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
165 changes: 165 additions & 0 deletions research/order-seventeen-clifford/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,165 @@
# Order seventeen and the sharp rank-eight response cap

Prepared by apoursh with substantial OpenAI Codex assistance. Apache-2.0.
The upper bound and Bell-spectator method are credited to Nick Petalas's
[PR6192](sources/pr6192-PROOF.md), head
`91df334ccd2a4bc9959e311f118ac8aaf967ffc2`; its Pauli-character step retains
Ross Richardson's credit. The new contribution is the explicit order-17
matrix, signed-generator certificate and rank-eight specialization.

## 1. Conventions and the four-qubit construction

Index the computational basis by integers `x=sum_j 2^j x_j`, with bits
`j=0,1,2,3`. Matrix products act rightmost first. Let

```text
Hg = (1+i)/2 [[1,1],[1,-1]], S = diag(1,i),
A = [[0,0,0,1], [0,0,1,0], [0,1,0,1], [1,0,1,1]],
U = i S_3 CZ_03 CZ_12 CZ_23 Hg^(tensor 4).
```

All factors are Gaussian-dyadic unitaries normalizing the Pauli group;
therefore U is a Gaussian-dyadic Clifford. Because `Hg=e^(pi i/4)H`,
its fourth tensor power is `-H^(tensor 4)`. Consequently every matrix
entry is given explicitly by

```text
U[x,y] = (-i/4) i^(x_3)
(-1)^(x_0 x_3 + x_1 x_2 + x_2 x_3 + sum_j x_j y_j).
```

The prefactor `-i` is part of the assertion. Replacing U by iU preserves
Cliffordness but changes its seventeenth power to iI and destroys Phi17.

Here is a complete integer certificate specification. Set `G=4U`, let
`P_0=I16` and `P_(j+1)=P_j G`, all over Gaussian integers. Exact
multiplication gives

```text
G G* = 16 I16,
P_17 = 4^17 I16,
sum_(j=0)^16 4^(16-j) P_j = 0,
Tr(G) = -4.
```

These are equalities of all 256 entries, not sampled vectors or floating
point residuals. `independent.py` evaluates exactly this unreduced integer
recurrence. `verify.py` separately reduces a common denominator after each
Gaussian matrix operation, and checks all powers U^j for j=0,...,17.
It additionally finds denominator four and trace -1 for every j=1,...,16.
Thus U has exact order 17 and satisfies the actual whole-bank primitive
relation `Phi17(U)=I+U+...+U^16=0`. Exact order follows because 17 is prime
and U is not the identity, as its trace is -1 rather than 16.

This is a complete finite certificate backed by explicit arithmetic, not
a formal-kernel verification or an extrapolation to unknown matrices.

## 2. Signed Pauli action

Use `X^x Z^z` in that order and Hermitian representatives
`P(x,z)=i^(sum_j x_j z_j) X^x Z^z`. Write binary labels as columns
`(x_0,...,x_3,z_0,...,z_3)`. Direct conjugation gives

```text
U X_j U* = Z_j,
U Z_j U* = i^(A_jj) X_j product_k Z_k^(A_kj).
```

Each displayed image is the positive Hermitian representative. Both
checkers verify all eight generator equalities as full complex matrices,
including their phases. Hence the induced binary symplectic map is

```text
S_U = [[0,I4],[I4,A]],
row masks = [16,32,64,128,129,66,164,216].
```

Its order is 17, its native upper-right block is `I4` of rank four,
and `rank(S_U-I8)=8`. For the last claim, a vector in its kernel has
`x=z` and `Ax=0`. A is invertible: `Av=(v3,v2,v1+v3,v0+v2+v3)`
vanishes only for v=0. Thus the kernel is trivial. These facts are also
checked over F2 without using floating point ranks.

