Robust nonlocal trace and extension theorems

Basic Information

• Paper ID: 2305.05735
• Title: Robust nonlocal trace and extension theorems
• Authors: Florian Grube, Moritz Kassmann (Universität Bielefeld)
• Classification: math.AP (Analysis of PDEs)
• Publication Date: May 2023 (v1), June 2023 (v2)
• Paper Link: https://arxiv.org/abs/2305.05735

Abstract

This paper establishes trace theorems and extension theorems for Sobolev-type function spaces applicable to nonlocal Dirichlet and Neumann problems, including fractional p-Laplacians. These results are robust with respect to the order of differentiability, maintaining consistency with classical trace and extension theorems in this sense.

Research Background and Motivation

Core Problem

The paper studies nonlinear nonlocal equations on bounded domains, such as: $(-\Delta)_p^s u = f \text{ in } \Omega, \quad u = g \text{ in } \mathbb{R}^d \setminus \Omega$

where the fractional p-Laplacian is defined as: $(-\Delta)_p^s u(x) = (1-s) \text{p.v.} \int_{\mathbb{R}^d} \frac{|u(x)-u(y)|^{p-2}(u(x)-u(y))}{|x-y|^{d+sp}} dy$

Research Significance

1. Essential characteristics of nonlocal operators: Due to the nonlocal nature of the operator, problem (1.1) requires specifying values $u(x)$ on $\mathbb{R}^d \setminus \Omega$ for well-posedness
2. Limitations of existing methods: Traditional assumptions such as $g \in W^{s,p}(\mathbb{R}^d)$ impose unnatural restrictions, since problem (1.1) does not involve any regularity of $g$ in $\mathbb{R}^d \setminus \Omega$, only requiring certain weighted integrability
3. Missing continuity: Existing theory lacks continuity as $s \to 1^-$, failing to recover classical local theory

Limitations of Existing Approaches

• Homogeneous problems ($g=0$) are relatively simple and have been studied
• Treatment of nonzero data $g$ typically assumes $g$ is regular on the entire space $\mathbb{R}^d$, such as $g \in W^{s,p}(\mathbb{R}^d)$
• Compromise approaches like $g \in W^{s,p}(\Omega_\epsilon) \cap L^p(\mathbb{R}^d; (1+|x|)^{-d-sp}dx)$ remain unnatural
• As $s \to 1^-$, the classical trace space $W^{1-1/p,p}(\partial\Omega)$ cannot be recovered

Research Motivation

Construct trace spaces $T^{s,p}(\Omega^c)$ suitable for nonlocal variational methods such that:

1. Naturally characterize the regularity of external data $g$
2. Be robust with respect to parameter $s$
3. Recover classical trace and extension theorems as $s \to 1^-$

Core Contributions

1. Introduction of new trace spaces: Define function space $T^{s,p}(\Omega^c)$ (defined on $\Omega^c$) equipped with measure $\mu_s(dx) := 1_{\Omega^c}(x)(1-s)d_x^{-s}(1+d_x)^{-d-s(p-1)}dx$ where $d_x := \text{dist}(x, \partial\Omega)$
2. Proof of robust trace theorem (Theorem 1.2, $p>1$): The trace operator $\text{Tr}_s: V^{s,p}(\Omega|\mathbb{R}^d) \to T^{s,p}(\Omega^c)$ is continuous linear, and there exists a continuous linear right inverse (extension operator) $\text{Ext}_s: T^{s,p}(\Omega^c) \to V^{s,p}(\Omega|\mathbb{R}^d)$, with continuity constants depending only on $\Omega$, lower bounds of $s$, and upper and lower bounds of $p$
3. Extension to $p=1$ case (Theorem 1.3): Prove that the trace operator $\text{Tr}_s: V^{s,1}(\Omega|\mathbb{R}^d) \to L^1(\Omega^c; \mu_s)$ is continuous, and there exists a continuous linear right inverse $\text{Ext}_s: T^{s,1}(\Omega^c) \to V^{s,1}(\Omega|\mathbb{R}^d)$
4. Limiting behavior theorem (Theorem 1.4): Prove that as $s \to 1^-$:
   • $\|\text{Tr}_s u\|_{L^p(\Omega^c;\mu_s)} \to \|\gamma u\|_{L^p(\partial\Omega)}$ ($u \in W^{1,p}(\mathbb{R}^d)$)
   • $[\text{Tr}_s u]_{T^{s,p}(\Omega^c)} \to [\gamma u]_{W^{1-1/p,p}(\partial\Omega)}$ ($u \in W^{1,p}(\mathbb{R}^d)$)
   • Recover classical trace spaces and Besov spaces
5. Application to Dirichlet problems (Corollary 1.7): Using the new trace and extension theory, establish well-posedness and energy estimates for problem (1.1)

