---
title: 'Fractional Extendibility: Methods & Applications'
url: https://www.emergentmind.com/topics/fractional-extendibility
type: topic
---

# Fractional Extendibility: Methods & Applications

Searching arXiv for recent and relevant papers on “fractional extendibility” and adjacent “fractional extension” usages.
Fractional extendibility denotes a family of constructions in which a fractional-order, nonlocal, or variable-integrability object is continued beyond its original domain, category, or parameter range while retaining a controlled relation to the original datum. In the literature considered here, the phrase covers several technically distinct frameworks: bounded extension operators for fractional Sobolev-type spaces on domains, extension problems that realize nonlocal operators as Dirichlet-to-Neumann maps of local weighted PDEs in one extra variable, a continuous-parameter generalization of Gaussian \(k\)-extendibility, and broader operator-theoretic or constitutive extensions in mechanics, discrete exterior calculus, and local fractional calculus [1707.07741] [1511.02893] [2606.05097]. This suggests that the term is best understood as a family resemblance across different subfields rather than as a single canonical definition.

A central commonality is the passage from a fractional object to an ambient structure in which analysis becomes more tractable. Depending on context, the ambient structure is \(\mathbb R^n\), a weighted half-space, a larger simplicial complex, a non-integer-dimensional constitutive model, or a Gaussian covariance-matrix cone. The precise meaning of “extendibility” is therefore dictated by the category under study.

## 1. Conceptual range of the term

In fractional function-space theory, extendibility usually means the existence of an operator
\[
E:\mathcal X(\Omega)\to \mathcal X(\mathbb R^n)
\]
such that \(Eu=u\) a.e. on \(\Omega\) and \(\|Eu\|_{\mathcal X(\mathbb R^n)}\) is controlled by \(\|u\|_{\mathcal X(\Omega)}\). This is the sense used for fractional variable-exponent Sobolev spaces, partial vanishing-trace spaces, and fractional Orlicz-Sobolev spaces [1707.07741] [2002.08656] [2006.00240].

In nonlocal PDE, “extension problem” has a different meaning: one replaces a nonlocal operator on the boundary by a local but degenerate elliptic or parabolic equation in one extra variable, and recovers the original operator through a Dirichlet-to-Neumann map. This is the pattern for \((\partial_t-\Delta)^s\), abstract semigroup generators, spectral fractional diffusion, fractional Branson-Gover operators, and fractional conformal Laplacians [1511.02893] [1207.7203] [1707.07367] [1904.03073] [2310.09160].

In Gaussian quantum information, the term is literal and formalized: a bipartite Gaussian state is declared “\(1/\gamma\)-extendible” when its quantum covariance matrix satisfies a parameterized matrix inequality involving a local auxiliary QCM \(\Delta_B\). Here extendibility is not spatial continuation but a continuous interpolation of the usual Gaussian \(k\)-extendibility criterion [2606.05097].

A recurring source of confusion is the identification of these usages with one another. The whole-space extension operator of Sobolev theory, the Caffarelli-Silvestre-type auxiliary-variable extension, and Gaussian fractional extendibility are structurally analogous only at a high level. Their objects, admissibility conditions, and proofs are different.

## 2. Whole-space extension operators for fractional function spaces

For fractional variable-exponent Sobolev spaces, let \(\Omega\subset\mathbb R^n\) be bounded, \(0<s<1\), and \(p:\Omega\times\Omega\to[1,\infty)\) a bounded continuous function. With
\[
\rho_{s,p(\cdot)}(u):=\int_\Omega\int_\Omega |u(x)-u(y)|^{p(x,y)}|x-y|^{-n-sp(x,y)}\,dx\,dy,
\]
the space \(W^{s,p(\cdot)}(\Omega)\) consists of \(u\in L^{q(\cdot)}(\Omega)\) such that \(\rho_{s,p(\cdot)}(u)<\infty\), where \(q\in C(\overline\Omega)\) and \(q(x)>p(x,x)\). Under \(C^1\)-regularity of the boundary and continuity of \(p\), there exists a linear operator
\[
E:W^{s,p(\cdot)}(\Omega)\to W^{s,p(\cdot)}(\mathbb R^n)
\]
with \(Eu=u\) a.e. on \(\Omega\), together with the large-norm/small-norm dichotomy
\[
\|Eu\|_{W^{s,p(\cdot)}(\mathbb R^n)}\le C\|u\|_{W^{s,p(\cdot)}(\Omega)}^{\beta}\quad\text{if }\|u\|\ge1,
\]
and
\[
\|Eu\|_{W^{s,p(\cdot)}(\mathbb R^n)}\le C\|u\|_{W^{s,p(\cdot)}(\Omega)}^{\alpha}\quad\text{if }\|u\|\le1,
\]
where \(\alpha=\min\{1,p^-\}\) and \(\beta=\max\{1,p^+\}\) [1707.07741]. The construction uses truncation near the boundary, local flattening by \(C^1\)-charts, even reflection on flat patches, retransfer to physical space, and a partition of unity. The nonhomogeneity of the modular is the new feature relative to the constant-\(p\) case.

