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
72 changes: 72 additions & 0 deletions research/clifford-canonical-denominator/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,72 @@
# Canonical odd denominator at the first four-qubit width

Let $R=\mathbb Z[i,1/2]$. The odd denominator of a Gaussian-rational matrix $B$ is the least positive odd integer $d$ such that $dB$ has entries in $R$. Choose one Gaussian-dyadic unitary $U_a$ per scalar coset of the full four-qubit Clifford group, and put $N=12,128,668,876,800$. Let $A$ have the $N$ columns $\operatorname{vec}(U_a^{\otimes13})/4^{13}$. Write $K=A^*A$ and $J=K^{-1}A^*$.

**Theorem.** The odd denominator of $J$ is
$$
D=22{,}742{,}436{,}009{,}630{,}709{,}072{,}767{,}882{,}717{,}680{,}300{,}149{,}753{,}125.
$$
The least odd denominator among *all* norm-optimal left inverses is $470,624,175$ by [PR #7802](https://github.com/apoursh/integer-mult-bounds/blob/ebe2c019e398d73444e4f33f2d71bd5775795a01/research/clifford-dyadic-noise-floor/PROOF.md). The much larger $D$ measures the cost in odd factors of choosing the canonical operator.

## 1. Transfer to a central kernel

The [width-13 splitting](https://github.com/apoursh/integer-mult-bounds/blob/7f08a74037788311cc6b7da27297b25f1b188e45/research/clifford-bell-fibres/PROOF.md) supplies a matrix $L$ over $R$ with $LA=I$. For every positive odd integer $d$,
$$
dK^{-1}\in R^{N\times N}\quad\Longleftrightarrow\quad dJ\in R^{N\times 256^{13}} .
$$
The forward implication multiplies on the right by $A^*$ and the reverse implication multiplies on the right by $L^*$. This permits a calculation on the inverse Gram kernel. Changing the scalar lift of a column conjugates $K$ by a diagonal Gaussian unitary and multiplies the corresponding rows of $J$ by units, so the denominator is intrinsic to the projective alphabet.

Let $S=\mathrm{Sp}(8,2)$, the quotient by the Pauli subgroup, and let $V$ be the defining 16-dimensional Clifford module. In the scalar sector of $V^{\otimes13}$ the irreducibles are $V\otimes\rho$, one for every quotient irreducible $\rho$. Write $d_\rho=\dim\rho$ and $\lambda_\rho$ for the corresponding positive eigenvalue of $K$. The 81 rows, their multiplicities and their normalizations come from Ali Romman's threshold computation and Nick Petalas's projected character data. At width 13, $V^{\otimes13}=V\otimes W$ for an inflated $W$ with integer character $(-64)^k$ on symplectic elements having fixed-space dimension $k$. Hence the character of $W$ is rational, and Galois-conjugate quotient characters occur in $W$ with equal multiplicities. [Width-13 decomposition and spectrum](https://github.com/aliromman/integer-mult-bounds/blob/1398fbdca3536389d8ac3c8b3aaa408f03e8f6c2/research/clifford-threshold-conditioning/PROOF.md) · [Odd-sector character and phases](https://github.com/aliromman/integer-mult-bounds/blob/9a8ce06644700ffa576d99316450bcb52835d6dc/research/clifford-odd-width-spectrum/PROOF.md)

For $g=U_a^*U_b$ let $s$ be its symplectic image. Matrix-coefficient orthogonality gives the entries
$$
K^{-1}(a,b)=\frac{16\,\operatorname{tr}(g)}{N}
\sum_\rho \frac{d_\rho}{\lambda_\rho}\,\chi_\rho(s).
\tag{1}
$$
Indeed the projector for $V\otimes\rho$ has kernel $(16d_\rho/N)\operatorname{tr}(g)\chi_\rho(s)$. To multiply two such kernels one may average over all lifts in the finite Clifford group because the central phases contributed by the two factors at the intermediate index cancel. Schur orthogonality then gives idempotence within each block and zero products between distinct blocks. Their ranks $(16d_\rho)^2$ sum to $N$. In the Gram kernel $\operatorname{tr}(g)^{13}/16^{13}$ a multiplicity $m_\rho$ therefore gives $\lambda_\rho=N m_\rho/(16^{13}\cdot 16d_\rho)$. Replacing each block scalar by its reciprocal yields (1). The identity entry uses $\operatorname{tr}(I)=16$. All traces in (1) are evaluated with the displayed lift of $g$.

