{"id":"4715808e-344f-4a6b-a0a9-5eb553f3555e","arxiv_id":"1908.05047","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Bundled graph states achieve near-Heisenberg-limited phase estimation and retain a quantum advantage under iid dephasing and, on average, under a few qubit erasures.","lead":"The authors derive the quantum Fisher information of arbitrary graph states and introduce bundled graph states whose precision scales close to the Heisenberg limit while surviving dephasing and partial qubit loss. The result gives a concrete resource family for noise-tolerant quantum sensing in future quantum networks.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The erasure-robustness claim in the abstract is average-case only; explicit single-erasure patterns (bundled star, and bundled cyclic with k=4) fall below the SQL, so the unqualified statement is unsupported.","rationale":"The reader's weakest assumption already identified the average-vs-worst-case gap, and my reading confirms it as the most load-bearing issue. The QFI formula Eq. (7) and the bundled-state construction are supported by the stabilizer reasoning and are not the point of failure. The Appendix B counting issue is real but peripheral to the advertised metrology resource claim. The erasure claim is central to the abstract and is stated unconditionally, yet the paper's own metric is explicitly an average over erasure patterns. Moreover, the bundled star is conceded to fail after one erasure, and the bundled cyclic example with k=4 shows the same failure under Eq. D.5 even though the average formula would look favorable. A concrete single-erasure pattern suffices to falsify the unqualified claim, so the manuscript needs a qualification or correction rather than a new central construction. Because the reader already conditioned the verdict on this issue, the verdict remains CONDITIONAL; no adjustment to the reader's assessment is required.","tokens_in":9935,"tokens_out":19213,"duration_ms":201289,"concrete_test":"Evaluate the authors' exact formula Eqs. D.5-D.6 for the n=120 bundled cyclic graph with k=4 bundles of j=30 qubits and a single erased vertex in one bundle. The formula gives Q=2j=60, below the SQL of n-1=119, while Eq. 23 or D.9 predicts a much larger average. If this evaluation reproduces, the unconditional quantum-advantage-after-erasures claim in the abstract must be amended to an average-case statement, and the validity of Eq. 23 and D.9 for k=4 should be rechecked.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central robustness claim that bundled graph states maintain a quantum advantage after finite erasures is only derived for the QFI averaged uniformly over all (n choose e) erasure patterns (text after Eq. 23; Appendix D). The abstract and conclusion state it without that qualification. The exact single-erasure formula, Eqs. D.5-D.6, shows that the distinction matters. For an n-qubit bundled star with k bundles of size j=n/k, losing one leaf qubit gives Q=j, below the SQL of n-1 for any k>1; the paper's own Fig. 2 concedes this for stars. For the bundled cyclic family, take k=4 bundles of size j=n/4 and erase one vertex in one bundle. Using the authors' Eq. D.5, the erased vertex's bundle and the opposite bundle have their shared neighborhoods fully contained in L, so their h_l vanish; the two adjacent bundles contribute j each, giving Q=2j=n/2, again below the SQL of n-1. Meanwhile the average formula Eq. 23 predicts a much larger value for this graph, indicating that the average formulas are either misapplied or incorrect for this configuration. Thus a fixed, known loss pattern can destroy the advertised advantage even while the average over loss patterns is favorable. The paper either needs to restrict the claim to average-case performance or to graphs and k values with proven worst-case QFI above the SQL; otherwise the abstract overstates the result.