{"id":"47cf376d-37be-4d32-8f86-74bd3293216f","arxiv_id":"2506.04530","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A quantum code corrects a noise subspace exactly when that subspace admits a special basis of partial isometries with orthogonal ranges, a criterion equivalent to Knill-Laflamme but with new structural consequences.","lead":"This paper reformulates quantum error correction in terms of inner products and noise bases built from partial isometries, proving new existence criteria for correcting codes. It offers a fresh operator-theoretic framework for code design, though the core criterion reduces to known Knill-Laflamme conditions.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main equivalence theorems hold; Example 4.13's largest-code non-existence proof is incomplete and must be replaced by a direct argument.","rationale":"I read the paper in good faith and identified what must be true for the central claim: that the existence of a C-decoding inner product or a C-decoding noise basis is equivalent to the Knill-Laflamme conditions. I checked the proofs of Theorems 2.3, 3.4, 3.6, and Corollary 3.5 step by step. The constructions are consistent, the positivity arguments are valid, and the Kraus operators in Theorem 3.4 do form a channel that annihilates the noise on the code. No fatal flaw appears in the main equivalence. The reader's stated weakest assumption, the non-negligible-noise quotient in Remark 3.3, is a legitimate caveat but not a true obstruction: a finite error set can always be replaced by its span and then quotiented by the negligible part, and the constructions of Sections 4 and 5 apply to the quotient. I therefore do not consider it the most load-bearing concern. The genuine gap is Example 4.13. The paper's contradiction is asserted rather than proved: it relies on the specific basis {I,V,V^2} instead of establishing that no basis of N2 can satisfy (3.9) for a⊕b. My own analytical check for that example indicates the non-existence conclusion is actually correct, but the proof needs to be rewritten. Because the reader already flagged this example in the rationale and issued a conditional verdict, my stress-test does not move the verdict; it reinforces the existing conditional status. The agreement is partial because I do not share the reader's emphasis on the quotient as the weakest assumption; I identify the incomplete Example 4.13 as the more concrete and load-bearing issue.","tokens_in":17904,"tokens_out":50932,"duration_ms":445783,"concrete_test":"Repair Example 4.13 by a direct linear-algebra check. For S = span{e0,e1,e5,e8}, write N = aI + bV + cV^2 from N2 and impose that N|S is an isometry, i.e. that the four vectors N e0, N e1, N e5, N e8 are orthonormal. Solve the resulting equations; they yield the isometry condition |a|^2+|b|^2+|c|^2 = 1 and \\bar{b}a + \\bar{c}b = 0. Then compute the cross-inner-product conditions for two such isometries to have orthogonal S-ranges. The solution set splits into a 2D subspace with b=0 and a single ray with fixed c/b for b≠0, so no three pairwise-orthogonal isometries exist. If this calculation is included, the example is correct; if it fails, the non-existence claim is false.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The core equivalence theorems (Theorems 3.4 and 3.6) are internally sound: the construction of a C-decoding inner product from a decoder, and the converse construction of a decoder from a C-decoding noise basis, follow from the stated assumptions under the non-negligible-noise hypothesis. The central claim therefore survives scrutiny. The load-bearing weakness is in Section 4.2, specifically Example 4.13, which is intended to prove that the largest correcting code need not exist. The proof as written argues that if a largest code existed, then a⊕b would be an N2-cc, and then observes that V(a⊕b)∩(a⊕b)=b, calling this a contradiction with the orthogonality condition (4.1). This is not a valid contradiction: (4.1) requires the existence of some C-decoding noise basis of N2, not that the particular basis {I,V,V^2} satisfies it. An intersection between V(C) and C for a single element of the noise space does not by itself preclude another linear basis of N2 from having orthogonal ranges. The paper gives no argument ruling out such a basis. The claim may be true, but the proof is incomplete and needs replacement.