Papers
Topics
Authors
Recent
Search
2000 character limit reached

Logarithmic Laplacian: Spectrum & Analysis

Updated 6 July 2026
  • The logarithmic Laplacian is a nonlocal operator emerging as the first-order expansion of the fractional Laplacian at s=0, featuring a hypersingular kernel and far-field tail contributions.
  • It adopts a distinct Dirichlet framework on bounded domains that yields a critical energy space with compact L² embedding, leading to a discrete spectrum and cases of negative first eigenvalues.
  • Recent studies extend its applications to nonlinear resonance, Fučík spectrum analysis, maximum principles, extension problems, and numerical approximations, underscoring its innovative role in spectral theory.

Searching arXiv for recent and foundational papers on the logarithmic Laplacian. Search query: "logarithmic Laplacian Chen Weth extension problem Fučík spectrum graphs" The logarithmic Laplacian, usually denoted LΔL_\Delta in the Euclidean setting and log⁡(−Δ)\log(-\Delta) in spectral formulations, is a borderline nonlocal operator obtained as the first-order expansion of the fractional Laplacian at s=0s=0. In RN\mathbb R^N, it is the pseudodifferential operator with Fourier symbol 2ln⁡∣ξ∣2\ln|\xi|, equivalently ln⁡(∣ξ∣2)\ln(|\xi|^2), and it admits a real-space representation with a near-field hypersingular kernel of order ∣x−y∣−N|x-y|^{-N}, a far-field tail term, and a local renormalization term. This combination makes it qualitatively distinct from both the classical Laplacian and the fractional Laplacian of positive order: it is a zero-order nonlocal operator, it is not scale invariant in the usual sense, and on bounded domains its first Dirichlet eigenvalue may be negative (Chen et al., 2017, Arora et al., 7 Jan 2026).

1. Definition and operator structure

For s∈(0,1)s\in(0,1), the fractional Laplacian satisfies

(−Δ)su(x)=u(x)+sLΔu(x)+o(s)as s→0+(-\Delta)^s u(x)=u(x)+sL_\Delta u(x)+o(s)\qquad \text{as }s\to0^+

in Lp(RN)L^p(\mathbb R^N), 1<p≤∞1<p\le \infty, for the regularity classes considered in the Euclidean theory. Accordingly, LΔL_\Delta is the pseudodifferential operator with Fourier multiplier 2ln⁡∣ξ∣2\ln|\xi|, or equivalently ln⁡(∣ξ∣2)\ln(|\xi|^2) (Chen et al., 2017, Arora et al., 7 Jan 2026).

A standard Euclidean integral representation is

LΔu(x)=cN∫B1(x)u(x)−u(y)∣x−y∣N dy−cN∫RN∖B1(x)u(y)∣x−y∣N dy+ρNu(x),L_\Delta u(x)=c_N\int_{B_1(x)}\frac{u(x)-u(y)}{|x-y|^N}\,dy -c_N\int_{\mathbb R^N\setminus B_1(x)}\frac{u(y)}{|x-y|^N}\,dy +\rho_Nu(x),

with

cN=π−N/2Γ(N/2),ρN=2ln⁡2+ψ(N/2)−γ,c_N=\pi^{-N/2}\Gamma(N/2),\qquad \rho_N=2\ln2+\psi(N/2)-\gamma,

where ψ=Γ′/Γ\psi=\Gamma'/\Gamma is the digamma function and γ=−Γ′(1)\gamma=-\Gamma'(1) is the Euler–Mascheroni constant. The kernel ∣x−y∣−N|x-y|^{-N} is the borderline case of hypersingular kernels, and the tail integral together with the local term ρNu(x)\rho_Nu(x) is intrinsic to the operator rather than an auxiliary correction (Arora et al., 7 Jan 2026).

This Euclidean operator is the whole-space counterpart of several related realizations. On bounded domains one usually imposes the nonlocal Dirichlet condition u=0u=0 in RN∖Ω\mathbb R^N\setminus\Omega, whereas in spectral formulations one writes log⁡(−Δ)\log(-\Delta) through functional calculus. A precise implication is that the same logarithmic symbol organizes Euclidean, bounded-domain, graph, and manifold theories, but the concrete realization depends on the ambient geometry and on whether one works with restricted exterior conditions or with spectral calculus (Chen et al., 2017, Chen, 24 Jun 2025).

