PureOne's picture
Publish verified v5.0.0 research release
b6f1c7f verified
|
Raw History Blame Contribute Delete
10 kB

Verification scope and reproduction — v5.0.0

Eve's Impossible Complex Frequencies and Perfect Number Quantization Riemannian Geometry
Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki

The release contains a 105-page manuscript with 81 numbered theorem, lemma, proposition, and corollary statements, comprising 31 theorems, 18 lemmas, 16 propositions, and 16 corollaries. The supplied v4 manuscript had 35 pages and 25 such statements. This is a statement inventory, not an assertion that all results are historically new. The manuscript contains nine figures.

The complete integrated Python/C++ verification and the separate isolated wheel installation check pass in the recorded environment. Written proofs, exact finite checks, numerical residuals, file integrity, and visual page review have separate roles.

Recorded evidence

Layer Executed result Evidence
Python tests 260 passed; zero failures, errors, or skips pytest.xml
Independent C++20 atlas 2,024 candidate matrices; 1,092 primitive matrices; 100 unimodular matrices; 10,920 matrix–modulus cases; 564,761 assertions cpp_report.json, exact_atlas.csv
Python cross-check of C++ output All 10,920 labels; 4,368 exact unit-weight minima; 200 certified lattice cases; 2,304 Smith subsystem cases python_report.json
New exact suite 2,101 explicit exact predicates; 30 moment-only reconstructions; 1,633 moment entries; three support-only reconstructions; 20 subsystem outputs v5/report.json
New numerical suite 2,555 explicit numerical predicates, separately labeled v5/numerical
Observation-only CLI replay Both moment-only and support-only command paths pass prepack_validation.json
Isolated wheel installation Version, author metadata, module origin, tomography, support inverse, Q6 switch, and refined tail certificate pass wheel_smoke.json
Manuscript 105 rendered pages, resolved references, no overfull boxes, exact title/author metadata, reviewed page layout pdf_preflight.json, manuscript_inventory.json

The C++ program independently audits the inherited arithmetic domain. It does not independently implement all new tomography, differential, or spectral results. An assertion count includes several checks within a case and must not be interpreted as a theorem count.

The new experiment counters likewise count explicit predicates or evaluations. Repeated evaluations across different configurations remain separate operations. The report states the denominator of each counter. No truth probability is inferred from these counts.

Exact observations, inputs, and source fixtures

The 30 files in tomography_inputs contain only dimension, atom bound, optional nonconstant mass, and rational moment entries. Source frequency fixtures are saved separately. The decoder does not receive those source matrices. It reconstructs a reduced model and revalidates every supplied moment.

The three files in support_inverse_inputs contain labeled distinct-image support coefficients, dimension, row count, and an external bound on every positive Smith invariant of every row subsystem. The bound is not automatically inferred from finite probes. The decoder explicitly returns supplied_bound_verified: false.

The worked depth_two_H4.json input uses only Q = 2, 3, 4, 5. It recovers the empty, first-row, second-row, and complete subsystem invariants (), (2), (4), (2). A separate counterexample shows how an unprobed torsion prime can conceal a false bound.

Exact certificate files use integer, rational, or Gaussian-rational arithmetic. Numerical files are stored under results/v5/numerical/ and state binary floating-point or 80-digit arithmetic. High precision is still numerical arithmetic, not an interval proof.

Extremal and stabilization checks

The exact envelope checks include the critical powers

[ P_c(4)=3/8,\qquad P_c(6)=137/452. ]

Both extremal architectures are checked at each crossing using complete exact small-alphabet distance enumeration. The general two-endpoint theorem and its discarded-branch margin have analytic proofs in the manuscript. A separate numerical illustration considers Q = 2 through 120 and 1,487 powers; it does not establish the universal result.

Six exact refined tail certificates are included:

Example Inherited sufficient Q0 Refined sufficient Q0
Frequencies (2,3) 31 21
Quartic selector 13 8
Mixed arithmetic 21 9
Triangle 10 4
Square unimodular 4 4
Sextic selector 57,461 3,067

The new suite checks 15 complete small grids for stabilization and 20 complete moving-wall grids, totaling 23,639 nonzero phase-difference evaluations. The sextic example uses three high-precision comparisons on its exact six-vector candidate shell. Its full 3,067² phase grid is not enumerated. The proof, relation identity, rational bounds, and certified nine-point enclosing box establish its all-tail result.

