{"id":"bec76b81-c10c-4707-b329-d3eb01edb6dd","arxiv_id":"2511.09821","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"A Liouvillian-spectrum noise-resilience index is introduced to pick circuit symmetries that avoid fast-decaying Pauli noise, with numerical and hardware demonstrations on IBM and D-Wave devices.","lead":"This paper introduces a noise-resilience metric for quantum circuits based on the slow-decaying modes of hardware noise, and uses it to choose circuit variants that tolerate noise better. It tests the idea on IBM superconducting chips and D-Wave annealers, where circuits aligned with the slower noise symmetry show less error.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (6)'s λ_M ignores expansion amplitudes and is identical for the two ansatze in the paper's own VQA benchmark; the claimed fidelity bound is therefore not established, and the actual discrimination is delegated to an unproven degeneracy count in the SM.","rationale":"The reader's weakest_assumption focused on the stabilizer-support trackability of Algorithm 1, which is a real limitation, but the more fundamental issue is at the level of the metric itself: Eq. (6) is not shown to bound any state error, and the paper's own VQA example shows that λ_M cannot distinguish the two ansatze. The actual discrimination is made by a separate degeneracy count in the Supplemental Material, which is never formulated as part of a rigorous error bound and which also ignores amplitudes. This supports the reader's REJECT verdict, but for a different primary reason than the one listed as weakest_assumption. The concern is load-bearing because the abstract's central promise—an efficiently computable noise vulnerability index that 'bounds errors' and indicates fidelity—is exactly what is not established. The proposed test would settle this by checking whether the finite, different trace distances for the two ansatze can possibly be functions of a quantity that is −∞ for both. It cannot, so the current manuscript does not support the headline claim.","tokens_in":14342,"tokens_out":16710,"duration_ms":187795,"concrete_test":"Recompute the noisy final state for the two ansatze in Sec. IV.B with q_y=0 using exact statevector simulation for n=8 and random parameter initializations. For each run, compute λ_M from Eq. (6) and the exact trace distance ||ρ_noisy−ρ_ideal||_1. Since both ansatze have λ_M = −∞ while their trace distances are finite and different, this directly falsifies any claim that λ_M alone bounds the error. An even more decisive analytical check: derive from Eq. (5) an explicit α-independent inequality of the form ||ρ_noisy−ρ_ideal||_1 ≤ f(λ_M); the presence of the α coefficients in every decay term shows no such inequality follows without amplitude information, so the burden falls on the paper to prove the missing bound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that the noise vulnerability index λ_M 'bounds errors in noisy implementations, with smaller values indicating greater fidelity'—is not supported by the framework. Eq. (6) defines λ_M as the minimum real part of the summed Liouvillian eigenvalues, minimized over all eigenmode sequences i1,...,iL. This quantity carries no information about the expansion coefficients α in Eq. (5), and for a layer-independent Pauli noise channel the eigenvalue set is the same for every layer, so λ_M is identical for any two circuits of equal depth, independent of the unitaries. The paper's own VQA example confirms this: with q_y=0, both a=x and a=y have λ_M → −∞ (Sec. IV.B). The discrimination the paper actually uses is a separate count of how many final right eigenvectors realize the minimum (Supplemental Material), but that count also ignores amplitudes: it counts support-set membership only. No theorem is given connecting either λ_M or the count to trace distance or fidelity. Thus the abstract's bounding claim is an overclaim; the metric may be a heuristic for selecting circuits in specific noise models, but it is not established as an error bound. Algorithm 1 inherits this problem: it produces an upper bound on a decay rate, not on the resulting state error.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a noise-resilience framework based on metastability in open quantum systems. It defines a metric λ_M, the minimum over decay rates of the noise superoperator eigenmodes appearing in the circuit evolution, and claims that this index bounds errors in noisy implementations such that smaller values indicate greater fidelity. The framework is applied to two settings: digital variational quantum algorithms (hardware-efficient ansatze) and analog adiabatic state preparation, supplemented by experiments on IBM superconducting processors and D-Wave annealers. The authors argue that noise-aware algorithm design can exploit metastable noise to achieve intrinsic resilience without full classical simulation, and they propose an efficient classical procedure (Algorithm 1) to upper-bound λ_M under assumptions on stabilizer inputs and Pauli-support tractability.","tokens_in":14607,"tokens_out":2762,"duration_ms":32454,"significance":"If the central claim were established, the paper would offer a practical, efficiently computable criterion for selecting noise-resilient quantum circuits, potentially valuable for NISQ algorithm design. The paper's strengths include explicit analytical examples, a concrete combinatorial