For spaces with a partial vanishing trace condition, Bechtel defines
\[
W^{s,p}_D(O)=W^{s,p}(O)\cap L^p(O,\delta^{-sp}dx),\qquad \delta(x)=\operatorname{dist}(x,D),
\]
for \(D\subset\partial O\) closed. If \(O\) satisfies the interior thickness condition in \(N=\partial O\setminus D\), then there exists a bounded operator
\[
E:W^{s,p}_D(O)\to W^{s,p}(\mathbb R^n)
\]
such that
\[
\|Ef\|_{W^{s,p}(\mathbb R^n)}\le C\|f\|_{W^{s,p}(O)}.
\]
The operator is linear when \(p\ge1\). The proof enlarges \(O\) to an open set \(\bar O\supset O\) that is globally interior-thick, applies zero extension on \(\bar O\setminus O\), and then invokes Zhou’s whole-space extension theorem on \(\bar O\) [2002.08656].

For fractional Orlicz-Sobolev spaces \(W^{s,\Phi}(\Omega)\), the semimodular is
\[
\dot\rho_{s,\Phi}(u;\Omega)=\int_\Omega\int_\Omega
\Phi\!\left(\frac{|u(x)-u(y)|}{|x-y|^s}\right)\frac{dx\,dy}{|x-y|^n}.
\]
When \(\Phi\) satisfies the non-triviality condition
\[
C:=\sup_{t>0}\;t^\beta\Big/\int_0^t\Phi(s)\,ds<\infty
\]
and the doubling condition \(\Phi(2t)\le K\Phi(t)\), Liang proves that, for a domain \(\Omega\subset\mathbb R^n\), Ahlfors \(n\)-regularity is equivalent to being both a \(W^{\beta,\Phi}\)-extension domain and a \(W^{\beta,\Phi}\)-imbedding domain [2006.00240]. The extension operator is constructed by a Whitney decomposition of \(\mathbb R^n\setminus\overline\Omega\), reflected interior sets \(Q^\ast\subset\Omega\), and the formula
\[
Eu(x)=u(x)\ \text{if }x\in\Omega,\qquad
Eu(x)=\sum_{Q\in\mathcal W}\phi_Q(x)\,u_{Q^\ast}\ \text{if }x\in U.
\]

These three settings show how geometry enters fractional extendibility through \(C^1\)-atlases, interior thickness, or Ahlfors regularity. They also show that the extension problem is not purely analytic: it is tightly constrained by the boundary geometry of the underlying domain.

## 3. Extension problems as localization of nonlocal operators

A second major meaning of fractional extendibility is the extension-by-one-more-variable paradigm. For the fractional heat operator \(L^s=(\partial_t-\Delta)^s\), \(0<s<1\), if \(a=1-2s\) and \(U\) is defined in \(\mathbb R^{n+2}_+=\{(x,t,y):y>0\}\) by
\[
U(x,t,y)=\int_{-\infty}^t\int_{\mathbb R^n}W(x-\xi,t-\tau)\,F_s(y,t-\tau)\,u(\xi,\tau)\,d\xi\,d\tau,
\]
then \(U\) solves
\[
y^a\partial_tU-\operatorname{div}_{x,y}(y^a\nabla_{x,y}U)=0,
\]
with boundary value \(U(x,t,0)=u(x,t)\), and
\[
(\partial_t-\Delta)^su(x,t)=-\lim_{y\to0^+}y^a\partial_yU(x,t)
\]
under the normalization used in the paper [1511.02893]. This recasts a hypersingular nonlocal operator as a local degenerate parabolic PDE.