## 2. A common upper bound and the diagonal divisor

Let $T$ be the least common multiple of the reduced numerators of the 52 distinct $\lambda_\rho$, and set $b_\rho=T d_\rho/\lambda_\rho\in\mathbb Z$. Write $\operatorname{odd}(x)$ for the positive odd part of a positive integer. The odd part of $N$ is $722,925$. Exact arithmetic on the 81 source rows gives

| Quantity | Value |
|---|---:|
| $T$ | 2674007761273451977985641706958295138125 |
| $\gcd_\rho b_\rho$ | 91268055040 |
| $\operatorname{odd}(\gcd_\rho b_\rho)$ | 85 |
| Common odd denominator of $16b_\rho/(NT)$ | $8505T=D$ |

This last denominator is an upper bound for all entries in (1). Characters of finite groups are algebraic integers, including the trace of $g$. Multiplying by $D$ clears the odd denominators of the scalar coefficients. Multiplying further by a suitable power of two gives a sum of algebraic integers. Each entry already belongs to $\mathbb Q(i)$ by inversion of $K$, so after this multiplication it is in $\mathbb Z[i]$, the ring of integers of $\mathbb Q(i)$. This argument allows quotient characters with nonrational values.

The diagonal entry is independent of $a$ and equals
$$
K^{-1}(a,a)
=\frac{1}{N}\sum_\rho\frac{(16d_\rho)^2}{\lambda_\rho}
=\frac{8084515602799097617265541791880056776818688}{7580812003210236357589294239226766716584375}.
$$
Its reduced odd denominator is $D/3$. The eigenvalue numerators alone account for $T=D/8505$.

## 3. The order-17 entry

Use the literal operator $U$ of [PR #6208](https://github.com/apoursh/integer-mult-bounds/blob/4e3f55105e436dfa364029f9032c90f6a239fbe0/research/order-seventeen-clifford/PROOF.md). It has trace $-1$, order 17, and a nonidentity symplectic image $s$ of order 17. The binary space $E$ of dimension eight is irreducible for $s$: the order of 2 modulo 17 is eight, so every nontrivial 17th root already needs degree eight. Consequently the linear centralizer of $s$ consists of multiplications by nonzero elements of $\mathbb F_{256}=\mathbb F_2[s]$. Adjoint for the symplectic form acts on that field as the sixteenth-power map, since it sends $s$ to $s^{-1}=s^{16}$. Multiplication by $u$ preserves the form exactly when $u^{17}=1$. Thus $|C_S(s)|=17$. The independent checker also enumerates all 256 linear maps commuting with the explicit binary action and finds exactly 17 symplectic ones.

Group the 81 quotient characters by their pair $(d_\rho,\lambda_\rho)$. Equal degree and multiplicity give equal eigenvalue, and conversely equal degree and eigenvalue give equal multiplicity. Section 1 shows that each such packet $B$ is Galois-stable; it may contain several orbits. There are 79 singletons and one packet of size two. The packet character $\psi_B=\sum_{\rho\in B}\chi_\rho$ takes integer values because it is both rational and algebraic integral.

Put $m=|B|$ and let $d$ be its common degree. On the cyclic subgroup generated by $s$, the rational character has equal multiplicities for all 16 nontrivial 17th roots. Indeed Galois automorphisms permute these root characters transitively and fix the restricted character. If the common multiplicity is $h$ and the trivial multiplicity is $h_0$, then $md=h_0+16h$ and $\psi_B(s)=h_0-h$. Combining this congruence with column orthogonality gives
$$
\psi_B(s)\equiv md\pmod {17},\qquad
|\psi_B(s)|^2\le m\sum_{\rho\in B}|\chi_\rho(s)|^2
\le 17m .
$$
Here $m\le2$, so the interval has length at most $2\sqrt{34}<17$. Each packet value is the unique integer in this interval with the required residue. Among the 80 packet values, five are $+1$, four are $-1$ and 71 are zero. Substitution in (1) with $U_a=I$, $U_b=U$ gives
$$
K^{-1}(I,U)=\frac{-309891064832}{22949540428586199553125}.
$$
This fraction is reduced. Its denominator has 3-adic valuation 12, one greater than that of the diagonal denominator. Their least common multiple is $D$. Together with Section 2 this proves the theorem. Rephasing either representative multiplies the off-diagonal entry by a fourth root of unity and preserves this divisor.