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies graph states as resources for single-parameter phase estimation under the Hamiltonian H = (1/2)∑X_i. It derives a closed-form expression for the quantum Fisher information (QFI) of any graph state with no isolated vertices, Q(G) = ∑_l v_l^2, where v_l are the sizes of disjoint groups of vertices with identical neighborhoods (Eq. 7). It then introduces a \"bundling\" construction that converts any k-qubit graph state into an n-qubit graph state with Q ≥ n^2/k, approximately Heisenberg-limited (Eq. 15). The paper also analyzes robustness under iid dephasing and finite erasures, derives an approximate QFI formula for dephasing (Eq. 20) and an average-QFI formula for single erasures (Eq. 23), provides a protocol for saturating the QFI with single-qubit measurements under stated conditions, and gives a lower bound on the number of stabilizer states useful for metrology (Appendix B).","tokens_in":10118,"tokens_out":15259,"duration_ms":152864,"significance":"The characterization Q(G) = ∑_l v_l^2 is elegant, correct, and immediately useful for identifying metrologically valuable graph states; the bundled construction is a simple and effective way to achieve near-Heisenberg scaling. The paper ships explicit derivations in the appendices, including exact erasure formulas and a dephasing approximation, and it proposes a concrete single-qubit measurement protocol. These are genuine strengths. If the noise-robustness claims are restricted to the regimes actually proven (average-case erasure robustness and dephasing in the regime where Eq. 19 holds), the paper will be a solid contribution to the practical quantum-metrology literature. As written, however, the abstract and conclusion overstate the robustness results.","major_comments":[{"comment":"The abstract and conclusion state that bundled graph states \"maintain a quantum advantage after being subjected to ... finite erasures\" without qualification. The only derivation in the manuscript is for the QFI averaged uniformly over all erasure patterns, defined after Eq. (23) and evaluated in Appendix D (Eqs. D.7–D.9). For a fixed erasure pattern the exact formula Eq. (D.5)–(D.6) can give values far below this average: for a bundled star with equal bundles of size n/k, losing one leaf qubit gives Q = n/k, below the standard quantum limit n−1 for k > 1 (as Fig. 2 concedes for stars), and for a bundled cyclic graph with k = 4, erasing one vertex in one bundle yields Q = n/2, also below n−1. Thus the advertised robustness is an average-case property, not a worst-case guarantee. The abstract, conclusion, and any further statements should either explicitly say \"on average over erasure patterns\" or restrict the claim to graphs and erasure patterns for which a worst-case bound is proven.","section":"Abstract; §3, after Eq. (23); Appendix D"},{"comment":"The dephasing result is an approximation whose validity is controlled by Eq. (19), (2p(1−p)+1/2)^{N_l} ≈ 0. When the shared neighborhoods N_l are small or when p is not very small, the correction term in Eq. (C.5) is non-negligible; the paper itself notes a change of slope for bundled star graphs with large k after Eq. (22). Therefore Eq. (21) and the abstract's claim of robustness under iid dephasing should be presented as holding in the parameter regime where Eq. (19) is valid, rather than as an unconditional property of bundled graph states.","section":"§3, Eqs. (19)–(21)"},{"comment":"The single-qubit measurement protocol is derived for small θ and for states possessing a stabilizer consisting only of Y and Z operators. The derivation is sound in that regime, but the text should state more explicitly that the error-propagation formula Δθ² ≈ 1/Q(G) is a small-θ approximation and that for states failing the stabilizer condition the appended-qubit construction changes the effective measurement; the current presentation is slightly too terse to make the domain of validity of Eq. (25) unambiguous.","section":"§4, Eqs. (24)–(25)"}],"minor_comments":[{"comment":"The affiliation contains missing accents: \"Laboratoire dInformatique\" and \"Universit\" should be \"Laboratoire d'Informatique\" and \"Université\".","section":"Affiliation line"},{"comment":"The phrase \"To obtain To obtain any sort of meaningful value\" contains a duplicated \"To obtain\".","section":"Appendix D, first paragraph"},{"comment":"The summation index in Eq. (24) runs from i = 0 in two places; it should run from i = 1 to n consistently with Eq. (1).","section":"Eq. (24)"},{"comment":"The author name \"Ozmenaic\" in the main text (near Eq. (12)) is a misspelling of \"Oszmaniec\" (reference [7]).","section":"References and text"},{"comment":"The notation \"lognQ\" in the caption is ambiguous; the base of the logarithm and the argument should be made explicit, e.g., \"log_{10}(Q)\" or \"ln Q\".","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"The central QFI characterization and the bundling construction are solid and likely novel enough for publication in a quantum-information journal. The main required changes are to qualify the erasure- and dephasing-robustness claims to their proven regimes and to fix the identified presentation issues. I do not see grounds for rejection, but the abstract currently promises more than the theorems deliver."