2. Dirichlet framework on bounded domains

For a bounded domain Ω⊂RN\Omega\subset\mathbb R^N with C1,1C^{1,1} boundary, the nonlocal Dirichlet problem is posed with

u=0in RN∖Ω.u=0 \qquad \text{in }\mathbb R^N\setminus\Omega.

Using

k(z)=1B1(z)∣z∣N,j(z)=1RN∖B1(z)∣z∣N,k(z)=\frac{\mathbf 1_{B_1}(z)}{|z|^N},\qquad j(z)=\frac{\mathbf 1_{\mathbb R^N\setminus B_1}(z)}{|z|^N},

the natural energy space is

H(Ω)={u∈L2(Ω):u=0 in RN∖Ω, ∬R2N∣u(x)−u(y)∣2k(x−y) dx dy<∞}.\mathbb H(\Omega)=\Bigl\{u\in L^2(\Omega):u=0 \text{ in }\mathbb R^N\setminus\Omega,\ \iint_{\mathbb R^{2N}}|u(x)-u(y)|^2k(x-y)\,dx\,dy<\infty\Bigr\}.

Its basic quadratic form is

E(u,v)=cN2∬R2N(u(x)−u(y))(v(x)−v(y))k(x−y) dx dy,\mathcal E(u,v)=\frac{c_N}{2}\iint_{\mathbb R^{2N}}(u(x)-u(y))(v(x)-v(y))k(x-y)\,dx\,dy,

and the logarithmic quadratic form is

EL(u,v)=E(u,v)−cN∬R2Nu(x)v(y)j(x−y) dx dy+ρN∫RNuv dx.\mathcal E_L(u,v)=\mathcal E(u,v)-c_N\iint_{\mathbb R^{2N}}u(x)v(y)j(x-y)\,dx\,dy+\rho_N\int_{\mathbb R^N}uv\,dx.

When uu vanishes outside Ω\Omega, one also has

EL(u,u)=cN2∬Ω×Ω(u(x)−u(y))2∣x−y∣N dx dy+∫Ω(hΩ(x)+ρN)u2(x) dx,\mathcal E_L(u,u)=\frac{c_N}{2}\iint_{\Omega\times\Omega}\frac{(u(x)-u(y))^2}{|x-y|^N}\,dx\,dy+\int_\Omega (h_\Omega(x)+\rho_N)u^2(x)\,dx,

with

hΩ(x)=cN(∫B1(x)∖Ω∣x−y∣−N dy−∫Ω∖B1(x)∣x−y∣−N dy).h_\Omega(x)=c_N\Bigl(\int_{B_1(x)\setminus\Omega}|x-y|^{-N}\,dy-\int_{\Omega\setminus B_1(x)}|x-y|^{-N}\,dy\Bigr).

This formula makes explicit the joint role of the interior interaction, the tail, and the geometry-dependent potential hΩh_\Omega (Arora et al., 7 Jan 2026).

A basic structural fact is that H(Ω)↪L2(Ω)\mathbb H(\Omega)\hookrightarrow L^2(\Omega) compactly. At the same time, the logarithmic Laplacian is a genuinely borderline operator: the natural energy space does not embed continuously into Lr(Ω)L^r(\Omega) for r>2r>2. Instead, the sharp replacement is Orlicz-type control,

H(Ω)↪Lφ(Ω),φ(t)≈t2log⁡(e+t),\mathbb H(\Omega)\hookrightarrow L^\varphi(\Omega),\qquad \varphi(t)\approx t^2\log(e+t),

with compact embedding into smaller Orlicz spaces. These embeddings are central in isolation arguments for the Fučík lines and in nonlinear variational analysis (Arora et al., 7 Jan 2026).

3. Spectral theory and eigenvalue asymptotics

The Dirichlet eigenvalue problem is

LΔu=λuin Ω,u=0in RN∖Ω.L_\Delta u=\lambda u \quad \text{in }\Omega,\qquad u=0 \quad \text{in }\mathbb R^N\setminus\Omega.

