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
78 changes: 78 additions & 0 deletions research/clifford-general-width/PROOF.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,78 @@
# A linear sufficient width for dyadic Clifford tensor interpolation

Let $R=\mathbb Z[i,1/2]$, $d=2^n$, and choose a Gaussian-dyadic unitary matrix $U_a$ in each projective class of the full $n$-qubit Clifford group modulo scalar powers of $i$. The raw observation matrix $C_f$ has columns $\operatorname{vec}(U_a^{\otimes f})$. Reordering its coordinates produces $\operatorname{vec}(U_a)^{\otimes f}$. All inputs and outputs below are complex-linear, including when the mixture coefficients are arbitrary complex numbers.

**Theorem.** There is a specified matrix $B_f$ over $R$ satisfying $B_f C_f=I$ for every $f\ge F(n)$, where

| Qubits | Sufficient width $F(n)$ |
|---|---:|
| $n=1,2$ | 5, 9 respectively |
| odd $n\ge3$ | $3n+2$ |
| even $n\ge4$ | $3n+1$ |

For the normalized columns $d^{-f/2}\operatorname{vec}(U_a^{\otimes f})$, multiply $B_f$ by $d^{f/2}$. This gives an $R$-left inverse whenever $nf$ is even. For odd $nf$ the normalization uses a square root of two, and the same formula is over $R[\sqrt2]$. In that case every entry of the normalized map lies in $\sqrt2\,\mathbb Q(i)$. Multiplying by an $R$-matrix keeps every entry in this same rational subspace, which excludes $1$ because $\sqrt2\notin\mathbb Q(i)$. Thus the raw and normalized ring conclusions genuinely differ. Every specialization of the raw identity to a residue field of odd characteristic still splits. Optimality for $n\ge5$ is open here.

The Bell-fibre decomposition and four-dimensional character corners are due to apoursh in [#7799, Sections 1–3](https://github.com/apoursh/integer-mult-bounds/blob/7f08a74037788311cc6b7da27297b25f1b188e45/research/clifford-bell-fibres/PROOF.md). The argument below includes both pair parities and constructs pure and mixed subspaces in arbitrary dimension. [#6714, three-qubit framework](https://github.com/alephnan/integer-mult-bounds/blob/f7f32a18dde54e67624a9ed9cc7e1b266d9585c0/research/clifford-three-qubit-threshold/PROOF.md) · [#7766, four-qubit spectrum](https://github.com/aliromman/integer-mult-bounds/blob/d7c2dd1868c70f0addce2c0a4d3220149452f14b/research/clifford-four-qubit-threshold/PROOF.md)

## 1. The two Bell contractions

Write $V=X\oplus L=\mathbb F_2^n\oplus\mathbb F_2^n$, $W_{x,z}=X^xZ^z$, and $q(x,z)=x\cdot z$. The symplectic pairing is the polar form of $q$, written $[u,v]$. Let $G=\mathrm{Sp}(V)$. Conjugation and congruence respectively act by

```text
U W_t U* = γ_U(t) W_(St)
U W_t Uᵀ = θ_U(t) W_(L_S t), L_S t = St+b_S .
```

Both phases are fourth roots of unity. The first identity is the Clifford definition. Its map to $G$ is onto since symplectic transvections lift by $(1+i)(I+iP)/2$ for Hermitian Paulis $P$. For an element fixing all labels the conjugation signs form a character. Removing the corresponding Weyl leaves a scalar by Pauli irreducibility. A scalar in $R$ of complex modulus one is a power of $i$, by descent in $a^2+b^2=4^j$. The projective kernel is therefore exactly the Weyls. Since the Weyls are real, $U,\bar U$ have the same linear label action. Thus $U U^T$ is a Weyl up to phase, which gives the affine label action in the second identity. Applying transpose gives $q(L_S t)=q(t)$.