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a 'C-decoding inner product' on a noise subspace N, defined by the condition that P_C N^* M P_C = phi(N,M) P_C, and proves an equivalence between the existence of such an inner product and the existence of a decoding channel for the code C (Theorems 3.4 and 3.6). It then characterizes code-correcting noise bases in terms of partial isometries with mutually orthogonal ranges, uses the von Neumann-Wold decomposition to build such bases from powers of a partial isometry, gives a criterion for 1-dimensional codes for cyclic shift and clock operators, and extends these constructions to reducing subspaces.","tokens_in":18161,"tokens_out":42925,"duration_ms":407858,"significance":"If fully correct, the paper would provide a clean operator-theoretic perspective on quantum error correction: the noise-basis formulation of the Knill-Laflamme conditions is elegant and leads to concrete constructions, notably Theorem 4.10 on the largest correctable code for powers of a unilateral shift and Theorem 4.14 for cyclic shifts. The proofs of the central equivalence theorems in Section 3 are detailed and internally consistent under the stated non-negligible-noise assumption. However, the claimed non-existence example in Section 4.2 is not proved as written, and Section 5 contains a dimensional error in Proposition 5.2 and Theorem 5.7. These issues affect two of the paper's advertised contributions, so the manuscript needs substantial revision.","major_comments":[{"comment":"The argument that a direct sum a⊕b cannot be an N_2-cc is incomplete. The orthogonality condition (4.1) requires the existence of a basis of N_2 whose ranges are mutually orthogonal, and the observation that V(a⊕b) ∩ (a⊕b) = b for the single operator V does not rule out some other basis {N_1,N_2,N_3} of N_2 satisfying (4.1). A valid proof should use the necessary condition of Corollary 3.5(iii) or Theorem 2.3(ii), for example by showing that P_{a⊕b} V P_{a⊕b} is not a scalar multiple of P_{a⊕b}. As written, the example does not establish the claimed non-existence of a largest N_2-correcting code.","section":"Section 4.2, Example 4.13"},{"comment":"Proposition 5.2 is false as stated for k>1: if each N_s has dimension n, then ⊕_{s=1}^k N_s has dimension k n, while the set {⊕_{s=1}^k N_{sj}}_{j=1}^n contains only n operators and therefore cannot be a basis of ⊕ N_s. The correct statement is that this set is a C-decoding noise basis of the subspace span{⊕ N_{sj} : j=1,...,n}. Consequently, Theorem 5.7's claim that {⊕_{s=1}^k T_s^r}_{r∈Z_m} is a noise basis of ⊕ N_s is incorrect when k>1 and m_s > m; it is a basis only of the diagonal subspace span{⊕ T_s^r : r∈Z_m}. This invalidates the stated generalization in the last section and needs to be corrected.","section":"Section 5, Proposition 5.2 and Theorem 5.7"}],"minor_comments":[{"comment":"In the proof of (iii)⇒(i), the displayed chain 'φ(N,M)PC = ... = φ(M,N)PC' is not correct as written; taking adjoints gives φ(M,N) = \\overline{φ(N,M)}, since the inner product is anti-linear in the first argument. The conclusion that φ is an inner product is salvageable if 'symmetric' is replaced by 'conjugate-symmetric'.","section":"Section 3, Corollary 3.5"},{"comment":"The operator V is called a 'partial unitary' but it is not one: its kernel is span{e12,e13,e14} while its range contains e12,e13,e14, so supp V ≠ ran V. The decomposition into unitary and completely non-unitary parts given later in the example is consistent with V being a partial isometry, not a partial unitary.","section":"Section 4.2, Example 4.13"},{"comment":"The notation 'Z_3 = {0,1,2,3}' is a typo; the index set should be Z_4.","section":"Section 4.2, Example 4.13"},{"comment":"In the proof, the sentence 'Note by (4.4) that {V^j}_{j=0}^t is a basis of N' should refer to N_t, not N.","section":"Section 4.2, Theorem 4.10"},{"comment":"The inner product formula should read φ(N,M) = ∑_{j=1}^{n-1} n_j \\overline{m_j}; the displayed formula omits the conjugation.","section":"Section 3, Theorem 3.6, Eq. (3.10)"},{"comment":"The symbol U is used both for the shift operator and for the noise subspace U = span{U^r}_{r∈Z_m}; this is confusing and should be disambiguated.","section":"Section 4.3"},{"comment":"The definition of 'largest N-correcting code' should explicitly require that the largest code C itself is an N-cc; the proof of Remark 4.12 uses this property via Remark 3.2.","section":"Section 4.2, Definition 4.9 and Remark 4.12"},{"comment":"There are numerous typos, including 'Krauss' for 'Kraus', 'c.f.' for 'cf.', a duplicated MSC code '15A63', and 'B(C)-cc' in the paragraph after Corollary 3.7.","section":"General"},{"comment":"The decoder formula in Corollary 5.8 is for the noise subspace span{⊕_{s=1}^k U_s^r}_{r∈Z_m}, not for the full direct sum ⊕ N_s; the