Skip to content
OpenContext
Contribution toProblem 5: Test problem: Hilbert–Schmidt inradius of the multiqubit stabilizer polytopeunder revision
0

Partial resultOpenContext 5.5

A weaker bound on the stabilizer-inradius conjecture

by Rafael Wagner · accepted 18 Nov 2026 · partially AI-assisted

Result

Using Claude Fable 5.2 and a few prompts, it was possible to prove the following weaker statement:

Proposition: Let ρ∈D(C2⊗⋯⊗C2⏟n times)\rho \in \mathcal{D}(\underbrace{\mathbb{C}^2 \otimes \cdots \otimes \mathbb{C}^2}_{n\text{ times}}) be an arbitrary nn-qubit state, with n≥2n\geq 2. Denote, furthermore, dn=2nd_n = 2^n and the function c(⋅):N→Qc_{(\cdot)}:\mathbb{N}\to\mathbb{Q} given by

cn=2dn+77dn+2>27.c_n=\frac{2d_n+7}{7d_n+2}>\frac{2}{7}.

Then,

Tr[ρ2]≤1dn−cn  ⟹  ρ∈STABn.\mathrm{Tr}[\rho^2] \leq \frac{1}{d_n-c_n} \implies \rho \in \mathrm{STAB}_n.

In particular, for all n≥2n \geq 2 it holds that

Tr[ρ2]≤1dn−27  ⟹  ρ∈STABn.\mathrm{Tr}[\rho^2] \leq \frac{1}{d_n-\frac{2}{7}} \implies \rho \in \mathrm{STAB}_n.

Note that this is only a partial result because the conjecture claims that the implication holds with cn=12c_n=\frac{1}{2} for arbitrary nn.

Background

The key technical idea is to use the fact that one can investigate the easier properties associated with the convex cones generated by the set of states and the stabilizer polytopes, as opposed to the sets themselves. In this case, one replaces rotations and the more complicated geometry of the inradius by properties of cone-preservation and related ideas.

Isotropic Pauli subgroups

Definition (isotropic Pauli subgroup). Let Pn\mathcal{P}_n be the set of 4n4^n Pauli strings, i.e. tensor products of Pauli matrices I,X,Y,Z\mathbb{I}, X, Y, Z on nn qubits without phases. A subset T⊆PnT \subseteq \mathcal{P}_n is an isotropic Pauli subgroup if

  • (1) it contains the identity, i.e., if I∈T\mathbb{I} \in T;
  • (2) it is Abelian, i.e. any two elements of TT commute;
  • (3) it is closed under multiplication up to a sign, i.e., ∀R,R′∈T\forall R,R' \in T, one has that either RR′∈TRR' \in T or −RR′∈T-RR' \in T.

Such a subgroup has ∣T∣=2k\vert T \vert = 2^k elements for some 0≤k≤n0 \leq k \leq n. A sign pattern τ\tau assigns an eigenvalue τ(R)=±1\tau(R) = \pm 1 to each R∈TR \in T consistently with multiplication (so, for example, τ(I)=1\tau(\mathbb{I})=1 and τ(R)τ(R′)=±τ(R′′)\tau(R)\tau(R') = \pm \tau(R'') if RR′=±R′′RR'= \pm R'', with the same sign on both sides). For any fixed TT there are finitely many sign patterns, namely 2k2^k of them. The operator

Πτ=1∣T∣∑R∈Tτ(R)R\Pi_\tau = \frac{1}{\vert T\vert} \sum_{R \in T}\tau(R)R

is the projector onto the corresponding joint eigenspace, which has dimension 2n−k2^{n-k}. The centraliser C(T)C(T) is the set of Pauli strings commuting with every element of TT.

Stabilizer codes

These constructions are very common in quantum error correction (QEC). An important family of quantum codes is that of stabilizer codes. Each pair (T,τ)(T,\tau) defines a stabilizer group, and Πτ\Pi_\tau projects onto its stabilizer code.

Definition (stabilizer code space). Given an isotropic Pauli subgroup TT with ∣T∣=2k\vert T \vert = 2^k and a sign pattern τ\tau, the stabilizer code space is the joint eigenspace given by

CT,τ={∣ψ⟩ : R∣ψ⟩=τ(R)∣ψ⟩ for all R∈T}=range[Πτ].\mathcal{C}_{T,\tau} = \left\{ \vert \psi \rangle \ : \ R\vert \psi \rangle =\tau(R)\vert \psi \rangle \text{ for all }R \in T \right\} = \mathrm{range}[\Pi_\tau].

