{"id":"5d11c03c-ae3b-4f07-b0f6-b74f57bb6595","arxiv_id":"2507.12326","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The authors combine noise symmetries (iid, isotropic, Choi) with the extension symmetry of a known AQEC bounding hierarchy, add a measurement-based rounding that converts outer bounds into certified codes, and demonstrate the framework on three-qubit codes under symmetric noise.","lead":"The paper develops computational tools for approximate quantum error correction (AQEC) under symmetric noise: a rounding procedure that converts bounding solutions into concrete encoder-decoder pairs with certified fidelity, and a symmetry-based dimension reduction that makes higher levels of a known SDP bounding hierarchy tractable. The methods are demonstrated on small three-qubit codes under depolarizing and amplitude-damping noise.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Prop. 5.1's averaging proof verifies tr_P[ρ]=1_L/d_L⊗ρ (Eq. 64) rather than the hierarchy constraints tr_L[ρ]=1_P/d_P⊗ρ and tr_{\\bar P^{(n)}}[ρ]=...⊗1_{\\bar L}/d; Theorem 5.3's symmetry reduction is unproven as written.","rationale":"The reader's weakest assumption identifies the same load-bearing point: Proposition 5.1's enlarged-invariance proof, as written, checks the wrong partial-trace constraints. This is indeed the gate through which all of Section 5 passes: Theorem 5.3 asserts equality of optimal values for the symmetry-reduced SDP only if the averaged optimizer still satisfies the level-n hierarchy constraints (iii) and (iv). The displayed computation in Eqs. (64)-(65), and the corresponding statements in Lemma 5.2 (68)-(69) and Theorem 5.3 (71), use tr_P[ρ] = 1_L/d_L ⊗ ρ_bar instead of tr_L[ρ] = 1_P/d_P ⊗ ρ_bar and tr_{\\bar P^{(n)}}[...] = ... ⊗ 1_{\\bar L}/d. That is not a cosmetic issue: a reader cannot verify the main reduction from the text, and an implementation following the printed equations would enforce the wrong constraints. However, a careful check shows the intended statement is very likely correct: for the diagonal actions used in Section 5, the averaging unitaries on the traced systems cancel inside the partial trace, so the true constraints are preserved. Thus this is an addressable exposition/typo problem rather than a refutation of the central claim. The block decomposition in Subsection 5.4.1 also independently checks out (the dimensions sum to 4096), and Theorem 3.1's de Finetti rounding is structurally sound. The unresolved numerical reproducibility and the sketchy proof of Theorem 7.6 are secondary; the decisive issue is the missing verification of the actual hierarchy constraints in Prop. 5.1. Since the concern is real but fixable, the reader's CONDITIONAL verdict is appropriate and no verdict change is needed.","tokens_in":35778,"tokens_out":23935,"duration_ms":248953,"concrete_test":"Independently re-derive Prop. 5.1 for each of the three symmetry actions (S_m diagonal on P and all \\bar P^{(i)}; Choi V diagonal on P and all \\bar P^{(i)}; U diagonal on L and all \\bar L^{(i)}) and check the two actual hierarchy constraints: tr_L[\\tildeρ] = 1_P/d_P ⊗ \\tildeρ_bar and tr_{\\bar P^{(n)}}[\\tildeρ_bar] = \\tildeρ_bar^{(n-1)} ⊗ 1_{\\bar L^{(n)}}/d_{\\bar L^{(n)}}. If both hold, correct Eqs. (64)-(65) and Lemma 5.2 accordingly; Theorem 5.3 then stands. If either fails for any symmetry, the symmetry-reduced SDP is not equivalent to (63) and the numerical outer bounds in Section 6 are not certified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Proposition 5.1's enlarged-invariance claim: an optimizer of (63) can be replaced by one invariant under the diagonal iid/Choi/isotropic actions without changing the value, so that Theorem 5.3 can restrict to the commutant. The proof, however, verifies the wrong identities. Eq. (64) computes tr_P[\\tildeρ] and concludes it equals 1_L/d_L ⊗ \\tildeρ_bar; Eq. (65) computes tr_{PL}[\\tildeρ] = \\tildeρ_bar. Neither is among the level-n constraints (8)/(63), which require tr_L[\\tildeρ] = 1_P/d_P ⊗ \\tildeρ_bar and tr_{\\bar P^{(n)}}[\\tildeρ_bar] = \\tildeρ_bar^{(n-1)} ⊗ 1_{\\bar L^{(n)}}/d_{\\bar L^{(n)}}. The same P/L swap appears in Lemma 5.2 (68)-(69), and the constraints in Theorem 5.3 (71) inherit this swap (ψ2/tr_P/1_L terms). If the averaging does not preserve the true constraints, the reduced SDP optimizes over a different feasible set and Theorem 5.3's equality of optimal values fails. Indications are that the correct constraints are in fact preserved—the averaging unitaries on the traced system cancel inside the partial trace—so this is likely a fixable typo, but the claim is unproven as written. Section 7 makes clear the diagonal action is essential: Example 7.2 shows the same groups with a non-diagonal action generate a larger group W=UV for which Definition 7.3 fails; thus the proof must be re-run for the diagonal action with the correct trace constraints.