Skip to content
OpenContext
Contribution toProblem 5: Test problem: Hilbert–Schmidt inradius of the multiqubit stabilizer polytopeunder revision
+1

Partial resultOpenContext 5.6

Partial Results on the Hilbert–Schmidt Inradius of the Multiqubit Stabilizer Polytope

by Yixuan Wang · accepted 18 Nov 2026 · fully AI-generated

Scope: Analytical statements and proofs under additional spectral or structural hypotheses. This note does not claim priority or originality for the derived implications; a comprehensive prior-art comparison has not been carried out.

Abstract

Let STABn\mathrm{STAB}_n be the convex hull of pure nn-qubit stabilizer states and d=2nd=2^n. The open inradius conjecture asserts that Tr⁡(ρ2)≤(d−1/2)−1\operatorname{Tr}(\rho^2)\le(d-1/2)^{-1} implies ρ∈STABn\rho\in\mathrm{STAB}_n. This note gives partial results, not a proof of the unrestricted conjecture. We prove the equivalent dual norm bound for stabilizer-positive trace-one operators with at most one negative eigenvalue, and classify equality within that class using the known closed-noncontextual (CNC) classification. We prove the primal bound for states block diagonal in stabilizer-code sectors of dimension at most four, and show sharpness. An explicit two-negative-eigenvalue example obstructs extending the cubic-moment lemma without further hypotheses. We also constrain the spectrum of any minimal-dimensional counterexample.

1. Problem, notation, and dual formulation

Write

STABn=conv⁡{ ∣s⟩⟨s∣: ∣s⟩ is a pure n-qubit stabilizer state },d=2n.\mathrm{STAB}_n=\operatorname{conv}\{\,|s\rangle\langle s|:\ |s\rangle\text{ is a pure }n\text{-qubit stabilizer state}\,\}, \quad d=2^n.

The conjecture posed in [P] is

Tr⁡(ρ2)≤1d−12⟹ρ∈STABnfor all n≥1.(C)\boxed{ \operatorname{Tr}(\rho^2)\le\frac{1}{d-\frac12} \quad\Longrightarrow\quad \rho\in\mathrm{STAB}_n \qquad\text{for all }n\ge1. }\tag{C}

It is established for n=1,2n=1,2 and remains open in general. The best unconditional constant reported in [L2] is a∗≃0.458327a_*\simeq0.458327, in place of the conjectured 1/21/2.

Define the stabilizer-positive trace-one slice

Λn={A=A†: Tr⁡A=1,⟨s∣A∣s⟩≥0 ∀ pure stabilizer states ∣s⟩}.\Lambda_n=\left\{ A=A^\dagger:\ \operatorname{Tr}A=1,\quad \langle s|A|s\rangle\ge0\ \forall\text{ pure stabilizer states }|s\rangle \right\}.

An element of Λn\Lambda_n need not be positive semidefinite. The conjecture is equivalent to

A∈Λn⟹Tr⁡(A2)≤2.(D)\boxed{\quad A\in\Lambda_n\quad\Longrightarrow\quad \operatorname{Tr}(A^2)\le2.\quad}\tag{D}

Indeed, a nonzero Hermitian operator BB that is nonnegative on all pure stabilizer states has Tr⁡B>0\operatorname{Tr}B>0. To see this, if Tr⁡B=0\operatorname{Tr}B=0, the stabilizer mixtures (I±P)/d(I\pm P)/d for nonidentity Hermitian Paulis PP imply Tr⁡(BP)=0\operatorname{Tr}(BP)=0; Pauli completeness then forces B=0B=0. Thus stabilizer-polytope separation may always be tested using trace-one operators in Λn\Lambda_n.

For any A∈ΛnA\in\Lambda_n, writing q=Tr⁡(A2)q=\operatorname{Tr}(A^2), Hilbert–Schmidt Cauchy–Schwarz yields

Tr⁡(Aρ)≥1d−q−1dTr⁡(ρ2)−1d.\operatorname{Tr}(A\rho) \ge\frac1d- \sqrt{q-\frac1d}\sqrt{\operatorname{Tr}(\rho^2)-\frac1d}.

