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
162 changes: 162 additions & 0 deletions research/g20-long-conjugation-obstruction/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,162 @@
# Longer fixed-route G20 conjugations have no paid-moment gain

## Claim and boundary

Fix Ali Romman's final PR5708 `WORD.json.gz` and `PATHS.json`, pinned in
[SOURCES.json](SOURCES.json) at main
`2485805a6d47f36ea83ddde6327d453ac82d44dc`. Consider a single A2 conjugation
with at least four noncommuting contacts, with each correction kept at its
contact's original frame. Every such conjugation either encounters a missing
published route or has nonnegative paid-query moment change for every
`0 < t < 1`. The statement covers all contact lengths present on this word,
including its maximum of 33, rather than a chosen upper cutoff.
Among emitting candidates the minimum moment change is exactly zero for
every `0<t<1`. For example, zero-based scalar positions 9,399 and 10,815
bound a four-contact rewrite of `ADD(560,315)` with an unchanged histogram.

This is a finite certificate with a scoped algebraic sign proof. It extends
unpaidmoon's PR6290 question and fixed-contact-frame definition. PR6290's
one-to-three-contact result is cited, not re-established here. No assertion
about a changed route table, a sequence of rewrites or another word follows.

## Exact candidate class

Remove `Q` and `ZERO_Q` records from the pinned word to obtain 11,930 scalar
records on 685 roles. Write `G(a,b)` for `ADD(a,b)`, the characteristic-two
map `x_a <- x_a + x_b`. Choose any two occurrences of the same ordered
pair `(a,b)`, with positions `left < right`. Their frames may differ.

The open interval must contain no `INIT` or `ERASE` of `a` or `b`. A contact
is an ADD with destination `b` or source `a`. Every contact must have one
of the following forms, with the indicated correction:

| Contact | Constraint | Correction |
|---|---|---|
| `ADD(b,s)` | `s != a` | `ADD(a,s)` |
| `ADD(d,a)` | `d != b` | `ADD(d,b)` |

In particular, the reversed gate `ADD(b,a)` is outside this class. Require
at least four contacts. Delete the two outer gates and insert the correction
immediately after each contact, at that contact's original frame. All other
records retain their order and frame. These are the only changes allowed.

Completeness is finite and explicit. Group every ADD position by its ordered
operand pair, visit every ordered choice of two increasing positions in each
group, inspect every intervening record, and retain exactly the intervals
satisfying the conditions above. There is no contact cutoff or node budget.
The resulting domain has exactly 1,503 intervals. The certificate lists every
interval's endpoints, operands, contact count and outcome.

## Scalar endpoint

For each valid contact `C`, direct multiplication of elementary matrices gives
`G C G = C R`, where `R` is the indicated correction. The correction commutes
with its contact. Every noncontact commutes with `G`, including an erase or
initialization of a different role. Since `G^2 = I`, conjugation distributes
over the entire interval. Thus deleting the two outer gates and replacing
every contact by `C R` preserves the scalar map for any contact count.

The verifier also checks the complete endpoint separately for every one of
the 1,503 changed words, including those that later fail route emission.
Each row is a 685-bit coefficient mask, initially `e_i`. ADD is exact XOR;
only owned roles may be erased, and INIT requires their whole row to be zero.
The expected final row is zero on owned roles. On other roles it is `e_i`,
with the additional term `e_(i-111)` for `111 <= i < 222`.
This tests every independent scalar input column, not a sample of payloads.
The native address action remains a cited predecessor premise.

## Route contract and certificate

Use precisely the source emitter's rules. An ADD with zero source row is
omitted. Otherwise move its destination and source to its frame, in that
order. A published one-edge route incurs its rank once when the moved row
is nonzero, and zero when that row is zero. A missing edge rejects emission
even for a zero row. An uninitialized owned zero row may be initialized.
At the end return every public role to its final frame and require every
owned role to be zero. Owned cleanup has the source's zero-query convention.

The unchanged word provides the necessary reference histogram
`{1:1281, 2:1338, 3:58, 4:57, 5:102, 6:10}`. This reference is not a new
predecessor audit or a claim of full supplier admission. Every changed word
that emits receives a freshly reconstructed complete paid-query histogram.
The signed difference is new minus reference.

