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
1 change: 1 addition & 0 deletions research/one-edge-word-obstruction/.gitattributes
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
retained/** -text whitespace=cr-at-eol
37 changes: 37 additions & 0 deletions research/one-edge-word-obstruction/CREDITS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
# Sources and credit

Ali Romman (@aliromman) prepared this proof, exact checker and package
with substantial OpenAI Codex assistance. New material is Apache-2.0.
No independent human or formal proof-kernel acceptance is claimed.

Nick Petalas (@petalas) supplied the four-projector alphabet and its paid-word
question in [PR #7749](https://github.com/CrocSwap/integer-mult-bounds/pull/7749),
at the [source proof](https://github.com/petalas/integer-mult-bounds/blob/6bd8a384d64b6da2334bc94f6a97cdd74d710780/research/even-one-edge-maximizer/README.md)
and [unchanged checker](https://github.com/petalas/integer-mult-bounds/blob/6bd8a384d64b6da2334bc94f6a97cdd74d710780/research/even-one-edge-maximizer/verify.py).
All five package files are retained byte-exact. The word proof uses the
displayed matrices directly; it does not require a new geometric maximum.
The source notices credit Ali Romman's
[PR #7707](https://github.com/CrocSwap/integer-mult-bounds/pull/7707)
and [PR #7692](https://github.com/CrocSwap/integer-mult-bounds/pull/7692)
for the comparison pool, even maximum and matching theorem.

The paid-generated augmentation and commutator approach follows
[PR #7711](https://github.com/CrocSwap/integer-mult-bounds/pull/7711),
[retained proof](https://github.com/aliromman/integer-mult-bounds/blob/ac185e4529a01156f7b574555b50827f8cb645bb/research/shared-hub-triple-word-obstruction/PROOF.md).
Its notices credit Jerry Jiaheng Chen (@CubeJerry) and
[PR #6782](https://github.com/CrocSwap/integer-mult-bounds/pull/6782)
for augmentation chronology. The matrices and chronology conditions here
are for the distinct four-projector alphabet.

The finite Laurent engines adapt
[PR #7763](https://github.com/CrocSwap/integer-mult-bounds/pull/7763),
[retained checker](https://github.com/aliromman/integer-mult-bounds/blob/3cb09504f8c1b6b728d8ce90b76ff938c06fbe5d/research/shared-hub-seven-call/verify.py)
and [retained credit](https://github.com/aliromman/integer-mult-bounds/blob/3cb09504f8c1b6b728d8ce90b76ff938c06fbe5d/research/shared-hub-seven-call/CREDITS.md).
That notice retains Nick Petalas's earlier bit-encoded Laurent method.
These two files are credit dependencies; the seven-call replay is not run.

The new result is the private-writer lemma and the all-length three-relation
classification, with exact R/S cancellation and a complete rank-power
inequality. `SOURCE-PINS.json` authenticates all eight retained files.
`SOURCES.json` records canonical main selection and catalogue provenance.
Content pins do not assert unmerged ancestry.
218 changes: 218 additions & 0 deletions research/one-edge-word-obstruction/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,218 @@
# Three relations leave only two productive labels

Ali Romman with substantial OpenAI Codex assistance. Apache-2.0.

Use Nick Petalas's alphabet from
[PR #7749](https://github.com/CrocSwap/integer-mult-bounds/pull/7749),
in coordinates a,b,f,g:

```text
P=(1,3), Q=(3,2), R=(5,9), S=(11,6).
```

The pair (x,y) denotes xy^T over F2. Each call is I+([z]-1)A for one
of these projectors. Its label z belongs to the torsion-free abelian
group H generated by all m paid labels. Equivalently H=Z^m/L, with
L saturated. Later calls act on the left. Every type is present, every
paid label is nonzero, and rank(L)<=3. Require a diagonal endpoint on
all four original coordinates; arbitrary changes of register basis
are not included.

**Theorem.** The word is diagonal exactly when its chronology satisfies
the following three conditions and its exact type totals satisfy (1).

1. No P or Q call lies between the first and last R calls.
2. No Q call lies between the first and last S calls.
3. After deleting P and Q, the R/S word has at most three runs. Thus one
type is a single block inside the other type, or the types form two blocks.

```text
total(P)=total(Q)=u, total(R)=total(S)=0. (1)
```

The full endpoint is diag([u],[u],1,1). The two fresh coordinates have
identity targets. This classification holds at every finite word length
under the stated three-relation contract.

For rational matrix labels and every 0<t<=1, the target rank-power sum is
at most the paid rank-power sum. No favorable rank-power tariff is available
in this class, even before physical costs are added. This is not a universal
word obstruction for the alphabet when more relations are allowed.

