Deterministic Geometric Decoupling: Cross-Component Isolation, Leakage Scaling and Plateau Dimension Law from k-Step Graph Theory
This paper adds one postulate only: transitions connect strictly lattice sites separated by a fixed step k. Decomposability and leakage then reduce to a single integer g = gcd(L, k): how many blocks arise, how long each block is, and whether cross-block operators commute are all read off from it, with no further structural assumption. Three conclusions follow: the number of components equals g exactly; each component has length L/g; the Hamiltonian is block diagonal, cross-component operators commute, and the cross-component OTOC vanishes identically at all times (an algebraic identity, not a physical discovery, with numerics used only as a workflow check). When a single cross-component link of strength ε is retained, the leakage grows as ε² in the early-time window (measured log-log slope 2.000) and holds up to ε of order λ; reconnecting the g components into one piece requires at least g−1 links forming a spanning tree on the quotient graph. Four closed-form quantities: the plateau value G∞=1+Σm cos²θm|φm(s)|² equals 5/4 for a source at a component endpoint and 3/2 at the centre, independent of k, L, component length and intra-component hopping amplitude, which combine into the dimension law G∞=1+d/4 (endpoint) and 1+d/2 (centre), verified for d=1 to 6 to within −0.10% to −0.14%, plus a change-of-initial-state prediction that outsiders can check themselves; the amplification coefficient of a single cross-component bond has a three-factor general form, matching exact diagonalisation to within 0.1% over 11 combinations of separation and source position; the isolation time is t*=2√θ/(ε√G), with its exponent fixed by the single matrix element of the stitching bond rather than fitted (theory −1, measured −1.024 to −1.032, consistent across three sizes), and five threshold values agree to within 0.12%; the leading asymptotic term (5/16)t² of f yields the closed form t*=4√θ/(√5ε). Parallel links split into three cases by landing point: same-site coherent t*∝m⁻¹, cross-component incoherent t*∝m⁻¹ᐟ², and different-site same-component which barely accelerates; the measured −0.693 lies between the two limits, as it must. Two readings independent of any threshold: the mutual information between two sets of sites in different components is strictly zero at ε=0 and increases monotonically with ε, with IAB/ε² between 2.2 and 2.9; and for a single bond joining corresponding sites of two components, the long-time occupation of the joined component is two-valued: 1/2 or 3/4 on a periodic ring, with the criterion 2(p−j₀)≡0 (mod n), and exactly 1/2 on an open chain if and only if gcd(p+1, n+1)=1; both are independent of ε. Leakage saturates to O(1) at long times, so the decoupling here buys a delay, not permanent protection. Three diagnostic traps: at strict decoupling the spectrum is g-fold degenerate so the level-spacing-ratio criterion breaks down; a zero without a positive control cannot be distinguished from a wrongly coded coupling; and data labelled "after decoupling" must use sizes with L a multiple of k. The main text keeps only three applications read directly off the theorems (error containment and the Krylov wall, with a common explanation), while extensions requiring new simulations are placed in the appendix; one conditional extension states that the blocking belongs to the transition graph alone, and that with arbitrary local interactions the conserved object is the Zn grading (n=L/g), provided the interactions are invariant under the k-step shift and periodic boundaries are used, with the criterion [H,U]=0 and with open boundaries breaking it (measured cross-block leakage 2.45%). What this paper delivers is a deterministic, reproducible, independently recomputable isolation platform that any group can check or refute line by line with small-scale numerics; its domain is graph connectivity, not spacetime emergence.
Authors
- Wenjun Luo (ORCID: https://orcid.org/0009-0002-2082-5651)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24420307
- Primary Topic
- Quantum many-body systems
- Type
- preprint