{"id":"06caa270-df23-44b4-a199-0f5222640584","arxiv_id":"2412.10318","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The bucket-brigade QRAM retains polylogarithmic query infidelity under arbitrary initialization, spatially correlated noise, and coherent noise, with a delayed Pauli twirling scheme restoring quadratic scaling.","lead":"Quantum random access memory (QRAM) is a proposed component that helps quantum computers read classical data, but it is hard to build with low errors. This paper proves that the bucket-brigade QRAM design stays accurate under more realistic noise such as imperfect initialization, correlated errors, and coherent errors, and it proposes a twirling method to suppress the worst coherent errors.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Coherent-noise bound (Theorem 6) rests on an unproved and likely mistyped inequality (Eq. 54); the step to Eq. 55 is not derived, leaving the O(n^4) scaling claim unsupported.","rationale":"The reader's weakest assumption was the noiseless address/bus registers, but that limitation adds only O(nε) infidelity and preserves the polylogarithmic scaling, so it is not the most load-bearing risk. The more serious gap is the proof of Theorem 6: the paper's extension to coherent errors is one of its three headline contributions, yet the proof relies on an unproved and likely mistyped inequality, Eq. (54), and an unexplained squaring of exponents in Eq. (55). The reader also flagged Eq. (54) in their rationale, which is why I partially agree. The arbitrary-initialization theorem (Theorem 4) is intricate but structurally sound, and the correlated-noise analysis, while approximate, introduces only O(ε^2 L) corrections that do not change the n-scaling. Thus the coherent-noise bound is the place where the central claim is least secure. A rigorous derivation or numerical verification for small n would resolve the concern. The appropriate verdict remains CONDITIONAL: the paper should not be fully accepted until Theorem 6 is substantiated. This does not change the reader's verdict, so I mark it UNCHANGED.","tokens_in":22790,"tokens_out":17685,"duration_ms":162545,"concrete_test":"Independently re-derive the proof of Theorem 6 for the special case K0 = exp(iκZ) (single-qubit coherent rotation), following the structure of Hann et al. Appendix D and tracking the non-Hermitian phase through the partial trace. If the resulting infidelity bound is not O((τ+1)^2(n+1)^2 ε), Theorem 6 is unsupported. Complement with a numerical simulation of a bucket-brigade QRAM for n = 2..6 with this noise applied to every router after every gate, using the GHZ address state of Eq. (49); if the infidelity does not scale as O(n^4 ε), the claim is falsified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that QRAM remains polylog-robust under arbitrary initialization, correlated, and coherent noise. The weakest link is the coherent-noise bound (Theorem 6), whose proof is a sketch that depends on Eq. (54). As printed, Eq. (54) is false: for near-identity K0, min_psi |Re<K0^{⊗n}>| is close to 1 and cannot be ≤ n^2 ε. The intended inequality is presumably 1 − min_psi |Re<K0^{⊗n}>| ≤ n^2 ε + O(κ^4), which holds for aligned states, but even then the jump to Eq. (55) — replacing n+1 and τ+1 by their squares in the exponent — is asserted without derivation. The non-Hermitian part V0 introduces phase correlations across routers and time steps that are not analyzed; the statement that 'the rest of the proof of Theorem 2 follows' is not credible because the proof in Hann et al. relies crucially on K0 being Hermitian. If this gap cannot be filled, the claim of maintaining polylogarithmic scaling for coherent errors is unverified, and the motivation for the twirling section weakens. The noiseless address/bus assumption, by contrast, would only add O(nε) infidelity and does not threaten the polylog scaling, so it is not the most load-bearing issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the bucket-brigade QRAM under generalized noise models. It claims: (i) arbitrary initialization of the routing tree does not destroy polylogarithmic infidelity scaling, with a query-doubling protocol making reset protocols unnecessary (Theorem 4, Corollary 4.1); (ii) spatially correlated quasi-local noise preserves polylogarithmic scaling via a cluster expansion and coarse graining (Section III B); (iii) arbitrary single-router coherent noise