Integral Fractional Laplacian Problems

• Integral fractional Laplacian problems are defined by a nonlocal operator with significant roles in anomalous diffusion, probability, and PDE theory.
• They employ representations like the Caffarelli–Silvestre extension to handle analytic regularity and capture boundary singularities in polygonal and polyhedral domains.
• The detailed regularity framework drives advanced numerical methods, including hp-FEM and adaptive mesh refinement for exponential convergence.

The integral fractional Laplacian is a nonlocal, singular operator arising in numerous contexts including anomalous diffusion, probability, PDE theory, and nonlocal mechanics. Its study on bounded domains, especially with analytic regularity, boundary singularities, and geometric complexity (vertices, edges, faces), has motivated the development of sophisticated theoretical and numerical frameworks.

1. Definition and Fundamental Representations

For a sufficiently smooth function $u$ on $\mathbb{R}^d$ and order $s\in(0,1)$, the integral fractional Laplacian is defined by

$$(-\Delta)^s u(x) = C(d,s)\;\mathrm{P.V.}\int_{\mathbb{R}^d} \frac{u(x)-u(z)}{|x-z|^{d+2s}}\,dz,$$

with $$C(d,s) = -2^{2s}\Gamma(s+d/2)/[\pi^{d/2}\Gamma(-s)]$$ in canonical normalization (Faustmann et al., 2021).

An equivalent local representation employs the Caffarelli–Silvestre extension: for $\alpha = 1-2s$,

$$\begin{cases} -\mathrm{div}(y^\alpha \nabla U) = 0 & \text{in }\Omega \times (0,\infty),\ U = 0 & \text{on }\partial\Omega \times [0,\infty),\ U(x,0) = u(x) & \text{on } \Omega, \end{cases}$$

with $$(-\Delta)^s u(x) = -d_s \lim_{y\to0^+} y^\alpha \partial_y U(x,y)$$ for $d_s = 2^{2s-1}\Gamma(s)/\Gamma(1-s)$ (Faustmann et al., 2021, Faustmann et al., 2023).

In polygonal or polyhedral domains, explicit distance-weight functions ($r_v(x)$, $r_e(x)$) are used to capture singularity structure at vertices, edges, and faces.

2. Weighted Analytic Regularity and Geometric Localization

The main technical advance is a sharp characterization of weighted analytic regularity for solutions to the Dirichlet problem in polygons (and their 3D extensions in polyhedra). For analytic right-hand-side $f$ with derivative bounds of the form $$\sum_{|\gamma|=m}\|\partial^\gamma f\|_{L^2} \le \gamma_f^{m+1} m^m$$, the extension solution $U$ satisfies, for any multi-index $\beta$ of order $p$ and for each geometric neighborhood (vertex, edge-vertex, edge, and interior core), weighted $L^2$ bounds of the type (Faustmann et al., 2021):

• In a vertex cone $\omega_v$:

$$\left\| r_v^{p-\frac{1}{2}-s+\varepsilon} D_x^\beta U \right\|_{L^2_\alpha(\omega_v^+)} \le C_\varepsilon \gamma^{p+1} p^p,$$

• In an edge-vertex wedge $\omega_{ve}$:

$$\left\| r_e^{-\frac{1}{2}-s+\varepsilon} r_v^{p+\varepsilon} D_{x_\perp} D_{x_\parallel}^{\beta} U \right\|_{L^2_\alpha((\omega_{ve})^{H/4})} \le C_\varepsilon \gamma^{p+1} p^p,$$

• Pure edge and interior neighborhood estimates follow similarly.

Passing to the trace $y\to 0$, analogous bounds hold for $u$ itself.

Crucial elements of the proof include: localization via the extension formulation, difference-quotient regularity shifts, local weighted Caccioppoli inequalities, dyadic coverings of singular sets, and bootstrapping arguments that systematically iteratively amplify derivative bounds up to analytic (factorial-growth) rate.

These results have been extended to polyhedra in $\mathbb{R}^3$, with further partitioning of neighborhoods (vertex-edge-face etc.), maintaining analytic control in weighted Sobolev norms dictated by distance to the singular set (Faustmann et al., 2023).