Mathematical preliminaries

Lemma: Let TT be an isotropic Pauli subgroup with ∣T∣=2k\vert T\vert = 2^k, and Πτ\Pi_\tau the projector onto its joint eigenspace with sign pattern τ\tau. If AA is non-negative on all nn-qubit stabilizer states, i.e., AA is a Hermitian operator on nn qubits with

⟨s∣A∣s⟩≥0\langle s|A|s\rangle \ge 0

for every nn-qubit stabilizer state ∣s⟩|s\rangle, then there is a Clifford unitary UU and an operator BB on n−kn-k qubits such that

  • (a) U ΠτA Πτ U†=∣0k⟩⟨0k∣⊗BU\,\Pi_\tau A\,\Pi_\tau\,U^\dagger = |0^k\rangle\langle 0^k|\otimes B;
  • (b) ⟨t∣B∣t⟩≥0\langle t|B|t\rangle \ge 0 for every (n−k)(n-k)-qubit stabilizer state ∣t⟩|t\rangle.

In other words, the block of AA on the code space is, up to a Clifford change of basis, an (n−k)(n-k)-qubit operator BB that satisfies the same kind of hypothesis as AA. We can also gain some intuition for why such a result might be useful. From (a), we have that

Tr[B]=Tr[ΠτA]\mathrm{Tr}[B] = \mathrm{Tr}[\Pi_\tau A]

and

Tr[B2]=Tr[(ΠτAΠτ)2].\mathrm{Tr}[B^2] = \mathrm{Tr}[(\Pi_\tau A\Pi_\tau)^2].

From (b), any inequality Tr[B2]≤M (Tr[B])2\mathrm{Tr}[B^2] \le M\,(\mathrm{Tr}[B])^2 known for (n−k)(n-k)-qubit operators satisfying the positivity constraint holds for BB. Together they imply that

Tr[(ΠτAΠτ)2]≤M (Tr[ΠτA])2.\mathrm{Tr}[(\Pi_\tau A\Pi_\tau)^2] \le M\,\big(\mathrm{Tr}[\Pi_\tau A]\big)^2.

This is the way the lemma enters the final theorem.

Proof. We want to show that there is a Clifford unitary UU and an operator BB on n−kn-k qubits such that

UΠτAΠτU†=∣0k⟩⟨0k∣⊗B,U \Pi_\tau A \Pi_\tau U^\dagger = \vert 0^k \rangle \langle 0^k \vert \otimes B,

where ⟨t∣B∣t⟩≥0\langle t \vert B \vert t \rangle \geq 0 for every (n−k)(n-k)-qubit stabilizer state ∣t⟩\vert t \rangle. The argument has four steps. First, we move the code space to a standard position. Pick independent generators R1,…,RkR_1,\ldots, R_k of TT and set gi=τ(Ri)Ri. g_i=\tau(R_i)R_i. These are kk independent, commuting, Hermitian Pauli operators, and they allow us to rewrite Πτ\Pi_\tau as

Πτ=∏i=1k12(I+gi).\Pi_\tau=\prod_{i=1}^k\tfrac12(\mathbb{I}+g_i).

It is a standard fact about the Clifford group that any such list can be mapped to Z1,…,ZkZ_1,\ldots,Z_k: there is a Clifford unitary UU with

UgiU†=ZiU g_i U^\dagger = Z_i

for all ii. (In more mathematical terms, Cliffords realise every symplectic transformation of the Pauli labels together with every choice of signs, and any kk independent commuting Paulis can be completed to a symplectic basis.) Consequently

U Πτ U†=∏i=1k12(I+Zi)=∣0k⟩⟨0k∣⊗In−k.U\,\Pi_\tau\,U^\dagger=\prod_{i=1}^k\tfrac12(\mathbb{I}+Z_i)=|0^k\rangle\langle0^k|\otimes \mathbb{I}_{n-k}.

Now, let A′=UAU†A'=UAU^\dagger. For any nn-qubit stabilizer state ∣s⟩|s\rangle,

⟨s∣A′∣s⟩=⟨U†s∣A∣U†s⟩≥0.\langle s|A'|s\rangle=\langle U^\dagger s|A|U^\dagger s\rangle\ge0.

Here the assumption on the positivity of AA is used: Cliffords map stabilizer states to stabilizer states, so U†∣s⟩U^\dagger|s\rangle is an nn-qubit stabilizer state, and AA is non-negative on it by assumption. So A′A' satisfies the same positivity requirements as AA. Let B=(⟨0k∣⊗In−k) A′ (∣0k⟩⊗In−k)B=(\langle0^k|\otimes \mathbb{I}_{n-k})\,A'\,(|0^k\rangle\otimes \mathbb{I}_{n-k}), an operator on the last n−kn-k qubits. Then, using our findings above,

U ΠτA Πτ U†=(∣0k⟩⟨0k∣⊗In−k) A′ (∣0k⟩⟨0k∣⊗In−k)=∣0k⟩⟨0k∣⊗B.U\,\Pi_\tau A\,\Pi_\tau\,U^\dagger=(|0^k\rangle\langle0^k|\otimes \mathbb{I}_{n-k})\,A'\,(|0^k\rangle\langle0^k|\otimes \mathbb{I}_{n-k})=|0^k\rangle\langle0^k|\otimes B.

