure spaces.

Methodology: Heat Semigroups as Regularizers

To bypass the limitations of the doubling condition, this work moves away from traditional geometric tools like maximal functions and spherical averages. Instead, the authors employ heat semigroups ($R_t$) as the primary mechanism for regularization.

1. Heat Flow Regularization

Rather than averaging a function over a local ball to smooth it, the authors use heat flow—the natural diffusion of a value over time—to regularize Sobolev functions. This approach is more intrinsically linked to the underlying geometry and curvature of the space than rigid geometric balls.

2. The Representation Formula

A key innovation in this paper is a novel representation formula. The authors express the difference between a function and its heat-regularized version using the square root of the Laplacian operator ($\sqrt{-L}$). By analyzing a specific kernel $K(s, t)$, they demonstrate that the "oscillation" of a function can be controlled by its gradient, even in spaces where volume grows exponentially or dimensions are infinite.

3. Application Frameworks

The technique is successfully applied across three distinct mathematical settings:

  • Da Prato’s Sobolev spaces: Gaussian measures on Hilbert spaces.
  • Wiener spaces: Utilizing Malliavin calculus.
  • $RCD(K, \infty)$ spaces: Metric measure spaces characterized by a lower bound on Ricci curvature.

Key Results and Quantitative Estimates

Quantitative Lusin Approximation

The authors establish a robust pointwise estimate for Sobolev functions in Gaussian, Wiener, and $RCD$ spaces: $$|f(x) - f(y)| \leq C_p |x - y| (g(x) + g(y))$$ where $g$ is an $L^p$ function directly related to the gradient of $f$. This inequality provides the necessary control to treat Sobolev functions as Lipschitz-like entities in non-doubling environments.

Stability of Lagrangian Flows

One of the most significant applications of this approximation is in the study of regular Lagrangian flows—the trajectories of particles within a vector field. In Theorem 5.2, the authors provide the first quantitative stability estimates for these flows in Gaussian settings. They prove that if two vector fields $b$ and $\bar{b}$ are close in $L^1$, the distance between their resulting flows is bounded by a term proportional to: $$\frac{1}{|\log \|b - \bar{b}\| |}$$ This logarithmic bound ensures that the paths of particles do not diverge erratically under slight perturbations of the driving field.

Discussion and Limitations

While this work provides a significant leap forward, it also identifies areas for future research. In the $RCD(K, \infty)$ setting, the results are currently limited to $p=2$. This restriction exists because "Riesz inequalities"—which relate the Laplacian to the gradient—have not yet been fully established for general $p$ in these specific spaces. Furthermore, the results in $RCD$ spaces currently require a lower bound on curvature ($K$).

By connecting abstract functional analysis with the dynamics of flows, this paper provides a framework for studying the well-posedness and stability of systems in general metric measure spaces where previous methods were only qualitative.

Summary updated

Abstract

We establish new approximation results, in the sense of Lusin, of Sobolev functions by Lipschitz ones, in some classes of non-doubling metric measure structures. Our proof technique relies upon estimates for heat semigroups and applies to Gaussian and $RCD(K, \infty)$ spaces. As a consequence, we obtain quantitative stability for regular Lagrangian flows in Gaussian settings.
BibTeX
@inproceedings{ambrosio2018lusin,
  author = {Luigi Ambrosio and Elia Brué and Dario Trevisan},
  title = {{Lusin-type approximation of Sobolev by Lipschitz functions, in Gaussian and $RCD(K,\textbackslash\{\}infty)$ spaces}},
  booktitle = {Advances},
  year = {2018},
  doi = {10.1016/j.aim.2018.09.033},
}
Figure