Skip to content

Prove sharp shared Fourier error on arbitrary unit inputs (no κ) - #7512

Open
petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w2-1113b-scoped-obstruction
Open

petalas wants to merge 1 commit into
CrocSwap:mainfrom
petalas:research/w2-1113b-scoped-obstruction

Conversation

@petalas

@petalas petalas commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Scoped written proof with exact finite corroboration. For every finite product of cyclic groups and every nonzero order-two translation, the shared phase-weighted three-pass Fourier routine has sharp relative induced bilinear l2 error

min_w ||B_w-B||_bil / ||B||_bil = sqrt(2-8/(3 sqrt(3)))
||B||_bil = N^(-3/2)
D(w)^2 = max_y {2-2 Re[conj(average_x c(x)w(x)) c(y)w(y)^2]}

The inputs are arbitrary complex unit-l2 pairs. Equal normalized characters attain the maximum columns; the inherited phase choice makes all columns equal and attains the optimum. This settles Open next question 2 of #6307.

The new norm reduction, extremizers and stronger-input theorem were produced by Nick Petalas. Nick Petalas #6307 retains credit for the coefficient-error lower bound, scalar moment inequality and phase choice, which are cited premises rather than re-verification targets. bbMisk #1477 and RossRichardson #940 retain the weighted-layout and convolution-contract credit. Cauchy-Schwarz, character orthogonality and elementary induced-norm arguments are classical. No new kappa or selected-frontier change is claimed.

Only research/shared-fourier-unit-input-error/ is added. Advance CLAIM.

Evidence

Before, #6307 established coefficient and basis-input errors and explicitly left the arbitrary-unit-input norm unresolved. After, the written proof gives the exact induced norm for every unit phase function and settles its sharp minimum.

  • A fresh literal exact-arithmetic three-pass interpreter and direct group-addition convolution check 1,058 phase assignments and 4,274 equal-character input pairs. Eighteen minimizing fixtures cover six group shapes through size sixteen. Another 72 unit-input pairs use distinct characters and real or complex superpositions.
  • Ten controls check three circuit mutations, maximum versus mean, and six exact-reason domain rejections. The maximum-versus-mean fixture gives squared coefficient error 7/4 and squared unit-input error 5/2, so the two norms differ for general phases.
  • Warning-strict normal and optimized direct runs pass with identical generated six-file hash-bound receipts, taking 2.56 s and 2.51 s locally. Both registered checks pass; the complete registered runner takes 5.27 s. Each command has a 120 s timeout. These timings bound replay cost and make no speedup claim.
  • Source metadata, exact committed bytes, owner-execute modes and declared source history pass at 0ab4b77b602ecf603a1c6b3e87f004efce82951a. Syntax inspection finds 25 functions, at most thirty physical lines. Optimized isolated replay also passes from a six-file Git-free export with an empty PATH.

From the repository root, using CPython 3.12 and the standard library:

/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/shared-fourier-unit-input-error/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -O -B research/shared-fourier-unit-input-error/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/shared-fourier-unit-input-error
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/shared-fourier-unit-input-error --check-source-history --check-modes

Another CPython 3.12 path may replace the displayed interpreter. Expected markers are PASS_SHARED_FOURIER_UNIT_INPUT_ERROR, PASS_REGISTERED_RESEARCH_PACKAGE and PASS_RESEARCH_COMMIT_BYTES. Verification writes no files. Receipts are generated, not hand-written. No predecessor scientific verifier was imported or executed.

Limitations and unrun obligations

The universal assertion is the written proof and uses #6307's cited lower bound. Finite replay corroborates the new reduction and upper bound over the declared small domains. Relative operator error divides by the norm of the desired bilinear map, not by the possibly zero output at each input pair.

Other translation orders, physical phase preparation, rounded arithmetic, routing, dirty restoration, storage, compiler closure, analytic recovery and the complete supplier bill remain unsettled. The five selected framework assumptions retain their status. Selected cold replay, predecessor scientific suites, Lean, Docker and a literal physical caller were unrun. No supplier admission or complete multiplication theorem follows. Independent human and formal proof-kernel review are unclaimed.

Open next questions

  • Find a matching sharp unit-input bound and attaining phase choice for a translation of order three. Start with its exact column formula; the nonbinary minimax inequality is missing.
  • Price a literal rounded caller using approximations to the attaining phases. Freeze its grid and norm contract, then charge coefficient preparation, routing, restoration and the complete recursive invoice.

No further question is reserved.

Related open work

No open PR studying this shared translated Fourier routine or its induced bilinear norm was found in the live targeted PR searches and the source-bound overlap snapshot. The direct predecessors #6307, #1477 and #940 are merged. Searches do not cover unpublished work or establish global novelty. No additional computations checked another contribution. The completion refresh read this account's seven other open PRs created since the advance CLAIM, #7511 and #7513 through #7518. None studies this shared translated Fourier routine or computes its induced bilinear norm, so no same-object entry or shared numerical comparison was added.

CI notes

CI did not certify this package. At unchanged head 0ab4b77b602ecf603a1c6b3e87f004efce82951a, run 43442255453, plan job 131942569460 failed in Onboarding helper tests. I read the completed checks once and inspected the actual job log. The step ran 760 tests with exactly one failure and 24 skipped:

test_research_current_status.CurrentCiNavigationTests.test_current_ci_section_matches_selected_record
AssertionError: '`make frontier7000-verify`' not found

The assertion is at line 58 of tests/test_research_current_status.py. On canonical base main e54523517d0f03e58352e647c403a5af3e525a57, certificates/selected-result.json names make frontier7000-verify, while docs/ci-verification.md still describes #6030 and make frontier6030-verify. I read both base files directly and confirmed that those files and the test are byte-identical at this PR head. This package did not cause the mismatch.

The dependent package replay job 131945244632 was skipped because plan failed. Run metadata independently confirms this exact head, failed plan and skipped replay. All passing scientific and source-binding checks above remain local evidence.

I independently inspected the failed-job logs for rohanarun's run 42905094314, dillona's run 42903313790, and petalas runs 43438133981, 43442114549 and 43442278203. Each log reports that same sole failing test, missing-command assertion, 760-test total and 24 skips. No fix, retrigger commit, push or workflow rerun was made; maintainers own the repair of main.

Merge Danger

Door: two-way.

Blast Radius: isolated.

Six new package files; no shared code, predecessor package, source archive or selected record changes.

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