Skip to content
Draft
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
291 changes: 291 additions & 0 deletions research/no-edge-dim5-irreducible-seven/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,291 @@
# An irreducible no-edge image of dimension 7 at pattern dimension 5

Prepared by unpaidmoon with substantial Cursor Grok assistance. Apache-2.0.
The fiber lemma is cited and not reproved. Everything else is proved here.

This is a written existence proof for one explicit pool. It is not a new κ.
The selected frontier stays `363005817/100000000000`. It is not an upper
bound on every pattern-dimension-5 pool, and it is not a census of those
pools. The 127-vector scan in `verify.py` checks this one image; it does
not search pattern spaces.

## 1. Statement

Work over F2. For a matrix N, write off(N) for N with its diagonal deleted.
For ambient width W, let P be a linear space of zero-diagonal W×W matrices
containing no matrix unit. The complete no-edge pool is

```text
C(P) = {x y^T : y·x = 1, off(x y^T) is a nonzero element of P}.
```

Every element is a rank-one idempotent. Coordinate projectors have zero
pattern, so they are not in C(P). Write V(P) for the span of the images of
C(P). The action is reducible when some subspace 0 < U < V(P) is invariant
under every pool projector.

**Theorem.** For every W ≥ 7 there is a 5-dimensional unit-free pattern
space whose complete pool has seven projectors, joint image of dimension 7,
and irreducible action.

Explicitly, on coordinates {0,1,2}, {3,4,5} and {6},

```text
(e0+e2)(e1+e2)^T,
(e1+e2)(e0+e2)^T,
(e0+e1+e2)(e0+e1+e2)^T,
(e3+e5)(e4+e5)^T,
(e4+e5)(e3+e5)^T,
(e3+e4+e5)(e3+e4+e5)^T,
(e2+e5+e6)(e2+e5+e6)^T.
```

Let z = e2+e5+e6 and let P be the span of the seven off-diagonal patterns,
padded by zero rows and columns when W > 7. Then C(P) is exactly these
seven projectors, dim V(P) = 7, and the action on V(P) is irreducible.

In particular, a complete no-edge pool of pattern dimension 5 need not have
total joint image at most 6, and its irreducible image can have dimension 7.
Both answers use this one pool. The argument uses completeness, the absence
of matrix units, and pattern dimension exactly 5 for this P. It does not
bound any other pattern space, and it does not price a word.

## 2. The seven patterns

Write A = {0,1,2} and B = {3,4,5}. The first three projectors are the
normal-form triple on A. Put

```text
u = e1+e2, v = e0+e2, o = e0+e1+e2.
```

Then v·u = 1 and o·o = 1, so u v^T, v u^T and o o^T are rank-one
idempotents. Their off-diagonal patterns are

```text
p_u = E10 + E12 + E20,
p_v = E01 + E02 + E21,
p_o = E01 + E02 + E10 + E12 + E20 + E21.
```

These have weights 3, 3 and 6. The first two are disjoint and their sum is
the third, so they are the three nonzero elements of a 2-dimensional space
Q_A supported on the off-diagonal positions of A. The same formulas shifted
by 3 give a 2-dimensional space Q_B on block B, with weights 3, 3 and 6.
The two blocks use disjoint positions, so H = Q_A ⊕ Q_B is 4-dimensional.

The seventh pattern comes from z = e2+e5+e6. This vector has weight 3, so
z·z = 1 and z z^T is a rank-one idempotent. Its off-diagonal pattern is

```text
p_* = E25 + E26 + E52 + E56 + E62 + E65.
```

It has weight 6. None of its six positions lies in an off-diagonal A-block
or B-block: each position meets coordinate 6 or crosses from A to B. Thus
p_* is outside H. Let P = H ⊕ span(p_*). Then dim P = 5.

A general element is q_A + q_B + λ p_* with q_A in Q_A, q_B in Q_B and
λ in F2. If λ = 0 and the element is nonzero, its weight is the sum of a
weight in {0,3,6} from each block, not both zero, hence at least 3. If
λ = 1, all six positions of p_* survive, because neither block meets them,
so the weight is at least 6. Every nonzero element has weight at least 3.
In particular P contains no matrix unit.

## 3. Support

Let x y^T be a pool element, so y·x = 1 and the pattern p = off(x y^T) lies
in P and is nonzero. Call p even when λ = 0 and odd when λ = 1.

Every matrix in P is zero in every row and every column indexed by
t ≥ 7. If x_t = 1, then y_j = p_tj = 0 for all j ≠ t, so y = e_t and
y·x = 1 forces y_t = 1. The pattern is then the off-diagonal part of row t,
which is zero, contradicting membership in the pool. Thus x_t = 0. If
y_t = 1, column t of the pattern is x with bit t cleared, hence x = e_t.
Combined with x_t = 0 this gives x = 0, which is impossible. Thus y_t = 0.
Both factors are supported on {0,1,2,3,4,5,6}.

