{"id":"8fd73c7b-02c0-4c9a-83b7-68a7b49d3ad3","arxiv_id":"2508.20647","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A group-theoretic construction yields multimode rotationally symmetric bosonic codes with linear-optical Pauli gates, and two-mode binomial instances that correct correlated dephasing exactly while improving dephasing performance without sacrificing loss correction.","lead":"This paper constructs a new family of multi-mode bosonic codes whose logical Pauli gates are implemented by beam splitters, and it analyzes two-mode binomial instances under photon loss and dephasing. It reports that these codes correct correlated dephasing exactly and, unlike single-mode rotationally symmetric codes, avoid a trade-off between dephasing and loss protection for the studied parameters.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"N=4 dephasing 'KL satisfied' claim is contradicted by Appendix E; first-order KL needs re-derivation.","rationale":"The reader identified the N=2 dephasing KL violation and the reliance on numerical SDP, but did not flag the internal inconsistency between the main-text claim of KL satisfaction for N=4 and Appendix E's demonstration that the diagonal KL entries differ for finite N. This is load-bearing because the N=4 analytic recovery is the key non-numerical support for the no-trade-off claim. If the KL claim is false, the paper's central quantitative assertion is not established by the analytic arguments and depends entirely on the unreleased numerical pipeline. The paper is otherwise coherent: the group-theoretic construction, the correlated-dephasing correction circuit in Appendix G, and the loss recovery maps appear internally consistent. The concern is not an ad hominem or a disagreement with consensus; it is a concrete internal inconsistency that can be settled by a direct KL matrix calculation. The verdict remains CONDITIONAL because, even if the KL claim fails, the numerical performance figures could still be correct, and the correlated-dephasing exactness is independent. The condition is to correct the KL statements and provide the numerical pipeline or a corrected analytic proof.","tokens_in":24994,"tokens_out":34427,"duration_ms":286452,"concrete_test":"Independently compute the first-order KL matrix for N=4 dephasing using the codewords in Eqs. (C28)-(C29) and the first-order Kraus operators E0 = I - (γ1/2)(b1†b1)² - (γ2/2)(b2†b2)², E1 = √(γ1)b1†b1, E2 = √(γ2)b2†b2 at δ=π/4, ϕ=π/(2N). Evaluate the 2×2 matrices ⟨i|Ea†Eb|j⟩ for i,j,a,b∈{0,1}. If ⟨0|Ea†Eb|1⟩ ≠ 0 or ⟨0|Ea†Ea|0⟩ ≠ ⟨1|Ea†Ea|1⟩ for any a, the claimed KL satisfaction is false and the Appendix D recovery proof must be re-examined.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central no-trade-off claim rests on the analytic dephasing recovery for K=2, N=4. The main text and Appendix D state that, for optimal angles δ=π/4,3π/4, the Knill-Laflamme conditions are satisfied for dephasing up to first order. However, Appendix E computes the KL matrix for the continuous dephasing channel and explicitly finds that the diagonal entries are unequal for finite N (Eq. E9), with equality only in the limit N→∞ or for correlated dephasing θ1=θ2. This is an internal contradiction: if the diagonal KL entries differ, the KL conditions cannot be satisfied for finite N, so the first-order recovery proof in Appendix D requires an additional error-distinguishing measurement (the P_L/P_E measurement in Eqs. D17-D22). That measurement is not part of the standard KL framework and its verification in Eqs. D23-D26 relies on a decomposition (Eq. D1) whose operators M0^(1), M1, M2 have noise-parameter scaling that appears inconsistent with a first-order expansion: they are O(γ), making their squares O(γ²), whereas the first-order channel contains O(γ) jump terms. If the N=4 analytical dephasing recovery is not valid, the main quantitative evidence for 'no trade-off' reduces to the unreproducible SDP numerics, weakening the abstract's unqualified claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a family of multimode rotationally symmetric bosonic codes constructed by inverting the group-theoretic framework of Denys and Leverrier: one first chooses logical Pauli gates implemented by passive linear optics and then derives codewords. The main instances studied are two-mode binomial codes with codewords given by Eqs. (6)-(7), for which the paper claims improved dephasing protection and, unlike single-mode RSB codes, no trade-off between dephasing and loss protection. The paper also claims exact correction of correlated dephasing via a four-mode circuit, and qudit encoding in arbitrary dimensions. The analytical