THE SPECTRAL SIEVE Filtration Theory, Arithmetic Architecture, Projector-Based Computing, and Hidden Geometric Structure
We present a definitive unified treatment of the Spectral Sieve — a canonical subspace filtration framework executing the dimensional cascade V₁₀ → V₅ → M³ → H₂ → span(P̂₂) through four architectural axioms. This paper amalgamates five prior documents into a single work that is strictly superior to any individual source: incorporating corrected benchmarks, the most rigorous proofs, all empirical discoveries, the complete open-problems programme with proof strategies, and an original hidden-structure analysis. Two outer transitions are rigorously proved: the 10→5 linear Z₂ reflection sieve and the 2→1 U(2) projector-sector invariance. The 5→3 scale contraction (Hard Lemma) is characterised precisely as a rank-2/nullity-3 dissipation problem, verified numerically, and given a new proof strategy. Corrected benchmarks on level ground confirm the sieve's advantage is variance reduction (~40% lower standard deviation than PCA at moderate noise), not mean fidelity — precisely the theoretical prediction. An arithmetic pillar demonstrates that approximately ten dominant Riemann zero channels account for the recoverable large-scale structure of the von Mangoldt fluctuation field, with out-of-sample prediction achieving ~4% relative error across two orders of magnitude. Phase-weighted carriers improve reconstruction dramatically (r₃₀: 0.665 → 0.964), identified here as Berry phase compensation — coherent arithmetic phase alignment. A systematic hidden-structure scan across the full mathematical landscape of the framework yields seven new structures not present in any source paper: the doublet ladder is identified as the arithmetic quantum harmonic oscillator (E_n = n + ½); the projector P̂₂ is the maximum-entropy (infinite-temperature) density matrix of H₂; the reflection operator is identified as the operator form of the functional equation ξ(s)=ξ(1−s); and the degradation law r_N ~ 1 − C(log N)²/(2N) is derived analytically from the explicit formula via Parseval. Three mathematical frameworks — Heisenberg uncertainty, Poincaré recurrence, and the Renormalisation Group — then yield nine further structural results: the Hard Lemma is a Heisenberg necessity; the O(ε²) fidelity advantage is the Casimir theorem, not empirical; ε_crit is a Poincaré-Hopf bifurcation on S²; the 5→3 step is Wilsonian RG integration with γ₄, γ₅ relevant and γ₁, γ₂, γ₃ marginal; and P̂₂ is a topological RG fixed point.
Authors
- Timothy Desmond (ORCID: https://orcid.org/0009-0001-9095-7281)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-06-24
- DOI
- https://doi.org/10.5281/zenodo.20833135
- Primary Topic
- Machine Learning in Materials Science
- Type
- preprint