analysis in the Supplemental Material, and real-device benchmarks on IBM and D-Wave hardware. However, the central theoretical claim—that λ_M bounds the error between noisy and ideal states—is not supported by the formalism. The metric is defined as a minimum over decay rates and deliberately ignores expansion amplitudes, so it cannot by itself bound trace distance, fidelity, or cost-function error. The paper's own VQA example illustrates this: both ansatze have λ_M = -∞, and the actual discrimination comes from an ad hoc degeneracy count in the Supplemental Material. Consequently, the proposed metric functions as a heuristic for comparing circuit symmetries under specific noise models, but the claimed general error bound is not established. The efficient upper-bound procedure also rests on restrictive assumptions about Pauli-support growth that are not shown to hold for generic variational circuits","major_comments":[{"comment":"The abstract and Sec. III claim that λ_M 'bounds errors in noisy implementations, with smaller values indicating greater fidelity.' Eq. (6) defines λ_M as the minimum real part of summed Liouvillian eigenvalues over eigenmode sequences, explicitly 'irrespective of the amplitude of its associated eigenvector.' The actual state error in Eq. (5) is a sum of terms α exp(Σλ) r, and a mode with the smallest (most negative) decay rate may have negligible amplitude, while modes with larger decay rates may dominate. No theorem is given connecting λ_M to trace distance or fidelity. As stated, λ_M is a worst-case decay-rate indicator, not an error bound, and the central claim is therefore an overclaim.","section":"Sec. III, Eq. (6)"},{"comment":"The paper's own VQA benchmark contradicts the use of λ_M as a discriminating resilience metric. With q_y=0, both the a=x and a=y ansatze have λ_M → -∞ (Sec. IV.B), so λ_M itself does not separate the circuits. The discrimination is instead made by counting the number of final right eigenvectors corresponding to the minimum, as described in the Supplemental Material. That count also ignores amplitudes (it only tracks support-set membership), and no theorem connects this degeneracy count to the gradient decay, cost error, or fidelity. Thus the claim that 'parameterized circuits that minimize λ_M are also the ones that mitigate NIBPs the most' is not established by the presented analysis.","section":"Sec. IV.B and Supplemental Material"},{"comment":"The efficient upper-bound procedure assumes that non-Clifford gates 'do not change the set of Pauli strings that generate the qubits state at each layer,' as in the hardware-efficient ansatz. For generic variational circuits with non-Clifford rotations, the Pauli-support set can grow exponentially, so the claimed polynomial-time computability does not extend to the broad class of algorithms suggested in the paper. Moreover, Algorithm 1 returns an upper bound on a decay rate, not on the resulting state error; even if λ̃_M is efficiently computable, it inherits the same amplitude-blindness problem as λ_M. The text should state this limitation explicitly and avoid implying that the procedure certifies fidelity.","section":"Sec. VI, Algorithm 1"},{"comment":"The experimental sections are presented as validation, but the connection to the theoretical metric is loose. In the IBM experiment, the noise parameters q_x=q_z=0.5, q_y=0 are chosen rather than measured or fitted from the device, and no statistical uncertainty or device calibration data are given. In the D-Wave experiment, the paper admits (Sec. IX) that the relevant noise is non-unital, whereas the theoretical framework assumes unital noise; the observed forward/reverse asymmetry is interpreted through symmetry arguments but is not quantitatively linked to λ_M or to the metastability criterion. These experiments illustrate a plausible phenomenon but do not validate the central bound claim.","section":"Secs. V and IX"}],"minor_comments":[{"comment":"Eq. (16) states |1⟩⟨1| = I - σ_z, which is missing the factor 1/2. This appears to be a typo, but it affects the subsequent analytical comparison of fidelities and should be corrected.","section":"Sec. VIII.B"},{"comment":"There are several typographical errors, e.g., 'Inizialitizing' in Sec. IV.B and 'dicussions' in the Acknowledgements. These should be fixed before publication.","section":"Throughout"},{"comment":"The naming 'upper bound to noise resilience' is confusing because λ_M is non-positive and larger values (closer to zero) correspond to better resilience. The 'upper bound' is actually an upper bound on the decay rate (or a lower bound on resilience). Clarify the direction of the bound.","section":"Sec. VI, Algorithm 1"},{"comment":"The notation α^1_i1 α^2_i1,i2 ... α^L_iL-1,iL is never fully defined recursively; in particular, the action of U_k on a right eigenmatrix r^{k-1} is described only in words. A precise definition of α^k in terms of the biorthogonal basis would improve rigor.","section":"Sec. III, Eq. (5)"}],"recommendation":"reject","confidential_remarks":"The paper has a useful intuition about aligning algorithmic symmetries with noise, and the experimental data may be of interest, but the central formal claim—that the proposed metric bounds errors—is not supported by the definitions. The failure of λ_M to separate the two ansatze in the paper's own main example is a significant issue, and the fallback degeneracy count is not proved to be a valid error proxy. Because the paper's headline contribution is the error-bound claim, and because fixing it would require either adding amplitude information (defeating the no-simulation purpose) or substantially re-scoping the claims to a heuristic, I recommend rejection. The authors might be able to reframe the work as a heuristic noise-symmetry analysis in a future version, but as it stands the central claim is not established."