The inherited suite additionally reruns twelve complete numerical grids around four older thresholds, twelve exact rational metric realizations, and the cographic K5 example at exact-energy alphabets.

Spectral checks

The new 24-case spectral suite builds the dense magnetic matrices using the inherited edge implementation, then independently compares the new sector decomposition and calibrated reconstruction. The largest dense matrix has dimension 40. The theoretical trace formula is not substituted for a measured dense trace to manufacture a recovery test.

Observed maxima include approximately 4.49e-13 absolute calibrated recovery error, 9.77e-15 absolute eigenvalue-list error, and 1.03e-14 relative determinant error. These are observed residuals in the stated finite corpus, not certified uniform bounds. The inherited 24-case spectral suite is retained and rerun separately.

Reproduce the release

From the extracted release root:

python -m pip install -r requirements.txt
python examples/demo_v5.py
python tools/verify_all.py --output ../eve-verification.json
python tools/smoke_wheel.py --output ../eve-wheel-verification.json

The integrated command checks the file manifest when present, copies the release to a temporary directory, runs all tests, rebuilds the C++ atlas, reruns both experiment suites, replays two observation-only CLI inputs, and runs both demonstrations. It does not rewrite the supplied evidence.

Full arithmetic scope requires a C++20 compiler. CXX can specify a compatible compiler executable. --skip-cpp uses the included atlas and reports partial Python scope. The provided Windows launchers are conveniences; the recorded execution was on Linux.

Run without -O or PYTHONOPTIMIZE. The wrapper rejects optimized execution because inherited experiment predicates include assertions.

The wheel check installs with --no-index --no-deps into an isolated target, reuses the explicitly located installed NumPy/SymPy/mpmath dependencies, and verifies that the imported evefreq comes from the target rather than the source tree. It tests packaging in the recorded environment; it is not a claim to have tested every supported platform.

Replay observations directly

python tools/experiments_v5.py --replay-moments results/v5/tomography_inputs/two_dimensional_mixed_unknown_mass.json
python tools/experiments_v5.py --replay-support results/v5/support_inverse_inputs/depth_two_H4.json

Regenerate figures and PDF

python -m pip install -r requirements-figures.txt
python tools/make_figures.py
python tools/make_figures_v5.py
python tools/build_pdf.py

The PDF build additionally requires a LaTeX distribution. It performs three compilation passes and rejects unresolved references or overfull boxes. The supplied release was also rendered with Poppler at 100 dpi and reviewed page by page. All 105 pages have readable content; no pages were inserted merely to reach the requested length.

Environment, budgets, and integrity

The recorded environment uses Linux, Python 3.12.14, NumPy 2.3.5, SymPy 1.14.0, mpmath 1.3.0, pytest 9.1.1, and Matplotlib 3.10.8. requirements-verified.txt records those versions without claiming to be a cross-platform lockfile. Ordinary installation requirements remain in requirements.txt and pyproject.toml.

Matrix-subset enumeration, full support reconstruction, moment tables, direction searches, and lattice/grid enumeration have explicit budgets. Exceeding a budget raises an error; it does not silently convert a partial search into a complete certificate. Complexity can be exponential.

The prepack report was generated before the final hash manifest, so its integrity field records that stage. After completing all files, the final manifest is generated and checked separately. It covers the supplied bytes, including the prepack report itself, without a circular self-hash. The manifest cannot include its own hash.

Rebuilding figures or the PDF can change file bytes and metadata. A mismatch then means that the local files differ from the sealed release. It does not by itself prove a mathematical error. Preserve the original archive if comparison against released bytes is needed.

The release has written proofs and reproducible internal checks. It does not claim proof-assistant formalization, independent expert acceptance, exhaustive historical-priority clearance, or a probability of universal correctness. See PROOF_MAP.md for review entry points and CLAIMS_AND_PRIOR_ART.md for the claim boundaries.