## 4. Independent local check

The second checker reconstructs all $\lambda_\rho$ using the fixed-space projection in [PR #7747](https://github.com/petalas/integer-mult-bounds/blob/ff12ca358d601a21cd4b1f016b8a3a969ac44016/research/clifford-four-qubit-width-nine/PROJECTION.json) and the integer character of $W$. It obtains $b_\rho\equiv0\bmod 3$ except in the two rows of degrees 35 and 238, where both residues equal one. Both degrees occur once, so their characters are rational. Their values at $s$ are $1$ and $0$, respectively, by the congruence and bound above. In the algebraic-integer ring containing all character values, the numerator of $NTK^{-1}(I,U)$ is therefore congruent to $-16\equiv2\bmod 3$. Thus it is prime to 3 at the rational level, while $NT$ contains $3^{12}$. This verifies the missing prime power without using the reported spectrum or the other 78 packet values at $s$.

The construction classifies common odd denominators of entries. An implementation may factor or share divisions differently. The result gives no operation count, fixed-tape algorithm, or improvement of the selected multiplication exponent.
31 changes: 31 additions & 0 deletions research/clifford-canonical-denominator/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
# Exact canonical denominator for the four-qubit Clifford decoder

At tensor width 13, the least positive odd integer that clears all odd denominators of the canonical Moore–Penrose decoder is **22,742,436,009,630,709,072,767,882,717,680,300,149,753,125**. The coefficient ring is $\mathbb Z[i,1/2]$, and the columns are the normalized tensor powers of one Gaussian-dyadic unitary representative of every projective four-qubit Clifford. The sharp odd denominator among *all* optimal-norm decoders is only 470,624,175. These two values answer different arithmetic questions about the same observations. [Proof](PROOF.md) · [Optimal-norm denominator and handoff, #7802](https://github.com/apoursh/integer-mult-bounds/blob/ebe2c019e398d73444e4f33f2d71bd5775795a01/research/clifford-dyadic-noise-floor/README.md)

The dyadic splitting identifies the canonical decoder's denominator with that of its inverse Gram matrix. A character expansion bounds every entry. Two explicit inverse-Gram entries then certify the full bound.

| Relative operator | Reduced odd denominator |
|---|---:|
| Identity (diagonal entry) | 7580812003210236357589294239226766716584375 |
| The order-17 operator of #6208 | 22949540428586199553125 |

The least common multiple of these denominators is the stated 44-digit answer. The order-17 entry supplies the additional factor of three missing from the diagonal. Its computation needs only representation degrees, multiplicities and the symplectic centralizer, without expanding the twelve-trillion-column Gram matrix. This is a coefficient-denominator result. A multiplication supplier and a price for physical divisions remain separate obligations. [Order-17 circuit and fixed scalar phase](https://github.com/apoursh/integer-mult-bounds/blob/4e3f55105e436dfa364029f9032c90f6a239fbe0/research/order-seventeen-clifford/PROOF.md)

## Reproduce

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

~~~sh
python3.12 -B scripts/research_package.py --package research/clifford-canonical-denominator
~~~

The primary checker reads the 81 rows of the exact width-13 spectrum, groups them into 80 rational character packets, reconstructs the order-17 column values and checks both reduced fractions. The independent checker generates the operator's phases from its gates, reconstructs its eight binary generator images, enumerates its full 17-element symplectic centralizer, and checks the missing $3^{12}$ directly using the independent fixed-space projection file. All 52 spectral levels contribute to the upper bound. The diagonal-only and eigenvalues-only denominators both fall short. [Numerical sources, original Git blobs and SHA-256 hashes](SOURCE-PINS.json)

## Credit and scope

Ali Romman supplied the threshold spectrum, odd-sector normalization, and earlier canonical odd-division obstruction. Nick Petalas supplied the quotient character projection from the ATLAS/GAP Character Table Library of Thomas Breuer. The Clifford conventions retain alephnan's Pauli-sector work. The sharp width-13 splitting, the noncanonical norm-optimum distinction, and the explicit order-17 representative are earlier contributions of apoursh. Classical character orthogonality and cyclotomic restriction are used throughout. This contribution is the matching universal bound and concrete two-entry certificate for the canonical decoder. Prepared by **apoursh** with substantial OpenAI Codex assistance. Apache-2.0. [Canonical obstruction, #7792](https://github.com/aliromman/integer-mult-bounds/blob/9321c67e29c3c41a188154bc5c2add7d7802b694/research/clifford-odd-division-obstruction/PROOF.md) · [Width-13 splitting, #7799](https://github.com/apoursh/integer-mult-bounds/blob/7f08a74037788311cc6b7da27297b25f1b188e45/research/clifford-bell-fibres/PROOF.md) · [Research claim](https://github.com/CrocSwap/integer-mult-bounds/discussions/1166#discussioncomment-19754792)

