Fractional Sobolev Seminorms: Theory & Applications

• Fractional Sobolev seminorms are nonlocal integral measures that assess function differentiability through pairwise differences.
• They are fundamental in analyzing nonlocal equations, Markov jump processes, and variational models, ensuring robust convergence properties.
• Discrete and continuous formulations maintain equivalence to the classical Gagliardo seminorm under inhomogeneous and anisotropic settings.

A fractional Sobolev seminorm is a nonlocal quadratic or p-th power functional that quantifies the differentiability of a function in an integral sense over pairs of points, instead of via classical derivatives. For $f$ in $L^2(\mathbb{R}^d)$ and $s \in (0,1)$, the classical Gagliardo (Slobodeckiĭ) seminorm takes the form

$$[f]^2_{H^s(\mathbb{R}^d)} = \iint_{\mathbb{R}^d \times \mathbb{R}^d} |f(x) - f(y)|^2 |x-y|^{-d-2s}\, dx\,dy,$$

defining the natural quadratic form structure underlying the Hilbert space $H^s(\mathbb{R}^d)$. Fractional Sobolev seminorms—and their generalizations—appear in the analysis of nonlocal equations, probabilistic jump processes, and as energy functionals for a broad class of nonlocal variational models.

1. Core Definitions and Variants

Fractional Sobolev seminorms are typically defined via integral difference quotients over pairs of points, using spatial kernels encoding the degree of nonlocality and the order of differentiability.

Gagliardo (Slobodeckiĭ) Seminorm:

$$[f]^2_{H^s(\mathbb{R}^d)} = \iint_{\mathbb{R}^d\times\mathbb{R}^d} |f(x) - f(y)|^2\, |x - y|^{-d - 2s}\, dx\, dy,$$

for $0 < s < 1$.

Generalized Quadratic Forms:

Consider quadratic forms of the type

$$Q_s(f) = \frac{1}{2}\iint_{\mathbb{R}^d\times\mathbb{R}^d} (f(x) - f(y))^2 k(x,y)\, dx dy,$$

where the kernel $k(x,y) \geq 0$ has controlled two-sided comparability with the canonical $|x-y|^{-d-2s}$ kernel, possibly restricted to a spatially dependent set of directions. These forms arise in models where, for each $x$, increments are restricted to subsets $V^{\Gamma}[x] \subset \mathbb{R}^d$ determined by an inhomogeneous family of double-cones $\Gamma$, see below for details (Bux et al., 2017).

Double-Cone Restricted Nonlocal Kernels:

Let $\Gamma$ be a measurable configuration assigning to each $x \in \mathbb{R}^d$ a double cone $V(\cdot)$ (parameterized by aperture $\theta(x)\in (0, \pi/2]$ and axis $v(x)\in S^{d-1}$). The quadratic form

$$Q_s(f) = \frac{1}{2}\iint_{\mathbb{R}^d \times \mathbb{R}^d} (f(x) - f(y))^2\, k(x, y)\, dx dy,$$

is defined with $k$ constrained by

$$\Lambda^{-1} \Big(1_{V^{\Gamma}[x]}(y) + 1_{V^{\Gamma}[y]}(x)\Big) |x-y|^{-d-2s} \leq k(x,y) \leq \Lambda |x-y|^{-d-2s},$$

where $\Lambda \geq 1$ denotes the uniform ellipticity parameter (Bux et al., 2017).

2. Key Comparability and Equivalence Results

A central structural property is that such generalized quadratic forms, even under strong inhomogeneity and non-isotropy in their jump geometry, remain comparable to the classical Gagliardo seminorm. Specifically:

Local Comparability Theorem:

For any ball $B\subset \mathbb{R}^d$ and $f\in L^2(B)$,

$$\iint_{B\times B}|f(x)-f(y)|^2|x-y|^{-d-2s}\,dx\,dy \leq c \iint_{B\times B}|f(x)-f(y)|^2\,k(x,y)\,dx\,dy,$$

with $c = c(d, \theta_0, \Lambda)$ determined by the minimal cone aperture $\theta_0$, dimension, and the kernel ellipticity ratio (Bux et al., 2017). The reverse bound follows from the upper bound on $k$.