The same reading applies to coordinate 6 when p is even: an even pattern
has zero row 6 and zero column 6, so x_6 = y_6 = 0 by the same two
sentences. An even pattern also has no cross entry between A and B.

## 4. The odd coset

Every odd pattern has λ = 1, so it contains all six positions of p_*. Those
entries are x_i y_j. Therefore

```text
x2 y6 = x6 y2 = x5 y6 = x6 y5 = x2 y5 = x5 y2 = 1,
```

and x_i y_6 = 0 for every i outside {2,5,6}, because every matrix in P has
zero column 6 outside rows 2 and 5. The first family gives
x2 = x5 = x6 = y2 = y5 = y6 = 1. The second gives x_i = 0 outside
{2,5,6}. Thus x = z. Row 6 of an odd pattern is zero outside columns 2 and
5, and x6 = 1, so y_j = 0 outside {2,5,6}. Thus y = z.

The projector is z z^T. Its pattern is exactly p_*, whose blocks inside A
and inside B are zero. An odd pattern with a nonzero block summand is
therefore not the pattern of this projector. It has no other candidate,
because the six entries already force the factors. The only odd pool element
is z z^T.

## 5. The even coset

Let p = q_A + q_B be even and nonzero, and let x y^T realize it. Section 3
puts both factors on A ∪ B, with no cross entry: x_a y_b = 0 and
x_b y_a = 0 for all a in A and b in B.

The internal A-pattern is the restriction of x y^T. If it is nonzero, some
x_a = y_a' = 1, so both the A-part of x and the A-part of y are nonzero.
From x_a = 1 and the cross equations, every B-bit of y is zero. From
y_a' = 1, every B-bit of x is zero. The internal B-pattern is then zero.
The symmetric argument says that a nonzero B-pattern forces the A-pattern
to vanish. A pattern with both blocks nonzero is unrealized.

So p lies in Q_A or in Q_B. Consider Q_A. Its three nonzero elements are
p_u, p_v and p_o, of weights 3, 3 and 6, and the three displayed projectors
realize them. None is a matrix unit. By the fiber lemma of #5921, a
realizable nonzero off-diagonal pattern that is not a matrix unit has
exactly one rank-one idempotent lift. Those three lifts are the first three
projectors, and the same holds for Q_B. There is no further even pool
element.

## 6. The pool and the image

Sections 4 and 5 account for every nonzero pattern in P. For every W ≥ 7,

```text
C(P) = the seven projectors in Section 1.
```

The first three images are u, v and o. Then u+o = e0, v+o = e1 and
u+v+o = e2, so they span the coordinate space L = span(e0,e1,e2). The next
three span R = span(e3,e4,e5) by the same sums. These two coordinate spaces
meet only at zero, and z is not in L ⊕ R because of e6. Therefore

```text
V(P) = L ⊕ R ⊕ span(e6), dim V(P) = 7.
```

Section 7 shows that the block covectors span L* ⊕ R* and that ζ is nonzero
on e6, so the seven covectors span the dual of V(P). The common kernel is
zero. That is not yet irreducibility.

## 7. Irreducibility

Identify V(P) with F2^7 on the ordered coordinates
(e0,e1,e2,e3,e4,e5,e6). Write a vector as ℓ + r + γ e6 with ℓ in L and
r in R. The linking covector is ζ = e2* + e5* + e6*, so

```text
ζ(ℓ + r + γ e6) = ℓ2 + r5 + γ.
```

The three block-A covectors, in coordinates (e0,e1,e2), are the rows

```text
0 1 1
1 0 1
1 1 1.
```

Adding row 0 to row 1 produces (1,1,0); adding that to row 2 produces
(0,0,1); back-substitution produces the identity. They are a basis of L*.
The block-B covectors are the same matrix on (e3,e4,e5), hence a basis of
R*. Each block image meets z in one coordinate: u, v and o each meet z in
e2, and each B-image meets z in e5. Therefore ζ is 1 on all six block
images. The block covectors are the swapped copies u*, v*, o* and their
translates, so each block covector is likewise 1 on z. Also ζ(z) = 1.

Let U be a nonzero invariant subspace.

If some w in U has ζ(w) = 1, the linking projector sends w to z, so z is in
U. Every block covector is 1 on z, so every block projector sends z to its
image, and all six images lie in U. The three A-images span L and the three
B-images span R, so L ⊕ R is contained in U. Then e2 and e5 lie in U, and
z + e2 + e5 = e6 lies in U. Hence U = V(P).