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper revisits the extendability-based SDP hierarchy for approximate quantum error correction (AQEC) introduced by Berta et al. and contributes two main developments. First, it proposes a measurement-based rounding scheme that converts outer-hierarchy optimizers into feasible inner encoder-decoder pairs with a convergence guarantee comparable to the outer hierarchy. Second, it develops a symmetry-reduction framework for AQEC that combines the permutation symmetry of n-extendable states with symmetries of the noise model, specifically iid, isotropic, and Choi symmetries, and applies it to compute outer bounds for encoding one logical qubit into three physical qubits for depolarizing and amplitude-damping noise. The paper also contains a general discussion of when two symmetry groups can be combined for SDP reduction, formulated through the condition that the set UV is a group. The numerical section reports Level-1 and Level-2 outer bounds for the three-qubit examples.","tokens_in":36052,"tokens_out":7862,"duration_ms":86515,"significance":"If the proof defects described below are repaired, this is a useful and timely contribution. The rounding theorem is an original mechanism with explicit convergence bounds, and the explicit representation-theoretic block decomposition in Section 5.4.1 is a valuable concrete data point: the block dimensions sum to 4096, showing internal consistency. The numerical examples demonstrate that nontrivial Level-2 bounds for small symmetric codes can be obtained, which was previously intractable with generic SDP solvers. The general criterion for combining symmetry groups in Section 7 addresses a genuine methodological question. However, the load-bearing symmetry-reduction claim currently rests on a proof that verifies the wrong constraint identities, so the mathematical core of the paper needs a substantial repair before the claims can be accepted.","major_comments":[{"comment":"The averaging argument proves the wrong constraint identities. Equation (64) verifies tr_P[tilde rho] = 1_L/d_L ⊗ tilde rho_{barP barL} and Eq. (65) verifies tr_PL[tilde rho] = tilde rho_{barP barL}, but the level-n constraints in Eq. (63) require tr_L[rho] = 1_P/d_P ⊗ rho_{barP barL} and tr_{barP^(n)}[rho_{barP barL}] = rho_{barP barL}^{(1..n-1)} ⊗ 1_{barL^(n)}/d_{barL^(n)}. The proof therefore does not establish that the averaged state lies in Sigma^n_prod, and without that Proposition 5.1 and hence Theorem 5.3 are unproven. Since the averaging unitaries act on the systems over which the required partial traces are taken, the intended preservation is plausible, but the proof must be rerun with the correct constraints.","section":"Section 5.1, Proposition 5.1"},{"comment":"Equations (68)-(69) in Lemma 5.2 state tr_P[rho] = 1_L/d_L ⊗ rho_{barP barL} and tr_L[rho_{barP barL}] = rho_{barP barL}^{(1..n-1)} ⊗ 1_P/d_P; neither is a hierarchy constraint. The hierarchy requires tr_L on the full state with 1_P/d_P on the P-side and tr_{barP^(n)} on the reduced state with 1_{barL^(n)}/d_{barL^(n)}. The constraints in Eq. (71) inherit this swap, so the claimed equivalence between the symmetry-reduced SDP and the level-n relaxation is not established. These statements are load-bearing for the numerical results in Section 6 and must be corrected.","section":"Section 5.2, Lemma 5.2 and Theorem 5.3"},{"comment":"The dimension factor in the proof is inconsistent: Eq. (25) and the original program use d_P^2, while Eq. (31) repeatedly uses d_{barL barP}^2. The measurement parameter m is also not fixed: the proof constructs an m-fold product measurement and then states that all 2≤m≤n are iterated, but the object sigma[n] is never defined as a function of m, and the relationship between m and the extension level n is left implicit. The rounding scheme is conceptually sound, but the proof needs a clean statement with consistent dimensions before Theorem 3.1's guarantee is certified.","section":"Section 3.1, Theorem 3.1"},{"comment":"The proof of the group extension structure is not rigorous as written. The map φ: G → Sym(barP_{1...m} × ... × barP^(n)_{1...m}) is stated to be surjective onto S_m≀S_n without a construction, and the kernel computation refers to 'ker(G→S_m≀S_{n-1})' although the preceding map is to S_m≀S_n. The conclusion G ≅ A_m ⋊ (S_m≀S_n) may be correct, but as written the proof does not establish the order of the kernel, which Corollary 7.7 relies on. This should be repaired for the general framework of Section 7 to be reliable.","section":"Section 7.2, Theorem 7.6"}],"minor_comments":[{"comment":"The notation S(H_LP, barLbarP) is not standard; the states should be density