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core of this paper is good and worth engaging with. Eq. (7) — Q(G) = sum_l v_l^2 for graph states with no isolated vertices — is a clean, genuinely useful characterization, and the bundled graph-state construction that achieves Q ~ n^2/k is a natural but new idea that builds on existing graph-state technology. The local-measurement protocol achieving 1/Q with fixed Y/Z measurements is a nice practical add-on. The derivations are straightforward and honest: they start from standard QFI formulas and stabilizer facts, no fitted parameters, no circularity. The dephasing analysis is also carefully done, with the approximation condition in Eq. (19) stated explicitly. I would trust the central construction. The soft spot is the robustness claim. The abstract says bundled graph states \"maintain a quantum advantage after being subjected to iid dephasing or finite erasures.\" That is not what is proven. The erasure result is explicitly an average over all (n choose e) loss patterns — the paper says so right after Eq. (23) — and the average can be favorable while specific loss patterns destroy the advantage. The stress-test example is correct: for a bundled star, losing one leaf leaves Q = n/k, below the SQL; for a bundled cycle with k=4, one erasure gives Q = n/2, below the SQL. The paper's own text concedes the star case but not the cyclic case, and the abstract makes no qualification at all. This is fixable by rewriting the abstract and conclusion to say \"on average over erasure patterns\" or by proving worst-case bounds for specific families. It is not fixed in the main text. A second, smaller issue is the counting argument in Appendix B. As written, the lower bound appears to count choices of generators rather than distinct stabilizer states, and the assertion that the constructed group contains no ±X_i or −X_iX_j stabilizers is stated without proof. This appendix should be either corrected or downgraded to a less prominent claim. The dephasing claim is fine in context — the approximation condition is stated — but the abstract also makes that sound more universal than it is. Bottom line: this paper deserves a serious referee. The central QFI formula and bundled construction are solid and should be published after the abstract and conclusion are corrected to match the average-case erasure result, and after the counting appendix is fixed or softened. I would bring it to a reading group and would cite the QFI characterization if I worked in graph-state metrology.","headline":"The QFI formula and bundled graph-state construction are solid and useful; the abstract's unqualified erasure-robustness claim is average-case only and should be fixed before publication.","tokens_in":735,"tokens_out":821,"would_cite":true,"duration_ms":55032,"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":"Any graph state's quantum Fisher information is a sum of squared twin-class sizes; bundled graph states from this identity approach the Heisenberg limit and handle dephasing and small erasures.","keywords":["quantum metrology","graph states","quantum Fisher information","Heisenberg limit","phase estimation","stabilizer states","dephasing noise","erasure robustness"],"falsifier":"Compute the exact quantum Fisher information of a bundled cyclic graph after erasing one entire bundle together with its neighbours, for a fixed pattern rather than an average; if that value falls below the standard quantum limit for $e\\le 3$ while the average over all patterns stays above it, the unqualified claim of robustness to finite erasures fails. Similarly, for a bundled graph whose shared neighbourhood size $N_l$ is small, evaluate the exact dephasing QFI at $p=0.15$ and compare with $(1-2p)^2Q(G)+4np(1-p)$; a large deviation would show the approximation's boundary.","tokens_in":9642,"feed_emoji":"🎯","tokens_out":12372,"duration_ms":109891,"temperature":0.7,"pith_summary":"This paper claims that the usefulness of a graph state for phase estimation is encoded in its symmetries: for any graph state with no isolated vertices, the quantum Fisher information equals the sum of squared sizes of the groups of qubits with identical neighbourhoods. Using this identity, the authors construct