Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions research/sliding-window-inverse/.gitignore
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
__pycache__/
*.pyc
3 changes: 3 additions & 0 deletions research/sliding-window-inverse/INPUTS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
{
"inputs/gaussian_check_setup.py": "46e234fbadbbf934cb02f609d7aa14355789c42f82e384cf624d31e57bd05a3a"
}
109 changes: 109 additions & 0 deletions research/sliding-window-inverse/LIBRARY-BRIDGES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,109 @@
{
"schema_version": 1,
"concepts": [
{
"id": "schur-elimination",
"source": "Mathproof/LinearAlgebra/SchurBlockElimination.lean",
"sha256": "b865c41950e67c86d3e2961e5a5682d0cbf1a49ad9e11612730e0c198c16ddb3",
"declarations": [
"schur_block_system_iff"
],
"role": "Deletion and reattachment of one coordinate; requires the actual two-sided inverse hypotheses.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "robust-frames",
"source": "Mathproof/Meta/RobustRealTIFrames20261006.lean",
"sha256": "47a7f3586e6b1be6497e237df7cd37c73fae5107b69f7242e5e128a2eb7e918b",
"declarations": [
"Frame.norm_apply",
"relative_frame_error"
],
"role": "Permutation-frame transport. Positive diagonal weights are separately justified by a pullback norm, not treated as orthogonal.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "tensor-packing",
"source": "Mathproof/Meta/RobustRealTIRecovery20261006.lean",
"sha256": "bffdfcd0b61f4f116c2f7e24566cbbef608fe127f2215fb3d225ad1218ce73d6",
"declarations": [
"pack_norm",
"pack_sub"
],
"role": "Finite tensor layouts preserve squared energy; real/imaginary and correlated payload columns share one exact map.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "block-energy",
"source": "Mathproof/Computability/IntegerTensorBlockNorm20261015.lean",
"sha256": "52a61d4d8ed0897f097001b9aac1d1ea46b57a0deec5598734b6badae134c420",
"declarations": [
"block_energy_prefix",
"total_energy_bound"
],
"role": "Maximum local depth controls a direct sum without an independence premise; inverse gaps yield explicit local factors.",
"catalog_status": "NOT_FOUND",
"catalog_hash_matches_current": false,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "word-embedding",
"source": "Mathproof/LiGuanCNOTFallout20260905.lean",
"sha256": "1b6b13e220370fee7a233ba1ee6b66b7eb9d981e8daa9042da2342eb982972da",
"declarations": [
"word_map",
"within_embedding_iff"
],
"role": "The generic monoid portion transports an unchanged finite word through fixed injective ports; no characteristic-two recorder theorem is used over rationals.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "footprint-frame",
"source": "Mathproof/Computability/MultitapeFrameRule20261014.lean",
"sha256": "e5aa9ec056340c67bc6e382eee0a859e6903d9944d3a0e021f2a24a12459cb66",
"declarations": [
"FootprintedRuns.frame"
],
"role": "Arbitrary exterior preservation is a design precedent; the finite-memory API is proved directly, not claimed as a Lean multitape instance.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "locality",
"source": "Mathproof/Computability/MultitapeLocality20261014.lean",
"sha256": "4e5df1bdd9764d335424acb2a3aad8364f647892e632d8b09aec85e45cfb5b31",
"declarations": [],
"role": "Local scanned-region dependence motivates the exact immutable-source/port contract; not a new tape execution theorem.",
"catalog_status": "ARTIFACT_PRESENT",
"catalog_hash_matches_current": true,
"evidence": "Current source read; no new kernel audit of this combined informal application."
},
{
"id": "woodbury",
"source": "research/woodbury-reactor-20260920/Woodbury.lean",
"sha256": "8f269e268a026bc659e33f85e8873713b325f618ef0dae9f7d469e911730b27d",
"declarations": [
"woodbury_two_sided"
],
"role": "Two-sided inverse of the 2x2 core gives the exact full rank-two replacement.",
"evidence": "Historical typed DAG operation exists; current source read. Not a fresh audit of this new result.",
"dag_snapshot": "5cafc7e755c67429021b07c755ddd70e5fef81e1f478c9e6d14f298cca131835",
"dag_type_node": "837a84e5d279c06003451565aefc1517bec1e6667bba5fc0a71ce315fc9002ed",
"historical_axioms": [
"propext",
"Classical.choice",
"Quot.sound"
]
}
],
"scope": "Discovery provenance, exact hypotheses checked against written application. Source presence or historical artifacts are not a fresh kernel proof."
}
60 changes: 60 additions & 0 deletions research/sliding-window-inverse/LIBRARY-BRIDGES.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,60 @@
# Eight shared-library bridges and nine mathematical concepts

