Formal Manuscript Manuscript · 2025

Why Scaling Stops: Log-Periodic Morphological Variance Under Competing Power-Law Constraints

InTelluric / Alnitak Group — Los Angeles, CA

Abstract

A striking asymmetry appears across both biology and engineering: scaling by multiplicity persists, but scaling by recursive refinement of the same subunit repeatedly halts at narrow, domain-specific bands. Large organisms are made of standard-size cells, not recursively smaller cells. Chips improved by adding more transistors, not by making transistors out of smaller transistors. The breakdown of classical MOSFET scaling is an engineering instance of the same structural puzzle: continued refinement failed not from insufficient cleverness but from physics — thermal-voltage limits (kT/q), subthreshold swing constraints (the 60 mV/decade thermionic floor), leakage and variability constraints — requiring architectural pivots at FinFET, Tri-Gate, GAA, and 3D stacking rather than simple shrinks.

This manuscript formalizes a general claim. For systems governed by a finite set of competing constraints that can be modeled as power laws of a characteristic scale L, if the constraint hierarchy recurs under renormalized rescaling — a discrete scale invariance (DSI) condition — then any bounded observable measuring "morphological freedom" must be periodic in ln L and therefore admits a log-periodic oscillatory representation. The core object is a dimensionless morphological variance V(L) ∈ [−1, 1]: a bounded proxy for how many qualitatively distinct configurations are feasible at a given scale while satisfying all active constraints. V(L) = +1 marks innovation bands where constraint competition yields many viable morphologies; V(L) = −1 marks collapse bands where one constraint dominates and geometry pins to a small set of modes.

The mathematical derivation is a chain of short steps: power-law constraints have log-linear intersection scales; if the constraint family recurs under rescaling, the recurrence yields DSI; DSI forces log-periodicity; any reasonable bounded periodic function admits a Fourier series; keeping the first harmonic as a parsimony truncation gives V(L) ≈ A₁ sin(2π/ln λ · ln(L/L₀) + φ₀). This is not a statistical fit — it is a forced mathematical form under DSI. The model is falsifiable: it predicts that in domains where DSI holds, pivot scales (constraint-competition crossovers) should be log-periodically spaced with extractable ratio λ. A domain where sequential pivots are not log-periodically spaced refutes DSI for that domain.

The empirical domain survey deliberately treats each case procedurally: identify the characteristic scale L, anchor the primitive L₀ empirically, specify the competing constraints as power laws, locate the constraint-equality pivot, compute V(L). Across capillary length (surface tension vs. gravity), mountain height (lithostatic stress vs. material strength), potato radius (central pressure vs. strength), Euler buckling (critical buckling stress vs. yield stress), Hall–Petch crossover (dislocation strengthening vs. grain-boundary softening), turbulence onset (inertial vs. viscous forces), and eukaryotic cell size (diffusive transport time vs. metabolic demand), the variance peaks predict pivot scales that match the empirically observed regime boundaries — in each case from first principles, without fitting V.

The manuscript also examines the conjectured universality of the golden ratio φ ≈ 1.618 as the scale ratio λ. Two independent arguments support treating φ as a natural attractor: the Fibonacci recursion mechanism (if architecture changes must reuse the previous two scale states, the scale ratio converges to φ by the recursion L_{n+1} = L_n + L_{n−1}), and Diophantine robustness (the golden ratio is extremally irrational in the continued-fraction sense, making it the most robust choice against resonance among constraint-exponent intersections). The paper draws a strict line: log-periodicity requires DSI, but DSI does not require λ = φ. The universality of φ is treated as an empirical claim subject to falsification, not as a derivable consequence of the framework.

Key Terms

log-periodic morphological variance discrete scale invariance power-law constraint competition morphological freedom innovation band collapse band recursive refinement MOSFET scaling breakdown FinFET transition kT/q thermionic limit subthreshold swing cell size invariance endosymbiotic constraint capillary length mountain height limit potato radius Euler buckling Hall-Petch crossover nanocrystalline inverse Hall-Petch turbulence onset Reynolds number pivot constraint hierarchy recurrence Fibonacci recursion golden ratio attractor Diophantine robustness V(L) Fourier series renormalized rescaling log-linear intersection scale variance peak criterion primitive subunit anchor

Key Findings

—The biological question "why don't cells make smaller cells" and the semiconductor question "why did MOSFET scaling require architectural pivots rather than continued shrinks" are structurally identical problems. Both are instances of recursive refinement being forbidden at a constraint-competition boundary — the same mathematical mechanism in two unrelated domains. The shared diagnosis is constraint physics, not design failure.
—Discrete scale invariance (DSI) — the condition V(λL) = V(L) for some fixed λ — forces V(L) to be log-periodic. This is a mathematical consequence, not a parameterization choice. If DSI holds for a domain's constraint family, log-periodicity follows by Fourier analysis of the resulting periodic function. The first harmonic gives the working approximation; additional harmonics are corrections that the model treats as predictive tests rather than fit parameters.
—Across six empirically distinct domains spanning fluid mechanics, structural engineering, materials science, planetary science, semiconductor scaling, and cell biology, constraint-competition crossover scales computed from first principles correctly identify the centers of empirically observed collapse bands (pivot scales where regime changes are documented). In each case, L₀ is anchored to the system's primitive subunit rather than fit to data.
—The golden ratio φ ≈ 1.618 is a candidate universal attractor for the scale ratio λ via two independent arguments: Fibonacci recursion in minimal-memory architectural hierarchies (where the only information available for the next scale is the previous two) and Diophantine robustness (φ is extremally irrational, making it maximally resistant to resonance among constraint-exponent intersections at rational multiples of ln λ). These are structural arguments, not fits.
—The model is falsifiable in a specific, non-vacuous sense. It predicts that in domains where DSI holds, sequential pivot scales should be log-periodically spaced with ratio λ extractable from any two consecutive pivots, and that subsequent pivots should occur at integer multiples of ln λ in log-scale. A domain where this spacing fails is evidence against DSI for that domain. Transistor scaling provides a concrete test: FinFET, GAA, and 3D stacking pivots should show recoverable λ if the semiconductor constraint family satisfies DSI.

Formal Statement

Log-periodic morphological variance V(L) as forced by discrete scale invariance under competing power-law constraints.

// Morphological variance (first-harmonic working form) V(L) = sin( (2πk / ln λ) · ln(L / L₀) + φ₀ ) // Parameters — L₀ and k are not fit knobs L₀ = primitive subunit scale // empirically anchored (e.g., silicon lattice: 5.43 × 10⁻¹⁰ m) k = harmonic index (integer, 1–3) // upper bound on simultaneously comparable constraints λ = discrete scale ratio (λ > 1) // conjectured universal attractor: φ ≈ 1.618 (golden ratio) φ₀ = phase convention // fixed by first observed pivot; subsequent pivots become tests // DSI condition that forces log-periodicity V(λL) = V(L) for all L > 0 // Interpreting V(L) V = +1 → innovation band // many feasible morphologies; constraint competition permits diversity V = −1 → collapse band // one constraint dominates; geometry pins to few modes V = 0 → constraint pivot // crossover scale; C_i(L) = C_j(L) here