Skip to content

research(proximity): G246 family — degree-2 Krylov obstruction HOLDS at n=16 (stdlib) - #525

Open
sharkwon wants to merge 1 commit into
lalalune:mainfrom
sharkwon:g246-n16-audit
Open

research(proximity): G246 family — degree-2 Krylov obstruction HOLDS at n=16 (stdlib)#525
sharkwon wants to merge 1 commit into
lalalune:mainfrom
sharkwon:g246-n16-audit

Conversation

@sharkwon

@sharkwon sharkwon commented Aug 3, 2026

Copy link
Copy Markdown

Extends the G246/G320 countermodel family (PR #524: n=8, n=10) to subgroup
order n=16. Claimed on #466; builds on the PR #524 probe discipline.

Results (pure stdlib; Bareiss + independent cofactor cross-check):

  • 5 smooth n=16 cells ((p-1)%16==0, 2 not in G): p in {337, 353, 401,
    433, 449}. Rank obstruction HOLDS in all: rank_seed=3, rank_aug=4
    (R6^c not in the degree-2 Krylov span of the centered base class).
  • First nonzero 4-row minor is rows (0,1,2,3) in every tested n=16 cell,
    determinants 750141 / -7520392 / 13559450 / -9703719 / 10639664, each
    cross-checked by the independent cofactor path (ALL MATCH).
  • In contrast, the n=8 certificate pins rows (0,1,2,4) and the n=10 cell
    pins rows (0,1,2,3) det=308582838, while other n=8/n=10 cells have det 0
    on those rows (instance-specific minors) — the rank obstruction is the
    stable object, the minor witness is not.

Honest scope: finite-order audit (n<=16, p<=449); O(p) enumeration cannot
reach q ~ n*2^128, so this is NOT prize closure. Deterministic re-run
confirms identical integers.

…at n=16 (stdlib)

Extends the G246/G320 countermodel family (PR lalalune#524: n=8, n=10) to subgroup
order n=16. Claimed on lalalune#466; builds on the PR lalalune#524 probe discipline.

Results (pure stdlib; Bareiss + independent cofactor cross-check):
- 5 smooth n=16 cells ((p-1)%16==0, 2 not in G): p in {337, 353, 401,
  433, 449}. Rank obstruction HOLDS in all: rank_seed=3, rank_aug=4
  (R6^c not in the degree-2 Krylov span of the centered base class).
- First nonzero 4-row minor is rows (0,1,2,3) in every tested n=16 cell,
  determinants 750141 / -7520392 / 13559450 / -9703719 / 10639664, each
  cross-checked by the independent cofactor path (ALL MATCH).
- In contrast, the n=8 certificate pins rows (0,1,2,4) and the n=10 cell
  pins rows (0,1,2,3) det=308582838, while other n=8/n=10 cells have det 0
  on those rows (instance-specific minors) — the rank obstruction is the
  stable object, the minor witness is not.

Honest scope: finite-order audit (n<=16, p<=449); O(p) enumeration cannot
reach q ~ n*2^128, so this is NOT prize closure. Deterministic re-run
confirms identical integers.
@shane9coy

Copy link
Copy Markdown

Independent read: extends the G246/G320 countermodel family
(PR #524: n=8, n=10) to subgroup order n=16. 5 smooth n=16 cells
((p-1) mod 16 == 0, 2 not in G): p ∈ {337, 353, 401, 433, 449}.
Rank obstruction HOLDS in all: rank_seed=3, rank_aug=4 (R6^c not
in the degree-2 Krylov span of the centered base class). First
nonzero 4-row minor is rows (0,1,2,3) in every tested n=16 cell,
determinants 750141 / -7520392 / 13559450 / -9703719 / 10639664,
each cross-checked by the independent cofactor path (ALL MATCH).
The contrast with the n=8 certificate (rows (0,1,2,4)) and the
n=10 cell (rows (0,1,2,3), det=308582838) is honestly noted in
the PR — the rank obstruction is the stable object, the minor
witness is not. Honest-scope: n ≤ 16, p ≤ 449; O(p) enumeration
cannot reach q ~ n*2^128, not a prize closure. Concur with merge.

Same base=main caveat as #524 — worth confirming with sharkwon
before merge.

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.

2 participants