Skip to content

Prove the sharp physical prefix range on the role-1256 graph - #7725

Open
petalas wants to merge 2 commits into
CrocSwap:mainfrom
petalas:research/w1-1118b-frontier-step
Open

petalas wants to merge 2 commits into
CrocSwap:mainfrom
petalas:research/w1-1118b-frontier-step

Conversation

@petalas

@petalas petalas commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Summary

The sharp physical component range of an ordered prefix on one nine-axis tile of the role-1256 graph is 51,863,149/8. This is a scoped proof with a finite certificate. It covers independent real and imaginary entrance components in $[-1,1]$. The real output at address zero attains the bound after donor 103886, the eleventh term of row 156140 in chart 127. Proof · Receipt

For each entrance, sum its signed scalar contributions, take the absolute value, and multiply by its chart gain $2^{\lceil d/2\rceil}$ at Fourier rank $d$. The scalar maximum from #7709 times the worst nine-axis gain gives $53{,}895{,}757/8$. Keeping the individual chart gains tightens that ceiling by 254,076.

Evidence and reproduction

The checker visited 172,414 prefixes in 7,644 rows. A backward rational expansion of the winning prefix agreed on all 448 nonzero entrance coefficients. Evaluating exact adjoint frame rows gave the attaining physical signs. aliromman posed the chart question and supplied the normalized prefix method. Clear1nsight provided the graph and physical chart contract from SYLforge's parent word. Original claim

From the repository root, these commands ran locally on arm64 macOS with CPython 3.12.15. The standalone verifier took 1.71 seconds and the package replay took 2.07 seconds.

PYTHON=python3.12
PKG=research/relay1256-physical-prefix-range
"$PYTHON" -I -W error -B "$PKG/verify.py"
"$PYTHON" -B scripts/research_package.py --package "$PKG"
"$PYTHON" -B scripts/check_research_commit.py --package "$PKG" --check-source-history --check-modes

The runs returned PASS_RELAY1256_PHYSICAL_PREFIX, PASS_REGISTERED_RESEARCH_PACKAGE and PASS_RESEARCH_COMMIT_BYTES. The final check covered 19 files, their committed bytes and executable modes, and base ancestry. The reproduction commands accept any CPython 3.12 executable.

Limits

The proof covers exact row accumulators on a single forward tile. Active tensor repetitions, paid-child interiors, numerator temporaries, rounding, correlated caller inputs and a complete root-width budget remain open. The selected cold replay, full predecessor suites and formal proof were unrun. The package gives no new κ or supplier admission.

Open next questions

  • Which prefix maximizes the physical component for repeated active chart factors?
  • Can a signed caller stage realize the local extremum under its entrance correlations?

Related open work

  • #7704 supplies the fixed role-1256 graph, operator phases and physical entrance conventions used here.
  • #7709 gives the normalized scalar maximum. Its extremizing location also maximizes the physical component under this input contract.
  • #7708 recodes the role-1256 graph. Its coefficients and prefixes change.
  • #7719 checks the recoded bank bases from Conditional κ ≥ 0.00374039009039 (+0.27743% over #7704) by recoding the role-1256 relay cuts #7708 against their parent.
  • #7712 splits three helpers on the recoded graph.
  • #7713 changes the helper interface on the recoded graph.
  • #7702 bounds integer temporaries for row-atomic arithmetic on the different chart-595 graph.
  • #7700 supplies the native parent word for the role-1256 contraction.

CI notes

This PR carries the shared CI-doc and replay-gap sync used by other open PRs. Earlier passing run

All six current-status tests and the registered historical replay-gap check passed locally. All three hosted checks passed on 72986db329dc83b77b9e69b25396c3da6d1a94bd, including the plan, the physical-prefix replay and the historical replay-gap replay. CI run 44369383245

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