Skip to content

Show four-stage K=4 schedules miss the Theorem S bound over F2 (no new κ) - #125

Draft
unpaidmoon wants to merge 1 commit into
cursor/upstream-626e936-base-a140from
cursor/theorem-s-k4-s4-a140
Draft

unpaidmoon wants to merge 1 commit into
cursor/upstream-626e936-base-a140from
cursor/theorem-s-k4-s4-a140

Conversation

@unpaidmoon

Copy link
Copy Markdown
Owner

Change and mathematical scope

Scoped negative result in the F2 multi-pair stage model. No κ change and no selected-frontier update. The selected record stays κ = 781451/250000000 from certificates/selected-result.json (source PR CrocSwap#6030 and its manifest), with the five framework assumptions still in force.

Theorem S gives N ≥ 2K + ceil(2K/(s−1)) pair-calls for s stages on K bank pairs. For K = 4 and s = 4 the bound is N = 11, while parallel B1 chains attain N = 12 at s = 3. Over F2, in the model of research/multipair-schedule-bound (bridges quotiented out; a stage [[I, H], [E, I]] with HE = EH = 0), no 4-stage schedule reaches the goal at cost 11. Every 4-stage schedule therefore has N ≥ 12. Four stages do not beat three stages at this size, and the minimum over schedules with at most four stages is 12.

The proof is the equality-signature skeleton of Theorem S (18 edges, waste exactly 3) plus an exhaustive GL(X)×GL(Y) orbit search along those edges. Stage 1 is one orbit per rank signature. The four reachable stage-2 orbits have no legal equality child, so the goal is unreachable. Dimension counts still allow waste 3; the obstruction is F2 linear algebra.

Remaining obligations: the same question over a field other than F2, and attainment at the predecessor's historical s = 8 price. This package does not rerun the selected supplier.

Task and overlap

Validation

CPython 3.12, standard library only. Metadata pins and the registered hook passed:

python3 scripts/research_package.py --package research/theorem-s-k4-s4 --check-only
python3 scripts/research_package.py --package research/theorem-s-k4-s4

The hook reruns research/theorem-s-k4-s4/verify.py (about 40s here) and checks:

  • K = 2 exhaustive frontier: cost 6 at exactly 3 stages and exactly 4 stages, matching the published sched_search frontier.
  • K = 2 equality: direct search and orbit search both find a cost-6 witness; replay reaches (Y, X); a one-bit corruption of the last H misses the goal.
  • K = 3 equality: both searches find no cost-8 schedule, matching the predecessor's unrestricted negative search.
  • K = 4: 18 equality edges; stage-1 orbits of sizes 66150, 11025 and 66150; 23552 stage-2 states in four orbits; zero stage-3 equality children.
  • |GL(n,2)| for n = 2, 3, 4. Stabilizer enumeration matches brute force for every signature at K ≤ 3. Every K = 4 stabilizer element used for canonization preserves the standard flag, so distinct orbits are not merged.

Not run: selected-supplier replay, Lean, Docker, and any search outside F2. The predecessor's chain replay is cited for odd-stage attainment of N = 12 at s = 3; it is not re-executed here.

Credit

Theorem S, the stage model, the K = 2 frontier, the K = 3 negative search, and odd-stage chain attainment are from themilkman216, with substantial Claude assistance, in research/multipair-schedule-bound (PR CrocSwap#2052), generalising research/cross-shear-stage-bound and citing research/format-limits-corrected. The T74 prices there come from PR CrocSwap#1800 and its lineage.

Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.

Open next questions

Linked from research/theorem-s-k4-s4/README.md and TASK.json:

  • Does K = 4, s = 4 attain N = 11 over a field other than F2? The integer skeleton still reaches waste 3. A first step is the same four stage-2 transitions over F3.
  • Is the bound attained at s = 8 on the bank count used in the predecessor's historical T74 pricing? K = 4 does not decide that case.

Workflow feedback (optional)

The equality skeleton is small enough that the F2 decision is an ordinary Python orbit search once stage 1 is quotiented by GL(X)×GL(Y). The predecessor's note that K = 4 needs a compiled search can be retired for this one instance.

Open in Web Open in Cursor 

The F2 multi-pair model has no 4-stage schedule of cost 11. Chains already
attain 12 at s=3, so four stages do not beat three at this size. No new kappa.

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.

2 participants