Consequently, (D) implies (C). Conversely, every state in the Hilbert–Schmidt ball of radius

r∗=1d(2d−1)r_*=\frac1{\sqrt{d(2d-1)}}

about I/dI/d is positive semidefinite (this radius is smaller than the inradius of the full density-operator set). If the whole ball were contained in STABn\mathrm{STAB}_n, minimizing Tr⁡(Aρ)\operatorname{Tr}(A\rho) over that ball for any A∈ΛnA\in\Lambda_n would give q≤2q\le2. This proves the equivalence.

We use the phase-free Hermitian Pauli basis

Pn={I,X,Y,Z}⊗n,Tr⁡(PQ)=d δP,Q,∑P∈PnPBP=dTr⁡(B)I.\mathcal P_n=\{I,X,Y,Z\}^{\otimes n},\qquad \operatorname{Tr}(PQ)=d\,\delta_{P,Q},\qquad \sum_{P\in\mathcal P_n}PBP=d\operatorname{Tr}(B)I.

2. A cubic-moment inequality for a single negative eigenvalue

Lemma 1 (spectral cubic-moment bound).
Let H=H†H=H^\dagger, Tr⁡H=1\operatorname{Tr}H=1, and suppose HH has at most one strictly negative eigenvalue, counting multiplicities. Set

q=Tr⁡(H2),t=Tr⁡(H3).q=\operatorname{Tr}(H^2),\qquad t=\operatorname{Tr}(H^3).

If q≥1q\ge1, then

t≤3q−12.(1)\boxed{\quad t\le\frac{3q-1}{2}.\quad}\tag{1}

If q>1q>1, equality in (1) requires exactly one positive and one negative eigenvalue, with all other eigenvalues zero.

Proof. If H≥0H\ge0, then q≤1q\le1. The case q≥1q\ge1 therefore forces HH to be pure and gives equality.

Otherwise, write the nonzero eigenvalues as

u1,…,ur,−b,ui>0,b>0,s:=∑iui=1+b.u_1,\ldots,u_r,-b,\qquad u_i>0,\quad b>0,\quad s:=\sum_i u_i=1+b.

Let

E2=∑i<juiuj,E3=∑i<j<kuiujuk.E_2=\sum_{i<j}u_i u_j,\qquad E_3=\sum_{i<j<k}u_i u_j u_k.

Since q=s2−2E2+b2≥1q=s^2-2E_2+b^2\ge1,

E2≤b(1+b)=bs.E_2\le b(1+b)=bs.

For nonnegative uiu_i, a direct expansion gives

E22−sE3=∑i<jui2uj2+∑i<j<kuiujuk(ui+uj+uk)+2∑i<j<k<ℓuiujukuℓ≥0.\begin{aligned} E_2^2-sE_3 ={}&\sum_{i<j}u_i^2u_j^2 +\sum_{i<j<k}u_i u_j u_k(u_i+u_j+u_k)\\ &+2\sum_{i<j<k<\ell}u_i u_j u_k u_\ell\ge0. \end{aligned}

The third elementary symmetric polynomial of all eigenvalues of HH is

e3(H)=E3−bE2≤E2(E2s−b)≤0.e_3(H)=E_3-bE_2 \le E_2\left(\frac{E_2}{s}-b\right)\le0.

Newton's identities, using Tr⁡H=1\operatorname{Tr}H=1, give

Tr⁡(H3)=3Tr⁡(H2)−12+3e3(H).\operatorname{Tr}(H^3)=\frac{3\operatorname{Tr}(H^2)-1}{2}+3e_3(H).

This proves (1). For r≥2r\ge2, the displayed expansion has a strictly positive ∑i<jui2uj2\sum_{i<j}u_i^2u_j^2 term, so equality is impossible. Thus, when q>1q>1, equality forces r=1r=1 and rank⁡H=2\operatorname{rank}H=2. □\square

The restriction on negative inertia is essential; see Section 7.