Use $k$ ordered Bell pairs and one singleton, $f_0=2k+1$. For input labels $t=(t_1,\ldots,t_k)$, an output $t'$ is hit by column $U$ exactly when $t'_j=L_S t_j$ for each $j$. The block obtained from that column is then $β U$, with $β=\prod_j\theta_U(t_j)$. Indeed input and output Bell vectors are the normalized vectorizations of the indicated Weyls. **Over $R$** each complete input/output Bell matrix element has coefficient $1/d$ times products of Weyl entries, since the two square roots cancel within a pair. This applies also to odd $n$. Write $w=\sum_j t_j$, $w'=\sum_j t'_j$ as binary sums.

| Pair count | Output sum | Fixed correction to a hit block $M$ | Result on $M=β U$ |
|---|---|---|---|
| $k$ even | $w'=Sw$ | $W_{w'} M W_w^{-1}$ | $σ U$, with $σ=β/\gamma_U(w)$ |
| $k$ odd | $w'=L_S w$ | $W_{w'}^{-1}M W_w$ | $σ\bar U$, with $σ=β\theta_U(w)$ |

For the odd row use $U W_w U^T=\theta_U(w)W_{w'}$ and $(U^T)^{-1}=\bar U$. In particular the fixed linear operation on an observation produces $σ\bar U$ on a column and $cσ\bar U$ on that column multiplied by $c$. It never conjugates $c$.

For either row of the table $σ=σ_S(t)$ is independent of the Pauli and scalar lift. Left multiplication of $U$ by a real Weyl multiplies $β$ by $(-1)^{[p,w']}$. The corresponding singleton phase in either row receives exactly that sign as well. In the even case a scalar $zU$ gives the extra factor $z^{2k}=1$. In the odd case the extra factor is $z^{2k+2}=1$. The phases of an ordinary or congruence action compose on their respective labels, so $σ_{S_1S_2}(t)=σ_{S_1}(L_{S_2}t)σ_{S_2}(t)$. This is where both parity and the direction of the right and left corrections enter.

## 2. A seed whose point stabilizer has dyadic order

Let $s=\lfloor n/2\rfloor+1$ except that $s=1$ at $n=2$. In the standard symplectic basis use the seed
```text
T = (0,Z_1,...,Z_n,X_(s+1),...,X_n),
k = 2n-s+1, f_0 = 4n-2s+3.
```

Fixing the seed’s zero label forces $b_S=0$. Fixing $L$ pointwise forces $S$ to be the shear $(x,z)\mapsto(x,z+Bx)$, $B=B^T$. Preserving $q$ forces $B$ to have zero diagonal, and the displayed $X$ labels force its last $n-s$ columns and rows to vanish. Conversely these shears are all products of real controlled-$Z_{ab}$ with $a<b\le s$. Their lifts commute with each Weyl in the seed and $W_w$ (the latter is a signed product of seed Weyls), and have $UU^T=I$. They have zero affine offset and trivial $σ$ on the seed for either parity. This identifies the stabilizer $C$ of order $2^{\binom s2}$.

The orbit of $T$ is $G/C$. For a tuple $t_i$ in that orbit choose a transporter $g_i$ and set $j_i=σ_{g_i}(T)$. Triviality on $C$ makes $j_i$ independent of the choice. For $L_S t_i=t_j$ the phase $σ_S(t_i)=j_j/j_i$ is known. Multiplying the corrected observation block by $j_i/j_j$ yields the sum of
```text
M_S = sum_(a over S) c_a U_a (k even),
M_S = sum_(a over S) c_a conjugate(U_a) (k odd)
```
over $S\in g_j C g_i^{-1}$. Every left-right translate of $C$ occurs this way. Writing $e_C=\sum_{c\in C}c/|C|$ in $R[G]$, it remains to produce $1\in R[G]e_C R[G]$. Then every single-fibre indicator follows by left translation. The averaging denominator is a power of two.

## 3. Character corners in arbitrary dimension

Take $A=\operatorname{Sym}_n(\mathbb F_2)$, the elementary abelian group of all shears fixing $L$ pointwise, allowing nonzero diagonal. Its signs are indexed by binary quadratic forms $Q$ on $L$, with polar form $λ$. Specifically $χ_Q(uu^T)=(-1)^{Q(u)}$ and $χ_Q(uv^T+vu^T)=(-1)^{λ(u,v)}$. The diagonal and cross terms provide exactly $n(n+1)/2$ independent bits. The idempotents $e_Q=|A|^{-1}\sum_{a\in A}χ_Q(a)a$ are orthogonal and sum to $1$. A linear change of variables in $L$, extended contragrediently to $X$, normalizes $A$ and transports these idempotents.