yields an O(n^4) infidelity bound (Theorem 6); and (iv) a delayed Pauli twirling scheme restores O(n^2) scaling (Theorem 7). The paper also reviews the bucket-brigade architecture and the prior Hann et al. result, and discusses the implications for reset-free and twirled QRAM designs.","tokens_in":23011,"tokens_out":12366,"duration_ms":117780,"significance":"If the theorems are correct, the paper meaningfully extends the known noise resilience of bucket-brigade QRAM: it supports reset-free operation, robustness to crosstalk-like correlated errors, and a concrete twirling scheme that exploits the compute-uncompute symmetry. The structural observations about branch-permutation symmetry and wait-state confinement are valuable and go beyond a simple restatement of [10]. The paper also gives credit where due by building on a published external benchmark rather than circularly reducing to its own assumptions. However, the proof of the coherent-noise extension is incomplete, and the correlated-noise and twirling results are only sketched; the central claims therefore need additional rigorous support before the paper can be accepted.","major_comments":[{"comment":"The proof of Theorem 6 is not valid as written. Equation (54) has the wrong inequality direction: for K0 = exp(iκZ) with small real κ, Definition 2 gives ε = 1 - cos²κ ≈ κ², while min_ψ |Re⟨K0^{⊗n}⟩| = |cos κ|^n ≈ 1 - nκ²/2. For n=2 and κ=0.1 this gives LHS ≈ 0.995 and RHS ≈ 0.04, so the printed inequality cannot hold. The intended statement is presumably a lower bound of the form min_ψ |Re⟨K0^{⊗n}⟩| ≥ 1 - O(n²ε), but even with that correction the inference to Eq. (55) is not derived. The non-Hermitian part V0 introduces phase correlations across routers and time steps that are absent in the Hermitian-K0 proof of Theorem 2 in [10], so the statement 'the rest of the proof of Theorem 2 follows' is not a substitute for a derivation. Footnote [30] itself notes that the non-Hermitian case in [10] required ε to scale inversely with n; Theorem 6 claims to overcome this, and the proof must show explicitly why that restriction is removed. Because Theorem 6 is the basis for the coherent-noise O(n^4) claim and for the motivation of Section IV, this gap is load-bearing.","section":"Section III C, Eqs. (54)-(55)"},{"comment":"The cluster expansion for correlated noise is an uncontrolled approximation. The text states that 'to order O(ε²), we can commute the noise channels' and then uses the resulting ordering in Eq. (45), but no bound on the commutator remainder is provided. Similarly, Eq. (46) asserts additivity of infidelities across coarse-grained trees without a proof, and Eq. (48) defines ε_d with an ambiguous summation index ('Et' is not defined). Since the polylogarithmic scaling for spatially correlated errors is one of the paper's headline results, these steps need to be turned into a rigorous inequality or explicitly identified as a conjecture.","section":"Section III B, Eqs. (43)-(47)"},{"comment":"Theorem 7 is stated without a proof. The paragraphs around Algorithm 1 and the discussion of nested twirls give a plausible mechanism, but they do not demonstrate that the delayed twirling operators, which must be correlated between the two queries to avoid extra copying, produce a Pauli channel with the claimed constants, nor that the in-situ SWAP corrections or classical memory reshuffling do not introduce un-twirled errors at a level that alters the scaling. Equations (61) and (62) are therefore unsupported. As the main suppression result, this requires a detailed derivation.","section":"Section IV B, Theorem 7"}],"minor_comments":[{"comment":"The displayed equality E_χ[|⟨ψout|Π(Vχ)|ψout⟩|²] = E_χ[∥Π(Vχ)|ψout⟩∥] is dimensionally inconsistent; it should presumably involve ∥Π(Vχ)|ψout⟩∥² or an equivalent expression.","section":"Section II, Eq. (24)"},{"comment":"Eq. (59) is an empty numbered display; please remove the placeholder or fill in the intended statement.","section":"Section IV A, around Eq. (59)"},{"comment":"The column headings and row entries are confusing, e.g., 'Two-level Yes No No O(n^6)' and 'Either No No Yes O(n^6)'; please define all combinations and explain how the entries follow from the theorems.","section":"Table I"},{"comment":"The notation 'T ∈ P2 and M = CXP†CX ∈ P2' is not defined; please spell out P2 and the action of CX.","section":"Section IV B, Algorithm 1"},{"comment":"The summation