## 3. Eight-qubit saturation

Let D be the four active qubits and E four identity spectators. Put

```text
V = (product_(j=0)^3 CNOT(j -> j+4)) (Hg^(tensor 4) tensor I16),
W = V^-1 (U tensor I16) V.
```

V and its inverse are Gaussian-dyadic Clifford circuits. Polynomial
evaluation commutes with tensoring an identity and with conjugation, so

```text
Phi17(W) = V^-1 (Phi17(U) tensor I16) V = 0.
```

This identity is on all 256 coordinates; the spectators need not be
initialized. It does not assert that their state is restored by a caller:
W generally couples the two registers. The proof avoids an unnecessary
dense 256-dimensional power computation by using the already certified
16-dimensional identity and these exact algebraic operations.

To compute the native block, order labels as `(x_D,x_E,z_D,z_E)`.
For input Z label `(d,e)`, V sends `(0,0,d,e)` to `(d,d,e,e)`.
The embedded active map sends this to `(e,d,d+Ae,e)`.
Applying V^-1 gives output X label

```text
(d+(A+I)e, d+e).
```

Therefore `B_W=[[I,A+I],[I,I]]`. If it kills `(d,e)`, the second
equation gives d=e and the first then gives Ae=0, so d=e=0.
Thus its rank is eight. Both checkers also build V from its individual
binary Hadamard/CNOT actions, perform the complete conjugation and
recover this block. The primary checker verifies symplecticity, order
and defect rank eight. Additional identity spectators preserve both
the polynomial relation and native rank eight for every M>=8.

## 4. Why eight is the cap

This is the n=17 specialization of Petalas's theorem, not a new general
upper-bound argument. A whole-bank Gaussian-rational Clifford T on M
qubits satisfying Phi17(T)=0 has characteristic polynomial `Phi17^t`,
where `t=2^(M-4)`. Indeed Phi17 is irreducible over Q(i), since
`[Q(zeta68):Q(i)]=phi(68)/2=16`. Thus M>=4 and `Tr(T)=-2^(M-4)`.

On the fixed space K of its binary Pauli action S_T, the signs of
Hermitian Pauli conjugation form a character. Their sum is the trace
of the adjoint action, namely `|Tr(T)|^2=2^(2M-8)>0`. A character
sum is either zero or |K|, hence `dim K=2M-8` and
`rank(S_T-I)=8`. The native B block is a submatrix of S_T-I, so
its rank is at most `min(M,8)`. Section 3 attains eight for every
M>=8. Four is the least possible active width; eight is the least
ambient width that can have an 8-by-8 full-rank native block.

The complete cited proof, including the character argument and its
classical field-degree source, is retained in
[the predecessor document](sources/pr6192-PROOF.md). Its pending status
does not imply maintainer acceptance. The concrete identities above
are checked independently of predecessor executable code.

## 5. Boundaries and failure controls

Finite order alone is insufficient. U tensor U has order 17, but

```text
Tr(Phi17(U tensor U)) = 16^2 + 16(-1)^2 = 272 != 0.
```

Other controls show that removing a CZ edge or omitting one polynomial
term fails Phi17; identity spectators without V leave native rank four;
an asymmetric proposed phase matrix is not symplectic; and the ordinary
companion matrix of Phi17 satisfies the polynomial but is not unitary.
These distinguish actual hypotheses rather than testing only that the
code can reproduce its own constants.

The displayed W uses 12 H-type factors (including inverse Hg), eight
CNOTs, three CZs, one S and a scalar phase i. This is only an algebraic
factor inventory. Whether those factors are native, how they align with
an actual producer/consumer, and their complete paid and target bill
are not established. There is no ownership/restoration certificate,
new supplier, new kappa or full multiplication theorem here.
149 changes: 149 additions & 0 deletions research/order-seventeen-clifford/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,149 @@
# An explicit order-seventeen Gaussian-dyadic Clifford

**Scoped construction and exact finite certificate; no new multiplication bound.**

This answers the order-17 existence question in Nick Petalas's pending
[PR6192](https://github.com/CrocSwap/integer-mult-bounds/pull/6192), pinned at
`91df334ccd2a4bc9959e311f118ac8aaf967ffc2`. We give a four-qubit Clifford
whose actual matrix satisfies `Phi17(U)=0`, then apply that predecessor's
Bell-spectator method to attain its native `B`-rank cap of eight on eight
qubits. The construction works over `Z[i,1/2]`, including its global phase.