Its spectrum is discrete: λ1L<λ2L≤⋯≤λkL→+∞.\lambda_1^L<\lambda_2^L\le \cdots \le \lambda_k^L\to+\infty. The first eigenvalue admits the Rayleigh characterization

λ1L=inf⁡u∈H(Ω)∖{0}EL(u,u)∥u∥L2(Ω)2,\lambda_1^L=\inf_{u\in\mathbb H(\Omega)\setminus\{0\}} \frac{\mathcal E_L(u,u)}{\|u\|_{L^2(\Omega)}^2},

and the first eigenfunction can be chosen strictly positive in Ω\Omega; moreover, λ1L\lambda_1^L is simple. A distinguishing feature is that, unlike the Laplacian or fractional Laplacian, λ1L\lambda_1^L may be negative (Arora et al., 7 Jan 2026).

The second eigenvalue admits a mountain-pass characterization on the unit L2L^2-sphere

P={u∈H(Ω):∫Ωu2 dx=1},\mathcal P=\Bigl\{u\in\mathbb H(\Omega):\int_\Omega u^2\,dx=1\Bigr\},

namely

λ2L=inf⁡γ∈Γmax⁡u∈γ[−1,1]EL(u,u),\lambda_2^L=\inf_{\gamma\in\Gamma}\max_{u\in\gamma[-1,1]}\mathcal E_L(u,u),

where Γ\Gamma is the family of continuous paths on P\mathcal P connecting −ϕ1-\phi_1 to ϕ1\phi_1. This characterization is obtained through the construction of the first nontrivial Fučík curve and implies, in particular, that all eigenfunctions associated with λ>λ1L\lambda>\lambda_1^L are sign-changing (Arora et al., 7 Jan 2026).

Beyond the principal spectral structure, sharp asymptotic information is available for the spectral realization H=12log⁡(−Δ)\mathcal H=\frac12\log(-\Delta) on open sets of finite measure. One has a Weyl law

lim⁡λ→∞e−dλN(λ)=∣Ω∣(2π)d∣Bd∣,\lim_{\lambda\to\infty} e^{-d\lambda}N(\lambda)=\frac{|\Omega|}{(2\pi)^d}|B_d|,

and equivalently

λn(Ω)∼1dlog⁡ ⁣(nα(Ω)),α(Ω)=∣Ω∣(2π)d∣Bd∣,\lambda_n(\Omega)\sim \frac1d\log\!\left(\frac{n}{\alpha(\Omega)}\right), \qquad \alpha(\Omega)=\frac{|\Omega|}{(2\pi)^d}|B_d|,

together with Berezin–Li–Yau-type upper bounds on Riesz means and Faber–Krahn minimization of the first eigenvalue by balls at fixed volume (Laptev et al., 2020). Complementary bounds for the Dirichlet logarithmic Laplacian include Li–Yau-type lower bounds and Kröger-type upper bounds for sums of the first kk eigenvalues, as well as the asymptotic relation

lim⁡k→∞λk(Ω) N2ln⁡k=1\lim_{k\to\infty}\lambda_k(\Omega)\,\frac{N}{2\ln k}=1

in the normalization used there (Chen et al., 2020).

4. Fučík spectrum and nonlinear resonance

For the logarithmic Laplacian, the Fučík spectrum is

ΣL={(α,β)∈R2: LΔu=αu+−βu− in Ω, u=0 in RN∖Ω, has a nontrivial solution},\Sigma_L=\Bigl\{(\alpha,\beta)\in\mathbb R^2:\ L_\Delta u=\alpha u^+-\beta u^- \text{ in }\Omega,\ u=0 \text{ in }\mathbb R^N\setminus\Omega, \text{ has a nontrivial solution}\Bigr\},

with u±=max⁡{±u,0}u^\pm=\max\{\pm u,0\}. The recent analysis of this spectrum identifies the two “trivial” lines

λ1L×R,R×λ1L\lambda_1^L\times\mathbb R,\qquad \mathbb R\times\lambda_1^L

as subsets of ΣL\Sigma_L, and proves that they are isolated in the spectrum (Arora et al., 7 Jan 2026).

