Energy Retention under Domain Truncation in Heterogeneous Fractional Elasticity: Sharp Half-Space Constants in All Dimensions, Multi-Directional Absorption, and a Capacitary Obstruction
The paper under discussion, authored by Sami Mehennaoui of the Université 20 août 1955‑Skikda and entitled “Energy Retention under Domain Truncation in Heterogeneous Fractional Elasticity: Sharp Half‑Space Constants in All Dimensions, Multi‑Directional Absorption, and a Capacitary Obstruction”, addresses a question left explicitly open by Evgrafov and Bellido in their analysis of the Eringen nonlocal elasticity model with Riesz attenuation kernel. In Eringen’s theory the pointwise stress–strain law is replaced by a convolution with a smoothing kernel, which captures size effects and removes artificial stress singularities; when the kernel is of Riesz type the nonlocal strain becomes a fractional gradient and the model becomes tractable in fractional Sobolev spaces. Evgrafov and Bellido proved well‑posedness for heterogeneous bodies under the assumption that the outer integration in the elastic energy runs over the whole space, which physically demands that the body be embedded in an infinite elastic medium with everywhere elliptic moduli. The present work asks whether that outer domain may be reduced to the body Ω itself while coercivity is preserved. The two models are genuinely different: even if the displacement vanishes outside Ω, its nonlocal strain does not, so the discarded exterior term is nonzero. The paper quantifies that discarded term exactly, determines on which geometries and in which ranges of the fractional order s the truncated model remains well posed, and identifies the precise mechanism by which it can fail. Its contributions are of five kinds: a structural strain identity, sharp leakage constants in model geometries, a multi‑directional absorption principle giving dimension‑free coercivity on slabs and polytopes, a complete analysis of the truncated null space with a sharp injectivity threshold on the ball, and a capacitary obstruction showing that the failure is a boundary‑regularity phenomenon rather than a failure of positivity. The general bounded Lipschitz case is deliberately not claimed; it is reduced to a single localization statement and stated as an open problem. The work is classified under 35R11, 35A15, 45E10, 46E35, 74B05, 74G25. The setting is fixed with careful attention to normalization, because the whole argument depends on exact constants. The Fourier convention is û(ξ) = ∫ u(x) e^{−i x·ξ} dx with Plancherel identity ‖u‖²_{L²(ℝⁿ)} = (2π)^{−n} ∫ |û(ξ)|² dξ, and the homogeneous seminorm is [u]²_s = (2π)^{−n} ∫ |ξ|^{2s} |û(ξ)|² dξ. The Riesz potential I_α of order α ∈ (0,n) is the Fourier multiplier with symbol |ξ|^{−α}, equivalently convolution with the kernel c_{n,α} |x−y|^{α−n}, where c_{n,α} = Γ((n−α)/2) / (2^α π^{n/2} Γ(α/2)); with this choice I_α I_β = I_{α+β} whenever the composition makes sense. The fractional gradient D^s u is the multiplier with symbol i ξ |ξ|^{s−1} times û, written in integral form as a principal value of ∫ (u(x) − u(y)) ⊗ (x−y) / |x−y|^{n+s+1} dy with a constant chosen exactly so that the symbol is i ξ |ξ|^{s−1}; the symmetric part ℰ^s u = ½(D^s u + (D^s u)^T) takes values in the space of symmetric n×n matrices with the Frobenius product, and the fractional divergence div^s u has symbol i ξ·û |ξ|^{s−1}. Because I_{1−s} ∘ ∇ = D^s at the level of symbols, the Riesz potential of order 1−s of the classical symmetric strain ε(u) = ½(∇u + ∇u^T) is exactly ℰ^s u; in the adopted normalization the strain identity carries constant one in every dimension. Remark 2.5 records that if the unnormalized integral convention is used in dimension one, the identity instead carries an explicit Gamma‑function constant, namely Γ(1−s) sin(πs/2)/s, whose values are tabulated as 2.5066283 at s = 0.5, 2.7210594 at s = 0.55, 2.9908813 at s = 0.6, 4.4661690 at s = 0.75 and 10.4404229 at s = 0.9; the equivalence with the normalized convention is verified in the appendix by duplication and reflection formulas. Two boundedness facts are proved in the precise form in which they are true. Lemma 2.1 is a Schur bound: if Ω has diameter at most d, then the truncated operator T_Ω = 1_Ω I_{2(1−s)} 1_Ω satisfies ‖T_Ω f‖_{L²(Ω)} ≤ (ω_{n−1} d^{2−2s} / (2−2s)) ‖f‖_{L²(Ω)}, where ω_{n−1} is the surface measure of the unit sphere; the proof observes that for x,y ∈ Ω the kernel coincides with the truncated kernel, which is integrable with mass ω_{n−1} d^{2−2s}/(2−2s), and applies Young’s inequality, restriction to Ω only decreasing the norm. Lemma 2.2 is a global bound on bounded supports obtained from the Hardy–Littlewood–Sobolev inequality combined with Hölder’s inequality on Ω; it requires the potential order 2(1−s) to be at most n/2, that is s ≥ 1 − n/4, a condition automatic in dimension n ≥ 4 and a genuine restriction in dimensions one, two and three. Remark 2.3 shows that this restriction cannot be removed: taking f = 1_{B₁}, the potential decays like |x|^{2−2s−n} at infinity, so its global L² norm diverges precisely when 4−4s ≥ n while ‖f‖_{L²} stays bounded. Remark 2.4 adds that the potential is not bounded from L²(Ω) to