Exactly 1,450 intervals encounter a missing route. For each, the certificate
records the first required ordered frame pair absent from the table. These
are rejection witnesses relative to this table, not proofs that the native
geometry could never supply another route. The other 53 intervals emit.
Their entire signed histogram classification is:

| New minus reference | Count | Sign for `0 < t < 1` |
|---|---:|---|
| `2H1 - H2` | 15 | Strict loss |
| `3H1 + 2H2` | 10 | Strict loss |
| `2H2 - H4` | 5 | Strict loss |
| `2H1 + H2 + H3` | 4 | Strict loss |
| `H1 + 3H2` | 2 | Strict loss |
| `H1 + 2H2 - H5` | 2 | Strict loss |
| `H1 + H3 - H4` | 1 | Strict loss |
| Zero | 14 | Neutral |

Here `Hr` contributes `r^t`. All coefficients in three of the nonzero
shapes are positive, so their moment changes are positive. In each of the
other four shapes, one negative job of rank `R` is replaced by at least
two positive jobs whose ranks sum to `R`. For positive `a,b` and `0<t<1`,
`(a+b)^t < a^t+b^t`. For example, divide by `(a+b)^t` and use
`u^t > u` for `0<u<1` on `u=a/(a+b)` and `1-u`.
Iteration proves strict positivity for any such split. This covers the
whole open exponent interval without floating-point sampling or root search.

The complete canonical candidate certificate has SHA256
`577323306b7b7f2547ca660e6bbef1badc953959e2f7cd20de9e1f002e254889`.
Canonical bytes are UTF-8 from `json.dumps(result, sort_keys=True,
separators=(",", ":"))`. The optional generated receipt includes that whole
certificate and hashes of all package files and live inputs.

## Controls and limits

The verifier checks all 256 four-contact words over the two allowed contact
orientations on four roles. It checks their complete input-column endpoints
and rejects deletion of the first correction in each. Reversed contacts,
erased operands and short intervals are explicit exclusion controls.
The exact sign rule rejects the profitable shape `H1 - 2H2 + H3`, whose
negative sign follows from strict concavity. Routing controls distinguish
successful charging, a missing edge and a zero source.
The proof checks use explicit exceptions and remain active under `-O`.

This result concerns one rewrite of the final word and the paid-query part
of its bill. It provides no new productive target, ordinary BIT saving or
whole multiplication exponent. No global caller, adjoint/native address
replay, finite compiler, scalar allowance, physical storage, precision,
outer ledger, uniform program or rounded multiplier is certified here.
The five selected framework assumptions remain. Additional routes, correction
coframes and sequences through neutral words are not excluded.

## Implementation constraints

The verifier uses structured iteration without recursion. Its inputs are
hash-pinned before decompression and parsing. Each compressed input is below
two million bytes, the scalar dimensions are checked as 11,930 by 685, the
candidate list has a fail-closed capacity of 4,096, and XOR rows have at most
685 bits. Every pair search has the finite upper bound `C(11930,2)`; every
interval and replay has the corresponding fixed record bound. One changed
word is live at a time, with no more than `11930-2+33` records on this source.
No input limit truncates the claimed domain.

Power-of-ten rule 3 deviation: CPython and its standard JSON and gzip readers
allocate dynamically. This standalone finite checker cannot preallocate those
runtime internals. The exact authenticated inputs, checked dimensions and
candidate capacity bound its live data; the normal and optimized complete
checks exercise those bounds. A future unpinned input interface would require
bounded streaming parsers before relaxing this contract. All functions fit
within 60 physical lines. Warning-strict normal and optimized execution are
the available diagnostics; no static-analysis dependency is introduced.

## Credit

The new longer-contact result and standalone certificate were produced by
Nick Petalas. Unpaidmoon posed the four-or-more-contact question in PR6290.
Ali Romman supplied PR5708's word, emitter and scalar A2 construction, following
Dugongue, ai-nanachi and the native/controller/helper/ordinary/stock lineage
retained in that package. The A2 identities and strict subadditivity are
classical and are not claimed as new. No predecessor file or credit is changed.
97 changes: 97 additions & 0 deletions research/g20-long-conjugation-obstruction/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,97 @@
# Longer fixed-route G20 conjugations attain a minimum paid-moment change of zero

Every single A2 conjugation with four or more contacts on the final
[PR5708 word](https://github.com/CrocSwap/integer-mult-bounds/pull/5708)
either encounters a missing published route or fails to improve the paid-query
moment for any exponent `0<t<1`. Corrections stay at their original contact
frames. Among the emitting candidates the minimum change is exactly zero,
attained by 14 words. No new kappa or supplier admission is claimed.

This answers the longer-contact, fixed-frame part of open question 2 in
[unpaidmoon's PR6290](https://github.com/CrocSwap/integer-mult-bounds/pull/6290),
head `9d2916b07402aca2133ab315230b9c761cc9eff7`. The one-to-three-contact
result remains that predecessor's contribution. This package enumerates only
the new four-or-more-contact domain. See the model and sign proof in
[PROOF.md](PROOF.md).

## Exact result and evidence

| Outcome | New candidates |
|---|---:|
| Missing published route | 1,450 |
| Emitted, strictly larger moment throughout `0<t<1` | 39 |
| Emitted, identical paid histogram | 14 |
| Total | 1,503 |

All 1,503 changed words retain the exact scalar endpoint on all 685 independent
input columns. Their contact counts range from 4 to 33. The enumeration has
no contact cutoff: it visits every equal-gate interval on the fixed 11,930-record
scalar word. Each missing-route record gives the first absent ordered frame
pair, and each emission gives its complete signed paid histogram.
Every strict-loss shape is positive by positivity or strict subadditivity;
no floating-point or exponent sampling is used.

