Source
Lorenzo Grimaldi, Pablo Villegas, Alessandro Vezzani, Raffaella Burioni, Davide Cassi and Andrea Gabrielli (v1 2025, v2 2026), “Asymptotic versus mesoscopic spectral dimensions in networks and inhomogeneous structures”, Physical Review E 114, L022301, LIT-028, read in full with the Supplemental Material in NOTE-005, v1 compared.
What was actually shown
Two exponents. The spectral dimension d_s is defined by the density of states, ρ(λ) ∼ λ^(d_s/2 − 1) as λ → 0. The Fiedler dimension d_g is defined by finite-size gap scaling, λ₁ ∼ N
The reduction (SM Eqs. 6–10). For a bundled graph the Laplacian is L = L_f ⊗ I + L_b ⊗ P₀. In the base eigenbasis it splits exactly into N_b fiber problems, each with a single-site impurity of strength ℓ^b, one base eigenvalue (Eq. 7). First-order perturbation about the fiber’s constant mode gives λ_{n1} = ℓ^b_{n1}/N_f (Eq. 10). NOTE-005 confirms the derivation, observes that the Rayleigh quotient makes it an upper bound on each sector’s lowest eigenvalue, and finds the ratio to the exact gap 0.92, 0.96, 0.98, 0.99 on the comb for L = 40 to 320.
The rule (Eqs. 5–6). With N_b ∼ L^(d_f,b), N_f ∼ L^(d_f,f) and the base gap ∼ N_b^(−2/d_g,b): d_g = 2(d_f,f + d_f,b)/(d_f,f + 2d_f,b/d_g,b). It holds when the base sector sets the gap, i.e. unless 2d_f,f/d_g,f > d_f,f + 2d_f,b/d_g,b (SM Eq. 14), a condition the main text omits. d_g = d_g,b only when d_g,b = 2 or the fiber is finite. Meanwhile d_s = d_s,f (Cassi–Regina; Burioni–Cassi). So the comb, a ring of rings, has d_s = 1 and d_g = 4/3 (NOTE-005’s numerics give a local exponent of 1.349); a d-dimensional lattice base with ring fibers has d_s = 1 and d_g = 2(1 + d)/3, which the gap scaling of Fig. 2 follows for d = 1–4.
Homogeneous cases agree. On the Sierpinski gasket the C(τ) plateau gives d_s/2 = 0.684(2) and the gap gives d_g/2 = 0.683(1), against the exact 0.683; the carpet gives 0.915(4) and 0.906(4) (v1 error bars).
Dynamics (SM S6–S7). The return probability decays as t^(−d_s/2) at intermediate times and is cut off at τ_eq ∼ N^(2/d_g). So two networks with the same d_s, comb and brush (d_g = 4/3 against 2), equilibrate on parametrically different times.
What could have come out otherwise. The gap could have followed the fiber’s exponent, or the two exponents could have stayed equal as in the homogeneous fractals. Exact numerics on lattice-based bundles fall on the predicted curve, with deviations still about 10⁻² at 10⁶–10⁷ nodes for d = 3 and 4.
What this does not say
- That composite networks in general decouple the two dimensions. The proof covers one fiber type per base node, one attachment point, no disorder and a common scaling length. “A generic geometric property of composite architectures” (p. 3) is an assertion; the outlook itself calls the construction “a minimal controlled setting”.
- That the paper’s composite finite-size scaling forms are right. SM Eqs. 25 and 29 carry the homogeneous prefactors; NOTE-005 derives N^(−1) for G and N⁰ for C_V, and the comb confirms it. The brush, which the figure plots, has d_s,b = d_g and cannot show the error.
- That homogeneous parts can never be fiber-controlled. v2 asserts it; v1’s one-line reason (2d_f/(d_f + 1) ≤ d_s ≤ d_f) was dropped.
- Anything reproducible about the tree and Barabási–Albert bases. Their construction and parameters are not given, and BA’s “effective d_g,b = 2” is asserted.
- That the mechanism explains polymer rheology. The link is an analogy; no data are analysed.
Connections
- Higher-order networks (THEORY-598, from LIT-091). Battiston et al. report that the spectral dimension of higher Hodge Laplacians grows with order in network-geometry complexes, and that synchrony is stable only for d_S > 4. They use “spectral dimension” in the density-of-states sense. An argument that joins a d_S threshold to a gap-controlled relaxation time should check which exponent it means; that is NOTE-057’s inference from reading the two together.
- Ollivier–Ricci flow (THEORY-456, from LIT-004) is the record’s other account of how a network’s geometry is read off its edges; it does not use the spectrum.