over 'Et' should be over time steps and/or error clusters; please make the index explicit.","section":"Section III B, Eq. (48)"},{"comment":"The phrase 'under mixed-unitary noise, so does the infidelity' is not defined; the theorem itself is stated for Bernoulli noise, so this sentence should be clarified.","section":"Section III A, proof of Theorem 4"}],"recommendation":"major_revision","confidential_remarks":"The coherent-noise proof is the make-or-break point. If the authors can correct Eq. (54) and provide a complete derivation of Eq. (55) and of Theorem 7, the paper would be a solid contribution. I would not require new numerics, but the proof sketches in Sections III B and IV B need to be made rigorous. The self-citation to [10] is appropriate and does not by itself raise concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real extension of the Hann et al. QRAM resilience program, and the arbitrary-initialization and correlated-noise results look solid. But the coherent-noise theorem (Theorem 6) has a hole in the proof as written: Eq. (54) is either mistyped or false for near-identity K0, and the jump to Eq. (55) is asserted without derivation. That part needs work before the headline claim about polylog scaling under coherent noise can be trusted.\n\nWhat's actually new: Theorem 4/Cor 4.1 for arbitrary router initialization (including qudits) is a clean advance and the query-doubling argument is plausible. Section III.B's coarse-graining construction for quasi-local correlated noise is clever and the error-rate rescaling argument is a reasonable extension of the prior framework. The delayed-twirling idea in Section IV is genuinely QRAM-specific and addresses a real gap in randomized compiling.\n\nThe paper is honest about leaning on [10]—the authors cite it throughout, and its published status makes that acceptable. The reuse of the same definitions and counting argument is not a problem per se; the new work is in the extensions.\n\nWhere it's soft: Theorem 6 is the load-bearing piece for the coherent-noise claim, and the proof sketch is too compressed. As printed, (54) has the inequality going the wrong way—min_psi |Re<K0⊗n>| is close to 1 for small kappa, not bounded above by n^2 epsilon. The intended bound is presumably on 1 − min_psi |...|, but that still doesn't get you to (55) without analyzing the non-Hermitian part V0 and its phase correlations across routers and time steps. The proof of Theorem 2 in Hann et al. relies on K0 being Hermitian, so 'the rest follows' is not credible as stated. The correlated-noise section also uses an O(epsilon^2) commutation approximation without fully justifying it, and Theorem 7's proof is more of a sketch. These are fixable gaps, but they're real. The noiseless address/bus assumption is actually a minor issue—adding O(n epsilon) would not break the polylog scaling.\n\nBottom line: this deserves a serious referee. The arbitrary-init and correlated-noise results are likely correct, and the twirling protocol is worth engaging. The coherent-noise theorem needs a complete proof or a numerical check before the paper's central claim is accepted.","headline":"Solid extensions of QRAM noise-resilience, but the coherent-noise theorem has a proof gap that must be fixed.","tokens_in":23580,"tokens_out":2834,"would_cite":true,"duration_ms":631499,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves bucket-brigade QRAM keeps polylogarithmic error growth under arbitrary router initialization, spatially correlated noise, and coherent errors, with a twirling protocol that restores the tighter quadratic scaling.","keywords":["bucket-brigade QRAM","noise resilience","coherent errors","quantum random access memory","Pauli twirling","initialization errors","correlated noise","query fidelity"],"falsifier":"Simulate or measure a two-level bucket-brigade QRAM with routers initialized to $\\left|+\\right\\rangle^{\\otimes N}$, no resets between queries, and per-router coherent rotations $e^{i\\kappa Z}$ for fixed small $\\kappa$; the paper predicts $1-F$ scales as $O(n^4)$, so observing exponential or super-polynomial growth in $n$ would falsify the resilience claim. Including comparable noise on the address and bus qubits should make the stated bounds fail, since the theorems do not cover it.","tokens_in":22516,"feed_emoji":"⚡️","tokens_out":9746,"duration_ms":86202,"temperature":0.7,"pith_summary":"Quantum random access