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
1,196 changes: 1,196 additions & 0 deletions research/clifford-five-qubit-modular-seed/CHARACTERS.json

Large diffs are not rendered by default.

58 changes: 58 additions & 0 deletions research/clifford-five-qubit-modular-seed/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,58 @@
# The raw width-15 Gaussian-dyadic splitting

Let $R=\mathbb Z[i,1/2]$, $G=\mathrm{Sp}_{10}(2)$, and $C$ be the elementary abelian group of 64 alternating shears supported on the first four of the five binary coordinates. For the full five-qubit Clifford alphabet, choose one Gaussian-dyadic unitary lift $U_a$ of each projective class. Let $A_f$ have columns $\operatorname{vec}(U_a^{\otimes f})$, without normalization. The minimum integer width with an $R$-linear left inverse of $A_f$ is **15**.

## From one Bell seed to an integral splitting

Use the seven Bell labels $(0,Z_1,Z_2,Z_3,Z_4,Z_5,X_5)$ and one unpaired tensor factor. Their simultaneous stabilizer in the symplectic label group is $C$. The real controlled-$Z$ lifts of its six generators act trivially on the corrected seed phase. Every left-right translate of the subgroup sum then occurs as a Bell observation after an $R$-linear correction. The explicit correction for seven pairs uses the conjugate singleton $\bar U_a$ and acts linearly on mixture coefficients. All Bell contractions and the final Pauli trace division by 32 use only Gaussian dyadics. These are the seven-pair seed and interpolation implication proved in Sections 1, 2 and 4 of #7803. [Pinned Bell correction and recovery](https://github.com/apoursh/integer-mult-bounds/blob/0fefff647d869054f2e2c4eef3cdcb6e3c60c573/research/clifford-general-width/PROOF.md)

Put $e=64^{-1}\sum_{c\in C}c$ and $I=R[G]eR[G]$. If $1\in I$, expand it in translated subgroup averages, recovering a single symplectic fibre from the seed observations. Translation gives the remaining fibres and trace pairing within a fibre recovers every coefficient. It suffices to show $R[G]/I=0$.

For any field $k$ of odd characteristic, the ideal $k[G]e k[G]$ equals $k[G]$ if and only if every simple $k[G]$-module has a nonzero $C$-fixed vector. Indeed, $eM=M^C$. Every nonzero quotient of the group algebra by a two-sided ideal has a simple module, obtained by taking a maximal left ideal of the quotient. An obstruction therefore supplies a simple module killed by $e$, and conversely any such module witnesses a proper ideal. This argument also applies in the nonsemisimple characteristics.

The module $R[G]/I$ is finite over $R$. Every maximal ideal of $R$ has a finite residue field of odd characteristic. The two-sided ideal commutes with base change at the level of the quotient, by right exactness and by using the finite set of generators $g e h$. An algebraic closure of a residue field is a faithfully flat extension of that field. Thus the vanishing of the quotient over the algebraic closure implies that its residue fibre vanishes. Nakayama's lemma then gives vanishing at each maximal ideal. A nonzero finite module has a maximal ideal in its support, proving the required global implication. This establishes existence of Gaussian-dyadic coefficients for the seed expansion.

## The ordinary and modular tests

The three relevant classes are the identity, a rank-two alternating shear $c_2$, and a rank-four alternating shear $c_4$. Their multiplicities in $C$ are $1,35,28$. For any Brauer character $\varphi$ in odd characteristic define
$$
b(\varphi)=\frac{\varphi(1)+35\varphi(c_2)+28\varphi(c_4)}{64}.
$$
The restriction to the elementary abelian 2-group is a sum of sign characters. Its eigenvalues lift uniquely to $\pm1$ in characteristic zero, so this exact integer is the dimension of the fixed subspace, including in characteristics that divide the dimension. It is additive on short exact sequences.