L²(ℝⁿ) in general because its symbol blows up at the origin while the Sobolev weight vanishes there; testing on a single frequency gives the exact ratio. Only the truncated bound of Lemma 2.1 and Lemma 2.2 in its legitimate range are used in the positive results, together with the strain bound of Lemma 3.1. The function spaces are defined without any boundary regularity. For an open Ω, H^s_Ω denotes the closure of C_c^∞(Ω) in the H^s(ℝⁿ) norm; every element vanishes almost everywhere on the complement, since the set of such functions is closed and contains the test functions. The converse holds for Lipschitz Ω but is never used. On bounded Ω the fractional Poincaré inequality ‖u‖²_{L²} ≤ C [u]²_s holds by positivity of the first eigenvalue of the fractional Dirichlet Laplacian, and the embedding into L² is compact. Definition 2.6 introduces the admissible sources: the closure of C_c^∞(Ω) in the negative‑order norm H^{−(1−s)}(ℝⁿ); every admissible source is a distribution supported in the closure of Ω, and both sides of every inequality proved later are continuous in that norm. Lemma 2.7 shows that this class contains every L² function with compact support in the open set Ω, proved by mollification and translation in an interior direction together with Lemma 2.2 to convert L² convergence into negative‑order convergence, and every strain ε(u) with u ∈ H^s_Ω, proved by Lemma 3.1. The coefficient tensor C is assumed symmetric and uniformly elliptic, λ|A|² ≤ C(x)A·A ≤ Λ|A|² for almost every x and all symmetric A, with 0 < λ ≤ Λ; no continuity of C is assumed anywhere in the paper. Section 3 establishes the structural core. Lemma 3.1 gives the sharp global fractional Korn bounds ½ [u]²_s ≤ ‖ℰ^s u‖²_{L²(ℝⁿ)} ≤ [u]²_s, together with the exact Fourier identity |ℰ̂^s u|² = ½ |ξ|^{2s−2} (|ξ|² |û|² + |ξ·û|²) and the identity ‖div^s u‖² = (2π)^{−n} ∫ |ξ|^{2s−2} |ξ·û|² dξ. The proof expands the Frobenius norm of the symmetrized symbol, observes that the cross terms combine into |ξ·û|², and uses |ξ·û| ≤ |ξ||û|. Both constants are sharp: the lower bound is attained by transverse fields with ξ·û = 0 and the upper bound by longitudinal fields; in dimension one both inequalities are equalities. Lemma 3.2 shows that for bounded Ω and u ∈ H^s_Ω the strain ε(u) belongs to H^{s−1}(ℝⁿ) with norm controlled by [u]_s, is supported in the closure of Ω, and lies in the closed subspace of such supported distributions; consequently its pairing with any function vanishing almost everywhere on Ω̄ is zero. Support in a closed set is preserved under distributional limits, so no boundary layer appears and no hypothesis on ∂Ω is needed; Remark 3.3 notes that for Lipschitz Ω and s > ½ one could argue instead that no nonzero distribution supported on the hypersurface ∂Ω lies in H^{s−1}. Lemma 3.4 extends the truncated map, initially defined on smooth compactly supported fields by an absolutely convergent integral, uniquely to a bounded operator on the admissible source space, by integrating by parts to transfer the derivative from u to the kernel and invoking Lemma 3.1. Theorem 3.5 is the strain identity: for bounded Ω, s ∈ (0,1) and u ∈ H^s_Ω, the truncated Riesz potential of the strain is supported in Ω̄, carries no boundary layer, and coincides inside Ω with the restriction of the whole‑space symmetric fractional gradient, with constant one; hence the truncated energy is exactly ∫_Ω C ℰ^s u · ℰ^s u. Remark 3.6 emphasizes that this selects the whole‑space operator, not the regional operator obtained by restricting the inner integral to Ω; the two differ by a generally nonzero term singular at the boundary. Section 4 computes the sharp exterior leakage on the half‑line, where the kernel of I_{1−s} is a positive multiple of |x−y|^{−s} and the constant cancels in all ratios. The Mellin transform M f(z) = ∫₀^∞ f(x) x^{z−1} dx becomes, after the substitution x = e^t, a Fourier transform in the logarithmic variable, so Plancherel gives ∫₀^∞ |f|² dx = (1/2π) ∫_ℝ |M f(½ + iτ)|² dτ. The interior operator T_in f(x) = ∫₀^∞ |x−y|^{−s} f(y) dy for x > 0 and the exterior operator T_out, the same integral evaluated at x < 0, are Mellin multipliers with a common shift z ↦ z + 1 − s. Lemma 4.1 establishes the factorization by Fubini and the substitution y = xt, valid in the strip 0 < Re z < s: at the origin the integrand behaves like a power with exponent Re(s−z)−1, at infinity like a power with exponent −1−Re z, and the singularity at t = 1 is integrable because s < 1. Lemma 4.2 evaluates the symbols in closed form: the interior symbol is Γ(1−s)[Γ(z)/Γ(z+1−s) + Γ(s−z)/Γ(1−z)], obtained by splitting at t = 1 and using the Beta integral together with the substitution t ↦ 1/t, while the exterior symbol is Γ(z)Γ(s−z)/Γ(s). Lemma 4.3 divides them and applies the reflection formula twice and a sum‑to‑product identity to obtain the exact ratio m_out(z)/m_in(z) = cos(πs/2) / cos(π(z − s/2)). On the Plancherel line z =
Authors
- Mehennaoui Sami (ORCID: https://orcid.org/0009-0000-2483-7604)
Institutions
- University of Skikda (DZ)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.5281/zenodo.24310203
- Primary Topic
- Nonlocal and gradient elasticity in micro/nano structures
- Type
- preprint