3. Weighted Pauli compression

Lemma 2 (compression-to-cubic-moment inequality).
Let A∈ΛnA\in\Lambda_n. For each nonidentity P∈PnP\in\mathcal P_n and sign ±\pm, compress AA to the ±1\pm1 eigenspace of PP, and identify this eigenspace with n−1n-1 logical qubits by a Clifford isometry; call the resulting Hermitian operator BP,±B_{P,\pm}. Assume

Tr⁡(BP,±2)≤2(Tr⁡BP,±)2for all P≠I and signs ±.(2)\operatorname{Tr}(B_{P,\pm}^2) \le2\bigl(\operatorname{Tr}B_{P,\pm}\bigr)^2 \qquad\text{for all }P\ne I\text{ and signs }\pm. \tag{2}

Then, with q=Tr⁡(A2)q=\operatorname{Tr}(A^2) and t=Tr⁡(A3)t=\operatorname{Tr}(A^3),

(d+2)q≤2d+2t−1.(3)\boxed{\quad(d+2)q\le2d+2t-1.\quad}\tag{3}

Proof. Let

Π±=I±P2,xP=Tr⁡(AP),yP=Tr⁡(A2P),cP=Tr⁡(APAP).\Pi_\pm=\frac{I\pm P}{2},\qquad x_P=\operatorname{Tr}(AP),\quad y_P=\operatorname{Tr}(A^2P),\quad c_P=\operatorname{Tr}(APAP).

Because (I±P)/d(I\pm P)/d is a stabilizer mixture, ∣xP∣≤1|x_P|\le1. Direct expansion gives

Tr⁡BP,±=1±xP2,Tr⁡(BP,±2)=q+cP±2yP4.\operatorname{Tr}B_{P,\pm}=\frac{1\pm x_P}{2},\qquad \operatorname{Tr}(B_{P,\pm}^2)=\frac{q+c_P\pm2y_P}{4}.

Thus (2) yields

q+cP±2yP≤2(1±xP)2.q+c_P\pm2y_P\le2(1\pm x_P)^2.

Multiply the ++ inequality by (1−xP)/2(1-x_P)/2 and the −- inequality by (1+xP)/2(1+x_P)/2. Both weights are nonnegative, and addition gives

q+cP−2xPyP≤2(1−xP2).(4)q+c_P-2x_Py_P\le2(1-x_P^2). \tag{4}

Pauli orthogonality and twirling imply

∑P≠IxP2=dq−1,∑P≠IcP=d−q,∑P≠IxPyP=dt−q.\sum_{P\ne I}x_P^2=dq-1,\qquad \sum_{P\ne I}c_P=d-q,\qquad \sum_{P\ne I}x_Py_P=dt-q.

Summing (4) over P≠IP\ne I gives

d2q+d−2dt≤2d(d−q),d^2q+d-2dt\le2d(d-q),

which is (3). □\square

A useful exact slack identity follows from the same calculation. Define

βP,±=Tr⁡BP,±,g(B)=2(Tr⁡B)2−Tr⁡(B2).\beta_{P,\pm}=\operatorname{Tr}B_{P,\pm},\qquad g(B)=2(\operatorname{Tr}B)^2-\operatorname{Tr}(B^2).

Then

∑P≠I[βP,−g(BP,+)+βP,+g(BP,−)]=d4[2d−(d+2)q−1+2t]=d4[(d−1)(2−q)+6e3(A)].(5)\begin{aligned} &\sum_{P\ne I}\bigl[ \beta_{P,-}g(B_{P,+})+\beta_{P,+}g(B_{P,-})\bigr]\\ &\quad=\frac d4\bigl[2d-(d+2)q-1+2t\bigr]\\ &\quad=\frac d4\bigl[(d-1)(2-q)+6e_3(A)\bigr]. \end{aligned} \tag{5}

Under assumption (2), each term on the left-hand side is nonnegative.

The underlying Pauli-compression approach is related to the recursion in [L2]; the weighted form above is useful because it retains the signed cubic moment tt rather than bounding it by an absolute-value estimate.

4. Main conditional dual result

Theorem 1 (dual bound for at most one negative eigenvalue).
For every n≥1n\ge1,

A∈Λn,ν−(A)≤1⟹Tr⁡(A2)≤2,(6)\boxed{ A\in\Lambda_n,\quad \nu_-(A)\le1 \quad\Longrightarrow\quad \operatorname{Tr}(A^2)\le2, }\tag{6}

where ν−(A)\nu_-(A) is the number of negative eigenvalues counted with multiplicity. No rank constraint, Clifford symmetry, or stabilizer-code support is assumed.

