Skip to content
Open
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
33 changes: 33 additions & 0 deletions research/ambient-rank-library/NOTICE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,33 @@
# Attribution and assistance

Prepared for **@CubeJerry** with substantial OpenAI assistance in the
proof, evaluator, independent author-side arithmetic oracle, mutation
controls and writing. The lead agent's separately recorded review is
additional AI evidence, not human or formal-kernel acceptance.

Credit Jacob Alan Sussman for the multinomial integer compiler and
Theorem 1G; zxv and the retained #4436/#4359 lineage for the chronological
flow and finite-precompilation argument; Mihir Srivastava for #4507's
least per-cut batching theorem; the authors and retained notices of
#4638/#5071 for the plateau divisor and separated symbolic recurrence.
Credit @CubeJerry and the selected #6916/#7000 lineage for the actual
independent ambient-cap geometry and its source-bound application.

`ambient_library.py` adapts the algorithm of @CubeJerry's #7345 at
`022461f115369efd988ef8c459907f5544bb8bb3`. The original `library.py` is
retained byte-for-byte as `inputs/default_library.py`, with its proof as
`reference/default-library-PROOF.md`. This adaptation and the default
regression are not independent reproductions of that implementation.
The new `literal_oracle.py` imports neither evaluator, uses explicit
role/path/port occurrences, prime valuations and rational costs, and
serves as the separate arithmetic check on bounded fixtures.

The root's leaf/owner companion at commit
`81f0b0b9a920bd1c9da583cbc3612efd5e75ea94` motivated the valid-library
interface; it is not imported here and has no claimed upstream PR number.

Frozen references retain their original notices and attribution. New
material is provided under Apache-2.0. No predecessor result, default
width asymptotic, selected source admission or normalized kappa improvement
is claimed as new. Mathematical novelty is the explicit ambient quantitative
construction/budget and scoped uniform-output obstruction described here.
129 changes: 129 additions & 0 deletions research/ambient-rank-library/OVERLAP.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,129 @@
{
"base": "e54523517d0f03e58352e647c403a5af3e525a57",
"date": "2026-11-12",
"scope": "Exact quantitative separated rank libraries at A>=m, plus a uniform evaluated-output obstruction; no native instruction emitter or new kappa.",
"queries": [
{
"query": "\"ambient\" \"library\"",
"tool": "github_search_prs",
"sort": "updated descending",
"results": [
{
"number": 7345,
"title": "Generate exact chronological compiler library tables with product aggregation and explicit binary and tape cost bounds",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7345"
},
{
"number": 7331,
"title": "Prove one crossing path keeps the proper-terminal rank-bill maximum and four break it (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7331"
},
{
"number": 7293,
"title": "Bound fixed BIT destination-chart reuse (0% κ change)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7293"
},
{
"number": 7263,
"title": "Attain exact factor-one Stein bounds with at most 2r terms (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7263"
},
{
"number": 7254,
"title": "Construct polynomial two-capacity sampling without prescribed images (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7254"
},
{
"number": 7248,
"title": "Prove four independent full guards beat their product over every field",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7248"
},
{
"number": 7205,
"title": "Prove the first central Schur state collision at pivot size four (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7205"
},
{
"number": 7193,
"title": "Prove the sharp rank-two bound for five address-twist contributions (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7193"
}
]
},
{
"query": "\"ambient\" \"rank-library\"",
"tool": "github_search_prs",
"sort": "updated descending",
"results": [
{
"number": 7345,
"title": "Generate exact chronological compiler library tables with product aggregation and explicit binary and tape cost bounds",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7345"
},
{
"number": 6384,
"title": "Prove a strictly weaker mirror-capacity criterion (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/6384"
},
{
"number": 5097,
"title": "Prove exact singular partial Pauli completion lengths",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/5097"
},
{
"number": 4690,
"title": "Add charged Gaussian contact, phase and Pauli gauge searches (no new κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/4690"
}
]
},
{
"query": "\"ambient-cap\" \"width\"",
"tool": "github_search_prs",
"sort": "updated descending",
"results": [
{
"number": 6208,
"title": "Construct order-17 Gaussian-dyadic Clifford attaining native rank eight (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/6208"
}
]
}
],
"precise_comparisons": [
{
"pr": 7345,
"commit": "022461f115369efd988ef8c459907f5544bb8bb3",
"files": [
"research/quantitative-rank-library/PROOF.md",
"research/quantitative-rank-library/library.py"
],
"finding": "Explicitly A=m; general ambient and selected A=153^1920 outside its theorem."
},
{
"pr": 4638,
"canonical_file": "research/chronological-width-bound/README.md",
"finding": "Default m-dependent contraction and plateau visits; general A divisor rederived, not imported unchanged."
},
{
"pr": 6916,
"canonical_file": "research/p104-shared-native-helper-splits/retained/compiler/PROOF.md",
"finding": "Already proves general ambient flow and qualitative finite construction; the new claim is quantitative evaluated library and explicit parameter dependence."
},
{
"prs": [
4507,
5071
],
"canonical_file": "research/symbolic-integer-batch-width/PROOF.md",
"finding": "Least chronological recurrence and singleton cuts inherited; not a new denominator rule."
}
],
"classification": "No exact superseding ambient quantitative claim identified in this bounded search; this is not exhaustive novelty evidence.",
"exclusions": [
"Full scalar/phase/adapter emitter assigned separately",
"Fixed-template outer absorption already covered by retained sources",
"No selected practical materialization or kappa improvement"
],
"coordination": "Lead orchestrator maintains current claims and wave manifest; no independent placeholder issue or upstream PR."
}
1 change: 1 addition & 0 deletions research/ambient-rank-library/PR-TITLE.txt
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
Construct chronological rank libraries at independent ambient caps and prove the quadratic explicit width-output obstruction
107 changes: 107 additions & 0 deletions research/ambient-rank-library/PR.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,107 @@
## Result and why it matters

The exact rank-library generator in #7345 explicitly assumes that the
ambient cap equals the largest productive target. The selected compiler
instead uses the actual cap `A=153^1920`, above its maximum target.
This package proves and implements the evaluated least-chronological
library for every fixed separated rule with independent `A>=m`, retaining
every individual exit divisor and all recursive costs.