Global Norm Equivalence:

For $f\in L^2(\mathbb{R}^d)$,

$$c [f]_{H^s(\mathbb{R}^d)}^2 \leq Q_s(f) \leq C [f]_{H^s(\mathbb{R}^d)}^2,$$

with $c$, $C>0$ as above, yielding norm equivalence at the space level.

Discrete Analogue (On $\mathbb{Z}^d$):

For discrete kernels $\omega(x,y)$ with an analogous two-sided comparability condition on bonds (again possibly restricted to geometric cone configurations), the corresponding quadratic form

$$E_\omega(f) = \frac{1}{2}\sum_{x,y \in \mathbb{Z}^d, |x-y|>R_0} (f(x)-f(y))^2 \omega(x,y)$$

has local and global comparability to the discrete Gagliardo form, with constants controlled explicitly in terms of the model parameters (Bux et al., 2017).

3. Proof Techniques: Discretization and Renormalization

The equivalence proofs proceed via multiscale geometric and combinatorial constructions, reflecting the fundamentally nonlocal nature of the seminorm:

• Discrete Approximation: The continuous kernel $k(x,y)$ is approximated by a discrete kernel $\omega_h(x,y)$ representing connections between $h\mathbb{Z}^d$ grid points, obtained by averaging $k$ over associated cubes $A_h(x)$. Uniform covering and connectivity properties of the cone configurations are used.
• Chain-of-Blocks and Renormalization: The main step involves demonstrating that, despite possible inhomogeneity in the cone geometry, all pairs in a block can be connected via paths of controlled length and comparability of each jump. This employs inductive covering arguments and cone-intersection lemmas.
• Limiting Passage: By letting the discretization scale $h\to 0$, and using Lebesgue differentiation and dominated convergence, comparability at finite resolution is upgraded to equivalence of the full continuous seminarms.

4. Geometric and Analytic Applications

1. Markov Jump Processes and Dirichlet Forms:

The forms $Q_s$ (and their generalizations) are precisely the Dirichlet forms of nonlocal symmetric strong Markov jump processes with jumping kernel $k(x,y)$. The comparability to the isotropic stable process ensures well-posedness, regularity, and spectral theory for the associated generators, and underlies qualitative properties such as transience/recurrence and heat kernel estimates.

2. Discrete-to-Continuum Limits:

The discrete approximation scheme delineates a general framework for establishing continuum-limits of discrete models (e.g., in probability, peridynamics, or network systems), guaranteeing that nonlocal quadratic forms arising from inhomogeneous microstructures converge to classical $H^s$-based theories at the macroscopic scale.

3. Quantitative Constants:

The precise structural dependence of the seminorm equivalence constants on geometric parameters is given explicitly: e.g., for worst-case jump chains of $N$ steps and multiplicity $M$,

$$c \sim \left(2\Lambda \lambda^{d+2s} N M \right)^{-1},$$

where $\lambda$ encodes the block-diameter to jump-length ratio (Bux et al., 2017).

5. Connections to Broader Theory and Open Questions

Fractional Sobolev seminorms, beyond the Gagliardo form, admit significant generalizations:

• Directionally restricted or anisotropic seminorms, as in conical, sectorial, or even stochastic settings.
• Inhomogeneous models without global regularity on the kernel or geometry, such as non-uniform cones or robust, location-dependent jump structures.
• Discrete analogues relevant for modeling lattice systems, interacting particle dynamics, and high-dimensional Markovian processes.

The exposition in (Bux et al., 2017) demonstrates that, within this highly general and inhomogeneous setting, the geometric and analytic core of $H^s$ spaces is robust, and optimal equivalence to the classical Gagliardo structure is maintained under mild ellipticity and minimal aperture conditions.

These findings also suggest further research directions, such as the full characterization of precise constants under various geometric constraints, the extension to more general nonlocal interactions, and the analysis of convergence rates in discretizations for computational applications.

6. Summary Table of Key Objects

Quadratic Form	Kernel Structure	Equivalence to Gagliardo
$Q_s(f)$	$k(x,y)$, 2-sided, inhom.	$c [f]^2_{H^s} \leq Q_s(f) \leq C [f]^2_{H^s}$
$E_\omega(f)$	$\omega(x,y)$, discrete, cone-restricted	Locally and globally comparable
Classical $[\,\cdot\,]_{H^s}$	$|x-y|^{-d-2s}$, isotropic	—

References:

• “Quadratic forms and Sobolev spaces of fractional order” (Bux et al., 2017)