If instead ζ vanishes on every vector of U, none of the six block images can
lie in U, because ζ is 1 on each of them. An invariant subspace that does not
contain a block image cannot be hit by that block's covector. All three
A-covectors therefore vanish on U. They span L*, so the L-component of every
vector in U is zero. The same holds for R. The relation ζ = 0 then forces the
e6-coordinate to vanish as well. Thus U = 0, contradicting the choice of U.

Every nonzero invariant subspace equals V(P). The action is irreducible.

Deleting the linking projector leaves the first six. Their covectors vanish
on the opposite block, and each block's projectors map into that block, so
L is a proper nonzero invariant subspace. The link is what makes the
seven-projector action irreducible. A single normal-form block is already
reducible for the same reason; the verifier uses the deleted pool, rather
than that smaller block, as the control.

## 8. The parity construction does not decide the question

The question asks about the parity construction one dimension above the
pattern-dimension-4 example. That construction is the even-weight column,
and it stays at total image 6.

Let S be the even-weight subspace of the coordinate space on {1,2,3,4,5,6}.
It is the kernel of the all-ones functional, so dim S = 5, and every nonzero
vector of S has weight at least 2. Let P_even be the 5-dimensional space of
matrices s e0^T with s in S. A nonzero element has weight at least 2, so
P_even contains no matrix unit.

A pool element x y^T has its only possible nonzero column in column 0. If
y_t = 1 for some t ≠ 0, column t of the pattern is zero, so x = e_t, and
column 0 is then 0 or e_t. The pattern is 0 or the matrix unit E_t0. Neither
lies in the pool. Thus y = e0 and x_0 = 1. The pattern is the vector x + e0
placed in column 0, so x = e0 + s for some nonzero s in S. These 31
projectors are exactly the pool, at every width at least 7. Their images
are e0 + s. Differences span S, and e0 is outside S, so the image has
dimension 6. Every covector is e0. The space S lies in the image and is
orthogonal to e0, so every projector sends S to 0. Thus S is a proper
nonzero invariant subspace. The parity pool is reducible and has total
image 6.

It therefore does not show that the total image can exceed 6, and it does
not show that an irreducible image can exceed 5. Section 1 is the pool that
decides both questions.

## 9. What this does not claim

The selected κ and its framework assumptions are unchanged. No rational
address, native word, supplier admission, or multiplication identity is
produced.

The theorem does not say that every complete no-edge pool of pattern
dimension 5 has joint image at most 7, and it does not say that 7 is the
only irreducible dimension above 5. The edge-quotient bound of #6782, in
the zero-edge case, still gives only dim V ≤ 2^5 − 1 = 31. A search through
one family of extensions is not a substitute for that upper bound, and none
is claimed here.

The 127 nonzero vectors of this image form a finite set. The verifier
closes every one under the seven projectors and finds that each closure is
all of V(P). That census matches Section 7; it is not a search over pattern
spaces. The width scans certify this pool and the parity pool only.

Question 2 of #7139, the coordinate classification of the dimension-6 pools
at pattern dimension 4, is not this result. Question 3 of #7139, paid-rank
accounting after a relation split, remains question 1 of #6921 and is not
answered here. The original card still asks for a new register alphabet;
that gate is unmet.

## Open next questions

1. **Is 7 an upper bound?** Does every complete no-edge pool of pattern
dimension 5 have total joint image at most 7? The witness shows that 6
is too small, and the zero-edge quotient bound is still 31. The missing
step is to show that a unit-free one-dimensional extension of a
dimension-6 pattern space adds at most one image dimension. This package
does not prove that estimate.

2. **Is the dimension-7 pool unique up to coordinates?** Is every
irreducible complete no-edge pool of pattern dimension 5 and joint-image
dimension 7 a coordinate permutation of the seven projectors in Section
1? The existence proof does not identify the pools. The first step is to
match the two normal-form blocks and the linking vector. Nothing in
Section 7 classifies other links.
107 changes: 107 additions & 0 deletions research/no-edge-dim5-irreducible-seven/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,107 @@
# No-edge pattern dimension 5 has an irreducible image of dimension 7

**Result.** For every ambient width at least 7, the seven projectors

```text
(e0+e2)(e1+e2)^T,
(e1+e2)(e0+e2)^T,
(e0+e1+e2)(e0+e1+e2)^T,
(e3+e5)(e4+e5)^T,
(e4+e5)(e3+e5)^T,
(e3+e4+e5)(e3+e4+e5)^T,
(e2+e5+e6)(e2+e5+e6)^T
```

are the complete pool of their pattern span. That span is five-dimensional
and contains no matrix unit. The joint image is seven-dimensional and the
action is irreducible. **No new κ.** The selected value remains
`363005817/100000000000`.