statement should say so explicitly to avoid the dimensional ambiguity noted above.","section":"Section 5, Corollary 5.8"}],"recommendation":"major_revision","confidential_remarks":"The central equivalence theorems in Section 3 are essentially a reformulation of the Knill-Laflamme conditions, so the paper's real novelty lies in the noise-basis examples and in Section 5. The dimensional error in Proposition 5.2/Theorem 5.7 and the incomplete argument in Example 4.13 are fixable but currently undermine two of the paper's advertised contributions. I would also ask the authors to correct the conjugate-symmetry slip in the proof of Corollary 3.5 and the mislabeling of V as 'partial unitary'."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The central equivalence theorems hold up, but the paper oversells them as new, and Example 4.13's proof of non-existence of a largest correcting code is incomplete as written.\n\nWhat is actually new: the code-decoding noise basis framework turns Knill-Laflamme into an explicit construction tool. Given a basis of the noise subspace satisfying the partial-isometry/orthogonality condition (4.1), you get a decoding channel for free. That is a clean reformulation, and the derivation is rigorous. The genuinely new results are Theorem 4.10, which characterizes the largest correcting code for the span of powers of a unilateral shift, and the reduction-subspace generalizations in Section 5. Those are worth having.\n\nWhere it gets soft: the \"new necessary and sufficient conditions\" in Sections 2 and 3 are, as you noted, restatements of Knill-Laflamme. The C-decoding inner product condition (2.1) is essentially the KL condition in disguise, so Theorem 3.4 is a rephrasing rather than an independent criterion. The authors should tone down the novelty language. That is a framing issue, not a mathematical one.\n\nThe load-bearing flaw is Example 4.13. It aims to prove that the set of N2-correcting codes need not have a largest element. The argument shows that for the specific basis {I,V,V^2}, the candidate code a⊕b fails the orthogonality condition. But (4.1) only requires existence of some C-decoding noise basis, not that the generating set you happen to look at satisfies it. Since C and F are shown to be N2-cc, they each possess some good basis; the example does not rule out a basis that works for a⊕b. So the non-existence proof is incomplete. There are also typos in the example (Z3 vs Z4, and L = span{e4,e5,e6} looks like it should be span{e0,e5,e6}) that make it hard to verify even the setup. The claim may be true, but it needs a direct argument, not this. The rest of Section 4 and Section 5 do not depend on this example, so the damage is contained.\n\nOne caveat: the main construction theorems are stated for the quotient noise after removing negligible operators. Applying them to a raw error set requires checking that the set is closed under the quotient. The paper notes this in Remark 3.3 but could be clearer in the examples.\n\nWho it's for: quantum error correction theorists who want an operator-theoretic perspective and explicit decoder constructions for shift-type noise. It is not a breakthrough, but it is a solid piece of work if Example 4.13 is fixed and the novelty claims are tempered.\n\nRecommendation: send it to peer review. The core proofs are careful, the new results on unilateral shifts are real, and the flaws are fixable.","headline":"Core equivalence theorems are sound but are restatements of Knill-Laflamme; Example 4.13's largest-code non-existence proof is incomplete and needs a direct argument.","tokens_in":18688,"tokens_out":4521,"would_cite":false,"duration_ms":41545,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P70","94B60","15A63","81P55"],"pacs":[],"model":"deepseek-v4-flash","headline":"Correctability of a quantum code is equivalent to the existence of a decoding inner product on the noise space.","keywords":["quantum error-correcting codes","inner products","partial isometry error bases","quantum channels","Knill-Laflamme conditions","von Neumann-Wold decomposition","shift and clock operators","largest correcting code"],"falsifier":"Enumerate all pairs $(C,\\mathcal{N})$ in a small Hilbert space, say $\\dim H = 3$, that satisfy $N P_C \\neq 0$ for every nonzero $N\\in\\mathcal{N}$, and compare the Knill-Laflamme condition (3.8) with the existence of a basis of $\\mathcal{N}$ satisfying (3.9); a single pair for which the first holds and the second fails would refute Theorem 3.6. The paper itself notes that without the assumption the statement degenerates: for $H=\\mathbb{C}^2$, $C=\\operatorname{span}\\{e_0\\}$, $\\mathcal{N}=\\operatorname{span}\\{I, |e_1\\rangle\\langle e_1|\\}$, the correction equation holds with $\\Phi=\\mathrm{id}$, but no $C$-decoding inner product exists because $|e_1\\rangle\\langle e_1|$ annihilates $C$.","tokens_in":17726,"feed_emoji":"⚛️","tokens_out":14930,"duration_ms":149650,"temperature":0.7,"pith_summary":"The paper's central claim is that quantum error correction can be characterized by a special inner product on the noise space: a code $C$ corrects a noise subspace $\\mathcal{N}$ exactly when $\\mathcal{N}$ carries a $C$-decoding inner product, equivalently when $\\mathcal{N}$ has a basis of partial isometries whose images of $C$ are mutually orthogonal. This recasts the Knill-Laflamme conditions in operator language and makes code construction a search for such a basis, from which an explicit decoding channel is written down. The authors construct these bases for partial isometries via the von Neumann-Wold decomposition, for cyclic unitary shift and clock operators, and for direct sums over reducing subspaces, and they show that a largest correcting code need not exist. If the characterization holds, the dimension restriction $\\dim\\mathcal{N}\\cdot\\dim C\\leq\\dim H$ and the trade-off between code size and noise size follow directly from the structure of the noise space.","feed_headline":"Quantum error correction reduced to a single inner product","feed_subtitle":"A code is decodable for a noise space exactly when that space carries a special inner product; explicit error bases follow.","key_machinery":"The load-bearing object is the $C$-decoding inner product $\\varphi$ on the noise subspace $\\mathcal{N}$, defined by $\\eta N^* M \\rho = \\varphi(N,M)\\eta\\rho$ for all code states $\\eta,\\rho$; an orthonormal basis of $(\\mathcal{N},\\varphi)$ is precisely a $C$-decoding partial isometry error basis, meaning each $N_j P_C$ is a partial isometry onto $C$ and the ranges of distinct elements are orthogonal. The von Neumann-Wold decomposition, which writes any partial isometry as a direct sum of a unilateral-shift (completely non-unitary) part and a unitary part, is the tool that turns powers of one operator into such bases, with the wandering space $L = H \\ominus \\operatorname{ran} V$ controlling which subspaces are correctable. For cyclic unitary operators, shift and clock operators generate one-dimensional correctable codes spanned by rays with flat coefficients in the appropriate basis, and Proposition 5.2 assembles these bases across reducing subspaces by taking direct sums.","core_discovery":"The central result, Theorem 3.6, states that a code $C$ is an $\\mathcal{N}$-correcting code if and only if the noise subspace $\\mathcal{N}$ has a basis $\\{N_j\\}$ such that each $N_j P_C$ is a partial isometry with range $C$ and the ranges for distinct $j$ are orthogonal, where $P_C$ denotes the orthogonal projection onto $C$. This is equivalent to the existence of the $C$-decoding inner product $\\varphi$ on $\\mathcal{N}$, defined by $\\langle Nu,Mv\\rangle = \\varphi(N,M)\\langle u,v\\rangle$ for code vectors $u,v$, and both are equivalent to the standard Knill-Laflamme conditions. Given such a basis, the decoding channel is $\\Phi(\\rho)=\\sum_j P_C N_j^*\\rho N_j P_C$, completed by one extra Kraus operator that annihilates the noise on $C$, and $\\varphi(N,N)=\\operatorname{tr}(\\rho N^*N)$ gives the noise weight. The paper uses the von Neumann-Wold decomposition to build these bases from powers of a single partial isometry, the shift and clock operators for the cyclic unitary case, and direct sums over reducing subspaces for composite systems.","pith_inferences":["Editorial inference: because condition (3.9) is purely about operator ranges, it suggests a numerical search strategy for correcting codes that optimizes over $C$ to make a candidate basis of $\\mathcal{N}$ into partial isometries with orthogonal ranges, rather than constructing stabilizer structures first.","Editorial inference: the quotient by negligible noise in Remark 3.3 implies that physical error models with operators that vanish on the code should be described by equivalence classes of noise operators; a testable extension is to re-run the constructions for such quotient noise spaces and check whether the same explicit decoders emerge.","Editorial inference: the reducing-subspace theorem suggests heterogeneous multi-partite codes whose factors have different local dimensions, with the shortest period $m=\\min m_s$ governing the common correction capability; this could be tested by constructing direct-sum