Skip to content

Prove the global ceiling for simultaneous helper restrictions on the selected BIT word - #7733

Open
alephnan wants to merge 2 commits into
CrocSwap:mainfrom
alephnan:codex/bit-simultaneous-helper-ceiling
Open

alephnan wants to merge 2 commits into
CrocSwap:mainfrom
alephnan:codex/bit-simultaneous-helper-ceiling

Conversation

@alephnan

@alephnan alephnan commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

No simultaneous choice of block-local linear helper restrictions improves the selected BIT caller’s raw crossing. Over all 437 helpers in its nine boundary groups, in either word orientation, the supremum of price savings with a strictly positive complete margin lies between 0.0038525393934 and 0.0038525393935. PR7001 already reported this interval for the original caller. This contribution proves its optimality over the entire specified family. At the project’s 0.01 target, even the best signed margin is strictly between −2355.302 and −2355.301. The selected conditional multiplication saving is 0.00363005817. Proof, including the full-interval ceiling · Exact certificate · Selected application review

At price $r^t$, for $\log_2(5/3)\le t\le1$, the largest change in the complete 36-call margin is exactly

$$ 36\max{0,\ 1+4^t-6^t}. $$

Setting helper 341 to zero attains the second branch. It removes one rank-one and one rank-four payment per invocation and surrenders one rank-six target. An edit first improves the old margin when the price-saving parameter $1-t$ exceeds 0.2517782975220009…. The complete bill is already negative there. The proof separately bounds the entire remaining range, including this apparent opportunity.

Why the one-group exclusions do not add automatically. The forward bank at chart 50, clock 19881, contains $x_{331}+x_{332}+x_{333}$. The equations $x_{331}=0$ and $x_{332}+x_{333}=0$ belong to different groups. Together they make the bank reconstructible, while either equation alone still requires payment. The proof accounts for such interactions with chronological rank increments, projects the mixed final spaces onto the 410-dimensional group, and classifies the optimizing tuples. Predecessor’s one-group theorem · Receiver and pivot verifier

Reproduction. From the repository root, CPython 3.12 and its standard library suffice.

python3.12 -B scripts/research_package.py --package research/bit-simultaneous-helper-relations

Both registered checks pass. One recomputes all 636 rational frames, 2,374 directed paid chart pairs and 3,530 bank prefixes per orientation. The other solves the column nullspaces from a separate word interpreter, checks all 820 large-block coordinate prices in order, and directly enumerates 64,882 small-block subspaces across both orientations. The Windows CPython 3.12.2 replay, Linux CPython 3.14.7 replay, PR7001’s original BIT application, committed-byte/source-history preflight and a Git-free export replay passed. Independent implementation · Whole-invoice arithmetic · Replay guide

The construction grants all historical same-chart receiver banks and scalar XOR for free. Boundary equations are block-local, the paid forms and routes are fixed, and the two directions are analyzed separately. A changed causal delivery after a restriction is an open next step. The native all-column, owned-erasure, compiler, precision/tape and analytic-recovery assumptions remain inherited. Full selected cold replay and human or formal-kernel review are outside the present verification.

Canonical main and the fork were synchronized at e54523517d0f03e58352e647c403a5af3e525a57. This PR changes only the new research package. The upstream workflow ran 760 onboarding tests and reported one failure. The shared CI guide points to frontier6030 while the selected record requires make frontier7000-verify. Package replay was skipped. The same navigation failure reproduces on the base. The package selector identifies exactly this contribution, and the registered replay passes locally. A second Linux replay unpacked the final Git-free export with read-only permissions, ran both scientific checks, and confirmed all 15 file digests afterward. Upstream check log · Existing CI repair · Coordination handoff

Ali Romman supplied the single- and one-group bank models and posed the cross-group question. Jerry Chen / CubeJerry supplied PR7000 and PR7001, DreamingOfClouds the paired caller lineage, and zxv and the retained authors the native word. AlephNaN (@alephnan) prepared the proof and both implementations with substantial OpenAI Codex assistance. Predecessor and source credit

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.

1 participant