To conclude, let ∣t⟩|t\rangle be a stabilizer state of n−kn-k qubits, with stabilizer generators h1,…,hn−kh_1,\dots,h_{n-k}. Then ∣0k⟩⊗∣t⟩|0^k\rangle\otimes|t\rangle is stabilized by the nn independent commuting Paulis Z1,…,Zk, I⊗h1,…,I⊗hn−kZ_1,\dots,Z_k,\ \mathbb{I}\otimes h_1,\dots,\mathbb{I}\otimes h_{n-k}, so it is an nn-qubit stabilizer state. Hence, by the above (that is, by the positivity requirements in the assumption of our lemma applied to this particular nn-qubit stabilizer state),

⟨t∣B∣t⟩=(⟨0k∣⊗⟨t∣) A′ (∣0k⟩⊗∣t⟩) ≥ 0.\langle t|B|t\rangle=\big(\langle0^k|\otimes\langle t|\big)\,A'\,\big(|0^k\rangle\otimes|t\rangle\big)\ \ge\ 0 .

This concludes the proof.

Proof of the main result

Let us write some definitions in order to state the main result differently. From now on we write d:=dn=2nd := d_n = 2^n. Write A=1d∑PaPPA = \tfrac1d\sum_P a_P P over the nn-qubit Paulis, with aI:=Tr[A]a_{\mathbb{I}} := \mathrm{Tr}[A]. Let Λn\Lambda_n be the set of Hermitian operators AA with Tr[A]=1\mathrm{Tr}[A] = 1 and ⟨s∣A∣s⟩≥0\langle s|A|s\rangle \ge 0 for every stabilizer state ss. This is known as the Λ\Lambda-polytope. Furthermore, define the number

Mn=max⁡A∈ΛnTr[A2].M_n = \max_{A\in\Lambda_n}\mathrm{Tr}[A^2].

We know that ρ∈STABn\rho \in \mathrm{STAB}_n iff Tr[Aρ]≥0\mathrm{Tr}[A\rho]\ge 0 for all A∈ΛnA\in\Lambda_n (this is standard convex duality; the polytope Λn\Lambda_n was introduced in the work of Zurel, Okay, and Raussendorf). If we write ρ=I/d+Y\rho = \mathbb{I}/d + Y with Tr[Y]=0\mathrm{Tr}[Y]=0, we can take any A∈ΛnA\in\Lambda_n and then, since Tr[A]=1\mathrm{Tr}[A]=1 and Tr[Y]=0\mathrm{Tr}[Y]=0, we find

Tr[Aρ]=1d+Tr[(A−Id) Y]  ≥  1d−∥A−Id∥2 ∥Y∥2  ≥  1d−Mn−1d  ∥Y∥2.\mathrm{Tr}[A\rho]=\frac{1}{d}+\mathrm{Tr}\big[(A-\tfrac{\mathbb{I}}{d})\,Y\big]\;\ge\;\frac{1}{d}-\big\|A-\tfrac{\mathbb{I}}{d}\big\|_2\,\|Y\|_2 \;\ge\; \frac1d-\sqrt{M_n-\tfrac1d}\;\|Y\|_2.

Above we have used the Cauchy–Schwarz inequality and ∥A−I/d∥22=Tr[A2]−1d≤Mn−1d\|A-\mathbb{I}/d\|_2^2=\mathrm{Tr}[A^2]-\tfrac{1}{d}\le M_n-\tfrac{1}{d}. The right-hand side is non-negative as soon as ∥Y∥22≤1/(d(dMn−1))\|Y\|_2^2\le 1/\big(d(dM_n-1)\big). Because Tr[ρ2]=1d+∥Y∥22\mathrm{Tr}[\rho^2]=\tfrac1d+\|Y\|_2^2, this condition reads Tr[ρ2]≤Mn/(dMn−1)=1/(d−1/Mn)\mathrm{Tr}[\rho^2]\le M_n/(dM_n-1)=1/(d-1/M_n).

Tr[Aρ]≥1d−Mn−1d ∥Y∥2.\mathrm{Tr}[A\rho] \ge \tfrac1d - \sqrt{M_n - \tfrac1d}\,\|Y\|_2.