## Open next questions

1. At which larger odd widths do the same two inverse-Gram entries certify the canonical denominator? Start at widths 15 and 17 using the eigenvalue polynomials from #7788 and the appropriate defining or conjugate trace. Additional primes may require another symplectic class.
2. Can an exact implementation realize the denominator with a short common division schedule? Factor the proven denominator into its prime powers and bind the proposed schedule to an actual coefficient-access layout. The two-entry certificate establishes arithmetic necessity and leaves storage, movement and operation counts open.
28 changes: 28 additions & 0 deletions research/clifford-canonical-denominator/SOURCE-PINS.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
{
"files": [
{
"pr": 7747,
"author": "petalas",
"head": "ff12ca358d601a21cd4b1f016b8a3a969ac44016",
"path": "research/clifford-four-qubit-width-nine/PROJECTION.json",
"url": "https://github.com/petalas/integer-mult-bounds/blob/ff12ca358d601a21cd4b1f016b8a3a969ac44016/research/clifford-four-qubit-width-nine/PROJECTION.json",
"use": "Independent recovery of the 81 width-13 eigenvalues from fixed-space character projections.",
"bytes": 7054,
"sha256": "0a36754668b66480e5f968c117a4532bc3a5eaab5c9d8448f14a5d0321c2efae",
"git_blob": "406f84b46901df771050eac3f367413cc6529ba1",
"local_path": "sources/sp8-projection.json"
},
{
"pr": 7772,
"author": "aliromman",
"head": "1398fbdca3536389d8ac3c8b3aaa408f03e8f6c2",
"path": "research/clifford-threshold-conditioning/RESULT.json",
"url": "https://github.com/aliromman/integer-mult-bounds/blob/1398fbdca3536389d8ac3c8b3aaa408f03e8f6c2/research/clifford-threshold-conditioning/RESULT.json",
"use": "Primary full-alphabet width-13 spectral blocks.",
"bytes": 28857,
"sha256": "d032d5e63820f4ffe246d21adcd8d272f04dfd635a6fc519c4a81a366ca41d49",
"git_blob": "039f22ba0bfec63c02fb3a97959766b8ec3a1bfa",
"local_path": "sources/width13-spectrum.json"
}
]
}
69 changes: 69 additions & 0 deletions research/clifford-canonical-denominator/SOURCES.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,69 @@
{
"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": 6208,
"head": "4e3f55105e436dfa364029f9032c90f6a239fbe0",
"author": "apoursh",
"title": "Construct order-17 Gaussian-dyadic Clifford attaining native rank eight (no κ)",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/6208"
},
{
"pr": 7747,
"head": "ff12ca358d601a21cd4b1f016b8a3a969ac44016",
"author": "petalas",
"title": "Prove the exact four-qubit Clifford Gram rank at width nine",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7747"
},
{
"pr": 7772,
"head": "1398fbdca3536389d8ac3c8b3aaa408f03e8f6c2",
"author": "aliromman",
"title": "Determine sharp four-qubit Clifford threshold conditioning",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7772"
},
{
"pr": 7788,
"head": "9a8ce06644700ffa576d99316450bcb52835d6dc",
"author": "aliromman",
"title": "Prove exact Clifford Gram endpoints at every odd width",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7788"
},
{
"pr": 7792,
"head": "9321c67e29c3c41a188154bc5c2add7d7802b694",
"author": "aliromman",
"title": "Prove odd-division obstruction for canonical Clifford decoding",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7792"
},
{
"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": 7802,
"head": "ebe2c019e398d73444e4f33f2d71bd5775795a01",
"author": "apoursh",
"title": "Determine the sharp odd denominator for noise-optimal four-qubit Clifford decoding",
"url": "https://github.com/CrocSwap/integer-mult-bounds/pull/7802"
}
]
}
Loading
Loading