bundled graph states by replacing each vertex of a seed graph with several qubits that all share the same neighbourhood, reaching a QFI of at least $n^2/k$, close to the Heisenberg limit. They show that under independent dephasing the QFI keeps the form $(1-2p)^2 Q(G)+4np(1-p)$, so bundled states beat the standard quantum limit for dephasing probability up to about $0.2$. For erasures, they compute the QFI averaged over all patterns of $e$ lost qubits and find that bundled cyclic graphs retain an advantage on average for up to three erasures, while bundled star graphs do not survive one. Because graph states are already produced and distributed on many platforms, these claims would make robust quantum-enhanced sensing a near-free add-on to existing graph-state technology.","feed_headline":"Graph states with twin qubits approach the Heisenberg limit","feed_subtitle":"A twin-neighbourhood formula predicts each graph's phase-estimation precision, and the best designs keep their edge under noise.","key_machinery":"The machinery is the partition of graph vertices into twin classes: sets of vertices with identical neighbourhoods. For a graph state with no isolated vertices, a class of size $v_l$ contributes $v_l^2$ to the quantum Fisher information because every pair in the class corresponds to a stabilizer $X_iX_j$; summing classes gives $Q(G)=\\sum_l v_l^2$. The bundled graph state construction amplifies this effect by replacing each vertex with $n_i$ mutually twin qubits, which pushes the QFI to $\\sum_i n_i^2$. A second mechanism, true twins ($N(i)\\cup\\{i\\}=N(j)\\cup\\{j\\}$) together with local Clifford operations, produces the extended formula $Q=\\sum_l v_l^2+\\sum_m u_m^2-n$ and explains why complete graphs reach $Q=n^2$. The same twin-class bookkeeping drives the dephasing and erasure formulas, where the QFI decomposes over classes $V_l$ and their shared neighbourhoods $N_l$.","core_discovery":"For phase estimation with Hamiltonian $H=\\sum_{j=1}^n X_j$, the paper proves that an $n$-qubit graph state with no isolated vertices has $Q(G)=\\sum_l v_l^2$, where $V_l$ are the disjoint sets of vertices sharing the same neighbourhood; the identity follows because $X_iX_j$ stabilizes the graph state exactly when $N(i)=N(j)$, so the QFI counts twin pairs. Extending the same counting to true twins and local Clifford operations gives $Q(G_{\\mathrm{LC}})=\\sum_l v_l^2+\\sum_m u_m^2-n$, recovering $Q=n^2$ for complete graphs. The bundled construction replaces vertex $i$ by $n_i$ qubits with a common neighbourhood, yielding $Q(G_{\\mathrm{bundle}})\\ge \\sum_i n_i^2 \\ge n^2/k$ for a $k$-vertex seed graph, i.e. near-Heisenberg scaling. Under iid dephasing the derived closed form is approximately $(1-2p)^2Q(G)+4np(1-p)$ when shared neighbourhoods are large, and the paper shows bundled cyclic and star graphs beat the SQL for $p\\le 0.2$. For $e$ erasures the paper computes the average QFI over all $\\binom{n}{e}$ loss patterns; bundled cyclic graphs stay above the SQL on average for $e\\le 3$, whereas bundled star graphs drop below after one erasure. Finally, the authors show the optimal $1/Q$ precision can be attained by measuring a stabilizer built from $Y$ and $Z$ operators, or by appending one extra qubit connected to a suitable set, for small phases.","pith_inferences":["Because $Q(G)$ is fixed entirely by the sizes of twin classes, any graph state can be screened for metrological usefulness by counting neighbourhood-equivalence classes alone, without simulating the state or optimizing measurements.","The erasure results are averages over all $\\binom{n}{e}$ loss patterns; for adversarial or correlated losses, a worst-case analysis could give a very different picture, and the paper does not provide that guarantee.","One natural extension, not pursued here, is to protect the twin-class structure with graph-state quantum error correction so the stabilizer pairings survive after correction; this could turn the average erasure advantage into a guaranteed one."],"forward_implications":["Cluster states have QFI $n$, confirming that unmodified cluster states give no scaling advantage, while star (GHZ-type) states have QFI $(n-1)^2+1$ and complete graphs reach $n^2$ after local Clifford operations.","Bundling any $k$-qubit seed graph with no isolated vertices produces an $n$-qubit state with $Q\\ge n^2/k$, so a constant number of bundles gives near-Heisenberg scaling for arbitrarily large $n$.","Under iid dephasing with probability $p$, bundled