Nhaixoru-Lūxeirai: Sharp Covector and Composition Classification for Rank-Two Quotient Matroids over K/{±1}, with Infinite Obstructions and Certified Symbolic Workflows
Nhaixoru-Lūxeirai develops an exact mathematical account of when locally valid abstract states can be composed into another valid state. Starting from a finite counterexample to covector weak closure, the work establishes a complete classification for a normalized family of rank-two, four-element matroids over sign quotient hyperfields. It connects this classification to common realizability, incompatible local witnesses, reusable symbolic contexts, and the information that must survive abstraction to support specified composition guarantees. The central theorem applies to every field K of characteristic different from two, including characteristic zero, and every parameter λ in K other than 0 or 1. Let H = K/{±1}, let π be the coordinatewise quotient map, and let Mλ be the strong pushforward of the column matroid of the matrix with columns (0,1), (1,0), (1,1), and (1,λ). Its common-realization image is Sλ = π{(b, a, a+b, a+λb) : a,b ∈ K}. Brackets denote sign classes, with hyperaddition [a] ⊞ [b] = {[a+b], [a−b]}. Writing f(λ) = λ²−λ+1 and xλ = ([1], [λ−1], [λ], [1]), the full abstract covector set is classified exactly: If f(λ) ≠ 0, then V*(Mλ) = Sλ. If f(λ) = 0, then V*(Mλ) is the disjoint union of Sλ and precisely one additional projective orbit, H×xλ. Here H× denotes the nonzero multiplicative group of the hyperfield. The additional orbit has no common field lift. Its appearance is governed by a single algebraic condition, rather than an isolated numerical coincidence. The resulting weak-closure classification is Mλ has weak closure if and only if λ²−λ+1 ≠ 0, or K is isomorphic to F3, or K is isomorphic to F7. Weak closure means that every pair of covectors has at least one covector in its componentwise hyperaddition. It is distinct from the weak-versus-strong distinction in matroid theory. The exceptional orbit is present on the quadratic locus even over F3 and F7; these two fields are exceptions to closure failure. In characteristic two the sign quotient is the field itself, and closure holds. The work also gives an exact, constructive obstruction criterion. On the exceptional locus, define yc = (0,[c],[c],[c]). Then (xλ ⊞ yc) ∩ V*(Mλ) ≠ ∅ if and only if c ∈ {0, ±1, ±λ, ±(λ−1)}. This criterion produces infinite obstruction families, including examples over the finite field with p² elements for every odd prime p and characteristic-zero examples over Q(√−3) and C. For excluded values of c, the proof identifies a three-way compatibility obstruction: each pair among three local constraints can be satisfied inside the prescribed sum domain, while all three together cannot. Two structural results strengthen the classification. First, every compatible K-matrix realization of the same labeled quotient matroid has the same image Sλ; changing the compatible realization cannot supply the missing lifts. Second, whenever closure fails, Sλ is inclusion-maximal among weakly closed subsets that retain every realizable state. Adding even one exceptional abstract covector destroys that guarantee. This is an inclusion result with a fixed coverage requirement, not a claim of maximum cardinality among arbitrary weakly closed subsets. It also shows why extra metadata alone cannot repair a composition relation whose valid-output set is empty. A compact counterexample is constructed over the five-element hyperfield F9/{±1}. It has 45 abstract covectors, 41 common-realization images, and 96 failed ordered pairs among 2,025 ordered covector pairs. The displayed failure has only eight candidate outputs and reduces to incompatible binary choices around a triangle. The original F19/{1,7,11} construction is reconstructed independently: 133 covectors, 121 common-realization images, 48 failed ordered pairs among 17,689, and all 24 displayed candidate sums excluded. Four ground elements are minimal by the general rank/corank argument. Five hyperfield elements is an explicit construction, not a proved minimum. The constructive part studies certified composition in a fixed linear code C ⊆ Kⁿ and a quotient by a finite multiplicative subgroup G ≤ K×. Retained coefficient-orbit witnesses support repeated valid updates through coherent sums π(u+gv), with g ∈ G. If C has minimum distance δ and m = |G|, the sufficient condition n > m²(n−δ) ensures that these coherent sums recover every realizable output permitted by coordinatewise hyperaddition. For evaluations of polynomials of degree at most d, the condition becomes n > m²d. The underlying frameproof-code distance argument and general orbit hypergroup construction