Hence Tr[ρ2]≤1d−1/Mn  ⟹  ρ∈STABn,\mathrm{Tr}[\rho^2] \le \frac{1}{d - 1/M_n} \;\Longrightarrow\; \rho\in\mathrm{STAB}_n, and the conjecture is exactly that Mn=2M_n = 2 for all nn. With the Lemma, we can now prove the result through a re-statement in terms of MnM_n, namely the bound Mn≤1/cnM_n \leq 1/c_n.

Theorem 2. For all n≥2n \ge 2,

Mn  ≤  7d+22d+7  <  72.M_n \;\le\; \frac{7d+2}{2d+7} \;<\; \frac72 .

Proof. Take A∈ΛnA\in\Lambda_n and an isotropic TT with ∣T∣=2n−2=d/4\vert T \vert = 2^{n-2} = d/4, so that its centraliser has ∣C(T)∣=4d|C(T)| = 4d. By the Lemma, each block ΠτAΠτ\Pi_\tau A\Pi_\tau corresponds to a two-qubit operator BB that is non-negative on all two-qubit stabilizer states, with trace qτ=Tr[ΠτA]q_\tau = \mathrm{Tr}[\Pi_\tau A].

The two-qubit case of the conjecture, M2=2M_2 = 2, is known from the exhaustive enumeration of the facets of the two-qubit stabilizer polytope (see Reichardt); it gives Tr[B2]≤2 Tr[B]2\mathrm{Tr}[B^2] \le 2\,\mathrm{Tr}[B]^2 (if Tr[B]=0\mathrm{Tr}[B] = 0 then B=0B = 0, and otherwise B/Tr[B]∈Λ2B/\mathrm{Tr}[B] \in \Lambda_2). Hence Tr[(ΠτAΠτ)2]≤2qτ2\mathrm{Tr}[(\Pi_\tau A\Pi_\tau)^2] \le 2q_\tau^2. Summing over τ\tau:

  • Left side: ∑τTr[(ΠτAΠτ)2]=Tr[(1∣T∣∑R∈TRAR)2]=1d∑Q∈C(T)aQ2\sum_\tau \mathrm{Tr}[(\Pi_\tau A\Pi_\tau)^2] = \mathrm{Tr}\big[\big(\tfrac{1}{|T|}\sum_{R\in T}RAR\big)^2\big] = \tfrac1d\sum_{Q\in C(T)}a_Q^2, because averaging over TT keeps exactly the Paulis commuting with TT.
  • Right side: qτ=1∣T∣∑R∈Tτ(R)aRq_\tau = \tfrac{1}{|T|}\sum_{R\in T}\tau(R)a_R, so ∑τqτ2=1∣T∣∑R∈TaR2\sum_\tau q_\tau^2 = \tfrac{1}{|T|}\sum_{R\in T}a_R^2.

Therefore

∑Q∈C(T)aQ2  ≤  8∑R∈TaR2.\sum_{Q\in C(T)} a_Q^2 \;\le\; 8\sum_{R\in T} a_R^2 .

Now average this inequality over all such TT. The symplectic group is transitive on non-identity Paulis, so each Q≠IQ \ne \mathbb{I} lies in C(T)C(T) for a fraction (4d−1)/(d2−1)(4d-1)/(d^2-1) of the TT's and in TT for a fraction (d/4−1)/(d2−1)(d/4-1)/(d^2-1). With Σ=∑Q≠IaQ2\Sigma = \sum_{Q\ne \mathbb{I}}a_Q^2 and aI=1a_{\mathbb{I}} = 1:

1+4d−1d2−1Σ  ≤  8+2d−8d2−1Σ⟹Σ≤7(d2−1)2d+7.1 + \frac{4d-1}{d^2-1}\Sigma \;\le\; 8 + \frac{2d-8}{d^2-1}\Sigma \quad\Longrightarrow\quad \Sigma \le \frac{7(d^2-1)}{2d+7}.

Finally Tr[A2]=(1+Σ)/d≤(7d+2)/(2d+7)\mathrm{Tr}[A^2] = (1+\Sigma)/d \le (7d+2)/(2d+7). Since this gives 1/Mn≥(2d+7)/(7d+2)=cn>2/71/M_n \ge (2d+7)/(7d+2) = c_n > 2/7, the implication above yields the Proposition.

Other results in the literature

The most notable recent result which is better than the above is the result shown recently by Liu and Liu in ther arXiv preprint:2609.03944v3 titled "On the geometry, typicality, and detectability of quantum magic".

Their theorem 1 shows that one can get much better than the bound cnc_n above. Their value is ≈0.458\approx 0.458.

Permanent link:open-context.jan-bulling.com/contribution/5.5

Discussion

Sign in to join the discussion.

No comments yet. Questions about this result, checks of its steps and pointers are welcome here.