1 Definition and Basic Properties

1.1 Fourier-transform definition on \(\mathbb{R}^n\)

For \(0<s<1\) and a sufficiently regular function \(u\) on \(\mathbb{R}^n\) (e.g., \(u\in \mathcal{S}(\mathbb{R}^n)\), the Schwartz class), the fractional Laplacian is defined by \[

(-\Delta)^s u \;=\; \mathcal{F}^{-1}\!\big(\xi^{2s}\,\widehat{u}(\xi)\big),

\] where \(\widehat{u}\) denotes the Fourier transform and \(\mathcal{F}^{-1}\) its inverse. This definition extends by density to broader classes of functions using appropriate Sobolev-space frameworks.

1.2 Singular-integral (kernel) formulation

On \(\mathbb{R}^n\), for \(0<s<1\) and suitable \(u\), one can write \[ (-\Delta)^s u(x)

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

\]

where P.V. denotes the Cauchy principal value. The integrand reflects a balance between the difference \(u(x)-u(y)\) and the long-range weight \(x-y^{-(n+2s)}\).

1.3 Normalization constants and conventions

The constant \(C_{n,s}\) depends on the chosen convention and ensures that the Fourier and integral definitions agree. A common choice is \[ C_{n,s}=\frac{2^{2s}s\,\Gamma\!\left(\frac{n}{2}+s\right)}{\pi^{n/2}\Gamma(1-s)}. \] Different texts may use equivalent constants that shift between conventions for the Fourier transform normalization and the integral kernel.

1.4 Basic symmetry, scaling, and homogeneity

The operator is translation-invariant and rotation-invariant on \(\mathbb{R}^n\). It is also homogeneous under scaling: if \(u_\lambda(x)=u(\lambda x)\), then \ (-\Delta)^s u_\lambda (x)=\lambda^{2s}\, [(-\Delta)^s u, \] capturing the fact that it behaves like an operator of order \(2s\).

1.5 Nonlocality and comparison with the classical Laplacian

In contrast to the classical Laplacian \(-\Delta u(x)\), which depends on the local behavior of \(u\) near \(x\), the fractional Laplacian involves contributions from all points in \(\mathbb{R}^n\) (or all points outside a domain in bounded-domain settings). The kernel’s polynomial decay means far-field influence remains present, though it becomes weaker with distance.

2 Connections with Function Spaces

2.1 Fractional Sobolev spaces \(H^s(\mathbb{R}^n)\)

For \(s\in(0,1)\), the fractional Sobolev space \(H^s(\mathbb{R}^n)\) can be defined via the Fourier transform: \[

\|u\|_{H^s}^2 = \int_{\mathbb{R}^n}(1+\xi^2)^s\,\widehat{u}(\xi)^2\,d\xi.

\] Equivalent descriptions exist using energy forms and Gagliardo seminorms, which are especially aligned with the singular-integral nature of \((-\Delta)^s\).

2.2 Energy form and Gagliardo seminorm

A central quantity is the Gagliardo seminorm: \[ [u]_{H^s(\mathbb{R}^n)}^2

=\int_{\mathbb{R}^n}\int_{\mathbb{R}^n}\frac{u(x)-u(y)^2}{x-y^{n+2s}}\,dx\,dy.

\] This expresses the “nonlocal variation” of \(u\) and is closely tied to the quadratic form associated with \((-\Delta)^s\).

2.3 Mapping properties and regularity expectations

In operator-theoretic terms, \((-\Delta)^s\) maps (in a weak sense) \(H^s(\mathbb{R}^n)\) to \(H^{-s}(\mathbb{R}^n)\), reflecting its order \(2s\). Regularity upgrades depend on data and domain geometry: when the right-hand side is smoother, solutions typically gain fractional derivatives, though boundary effects can limit smoothness.

2.4 Weak formulations and variational setting

A common variational approach uses the bilinear form \[ \mathcal{E}(u,v)=\frac{C_{n,s}}{2}\int_{\mathbb{R}^n}\int_{\mathbb{R}^n}

\frac{(u(x)-u(y))(v(x)-v(y))}{x-y^{n+2s}}\,dx\,dy.

\] One seeks \(u\) such that \(\mathcal{E}(u,\varphi)=\langle f,\varphi\rangle\) for test functions \(\varphi\). This framework accommodates nonsmooth functions and makes existence and uniqueness arguments possible.

2.5 Fractional Laplacian as an operator between Sobolev spaces

Within the weak framework, \((-\Delta)^s\) can be realized as the operator associated with \(\mathcal{E}\). For appropriate classes of \(u\), the identity \(\langle (-\Delta)^s u,\varphi\rangle=\mathcal{E}(u,\varphi)\) holds, where \(\langle\cdot,\cdot\rangle\) denotes duality between \(H^{-s}\) and \(H^s\). This clarifies how the operator acts even when pointwise evaluation is not meaningful.