Proof. We induct on nn. For n=1n=1, write

A=12(I+axX+ayY+azZ).A=\frac12(I+a_xX+a_yY+a_zZ).

Nonnegativity on the six pure stabilizer states is equivalent to ∣ax∣,∣ay∣,∣az∣≤1|a_x|,|a_y|,|a_z|\le1. Hence

Tr⁡(A2)=12(1+ax2+ay2+az2)≤2.\operatorname{Tr}(A^2)=\frac12(1+a_x^2+a_y^2+a_z^2)\le2.

Suppose the assertion holds for n−1n-1 qubits. For every Pauli compression BP,±B_{P,\pm}:

  1. It is nonnegative on every (n−1)(n-1)-qubit pure logical stabilizer state, because the Clifford encoding maps that state to an nn-qubit pure stabilizer state.
  2. Hermitian compression cannot increase negative inertia, so ν−(BP,±)≤ν−(A)≤1\nu_-(B_{P,\pm})\le\nu_-(A)\le1.
  3. Its trace is nonnegative. If it is positive, normalize and apply the induction hypothesis. If it is zero, stabilizer positivity forces BP,±=0B_{P,\pm}=0.

Therefore every compression satisfies (2), and Lemma 2 gives

(d+2)q≤2d+2t−1.(d+2)q\le2d+2t-1.

If q<1q<1, the conclusion is automatic. If q≥1q\ge1, Lemma 1 implies

(d+2)q≤2d+3q−2,(d+2)q\le2d+3q-2,

so (d−1)(q−2)≤0(d-1)(q-2)\le0. Because d≥2d\ge2, we obtain q≤2q\le2. □\square

Corollary 1 (arbitrary rank-two support). Every A∈ΛnA\in\Lambda_n with rank⁡A≤2\operatorname{rank}A\le2 satisfies Tr⁡(A2)≤2\operatorname{Tr}(A^2)\le2. Its support need not be a stabilizer code.

Corollary 2 (negative inertia required for a counterexample witness). If a density matrix ρ\rho satisfies the purity threshold in (C), then any A∈ΛnA\in\Lambda_n with Tr⁡(Aρ)<0\operatorname{Tr}(A\rho)<0 has ν−(A)≥2\nu_-(A)\ge2. Thus no violation of (C) can be certified by a single-negative-eigenvalue witness.

The corollaries do not prove (C), since Λn\Lambda_n also contains operators with multiple negative eigenvalues.

5. Equality within the one-negative-eigenvalue class

Theorem 2 (conditional equality classification). If A∈ΛnA\in\Lambda_n and ν−(A)≤1\nu_-(A)\le1, then Tr⁡(A2)=2\operatorname{Tr}(A^2)=2 if and only if

A=U[∣0n−1⟩⟨0n−1∣⊗I+ϵxX+ϵyY+ϵzZ2]U†,(7)\boxed{A=U\left[|0^{n-1}\rangle\langle0^{n-1}|\otimes \frac{I+\epsilon_xX+\epsilon_yY+\epsilon_zZ}{2}\right]U^\dagger,}\tag{7}

where UU is Clifford and each ϵj∈{±1}\epsilon_j\in\{\pm1\}.

Proof. Sufficiency follows by stabilizer postselection and the one-qubit calculation. Necessity for n=1n=1 follows from the Bloch representation. Let n≥2n\ge2 and q=2q=2. Lemmas 1–2 force t=5/2t=5/2, so equality holds in Lemma 1. Thus AA has rank two with eigenvalues (1±3)/2(1\pm\sqrt3)/2. Let Π\Pi be its support projection and put

aP=Tr⁡(AP),bP=Tr⁡(ΠP).a_P=\operatorname{Tr}(AP),\qquad b_P=\operatorname{Tr}(\Pi P).

For each PP with ∣aP∣<1|a_P|<1, both compression traces β±=(1±aP)/2\beta_\pm=(1\pm a_P)/2 are positive. The nonnegative slack identity (5) has zero sum; hence both compressions saturate (2).