The standalone checker tests 256 four-contact identities and 256 missing
correction mutants, plus seven boundary, sign and routing controls. It uses
explicit exceptions under both normal and optimized Python. A generated
receipt can contain the complete candidate certificate and SHA256 bindings
for all six package files and every live input. Its canonical certificate
digest is `577323306b7b7f2547ca660e6bbef1badc953959e2f7cd20de9e1f002e254889`.

## Reproduce

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

```sh
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -B research/g20-long-conjugation-obstruction/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -I -W error -O -B research/g20-long-conjugation-obstruction/verify.py
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/research_package.py --package research/g20-long-conjugation-obstruction
/Users/nick/git/integer-mult-bounds/.venv/bin/python3.12 -B scripts/check_research_commit.py --package research/g20-long-conjugation-obstruction --check-source-history
```

Any CPython 3.12 interpreter can replace the local absolute path. Each direct
check prints `PASS_G20_LONG_CONJUGATION_OBSTRUCTION candidates=1503
missing_route=1450 emitted=53 neutral=14 strict_loss=39 min_contacts=4
max_contacts=33`. The registered runner prints `PASS_REGISTERED_RESEARCH_PACKAGE`.
The optional `--output /absolute/path/outside/checkout/receipt.json` writes a
generated, source-bound receipt. It never writes into the checkout.

Exact source commits, Git blobs and SHA256 pins are in
[SOURCES.json](SOURCES.json); the changed card scope and structured follow-ups
are in [TASK.json](TASK.json). The selected record and catalogue are historical
provenance. The three live inputs are authenticated before use. Neither the
predecessor verifier nor its larger caller is executed.

## Limitations

The result fixes the final PR5708 word, one conjugation, original correction
frames and the published routes. A missing route is not a geometric impossibility.
Further rewrites through neutral neighbors, added routes and different correction
frames remain open. The certificate concerns the finite scalar endpoint and
paid-query component, not a complete signed supplier invoice. Native/adjoint
address action, the global caller, scalar allowance, target admission, compiler,
precision, physical tape/storage and outer predicates were not replayed.
No uniform program, rounded multiplier or proof-kernel verification is supplied.
The five selected framework assumptions and selected kappa `781451/250000000`
remain unchanged. Independent human and formal proof-kernel review are unclaimed.

## Open next questions

1. Can one of the 14 new neutral rewrites expose a later paid saving? Generate
one listed neutral word, enumerate its changed intervals and price the
complete emitted bill. This matters because the theorem permits one rewrite
only. No neutral-neighbor search or caller admission is supplied here.
2. Can a new native route or a moved correction frame rescue a rejected interval?
Start with one recorded missing frame pair, construct its actual native
action, and charge routes, targets, scalar fees and restoration. These costs
and the complete physical caller remain unresolved.

## Credit

The new longer-contact result and certificate were produced by Nick Petalas.
Credit unpaidmoon for PR6290's question and fixed-frame definition, and
Ali Romman for PR5708's word, row-sensitive emitter and A2 construction,
following Dugongue, ai-nanachi and all retained native, controller, helper,
ordinary and stock sources. The standard A2 identity and subadditivity proof
are not claimed as new. No predecessor file or credit line is changed.
New work is Apache-2.0.
36 changes: 36 additions & 0 deletions research/g20-long-conjugation-obstruction/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,36 @@
{
"schema_version": 1,
"base_commit": "2485805a6d47f36ea83ddde6327d453ac82d44dc",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "provenance",
"git_blob": "d59b5035e7b4cbaa9d688a1d32dd0d6eb79c466a",
"sha256": "9b8d0eea95b3bf0dfcb7938a732a6439b87a2685dec2a5ab9e525b3b34debbfa"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
"sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
},
{
"path": "research/g20-conjugation-admission/PATHS.json",
"usage": "input",
"git_blob": "efbdf632746a971f9a593f57ccbb7d2cc67a6760",
"sha256": "f4dd58b78121040b1f054c0f7a626b25f6ef8d39f3de044d3456ea7dea8a6a5e"
},
{
"path": "research/g20-conjugation-admission/WORD.json.gz",
"usage": "input",
"git_blob": "3877a72d8a15adb03f6492f7d2fd8fa902ecb639",
"sha256": "8dd441fdb9c8923df06eb5ebfda37d03e4539f2305763e688d8c7025d196c4b1"
},
{
"path": "research/g20-conjugation-admission/construct.py",
"usage": "input",
"git_blob": "c0fe9d080b76171d8ed967cbad337c596d250a23",
"sha256": "2922052092d9323de2c715afb66dbd33fb58c7d0eeb3b9298bab5f7b2a930cb1"
}
]
}
Loading