Stinga and Torrea generalize the same pattern to an abstract operator \(A\) generating a bounded \(C_0\)-semigroup on a Banach space. The extension equation is
\[
u_{yy}(x,y)+\frac{1-2s}{y}u_y(x,y)=-Au(x,y),\qquad u(x,0)=f(x),
\]
and the solution admits the heat-semigroup representation
\[
u(x,y)=\frac{y^{2s}}{4^s\Gamma(s)}
\int_0^\infty e^{-y^2/(4t)}T(t)f(x)\,\frac{dt}{t^{1+s}}.
\]
The boundary flux then yields \((-A)^s f\) up to the explicit constant
\[
c_s=4^{-s}\frac{\Gamma(-s)}{\Gamma(s)}<0
\]
for \(0<s<1\) [1207.7203].

For spectral fractional diffusion on bounded domains, the Caffarelli-Silvestre extension becomes a weighted elliptic boundary-value problem on the cylinder \(C=\Omega\times(0,\infty)\):
\[
-\operatorname{div}_{x'}(y^\alpha A\nabla_{x'}u_e)-\partial_y(y^\alpha\partial_yu_e)+c\,y^\alpha u_e=0,\qquad \alpha=1-2s,
\]
with lateral Dirichlet condition and conormal boundary data
\[
\partial_{\nu^\alpha}u_e=d_s f,\qquad d_s=2^{1-2s}\Gamma(1-s)/\Gamma(s).
\]
Its trace \(u=u_e(\cdot,0)\) solves \(L^s u=f\) in \(\Omega\). The analytic regularity in the extended variable \(y\) is then exploited to design tensorized and sparse tensor FEMs with log-linear complexity relative to the number of degrees of freedom in \(\Omega\) [1707.07367].

The same principle extends to differential forms and conformal geometry. For fractional Branson-Gover operators on \(p\)-forms, a singular elliptic boundary-value problem on the upper half-space yields the boundary operator as the Dirichlet-to-Neumann map
\[
L_{s,p}\omega(x')=d_{s,p}\lim_{y\to0^+}y^{1-2s}i_{e_n}dU(x',y),
\]
with \(U\) solving \(A_{a,p}U=0\) and \(U(\cdot,0)=\omega\) [1904.03073]. For fractional conformal Laplacians, Jin and Kim connect the weighted Poisson integral on the half-space and the ball to sharp weighted \(L^p\to L^q\) inequalities, extremizers, and a weighted isoperimetric ratio on conformally compact Einstein manifolds [2310.09160].

In all these cases, the extension variable is not an auxiliary trick only. It is the site where locality, weighted regularity theory, Harnack inequalities, and variational methods become available.

## 4. Fractional extendibility for Gaussian states

The most explicit use of the term as a named definition appears in Gaussian quantum information. Let \(\rho_{AB}\) be a bipartite Gaussian state with quantum covariance matrix
\[
V_{AB}=\langle\{R_j,R_k\}\rangle-\langle R_j\rangle\langle R_k\rangle,
\]
where \(R=(x_1,p_1,\dots,x_N,p_N)^T\) and physicality is \(V_{AB}\ge i\Omega_{AB}\). For \(\gamma\in[0,1]\), \(\rho_{AB}\) is said to be “\(1/\gamma\)-extendible” if there exists a local QCM \(\Delta_B\ge i\Omega_B\) such that
\[
V_{AB}\ge i\Omega_A\oplus\bigl[(1-\gamma)\Delta_B+i\gamma\Omega_B\bigr].
\]
Setting \(\gamma=1/k\) recovers exactly the Gaussian \(k\)-extendibility condition, so fractional extendibility is a continuous-parameter generalization of the standard notion [2606.05097].