With little-endian qubits `0,1,2,3`, rightmost factors acting first, set

```text
Hg = (1+i)/2 [[1,1],[1,-1]], S = diag(1,i),
U17 = i S_3 CZ_03 CZ_12 CZ_23 Hg^(tensor 4).
```

The complete `16 x 16` certificate gives `U17^17=I`, `Phi17(U17)=0`,
trace `-1`, all eight signed Pauli-generator images, symplectic defect
rank eight and native `B` rank four. On four added identity spectators,
let `V` prepare four Bell pairs and put `W=V^-1(U17 tensor I16)V`.
Then `Phi17(W)=0`, and its native block is

```text
B_W = [[I, A+I], [I, I]], A = [[0,0,0,1],
[0,0,1,0],
[0,1,0,1],
[1,0,1,1]].
```

Since `A` is invertible over `F2`, `B_W` has rank eight. Adding identity
qubits preserves that rank, so the cap is attained for every ambient
width `M>=8`. See [PROOF.md](PROOF.md) for conventions, the finite
certificate, algebraic extension and credited upper-bound argument.

## Evidence and limits

`verify.py` uses exact Gaussian integer numerators with reduced denominators.
It checks all matrix entries, all sixteen nonidentity powers, all signed
Pauli-generator images, and the complete binary Bell conjugation. Seven
controls distinguish wrong phase, an omitted CZ, a truncated polynomial,
finite order without a primitive response, unused spectators, a nonsymplectic
prediction and a nonunitary cyclotomic companion.

`independent.py` was written separately, without importing the primary
checker. It uses unreduced Gaussian integers `G=4U17` and clears denominators
in the polynomial identity; it independently checks the eight Pauli images
and constructs the binary Bell circuit. Both checkers use explicit failures
that remain active under optimized Python. The eight-qubit polynomial
identity follows by tensoring and conjugation in the written proof; neither
checker expands all powers of a dense `256 x 256` matrix.

This is an explicit finite algebraic response, not an admitted supplier.
No source-bound caller, native legal schedule, ownership/dirty restoration,
complete paid/target invoice or multiplication theorem is supplied. Native
`B` rank is a binary algebraic invariant, not a physical gate-price receipt.
The selected conditional `kappa=781451/250000000`, its five framework
assumptions and selected evidence remain unchanged. No selected cold replay
was run. Earlier binding-only orientation was reused after confirming its
1227 bound inputs were unchanged from checkpoint `3f0152a32eeebac0df11946dbd50cabe5a7f7839` to this base;
that reuse is not a fresh scientific replay.

## Reproduce

From a full Git checkout, with Python 3.12 and no third-party dependencies:

```sh
python3.12 -B scripts/research_package.py --package research/order-seventeen-clifford --check-only
python3.12 -B scripts/research_package.py --package research/order-seventeen-clifford
```

The registered commands run both checkers normally and with `-O`, isolated
and with warnings treated as errors. Expected markers are
`PASS_ORDER_SEVENTEEN_CLIFFORD` and `PASS_INDEPENDENT_ORDER17`.
For a Git-free source export, run these direct commands from its root,
once normally and once with `-O` added:

```sh
python3.12 -I -B -W error research/order-seventeen-clifford/verify.py
python3.12 -I -B -W error research/order-seventeen-clifford/independent.py
```

[VERIFY.json](VERIFY.json) has no external live inputs: all mathematical
inputs are defined in the package. [SOURCES.json](SOURCES.json) pins the
base, original catalogue card and selected record as historical provenance.
The pending predecessor is not an ancestor dependency: its two cited source
documents are retained verbatim under `sources/` to make the precise theorem
and method available in an offline export. [SOURCE-PINS.json](SOURCE-PINS.json)
records their public source URLs, exact head and SHA256 digests, checked by
the primary verifier. Their bytes were also compared with the fetched head
before publication. No predecessor executable is imported or replayed.
The Git source-history preflight covers declared base provenance; it does
not authenticate the external predecessor-head label or validate its theorem.
Package-local changes already select this registered replay.