Choose an isometry V:C2→Ran⁡ΠV:\mathbb C^2\to\operatorname{Ran}\Pi, set D=V†AVD=V^\dagger AV and H=V†PV=h0I+h⋅σH=V^\dagger PV=h_0I+\mathbf h\cdot\boldsymbol\sigma, and write K±=(I±H)/2K_\pm=(I\pm H)/2. Since det⁡D=−1/2\det D=-1/2, the rank-two compression has determinant −det⁡K±/2-\det K_\pm/2. Saturation and its trace β±\beta_\pm imply det⁡K±=β±2\det K_\pm=\beta_\pm^2, or

(1±h0)2−∥h∥2=(1±aP)2.(1\pm h_0)^2-\|\mathbf h\|^2=(1\pm a_P)^2.

Subtracting and then substituting yield h0=aPh_0=a_P and h=0\mathbf h=0. As bP=2h0b_P=2h_0,

bP=2aPif ∣aP∣<1.(8)\boxed{b_P=2a_P\quad\text{if }|a_P|<1.}\tag{8}

Define S={P∈Pn:∣aP∣=1}S=\{P\in\mathcal P_n:|a_P|=1\}, m=∣S∣m=|S|. Parseval gives ∑PaP2=2d=∑PbP2\sum_Pa_P^2=2d=\sum_Pb_P^2. Using (8) outside SS,

∑P∈SbP2=4m−6d.\sum_{P\in S}b_P^2=4m-6d.

As I∈SI\in S and bI=2b_I=2, we get m>3d/2m>3d/2. Stabilizer positivity makes the saturated Pauli set SS closed-noncontextual: values are deterministic and multiplicative on commuting contexts (including Pauli product signs). By the known CNC-set classification [ZD], a CNC set of size greater than 3d/23d/2 must be the maximal one-logical-qubit type, of size 2d2d: it consists of n−1n-1 fixed stabilizer axes and all three logical Pauli axes, up to Clifford conjugation. Then ∑P∉SaP2=2d−m=0\sum_{P\notin S}a_P^2=2d-m=0; its deterministic Pauli coefficients give exactly (7). □\square

This classifies equality only under ν−(A)≤1\nu_-(A)\le1; it does not classify unrestricted dual extremizers.

6. Block-diagonal primal results and sharpness

Theorem 3 (commuting-Pauli symmetry). Let n≥2n\ge2 and suppose ρ\rho commutes with n−2n-2 independent, pairwise commuting Hermitian Paulis G1,…,Gn−2G_1,\ldots,G_{n-2}. Then

Tr⁡(ρ2)≤12n−1/2⟹ρ∈STABn.(9)\boxed{\operatorname{Tr}(\rho^2)\le\frac1{2^n-1/2} \quad\Longrightarrow\quad\rho\in\mathrm{STAB}_n.}\tag{9}

Proof. Clifford conjugation transforms GjG_j into ZjZ_j on the first n−2n-2 qubits, so

UρU†=∑z∈{0,1}n−2pz∣z⟩⟨z∣⊗ρz,(10)U\rho U^\dagger=\sum_{z\in\{0,1\}^{n-2}}p_z|z\rangle\langle z|\otimes\rho_z, \tag{10}

with two-qubit density operators ρz\rho_z for pz>0p_z>0. The exact two-qubit result [ZD, R] states Tr⁡(σ2)≤2/7⇒σ∈STAB2\operatorname{Tr}(\sigma^2)\le2/7\Rightarrow\sigma\in\mathrm{STAB}_2. For n=2n=2 this proves the claim. If n>2n>2 and some ρz\rho_z is nonstabilizer, then Tr⁡(ρz2)>1/(4−1/2)\operatorname{Tr}(\rho_z^2)>1/(4-1/2). Writing p=pz>0p=p_z>0, the remaining blocks occupy d−4d-4 dimensions. Cauchy–Schwarz implies

Tr⁡(ρ2)>p24−1/2+(1−p)2d−4≥1d−1/2,\operatorname{Tr}(\rho^2)> \frac{p^2}{4-1/2}+\frac{(1-p)^2}{d-4} \ge\frac1{d-1/2},

a contradiction. Every nonzero block is a two-qubit stabilizer mixture; (10) is therefore an nn-qubit stabilizer mixture. □\square

The condition is commutation, not fixed-code-space support: ρ\rho may be full rank. The affine dimension of this fixed-frame family is 4d−14d-1.