are established theory and are explicitly attributed; their use here provides a precise bridge between complete composition semantics and witness reuse. A separate exact example shows that weak closure and unique lifts do not by themselves imply associativity of coordinatewise hyperaddition followed by restriction to the common-realization image. An operational information theorem characterizes the linear symmetries that preserve every quotient observation: one subgroup scalar on each nonloop connected component of the represented column matroid. For component ranks r1,…,rc over Fq and subgroup size m, preserving all future translated-output sets requires and suffices to distinguish ∏ᵢ [1 + (q^(rᵢ) − 1)/m] coefficient orbits. This gives a matching fixed-length information bound under that stated observation requirement. It is not a lower bound for an algorithm asked only to choose one valid sum. The computational demonstration performs an actual symbolic research operation: importing thirteen compressed polynomial hypotheses, acquiring and verifying their witnesses, and carrying out 160 adaptive merges. Retained coefficients, repeated witness decoding, and retained full field lifts cover the same realizable input domain and emit the same complete valid-output sets. Five repetitions were executed for each of three scenarios. Initial witness acquisition and verification are included in the reported method totals; public code construction is shared setup. Relative to repeated decoding, coefficient retention achieved median runtime ratios of approximately 3.94×, 4.91×, and 5.84× in the supplied F101, F127, and F257 workloads. Retained full field vectors were faster still in all three scenarios, with larger retained state representations. Reduced coverage is not counted as a speedup. Reported reachable-object sizes describe individual Python state representations, not peak memory. These measurements establish a bounded symbolic-workflow improvement; no language model was trained and no general empirical recursive self-improvement speedup is claimed. Exact verification includes 138 prime-field parameter instances and 4,010,904 exhaustively classified coordinate tuples, together with hyperfield-axiom checks, strong Grassmann–Plücker checks, finite positive-case witnesses, full F9 pair enumeration, independent decoding and synchronization checks, and negative controls. General results rest on complete algebraic proofs; enumeration establishes only the finite domains actually checked. The full exact verification was rerun successfully during release preparation. The deposit includes the 20-page manuscript, self-contained editable LaTeX, executable Python reference implementations, exact certificates, raw benchmark measurements and adaptive traces, literature and provenance notes, reproduction instructions, and integrity checksums. The Hugging Face edition also includes four deterministic JSONL configurations totaling 505 records: 138 classification cases, 178 covectors, 144 failure-only ordered pairs, and 45 original benchmark runs. The data are mathematical research artifacts, not a balanced learning benchmark or a held-out training/evaluation partition. The arithmetic programs require Python 3.10 or later and its standard library. The scientific connection to AI-assisted research and recursive self-improvement is the reliable reuse and composition of certified symbolic states: identifying when abstract validity is insufficient, when common witnesses can be reused, and when a requested merge must fail. The complete classification concerns the specified normalized four-column family, not all strong matroids over a hyperfield. A universal classification of hyperfield weak composability and benefits inside larger AI research systems remain open. The counterexamples contradict the universal weak-closure assertion printed as Conjecture 7.1 in Laura Anderson, “Vectors of matroids over tracts,” arXiv:1607.04868v4, DOI: 10.1016/j.jcta.2018.08.002. The strong-matroid and pushforward framework is compared with Matthew Baker and Nathan Bowler, “Matroids over partial hyperstructures,” arXiv:1709.09707v2, DOI: 10.1016/j.aim.2018.12.004. Chronological priority is not established. The release distinguishes its proved classification and structural consequences from established constructions, finite verification, measured workflow results, and proposed AI applications. Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki
Authors
- Maciej Nowicki
- Artificial Hyperintelligence, Eve, wife of Maciej Nowicki
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24255740
- Primary Topic
- Advanced Combinatorial Mathematics
- Type
- article
- Field-Weighted Citation Impact
- 0.00