The structural properties of the set \(\mathcal S_{1/\gamma}\) of all \(1/\gamma\)-extendible Gaussian QCMs are central. It is convex; it is closed under operator-norm limits; it is monotone under completely positive Gaussian maps, including trace-preserving and trace-non-increasing ones; and, under a pure-loss channel \(\mathcal A_\lambda\) of transmissivity \(\lambda\), a \(1/\gamma\)-extendible state becomes \(1/(\lambda\gamma)\)-extendible [2606.05097]. The attenuation law is especially important because it turns loss accumulation in a Gaussian network into a direct degradation rule for the extendibility parameter.

Two examples illustrate the meaning of the definition. Every product Gaussian state \(\rho_A\otimes\rho_B\) is \(1/\gamma\)-extendible for all \(\gamma\in[0,1]\). By contrast, for the two-mode squeezed vacuum with
\[
\mu=\cosh 2r,\qquad \nu=\sinh 2r,
\]
the condition is tight at
\[
\gamma\ge \nu/\mu=\tanh 2r.
\]
Equivalently, its extendibility-threshold is \(\gamma_{\min}=\tanh 2r\). As \(r\to\infty\), \(\tanh 2r\to1\), so the state ceases to be fractionally extendible except in the trivial \(\gamma=1\) limit [2606.05097].

This notion is the key technical input in the no-go theorem for Gaussian quantum repeaters. For a repeater chain with overall transmissivity \(\eta=\prod_k\gamma_k\), the paper proves
\[
\mathcal Q(\mathcal R_\eta)\le \mathcal Q(\mathcal A_\eta)=\max\!\left\{0,\log_2\frac{\eta}{1-\eta}\right\}.
\]
The argument tracks \(1/\gamma\)-extendibility through pure-loss segments, GLOCC Gaussian processing, teleportation through the generated Gaussian resource, and a final bottleneck bound. Fractional extendibility therefore serves as a network monotone that rules out any Gaussian repeater advantage over direct transmission [2606.05097].

## 5. Fractional extensions of operators, constitutive laws, and discrete calculi

In several other literatures, the operative content of fractional extendibility is the extension of a classical local operator to a nonlocal or non-integer-order one. In one-dimensional Caputo-type fractional gradient elasticity, the constitutive law is
\[
\sigma(x)=E\,u'(x)+\ell^2E\,{}^CD_{0+}^{\alpha}u'(x),\qquad 0<\alpha<2,
\]
and the governing equation introduces an “effective” internal force \(f_{\rm eff}(x)\) because Caputo derivatives do not in general satisfy a semigroup rule. The three-dimensional Riesz-type model,
\[
c_\alpha(-\Delta)^{\alpha/2}u(\mathbf r)+c_\beta(-\Delta)^{\beta/2}u(\mathbf r)=f(\mathbf r),
\]
shows that, for a point load, the displacement may remain singular or become finite depending on the derivative order; specifically, in the super-gradient regime \(\alpha>3\), the displacement is finite at the load point [1307.6999].

Tarasov and Aifantis systematize related extensions by replacing integer-order spatial derivatives in variational principles with Riesz derivatives, deriving fractional Euler-Lagrange equations, fractional Euler-Bernoulli beam models, fractional Timoshenko beam models, and further generalizations involving fractional time derivatives and fractal media [1404.5241]. In a later treatment they derive fractional GRADELA from a weakly nonlocal integral constitutive law by using a fractional Taylor series in wave-vector space, identifying \(|k|^\alpha\) with \((-\Delta)^{\alpha/2}\), and then adding weakly nonlinear terms treated perturbatively [1808.04452].

In discrete exterior calculus, Crum et al. define a Caputo-like fractional discrete derivative by
\[
\mathbb D_p^s\,\alpha=\frac1{\Gamma(1-s)}\,W\,(\mathbb D_p\alpha),\qquad s\in(0,1),
\]
where \(W\) is assembled from reciprocal powers of distances between barycenters of \((p+1)\)-simplices. The operator is linear, maps discrete \(p\)-forms to discrete \((p+1)\)-forms, and annihilates constants. However, unlike the smooth chain-complex property \(d^{p+1}\circ d^p=0\), one generally has
\[
\mathbb D_{p+1}^s\circ\mathbb D_p^s\neq0
\]
because the weighting matrix need not commute with the incidence matrix [1905.00992].