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know up front: the paper's advertised tool—the λ_M noise vulnerability index—does not do what the abstract says. It is defined as a minimum decay rate over the circuit's Pauli modes, with no amplitude information, so it cannot bound the error or fidelity of a final state. The authors themselves show that in their main VQA example, both ansatze have λ_M → −∞; the actual discrimination comes from a separate count of how many eigenvectors realize that minimum, introduced only in the Supplemental Material. That count is also amplitude-blind. For layer-independent Pauli noise, λ_M depends only on depth, not on the circuit unitaries—another sign that it cannot be a circuit-specific error bound. So the central claim, that the index 'bounds errors' with smaller meaning greater fidelity, is unsupported by the math in the paper.\n\nWhat is genuinely new and good: the combinatorial analysis of how Pauli strings evolve under the CZ-layer, and the accompanying counting argument that selects the a=y ansatz, is a nice piece of work. The single-qubit analytical example (rotation about y vs x under σx-dominated noise) is clean and correct. The hardware data—IBM showing a=y more resilient, D-Wave showing reverse annealing less noisy than forward—are real observations, and the authors are candid that the D-Wave noise is non-unital and outside the framework. The efficient upper-bound idea (Algorithm 1) is a legitimate technical novelty, though its scope is narrow: it assumes a stabilizer input and non-Clifford gates that do not enlarge the Pauli-support set, and the proof of the bound is not given.\n\nThe soft spots beyond the main one: the experimental sections show a reproducible asymmetry but never measure metastability itself—no spectral gap, no time-scale separation. The figures lack error bars and no code or data are provided, which limits verification. Those are secondary; the primary issue is that the paper's own metric fails to separate the circuits it claims to separate, and no theorem connects the metric (or the degeneracy count) to fidelity.\n\nWho gets value from this: someone working on noise-aware ansatz design under known Pauli error models might find the counting heuristic useful, and the experimental asymmetry data are worth knowing. But as a theoretical claim about bounding errors, it does not hold up.\n\nRecommendation: this should go to peer review—the technical pieces deserve scrutiny and the experiments are checkable—but a referee should ask for major revisions or a substantial reframing. The paper would be stronger as a heuristic plus empirical study, not as a proven bound.","headline":"λ_M is a plausible heuristic but not a proven error bound; the real value is the Pauli-support counting and the hardware asymmetry data.","tokens_in":15186,"tokens_out":4509,"would_cite":false,"duration_ms":45068,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":["03.65.Yz","03.67.Lx"],"model":"deepseek-v4-flash","headline":"This paper claims that a cheaply computable noise index can rank quantum circuits by fidelity under hardware noise, and that the required metastable noise structure is present on today's devices.","keywords":["metastability","noise resilience","variational quantum algorithms","noise-induced barren plateaus","adiabatic state preparation","quantum annealing","Pauli twirling","Liouvillian spectrum"],"falsifier":"Measure fidelity decay under anisotropic Pauli noise (γx≫γy) preparing |1⟩ via an X-rotation versus a Y-rotation over time π; if the Y-rotation does not deliver the higher fidelity, the λ_M ordering is not predictive. Alternatively, run a circuit whose non-Clifford layers provably enlarge the Pauli support; if the upper bound still tracks errors, the support-trackability premise is not the operative mechanism.","tokens_in":14099,"feed_emoji":"⚛️","tokens_out":8280,"duration_ms":82999,"temperature":0.7,"pith_summary":"Metastability is the observation that noise does not attack all components of a quantum state uniformly: some decay channels relax almost immediately while others persist for a long time. The authors' key claim is that an algorithm's robustness is governed by how its state decomposition overlaps the fast-decaying channels, captured by a single noise vulnerability index λ_M defined as the minimum real part of the summed Liouvillian eigenvalues over the circuit's Pauli modes. They argue that circuits with larger (less negative) λ_M, and fewer modes at the minimum, produce final states closer to the ideal state, and they give an efficient recipe to upper-bound λ_M for stabilizer-initialized, layered circuits without simulating the algorithm. They verify the prediction numerically in variational circuits and analog adiabatic state preparation, and experimentally on superconducting gate processors and quantum annealers, where the symmetry-dependent error asymmetry matches the theory. If the claim holds, noise-aware