Fractional Laplacians: Definitions, Analysis & Applications

• Fractional Laplacians are nonlocal operators defined via Fourier, Riesz, and spectral methods that capture both analytic and probabilistic properties across various domains.
• They exhibit distinct boundary behaviors and underpin models in anomalous diffusion, image processing, and fluid dynamics.
• Numerical schemes like finite differences, FEM/SEM, and Monte Carlo methods address the challenges from their singular kernels and infinite stencil structures.

The fractional Laplacian $(-\Delta)^s$, for real $s>0$, is a canonical nonlocal operator of central importance in analysis, probability, partial differential equations, mathematical physics, numerical methods, and geometry. Its rigorous definition, analytic properties, distinct variants, and nonlocal boundary phenomena have far-reaching implications for theory and applications.

1. Definitions and Core Variants

The fractional Laplacian is primarily defined on $\mathbb{R}^n$ via three principal formulations, which are equivalent in the whole-space setting:

• Fourier Multipliers: For $u\in S(\mathbb{R}^n)$,

$$\mathcal{F}[(-\Delta)^s u](\xi) = |\xi|^{2s}\, \mathcal{F}[u](\xi).$$

• Singular Integral (Riesz) Form:

$$(-\Delta)^s u(x) = C_{n,s}\, \mathrm{P.V.} \int_{\mathbb{R}^n} \frac{u(x) - u(y)}{|x-y|^{n+2s}}\, dy,$$

where $$C_{n,s} = 2^{2s} s\,\frac{\Gamma(s+\frac{n}{2})}{\pi^{n/2} \Gamma(1-s)}$$ for $0 < s < 1$.

• Spectral (Functional Calculus) Definition: On a domain $\Omega\subset \mathbb{R}^n$, using eigenpairs $\{\lambda_j,\varphi_j\}$ of $s>0$0 with specified boundary condition,

$s>0$1

The operator domain is determined by the $s>0$2-based expansion.

On bounded domains, the above general framework differentiates into several inequivalent "fractional Laplacians" due to the treatment of boundary conditions and operator domains (Lischke et al., 2018, Nazarov, 2021, Musina et al., 2013):

Operator Type	Definition/Domain	Associated Process/Boundary
Restricted (integral/Dirichlet)	Riesz form, $s>0$3 on $s>0$4	Killed Lévy flight (absorbing)
Spectral (Dirichlet/Neumann)	Spectral calculus on $s>0$5 with local BCs on $s>0$6	Subordinate killed Brownian
Regional/Censored	Riesz integral restricted to $s>0$7	Reflected/censored process
Negative-order/dynamic	Extension/dual variational methods	Potential theory, inverse flows

In non-Euclidean settings, further generalizations include the Laplace–Beltrami setting, graph (discrete) Laplacians, and Laplacians on smooth bundles (Chien, 2022, Gerontogiannis et al., 6 Feb 2025, Benzi et al., 2019).

2. Functional Analytical and Probabilistic Structure

Fractional Laplacians are self-adjoint, nonlocal, and generate Dirichlet forms of the type

$s>0$8

On compact domains, the form domain and the corresponding function spaces differ depending on the variant: for the restricted case, $s>0$9 supported in $\mathbb{R}^n$0; for the spectral case, $\mathbb{R}^n$1 in the appropriate $\mathbb{R}^n$2 space respecting the boundary conditions (Nazarov, 2021, Musina et al., 2013).

Stochastically, the restricted Laplacian arises as generator of a killed symmetric $\mathbb{R}^n$3-stable process, whereas the spectral Laplacian comes from subordinating a stopped Brownian motion (Garbaczewski, 2018, Ivrii, 2016, Lischke et al., 2018). The regional variant is governed by censored (or reflected-and-killed) stable processes, giving rise to invariant measures distinct from the restricted case (Borthagaray et al., 2021, Garbaczewski, 2018). The distinction is due to the non-commutativity of stopping and subordination in stochastic process construction.

