{"id":"8d5850ef-af06-4d42-bf14-fe8598a9d98c","arxiv_id":"2505.00427","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For unitarily covariant codes, approximate correction of local erasure is possible exactly when the conditional min-entropy of the encoded Choi state is at most a simple function of the logical dimension and the allowed error probability.","lead":"Quantum codes that can run any gate while also surviving erased pieces are normally forbidden by the Eastin-Knill theorem. This paper proves a new rule for the approximate case: a single entropy number decides whether such a code can meet a target error rate, and that number can be computed by an optimization program.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Corollary 1's only-if direction relies on an unproved identification of per-gate transversality with U(d)-covariance under a genuine tensor-product representation; Lemma 3's twirled decoder needs such a representation.","rationale":"The reader's weakest_assumption identifies the same load-bearing point: Corollary 1 imports the equivalence between transversal implementation of the full unitary group and U(dL)-covariance under a tensor-product representation. My reading sharpens why this is load-bearing: Lemma 3's twirled decoder is only guaranteed covariant if the physical gate family used in the twirl is an actual representation, and Theorem 1's reduction to the covariant purification problem in Lemma 1 needs that covariance. If a code merely has per-gate local unitaries that agree with the logical action on the code subspace but not as a representation globally, the necessity proof does not go through. I found no internal inconsistency in the main entropic calculation: the twirl identity in Eq. (D8) is correct for the U^*⊗U representation, Lemma 4 checks out, and the W-state threshold in Corollary 2 is consistent with an optimal decoder that maps the erased component to the maximally mixed state. The paper's explicit decoder in Example 2 appears suboptimal relative to Corollary 2, since it achieves error Ne/n rather than (Ne/n)(1 - 1/dL), but this affects the example's 'optimal decoder' claim rather than the theorem. The missing epsilon domain is real but repairable. Overall the reader's CONDITIONAL verdict remains appropriate; the concern should be addressed by proving or precisely citing the transversality-covariance equivalence, and by stating the allowed epsilon range.","tokens_in":22607,"tokens_out":43547,"duration_ms":463368,"concrete_test":"Test the missing equivalence directly: take the W-state isometric encoder E and, for each U in U(dL), set V_U = (E U E†) ⊕ W_U, where W_U is a local unitary on the code complement chosen so that W_U W_V ≠ W_{UV}. This family satisfies the paper's per-gate transversality definition but is not a representation. Then check whether the twirled decoder R_G := ∫ dg U^g†_L R V^g_{P'} in Lemma 3 remains U(dL)-covariant and whether the concavity argument f(R_G) ≥ f(R) still goes through. If R_G is not covariant, the reduction to Lemma 1 requires the unstated equivalence. Alternatively, verify directly from Ref. [15] whether it proves the extension of per-gate transversal gates to a tensor-product unitary representation on the full physical space; if not, Corollary 1 should be restricted to codes satisfying Eq. (32) with a genuine representation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central iff claim for arbitrary codes with a universal transversal gateset depends on an equivalence that is asserted and cited to Ref. [15] rather than proved. Eq. (3) requires, for each U in U(dL), existence of local unitaries V_U satisfying E U = V_U E, but Lemma 3 and Theorem 1 need a single unitary representation g -> U^g_P such that E U^g_L = U^g_P E and N U^g_P = U^g_{P'} N. Corollary 1's proof says Eq. (32) 'encompasses' the transversal case, but if the V_U are chosen per gate and only constrained on the code subspace, they need not satisfy V_U V_W = V_{UW} on the full physical space. In that case the twirled decoder R_G = ∫ dg U^g†_L R U^g_{P'} in Lemma 3 is not manifestly G-covariant, so the reduction of epsilon-correctability to the G-covariant multi-state purification problem in Lemma 1 is not established. Thus the necessity direction of the entropic condition is unproved for codes whose transversal gates do not form a genuine tensor-product representation. A smaller, repairable gap is that the condition -log dL(1 - c epsilon) is undefined unless epsilon ≤ dL/(dL+1), which is not stated in Theorem 1 or Corollary 1.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a resource-theoretic framework for approximate quantum error correction and states a new necessary and sufficient condition for a U(d_L)-covariant encoder followed by U(d_L)-covariant noise to be epsilon-correctable in the worst-case pure-state fidelity sense. The condition is an upper bound on the conditional min-entropy of the Choi state of the noisy encoding, and it is claimed to hold for any code admitting a transversal implementation of the full unitary group when the noise is local erasure. The main tool is a multi-state purification lemma for the resource theory of asymmetry, and the paper applies the result to the W-state code, deriving the exact threshold epsilon >= (N_e/n)(1 - 1/d_L) and constructing an explicit decoder for known