Explicit Hermitian Hamiltonian for the Riemann Zeros from Modular Arithmetic Quantum Chaos and Multifractality from Z/6Z
José Ignacio Peinador Sala
TL;DR — The Core Finding Contrary to the initial expectation that arithmetic structures (e.g., the coprime-to-6 mask or the Möbius mask) generate distinct number-theoretic spectral signatures, this work provides numerical and formally certified evidence that the Non-Ergodic Extended (NEE) phase is entirely governed by the preservation of first-neighbor connectivity (d=1). Suppressing the d=1 hopping channel induces a topological collapse into strong spatial localization (D2≈0.288→0.045, a factor of 6.4), representing a single-particle spatial analogue of Hilbert Space Fragmentation (HSF). Furthermore, the deep arithmetic logic of the Möbius function acts as topological white noise, in compliance with the Möbius Randomness Principle (Sarnak conjecture): since 86.05% of the interaction variance is concentrated in short-range bonds (d≤5), deeper arithmetic channels are swamped by the finite-size GUE noise floor (σGUE∼1/N). Theoretical and Physical Framework Deterministic-stochastic hybrid Hamiltonian. Combines an exact Riemann–von Mangoldt diagonal potential (inverted deterministically via the Lambert WW function) with Power-Law Random Banded Matrix (PRBM) off-diagonal disorder. Sparse hopping constraints. Hopping is filtered by deterministic arithmetic sieves (coprime-to-6 mask Ξ6, square-free Möbius mask Ξμ, Liouville mask Ξλ) and controlled random Bernoulli ensembles. Single-particle Hilbert space fragmentation. Without d=1, adjacent spatial transitions (i→i+1) require higher-order virtual paths (i→i+5→i+1) that suffer severe energy detuning from the fluctuating diagonal potential. The one-dimensional metric graph shatters into decoupled spatial clusters. Coupling variance dominance. The power-law decay (d−1.5 in variance) concentrates 86.05% of the interaction weight in d∈{1,2,3,5}. Deeper arithmetic channels (d≥6) remain buried beneath the finite-size GUE noise floor. Parameter-free operating point. ν=0.75 (exact center of the critical window 0.5<ν<1); local resonance coupling analytically derived as ϵc=π2; Maslov phase 7/8. Mechanized Formal Verification in Lean 4 The algebraic, topological, and connectivity constraints governing the model are certified 100% sorry-free by the Lean 4 kernel (v4.34.1 with Mathlib4), across 13 Level N1 modules (23 theorems, 11 definitions, 34 total verified statements): WheelChannelStructure.lean — Confinement to active chiral sublattices (d≡1,5(mod6)). UnitCellMeasure.lean — Unit cell totient measure (∑d=16Ξ(d)=2 ⟹ φ(6)/6=1/3). ChiralPhaseAbsorption.lean — Invertibility of coprime residues modulo 6. KatoRellichRegime.lean — Self-adjointness domain window (1/2<ν<1). LocalResonanceCoupling.lean — Hybridization resonance condition (ϵc=π2). ChannelDensityContraction.lean — Strict density hierarchy (4/15<1/3<1/2<1). NonErgodicPhaseBound.lean — Non-triviality of the bipartite ceiling (1/4<1). PolyphaseFilterZ6.lean — Complete stopband annihilation for non-coprime channels. FirstNeighborPreservation.lean — Guaranteed base coordination degree (w1=2). DihedralQuantization.lean — Reflection involution and percolation threshold bounds. BipartiteInterferenceBound.lean — Certified upper bound D2≤1/4 under chiral equiprobability. ExpectedDegreeBound.lean — Mean degree lower bound (⟨deg⟩local≥2). FractalDimensionUpperBound.lean — Generalized Jensen inequality for the IPR and universal bound D2≤φ(m)/m. Empirical validation of Lean 4 axioms. Extensive GPU audits up to N=12 000 confirm that the chiral equiprobability axiom (c1=c5=1/2) holds strictly at the ensemble level (⟨W1−W5⟩=−0.000048≈0), with local mass fluctuations decaying monotonically as N→∞. Comprehensive Experimental Validation Matrix Observable Empirical Result Benchmark Physical significance Macroscopic alignment (R2) >0.999999 R2→1 Autonomous Weyl law tracking; zero empirical rescaling Bulk spacing ratio ⟨r⟩ 0.5891±0.0024 GUE 0.5996 / Poisson 0.3863 Intermediate non-ergodic extended statistics SFF ramp exponent γ 0.6086±0.0103 