results include first-order recovery maps for loss (N=2,4) and dephasing (N=4), numerical SDP benchmarks showing dephasing infidelity decreasing with N, and an appendix proving the correlated-dephasing circuit.","tokens_in":25231,"tokens_out":22270,"duration_ms":191131,"significance":"If the claims are valid, the code family is a meaningful step for bosonic QEC: it preserves linear-optics logical Pauli gates while improving dephasing resilience, and the exact correction of correlated dephasing is a clean, testable result. The group-theoretic inversion is a promising design principle, and the two-mode binomial instances are explicit enough for experimental follow-up. The analytical structure (Appendices A-D) and the correlated-dephasing proof (Appendix G) are the main strengths. However, several load-bearing analytical statements are internally inconsistent or insufficiently specified, and the qudit generalization appears to have a group-representation flaw. These issues need to be resolved before the central claims can be accepted.","major_comments":[{"comment":"The claim that the N=4 two-mode code satisfies the Knill-Laflamme conditions for dephasing up to first order is inconsistent with the recovery map actually presented. If the KL conditions held for the full set of first-order error operators, the standard recovery would consist of a projective syndrome measurement followed by a single unitary per syndrome; the additional projection P_L/P_E introduced in Eqs. (D17)-(D22) would be unnecessary. Its presence indicates that the (0,0)-syndrome error operators are not proportional on the code, so the KL conditions are not satisfied as stated. The authors should either prove the KL conditions for the complete error set or revise the main-text claim to state that the errors are correctable via a syndrome measurement augmented by a second-level measurement.","section":"Main text, paragraph after Eq. (7); Appendix D, Eqs. (D1)-(D26)"},{"comment":"The operators M0^(1) and M0^(2) are defined with scaling γ_i t, e.g. M0^(1)=γ1 t(a1†a1 cos²δ+a2†a2 sin²δ). With this scaling, the terms M0^(1) ρ M0^(1)† in Eq. (D1) are of order (γt)², whereas the first-order dephasing jump terms in Eq. (B14) are of order γt. The decomposition in Eq. (D1) therefore does not match the first-order channel expansion used in the verification (D23)-(D26). The correct first-order jump operators should carry √(γt) prefactors. This is not a mere notational slip: the recovery verification is the only analytical basis for the N=4 dephasing claim, so the derivation must be redone with consistent scaling.","section":"Appendix D, Eqs. (D1)-(D4)"},{"comment":"The claimed qudit encoding in arbitrary dimensions is not supported by the group-theoretic construction as written. For d>2, π(h)=U_BS exp(i2π/(Nd) a1†a1) U_BS† satisfies π(h)^{2N}=exp(i4π/d a1†a1), which is not the identity operator on the physical Hilbert space. Hence π is not a representation of the stated group G=⟨g,h|h^{2N}=e⟩. For d=2 the relation holds (since 4π/d=2π), which explains why the qubit examples work, but the arbitrary-dimension claim requires either redefining the group as ⟨g,h|h^{Nd}=e⟩ (for which π(h)^{Nd}=I) or explicitly stating that π is only required to reproduce the logical action on the codespace rather than being a full group representation.","section":"Abstract; Appendix A, Eqs. (1)-(3)"},{"comment":"The statement that the codes exhibit 'no trade-off between protection against dephasing and photon loss' is too strong. Fig. 2(b) shows the loss performance has an optimum near N=4 and degrades for larger N, similar to the single-mode case; the dephasing performance improves up to N=8 in Fig. 2(a). Thus the improvement is a removal of the dephasing side of the trade-off in the studied range, not a removal of the trade-off itself. The claim should be qualified to the range where loss performance is comparable to the single-mode code.","section":"Abstract and Conclusions; Fig. 2"}],"minor_comments":[{"comment":"The statement 'It can be verified that the Knill-Laflamme conditions for the loss channel, up to first order, are satisfied' omits the verification entirely. Since the N=4 loss recovery is one of the paper's main analytical results, the calculation should be shown or at least summarized.","section":"Appendix C, subsection 'K=2, N=4 binomial encoding'"},{"comment":"The projection P_E is defined using an unspecified state |E⟩, and the unitary in Eq. (D19) depends on it. The recovery map is therefore not completely specified. Please define |E⟩ explicitly and show that the states reachable by the first-order dephasing errors are in the support of P_L or P_E