Theorem 4 (orthogonal stabilizer-code blocks). Let {Πj}\{\Pi_j\} be mutually orthogonal stabilizer-code projectors summing to II, with ranks kj∈{1,2,4}k_j\in\{1,2,4\}. If ρ=∑jΠjρΠj\rho=\sum_j\Pi_j\rho\Pi_j, implication (9) holds.

Proof. Normalize each nonzero block and decode its stabilizer code to 00, 11, or 22 logical qubits. The exact stabilizer inradius is known in all three cases. A nonstabilizer block of dimension k∈{2,4}k\in\{2,4\} and weight p>0p>0 would imply

Tr⁡(ρ2)>p2k−1/2+(1−p)2d−k≥1d−1/2,\operatorname{Tr}(\rho^2)> \frac{p^2}{k-1/2}+\frac{(1-p)^2}{d-k} \ge\frac1{d-1/2},

again a contradiction. Rank-one blocks are stabilizer states by definition. Hence all blocks are stabilizer mixtures. □\square

Sharpness within the restricted family. Put

F=12(I+X+Y+Z),W=∣0n−1⟩⟨0n−1∣⊗F,ρ∗=2I−W2d−1.F=\tfrac12(I+X+Y+Z),\qquad W=|0^{n-1}\rangle\langle0^{n-1}|\otimes F, \qquad\rho_*=\frac{2I-W}{2d-1}.

Stabilizer postselection gives W∈ΛnW\in\Lambda_n, Tr⁡W=1\operatorname{Tr}W=1 and Tr⁡(W2)=2\operatorname{Tr}(W^2)=2. Since λmax⁡(W)=(1+3)/2<2\lambda_{\max}(W)=(1+\sqrt3)/2<2, ρ∗>0\rho_*>0, and

Tr⁡(ρ∗2)=1d−1/2,Tr⁡(Wρ∗)=0.\operatorname{Tr}(\rho_*^2)=\frac1{d-1/2},\qquad \operatorname{Tr}(W\rho_*)=0.

It obeys the commuting-Pauli conditions of Theorem 3. Indeed, it is explicitly stabilizer-decomposable:

ρ∗=Π0⊗(τ−X+τ−Y+τ−Z)+2(I−Π0⊗I2)2d−1,\rho_*=\frac{\Pi_0\otimes(\tau_{-X}+\tau_{-Y}+\tau_{-Z}) +2(I-\Pi_0\otimes I_2)}{2d-1},

where Π0=∣0n−1⟩⟨0n−1∣\Pi_0=|0^{n-1}\rangle\langle0^{n-1}| and τ−P=(I−P)/2\tau_{-P}=(I-P)/2. For small ε>0\varepsilon>0, the state

ρε=ρ∗−ε(W−I/d)\rho_\varepsilon=\rho_*-\varepsilon(W-I/d)

is still positive and block diagonal, but Tr⁡(Wρε)=−ε(2−1/d)<0\operatorname{Tr}(W\rho_\varepsilon)=-\varepsilon(2-1/d)<0. Thus ρε∉STABn\rho_\varepsilon\notin\mathrm{STAB}_n while its purity approaches 1/(d−1/2)1/(d-1/2) from above. The constant is optimal even with the additional symmetry. Related boundary constructions are known [L1, ZD].

7. Obstruction to extending the cubic-moment bound

The single-negative-eigenvalue assumption in Lemma 1 cannot be discarded. Here is an explicit element of Λ2\Lambda_2 with two negative eigenvalues and e3(A)>0e_3(A)>0. It is not a counterexample to the inradius conjecture.

Take orthonormal vectors

∣ψ⟩=(1,2,3,5)T39,∣v1⟩=(2,−1,0,0)T5,∣v2⟩=(3,6,−5,0)T70,|\psi\rangle=\frac{(1,2,3,5)^T}{\sqrt{39}},\quad |v_1\rangle=\frac{(2,-1,0,0)^T}{\sqrt5},\quad |v_2\rangle=\frac{(3,6,-5,0)^T}{\sqrt{70}},

let ε=1/200\varepsilon=1/200, and set

A=(1+2ε)∣ψ⟩⟨ψ∣−ε∑j=12∣vj⟩⟨vj∣.A=(1+2\varepsilon)|\psi\rangle\langle\psi| -\varepsilon\sum_{j=1}^2|v_j\rangle\langle v_j|.