## 1. A unique private writer cannot give a diagonal endpoint

Only R writes coordinate f, and its covector a+g has a nonzero coordinate
outside f. Suppose there is exactly one R call, at a nonzero label z.
Write its earlier word as U and later word as V. Both exclude R. Their
f rows are identity rows. The final f row is

```text
e_f^T + ([z]-1) (a+g)^T U.
```

The group algebra F2[H] is a Laurent polynomial domain. If the endpoint
were diagonal, each component outside f of (a+g)^T U would be zero.
Apply augmentation, which sends every group monomial to one. Every call
and U then become identity, so this row becomes (a+g)^T. It has nonzero
components outside f, a contradiction.

The same proof applies to S and g, since only S writes g and its covector
is b+f. Thus any diagonal word with all four types and nonzero labels
needs at least two R calls and two S calls. It needs at least six calls
in total, without any relation-rank assumption.

This floor is sharp. In H=Z^3, use independent nonzero u,r,s and the word
P(u),Q(u),R(r),R(-r),S(s),S(-s). It has exactly three saturated relations
and endpoint diag([u],[u],1,1). The R/S calls remain paid.

## 2. The minimal relation quotient

Let p,q,r,s be the four positive type counts. Section 1 permits us to
assume r,s>=2. The off-patterns have exactly three independent generators:
off(P)=off(Q), off(R), off(S). The degree-one off-diagonal endpoint
therefore forces

```text
sum(P forms)=sum(Q forms)=u,
sum(R forms)=0, sum(S forms)=0. (2)
```

These are three independent relations among the m call variables before
quotient. Since H is paid-generated and L is saturated, its degree-one
augmentation space has dimension m-rank(L), and its graded algebra is
the polynomial ring on that space. The limit rank(L)<=3 makes (2) the
complete set of degree-one relations; rank(L)=3 in every diagonal case.

Choose the common u, all but one variable of each P and Q type, and all
but one variable of each R and S type as coordinates. The P variables
are a basis of their p-dimensional span; the Q variables are a basis of
their q-dimensional span. The R variables have exactly one dependence,
their full sum zero, and the same is true of S. The R and S variable
spaces are disjoint from each other and from the P/Q variable spaces.
The latter two share only u.

This is a statement about the paid-generated group. An unrelated larger
ambient exponent group must not supply the augmentation coordinates.
Nor does degree one alone imply the exact integral identities (1).

## 3. Four commutators force the chronology

P,Q are orthogonal, as are P,S. The remaining four off-commutators
[P,R], [Q,R], [Q,S], [R,S] are independent modulo the pattern span.
Here [A,B]=AB+BA. The following finite certificate uses bit 4i+j for
matrix entry (i,j), after deleting the diagonal:

| Row | Hexadecimal encoding |
| --- | --- |
| off(P) | 0002 |
| off(R) | 0908 |
| off(S) | 6046 |
| off([P,R]) | 030a |
| off([Q,R]) | 0202 |
| off([Q,S]) | 2044 |
| off([R,S]) | 1098 |

Row elimination gives 6046,2044,1098,0908,030a,0108,0002, with distinct
leading positions 14,13,12,11,9,8,1. Thus all seven rows are independent.
The checker also tests all 127 nonempty subsets.

Compare the candidate chronology with the order P then Q then R then S.
The reference word's degree-two off-diagonal part lies in the pattern
span: its R/S totals have zero degree-one part, and P,Q commute. Each
interchange of two calls contributes the product of their linear forms
times the corresponding commutator. Independence makes each of the four
inversion polynomials zero.

For P/R, that polynomial is the sum over P calls of its variable times
the sum of variables of all earlier R calls. Independence of the P
variables and their products with the disjoint R space makes each such
R subset sum zero. The R variables have no proper nonempty zero-sum
subset. Hence each P call is before every R or after every R. The Q/R
and Q/S polynomials give the corresponding statements for Q. These
are conditions 1 and 2.

For R/S, form the r-by-s binary matrix M with entry one when S_j is
before R_i. The map from its rows and columns to the R/S variable
spaces has all-one kernels. Its tensor image is zero exactly when

```text
M_ij = alpha_i + beta_j.
```

One can check this by eliminating the last variable on each side: every
remaining coefficient is M_ij+M_i,s+M_r,j+M_r,s. In chronological order,
each row of M is a prefix of ones among the S calls. Two rows can differ
only by zero or by the all-one row. If one prefix is proper and nonempty,
every row has that same prefix, so all R calls form a single R block
between S blocks. Otherwise the prefixes are empty or full, so all S
calls form a single S block between R blocks. This is condition 3.