The central variational object is the constrained functional

Er(u)=EL(u,u)−r∫Ω(u+)2 dxE_r(u)=\mathcal E_L(u,u)-r\int_\Omega (u^+)^2\,dx

on the unit L2L^2-sphere P\mathcal P. If

c(r)=inf⁡γ∈Γmax⁡u∈γ[−1,1]Er(u),c(r)=\inf_{\gamma\in\Gamma}\max_{u\in\gamma[-1,1]}E_r(u),

then for every r≥0r\ge0,

(r+c(r),c(r))∈ΣL,(c(r),r+c(r))∈ΣL,(r+c(r),c(r))\in\Sigma_L,\qquad (c(r),r+c(r))\in\Sigma_L,

and these points form the first nontrivial Fučík curve

C={(r+c(r),c(r)),(c(r),r+c(r)):r≥0}.\mathcal C=\{(r+c(r),c(r)),(c(r),r+c(r)):r\ge0\}.

This curve intersects the diagonal at

(c(0),c(0))=(λ2L,λ2L),(c(0),c(0))=(\lambda_2^L,\lambda_2^L),

so the second eigenvalue is recovered as the diagonal crossing of the first nontrivial Fučík branch (Arora et al., 7 Jan 2026).

The function r↦c(r)r\mapsto c(r) is Lipschitz and non-increasing, with

0≤c(r)−c(r′)≤r′−r(r<r′),0\le c(r)-c(r')\le r'-r \qquad (r<r'),

and the parametrized branch r↦(r+c(r),c(r))r\mapsto (r+c(r),c(r)) is strictly decreasing in the sense established there. Moreover,

c(r)→λ1Las r→+∞.c(r)\to \lambda_1^L \qquad \text{as } r\to+\infty.

The same framework yields a nonresonance theorem: under asymptotic slope conditions lying between (λ1L,λ1L)(\lambda_1^L,\lambda_1^L) and a point (α,β)∈C(\alpha,\beta)\in\mathcal C, the nonlinear problem

LΔu=f(x,u)in Ω,u=0 in RN∖ΩL_\Delta u=f(x,u)\quad \text{in }\Omega,\qquad u=0 \text{ in }\mathbb R^N\setminus\Omega

admits at least one nontrivial solution by the Mountain Pass Theorem, using the Palais–Smale condition and the geometry induced by the Fučík curve (Arora et al., 7 Jan 2026).

5. Maximum principles, symmetry, and critical semilinear equations

For bounded Lipschitz domains, the maximum principle for the Dirichlet logarithmic Laplacian is governed by the sign of the first eigenvalue: LΔL_\Delta satisfies a maximum principle in Ω\Omega if and only if λ1(LΔ,Ω)>0\lambda_1(L_\Delta,\Omega)>0. This is qualitatively different from the Laplacian and from positive-order fractional Laplacians, because λ1\lambda_1 may be nonpositive on sufficiently large sets, while positivity is recovered for domains of sufficiently small measure (Chen et al., 2017).

Antisymmetric maximum principles and Hopf-type lemmas have been developed for LΔL_\Delta. In symmetric sets, these tools support a moving-planes argument showing that if Ω\Omega is bounded, convex in the direction e1e_1, and symmetric with respect to {x1=0}\{x_1=0\}, then every strictly positive weak solution of

LΔu=f(u)in Ω,u=0in RN∖ΩL_\Delta u=f(u)\quad \text{in }\Omega,\qquad u=0 \quad \text{in }\mathbb R^N\setminus\Omega

is symmetric with respect to {x1=0}\{x_1=0\} and decreasing in the e1e_1-direction; in balls, positive solutions are radial and radially decreasing. The same circle of ideas yields rigidity for a parallel surface problem, forcing the underlying sets to be concentric balls (Pollastro et al., 2024).

On the whole space, a critical semilinear equation involving the logarithmic Laplacian has a complete classification in the normalization stated there. When

LΔu=k ulog⁡u,u≥0in Rn,L_\Delta u = k\,u\log u,\qquad u\ge0 \quad \text{in }\mathbb R^n,

the only positive solutions occur at

k=4n,k=\frac4n,

and they are exactly

ux~,t(x)=βn(tt2+∣x−x~∣2)n2,t>0, x~∈Rn,u_{\tilde x,t}(x)=\beta_n\Bigl(\frac{t}{t^2+|x-\tilde x|^2}\Bigr)^{\frac n2}, \qquad t>0,\ \tilde x\in\mathbb R^n,

with

βn=2n2en2ψ(n2).\beta_n=2^{\frac n2} e^{\frac n2\psi(\frac n2)}.