3. Structure of Weighted Sobolev and Analytic Spaces

The analytic regularity theory is framed within weighted Sobolev and analytic spaces reflective of boundary geometry:

• Distance weights: $r_v(x) = \operatorname{dist}(x, v)$ for vertices $v$, $r_e(x) = \operatorname{dist}(x, e)$ for edges $e$.
• Domains are covered by overlapping zones: vertex cones, edge-vertex wedges, pure edge bands, and interior cores (Faustmann et al., 2021).
• Tangential derivatives $D_{x_\parallel}$ and normal $D_{x_\perp}$ are distinguished in edge and edge-vertex regions.
• Weighted analytic norms take the form $$\| r_v^{\mu+\beta} \partial_\tau^{\beta} U \|_{L^2(D)}$$ or more generally in the extension $$\| \cdot \|_{L^2_\alpha(D)} = \int_D y^\alpha |\cdot|^2 dx\,dy$$.

Factors defining analytic data (factorial growth in derivatives), and factorial/Gevrey bounds for solution derivatives, underpin the deep regularity structure.

4. Proof Strategy: Localization, Caccioppoli and Bootstrapping

The proof methodology includes:

• Reformulation as a local extension problem in $\Omega \times (0,\infty)$.
• Difference-quotient arguments providing $H^t$ shifts in tangential regularity.
• Weighted, localized Caccioppoli inequalities that control second derivatives on scaled balls/half-balls.
• Dyadic, geometric coverings that ensure locally finite overlap and permit iterative bootstrapping.
• Bootstrapping procedure for high-order bounds, with combinatorial control yielding explicit factorial growth constants.
• Trace-back of high-order derivative bounds using trace and Hardy inequalities, transferring control from $U$ to $u$.

These analytic regularity results enable tight a priori bounds for “hp”-finite element methods and mesh design in polygonal domains.

5. Key Formulas and Constants

Several explicit formulas are central:

• Fractional Laplacian: $$(-\Delta)^s u(x) = C(d,s)\,\mathrm{P.V.} \int (u(x)-u(z))|x-z|^{-(d+2s)}\,dz$$.
• Extension PDE: $-\mathrm{div}(y^\alpha \nabla U) = 0$ in $\Omega \times (0,\infty)$, $U|_{y=0} = u$, $U|_{\partial\Omega \times [0,\infty)} = 0$, $\alpha = 1-2s$.
• Weighted norm: $$\| v \|^2_{L^2_\alpha(D)} = \int_D y^\alpha |v|^2 dx\,dy$$ for $D \subset \Omega \times (0,\infty)$.
• Tangential derivatives: $D_{x_\parallel} = e_\parallel \cdot \nabla_x$, $D_{x_\perp} = e_\perp \cdot \nabla_x$.
• Analytic norm (data): $$\sum_{|\gamma|=m} \|\partial^\gamma f\|_{L^2(\Omega)} \le \gamma_f^{m+1} m^m$$, norm $$= \sup_m \gamma_f^{-m-1} m^{-m} \sum_{|\gamma|=m} \|\partial^\gamma f\|_{L^2}$$.
• Error bounds (vertex-edge case, $p=j+k$): $$\| r_e^{-1/2-s+\varepsilon} r_v^{p+\varepsilon} D_{x_\perp} D_{x_\parallel}^p u \|_{L^2(\omega_{ve})} \le C_\varepsilon \gamma^{p+1} p^p \| f \|_{analytic}$$.

Constants $\gamma$ depend only on data and geometry, with exponential convergence implications for “hp”-FEM strategies.

6. Implications for Numerical Methods and Further Applications

The weighted analytic regularity theory supplies explicit, constructive bounds for:

• Mesh grading and local polynomial degree allocation in finite element algorithms, enabling exponential convergence $O(\exp(-\sigma N^{1/4}))$ in total degrees of freedom $N$ (Faustmann et al., 2021).
• Adaptive mesh refinement steered by singularity structure tied to geometric features (vertices, edges).
• Boundary element method development suitable for high-order, nonlocal problems in polygons and polyhedra.

The analytic regularity framework also underpins further developments in nonlocal elliptic theory, regularity bootstrapping for PDE systems, and a posteriori error analysis in fractional elliptic problems on domains with singular boundaries. These advances form the technical foundation for high-accuracy and complexity-optimal solvers in integral fractional Laplacian problems.