Nonlocal Operators in Analysis and PDEs

• Nonlocal operators are integral operators whose action at a point depends on values over a region, effectively modeling long-range interactions and memory effects.
• They naturally arise in applications such as Lévy processes, anomalous diffusion, peridynamics, and fractional calculus, impacting nonlocal boundary value problems.
• Their rigorous analysis uses variational methods in Hilbert and Banach spaces to establish well-posedness, regularity, and to design efficient computational algorithms.

A nonlocal operator is a (typically integral) operator whose action at a point depends on values of a function over a region, rather than solely at that point or its infinitesimal neighborhood. Such operators naturally arise in the analysis of Lévy processes, anomalous diffusion, peridynamics, fractional calculus, and nonlocal boundary value problems. In analysis and PDE theory, nonlocal operators generalize differential operators, capturing long-range interactions, jumps, or memory effects. Their study encompasses a variety of settings: from classical fractional Laplacians to anisotropic and nonsymmetric forms, nonlinear and data-driven frameworks, and extends to the design of fast algorithms for computational implementation.

1. Integral Formulations and Structural Classes

A general linear nonlocal operator acting on scalar functions is defined as

$$\mathcal{L}u(x) = \lim_{\epsilon \to 0^+} \int_{|y-x|>\epsilon} (u(x) - u(y))\, k(x,y)\,dy,$$

with $k(x,y)$ a measurable, typically symmetric or controlled-asymmetric kernel which models the strength and nature of interactions between points $x$ and $y$ (Felsinger et al., 2013). Prototypical examples include the fractional Laplacian,

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

the integral generator of isotropic $\beta$-stable Lévy flights $\Delta^{\beta/2}$ (Deng et al., 2018, Bucur, 2017), and more general anisotropic or truncated operators,

$$\mathcal{L}u(x) = \sum_{k=1}^d \mathrm{P.V.}\int_{\mathbb{R}} (u(x + t e_k) - u(x))\, a_k(x,t) |t|^{-1-\alpha_k}\,dt,$$

where the jump kernel can be highly singular and direction-dependent (Chaker et al., 2018).

The nonlocal operator's kernel can be decomposed into symmetric and antisymmetric parts,

$$k_s(x,y)=\frac{1}{2}(k(x,y)+k(y,x)),\quad k_a(x,y)=\frac{1}{2}(k(x,y)-k(y,x)),$$

which is vital for analyzing well-posedness, maximum principles, and regularity (Kassmann et al., 2022). Many applications, such as obstacle problems for Markov jump processes, further admit drift terms or are driven by Lévy measures with varied singularities and exponential tails (Danielli et al., 2017).

2. Functional Analytic and Variational Foundations

The rigorous treatment of nonlocal operators is embedded in suitable Hilbert or Banach spaces, paralleling the Sobolev $H^s$ scale. For a kernel $k(x,y)$0 and open set $k(x,y)$1,

$k(x,y)$2

with corresponding quadratic form

$k(x,y)$3

This setting enables classical variational methods—Lax-Milgram, Fredholm alternative, coercivity via Gårding's inequality—to establish solvability and regularity of the Dirichlet and evolution problems, with nonlocal "boundary data" prescribed on the complement (not the classical boundary) of domains (Felsinger et al., 2013).

Recent advances extend this theory to nonlinear, Orlicz-type settings, employing convex functionals $k(x,y)$4 whose (possibly $k(x,y)$5-trace) subgradients generate nonlinear submarkovian semigroups with contractivity, comparison, and domination properties (Chill et al., 25 Jan 2026). This includes fractional $k(x,y)$6-Laplacians, random walk operators on graphs, and operators on metric-measure spaces.

3. Analytical Properties and Regularity Theory

Nonlocal operators exhibit a rich regularity theory, reflecting both the singularity and the nonlocality of the kernel. For symmetric kernels with integrability or power-type singularities, weak Harnack inequalities and interior Hölder continuity hold for solutions to associated integro-differential problems,

$k(x,y)$7

with explicit exponents dependent on the kernel orders and dimension [1803.