erasure.","tokens_in":22743,"tokens_out":23072,"duration_ms":227235,"significance":"If the stated equivalence between transversal universality and U(d_L)-covariance is justified, the paper provides a clean single-shot entropic criterion that is both necessary and sufficient, SDP-computable, and directly connected to the Eastin-Knill theorem. The multi-state purification lemma (Lemma 1) is a useful generalization of prior asymmetry-theory results and is of independent interest. The W-state example gives a concrete, falsifiable threshold and an explicit achievability construction. The appendices contain detailed proofs, and the derivations are largely self-contained. The main reservation is that the advertised scope of Corollary 1 is broader than what the proof establishes.","major_comments":[{"comment":"The proof of Corollary 1 identifies 'admits a transversal implementation of the full unitary group' with U(d_L)-covariance under a fixed tensor-product representation, as expressed in Eq. (32). This identification is asserted and cited to Ref. [15] but not proved. The per-gate definition in Eq. (3) only guarantees, for each U, existence of local unitaries V_U satisfying E U = V_U E; it does not guarantee that the V_U form a genuine group representation on the full physical space. Lemma 3 and Theorem 1 require a single representation g -> U^g_P satisfying E U^g_L = U^g_P E and N U^g_P = U^g_{P'} N. If the transversal implementations are chosen per gate and only constrained on the code subspace, the twirled decoder in Lemma 3 is not manifestly G-covariant, and the reduction of epsilon-correctability to the multi-state purification problem is not established. Thus the necessity direction of the entropic condition is unproved for codes whose transversal gates do not form a tensor-product representation. The authors should either prove the equivalence or explicitly restrict Corollary 1 to U(d_L)-covariant encoders, which is the setting of Theorem 1.","section":"Section IV C, Corollary 1; Section II B"},{"comment":"The inequality Hmin(L|P') <= -log d_L(1-c epsilon) is only defined for epsilon < d_L/(d_L+1), since 1-c epsilon becomes non-positive for larger epsilon. For epsilon >= d_L/(d_L+1) the condition is automatically satisfied by the general lower bound Hmin >= -log d_L, but this is not stated. The theorem and corollary should specify the domain of epsilon and clarify the behavior at the boundary.","section":"Theorem 1, Eq. (26); Corollary 1, Eq. (31)"}],"minor_comments":[{"comment":"The explicit decoder defined in Eq. (43) is not shown to be trace-preserving. In particular, the adjoint V(n-N_e)^\\dagger annihilates the state |d_L,...,d_L>, so applying the decoder to the component (N_e/n)|d_L,...,d_L>= (Ne/n)(1 - 1/dL) comes straight out of the theorem, and I verified the algebra. The explicit decoder is a nice extra, though it saturates the bound only in the large-dL limit; the paper's wording \"achieves the scaling\" is acceptable but should not be mistaken for finite-dL optimality.\n\nThe soft spots are real but manageable. The first is in Corollary 1. The paper identifies \"admits a transversal implementation of the full unitary group\" with U(dL)-covariance under a fixed tensor-product representation, citing [15] rather than proving it. That matters because the proof of the necessity direction uses a twirled decoder built from a single representation. If the per-gate transversal unitaries V_U are chosen independently, they need not satisfy V_U V_W = V_UW off the code subspace, so the reduction to the covariant problem is not automatic. As written, the corollary overstates its scope. The fix is straightforward: either state covariance as an explicit hypothesis in the corollary, or prove the equivalence. The second issue is minor: the inequality loses meaning unless epsilon < dL/(dL+1), and that domain is not stated in Theorem 1 or Corollary 1.\n\nWho gets value: anyone working on approximate QEC, covariant codes, or asymmetry purification. Lemma 1 is likely to be cited on its own. I would send this to a serious referee. The main theorem stands, and the corollary's gap is repairable; a revision that tightens the transversality-to-covariance step and states the epsilon range would make this a solid piece. I would not desk reject it.","headline":"Genuinely new iff condition for approximate erasure correction of covariant codes, with an exact W-state threshold; the transversal-code corollary needs a proved covariance equivalence.","tokens_in":23399,"tokens_out":6973,"would_cite":true,"duration_ms":71852,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P70","81P45"],"pacs":["03.67.Pp","03.67.