as claimed.","section":"Appendix D, Eq. (D18)"},{"comment":"No details are given for the SDP optimization (Hilbert-space truncation, solver, convergence tolerance), and no code is provided. Given that the dephasing results for N=6 and N=8 rest entirely on the numerics, please include these details or a link to a reproducible implementation.","section":"Fig. 2 and numerical methods"},{"comment":"There are several typographical and formatting artifacts, e.g. the garbled axis labels in Fig. 2 ('0 - 2 - 4t w o - m o d e') and inconsistent uses of 'Knill-Lafflame' versus 'Knill-Laflamme'. A careful proofreading pass is needed.","section":"Throughout the text"},{"comment":"The discussion around Eq. (E9) correctly notes that the diagonal KL entries for the continuous dephasing channel are unequal for finite N, with equality only for N→∞ or correlated dephasing. To avoid confusion with the first-order KL claim, please state explicitly that this inequality concerns the full (non-truncated) channel and does not by itself contradict a first-order KL analysis.","section":"Appendix E, Eq. (E9)"}],"recommendation":"major_revision","confidential_remarks":"The paper contains several promising ideas and a clean result for correlated dephasing, but the analytical core for the headline 'no trade-off' claim needs careful rework. The N=4 dephasing recovery has a scaling inconsistency and an unexplained extra measurement; the qudit generalization has a group-representation issue. These are fixable in a revision, but as written they prevent acceptance. I would also encourage the authors to release the SDP code and the N=4 loss KL calculation to support reproducibility."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a serious entry in bosonic QEC, and the referee should engage with it, but the analytical dephasing section needs a fix before the main claim is fully supported.\n\nWhat's genuinely new: the multimode RSB code family from the group-theoretic construction, with the beam-splitter degree of freedom, the qudit codewords Eq. (3), and the two-mode binomial instances Eqs. (6)-(7). The exact correction of correlated dephasing in Appendix G is a clean, independent result; the circuit commutation argument checks out. The loss recovery maps for N=2 and N=4 are also plausible; the KL checks for loss are straightforward and I didn't see a problem.\n\nThe main soft spot is Appendix D. The text says KL conditions are satisfied for N=4 dephasing at optimal angles, but the operators M0^(1), M1, M2 are defined with an extra factor of the noise strength (they scale as γ, not √γ). As written, the first-order channel expansion and the recovery verification D23-D26 don't match: terms that are O(γ²) are being used to correct an O(γ) channel. This looks like missing square roots rather than a fatally wrong idea, and the KL checks themselves are scale-invariant, but the recovery proof as printed is not rigorous. The numerical SDP results are then doing more work than the analytic section admits. A second related point: the abstract's \"no trade-off\" phrasing drops the \"up to optimal N\" qualifier that appears in the main text; that should be fixed.\n\nThe stress-test note claims Appendix E contradicts the first-order KL claim, but I don't think that's right: Appendix E is about the full channel at arbitrary dephasing strength, and its diagonal inequality does not directly invalidate first-order KL. The contradiction is not the real problem; the scaling inconsistency is.\n\nLesser issues: no code, data, or solver details for the SDP benchmarks, so the numerics aren't reproducible from the manuscript. One KL verification (N=4 loss) is omitted with \"It can be verified.\" For N=2, the paper is honest that dephasing KL fails and the improvement is numerical; that's fine.\n\nBottom line: the code family and correlated-dephasing correction are worth having. The independent-dephasing no-trade-off claim should be conditional until Appendix D is re-derived and the numerics released. I would send this to peer review with a request for major revision.","headline":"Solid group-theoretic code family with a clean correlated-dephasing result, but the analytic dephasing recovery needs re-derivation and the numerics need release before the no-trade-off claim is fully supported.","tokens_in":25856,"tokens_out":29896,"would_cite":true,"duration_ms":250637,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper introduces multimode rotationally symmetric bosonic codes in which, up to the optimal rotational order, dephasing protection improves without sacrificing photon-loss protection, and which correct correlated dephasing exactly.","keywords":["bosonic