Skip to content

Prove exact Clifford spectrum in the Bell central sector - #7789

Open
aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-096
Open

aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/consecutive-result-096

Conversation

@aliromman

@aliromman aliromman commented Nov 19, 2026 •

Copy link
Copy Markdown
Contributor

Enabling algebraic theorem: for every four-qubit Clifford tensor width f >= 14 with f = 2 mod 4, the exact minimum and maximum normalized Gram eigenvalues are H(-t) and H(t), where t=2^(-(f-2)/2) and H is an explicit degree-eight polynomial. Both endpoint eigenspaces have dimension 14,400. This supplies sharp coefficient-recovery and condition constants for the full alphabet. No new kappa or physical decoder is claimed.

Read the proof and explicit polynomial, all spectral blocks and interval certificates, checker, source pins and credits.

The scalar character iI -> -1 requires both actual Bell-line inertia groups. The odd-width formula cannot be substituted into this sector.

Stabilizer Quadratic orbit size Irreducible blocks
O8-(2).2 120 60
O8+(2).2 136 67

All 127 degree-eight spectral polynomials are computed in two ways: directly from stabilizer class sums and through the complete restrictions from Sp(8,2). All 1,143 coefficient identities agree. Dickson twisting reflects the full spectrum. The extreme rows are the minus group's trivial and Dickson rows; their labels swap with f mod 8, while the endpoint values remain H(-t) and H(t).

There are 251 strict endpoint comparisons on 0 < t <= 1/64, with 2,008 strictly positive Bernstein coefficients. A separate certificate proves H(-t)>0. Exact expansion checks every polynomial identity, so this covers the infinite width range. The exact Gram condition number is H(t)/H(-t); the tensor-map condition is its square root; the recovery norm is 1/sqrt(H(-t)). The sharp leading errors are 272t and 136t for the two condition numbers. Subalphabets inherit valid bounds without a subset sharpness claim.

Use Python 3.12 and the standard library:

python3.12 -B scripts/research_package.py --package research/clifford-bell-sector-spectrum

Both modes print this prefix and elapsed seconds:

PASS_CLIFFORD_BELL_SECTOR_SPECTRUM 127 blocks 251 order proofs 1143 restriction identities

The unchanged stabilizer and threshold source suites run. Complete trace and second-moment polynomial identities pass, including the even-sector Pauli-lift factor. All 22 widths 14,18,...,98 pass multiplicity, dimension and independent quadratic-orbit norm checks. A literal one-qubit calculation checks all 576 products and four complete Gram spectra. Six altered-data/domain controls reject. All 40 source pins, registered normal/optimized replay, syntax/indentation, committed bytes/history/modes, export and sealed read-only Git-free serial replay pass. 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.

Ali Romman (@aliromman) supplied PR #7771's Bell extensions and stabilizer data, PR #7788's all-width phase and polynomial method, and PR #7772's projective normalization. Nick Petalas (@petalas) supplied PR #7747's quotient projection. alephnan (@alephnan) supplied PR #6714's Pauli-sector framework. All retained source notices preserve the ATLAS, CTblLib, MatShep, Taylor and predecessor credit.

Published character tables and class fusions remain mathematical inputs; the archive is not re-extracted. The full conditioning and odd-spectrum source suites are not invoked: only their generic arithmetic helpers are used. No full four-qubit Gram, full three-qubit suite, selected cold replay, Lean or Docker is supplied. Widths divisible by four remain outside scope. Native preparation, physical inverse, precision, routing, owned recovery, stock and complete caller costs remain unknown. No independent human or proof-kernel acceptance is claimed. T(4)=13, all five assumptions and the selected frontier remain unchanged.

Base: canonical main e54523517d0f03e58352e647c403a5af3e525a57, priority revision 8. Prepared by Ali Romman with substantial OpenAI Codex assistance in research, implementation, verification and writing. Apache-2.0. Canonical claim.

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