codes with unequal block sizes and comparing their performance with the paper's decoders.","Editorial inference: if the inner-product characterization is right, it may offer an alternative route to stabilizer codes by taking Weyl operators as the noise basis and asking which subspaces satisfy (3.9), potentially generating codes not captured by the usual stabilizer tabulation."],"forward_implications":["The dimension bound $\\dim \\mathcal{N}\\cdot\\dim C\\leq\\dim H$ forces a trade-off between code size and noise size, and rules out any $B(H)$-correcting code except in trivial dimension one (Corollary 3.7).","Once a $C$-decoding noise basis is found, the decoding channel is explicit: Kraus operators $K_j = P_C N_j^*$ for $j=1,\\ldots,n-1$ plus a single complementary operator $K_n$ that annihilates the noise on $C$ (Theorem 3.6).","For a unilateral-shift partial isometry $V$ with wandering space $(L,m)$, every $C_t = L \\ominus \\ker V^t$ is an $\\mathcal{N}_t$-correcting code for $\\mathcal{N}_t = \\operatorname{span}\\{V^j: j\\leq t\\}$, and when $t=m$ the code $L \\ominus \\ker V^m$ is the largest $\\mathcal{N}_m$-correcting code (Theorem 4.10).","For shift and clock operators, the correctable codes are exactly one-dimensional rays spanned by flat superpositions in the appropriate basis, with explicit decoders given in Corollaries 4.15, 4.16, and 5.8.","The family of $\\mathcal{N}$-correcting codes need not have a largest member: Example 4.13 exhibits a partial unitary operator where two orthogonal largest-correcting codes cannot be summed."],"supporting_citations":[{"why":"Supplies the Knill-Laflamme criteria that Theorem 3.4 and Corollary 3.5 rephrase as the $C$-decoding inner product condition.","marker":"[7]"},{"why":"Provides the von Neumann-Wold decomposition used in Theorem 4.6 to split partial isometries into unilateral-shift and unitary parts.","marker":"[17]"},{"why":"Gives the representation of inner products by positive invertible operators, used in Lemma 2.6 and Theorem 2.7 to connect inner products to operator theory.","marker":"[4]"}],"fun_headline_variants":["Inner products define quantum error-correcting codes","A single inner product governs code decodability","Quantum codes via inner products on noise spaces","QEC conditions from one inner product"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The chain of equivalences rests on the non-negligible-noise assumption that every nonzero noise operator has a nonzero effect on the code, $N P_C \\neq 0$, so that the decoding inner product is non-degenerate rather than merely a positive semidefinite form.","fun_headline_variants_meta":{"raw":{"variants":["Inner products define quantum error-correcting codes","A single inner product governs code decodability","Quantum codes via inner products on noise spaces","QEC conditions from one inner product"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000341,"raw_usage":{"total_tokens":1848,"prompt_tokens":881,"completion_tokens":967,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":497,"completion_tokens_details":{"reasoning_tokens":911}},"tokens_in":497,"tokens_out":967,"duration_ms":9730,"temperature":1.0,"reasoning_tokens":911,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:50:35.160764+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate all pairs $(C,\\mathcal{N})$ in a small Hilbert space, say $\\dim H = 3$, that satisfy $N P_C \\neq 0$ for every nonzero $N\\in\\mathcal{N}$, and compare the Knill-Laflamme condition (3.8) with the existence of a basis of $\\mathcal{N}$ satisfying (3.9); a single pair for which the first holds and the second fails would refute Theorem 3.6. The paper itself notes that without the assumption the statement degenerates: for $H=\\mathbb{C}^2$, $C=\\operatorname{span}\\{e_0\\}$, $\\mathcal{N}=\\operatorname{span}\\{I, |e_1\\rangle\\langle e_1|\\}$, the correction equation holds with $\\Phi=\\mathrm{id}$, but no $C$-decoding inner product exists because $|e_1\\rangle\\langle e_1|$ annihilates $C$.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Knill-Laflamme criteria that Theorem 3.4 and Corollary 3.5 rephrase as the $C$-decoding inner product condition."},{"cited_title":"MR 2760647","cited_arxiv_id":null,"evidence_quote":"Provides the von Neumann-Wold decomposition used in Theorem 4.6 to split partial isometries into unilateral-shift and unitary parts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the representation of inner products by positive invertible operators, used in Lemma 2.6 and Theorem 2.7 to connect inner products to operator theory."}],"review_version":1}