codes","rotational symmetry","quantum error correction","multimode codes","binomial codes","linear optics","dephasing noise","qudits"],"falsifier":"Compute the optimal-recovery entanglement infidelity for the $N=4$, $K=2$ code with the optimal angles under pure dephasing with unequal mode rates $\\gamma_1\\neq\\gamma_2$; if the infidelity no longer improves over the single-mode code as $N$ grows, the no-trade-off claim is tied to the equal-rate local noise model. For the $N=2$ case, a closed-form recovery cannot exist at first order because of the Knill-Laflamme violation in Appendix D, so any claimed dephasing advantage should be checked against the numerical semi-definite program at stronger dephasing strengths.","tokens_in":24728,"feed_emoji":"🛡️","tokens_out":8455,"duration_ms":74632,"temperature":0.7,"pith_summary":"The paper aims to establish that rotationally symmetric bosonic codes do not have to choose between protecting against photon loss and protecting against dephasing. It constructs a family of multimode codes by a group-theoretic prescription that starts from the logical Pauli gates one wants to implement with linear optics, and shows that in the two-mode binomial instances the dephasing performance improves as the rotational order $N$ grows while the loss performance stays comparable to the single-mode code. It further claims that the entire two-mode family corrects correlated dephasing noise exactly. A sympathetic reader would care because this removes a known design trade-off and supplies bosonic qudits whose Pauli gates are simple passive-linear operations.","feed_headline":"Two-mode bosonic codes erase the dephasing-loss trade-off","feed_subtitle":"In the new two-mode binomial codes, dephasing protection grows with N while loss performance stays put.","key_machinery":"The machinery is a group-theoretic code construction: choose the group $G=\\langle g,h\\,|\\,h^{2N}=e\\rangle$, let the logical representation send $g$ to $X$ and $h$ to $Z$, and build the physical representation from passive linear optics—a beam-splitter network $\\hat{U}_{\\mathrm{BS}}$ followed by rotations such as $\\pi(h)=\\hat{U}_{\\mathrm{BS}}\\exp(i\\frac{2\\pi}{Nd}\\hat{a}_1^\\dagger\\hat{a}_1)\\hat{U}_{\\mathrm{BS}}^\\dagger$. The new element is the mode-mixing unitary $\\hat{U}_{\\mathrm{BS}}$, which rotates the noise operators into a basis where dephasing acts partly through generators $\\hat{G}^{\\pm}$ rather than only through individual photon numbers. The codewords are superpositions of Fock states in which each mode has only levels at multiples of $N$, cyclically shifted across the $d$ modes, so the full Pauli group is realized by beam splitters and rotations. Recovery uses modular number measurements followed by correction unitaries; the exact correlated-dephasing correction uses controlled-$X$ gates of the form $\\exp(i\\frac{\\pi}{2N}\\hat{n}_1\\otimes\\hat{G}^-_{34})$.","core_discovery":"The central claim is that adding a second mode, together with a tunable beam-splitter rotation, converts the single-mode rotationally symmetric binomial code into a code with simultaneous resistance to loss and dephasing. Concretely, for the two-mode $K=2$ code with codewords $|0_N\\rangle=\\hat{U}_{\\mathrm{BS}}\\frac{1}{\\sqrt{2}}(|0\\rangle+|2N\\rangle)\\otimes|N\\rangle$ and $|1_N\\rangle=\\hat{U}_{\\mathrm{BS}}|N\\rangle\\otimes\\frac{1}{\\sqrt{2}}(|0\\rangle+|2N\\rangle)$, choosing the beam-splitter angles $\\delta=\\pi/4$ or $3\\pi/4$ and $\\phi=\\pi/(2N)$ makes the Knill-Laflamme conditions hold to first order for both loss and dephasing when $N=4$, while for $N=2$ only loss admits the closed-form first-order recovery and dephasing is handled by numerically optimized recovery. The paper also proves that any code of the form in Eqs. (4)-(5) corrects arbitrary correlated dephasing of the form in Eq. (11) exactly, using the four-mode circuit in Fig. 3, and that the same construction encodes qudits of any dimension $d$ in $d$ modes. In the paper's own terms, this means the single-mode trade-off between dephasing and loss protection is resolved, up to the optimal value of $N$.","pith_inferences":["Because the advantage is demonstrated for independent equal-rate local noise, an immediate test is asymmetric rates: if the dephasing improvement persists when $\\gamma_1\\neq\\gamma_2$ or $\\kappa_1\\neq\\kappa_2$, the construction is more robust than the paper's stated model.","The exact correlated-dephasing argument relies on commutation of the total photon number with the mode-mixing generators; the same commutation suggests the circuit may extend to correlated loss or to more than two modes, though the