memory in the bucket-brigade architecture routes a query through a binary tree of small quantum routers, and earlier work showed that when errors hit individual routers, the query infidelity grows only polylogarithmically with memory size. This paper tries to establish that this resilience is not an accident of a narrow noise model: it survives arbitrary router initialization, spatially correlated noise over router clusters, and coherent (unitary) errors, with infidelity still polylogarithmic in all three cases. The price of coherent errors is a quadratic worsening of the bound, from roughly $O(n^2)$ to $O(n^4)$. The paper also proposes a delayed Pauli twirling scheme that exploits the query circuit’s compute–uncompute symmetry to turn coherent errors into stochastic Pauli noise and restore the $O(n^2)$ scaling. If right, these results mean QRAM hardware need not include per-router reset or measurement, simplifying near-term architectures.","feed_headline":"QRAM error growth stays polylog under realistic noise","feed_subtitle":"New proofs cover arbitrary router initialization, correlated errors and coherent errors; a twirling fix restores the tighter bound.","key_machinery":"The central object is the mutually coherent subspace $V'_\\chi$ of address states that remain both successful and mutually coherent for a given error configuration $\\chi$; the proof shows the noisy channel maps this subspace disjointly from its orthogonal complement, forcing the reduced output state to be rank one on that block (the analogue of Eq. 17). Two further mechanisms carry the argument: the time-reversal symmetry $V_t = V_{\\tau-t}$ of the query circuit, which lets a twirling correction applied at time $t$ be delayed to its conjugate partner at time $\\tau-t$, and a coarse-graining map that turns a connected cluster of corrupted routers into a single local error on a coarser tree, which is how correlated noise is handled.","core_discovery":"On the paper’s own terms, the bucket-brigade QRAM’s natural noise resilience, previously proved for local incoherent single-router noise, extends to three practically important classes of error. Theorem 4 and Corollary 4.1 show that even when routers are initialized to an arbitrary state, possibly mixed and of any dimension, a two-level QRAM using the doubled query circuit $Q' = Q\\,\\mathrm{CX}_{B,B'}\\,Q$ obeys $1-F \\le 4\\varepsilon(\\tau+1)(n+1)^2$, so resetting the tree between queries is unnecessary. Section III B shows that quasi-local spatially correlated errors preserve the polylogarithmic scaling after a coarse-graining of the tree, with only an $L$-dependent rescaling of the error rate. Theorem 6 generalizes the bound to arbitrary single-router noise, including coherent errors, giving $1-F \\le A\\varepsilon(\\tau+1)^2(n+1)^2 \\in O(n^4)$. Section IV’s delayed twirling protocol then tailors coherent noise to stochastic Pauli noise by using the circuit’s time-reversal symmetry, restoring $O(n^2)$ infidelity scaling.","pith_inferences":["A direct experimental test is to operate a two-level QRAM with routers left in random states and compare query fidelity with and without a reset step; the paper predicts no reset penalty at polylog order, a claim measurable on small devices.","If reset-free operation works, the dominant remaining error source will be address- and bus-register noise, which this paper explicitly brackets; extending the proof to those registers is the natural next step for the architecture.","The delayed-twirling idea should transfer to any circuit with compute–uncompute symmetry and non-Clifford gates, offering a route to randomized compiling beyond the Pauli-Clifford setting.","The $n^2 \\to n^4$ coherent-error penalty implies that QRAM error budgets in fault-tolerant designs should either twirl or otherwise decorrelate unitary errors before allocating error-correction overhead."],"forward_implications":["A QRAM can be queried repeatedly without resetting or measuring the router tree between queries; Corollary 4.1 says any qudit initialization, mixed or pure, retains the same polylog bound, removing a major hardware requirement.","Spatially correlated noise that is quasi-local, meaning supported on connected clusters of bounded size $L$, only rescales the effective error rate, so crosstalk and miscalibrated multi-router gates do not break the architecture’s resilience unless the correlations are long-range.","Coherent