The group order is $2^{25}3^6 5^2\cdot7\cdot11\cdot17\cdot31$. For a prime away from this order the ordinary character table gives the Brauer table over an algebraically closed residue field. In modular character theory these are all defect-zero blocks and their ordinary characters reduce irreducibly with unchanged lifted values. All 198 ordinary irreducibles have $b>0$, as found in #7806 and checked again from the three ordinary class columns here. For each of the five available odd-characteristic symplectic tables in CTblLib, the full modular computation gives the following results. [Ordinary predecessor census](https://github.com/apoursh/integer-mult-bounds/blob/b42e665721e76e69d05f24a19bdabee0e0f77c04/research/clifford-five-qubit-threshold/MOMENTS.md) · [CTblLib symplectic table record](https://www.math.rwth-aachen.de/~Thomas.Breuer/ctbllib/ctbltoc/data/S10%282%29.html)

| Characteristic | Irreducible Brauer rows | Least $b$ | Least $b$ on a nontrivial row |
|---:|---:|---:|---:|
| 5 | 162 | 1 | 15 |
| 7 | 187 | 1 | 15 |
| 11 | 195 | 1 | 15 |
| 17 | 192 | 1 | 15 |
| 31 | 195 | 1 | 15 |

At 3, compare the restrictions to the simple plus and minus orthogonal subgroups containing the same $C$. If a nonzero $G$-module had zero $C$-fixed dimension, the additivity and nonnegativity of $b$ would restrict every composition factor in each subgroup to one of its zero rows. At $c_2$ the normalized Brauer trace of such a module would have to be both $-1/31$ and $-67/1309$. The two rational numbers differ, and lifting the involution eigenvalues makes this an equality test in characteristic zero. This proves positivity in the last exceptional characteristic. [Subgroup embeddings, class fusions and all zero rows](THREE.md)

The global argument now gives the raw $R$-left inverse. Given a target row at width 15, tensor it with the linear functional $\operatorname{vec}(U_a)^*/32$ for each extra slot. This preserves its value one on the target and zero on the others, and it also shows that a left inverse at any width below 14 would give one at 14. Width 14 already has a nonzero complex kernel of dimension 6,954,742,423,552 by #7806. Since base change to $\mathbb C$ preserves a left inverse, this proves the lower bound as well. [Pinned width-14 kernel](https://github.com/apoursh/integer-mult-bounds/blob/b42e665721e76e69d05f24a19bdabee0e0f77c04/research/clifford-five-qubit-threshold/PROOF.md)

## A universal lower bound for the fixed dimension

The same data give a quantitative result. For every finite-dimensional $G$-module $M$ in odd characteristic, let $d=\dim M$, $b=\dim M^C$ and $t$ be the Brauer trace at $c_2$. Then
$$
b\ \ge\ \frac{6}{6919}d.
$$
To see this at 3, the 39 plus-group Brauer rows satisfy $31t+d\ge32b$ and the 35 minus-group rows satisfy $1309t+67d\le29920b$, including the respective zero rows. All three quantities are additive on restriction and passage to composition factors. Canceling the shared trace gives
$$
768d=31(1309t+67d)-1309(31t+d)
\le (31\cdot29920-1309\cdot32)b=885632b.
$$
The coefficient is sharp for the relaxation that matches only degree and the two shear traces across restrictions. For the plus group take $168\varphi_1+223\varphi_4$, and for the minus group take $6\varphi_{13}+7\varphi_7$. Their common triple is $(193732,-6076,1060)$ with $b=168$, attaining $6/6919$. Whether a symplectic-group module attains that fraction at 3 remains open.

Away from the group order the ordinary minimum of $b/d$ is $15/5797$. The same minimum holds in characteristics 5, 11, 17 and 31. In characteristic 7 it is $45/40393$. These are minima over all nonzero modules because every module has a composition series and $b/d$ is a dimension-weighted average of the simple ratios. Each of these minima exceeds $6/6919$. Extending the ground field to its algebraic closure preserves both dimensions since $e$ is an idempotent with image $M^C$.

## Reconstructing the finite table calculation

Write $O$ for the three selected columns of the ordinary character table, $D$ for the ordinary-to-modular decomposition matrix and $B$ for the same three columns of the table of irreducible Brauer characters. The three classes are regular for every odd prime, so $D B=O$. The numerical evidence stores $O$ and the positive entries of $D$, indexed from one. For each of the seven tables a sequence of integer pivots recovers all unknown rows of $B$. Substitution into every ordinary row checks the complete overdetermined system. Ordinary column orthogonality separately checks each group's order, the two centralizers and the vanishing cross products.