## Coordination and scope

Canonical main was pulled first at
`a5bff3bf133e482541c09eba4e773884bdb2b1b7`, priority revision 8.
The original `joint-query-controller` card is retained in [TASK.json](TASK.json).
Its scope is explicitly changed to this bounded algebraic subproblem: the
card's complete source/consumer acceptance gate remains unmet.
The canonical investigation is [Discussion 1166, comment 19273131](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19273131),
responding to the predecessor's released
[handoff](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19272674).

Overlap coverage was bounded: the local helper searched 30 of 8623 package
notes (8593 absent in this sparse checkout), without a fresh remote cache.
Live checks read the newest 15 handoff comments and all-state PR searches
for `Clifford seventeen`, `Phi17` and `order-17`; no competing explicit
instance was found in those results. This is not an exhaustive novelty audit.
The stopping rule was the first complete exact order-17 construction and
its rank certificate, without searching larger orders or proposing a
supplier-budget change.

## Open next questions

- **Can this response satisfy an actual source/consumer contract?** Its
algebraic feasibility removes one uncertainty, but does not remove paid
work. Start by identifying one current caller whose required whole-bank
action matches this exact matrix, then check native chronology, all
retained outputs and restoration. Blocker: no such caller is supplied here.
- **Can the rank-eight witness be implemented more cheaply than this Bell
conjugation in that caller's allowed alphabet?** A cheaper implementation
might affect a complete invoice. First establish the matched contract and
price the stated circuit, then compare one exactly equivalent circuit.
Blocker: without that contract and cost model, an algebraic gate count
cannot establish a saving. Do not expand the search before that gate.

The structured questions and blockers are also in [TASK.json](TASK.json).

## Credit

Prepared by **apoursh**, with substantial OpenAI Codex assistance for
construction, independently written exact checks, proof and packaging.
Nick Petalas retains credit for PR6192's odd-cyclotomic rank theorem and
Bell-spectator sharpness method; Ross Richardson retains the earlier
whole-bank cubic/Pauli-character argument (PR4081). Copied predecessor
documents retain their attribution and the repository's Apache-2.0 license.

Cyclic Clifford constructions are classical; for example, Oliver Kern,
Kedar S. Ranade and Ulrich Seyfarth give cyclic MUB constructions in
power-of-two dimensions in
[arXiv:0912.4661](https://arxiv.org/abs/0912.4661).
This package claims a reproducible explicit Gaussian-dyadic instance and
rank-eight certificate for the repository's handoff, not historical priority
for cyclic Clifford existence or Bell conjugation. The cited literature is
context, not a replay dependency. New material is Apache-2.0.
19 changes: 19 additions & 0 deletions research/order-seventeen-clifford/SOURCE-PINS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
{
"source_pr": 6192,
"author": "Nick Petalas",
"head": "91df334ccd2a4bc9959e311f118ac8aaf967ffc2",
"status": "pending predecessor; cited theorem and method, not independently replayed",
"license": "Apache-2.0",
"files": [
{
"path": "sources/pr6192-PROOF.md",
"source_url": "https://github.com/CrocSwap/integer-mult-bounds/blob/91df334ccd2a4bc9959e311f118ac8aaf967ffc2/research/odd-cyclotomic-response/PROOF.md",
"sha256": "324f494b016b4c63573cb6ed3fd14cedc258621cb17a5b9800165d921a1445d0"
},
{
"path": "sources/pr6192-README.md",
"source_url": "https://github.com/CrocSwap/integer-mult-bounds/blob/91df334ccd2a4bc9959e311f118ac8aaf967ffc2/research/odd-cyclotomic-response/README.md",
"sha256": "dab05bd081f14224bdbe2a0429b8f4389fb930a2922a7544781b184bd4d0f7ec"
}
]
}
18 changes: 18 additions & 0 deletions research/order-seventeen-clifford/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
{
"schema_version": 1,
"base_commit": "a5bff3bf133e482541c09eba4e773884bdb2b1b7",
"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"
}
]
}
Loading