Skip to content

Bind sharp radial phase-covariant Gaussian rounding to complete ports (no kappa) - #7698

Open
aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-037
Open

aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-037

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Result and scope

A sharp radial Gaussian provider now preserves phase covariance and the conjugate-inverse convention, with complete retained-port execution. The pair map attains the global radial supremum h/sqrt(2) while preserving exact sum, common translations and wire swap. It selects the individual-ring matching by conserved sum parity. Its negative provider is its componentwise conjugate and exact inverse.

Read the proof, exact receipt, provider, unchanged complete port and source pins. The actual retained two-/three-call C0/C2 programs keep all source monomials, additions, calls and both arbitrary dirty banks. Successful reverse execution returns both banks exactly, and the retained coordinate-error requests remain valid.

For N=8^f, set D=f*h*sqrt(N)/2. A complete native call changes field norm by at most D in absolute value. With entry norms X,Y, a k-call forward port has output bounds X+D and Y+3X+(k+2)D; the reverse bounds are X+D and Y+3X+(k-1)D. The pair and single-layer radial bounds are sharp. The complete-call and port envelopes are sufficient bounds, with no claim of optimality. Approximation error remains f*h*sqrt(N/2) per call.

No new kappa, selected-assumption discharge or accelerated native tariff is claimed. Fourth power and tensor commutation fail, with retained witnesses; the inverse reverses the literal stage order. A fixed multitape graph, routing, owned backing, recursive pricing, stock and selected caller admission remain open. All five selected assumptions remain. This completes the advance claim.

Evidence and direct implementation

Normal and optimized CPython 3.12.16 replay passed on macOS arm64. Checks cover 13122 pair directions, 52488 phase cases, 16900 complete ring points, all 139776 signed-input/provider cases at widths 2/3/4, and 68 sharpness indices through 4097 bits. Exactly 94080 signed cases fit and 45696 correctly reject output overflow.

There are 24 complete native calls, 672 exact fiber observables, 24 complete port executions with 60 actual paid cuts, 24 exact dirty returns and 12 aligned predecessor comparisons. Direct rational oracles check full port error and actual call inputs. All new radial envelopes pass. The primary port executions include 16720 pairs and 133760 stored range checks. Controls cover invalid data/signs, overflow, aliasing, wrong axes, wrong inverse order and excluded identities.

The runtime uses fixed comparisons, signs, parities and halves; it has no variable data division, growing table or matching search. Stored words keep the source width w, while a constant number of w+6 auxiliary words are charged. Including every pair, source wrapper, mix, range decision and owned auxiliary/route cleanup, the indexed bounds are O(fN*(w+f)) bit work and O(N*(w+f)) stored bits. These are constructive asymptotic bounds, not a measured speedup or tape-transition invoice. The old eight-word pair allocation is not claimed.

Metadata, source history, committed bytes, modes, fifteen-file export and archive integrity pass. A read-only optimized replay without Git or a usable PATH passes and preserves all sixteen payload/manifest hashes and modes. The copied ports.py retains its original trailing blank line; new-source whitespace checks exclude only that byte-exact retained file.

With CPython 3.12 and its standard library, from the repository root:

python3.12 -I -W error -B research/phase-covariant-radial-ports/verify.py
python3.12 -I -W error -B -O research/phase-covariant-radial-ports/verify.py
python3.12 -B scripts/research_package.py --package research/phase-covariant-radial-ports
python3.12 -B scripts/check_research_commit.py --package research/phase-covariant-radial-ports --check-source-history --check-modes

Expected markers:

PASS_PHASE_COVARIANT_RADIAL_PORTS
PASS_REGISTERED_RESEARCH_PACKAGE
PASS_RESEARCH_COMMIT_BYTES

Hosted CI did not certify this package. The plan job failed in the shared test_current_ci_section_matches_selected_record: the base guide names frontier6030 while the selected record requires make frontier7000-verify. The log reports 760 tests, one failure and 24 skipped; package replay was skipped. This package changes no shared CI file. Local replay remains the scientific evidence. Full predecessor package replays, selected cold replay, Lean and Docker were not run. Independent human and formal-kernel review are unclaimed. Whole-call range failure is not promised to be atomic; the inverse theorem covers successful range-valid executions.

Sources and credit

Prepared by @aliromman with substantial OpenAI Codex assistance. Ali Romman's PR #7696, pinned proof, supplies the global lower theorem and adapted ring code. It retains Nick Petalas's finite-witness credit and CubeJerry/Jerry Chen's half-grid and annular credit. The lower theorem is not newly claimed here.

Ali Romman's PR #6509, pinned proof, supplies the backend interface, actual retained code and adapted oracle method. Its coordinate requests and width proof are dependencies, not independently re-established results. @JoyboyBrian supplied the retained port work in PR #5485, with the PR #4743 native/error lineage. The copied notices remain intact.

@CubeJerry's PR #6758, pinned source, is prior phase restoration and complete-port binding for a distinct four-axis quotient map. No code or two-bit precision gain is transferred from it. Credit details identify all source roles and retained notices. Generic reversible rounding, phase restoration and this port interface are prior work. New work is Apache-2.0.

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Card uniform-transfer-audit, scope changed to this local provider and actual retained consumer. Current handoffs, source-specific intake, refreshed PR/Discussion overlap and the complete closed snapshot with its updated-order delta show no competing one-pair ring-provider claim. Sparse local and unpublished work remain coverage limits.

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