These are actual source statements or explicitly labelled design precedents.
The new sufficient proofs are fully written in PROOF.md; none relies on an
unavailable spectral oracle or an assumed unknown circuit. Exact current source
hashes and historical evidence distinctions are in LIBRARY-BRIDGES.json.

| Shared source | Reusable fact or concept | Concrete use here | Limit |
| --- | --- | --- | --- |
| SchurBlockElimination | two-sided Schur system elimination | principal inverse deletion and bordering | Already used in PR3851; no novelty claim for Schur. |
| woodbury-reactor / typed DAG | `woodbury_two_sided` | a two-column/two-row update and 2×2 inverse replace a dense inverse | Historical audited algebra; combined cache is informal. |
| RobustRealTIFrames | `relative_frame_error`, exact orthogonal frames | source-label permutations transport inverse/error without changing norm | Arbitrary positive D is NOT orthogonal. |
| RobustRealTIRecovery | `pack_norm`, coordinate bijection | tensor/column flattening preserves squared energy | Does not prove spectral recovery or supply frames. |
| IntegerTensorBlockNorm | `block_energy_prefix`, `total_energy_bound` | correlated blocks use maximum local amplification, not total unrelated work | Separate inverse applications, not an unbounded product. |
| LiGuanCNOTFallout | generic `word_map` and injective word transport | fixed caller-port embedding preserves the same local-map sequence | Uses only generic monoid lemmas, not F₂-only recorder algebra. |
| MultitapeFrameRule | `FootprintedRuns.frame` | direct finite-memory exterior-frame proof and dirty-value controls | This API is not a new Lean multitape program. |
| MultitapeLocality | finite-region observational dependence | immutable old-H and injective finite port contract | Design precedent, not an instantiated new execution certificate. |

The additional numerical concept is an actual residual-to-recovery certificate:
check the full rational residual against the exact weighted old-H matrix before
asserting a numerical tolerance. It follows the shared robust-recovery practice
of checking the returned object itself; the inverse-specific proof is new here.
The exact coefficient-height/minor argument and diagonal similarity cancellation
are retained from the Gaussian predecessor, not advertised as unused inventions.

The submitted cache, two-sided residual check, weighted rounding, port-map
implementation and tensor/block controls instantiate the first six algebraic
patterns and a direct finite-memory frame. The two multitape sources remain
explicit design precedents; their Lean execution theorems are not newly applied
to this host-language algorithm. We do not inflate that boundary into a formal
tape result merely to increase the source count.

## Search coverage and unused status

The existing authoritative source catalogue contains 401,769 registered modules.
Twenty broad source queries retrieved at most 100 hits each, with bounded
MathProof-specific follow-ups; twelve typed-DAG/symbol queries and six indexed
note searches were also performed. Sources include the federated project,
downloaded libraries and registered archive notes. This is broad retrieval,
not a complete inspection or semantic search of every registered statement.
The first unrestricted ranked source query was bounded after slow common-term
search; the bounded indexed replacement preserved explicit limits.

Actual source inspection then followed the eight bridges above, the local
Woodbury mission/proof, current Gaussian source and precise predecessor APIs.
The private search packets retain indexed paths, revisions, snapshots and misses.
Only the sanitized provenance needed for review is published here.

Schur, weights, minors and Neumann reasoning are demonstrably already present
in PR3851. The scoped Woodbury sliding-cache application, its weighted residual
and fixed-port/block contracts were not located in the inspected project sources
or the 575-open-PR body scan. That is a bounded overlap finding, not a guarantee
that nobody has applied any of these standard ideas anywhere in the project.