A complete no-edge pool of pattern dimension 5 therefore need not have total
joint image at most 6, and its irreducible image can have dimension 7. The
parity construction one dimension above the pattern-dimension-4 example still
has total image 6 and is reducible, so it does not decide either question.

## Scope and evidence

Catalogue card: `bit-new-register-model`, priority revision 8, base commit
`e54523517d0f03e58352e647c403a5af3e525a57`. Scope status: **changed**. The
card asks for a new register alphabet. This package answers open question 1
of #7139: whether a complete no-edge pool of pattern dimension 5 still has
total joint image at most 6, and whether its irreducible image can exceed 5.
The reason is recorded in [TASK.json](TASK.json). The original native
acceptance gate is unmet, and the selected frontier is unchanged. Question 2
of #7139, the coordinate classification at pattern dimension 4, and question
3, the paid-rank split, are not this result.

[PROOF.md](PROOF.md) is the argument. Sections 2–6 identify the pattern space
and prove that the complete pool is exactly these seven projectors at every
width at least 7. Section 7 proves the action is irreducible: a vector hit
by the linking covector generates the whole image, and a subspace killed by
that covector cannot contain a block image, which forces it to be zero.
Section 8 records the even-weight column of dimension 5: total image 6, with
the even-weight space invariant. That is the parity construction named in
the question. It is a comparison, not a bound.

[verify.py](verify.py) repeats the finite parts: the pattern-space weights,
the six linking entries on every odd coset vector, the dual bases, the
complete pools at widths 7, 8 and 9, the closure of all 127 nonzero image
vectors, the deleted-link control, and the parity pool at widths 7 and 8.
Deleting the linking projector leaves the coordinate block on {0,1,2}
invariant. The parity pool's even-weight line is invariant, while closing an
image vector under that pool spans the whole image, so the reducible control
is not an arbitrary seed. These checks certify the two explicit pools. They
do not enumerate five-dimensional pattern spaces. Sections 4–7 are what
cover every width.

Source pins are in [SOURCES.json](SOURCES.json). The catalogue is provenance
at the base commit, and the selected record is an input because the verifier
reads it. Full Git history is needed to resolve the pins. Nothing here reruns
the selected baseline as a scientific replay.

Overlap was checked on 2026-11-11. A local `find_overlap.py --offline` search
at base `e54523517d0f03e58352e647c403a5af3e525a57` covered 10178 notes with a
complete head inventory. The matches are the predecessor packages for pattern
dimension at most 4. The remote overlap cache was not available in that run.
A live GitHub search of `CrocSwap/integer-mult-bounds` for the pattern-dimension-5
no-edge question returned the predecessor chain and open #7094 and #7139, which
ask the question and do not answer it. Open #7138 is the weight-1 pattern-dimension-4
bound. Recent comments on discussion #1166, checked the same day, do not take
this question. No match is not a novelty certificate.

## Reproduce

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

```sh
python3 -B research/no-edge-dim5-irreducible-seven/verify.py
python3 -I -O -B research/no-edge-dim5-irreducible-seven/verify.py
python3 -B scripts/research_package.py --package research/no-edge-dim5-irreducible-seven --check-only
python3 -B scripts/research_package.py --package research/no-edge-dim5-irreducible-seven
```

The verifier prints `PASS_NO_EDGE_DIM5_IRREDUCIBLE_SEVEN`.

## Open next questions

1. **Is 7 an upper bound at pattern dimension 5?** Does every complete no-edge
pool of pattern dimension 5 have total joint image at most 7? This witness
shows that 6 is too small. The zero-edge quotient bound is still 31. The
smallest next step is the one-dimensional extension of a dimension-6
pattern space. A width-bounded search would not prove the bound.

2. **Is the dimension-7 pool unique up to coordinates?** Is every irreducible
complete no-edge pool of pattern dimension 5 and joint-image dimension 7 a
coordinate permutation of these seven projectors? The existence proof does
not identify the pools. The smallest next step is to match the two
normal-form blocks and the linking vector.

## Credit

The fiber lemma is Nick Petalas, #5921. The parity construction compared in
Section 8 is CubeJerry, #6838, moved up to pattern dimension 5. The normal-form
triple is the one classified in #6921; this proof uses only the displayed
projectors and the fiber lemma. #7139 asked the question and proved the
pattern-dimension-4 bound; that bound is not an input. #6782 supplies the
zero-edge quotient bound 31. Prepared by unpaidmoon with substantial Cursor
Grok assistance. New work is Apache-2.0.
Loading