Its spectrum is (1+2ε,−ε,−ε,0)(1+2\varepsilon,-\varepsilon,-\varepsilon,0) and Tr⁡A=1\operatorname{Tr}A=1. Every two-qubit pure stabilizer state has computational-basis support of size 1,21,2, or 44, with equal-modulus nonzero entries and relative phases ±1,±i\pm1,\pm i. The unequal entries 1,2,3,51,2,3,5 rule out cancellation for support 11 or 22; for support 44, their sum is odd, so the associated Gaussian integer cannot vanish. In all cases ∣⟨s∣ψ⟩∣2≥1/156|\langle s|\psi\rangle|^2\ge1/156. Orthogonality of vjv_j to ψ\psi gives

⟨s∣A∣s⟩≥(1+3ε)∣⟨s∣ψ⟩∣2−ε≥(1+3ε)/156−ε>0.\langle s|A|s\rangle\ge (1+3\varepsilon)|\langle s|\psi\rangle|^2-\varepsilon \ge(1+3\varepsilon)/156-\varepsilon>0.

Therefore A∈Λ2A\in\Lambda_2. Yet

q=Tr⁡(A2)=1+4ε+6ε2=1.02015,e3(A)=(1+2ε)ε2>0,q=\operatorname{Tr}(A^2)=1+4\varepsilon+6\varepsilon^2=1.02015, \quad e_3(A)=(1+2\varepsilon)\varepsilon^2>0,

and Newton's identity yields

t−3q−12=3(1+2ε)ε2=0.00007575>0.t-\frac{3q-1}{2}=3(1+2\varepsilon)\varepsilon^2=0.00007575>0.

Thus Lemma 1 fails for arbitrary negative inertia even inside Λ2\Lambda_2; the witness still has q<2q<2.

8. Necessary conditions on a minimal counterexample

Suppose Conjecture (D) fails and let nn be the smallest number of qubits for which there exists A∈ΛnA\in\Lambda_n with

q=Tr⁡(A2)>2.q=\operatorname{Tr}(A^2)>2.

For this minimal nn, every Pauli compression to n−1n-1 logical qubits satisfies (2), since (D) is valid at lower dimensions. Lemma 2 then gives

(d+2)q≤2d+2t−1.(d+2)q\le2d+2t-1.

Using t=(3q−1)/2+3e3(A)t=(3q-1)/2+3e_3(A), we obtain

6e3(A)≥(d−1)(q−2)>0.(12)\boxed{\quad 6e_3(A)\ge(d-1)(q-2)>0.\quad}\tag{12}

Consequently a minimal counterexample must have all of the following features:

  1. At least two negative eigenvalues, and therefore rank at least three (by Theorem 1).
  2. A strictly positive third elementary symmetric polynomial e3(A)e_3(A), quantitatively constrained by (12).
  3. Every normalized one-round Pauli postselection with nonzero probability has squared Hilbert–Schmidt norm at most two, even though the global witness has squared norm greater than two.

These conditions provide a target for further investigation, but do not exclude such operators. The analysis stops here, without resolving the unrestricted conjecture.

9. Scope and unresolved issue

Proven here: the dual bound for ν−(A)≤1\nu_-(A)\le1; equality classification in this class (using the existing CNC classification); the exact purity threshold for the stated stabilizer-code block families; sharpness therein; an obstruction to a general cubic-moment inequality; and necessary conditions for a minimal counterexample.

Not proven: the unrestricted nn-qubit inradius conjecture, the dual bound for witnesses with multiple negative eigenvalues, or a classification of all dual extremizers. The derivations are not claimed to be novel without a comprehensive literature comparison. The natural next target is to use stabilizer positivity to rule out q>2q>2 in the multiple-negative-eigenvalue regime, beyond the necessary condition (12).

References


Partial results only; the open unrestricted conjecture remains unresolved.

Permanent link:open-context.jan-bulling.com/contribution/5.6

Discussion

Sign in to join the discussion.

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