For an $s$-space $W\le V$ isotropic for the symplectic pairing $[\,,\,]$, $C_W$ generated by all pair shears on basis vectors of $W$ is conjugate to $C$. The pair shear at $u,v$ sends $t\mapsto t+[t,u]v+[t,v]u$. Extend an isotropic basis to a symplectic basis to see the conjugation. Thus every $e_{C_W}$ belongs to the ideal in question.

**Pure planes.** If $λ$ has rank $2h$, it admits a totally isotropic subspace of dimension $n-h$, by adjoining the radical to one side of a symplectic basis of the quotient. Whenever $n-h\ge s$, choose an $s$-dimensional subspace $W\le L$ of this kind. Then $C_W\le A$, $χ_Q|_{C_W}=1$ and $e_Qe_{C_W}e_Q=e_Q$. This covers all odd $n$, the exception $n=2$, and every degenerate polar form at even $n$.

**Mixed planes.** The remaining case is $n=2m\ge4$, $s=m+1$, $λ$ nondegenerate. Pick $r\ne0$ with $Q(r)=0$. On any $λ$-isotropic two-plane $Q$ is linear and hence has a nonzero zero. Put $H=r^{\perp_λ}\subset L$, whose radical under $λ|_H$ is precisely $\langle r\rangle$. Extend $r$ to a maximal $λ$-isotropic $m$-space $P\le H$ using $H/\langle r\rangle$. Choose $z\in L\setminus H$ and $x\in X$ dual to $z$ and annihilating $H$ in the original pairing. Take $W=P\oplus\langle x\rangle$, which is symplectically isotropic in $V$.

The pure pairs in $P$ generate a subgroup $K\le A$ of order $2^{\binom m2}$ on which $χ_Q=1$. The remaining pair shears, involving $x$, are precisely the $2^m$ transformations $T_p$ with $p\in P$ sending $z\mapsto z+p$ and fixing $H$ pointwise on $L$. On $X$ they act contragrediently. They normalize $A$. The explicit pair formula shows these commute, and together with $K$ they give the unique products $aT_p\in C_W$. A nonzero $p\in P$ distinct from $r$ lies outside the radical of $λ|_H$, so choose $y\in H$ with $λ(p,y)=1$. Then $λ(z,y)$ changes under $T_p$, hence $Q$ changes. At $p=r$, $Q(z+r)=Q(z)+Q(r)+λ(z,r)=Q(z)+1$. Thus exactly $K$ fixes the character in $C_W$. For $p\ne0$, $e_Q aT_p e_Q=0$ by orthogonality of distinct character idempotents. For $a\in K$, $e_Q a e_Q=e_Q$. Consequently
```text
e_Q e_(C_W) e_Q = 2^(-m) e_Q .
```
Multiplying by $2^m$ recovers $e_Q$. At $n=2$ an anisotropic quadratic form can have no nonzero zero, explaining the separate choice of $s$.

Summing these explicit pure or rescaled mixed corners gives the identity in the ideal. Expanding the idempotents expresses it using the measured coset sums and coefficients in $R$. Enumeration of finite groups, subspaces and transporters suffices to specify a particular decoder.

## 4. Individual coefficients and larger widths

For the recovered matrix $M_S$, put $V_a=U_a$ for even $k$, $V_a=\bar U_a$ for odd $k$. Return $\operatorname{Tr}(V_a^*M_S)/d$. Distinct classes in the same fibre differ by a nonidentity Weyl up to scalar, hence have zero trace pairing, and the target pairs to $d$. This proves the raw left-inverse identity at $F(n)$. At larger width, tensor each target row with one copy of $\operatorname{vec}(U_a)^*/d$ per extra slot. The extra factor evaluates to one on its target. Its values on other columns are immaterial because the original row vanishes there. Each algebraic operation is over $R$.

