Skip to content

Prove the F_2^6 difference basis has order 14 - #80

Draft
unpaidmoon wants to merge 1350 commits into
mainfrom
cursor/f2-six-difference-basis-5dba
Draft

unpaidmoon wants to merge 1350 commits into
mainfrom
cursor/f2-six-difference-basis-5dba

Conversation

@unpaidmoon

Copy link
Copy Markdown
Owner

Change and mathematical scope

Scoped census. No new κ. The selected conditional exponent stays 60403/20000000.

A difference basis of (\mathbb{F}_2^6) is a set whose pairwise distinct sums are all 63 nonzero vectors. No 12-point or 13-point set is a difference basis. The minimum order is 14, with 53,340 normalized witnesses. One witness is {0, 1, 2, 3, 4, 5, 6, 8, 16, 24, 32, 40, 48, 63}.

On the six-source connector, prefix differences can cover every nonzero mask only when there are at least 14 distinct prefixes. Revisits shrink that set, so every covering closed walk is simple and has length at least 14. The set-level query count (2 \cdot 63 + |P|) is at least 140, not (2 \cdot 63 + 12). The 14-point set is not a walk and not a priced program, so 140 is not shown to be attained.

Task and overlap

Base fbc9ea0249746acca4f504d4fda2b84e0e36cf0d. Card connector-alphabet-change, scope refined. Package research/f2-six-difference-basis.

This is item 10 of the ordered eligible list in the package README. It answers open question 2 of research/connector-triple-query-floor. It does not rerun that package's nine-point census and does not search its length-10 triple leaf question.

Nearest related work: Aaron Wang's research/connector-trio-cycle25-seed1b82df excludes one twelve-call profile (eleven singleton edges and one closing triple). This census excludes every 12-point and every 13-point set, whether or not the set is that profile. Ross Richardson's research/universal-connector-boundary supplies the paired-readout model.

Validation

CPython 3.12 and cc, from the repository root:

python3.12 -B scripts/research_package.py --package research/f2-six-difference-basis

Observed:

PASS_F2_SIX_DIFFERENCE_BASIS min_order=14 cover12=57 cover13=62 sets12=4187106 sets13=36288252 sets14=264385836 hits14=53340 query_floor=140 no_new_kappa=true
PASS_F2_SIX_DIFFERENCE_BASIS_INDEPENDENT hits12=0 best12=57 order14=1
PASS_REGISTERED_RESEARCH_PACKAGE research/f2-six-difference-basis outcome=completed

The twelve-point census is pure Python. The thirteen- and fourteen-point censuses are the compiled loop in census.c. The two coverage routines agree on all 4,187,106 normalized twelve-point sets and on the recorded extremal witnesses. Controls: the (\mathbb{F}_2^3) predicate finds order 5 and rejects order 4; a coordinate hyperplane misses the odd coset; deleting any point of the order-14 witness drops the cover.

Not run: Lean, Docker, the selected cold replay, and any search for a closed walk.

Credit

The question is open question 2 of unpaidmoon's research/connector-triple-query-floor. The five-source model is Ross Richardson's. The twelve-call profile exclusion is Aaron Wang's.

Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0. No independent human review and no formal-kernel acceptance are claimed.

Open next questions

  1. Does a simple closed walk of length 14 realize a 14-point difference basis with legal connector steps? First order the recorded witness so every edge has weight at most 3 and the last vertex returns to the first. An ordering is not yet a priced program.
  2. Does any closed length-10 five-source walk with one triple and multiplicity (4, 2, 2, 2, 2) attain leaf bill 120 on prices (1, 4, 10, 16, 28)? That remains open question 1 of the triple-source floor.
Open in Web Open in Cursor 

Ryanson Jonathan and others added 30 commits October 21, 2026 17:48
…ur section

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s' orientation

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…bles

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…2..5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… GL_4(F3)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… fields

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…L = 26, nested through L = 22