paper does not claim this.","The torus phase-distribution picture implies the beam-splitter angles are a tunable resource: optimizing them for a specific noise channel could yield further gains, and the same idea might transfer to translationally symmetric bosonic codes."],"forward_implications":["For the two-mode $K=2$ binomial codes at optimal angles, dephasing infidelity decreases as $N$ increases up to an optimal value, while loss performance remains comparable to the corresponding single-mode code; the single-mode trade-off is absent in this regime.","The full Pauli group on the encoded qubit or qudit is implemented by passive linear optics, so logical $X$ and $Z$ are available without auxiliary qubits.","Any two-mode RSB code in the family corrects arbitrary correlated dephasing of the form in Eq. (11) exactly, with a fixed four-mode circuit independent of the order $N$ and the encoding angles.","The construction supports qudit encoding in arbitrary dimension $d$, with order-$N$ rotational symmetry in each of $d$ modes.","The Hadamard, $S$, $T$, and $CZ$ gates can be implemented with Kerr interactions and gate teleportation, so the code family is compatible with a useful non-universal gate set."],"supporting_citations":[{"why":"Supplies the single-mode rotationally symmetric binomial codes and the loss-dephasing trade-off that the paper's two-mode codes are designed to beat.","marker":"[13]"},{"why":"Supplies the group-theoretic construction, inverted as code-from-gates, that this work extends by adding mode-mixing freedom.","marker":"[14]"},{"why":"Supplies the general framework of binomial bosonic codes and the Knill-Laflamme and recovery-map techniques used in the appendices.","marker":"[15]"},{"why":"The two-mode CLY code used as the earlier multimode baseline in the numerical comparison.","marker":"[26]"},{"why":"Supplies the semi-definite programming method used to find the optimal recovery maps in the numerical benchmarks.","marker":"[27]"},{"why":"Supplies the phase-measurement distinguishability and random-telegraph-noise analysis that the paper extends to two-mode codes.","marker":"[28]"}],"fun_headline_variants":["Two-mode bosonic codes break the dephasing-loss trade-off","Group-theoretic bosonic codes: no dephasing-loss trade-off","Multi-mode bosonic codes erase the dephasing-loss trade-off","Linear-optics bosonic codes with exact dephasing correction","Bosonic qudit codes: arbitrary dimension via linear optics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the noise is well described by independent, Markovian, equal-rate loss and dephasing in each mode, and that correcting to first order in the noise strengths is enough to establish the performance advantage.","fun_headline_variants_meta":{"raw":{"variants":["Two-mode bosonic codes break the dephasing-loss trade-off","Group-theoretic bosonic codes: no dephasing-loss trade-off","Multi-mode bosonic codes erase the dephasing-loss trade-off","Linear-optics bosonic codes with exact dephasing correction","Bosonic qudit codes: arbitrary dimension via linear optics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001168,"raw_usage":{"total_tokens":4870,"prompt_tokens":1024,"completion_tokens":3846,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":640,"completion_tokens_details":{"reasoning_tokens":3756}},"tokens_in":640,"tokens_out":3846,"duration_ms":25145,"temperature":1.0,"reasoning_tokens":3756,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:42:31.464950+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the optimal-recovery entanglement infidelity for the $N=4$, $K=2$ code with the optimal angles under pure dephasing with unequal mode rates $\\gamma_1\\neq\\gamma_2$; if the infidelity no longer improves over the single-mode code as $N$ grows, the no-trade-off claim is tied to the equal-rate local noise model. For the $N=2$ case, a closed-form recovery cannot exist at first order because of the Knill-Laflamme violation in Appendix D, so any claimed dephasing advantage should be checked against the numerical semi-definite program at stronger dephasing strengths.","supporting_citations":[{"cited_title":"Denys and A","cited_arxiv_id":null,"evidence_quote":"Supplies the group-theoretic construction, inverted as code-from-gates, that this work extends by adding mode-mixing freedom."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the general framework of binomial bosonic codes and the Knill-Laflamme and recovery-map techniques used in the appendices."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The two-mode CLY code used as the earlier multimode baseline in the numerical comparison."}],"review_version":2}