Counterexamples to the $L^p$-Calderón--Zygmund Estimate on Open Manifolds

Siran Li

Abstract

Based on a construction due to B. Güneysu and S. Pigola (Adv. Math. 281 (2015), pp.353–393), for each $p \in [1,\infty]$ and $m \in \mathbb{Z}_{\geq 2}$, we exhibit an $m$-dimensional Riemannian open manifold $\mathcal{M}$ on which the $L^p$-Calderón–Zygmund estimate

$$\|\nabla \nabla f\|_{L^p} \leq C_1 \|\Delta f\|_{L^p} + C_2 \| f\|_{L^p}$$

is false for any $C_1, C_2$ depending on $m$ and $p$. Therefore, one must impose further geometric conditions on the manifold to ensure the validity of the Calderón–Zygmund estimate.

Introduction

In the analysis of partial differential equations (PDEs) within Euclidean space $\mathbb{R}^m$, the Calderón–Zygmund estimate serves as a fundamental regularity result. It establishes that the $L^p$ norm of a function's full Hessian (the matrix of second derivatives, $\nabla \nabla f$) can be controlled by the $L^p$ norms of its Laplacian (the trace of the Hessian, $\Delta f$) and the function itself. Essentially, this estimate guarantees that if the "average" second derivative is well-behaved, then all individual second-order directional derivatives are similarly controlled.

While this estimate is a staple of flat-space analysis for $1 < p < \infty$, its validity on more general Riemannian manifolds—specifically open, non-compact, and complete manifolds—is not guaranteed. This paper demonstrates that without specific geometric constraints, the Calderón–Zygmund estimate can fail universally across all dimensions $m \geq 2$ and for the entire range of $p \in [1, \infty]$.

The Problem: Global vs. Local Geometry

On a general Riemannian manifold $(\mathcal{M}, g)$, the relationship between the Laplacian and the Hessian is mediated by the curvature of the manifold. In Euclidean space, the curvature is zero, and the estimate holds with universal constants. However, on an open manifold, the geometry can become increasingly "wild" as one moves toward infinity.

Previous research had identified specific instances where the estimate fails, such as for $p=2$ in two dimensions. The work presented here extends this failure to a much broader class of spaces, proving that the estimate is not a generic property of open manifolds but rather one that requires strict geometric regularity, such as bounded curvature or a positive injectivity radius.

Methodology: Warped Product Constructions

The proof relies on the explicit construction of a family of manifolds using warped products. The manifold $\mathcal{M}$ is defined by taking $\mathbb{R}^m$ and equipping it with a metric of the form:

$$g = dr^2 + \sigma^2(r) \text{can}^{m-1}$$

Here, $\text{can}^{m-1}$ represents the standard metric on the $(m-1)$-sphere, and $\sigma(r)$ is the "warping function" that dictates the manifold's radial geometry.

The Sawtooth Mechanism

The core of the counterexample lies in the choice of $\sigma(r)$. The author constructs a "sawtooth" warping function that oscillates with increasing frequency and amplitude. The critical insight is the decoupling of the components of the Calderón–Zygmund inequality:

  1. The Laplacian and $L^p$ Norm: These terms depend primarily on the magnitude of the warping function $\sigma(r)$. By carefully controlling the growth of $\sigma$, these terms can be kept relatively small.
  2. The Hessian: The Hessian contains terms involving the derivative $\sigma'(r)$. By making $\sigma$ oscillate rapidly, the author ensures that $\sigma'$ becomes extremely large at specific intervals.

By using localized versions of Green's functions (the fundamental solutions to the Laplace equation) as test functions, the author demonstrates that the Hessian term can be forced to "blow up" while the Laplacian and function norms remain bounded.

Key Results

The primary contribution of the paper is Theorem 1.1, which states that for any dimension $m \geq 2$ and any $p \in [1, \infty]$, there exists an open Riemannian manifold where the Calderón–Zygmund estimate is invalid for any constants $C_1, C_2$.

Furthermore, the paper provides a quantitative measure of this failure. In the constructed manifolds, the ratio of the Hessian's $L^p$ norm to the combined norms of the Laplacian and the function grows exponentially (specifically, at a rate comparable to $e^{k^{1000}}$ for large $k$). This massive divergence proves that no universal constants can satisfy the inequality across the entire manifold.

Conclusion and Geometric Implications

The counterexamples constructed in this study are characterized by "pathological" geometry. Specifically, they exhibit:

  • Unbounded Curvature: The Riemann curvature tensor is not bounded from above or below.
  • Vanishing Injectivity Radius: The manifold lacks a uniform lower bound on the size of local coordinate neighborhoods.

These results clarify the boundary between analysis and geometry. They demonstrate that the Calderón–Zygmund estimate is intrinsically linked to the global curvature bounds of the underlying space. While the estimate is known to hold on manifolds with bounded Ricci curvature and a positive injectivity radius, this work shows that such conditions are not merely technical conveniences but are strictly necessary to prevent the second derivatives from oscillating out of control.

Summary updated

Abstract

Based on a construction due to B. Güneysu and S. Pigola (\textit{Adv. Math.} \textbf{281} (2015), pp.353--393), for each $p \in [1,\infty]$ and $m \in \mathbb{Z}_{\geq 2}$, we exhibit an $m$-dimensional Riemannian open manifold $\mathcal{M}$ on which the $L^p$-Calderón--Zygmund estimate \begin{equation*} \|\nabla \nabla f\|^p_{L^p} \leq C_1 \|Δf\|_{L^p}^p + C_2 \| f\|_{L^p}^p \qquad \text{ for all } f \in C^\infty_c(\mathcal{M}) \end{equation*} is false for any $C_1, C_2$ depending on $m$ and $p$. Therefore, one must impose further geometric conditions on the manifold to ensure the validity of the Calderón--Zygmund estimate.
BibTeX
@article{li2019counterexamples,
  author = {Siran Li},
  title = {{Counterexamples to the $L^p$-Calderón--Zygmund Estimate on Open Manifolds}},
  journal = {Annals},
  year = {2019},
  doi = {10.1007/s10455-019-09688-3},
}

Loading related content...

Figure