3. Boundary Behavior, Comparison, and Extension Principles

Boundary handling is central to understanding the analytic and probabilistic behavior of fractional Laplacians (Nazarov, 2021, Musina et al., 2013, Musina et al., 2014).

• Boundary singularities: In the restricted (integral/Riesz) case, solutions near $\mathbb{R}^n$4 exhibit singularities of the form $\mathbb{R}^n$5 for $\mathbb{R}^n$6, whereas spectral solutions are smoother up to the boundary.
• Comparison of Operators: On convex domains and for $\mathbb{R}^n$7, the spectral operator is strictly greater than the restricted one on non-negative functions: for $\mathbb{R}^n$8, $\mathbb{R}^n$9 (in the distributional sense) (Nazarov, 2021, Musina et al., 2013, Nazarov, 2021).
• Extension Problems: The Caffarelli–Silvestre extension and its duals (Stinga–Torrea) realize the (positive or negative order) fractional Laplacian as a Dirichlet-to-Neumann (resp. Neumann-to-Dirichlet) map for a weighted elliptic PDE in one higher dimension (Musina et al., 2014). This connection is leveraged both analytically and numerically.

4. Spectral Properties, Weyl Laws, and Explicit Solutions

Spectral theory for fractional Laplacians on bounded domains is rich and subtle:

• Spectra and Eigenfunctions: Both restricted and spectral Laplacians have purely discrete spectra on bounded domains. The eigenvalues differ quantitatively, leading to different dynamics under equations such as the fractional heat equation (Ivrii, 2016, Dyda, 2011).
• Weyl Asymptotics: Two-term asymptotic expansions have been established for eigenvalue counting functions, extending classical Weyl laws to nonlocal operators. The second term captures boundary corrections and is relevant for heat kernel expansions and probabilistic exit times (Ivrii, 2016).
• Explicit Solutions and Harmonic Functions: Closed-form expressions for the fractional Laplacian of special functions—polynomials, Hermite/Laguerre functions, Zernike-type polynomials—are available (Gutleb et al., 2023, Dyda, 2011, Abatangelo et al., 2020). Harmonic and torsion functions for the operator can be constructed explicitly on balls and ellipsoids, with notable applications in constructing counterexamples to the weak maximum principle and benchmarking numerical schemes (Abatangelo et al., 2020, Gutleb et al., 2023).

5. Numerical Schemes and Computational Considerations

Numerical analysis for fractional Laplacians is nontrivial due to the operator's nonlocality and singular kernel. Contemporary approaches include:

• Finite Difference Discretizations: Grünwald–Letnikov, spectral, regularized central difference, and quadrature-difference schemes each provide a discretization with differing accuracy and nonlocality. Full infinite stencils arise, and the kernel decays algebraically as $u\in S(\mathbb{R}^n)$0 (Huang et al., 2016).
• FEM/SEM and Matrix Approaches: For the spectral variant, spectral element or finite element methods can leverage the operator's diagonal structure in eigenbasis, with extension problems (as in Caffarelli–Silvestre) handled by tensorized or hierarchic methods (Borthagaray et al., 2021, Lischke et al., 2018).
• Monte Carlo Methods: Walk-on-spheres algorithms are effective for the Riesz (restricted) operator, particularly for high-dimensional or complex geometries (Lischke et al., 2018).
• Preconditioners: Robust multilevel (BPX-type) preconditioners have been constructed for the integral, spectral, and regional Laplacians, yielding condition number bounds independent of the mesh refinement and the fractional power $u\in S(\mathbb{R}^n)$1 (Borthagaray et al., 2021).
• Frame/Spectral Bases: Development of explicit spectral/fame bases adapted to the operator enables efficient matrix representations and rapid convergence for smooth data (Gutleb et al., 2023).