matrices on H_LP and H_barLbarP, respectively.","section":"Section 2.2, Eq. (5)"},{"comment":"The informationally complete measurement M is not defined as a POVM; the normalization and the distortion constant c(d) should be stated explicitly.","section":"Section 3.1, Eq. (26)"},{"comment":"The text says '12 blocks of sizes' but the table lists multiplicities; the actual block dimensions should be given explicitly, or the entries should be called multiplicities.","section":"Section 5.4.1, Table (94)"},{"comment":"The sentence 'a several linear variables and four 2×2 SDP variables' is ungrammatical, and the proof of the LP claim in the trivial case shown in Eq. (101) is only sketched.","section":"Proposition 5.5"},{"comment":"The braced expressions in Eq. (121) are notationally garbled and should be cleaned up; the notation V in (113) also conflicts with the Choi-symmetry group V used elsewhere.","section":"Example 7.2, Eq. (121)"},{"comment":"No details are given on how the Level 2 curves were obtained, such as solver, precision, or confirmation of the block decomposition; adding this information would strengthen the numerical claim.","section":"Figure 4"}],"recommendation":"major_revision","confidential_remarks":"The repeated P/L swap in Proposition 5.1, Lemma 5.2, and Theorem 5.3 strongly suggests a systematic typographical error rather than a fundamental conceptual flaw, since the intended averaging likely does preserve the correct hierarchy constraints. However, the proof as written does not establish the main theorem, and the manuscript should not be accepted until this is fixed and the reduced SDP is rigorously shown to be equivalent to the level-n relaxation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth engaging with. It does three real things: it constructs a rounding scheme that converts outer-hierarchy optimizers into certified inner codes, it builds a symmetry-reduction framework that combines iid, isotropic, Choi, and extendability symmetries, and it gives a clean group-theoretic criterion for when such combined reductions are valid. The representation-theoretic tables check out (block dimensions sum to 4096), and the numerical examples, while modest, show the method producing outer bounds that were previously out of reach. This is a genuine step toward making AQEC bounds computable for small codes.\n\nThe soft spots are real but addressable. The stress-test note is on target: the proof of Proposition 5.1 verifies the wrong partial-trace identities. Equation (64) uses tr_P[ρ] = 1_L/d_L ⊗ ρ, but the hierarchy constraint (63) requires tr_L[ρ] = 1_P/d_P ⊗ ρ and a matching condition on the ¯P^{(n)} trace. The same P/L swap appears in Lemma 5.2 and is inherited by Theorem 5.3. As written, the symmetry reduction is unproven. That said, the averaging idea itself is sound: because the symmetries act diagonally on the P and L systems, the correct constraints are preserved by cyclicity of the partial trace under the twirl. The error is likely a typo, not a conceptual break. Section 7's Example 7.2 actually supports this reading by showing that the diagonal action is essential; the proof just needs to be re-run with the right trace constraints.\n\nTwo smaller issues. Theorem 3.1's statement is looser than the proof: it claims the same performance guarantee as Proposition 2.1(a), but the bound in (31) has an extra d^2 factor and a square root of log that don't match the clean poly(d_L,d_P)√(ln d/n) in the abstract. That is a presentation flaw, not a broken result. And the numerics come without code, data, or error bars, so a referee cannot independently verify the plots; for a proof-of-principle section this is forgivable but worth flagging.\n\nThe central argument holds up, modulo the fixable proof gap. This deserves a serious referee. I would send it to review and ask the authors to correct the P/L swap, tighten Theorem 3.1's constants, and release the numerical code.","headline":"A genuinely useful symmetry-reduction framework for AQEC, with a likely-fixable proof gap in the central averaging lemma.","tokens_in":36779,"tokens_out":1626,"would_cite":true,"duration_ms":21882,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","90C22","20C30","81P45"],"pacs":[],"model":"deepseek-v4-flash","headline":"Outer SDP bounds for approximate quantum error correction can be rounded into concrete codes, and noise symmetries can be merged with extendability to make fixed-level relaxations computable.","keywords":["approximate quantum error correction","semidefinite programming hierarchy","symmetry reduction","extendability","channel fidelity","rounding scheme","depolarizing channel","quantum de Finetti theorem"],"falsifier":"Take the non-commuting action of Example 7.2 (two-copy iid symmetry versus two-column extendability on six