It also proves a necessary qualification on uniform complexity. For the
fixed targets 2 and 3 and one paid rank-1 site, every A>=4 satisfies
`B_n=2^(floor((n-A)/2)+1)` on `A<=n<2A`. The width outputs on that plateau
alone contain exactly `floor((A-1)^2/4)+2A` bits. At `N=2A-1`, a uniform
near-linear explicit all-rank bit-output bound is therefore impossible.
This does not contradict the retained selected construction: its finite
template is fixed before multiplication input size varies.

## Change and exact scope

Adds only `research/ambient-rank-library/`, containing the proof, bounded
integer evaluator, separately written explicit-occurrence/prime-valuation
oracle, semantic controls, source pins, receipt and registered replay.

The ambient-cap flow is inherited from the retained #6916 compiler proof
§6. The new quantitative analysis proves uniform contraction by `(A-1)/A`,
an all-prime plateau divisor with exponent `A*ceil(log2 k_min)`, and
explicit binary/indexed and sequential-tape metadata costs. The default
constant m cannot be substituted for A: for A=64 the actual width B_127 is
2^32 while that false one-plateau divisor is 8.

For each fixed A and fixed histograms, evaluated row lengths remain
polylogarithmic in N, indexed construction is `N*polylog N`, and the
sequential companion is `N^2*polylog N`. The proof exposes every A-dependent
factor when A is supplied in binary. Initialization, arithmetic lengths,
sorting, child-table access, copying, head restoration and historical cells
are charged. These are metadata bounds, not native payload-movement claims.

## Novelty and predecessor credit

Credit #4507/#5071 for least chronological gcd/LCM integrality, #4638 for
the default plateau-visit argument, #6916 and the selected source lineage
for independent ambient geometry, and #7345 at
`022461f115369efd988ef8c459907f5544bb8bb3` for product aggregation and the
quantitative default-library framework. The new evaluator adapts that code;
its frozen original is used only for separately labelled regression checks.
The ambient extension and varying-A lower bound are the delta.

This supplies a valid numerical library premise for the separately
published [chronological leaf/owner locator](https://github.com/CubeJerry/integer-mult-bounds/tree/81f0b0b9a920bd1c9da583cbc3612efd5e75ea94/research/chronological-leaf-owner-index).
That companion is contextual credit, not a runtime dependency or an
invented upstream PR reference. Canonical tested base is
`e54523517d0f03e58352e647c403a5af3e525a57`; exact nine-file provenance is
retained in `SOURCES.json`.

## Reproduction and evidence

From the repository root with Python 3.12 and the standard library:

```sh
python -I -W error -B research/ambient-rank-library/verify.py
python -I -O -W error -B research/ambient-rank-library/verify.py
python -B scripts/research_package.py --package research/ambient-rank-library --check-only
```

Direct marker: `PASS_AMBIENT_RANK_LIBRARY`. Both direct commands compare
the complete computed hash-bound receipt without writing to the package.
The independent oracle explicitly constructs role words, paid paths and
target ports, then uses prime valuations for widths and exact rational
per-array costs. It imports no production evaluator. All fields agree on
1,267 ranks, 21,442 units, 31,124 cuts and 35,141 constraints; exact
all-prime divisibility holds at every checked rank. The lower-bound family
is checked on 4,000 plateau rows across 81 ambient values. Fifteen semantic
and domain controls reject. Separately, 146 default-A rows agree with the
frozen predecessor. Six frozen local source copies are authenticated.

The large-A test uses the number `153^1920` only with toy histograms and
evaluates bounds/bit-count descriptors. It materializes zero library rows.
Finite checks corroborate the written universal proof, not replace it.
The separate reviewer accepted the universal argument and checked 269 complete
rows using a positive source census and brute-force least replication, plus
260 plateau widths and three semantic mutations. Normal and optimized
receipts match. The unchanged review and checker are retained under
`independent/`; the root authenticated all nine exact commit/path bindings.

## Limitations, open obligations and kappa

**No new kappa.** Selected `0.00363005817` and all five assumptions remain.
The actual selected evaluated target/paid manifest and winning library are
not materialized. This package proves neither native correctness nor full
scalar/phase/adapter instruction emission, physical owner routing, owned
erasure, precision/tape admission or analytic recovery. It supplies metadata
to those interfaces. Open questions are source-specific gcd-sensitive
widths, actual selected manifest binding, and full instruction emission
with all physical costs. No selected cold replay, Lean, Docker, human or
proof-kernel acceptance is claimed.

Prepared for **@CubeJerry**, with substantial OpenAI assistance. Credit
Jacob Alan Sussman, zxv, Mihir Srivastava, the #4638/#5071 authors and all
retained native/outer lineage as detailed in `NOTICE.md`. Predecessor
notices remain. New material is Apache-2.0.

## Merge scope

One isolated package. No predecessor, shared implementation, fork main or
selected-frontier changes. This text is a proposed PR description; the run
publishes a checked fork branch for user review and does not submit an
upstream PR.
Loading
Loading