6. Applications, Extensions, and Open Directions

Fractional Laplacians appear in a diverse array of applications:

• Anomalous Diffusion, Fluid Dynamics, Image Processing: Equations involving $u\in S(\mathbb{R}^n)$2 model nonlocal diffusion, turbulent transport, interface motion, and anomalous dissipation (Cordoba et al., 2015, Lischke et al., 2018).
• Random Walks and Networks: Fractional graph Laplacians underlie nonlocal dynamics on networks and have direct interpretations via random walks with long jumps, superdiffusion, and consensus protocols in multi-agent systems (Benzi et al., 2019).
• Geometry and Inverse Problems: Fractional Laplace–Beltrami operators, and their noncommutative/fibered analogues (fractional connection Laplacians), connect to geometric and analytic inverse problems—global data from fractional source-to-solution maps determine geometry, bundles, and connections (Chien, 2022, Gerontogiannis et al., 6 Feb 2025).
• Nonhomogeneous and Generalized Operators: Inhomogeneous media lead to generalized fractional Laplacians, with divergence structure involving nonconstant coefficients via Riesz potentials (Zheng et al., 2023).
• Quantum Metrics and Noncommutative Geometry: Quantum Monge–Kantorovič metrics based on commutators with fractional Laplacians provide a link to the noncommutative geometry of fractals, group $u\in S(\mathbb{R}^n)$3-algebras, and quantum groups (Gerontogiannis et al., 6 Feb 2025).

Open problems highlighted in recent literature include a full characterization of inequalities and extremal functions for spectral forms under modulus, explicit control of boundary layers for all operator types, further development of fast and stable high-dimensional discretizations, and extensions to operators on general metric-measure spaces and noncommutative algebras (Nazarov, 2021, Gerontogiannis et al., 6 Feb 2025).

7. Tabular Comparison of Representative Fractional Laplacians

Operator Name	Kernel/Domain	Boundary	Stochastic Process
Riesz (integral)	$u\in S(\mathbb{R}^n)$4 (whole space)	Requires $u\in S(\mathbb{R}^n)$5 outside $u\in S(\mathbb{R}^n)$6 (volume constraint)	Symmetric $u\in S(\mathbb{R}^n)$7-stable process
Restricted Dirichlet	Riesz as above, with $u\in S(\mathbb{R}^n)$8 outside $u\in S(\mathbb{R}^n)$9	Nonlocal Dirichlet	Killed Lévy flight
Spectral Dirichlet	Spectral expansion in Dirichlet eigenbasis	Local Dirichlet	Subordinate killed Brownian motion
Regional (censored)	Riesz kernel restricted to $$\mathcal{F}[(-\Delta)^s u](\xi) = |\xi|^{2s}\, \mathcal{F}[u](\xi).$$0	"Nonlocal Neumann"/no explicit BC	Reflected/censored process

• (Musina et al., 2013) Musina, Nazarov, "On fractional Laplacians"
• (Borthagaray et al., 2021) BPX preconditioners for fractional Laplacians
• (Nazarov, 2021) Nazarov, "Variety of fractional Laplacians"
• (Lischke et al., 2018) G. Acosta et al., "What Is the Fractional Laplacian?"
• (Ivrii, 2016) Ivrii, "Spectral Asymptotics for Fractional Laplacians"
• (Dyda, 2011) Dyda, "Fractional calculus for power functions"
• (Huang et al., 2016) Huang, Oberman, "Finite difference methods for fractional Laplacians"
• (Garbaczewski, 2018) Kaleta, Lörinczi, "Fractional Laplacians and Levy flights in bounded domains"
• (Abatangelo et al., 2020) Abatangelo et al., "Fractional Laplacians on ellipsoids"
• (Gutleb et al., 2023) Gutleb, Papadopoulos, "Explicit fractional Laplacians and Riesz potentials"
• (Benzi et al., 2019) Estrada et al., "Nonlocal network dynamics via fractional graph Laplacians"
• (Chien, 2022) Chien, "An inverse problem for fractional connection Laplacians"
• (Gerontogiannis et al., 6 Feb 2025) Gerontogiannis, Mesland, "Ideal quantum metrics from fractional Laplacians"
• (Zheng et al., 2023) Borthwick, Nistor, Zeng, "A generalized fractional Laplacian"
• (Cordoba et al., 2015) Córdoba, Martínez "A pointwise inequality for fractional laplacians"