For k∈(0,+∞)∖{4n}k\in(0,+\infty)\setminus\{\frac4n\}, no positive solution exists (Chen et al., 2024).

A parallel bounded-domain nonlinear theory treats subcritical, critical, and supercritical logarithmic nonlinearities through sharp logarithmic Sobolev inequalities, a Pohozaev identity, and a Díaz–Saa type inequality. In particular, star-shaped domains support nonexistence results at and above the critical logarithmic threshold, while logistic-type problems with λ<0\lambda<0 admit a unique nontrivial nonnegative weak solution together with positivity and boundary estimates (Arora et al., 2024).

6. Extensions, fundamental solutions, and nonlinear analogues

A local extension problem for the Euclidean logarithmic Laplacian is now available. For u∈L0(RN)u\in L_0(\mathbb R^N), the unique extension wuw_u solves

−div⁡(t∇wu)=0in R+N+1,−lim⁡t→0+t∂twu=u,-\operatorname{div}(t\nabla w_u)=0 \quad \text{in }\mathbb R^{N+1}_+, \qquad -\lim_{t\to0^+} t\partial_t w_u=u,

and is represented by the Poisson formula

wu(x,t)=CN∫RNu(ξ)(∣x−ξ∣2+t2)N/2 dξ.w_u(x,t)=C_N\int_{\mathbb R^N}\frac{u(\xi)}{(|x-\xi|^2+t^2)^{N/2}}\,d\xi.

The boundary operator is

LΔu=2(ln⁡2−γ)u−2lim⁡t→0+(wu(⋅,t)+uln⁡t).L_\Delta u = 2(\ln2-\gamma)u-2\lim_{t\to0^+}\bigl(w_u(\cdot,t)+u\ln t\bigr).

After doubling the extension variable, the extended function becomes harmonic in RN+2∖(RN×{0})\mathbb R^{N+2}\setminus(\mathbb R^N\times\{0\}), and this leads to a weak unique continuation principle for LΔu=0L_\Delta u=0 (Chen et al., 2023).

On complete Riemannian manifolds, the logarithmic Laplacian admits a Bochner integral formula

log⁡(−Δ)f=∫0∞e−tf−etΔft dt\log(-\Delta)f=\int_0^\infty \frac{e^{-t}f-e^{t\Delta}f}{t}\,dt

on the logarithmic Sobolev space Hlog⁡(M)H^{\log}(M). Under a Ricci lower bound, one obtains the pointwise formula

log⁡(−Δ)specf(x)=∫MK1(x,y)(f(x)−f(y)) dvol(y)−∫MK2(x,y)f(y) dvol(y)+Γ′(1)f(x),\log(-\Delta)_{\mathrm{spec}} f(x)=\int_M K_1(x,y)(f(x)-f(y))\,d\mathrm{vol}(y) -\int_M K_2(x,y)f(y)\,d\mathrm{vol}(y)+\Gamma'(1)f(x),

with

K1(x,y)=∫01pt(x,y)t dt,K2(x,y)=∫1∞pt(x,y)t dt.K_1(x,y)=\int_0^1 \frac{p_t(x,y)}{t}\,dt,\qquad K_2(x,y)=\int_1^\infty \frac{p_t(x,y)}{t}\,dt.

The discrepancy between spectral and heat-kernel definitions is governed by the mass-loss function and therefore by stochastic completeness (Chen, 24 Jun 2025).