-a"],"model":"deepseek-v4-flash","headline":"A single conditional min-entropy bound is necessary and sufficient for a quantum code to support universal transversal gates and approximately correct local erasure.","keywords":["Eastin-Knill theorem","approximate quantum error correction","transversal gates","erasure noise","conditional min-entropy","resource theory of asymmetry","covariant quantum codes","W-state code"],"falsifier":"Compute $H_{\\min}(L|P/j)_{J(\\mathrm{tr}_{P_j}\\circ E)}$ for a $U(d_L)$-covariant code and compare it with $-\\log d_L(1-c\\varepsilon)$; a code that violates the inequality yet admits a decoder with worst-case infidelity $\\varepsilon$ would refute the necessity direction. A concrete candidate: find any decoder for the $W$-state code that reaches $\\varepsilon < (N_e/n)(1-1/d_L)$ under erasure, for instance $\\varepsilon<0.005$ with $n=100$ qutrits and one erased subsystem, since the paper claims no such decoder exists.","tokens_in":22253,"feed_emoji":"⚛️","tokens_out":13843,"duration_ms":123832,"temperature":0.7,"pith_summary":"This paper claims that the Eastin-Knill obstruction to combining quantum error correction with universal transversal gates has an exact single-shot approximate form. For any encoder that is covariant with respect to the full unitary group $U(d_L)$ and any covariant noise channel, the code is $\\varepsilon$-correctable if and only if a conditional min-entropy of the Choi state of the encoded-and-noised channel falls below $-\\log d_L(1-c\\varepsilon)$, with $c=(d_L+1)/d_L$. For local erasure this specializes to a condition on the Choi state after tracing out the erased subsystem. The condition is checkable by a semidefinite program, so it gives a rigorous yes-or-no test for whether a covariant code can meet a target error rate, and for the $W$-state code it evaluates to the exact threshold $\\varepsilon \\ge (N_e/n)(1-1/d_L)$. This turns a theorem that forbids an entire regime into a boundary that distinguishes achievable from unachievable approximate transversal codes.","feed_headline":"One entropy bound decides when quantum codes can use transversal gates","feed_subtitle":"A necessary-and-sufficient entropy test, solvable by SDP, fixes the exact error threshold of the W-state code.","key_machinery":"The machinery is the conditional min-entropy $H_{\\min}(A|B)_\\Sigma$, a single-shot entropy measuring how much uncertainty about $A$ remains given $B$, used as a resource monotone in the resource theory of asymmetry, together with the identification of covariant quantum error correction with multi-state purification. The key identity expresses the optimal overlap with a pure target under $G$-covariant channels as $2^{-H_{\\min}(B|A)_{\\Pi_G(\\psi^T\\otimes\\rho)}}$; for the compact group $U(d_L)$, Haar-twirl identities collapse the purification condition into a bound on the Choi state of $N\\circ E$. A functional additivity lemma splits the twirled operator into a flat, deterministic term and the Choi term, which is what produces the prefactor $c=(d_L+1)/d_L$, and the final condition is an SDP because conditional min-entropy is SDP-computable. In the erasure case the noise is simply a partial trace, which is covariant for any group, so the whole theorem transfers to the transversal-gate setting.","core_discovery":"The central claim is Theorem 1: if the encoding map $E_{L\\to P}$ and the noise $N_{P\\to P'}$ are both $U(d_L)$-covariant, then a recovery map achieving worst-case infidelity $\\varepsilon$ over all pure logical states exists if and only if $H_{\\min}(L|P')_{J(N\\circ E)} \\le -\\log d_L(1-c\\varepsilon)$, where $c=(d_L+1)/d_L$ and $J(\\cdot)$ is the Choi state. Corollary 1 applies this to erasure of the $j$-th subsystem, giving $H_{\\min}(L|P/j)_{J(\\mathrm{tr}_{P_j}\\circ E)} \\le -\\log d_L(1-c\\varepsilon)$ as the necessary and sufficient condition for a code to admit a transversal implementation of the full unitary group and approximately correct that erasure. The proof routes quantum error correction through the resource theory of asymmetry: under covariant encoding and noise the optimal decoder may be taken covariant, so finding a decoder is the same as finding a covariant purification of every logical state up to error $\\varepsilon$, and the multi-state purification lemma reduces this to a single conditional min-entropy. At $\\varepsilon=0$ the bound reproduces the exact Eastin-Knill impossibility, and the $W$-state code saturates the inequality, yielding the closed-form threshold of Corollary 2.","pith_inferences":["The paper leaves open whether the same purification machinery applies to discrete subgroups such as the Clifford group; if it does, the result would give necessary-and-sufficient single-shot thresholds for magic-state-injection-style fault tolerance, not just full unitary covariance.","Because the theorem is an if-and-only-if SDP, one could in principle optimize over encoders to minimize achievable $\\varepsilon$ under erasure; the paper does not perform such a search, but its formulation makes this a natural next step.","The fact that the $W$-state code saturates the bound suggests, though the paper does not claim, that saturating families may be characterized by a Choi state splitting into a uniform term plus a smaller covariant code; verifying this would give a structural classification of optimal approximate transversal codes."],"forward_implications":["For any $U(d_L)$-covariant code, the condition can be evaluated by solving an SDP, giving a certificate that a target worst-case error $\\varepsilon$ is either achievable or impossible before any decoder is constructed.","The $W$-state code corrects the erasure of $N_e$ subsystems exactly when $\\varepsilon \\ge (N_e/n)(1-1/d_L)$; with $n=100$ qutrits and one encoded qubit this puts the achievable single-erasure error at $0.005$.","Setting $\\varepsilon=0$ in the bound recovers the original Eastin-Knill no-go result, so the theorem contains the exact theorem as a limiting case rather than merely approximating it.","For erasure of any number $m