subsystems), compute the symmetry-reduced level-2 SDP in the commutant of the generated group $W$, and compare its value with the un-reduced level-2 relaxation; Proposition 5.1 and Theorem 5.3 predict equality only for diagonal actions, so a mismatch would disprove the enlarged-invariance claim.","tokens_in":35435,"feed_emoji":"🛡️","tokens_out":8345,"duration_ms":89601,"temperature":0.7,"pith_summary":"This paper targets the gap between outer bounds and real codes in approximate quantum error correction (AQEC): the extendability semidefinite programming (SDP) hierarchy gives converging upper bounds on the best channel fidelity, but its fixed levels are too large to evaluate. The authors claim two fixes work together. First, a measurement-based rounding scheme turns an optimal solution of the level-$n$ relaxation into a genuinely feasible encoder--decoder pair whose fidelity is within the same polynomial-in-dimension, $\\sqrt{\\ln d / n}$ gap as the outer bound. Second, they show that noise symmetries---permutation symmetry of identical copies, isotropic unitary symmetry, and Choi covariance---can be combined with the extendability permutation symmetry, provided the symmetries act diagonally on every extension copy. The payoff is that non-trivial outer bounds for one logical qubit encoded into three physical qubits under depolarizing or amplitude-damping noise become numerically accessible.","feed_headline":"Rounding turns error-correction bounds into real quantum codes","feed_subtitle":"Merging noise and extension symmetries shrinks the semidefinite program to small, solvable blocks.","key_machinery":"The objects doing the work are the constrained $n$-extendable states, the sets $\\Sigma^n_{\\mathrm{prod}}(LP:\\bar L\\bar P)$, whose constraints (positivity, permutation invariance, and two partial-trace equalities) encode Choi's theorem. The rounding machinery is an informationally complete measurement on the extension copies, which projects an outer solution onto a convex combination of product states while preserving the partial-trace constraints. The symmetry machinery is the enlarged diagonal action of the iid-symmetry group $S_m$, the isotropic unitary group $U(2)$, and the Choi covariance group $V$ on all extension copies, together with the extendability group $S_n$; Schur--Weyl duality and branching rules decompose the commutant into small isotypic blocks, and Proposition 5.1 supplies the group averaging that keeps the hierarchy constraints intact. The general criterion for merging an objective symmetry $U$ with a variable symmetry $V$ is that $W=\\{UV\\}$ be a group.","core_discovery":"On the paper's own terms, the central discovery is a pair of theorems. Theorem 3.1 states that from a level-$n$ optimizer of the outer hierarchy one can construct inner states $\\sigma^{(n)}_{LP\\bar L\\bar P}\\in\\Sigma^n_{\\mathrm{prod}}$ whose fidelity gap to the true channel fidelity closes as $c(d_{\\bar L\\bar P})\\sqrt{2\\ln 2\\,\\ln d_{LP}/n}$, matching the outer convergence rate; the construction measures the extended systems with an informationally complete measurement and forms a separable post-measurement state. Theorem 5.3 states that, when the noise symmetries are extended diagonally to all $n$ extension copies, the symmetry-reduced SDP in the commutant of the combined group $W$ has the same optimal value as the original level-$n$ relaxation; the proof rests on Proposition 5.1's group-averaging argument and a Hilbert--Schmidt basis of the commutant. The paper also shows that for a global depolarizing channel the reduction degenerates to a linear program at low levels, and Section 7 identifies a necessary condition for combining symmetry groups: the product set $W=\\{UV\\}$ must itself be a group.","pith_inferences":["The rounding scheme is not tied to AQEC: it should transfer to other constrained separability problems where the hierarchy constraints are partial-trace constraints and an informationally complete measurement is available.","A testable extension is to benchmark asymmetric noise (e.g., unequal depolarizing rates per qubit) where the iid symmetry is absent; the Section 7 criterion predicts that only diagonal actions of the remaining symmetries can be safely merged.","The linear-program reductions suggest that small symmetric codes can be exhaustively benchmarked against the best non-signaling-assisted bounds, which may reveal how much of the gap comes from the coding constraints rather than the noise model.","One could strengthen the rounding by constructing problem-adapted measurements with fewer outcomes; the paper notes such measurements give valid codes, though without the rigorous optimality guarantee."],"forward_implications":["Fixed level-2 outer bounds for one logical qubit encoded into three physical qubits