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One hundred twenty-five of the 216 targets meet the representation rank with a coordinate word, and eight of them have sharp floor 6.
On the CrocSwap#3954 scaffold, Pi P_enc Pi matches neither the adjacent data charts nor the identity-to-full-F helper transition, and every nonnegative price of the exchange stays over capacity by at least 44096.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
The size section pins log2 T within two bits of p/6. Charging one polynomial number of full-array factors is not a 2^{O(p)} cost, while one factor of T does determine an explicit working-precision coefficient.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Under CrocSwap#4141's digit bound and CrocSwap#3667's S_leaf=1 hypothesis, T(s) is a sufficient coarse-depth coefficient and the published reading-(a) value is exactly 80*T(s). The selected stage coefficient stays null.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…t axis

Scoped layout lemma for the CrocSwap#4100 Gaussian caller: the exposed field pads by
a factor strictly below two, the short-field transpose uses ceil(log2 a)
stages, and deleting the pad before the next axis keeps the live multiplier
below two. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
The CrocSwap#4100 dyadic carve keeps at least W retained coordinates between interiors. In boundary order the Schur complement then has bandwidth at most 2W-1, matching the one-interval cap and the written O(W) claim. A shorter separator lets the bandwidth grow with the number of interiors.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
At the pinned CrocSwap#4088 profile, one rank-(r*f) child keeps the scalar moment negative, while f copies of each scalar child make it positive for every integer f>=2. The recipe records stay scalar. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Price CRT reconstruction and four rotation networks by explicit sequential
bit-steps, and show they fit 2^{4 ell}(p+2)^{16} for every p >= 2 and ell >= 1
under the stated width and batch caps. The displayed 2^{-G} comparison follows
with the same constant. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
A square i-call block forces at least twice as many j-calls, so the inherited 11t/5 bound is not attained on the rays t_i=4 t_j and t_j=4 t_i, and the (7,1) count witness (7,13) is unrealizable. The other 20-call splits stay open. No new kappa.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
…th 26

The (5,24) census still fires (28 masks; 15 and 23 fall silent again), so
with the flat cap the answer is 26 and the n = 5 firing set is not monotone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…0 and 282/100.

Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…kappa)

Checks vendor/FINE-PRIME.md section 1 at PROOF.md's epsilon = 1 - 10^-30:
cited Trudgian Theorem 2, exact theta range, dominance and its explicit
threshold, search cost, and small exact prime counts. The step holds as
stated; CrocSwap#4029's d <= ell^334 domain does not cover this caller.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…a uniform supplier

Count the completed transforms that helpers leave on their own arbitrary
inputs as productive jobs of mixed rank, instead of restoring them. The
p3 controller's complex mixed root reaches 0.0023051. A finite integer
compiler built on Theorem 1G (CrocSwap#4268) turns the mixed rule into an
ordinary uniform supplier at an explicit rank. A delayed BIT cover with
one common weight, new precision and storage constants, and a complete
nine-row outer bill then give κ = 0.0023 with exact minimum slack
98849/50000000000. This is +32.87% in exponent saving over CrocSwap#4181.

research/kappa-rebuild holds the frozen research checkpoint, with its
claim bindings, constructions, audits and literature notes.
research/kappa-0023-productive-jobs holds the review package: the PR
description and bibliography, the import manifest, the sequential replay
runner and comparison controls, and fresh evidence. The remaining
research/kappa-* directories carry the exact predecessor dependencies and
their notices. Scoped .gitattributes entries preserve the delivered bytes.

Based on canonical main 12fa8c5 (priority revision 7). A fresh replay
passes 23 scientific stages and 59 artifact comparisons. The claim is
conditional on inherited premises and awaits external mathematical
review. The winning templates are specified but not materialized.
jwr404 and others added 30 commits October 21, 2026 18:05
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…rerun filled)

Follow-up to CrocSwap#3154. K579T Theorem K5, R579b Lemmas K5^-/K5^-2 (two implementations each), the non-root type list, the DEG2 mask model with a soundness proof, DEG2 implementation A (six-sets 0) and implementation C with the fixed oracle: six-set count 0 over 64 rc=0 shards covering every (type, orbit-rep shard) exactly once (coverage.py), K=3/K=4 controls found, oracle 400/400 vs impl A. No kappa claim.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
…ew scope [skip ci]

Scientific validation and selected-frontier admission remain open.
No normalized 12- or 13-point set covers every nonzero difference, and 53340 normalized 14-point sets do. The six-source set-level query floor is 140.

Co-authored-by: unpaidmoon <unpaidmoon@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.