No norm or running-time bound follows from just counting the Bell pairs. Even the character expansion has $2^{n(n+1)/2}$ terms before combining coefficients. A physical interface needs a concrete producer, precision, storage, movement and a full restoration bill.
42 changes: 42 additions & 0 deletions research/clifford-general-width/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
# Dyadic Clifford tensor decoders at linear width for every qubit count

For the full projective Gaussian-dyadic Clifford alphabet on $n$ qubits, the raw tensor observations determine every mixture coefficient by a linear function over $\mathbb Z[i,1/2]$ at all widths $f\ge F(n)$. At odd $n\ge3$ it suffices to take $F(n)=3n+2$. At even $n\ge4$ it suffices to take $F(n)=3n+1$. The construction also gives widths 5 and 9 on one and two qubits. At odd $nf$ the raw theorem covers the Gaussian-dyadic map. Unit-length normalization introduces a square root of two. [Proof and exact input convention](PROOF.md)

| Qubits $n$ | 3 | 4 | 5 | 6 | 7 | 8 |
|---|---:|---:|---:|---:|---:|---:|
| Sufficient dyadic width $F(n)$ | 11 | 13 | **17** | **19** | **23** | **25** |
| Earlier general complex width $4n+4$ | 16 | 20 | 24 | 28 | 32 | 36 |
| Earlier determinant-based dyadic width $4n2^n$ | 96 | 256 | 640 | 1,536 | 3,584 | 8,192 |

The middle row is Ali Romman's all-$n$ complex independence and conditioning result. The present proof gives no uniform numerical spectral enclosure at $F(n)$. The bottom row is apoursh's all-$n$ interpolation theorem. At fixed sizes, the known sharp complex thresholds are 5, 9, 10 and 13 for $n=1,2,3,4$ respectively. The four-qubit dyadic threshold 13 was established in #7799. The width-10 dyadic question at three qubits remains outside the present argument. [#5526, uniform complex bound](https://github.com/aliromman/integer-mult-bounds/blob/45f7de73b168721cdf7ec8c1769b1b594493716a/research/clifford-pointwise-faithfulness/PROOF.md) · [#7797, determinant interpolant](https://github.com/apoursh/integer-mult-bounds/blob/9d3237a6308228145f1c26dc349f18d5c857bb41/research/clifford-dyadic-left-inverse/PROOF.md) · [#6714, three-qubit proof and small-size lineage](https://github.com/alephnan/integer-mult-bounds/blob/f7f32a18dde54e67624a9ed9cc7e1b266d9585c0/research/clifford-three-qubit-threshold/PROOF.md) · [#7799, four-qubit splitting](https://github.com/apoursh/integer-mult-bounds/blob/7f08a74037788311cc6b7da27297b25f1b188e45/research/clifford-bell-fibres/PROOF.md)

There was already a route to linear dyadic width using the eight-qubit stabilizer argument of #7798. With $2n$ binary Choi coordinates, use $\lceil n/2\rceil$ disjoint copies of its dimension-four length-eight doubly-even witness code. The first $2n$ rows still cancel all quadratic phases. The same origin factor and subspace Möbius sum give width $8\lceil n/2\rceil+1$. For raw columns, rephase each Hadamard in the target's Bell transform with $(1+i)/2$ multiplying the unnormalized binary Hadamard. The target amplitude is then $(1+i)^n$, a unit of the dyadic ring, so the separator can be rescaled over that ring. This comparison extends the code argument in #7798, which established sharpness for all eight-qubit stabilizer rays. The Clifford-specific argument here uses fewer copies for every $n\ge3$. [#7798, Sections 1–3 and 5](https://github.com/apoursh/integer-mult-bounds/blob/0ae10f215780296ca50142fd36fdf904b8c1b146/research/clifford-width17/PROOF.md)