3 Spectral Theory and Alternative Realizations

3.1 Spectral definition for bounded domains

On a bounded domain \(\Omega\subset\mathbb{R}^n\), one can define an operator via the Dirichlet Laplacian \(-\Delta\) on \(\Omega\). If \(\{\phi_k\}\) are Dirichlet eigenfunctions with eigenvalues \(\{\lambda_k\}\), then for suitable \(u=\sum_k a_k\phi_k\), \[ (-\Delta)^s_{\Omega} u = \sum_k a_k\,\lambda_k^s\,\phi_k. \] This “spectral fractional Laplacian” emphasizes boundary conditions encoded through eigenfunctions.

3.2 Relation to eigenfunction expansions

The spectral representation ties regularity and behavior to the decay of coefficients \(a_k\) and growth of \(\lambda_k\). It also facilitates estimates, since \(\lambda_k^s\) scales eigenmodes by the fractional power.

3.3 Semigroup viewpoint (subordination) basics

Another realization uses semigroups. Let \(T(t)=e^{t\Delta}\) be the heat semigroup (with appropriate boundary conditions on \(\Omega\)). Fractional powers can be expressed through subordination: \[ (-\Delta)^s u \quad \text{is related to} \quad \int_0^\infty \big(u - T(t)u\big)\,\mu_t^{(s)}\,dt, \] where \(\mu_t^{(s)}\) is a probability density associated with stable subordinators. This perspective highlights how the fractional operator is built from repeated applications of the classical diffusion semigroup.

The integral form corresponds to the generator of a symmetric \(2s\)-stable Lévy process. Informally, such a process makes jumps rather than only moving continuously. The fractional Laplacian then measures the infinitesimal effect of these jumps on observables, explaining the nonlocal kernel and the long-range interactions.

4 Fundamental Solutions and Green’s Functions

4.1 Whole-space fundamental solution

In \(\mathbb{R}^n\), the fractional Laplacian has an explicit fundamental solution for suitable \(s\). Formally, the Riesz kernel yields \[

(-\Delta)^s \big(x^{-(n-2s)}\big)\propto \delta_0,

\] with proportionality constant depending on \(n\) and \(s\). This mirrors how the classical Laplacian’s fundamental solution produces the Newtonian kernel.

4.2 Riesz potential perspective

Associated with \((-\Delta)^s\) is the Riesz potential \(I_\alpha\), which acts like convolution with \(x^{-(n-\alpha)}\) for appropriate \(\alpha\). In many settings, \((-\Delta)^s\) and \(I_{2s}\) are inverses up to normalization on suitable function classes, giving an effective tool for representing solutions of nonlocal Poisson-type equations.

4.3 Heat kernel and long-time behavior heuristics

The fractional heat equation \(\partial_t u + (-\Delta)^s u=0\) has a kernel with heavy tails compared with the Gaussian heat kernel. Heuristically, large jumps correspond to slower spatial decay, while temporal decay follows scaling consistent with order \(2s\). These qualitative features influence how solutions disperse over time.

4.4 Boundary-influenced kernels in bounded domains

In bounded domains, Green’s functions and Poisson kernels depend strongly on how the fractional operator encodes the boundary. Different “fractionalizations” (spectral vs. integral with exterior conditions) lead to different kernel behavior near \(\partial\Omega\), affecting both magnitude and regularity of solutions.

5 Boundary Value Problems

5.1 Dirichlet-type problems (whole-space formulation)

For the integral (kernel-based) fractional Laplacian on \(\Omega\), a standard Dirichlet-type problem prescribes \(u\) on \(\mathbb{R}^n\setminus\Omega\). For instance, one sets \(u=g\) outside \(\Omega\) and solves for \(x\in\Omega\) in \[ (-\Delta)^s u = f \quad \text{in } \Omega, \] where the nonlocal operator couples values from outside to inside through the tail of the kernel.

5.2 “Spectral” versus “integral” notions on domains

On bounded domains, the operator defined spectrally generally differs from the integral-defined operator with exterior data. These differences manifest in boundary regularity, energy spaces, and maximum principles. As a result, one must specify which definition is intended when stating boundary value problems.

5.3 Exterior Dirichlet conditions and truncation effects

Because the operator samples values across the complement of \(\Omega\), changing exterior data can alter the interior solution even far from the boundary. Truncating the integral (for numerical methods or approximations) introduces errors that depend on how much of the kernel’s far tail is neglected.

5.4 Regularity near the boundary and boundary layers (overview)

Solutions typically exhibit reduced regularity at the boundary compared with the interior. A common feature is boundary layer behavior: the solution may scale like a power of the distance to \(\partial\Omega\), with exponents depending on \(s\) and on the boundary condition type. These effects are central in proving Hölder or Sobolev regularity results.