A later actual-source comparison found that RaD's retained
`review-reusable-banded-inverse.md` already uses Woodbury for a circular wrap
correction and shares setup across right-hand sides. That predecessor is
explicitly credited. It does not describe the one-index principal-set deletion/
bordering cache checked here. Our claim is this scoped composition, not first
use of Woodbury, residuals, setup reuse or weighted Gaussian inverses.
25 changes: 25 additions & 0 deletions research/sliding-window-inverse/NOTICE
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
Apache-2.0. New composition, exact host algorithm, proof and checks by Dugongue,
with substantial OpenAI Codex assistance. No independent human review claimed.

Unmodified vendored source:
research/physical-gaussian-core/gaussian_check_setup.py from Jose Manuel
(jmbarrancoml), PR3851 at 7ab67b4a84d28cd562b9711aa315ca5b47822b37.
Only its literal exact rational setup fixture is consumed by AST parsing.
Its physical implementation and native dependencies are not imported or rerun.

Retain the predecessor's attribution: Chafik Boukhalfa's PR489 local inverse
and PR481/486 boundary/halo work; the RaD project (hipotures) Gaussian,
phase-cell, microbox and reusable-banded-inverse reports; PR3787 word multiplier
at 9b5474f47b0ea9a3c76bb5b1b577fb36704815e0 including corrected PR3709;
Mason Hale's interval reference; Dugongue's scalar ABI precedent; Harvey and
van der Hoeven, the OpenAI manuscript and Douglas Colkitt's framework.
These source notices are retained credit, not a fresh validation of their full
physical/analytic claims. PR3763's chirp caller is a separate future integration
interface, not implemented or re-audited here.

The shared MathProof Schur, Woodbury, robust-TI frames/recovery, block norm,
CNOT/TI word, locality and frame sources retain their existing authorship and
provenance, identified by exact source hashes in LIBRARY-BRIDGES.json.
The matrix identities and elementary norm facts are standard mathematics.
Woodbury already appears in RaD's reviewed circular-wrap correction; this work
claims a moving-principal-window cache application, not its first project use.
34 changes: 34 additions & 0 deletions research/sliding-window-inverse/OVERLAP.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,34 @@
# Current source and overlap

Canonical main fetched at entry and at the final research checkpoint:
`20ecf2891d80b5acf18d980b2164fb44ea7faa90`, priority revision 6.
Current AGENTS.md, autoresearch.md and CONTRIBUTING.md were refreshed and compared
with the same previously read source revision. PR3851's current head remained
`7ab67b4a84d28cd562b9711aa315ca5b47822b37`. Its actual PROOF, setup fixture,
window/axis/matrix/production source, README and source manifest were inspected.

The original uniform-transfer-audit card is retained verbatim; the changed scope
and its actual sufficient conclusion are explicit in TASK.json. The selected
baseline bindings were checked in the preceding same-main campaign; no new
selected cold replay is claimed for this unrelated component.

A fully paginated REST snapshot scanned 575 open PR titles/bodies. No body match
for the scoped sliding principal-inverse cache was located. The repository
find_overlap.py also searched the snapshot and available local package notes.
Latest all-state updated records were scanned at entry and a final 30-record
checkpoint through PR3871. Start-here and handoff Discussion HTML was read, with
partial possible reply coverage and no complete GraphQL Discussion cache.
No exhaustive historical closed-PR, arbitrary-diff or unpublished coverage claim.

Actual-source comparison found a crucial existing use: RaD's retained
review-reusable-banded-inverse.md already uses Woodbury for circular-wrap repair,
plus setup sharing, rounded row residuals, inverse bounds and coefficient heights.
Its phase-cell report and PR3851 already use Schur and exact similarity weights.
Those tools are predecessors, not newly introduced mechanisms. The new delta is
one-index changes of the principal index set, with exact cache update, source-
label contracts, weighted residual recovery and a finite-memory port/energy API.
Neither new broad mathematical novelty nor first project use of the standard
identities is asserted. The scoped application was not found in the compared
sources. LIBRARY-BRIDGES.md records the broad DAG/source/note search and limits.

No exact new kappa, selected-frontier change, hosted CI or human review claimed.
Loading