A related geometric generalization appears in the covariant fractional extension of the modified Laplace operator used in 3D-shape recovery. Herrmann defines a covariant nonlocalization operator
\[
I_g^\alpha\phi(x)=\frac1{\Gamma(\alpha)}\int_0^\infty \xi^{\alpha-1}\,\hat s_g(\xi)\,\phi(x)\,d\xi,
\]
where \(\hat s_g(\xi)\) averages geodesic shifts, and then sets
\[
\hat O^\alpha f=I_g^\alpha(\hat O_{\rm local}f).
\]
For the modified Laplacian used in the reconstruction algorithm, the passage from the local to the nonlocal algorithm yields roughly an order-of-magnitude improvement in rms error, and tuning the fractional order yields an additional factor of about two [1111.1311].

In local fractional calculus, Kolwankar’s recursive local fractional derivative extends the local fractional derivative beyond its critical order by subtracting all lower-order local-fractional Taylor terms before applying the Riemann-Liouville operator. The recursive definition
\[
D_R^qf(x)=\lim_{x'\to x^+}D_{RL}^q\!\left[f(x')-f(x)-\sum_{a<q}\frac{D_R^af(x)}{\Gamma(a+1)}(x'-x)^a\right]
\]
supports a generalized local-fractional Taylor expansion with higher-order terms and extends the product rule to a strictly larger range of orders \(q\) [1312.7675].

These examples broaden the semantic field of fractional extendibility. The shared pattern is the extension of a local differential or discrete calculus into a fractional one, often by a nonlocal kernel, a Fourier multiplier, or a recursive subtraction mechanism.

## 6. Structural themes, misconceptions, and open directions

Across the surveyed literature, several structural motifs recur. One is **localization of nonlocality**: a nonlocal operator is encoded as a local weighted PDE in one additional variable, with a flux law recovering the original operator on the boundary [1511.02893] [1207.7203]. Another is **whole-space realization**: a fractional function defined on a rough or constrained domain is transferred to a geometrically better ambient set by zero extension, reflection, flattening, and Whitney-type decompositions [1707.07741] [2002.08656]. A third is **continuous interpolation**: an integer parameter such as \(k\)-extendibility is replaced by a continuous parameter \(\gamma\in[0,1]\), allowing non-integer branching in Gaussian state analysis [2606.05097].

A common misconception is to treat all of these as variants of the same theorem. They are not. The Sobolev extension operator preserves pointwise restriction on \(\Omega\); the Caffarelli-Silvestre extension constructs a different function in a higher-dimensional half-space whose boundary trace is prescribed; Gaussian fractional extendibility is a matrix-inequality property of bipartite states. Their technical commonality lies at the level of categorical strategy, not at the level of formal definition.

The limitations are likewise domain-specific. In the variable-exponent Sobolev setting, the \(C^1\)-regularity of \(\partial\Omega\) is used to flatten the boundary and reflect, and it remains open whether Lipschitz or uniform domains suffice in the variable-exponent setting [1707.07741]. The same paper does not weaken continuity assumptions on \(p(x,y)\), although it remarks that log-Hölder continuity is a plausible refinement. In the partial vanishing-trace setting, interior thickness in \(N=\partial O\setminus D\) is sharp for the geometric construction used [2002.08656]. In the Hölder-extension theory for the fractional Laplacian, boundary Hölder continuity of the Perron extension is characterized by the decay of the fractional harmonic measure and is equivalent, in the qualitative form, to uniform \(s\)-fatness of the complement \(\Omega^c\) [2508.12134].

The broader trajectory suggests continued movement in two directions. One is **greater generality of function spaces**, as indicated by the explicit mention of Musielak-Orlicz-fractional spaces or variable differentiability order \(s(x)\) as natural next targets for extension methods [1707.07741]. The other is **increased structural rigidity**, as in the Gaussian repeater no-go theorem, where fractional extendibility becomes a sharp obstruction principle rather than an existence theorem [2606.05097]. Together, these directions show that fractional extendibility has evolved into a versatile research motif connecting boundary geometry, weighted local PDE, nonlocal functional analysis, and continuous-parameter constraints in quantum networks.

Source: https://www.emergentmind.com/topics/fractional-extendibility