The Fourier coefficients of the restricted class function on $C$ provide a second set of integer constraints. The 64 signs have three orbits under change of basis on the four-space, indexed by ranks 0, 2 and 4. The sums of each representative sign over the three original rank classes are respectively $(1,35,28),(1,3,-4),(1,-5,4)$. For example, enumerate the 64 symmetric zero-diagonal bit matrices in off-diagonal order $(12,13,14,23,24,34)$. A representative sign is the parity on none, on coordinate 12, or on 12 plus 34. The replay checks divisibility by 64, nonnegativity and recovery of the full degree from these three multiplicities. GAP's native irreducible Brauer tables independently reproduce all 1,005 modular triples in their published ordering.

The theorem applies to raw observations. At odd width the Hilbert–Schmidt unit normalization contains $\sqrt 2$ when there are five qubits. A normalized inverse follows over the extended ring $R[\sqrt 2]$ by rescaling. None of these existence statements prices the extraction of all fibres, a coefficient-growth bound or a multiplication interface.
46 changes: 46 additions & 0 deletions research/clifford-five-qubit-modular-seed/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
# Five qubits split over the Gaussian dyadics at width 15

**Fifteen is the minimum raw tensor width for an exact Gaussian-dyadic left inverse of the full projective five-qubit Clifford alphabet.** The seven-Bell-pair seed already suffices at that width. Width 14 has a complex kernel, and #7803 previously gave an integral construction at width 17. The result closes the interval [15,17] from #7806. [Global proof](PROOF.md) · [Characteristic-three argument](THREE.md)

The obstruction is a simple symplectic-group module with no fixed vector for the same group of 64 shears. The published tables rule this out at 5, 7, 11, 17 and 31 in all 931 cases. At 3, the two orthogonal subgroups give incompatible trace ratios for an obstruction. Their tables have only three possible zero rows between them. Together with the ordinary calculation at primes away from the group order, these checks lift the subgroup-average identity to $\mathbb Z[i,1/2]$. The proof establishes existence of an exact raw inverse. A usable decoding circuit, a noise bound, the cost of its observations and any effect on the selected multiplication bound remain open.

The fixed space has dimension at least $6\dim(M)/6919$ for **every** finite-dimensional symplectic-group module in odd characteristic. The coefficient is sharp when matching the two orthogonal restrictions only on the identity and both shear classes. [Quantitative bound and matching witness](PROOF.md#a-universal-lower-bound-for-the-fixed-dimension)

## Reproduce the certificate

From the repository root run the registered CPython 3.12 checks. Their inputs are the compact numerical evidence and the code in this directory.

~~~sh
python3.12 -B scripts/research_package.py --package research/clifford-five-qubit-modular-seed
~~~

The primary checker recovers all 1,005 modular triples from exact decomposition equations, compares the two trace ratios and checks the ordinary class columns and the full restriction to the 64-element subgroup. A separate program generates all 64 binary shears, realizes the six generators as commutators in both orthogonal groups and counts the classes geometrically. [Retained numerical data](CHARACTERS.json) · [Primary checker](verify.py) · [Independent geometry check](check_orthogonal.py)

### Regenerating from the original sources

Two independent extraction routes reproduce the character calculation. The first uses the standard GAP interface with CTblLib 1.3.11. In this directory run

~~~sh
gap -r -T -A -q -b gap_rows.g > /tmp/clifford-five-brauer-rows.json
python3.12 -I -B verify.py --gap-rows /tmp/clifford-five-brauer-rows.json
~~~

The second route uses the original GAP archive for the ordinary characters and the published MOC decomposition-matrix PDFs for the seven modular tables. Put the original PDFs in one directory under their published names. Run the following from this package with Python and PyMuPDF. Both external methods were exercised during preparation, using GAP 4.14.0 and PyMuPDF 1.26.3 respectively. The registered replay needs only CPython and its standard library. [Release archive](https://www.math.rwth-aachen.de/~Thomas.Breuer/ctbllib/ctbllib-1.3.11.tar.gz) · [Archive and PDF SHA-256 pins](rebuild.py)

~~~sh
python -B rebuild.py /path/to/ctbllib-1.3.11.tar.gz /path/to/pdfs
python -B check_orthogonal.py /path/to/ctbllib-1.3.11.tar.gz
~~~