ansatz selection becomes a practical design rule for near-term hardware, no error correction overhead required.","feed_headline":"A noise metric predicts which quantum circuits stay accurate","feed_subtitle":"It bounds errors using Pauli noise modes, and hardware tests confirm the ordering.","key_machinery":"The noise vulnerability index λ_M (Eq. 6): the minimum over all Pauli-mode paths of the real part of the sum of Liouvillian eigenvalues encountered along the circuit, together with the multiplicity of that minimum. It quantifies worst-case exposure to fast-decaying noise modes. The supporting machinery is the Pauli-string decomposition of the state under Clifford layers (stabilizer support tracking), which lets Algorithm 1 accumulate logarithms of Pauli weights to upper-bound λ_M without simulating non-Clifford layers.","core_discovery":"The paper's central claim is that a quantum algorithm's noise sensitivity is governed by the Pauli strings present in its state and how those strings decay under the hardware noise. The index λ_M of Eq. (6) is the minimum real part of the summed Liouvillian eigenvalues over those strings; the authors show that larger (less negative) λ_M means better worst-case fidelity, while the count of strings hitting the minimum separates otherwise-tied circuits. They demonstrate this by counting σ_y-containing strings in hardware-efficient ansatzes (predicting and then observing higher resilience for a=y rotations), by an analytical single-qubit rotation where Y-axis beats X-axis under σ_x-dominated noi","pith_inferences":["The authors do not discuss using λ̃_M as an optimization target; a natural extension is to include it as a penalty in the classical parameter update of a variational circuit, steering parameters toward metastable regions of the noise spectrum.","The support-set trackability assumption is the practical bottleneck; for circuits whose non-Clifford gates do enlarge the Pauli support, one could probe whether truncating low-weight strings preserves the fidelity ranking, a testable approximation for generic ansatzes.","The annealing results involve non-unital noise, which the main formal framework excludes; if the symmetry-based predictions hold there, a non-unital generalization of Eq. (6) would likely extend the method to thermalization-dominated hardware.","A direct experiment could compare two circuits with identical λ_M but different multiplicities; the theory predicts the one with fewer vulnerable modes decays slower, separating worst-case indexing from average-case behavior."],"forward_implications":["Variational circuits can be screened for noise resilience before execution by computing λ̃_M, and the screening predicts the depth at which noise-induced barren plateaus erase the cost landscape.","A single-qubit analytical example shows that under σ_x-dominated noise, a Y-rotation achieves final fidelity roughly e^{-2εT} while an X-rotation gives e^{-γx T}, a directly testable ordering.","Adiabatic state preparation can be made resilient by choosing initial Hamiltonians whose instantaneous state support avoids the dominant noise strings, with the W-state example quantifying the gain.","Because Algorithm 1 never depends on the non-Clifford layers, the framework is compatible with quantum advantage in principle: resilience can be certified without a full classical simulation.","Gate-model and annealing hardware benchmarks indicate the symmetry-dependent error asymmetry is present in actual machines, making the design rule immediately actionable."],"fun_headline_variants":["Quantum noise metric predicts circuit accuracy","Pauli strings reveal which quantum circuits resist noise","New index bounds quantum errors, hardware-tested","Metastability in noise guides resilient quantum algorithms","Noise-aware design: metric predicts quantum circuit fidelity"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The efficiency claims rest on the state's Pauli support remaining small enough to track (stabilizer input plus non-Clifford gates that do not enlarge the support), and the error bound on the Pauli-twirled, unital Markovian description of the hardware noise being accurate.","fun_headline_variants_meta":{"raw":{"variants":["Quantum noise metric predicts circuit accuracy","Pauli strings reveal which quantum circuits resist noise","New index bounds quantum errors, hardware-tested","Metastability in noise guides resilient quantum algorithms","Noise-aware design: metric predicts quantum circuit fidelity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000143,"raw_usage":{"total_tokens":987,"prompt_tokens":701,"completion_tokens":286,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":445,"completion_tokens_details":{"reasoning_tokens":217}},"tokens_in":445,"tokens_out":286,"duration_ms":3461,"temperature":1.0,"reasoning_tokens":217,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T22:32:18.295458+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure fidelity decay under anisotropic Pauli noise (γx≫γy) preparing |1⟩ via an X-rotation versus a Y-rotation over time π; if the Y-rotation does not deliver the higher fidelity, the λ_M ordering is not predictive. Alternatively, run a circuit whose non-Clifford layers provably enlarge the Pauli support; if the upper bound still tracks errors, the support-trackability premise is not the operative mechanism.","supporting_citations":[],"review_version":1}