under iid depolarizing or amplitude-damping noise become computable, including cases that previously required intractable dense SDPs.","The rounding theorem yields explicitly constructible encoder-decoder pairs whose fidelity is within the same $O(\\sqrt{\\ln d / n})$ gap as the outer bound, giving certified warm starts for see-saw optimization.","For global depolarizing channels, the symmetry reduction makes low levels linear programs; for $m$-fold single-qubit depolarizing without extensions, the reduced program is a linear program.","The symmetry-reduced SDP has exactly the same optimal value as the original level-$n$ relaxation, so the reduced computation is still a valid outer bound.","When the combined symmetry set $W=\\{UV\\}$ is not a group, joint symmetry reduction fails, so the diagonal-action condition is not just convenience but necessary."],"supporting_citations":[{"why":"Supplies the extendability SDP hierarchy and its de Finetti-style convergence theorem, which both the outer bounds and the rounding proof rely on.","marker":"[3]"},{"why":"Formulates channel fidelity as a bilinear optimization over encoder-decoder Choi matrices and provides the see-saw inner bounds that the new rounding targets.","marker":"[31]"},{"why":"Provides the standard symmetry-reduction framework for SDPs via the regular *-representation, which Section 5 extends to combined symmetries.","marker":"[12]"},{"why":"Worked out extendability-symmetry reduction for AQEC; the paper's contribution is combining that with noise symmetries.","marker":"[9]"},{"why":"Gives the informationally complete measurement and distortion bound $c(d)=O(d)$ that controls the rounding fidelity gap.","marker":"[22]"},{"why":"Supplies the representation theory of the symmetric group and branching rules used to decompose the commutant into small blocks.","marker":"[34]"}],"fun_headline_variants":["Rounding turns QEC bounds into certifiable codes","Symmetry cuts SDP size for quantum error correction","Matching convergence: rounding and symmetry in AQEC","From outer bounds to inner codes: a rounding scheme","Practical QEC via symmetry-reduced SDPs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the noise symmetries can be made to act diagonally on every copy of the extended system, so that averaging an optimal solution over the enlarged group still satisfies all hierarchy constraints.","fun_headline_variants_meta":{"raw":{"variants":["Rounding turns QEC bounds into certifiable codes","Symmetry cuts SDP size for quantum error correction","Matching convergence: rounding and symmetry in AQEC","From outer bounds to inner codes: a rounding scheme","Practical QEC via symmetry-reduced SDPs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000212,"raw_usage":{"total_tokens":1423,"prompt_tokens":952,"completion_tokens":471,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":395}},"tokens_in":568,"tokens_out":471,"duration_ms":5764,"temperature":1.0,"reasoning_tokens":395,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:54:22.482923+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the non-commuting action of Example 7.2 (two-copy iid symmetry versus two-column extendability on six subsystems), compute the symmetry-reduced level-2 SDP in the commutant of the generated group $W$, and compare its value with the un-reduced level-2 relaxation; Proposition 5.1 and Theorem 5.3 predict equality only for diagonal actions, so a mismatch would disprove the enlarged-invariance claim.","supporting_citations":[{"cited_title":"Semidefinite programming hierarchies for constrained bilinear optimization","cited_arxiv_id":null,"evidence_quote":"Supplies the extendability SDP hierarchy and its de Finetti-style convergence theorem, which both the outer bounds and the rounding proof rely on."},{"cited_title":"Iterative Optimization of Quantum Error Correcting Codes","cited_arxiv_id":null,"evidence_quote":"Formulates channel fidelity as a bilinear optimization over encoder-decoder Choi matrices and provides the see-saw inner bounds that the new rounding targets."},{"cited_title":"Reduction of symmetric semidefinite programs using the regular∗ -representation","cited_arxiv_id":null,"evidence_quote":"Provides the standard symmetry-reduction framework for SDPs via the regular *-representation, which Section 5 extends to combined symmetries."},{"cited_title":"Efficient Approximation of Quantum Channel Fidelity Exploiting Symmetry","cited_arxiv_id":"2308.15884","evidence_quote":"Worked out extendability-symmetry reduction for AQEC; the paper's contribution is combining that with noise symmetries."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the representation theory of the symmetric group and branching rules used to decompose the commutant into small blocks."}],"review_version":1}