The symplectic PDFs are [mod 5](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/S10%282%29/S10%282%29mod5.pdf), [mod 7](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/S10%282%29/S10%282%29mod7.pdf), [mod 11](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/S10%282%29/S10%282%29mod11.pdf), [mod 17](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/S10%282%29/S10%282%29mod17.pdf) and [mod 31](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/S10%282%29/S10%282%29mod31.pdf). The orthogonal PDFs use the simple groups. [Plus group mod 3](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/O10%2B%282%29/O10%2B%282%29mod3.pdf) · [Minus group mod 3](https://www.math.rwth-aachen.de/~MOC/decomposition/tex/O10-%282%29/O10-%282%29mod3.pdf)

## Attribution and scope

The source character data belong to the ATLAS of Finite Groups, the Modular ATLAS, Gerhard Hiss and Thomas Breuer's GAP Character Table Library, with decomposition matrices published by the MOC project. CTblLib retains its GPL-3.0-or-later license. The package records numerical character values and decomposition multiplicities keyed to those published tables. The new proof and code were prepared by apoursh with substantial OpenAI Codex assistance under Apache-2.0.

The group-algebra reduction is from apoursh's raw Bell interpolation construction, with earlier Clifford representation and spectrum results by alephnan and Ali Romman. The complex lower bound and ordinary fixed-space calculation at five qubits are from #7806. The new result handles all odd residue characteristics and proves sharpness over the Gaussian dyadics. [Coordination claim](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19755739) [#7803 pinned Bell construction](https://github.com/apoursh/integer-mult-bounds/blob/0fefff647d869054f2e2c4eef3cdcb6e3c60c573/research/clifford-general-width/PROOF.md) · [#7806 pinned complex threshold](https://github.com/apoursh/integer-mult-bounds/blob/b42e665721e76e69d05f24a19bdabee0e0f77c04/research/clifford-five-qubit-threshold/PROOF.md)

## Open next questions

1. Can one exhibit a short translated-subgroup identity over the Gaussian dyadics? Start by constructing a sparse expression for 1 in the ideal of the seven-pair seed and bounding its support and coefficient sizes. The local-to-global existence argument supplies neither quantity.
2. Can an explicit five-qubit decoder at width 15 meet the complex noise target with a priced observation interface? Begin with one fibre and its required inputs, compare its row norm against the Gram enclosure in #7806, and account for acquisition, precision and storage. Recoverability alone does not establish a supplier.
55 changes: 55 additions & 0 deletions research/clifford-five-qubit-modular-seed/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,55 @@
{
"schema_version": 1,
"base_commit": "e54523517d0f03e58352e647c403a5af3e525a57",
"files": [
{
"path": "certificates/selected-result.json",
"usage": "provenance",
"git_blob": "75d77b328ea02ded535e7f23aee5b1e55d0bb1a8",
"sha256": "154c3d30455d778fd3343b6b4851f0fa4665f0f7625b0adb3675e2bafbbf6bf1"
},
{
"path": "docs/research/agent-task-catalog.json",
"usage": "provenance",
"git_blob": "3404fbdd5e7529a0674eef26114b5071e4586cec",
"sha256": "ead6cdc9d0af6d58e8a49744d8ec8c7c15e13d2ba93c276c434bcf9b51b93a81"
}
],
"cited_research": [
{
"pr": 6714,
"head": "f7f32a18dde54e67624a9ed9cc7e1b266d9585c0",
"author": "alephnan",
"title": "Prove the sharp three-qubit Clifford tensor threshold T(3)=10",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/6714"
},
{
"pr": 7766,
"head": "d7c2dd1868c70f0addce2c0a4d3220149452f14b",
"author": "aliromman",
"title": "Prove the sharp four-qubit Clifford tensor threshold T(4)=13",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7766"
},
{
"pr": 7799,
"head": "7f08a74037788311cc6b7da27297b25f1b188e45",
"author": "apoursh",
"title": "Prove the sharp Gaussian-dyadic four-qubit Clifford tensor width is 13 (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7799"
},
{
"pr": 7803,
"head": "0fefff647d869054f2e2c4eef3cdcb6e3c60c573",
"author": "apoursh",
"title": "Prove linear Gaussian-dyadic tensor widths for the full n-qubit Clifford alphabet",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7803"
},
{
"pr": 7806,
"head": "b42e665721e76e69d05f24a19bdabee0e0f77c04",
"author": "apoursh",
"title": "Prove the sharp five-qubit Clifford tensor threshold T(5) = 15",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7806"
}
]
}
Loading
Loading