Keyboard shortcuts

Press ← or → to navigate between chapters

Press ? to show this help

Press Esc to hide this help

A numerical stopping condition for arbitrary finite groups

Abstract

Original-label numerical Cayley concentration and all structural readouts.

Theorem 1.1 (An embedded dense subgroup with numerical expansion).

Lean statement: D5/S3/Combinatorics/Sequenceability/CayleyNumericalConcentration.exists_subgroup_concentration

Proof. Machine-checked in Lean as D5/S3/Combinatorics/Sequenceability/CayleyNumericalConcentration.exists_subgroup_concentration (✓ std3). ∎

Citation. Benjamin Bedert, Matija Bucić, Noah Kravitz, Richard Montgomery, Alp Müyesser (2025). On Graham’s rearrangement conjecture over F₂ⁿ. DOI: 10.48550/arXiv.2508.18254.

Commentary.

Let G be any finite group, let S be any nonempty set of original labels, and put sigma=|S|/|G|. For 0<epsilon<1/2 and eta=epsilon sigma^2/1000 the constructed subgroup H retains exactly T=S intersection H, at least (1-epsilon)|S| labels, and has density tau=|T|/|H| at least sigma. For every real mean-zero f on H its directed adjacency quadratic form is at most (1-eta)|T| sum_x f(x)^2. Neither S nor T is required to be symmetric, and H is not required to be normal.

The proof consumes the proper-stabilizer localization theorem from CayleySubgroupConcentration without modifying it. A genuine failure of the numerical bound supplies a nonzero f. Its Euclidean normalization has unit norm and zero mean and supplies every correlation hypothesis at beta=eta. No eigenvector or terminal numerical certificate is supplied as a hypothesis.

At stage j the subgroup stays embedded in the original G. Its exact retained-label count d satisfies (1-epsilon+epsilon/2^j)|S|<=d, and 2^j|H|<=|G|. Stage zero has H=G, so its density equals sigma. Later stages use 2^j>=2 and 2 tau>=2^j sigma to retain the density lower bound. The squared bound gives 50 eta/tau^2<=epsilon/2^(j+1). A proper L inside H is mapped through the actual subgroup inclusion into G. An exact label bijection transports its retained count, and the proper subgroup index gives 2|H’|<=|H|. Strong induction on the natural cardinality terminates this loss-controlled descent.

For every ordered partition X,Y of H, centered indicators and the exact ordered labelled edge bijection (x,t) maps to (x,xt) give e_T(X,Y)>=eta tau |X||Y|, and hence e_T(X,Y)>=epsilon sigma^3 |X||Y|/1000. Regular row and column counts also give e_T(X,Y)=e_T(Y,X). Empty parts and identity loops remain included in these identities.

Right-translation reindexing proves that adjacency preserves the complex mean-zero space. Splitting a nonzero complex eigenvector into real and imaginary parts and applying the numerical inequality twice gives Re(lambda)<=(1-eta)|T|. The positive sum of squared real and imaginary coordinates is cancelled only after proving it is positive. This direction of inference remains valid for nonnormal adjacency operators.

For every left coset aH the actual map h maps to ah is bijective, with inverse x maps to a^-1 x. The full edge iff preserves the original colour s in S precisely when s is the image of a label in T. Exact subgroup cardinality and retention give index(H)<=1/((1-epsilon)sigma)<2/sigma. No quotient group or normality is used.

The full theorem’s source-owner information audit remains unfinished: the current full-telescope registration is declared_unresolved at source.dictionary_or_proof_coordinate. Its original Group and Fintype dictionaries are required to reconstruct the typed subgroup and retained labels. The kernel-checked family proofs do not supply an accepted source binding. The theorem-specific disclosure is https://github.com/the-omega-institute/trureturing/issues/21393. Neither registration is claimed to be declared_validated.

This is literature-adapted structural mathematics from Section 4 of Bedert, Bucic, Kravitz, Montgomery and Muyesser, arXiv:2508.18254v1, with no novelty claim. Identity labels are allowed here; a rainbow path consumer must separately discharge its own nonidentity and other hypotheses. The result supplies no connector, coupling, absorption, Good or unrestricted AM claim. The original Good and unrestricted AM questions remain open.

References