errors are the least favorable of the extended classes: they raise the infidelity bound from $O(n^2)$ to $O(n^4)$, so unitary error buildup must be tracked separately from stochastic noise.","The delayed twirling protocol converts coherent errors into stochastic Pauli channels and restores $O(n^2)$ scaling, or $O(n^3)$ when classical memory reshuffling replaces SWAP corrections, making the tighter bound available without full error correction."],"supporting_citations":[{"why":"Supplies the base resilience theorem for local incoherent single-router noise and the proof techniques that Sections III and IV extend.","marker":"[10]"},{"why":"Introduces the bucket-brigade QRAM architecture and the routing operation that is the object of all the bounds.","marker":"[23]"},{"why":"The earlier challenge to bucket-brigade noise resilience that the theorems here are designed to settle.","marker":"[31]"},{"why":"Raises the reset-requirement concern that Corollary 4.1 addresses by showing arbitrary router initializations are tolerable.","marker":"[13]"},{"why":"Provides the Pauli twirling formalism used in Section IV to convert coherent errors to stochastic Pauli noise.","marker":"[29]"},{"why":"Supplies randomized-compiling and twirling-group background that the paper argues does not directly apply to QRAM, motivating the delayed twirling protocol.","marker":"[33]"},{"why":"Pseudo-twirling scheme for non-Clifford gates referenced as an alternative and extended by the paper’s edge-twirling variant.","marker":"[38]"}],"fun_headline_variants":["Bucket-brigade QRAM keeps polylog error under realistic noise","QRAM noise resilience extends to correlated and coherent errors","No reset needed: QRAM queries robust to initialization errors","Twirling restores tighter error bound for QRAM with coherent noise","QRAM polylog error survives spatially correlated noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central bounds assume that the address and bus qubits are noiseless throughout the circuit, so the polylogarithmic resilience is proven only for errors on the router qutrits; comparable noise on the address or bus registers would add an extra contribution not bounded by these theorems.","fun_headline_variants_meta":{"raw":{"variants":["Bucket-brigade QRAM keeps polylog error under realistic noise","QRAM noise resilience extends to correlated and coherent errors","No reset needed: QRAM queries robust to initialization errors","Twirling restores tighter error bound for QRAM with coherent noise","QRAM polylog error survives spatially correlated noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00073,"raw_usage":{"total_tokens":3294,"prompt_tokens":995,"completion_tokens":2299,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":2215}},"tokens_in":611,"tokens_out":2299,"duration_ms":14815,"temperature":1.0,"reasoning_tokens":2215,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:57:28.951003+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate or measure a two-level bucket-brigade QRAM with routers initialized to $\\left|+\\right\\rangle^{\\otimes N}$, no resets between queries, and per-router coherent rotations $e^{i\\kappa Z}$ for fixed small $\\kappa$; the paper predicts $1-F$ scales as $O(n^4)$, so observing exponential or super-polynomial growth in $n$ would falsify the resilience claim. Including comparable noise on the address and bus qubits should make the stated bounds fail, since the theorems do not cover it.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the base resilience theorem for local incoherent single-router noise and the proof techniques that Sections III and IV extend."},{"cited_title":"Asaka, K","cited_arxiv_id":null,"evidence_quote":"The earlier challenge to bucket-brigade noise resilience that the theorems here are designed to settle."},{"cited_title":"Sch¨ utzhold, Phys","cited_arxiv_id":null,"evidence_quote":"Raises the reset-requirement concern that Corollary 4.1 addresses by showing arbitrary router initializations are tolerable."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Pauli twirling formalism used in Section IV to convert coherent errors to stochastic Pauli noise."},{"cited_title":"Error-Mitigated Quantum Random Access Memory","cited_arxiv_id":"2403.06340","evidence_quote":"Pseudo-twirling scheme for non-Clifford gates referenced as an alternative and extended by the paper’s edge-twirling variant."}],"review_version":1}