GUE γ=1 Sub-diffusive spectral ramp characteristic of NEE FSS thermodynamic limit D2∞ 0.113±0.003 Anderson D2=0 39σ rejection of Anderson localization Chalker scaling β=2D2 β=0.5549±0.0020 2D2=0.5562±0.0054 0.2σ agreement between transport and fractal support Multifractal curvature ΔD D0−D2=0.730 Ergodic ΔD=0 Strict convexity of Dq for q∈[−3,4]; width Δα=3.726 First-neighbor removal D2=0.0448±0.0009 Preserved: 0.2882 Sixfold collapse into strong localization (HSF) Short-range rescue (d=1) D2=0.2849±0.0007 Unforced: 0.0556 Forcing d=1 restores the NEE phase in random masks Topological pinning (ε=εc) D2no/D2with≈0.17 Healed ratio →1→1 Collapse cannot be overcome at the critical resonance Riemann zeros control (SFF) γzeros=0.94±0.12 GUE γ=1γ=1 Real zeros are ergodic; the NEE ramp is intrinsic to the mask Repository Structure and Reproducibility The codebase is fully vectorized, GPU-accelerated (PyTorch cuSOLVER in complex128), and audited via an automated SHA-256 manifest. The notebook regenerates every figure and table of the manuscript from a fixed random seed (SEED = 42). PRBM_Modular_Lean/ — Standalone Lean 4 (v4.34.1) project containing all 13 verified .lean files and lakefile.lean. Modules 00a–00c — System setup, forensic audit of the 100,000-zero Riemann dataset, and vectorized Hamiltonian engine (complex128). Modules 01–03 — Exact diagonalization at N=10 000; local level statistics P(s) and ⟨r⟩; spectral form factor ensemble averaging (M=100, N=15 000). Modules 04–06 — Channel-density scaling and finite-size scaling; FSS collapse of P(d2,N) across N∈{2000,4000,8000,12000}; AIC/BIC model selection. Modules 07–09 — Anderson transition sweep (ν∈[0.5,3.0]); systematic mask comparison suite; Möbius equivalence and density sweep. Modules 10–12 — Full multifractal spectrum Dq,f(α); microscopic GPU diagonalization at N=12 000; quantum dynamics (MSD, survival probability, Shannon and entanglement entropies). Modules 13–15 — Riemann zeros control SFF; coupling sweep over ε/εc∈[0.2,4.0]; chiral equiprobability audit ⟨Wr⟩ across residue classes. Module 16 — Automated high-resolution vector export (PDF/PNG 300 DPI) and cryptographic manifest generation (MANIFEST_SHA256.txt). Open-Access Dual Licensing Source code (Lean 4, Python, PyTorch): released under the MIT License. Free for academic, educational, and commercial replication. Manuscript, figures, manifests, and data: released under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Declarations Author contributions. Conceptualization, methodology, software, formal analysis, formal verification, investigation, data curation, validation, writing (original draft and revision), visualization, project administration. Competing interests. The author declares no competing interests, financial or otherwise. Funding. No external funding was received for this research. Ethics. Ethical approval was not required, as this research is theoretical and numerical and involves no human participants or animal subjects. Generative AI disclosure. Google Gemini and DeepSeek were used exclusively as auxiliary tools for language editing, adversarial logical review ("red-teaming"), and code debugging. They were not used to produce any original scientific content: all model design, numerical experiments, theoretical derivations, and interpretations are the author's own work. Every AI-assisted output has been reviewed and validated by the author. Scientific Epilogue "In the beginner's mind there are many possibilities, but in the expert's there are few." — Shunyu Suzuki What began as an inquiry into whether number-theoretic structures dictate quantum chaos revealed a simpler, deeper, and more universal physical law: geometry and local connectivity precede arithmetic complexity. The code, formal proofs, and data deposited here are provided to ensure complete auditability, verifiable down to the individual logical inferences of the Lean 4 kernel. Archival snapshot: October 2026 · Certified Level N1 · Built with Lean 4, PyTorch, NumPy, SciPy, and Google Colab. Version: v4 ( submission EPJB-D-26-00793)
•Mar 31, 2026•0 citations