Nick Petalas's general complex lower bound $T(n)\ge2n+1$ for $n\ge2$ also bounds the dyadic threshold below. Several residue classes have stronger known obstructions, including $2n+3$ when positive $n$ is divisible by four. At $n=5$ the published lower bound is 13, leaving the dyadic threshold in $[13,17]$. [#5967, lower bounds and their lineage](https://github.com/petalas/integer-mult-bounds/blob/49404527570c90f4a19708004c81533398b5e7b5/research/clifford-threshold-linear-lower/README.md) · [#7766, Dickson obstruction](https://github.com/aliromman/integer-mult-bounds/blob/d7c2dd1868c70f0addce2c0a4d3220149452f14b/research/clifford-four-qubit-threshold/PROOF.md)

## Construction

The decoder uses $2n-s+1$ ordered Bell pairs and a singleton, where $s=\lfloor n/2\rfloor+1$ for $n\ge3$. An even pair count exposes a fibre sum of the representative matrices. An odd pair count exposes a fibre sum of their entrywise conjugates with the *original* mixture coefficients. This second case is needed, for example, at six qubits. A maximal isotropic subspace for a binary alternating form supplies most of the character projections. A mixed subspace supplies the missing projections for full-rank forms at every even qubit count from four upwards. The resulting identity over the ring recovers the fibre matrices from sums over subgroups of power-of-two order.

## Reproduce

Use CPython 3.12 and its standard library from the repository root.

```sh
python3.12 -B scripts/research_package.py --package research/clifford-general-width
```

- `verify_planes.py` covers all 33,866 quadratic characters at sizes $n=1$ through 5. Pure-subspace witnesses cover all linear parts of a given polar form together. Mixed witnesses are checked by actually translating the values of each quadratic form. The script also checks all 18 canonical alternating ranks at $n=6,7,8,9$, including all linear parts at maximal even rank. The combined checks include 768 mixed corners, 5,632 nonidentity character moves and two failure controls.
- `verify_twist.py` constructs all 96 one-qubit lifts by exact rational Gaussian matrix multiplication. It checks 32,640 ordered Bell tuples, including 6,528 with odd pair count, and 12,240 cocycle identities. On a mixture of all 24 projective classes with complex coefficients and individually rephased representatives, conjugating the coefficients corrupts all six tested blocks.
- `verify_blocks.py` constructs actual five- and six-qubit monomial Clifford matrices with arbitrary nonreal scalar lifts. It computes Bell matrix elements directly on a mixture spanning every element of a planted stabilizer coset with four chosen Pauli lifts per element, plus a disjoint competing coset. At $n=5$ it includes 32 and excludes 32 columns. At $n=6$ it includes 256 and excludes 256 columns. The resulting nonzero corrected matrices agree entry by entry with the planted ordinary- or conjugate-action fibre sums.

The written proof supplies the general dimension and full-alphabet quantifiers. The finite checks concern the algebraic decoder. A conditioning estimate at the new widths and a physical decoder still require separate arguments. This conclusion has no effect on the selected exponent.

## Sources and attribution

Ali Romman supplied the pointwise bound and four-qubit spectrum. Nick Petalas supplied the project-wide lower-bound comparison. MatShep supplied the sharp one- and two-qubit computations, and alephnan supplied the sharp three-qubit theorem and scalar-sector framework. The code-witness technique credits Ali Romman and Nick Petalas. The determinant, stabilizer and four-qubit Bell-fibre predecessors are due to apoursh and retain their cited lineage. The new steps here are the parity-dependent singleton correction and the general-dimensional subgroup average used for the width bound. The symplectic and character-idempotent arguments use classical linear algebra. Prepared by **apoursh** with substantial OpenAI Codex and separate agent assistance. Apache-2.0. Publication is independent of maintainer scientific review.

## Open next questions

1. At five qubits, determine whether one can decode below width 17. Width 15 permits seven Bell pairs and one singleton. For the same seed formula this leaves an alternating subgroup on four coordinates. Start by testing whether its subgroup average generates the unit ideal, or exhibit a representation in which it vanishes. An obstruction to this one orbit would still leave other decoders available.
2. For the normalized six-qubit source at width 19, bound a specified decoder's noise and evaluation costs. Start with the actual coset support after combining the mixed corners, and identify an input producer for those blocks. The present ideal identity alone supplies neither a Gram enclosure at 19 nor a physical read schedule.
Loading
Loading