Nonlocal operators admit maximum principles with appropriate assumptions. For example, if \((-\Delta)^s u \ge 0\) in \(\Omega\) and \(u\) satisfies a nonnegative exterior condition (in the integral sense), then \(u\) may be constrained from below inside \(\Omega\). Inequalities are formulated using the nonlocal structure and often require careful handling of the integral operator’s contributions across the boundary.

6 Calculus with the Fractional Laplacian

6.1 Product rules and commutator estimates (high-level)

Unlike the classical Laplacian, there is no simple identity for \((-\Delta)^s(uv)\) in terms of \((-\Delta)^s u\) and \((-\Delta)^s v\) alone. Instead, one uses fractional Leibniz rules and commutator estimates that bound the difference between the fractional Laplacian of a product and combinations of fractional Laplacians of factors. These results are key in nonlinear PDE analysis.

6.2 Chain rules for smooth nonlinearities (overview)

For compositions \(F(u)\) with smooth \(F\), there are inequalities controlling \((-\Delta)^s(F(u))\) through derivatives of \(F\) and fractional seminorms of \(u\). Such chain-rule-type statements typically take the form of estimates rather than exact equalities, reflecting the operator’s nonlocal character.

6.3 Integration by parts formulas (weak setting)

In the weak framework, integration by parts is expressed through the energy form \(\mathcal{E}\). For appropriate \(u\) and test functions \(v\), \[ \langle (-\Delta)^s u, v\rangle = \mathcal{E}(u,v), \] which replaces the classical divergence theorem. This identity supports variational methods and allows manipulation of weak solutions without requiring pointwise derivatives.

6.4 Fractional gradient/divergence analogues (overview)

Because the operator is not local, there is no universally adopted “fractional gradient” whose divergence exactly reproduces \((-\Delta)^s\). However, one can define nonlocal gradient-like quantities in bilinear forms or via extension methods, where the fractional operator emerges as a boundary trace of a local degenerate elliptic problem. These constructions are useful for deriving identities and estimates.

7 Computation and Numerical Approximation

7.1 Discretization of singular integrals

For the integral definition, numerical schemes must handle the singularity at \(y=x\) and the long-range tail. Common approaches include splitting the integral into near-field and far-field regions, using specialized quadrature for the singular part, and truncating the far-field contribution with error control.

7.2 Spectral methods via eigenfunction expansions

When eigenpairs of the Dirichlet Laplacian are available (or approximated), one can compute the fractional operator by applying \(\lambda_k^s\) to expansion coefficients. This can be efficient for smooth solutions and geometries where eigenfunctions are tractable, but it may be costly in high dimensions or complex domains.

7.3 Extension method (Caffarelli–Silvestre) as a computational tool

A practical technique is the extension method: one embeds the problem into a higher-dimensional degenerate elliptic PDE such that \((-\Delta)^s\) appears as a weighted normal derivative at the boundary of the extended domain. Numerically, this converts a nonlocal operator into a local problem with variable coefficients in one extra dimension, often improving stability and facilitating standard finite element discretizations.

7.4 Truncation, quadrature, and error considerations (overview)

All numerical methods must address approximation errors: truncation of the infinite domain, quadrature error near the singularity, and discretization error from grids or basis truncations. Error behavior typically depends on \(s\), solution regularity, and how the kernel tail is handled.

7.5 Benchmark examples and convergence diagnostics

To validate algorithms, practitioners use benchmark problems with known explicit solutions or high-accuracy reference computations. Convergence is assessed by comparing norms of the numerical error against mesh refinement and by monitoring how errors change with fractional order \(s\).

8 Applications and Model Equations

8.1 Fractional diffusion (anomalous transport) models

Fractional Laplacian operators often appear in diffusion models where transport is not governed by standard Brownian motion. The resulting dynamics show anomalous spreading: profiles evolve with rates and spatial decay different from classical diffusion, reflecting the influence of long jumps.

8.2 Fractional reaction–diffusion equations

Combining \((-\Delta)^s\) with reaction terms yields fractional reaction–diffusion systems that can model pattern formation and propagation where both local growth/decay and nonlocal transport coexist. Analysis frequently relies on variational formulations, maximum principles, and energy estimates.

In quantum-mechanical analogues, replacing the classical kinetic term by a fractional power of the Laplacian leads to Lévy-flight or nonlocal dispersive behavior. At a conceptual level, these models capture different dispersion mechanisms and correspond to generators of stable processes.

8.4 Nonlocal elliptic problems and variational applications

Many stationary nonlocal equations, such as fractional Poisson problems and fractional semilinear elliptic equations, admit a variational structure. The associated energy functional typically involves the Gagliardo seminorm, enabling the use of critical point theory and compactness arguments under suitable assumptions.

8.5 Common modeling assumptions and parameter choices

Modeling typically specifies \(s\) to tune jump intensity and decay rate of the kernel. Parameter selection depends on scaling laws, experimental fit, or nondimensionalization. Assumptions about boundary conditions—whole-space, exterior Dirichlet data, or spectral boundary encoding—must align with the physical meaning of how far-field interactions are treated.