On weighted graphs, the same semigroup formula becomes

log⁡(−Δ) u=∫0∞e−tu−etΔut dt\log(-\Delta)\,u=\int_0^\infty \frac{e^{-t}u-e^{t\Delta}u}{t}\,dt

in ℓ2(V,μ)\ell^2(V,\mu), and under stochastic completeness one obtains an explicit pointwise form with a short-range symmetric difference term Wlog⁡(x,y)W_{\log}(x,y), a long-range linear term W(x,y)W(x,y), and the constant Γ′(1) u(x)\Gamma'(1)\,u(x). For weighted lattice graphs, Wlog⁡W_{\log} has super-fast off-diagonal decay, whereas W(x,y)W(x,y) decays like d(x,y)−dd(x,y)^{-d}; accordingly, log⁡(−Δ)\log(-\Delta) fails to be bounded on ℓ2\ell^2 (Chen et al., 8 Jul 2025).

The logarithmic Laplacian also supports further structural extensions. An alternative construction of fundamental solutions, based on a division problem, proves that in dimensions d≥3d\ge3 there exists a fundamental solution and identifies the entire LΔL_\Delta-harmonic class in the Lizorkin setting as those distributions whose Fourier transforms are single-layer distributions supported on Sd−1S^{d-1}; the same approach gives low-dimensional existence results away from the origin (Lee, 25 Jun 2025). Higher logarithmic operators Lm\mathcal L_m with symbol (2ln⁡∣ξ∣)m(2\ln|\xi|)^m arise as higher ss-derivatives of (−Δ)s(-\Delta)^s and generate Taylor expansions in the order parameter (Chen, 2023). In the nonlinear direction, the logarithmic pp-Laplacian LΔpL_{\Delta_p} is the s→0+s\to0^+ derivative of the fractional pp-Laplacian, with a variational Dirichlet theory, a Faber–Krahn inequality, a boundary Hardy-type inequality, and a maximum principle whose validity depends on the sign of the first Dirichlet eigenvalue (Dyda et al., 2024). Subsequent work constructs an unbounded minimax sequence of eigenvalues for LΔpL_{\Delta_p} and develops sharp pp-logarithmic Sobolev inequalities and critical Orlicz embeddings for the associated energy space (Arora et al., 26 Dec 2025, Arora et al., 30 Oct 2025).

7. Numerical approximation

Two rather different numerical approaches have been developed for Dirichlet problems involving the logarithmic Laplacian. In one dimension, finite element analysis for the restricted Dirichlet problem uses recently established log-Hölder regularity and log⁡\log-weighted spaces to prove a rigorous error estimate

∥u−uh∥h(Ω)≤C ℓ(h)α∥f∥Y(Ω),\|u-u_h\|_{\mathfrak h(\Omega)}\le C\,\ell(h)^\alpha \|f\|_{\mathcal Y(\Omega)},

where ℓ(h)=∣ln⁡(min⁡{ρ0,h})∣−1\ell(h)=|\ln(\min\{\rho_0,h\})|^{-1}. A distinctive feature is that the logarithmic stiffness matrix can be obtained as the derivative at s=0s=0 of the fractional stiffness matrix, and the discrete eigenvalue problem converges to the continuous one (Hernández-Santamaría et al., 2023).

A Fourier-side sinc-basis method treats log⁡(−Δ)\log(-\Delta) directly as the multiplier with symbol log⁡(∣ω∣2)\log(|\omega|^2). In this framework, the discrete operator is assembled from the Fourier stencil, with singular-frequency treatment near ω=0\omega=0 by a Duffy transform. On disks BR⊂R2B_R\subset\mathbb R^2, the computed Dirichlet eigenvalues follow the scaling law

λ(ℓ)=2log⁡(R/Rℓ),\lambda^{(\ell)}=2\log(R/R_\ell),

and the numerical study reports the radii RℓR_\ell at which the ℓ\ell-th eigenvalue vanishes, as well as multiplicities consistent with rotational symmetry. For the Dirichlet problem with f≡1f\equiv1, the observed L2L^2 error decays approximately linearly in hh in the presented tests, although a rigorous convergence proof for the logarithmic Laplacian is explicitly stated to remain open (Dondl et al., 15 Sep 2025).

Across these developments, the logarithmic Laplacian appears not as a degenerate limit of more familiar operators, but as a structurally independent zero-order nonlocal object. Its tail contributions, Orlicz-scale embeddings, possible negativity of the principal eigenvalue, and sensitivity to stochastic completeness or exterior geometry are recurrent themes in the current theory.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (17)

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to Logarithmic Laplacian.