graph states have QFI approximately $(1-2p)^2Q(G)+4np(1-p)$, keeping them above the standard quantum limit for $p\\lesssim 0.2$ when shared neighbourhoods are large.","On average over all patterns of up to three erased qubits, bundled cyclic graphs retain a quantum advantage; bundled star graphs do not survive even a single erasure.","For small phases, the optimal precision $\\Delta\\theta^2=1/Q$ is achievable with a fixed stabilizer measurement, or with one extra appended qubit for arbitrary graph states."],"supporting_citations":[{"why":"Supplies the closed-form quantum Fisher information formula (Eq. A.6) that the paper applies to graph states.","marker":"[11]"},{"why":"Defines the quantum Fisher information as the Fisher information maximized over POVMs, the precision bound used throughout.","marker":"[30]"},{"why":"Establishes that unmodified cluster states do not provide a scaling advantage; the paper reproduces this as Q=n and uses it as a baseline.","marker":"[6]"},{"why":"Surveys which entangled states are good metrology resources and connects symmetry to usefulness, motivating the twin-class analysis.","marker":"[7]"},{"why":"Counts n-qubit stabilizer states, which is the basis for the lower bound on the number of useful stabilizer states.","marker":"[31]"},{"why":"Defines true twins, the graph-theoretic notion used in the local-Clifford extension of the QFI formula.","marker":"[32]"},{"why":"Provides the Cramér-Rao bound and the idea that the optimal POVM can be parameter-dependent, which the measurement section addresses.","marker":"[33]"}],"fun_headline_variants":["Twin-neighbourhood graphs approach Heisenberg limit","Bundled graph states achieve near-Heisenberg precision","Graph states with twin qubits beat noise in metrology","Bundled graph states retain metrological advantage","Near-Heisenberg precision from bundled graph states"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The robustness claims for erasures and dephasing rest on averaging over all erasure patterns and on shared neighbourhoods being large enough that the correction term $(2p(1-p)+1/2)^{N_l}$ is negligible; a fixed bad loss pattern or a small shared neighbourhood can remove the advantage.","fun_headline_variants_meta":{"raw":{"variants":["Twin-neighbourhood graphs approach Heisenberg limit","Bundled graph states achieve near-Heisenberg precision","Graph states with twin qubits beat noise in metrology","Bundled graph states retain metrological advantage","Near-Heisenberg precision from bundled graph states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001516,"raw_usage":{"total_tokens":6123,"prompt_tokens":1044,"completion_tokens":5079,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":660,"completion_tokens_details":{"reasoning_tokens":5005}},"tokens_in":660,"tokens_out":5079,"duration_ms":32703,"temperature":1.0,"reasoning_tokens":5005,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:27:39.951873+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the exact quantum Fisher information of a bundled cyclic graph after erasing one entire bundle together with its neighbours, for a fixed pattern rather than an average; if that value falls below the standard quantum limit for $e\\le 3$ while the average over all patterns stays above it, the unqualified claim of robustness to finite erasures fails. Similarly, for a bundled graph whose shared neighbourhood size $N_l$ is small, evaluate the exact dephasing QFI at $p=0.15$ and compare with $(1-2p)^2Q(G)+4np(1-p)$; a large deviation would show the approximation's boundary.","supporting_citations":[{"cited_title":"Friis, D","cited_arxiv_id":null,"evidence_quote":"Supplies the closed-form quantum Fisher information formula (Eq. A.6) that the paper applies to graph states."},{"cited_title":"Clark, C","cited_arxiv_id":null,"evidence_quote":"Defines the quantum Fisher information as the Fisher information maximized over POVMs, the precision bound used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Counts n-qubit stabilizer states, which is the basis for the lower bound on the number of useful stabilizer states."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines true twins, the graph-theoretic notion used in the local-Clifford extension of the QFI formula."},{"cited_title":"van Loock, C","cited_arxiv_id":null,"evidence_quote":"Provides the Cramér-Rao bound and the idea that the optimal POVM can be parameter-dependent, which the measurement section addresses."}],"review_version":1}