## 4. Exact cancellation, not just parity

If R is the inner block of the R/S chronology, conditions 1 and 3 make
its calls consecutive: neither P,Q nor S can occur between them. Combine
them into one call at their exact sum. Section 1 forces that sum to be
zero, so the block is identity. Only P can remain between S calls, by
condition 2. P commutes with S, so combine the S calls and use the same
private-row argument to force their exact total to zero.

If S is the inner block, condition 2 excludes Q and condition 3 excludes
R from its interval. Commute P past S and combine the S calls first.
Their total is zero by Section 1. Condition 1 then lets the R calls
combine, and their total is zero too. The same reasoning covers two
disjoint blocks.

Only P and Q remain. They commute and their product on a,b is

```text
[ [v], [v]+[w] ]
[ 0, [w] ],
```

where v and w are their exact totals. Diagonality forces [v]=[w], hence
v=w=u. This proves (1), necessity and the full endpoint. Conversely,
the same block reduction proves sufficiency from conditions 1--3 and (1).

Give the first target all P paid labels and the second target all Q paid
labels. Give the two identity targets empty subsets. Rank subadditivity,
monotonicity and subadditivity of x^t for 0<t<=1 give

```text
2 rank(u)^t <= sum_(P and Q calls) rank(z)^t
<= sum_(all paid calls) rank(z)^t.
```

The R/S cancellations are not free calls. Additional physical charges
cannot reverse this inequality inside this stated rank-power model.

## Verification and limits

All 35 multiplicity domains with 6<=m<=9, p,q>=1 and r,s>=2 are checked.
Recursive type-word generation and a separate position-subset enumeration
agree on every domain and its multinomial count. The 108,904 words each
receive an exact quadratic chronology check and a full Laurent matrix
product. Exactly 8,412 are diagonal, all with the stated four entries.
The minimal domain (1,1,2,2) has 46 passing orders out of 180.

The label fixtures use the universal exact quotient of (1). Every private
basis vector is a paid label, and u is an integral sum of P labels, which
authenticates the paid-generated free group and saturated three-relation
kernel. For these fixtures each exponent-support coordinate has width at
most two. Base-three substitution is injective on the whole support box.
A separate multivariate set engine checks 175 complete products.

Four hundred conjugated single-writer words check the private-row
obstruction, including nonzero even labels. Three nonzero even changes
to Q, R and S totals preserve the quadratic check but fail the exact
endpoint; a separate multivariate product checks each fault. This marks
why parity is insufficient. All eight retained files are authenticated,
and the PR7749 geometry verifier runs unchanged.

The all-length statement is the written proof, not a finite census of
all saturated lattices. More than three relations, changed projector
pools, arbitrary register bases, native implementation, physical storage
and restoration, and the complete caller invoice remain open. The full
PR7711/PR7763 word suites, other geometry suites, selected cold replay,
Lean and Docker were not run. There is no new kappa or selected-frontier
change; all five framework assumptions remain. No independent human or
formal proof-kernel acceptance or global literature priority is claimed.
35 changes: 35 additions & 0 deletions research/one-edge-word-obstruction/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
# Four-projector words at three exponent relations

At every finite length, a full diagonal word on the PR7749 alphabet with
all four types, nonzero labels and at most three saturated relations must
cancel both fresh-type totals. Its only targets are ([u],[u],1,1).
The complete rank-power tariff cannot improve for 0<t<=1. Six calls are
the sharp minimum for a nonzero complete-alphabet diagonal word.

Read [PROOF.md](PROOF.md), [RESULT.json](RESULT.json),
[verify.py](verify.py), [SOURCE-PINS.json](SOURCE-PINS.json) and
[CREDITS.md](CREDITS.md). The proof gives necessary and sufficient
chronology and exact-total conditions. It does not transfer geometric
image dimension into productive targets.

Run from the repository root with Python 3.12 and the standard library:

```sh
python3.12 -B scripts/research_package.py --package research/one-edge-word-obstruction
```

Both modes print `PASS_ONE_EDGE_WORD_OBSTRUCTION`. The source geometry
checker runs unchanged. No physical supplier, new kappa or selected-frontier
change follows. All five framework assumptions remain.

## Open next questions

- Can a fourth exponent relation change the productive endpoint? Start
with a precise paid-generated lattice and the full four-coordinate map.
Degree-two relations can change, but exact totals and physical costs
must still be checked.
- Can a different coordinate boundary make this alphabet useful? State
and charge the change of basis, dirty restoration and native consumer.
The present proof covers the original diagonal boundary only.

Ali Romman with substantial OpenAI Codex assistance. Apache-2.0.
Loading
Loading