Detailed Methodology

Task Definition

Study the mapping from energy space $V^{s,p}(\Omega|\mathbb{R}^d)$ to trace spaces on exterior domain $\Omega^c$: $V^{s,p}(\Omega|\mathbb{R}^d) := \{u: \mathbb{R}^d \to \mathbb{R} \text{ measurable} \mid [u]_{V^{s,p}(\Omega|\mathbb{R}^d)} < \infty\}$ where the seminorm is defined as: $[u]_{V^{s,p}(A|B)}^p := (1-s) \iint_{A \times B} \frac{|u(x)-u(y)|^p}{|x-y|^{d+sp}} dx dy$

Construction of Trace Spaces

Key Innovation: Design of measure $\mu_s$$\mu_s(dx) = 1_{\Omega^c}(x)(1-s)d_x^{-s}(1+d_x)^{-d-s(p-1)}dx$

This measure design has the following characteristics:

1. Behavior near boundary: The $(1-s)d_x^{-s}$ term ensures that as $s \to 1^-$, $\mu_s$ weakly converges to the Hausdorff measure on boundary $\partial\Omega$
2. Decay at infinity: The $(1+d_x)^{-d-s(p-1)}$ term ensures appropriate decay at infinity
3. Parameter robustness: The choice of coefficient $(1-s)$ makes the measure scaling natural as $s \to 1^-$

Definition of trace space $T^{s,p}(\Omega^c)$ (equation 1.6): $T^{s,p}(\Omega^c) := \{g: \Omega^c \to \mathbb{R} \text{ measurable} \mid \|g\|_{T^{s,p}(\Omega^c)} < \infty\}$ with norm: $\|g\|_{T^{s,p}(\Omega^c)}^p := \|g\|_{L^p(\Omega^c;\mu_s)}^p + [g]_{T^{s,p}(\Omega^c)}^p$

The seminorm is defined as (for $p>1$): $[f,g]_{T^{s,p}(\Omega^c)}^p := \iint_{\Omega^c \times \Omega^c} \frac{|f(x)-f(y)|^{p-2}(f(x)-f(y))(g(x)-g(y))}{((|x-y|+d_x+d_y) \wedge 1)^{d+s(p-2)}} \mu_s(dx)\mu_s(dy)$

Key design principles:

• The denominator term $(|x-y|+d_x+d_y) \wedge 1$ becomes $(|x-y|) \wedge 1$ near the boundary, recovering the classical Slobodeckij seminorm structure
• The weight in measure $\mu_s$ compensates for the dimensional change (from $d$ dimensions to $(d-1)$ dimensional boundary)

Proof Strategy for Trace Theorem

Section 3: Proof of trace embedding

The proof is divided into two propositions:

1. Proposition 3.9: Embedding of $L^p$ part $\|\text{Tr}_s u\|_{L^p(\Omega^c;\mu_s)} \leq C \|u\|_{V^{s,p}(\Omega|\mathbb{R}^d)}$
2. Proposition 3.10: Embedding of seminorm part $[\text{Tr}_s u]_{T^{s,p}(\Omega^c)} \leq C \|u\|_{V^{s,p}(\Omega|\mathbb{R}^d)}$

Key technical tools:

(a) Approximate trace inequality (Theorem 3.5): For $u \in W^{s,p}(\Omega)$, prove $\int_{\Omega} |u(x)|^p \tau_s(dx) + \int_{\Omega} \int_{\Omega} \frac{|u(x)-u(y)|^p}{((|x-y|+d_x+d_y) \wedge 1)^{d+s(p-2)}} \tau_s(dy)\tau_s(dx) \leq C \|u\|_{W^{s,p}(\Omega)}^p$

where $\tau_s(dx) = (1-s)d_x^{-s}1_{\Omega}(x)dx$

Proof method:

• Use interpolation theory of Bessel potential spaces
• Define $\alpha_0 := s(1+p)/(2p)$, $\alpha_1 := 1 + s/(2p)$
• Prove continuity for $H^{\alpha_i,p}(\mathbb{R}^d)$ (Lemma 3.4)
• Obtain result via real interpolation $[H^{\alpha_0,p}, H^{\alpha_1,p}]_\theta^p = W^{s,p}(\mathbb{R}^d)$

(b) Hardy-type inequality (Theorem 3.6, $p=1$ case): $(1-s) \int_{\Omega} \frac{|u(x)|}{d_x^s} dx \leq C\left(\|u\|_{L^1(\Omega)} + s(1-s) \int_{\Omega \times \Omega} \frac{|u(x)-u(y)|}{|x-y|^{d+s}} d(x,y)\right)$

Proof strategy:

• Use local Lipschitz property of boundary
• Apply optimal Hardy inequality on half-spaces (Theorem B.1)
• Handle global case via partition of unity

(c) Technical lemmas on distance functions:

• Lemma 3.7: Convert $d_x^{-s}$ to integrals over $\Omega$
• Lemma 3.8: Establish lower bounds using uniform interior cone condition

Proof of Proposition 3.9 ($L^p$ part):

1. Decompose $\Omega^c = \Omega_1^{\text{ext}} \cup \Omega^{\text{ext}}_1$
2. On $\Omega_1^{\text{ext}}$: Use Lemma 3.8 and Theorem 3.5/3.6
3. On $\Omega^{\text{ext}}_1$: Direct estimation using distance function lower bounds

Proof of Proposition 3.10 (seminorm part, $p>1$):

1. Decompose integral domain: $\Omega^{\text{ext}}_\rho \times \Omega^{\text{ext}}_\rho$, $\Omega^c \times \Omega^\rho_{\text{ext}}$, $\Omega^\rho_{\text{ext}} \times \Omega^c$
2. Far-field part ($\Omega^\rho_{\text{ext}}$): Use Proposition 3.9 and distance lower bounds
3. Near-field part ($\Omega^{\text{ext}}_\rho \times \Omega^{\text{ext}}_\rho$):
   • Apply Lemma 3.8 twice, introducing intermediate integration variables $z,w \in \Omega$
   • Apply Theorem 3.5 to estimate integral over $\Omega \times \Omega$
   • Key: Refined estimation of kernel function $a(z,w)$ (equation 3.19)

Proof Strategy for Extension Theorem

Section 4: Construction of extension operator

Core idea: Weighted averaging extension based on Whitney decomposition

(a) Whitney decomposition:

• Perform dyadic Whitney decomposition $\mathcal{W}(\Omega)$ of $\Omega$
• Each cube $Q$ satisfies: $\text{diam}(Q) \leq d(Q,\partial\Omega) \leq 4\text{diam}(Q)$
• Define $\mathcal{W}_{\leq\kappa}(\Omega)$ as cubes with side length $\leq 2^\kappa$, where $\kappa := \lfloor\log_2(\rho/\sqrt{d})\rfloor$, $\rho := \text{inr}(\Omega)/2 \wedge 1/2$

(b) Partition of unity: Construct smooth partition of unity $\{\phi_Q\}_{Q \in \mathcal{W}(\Omega)}$ satisfying $\sum_Q \phi_Q = 1_\Omega$

(c) Measure $\tilde{\mu}_s$: Define simplified measure (without decay at infinity): $\tilde{\mu}_s(dz) = 1_{\Omega^c}(z)\frac{1-s}{d_z^s} dz$ and normalization constant: $a_{Q,s} := \left(\tilde{\mu}_s(B_{6l_Q}(q_Q))\right)^{-1}$

Lemma 4.1 proves: $C^{-1}l_Q^{s-d} \leq a_{Q,s} \leq C l_Q^{s-d}$

(d) Definition of extension operator (equation 4.11):