{"id":"1ea4d619-0e18-4c21-8007-487b2a13aa11","arxiv_id":"2505.04743","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Classical purification of noisy Pauli-product states recovers magic and entanglement, with a noise floor that depends on when those resources are generated and on which circuit state is chosen.","lead":"By purifying noisy measurement data, the authors recover magic and entanglement from imperfect quantum states and show that states generating these resources later in a circuit survive noise better. This gives circuit designers a practical ordering rule and lets them reuse one robust state for many chemistry problems.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Purification to M=5 is load-bearing and unchecked: without verifying the spectral gap and dominant-eigenvector fidelity, recovered resource values and the path-ordering conclusion have no guaranteed referent.","rationale":"The reader identified the same load-bearing assumption, and I agree with that selection. The paper is an empirical and numerical study whose headline ordering claim is stated through coherent mismatch after purification; Eq. (6) is the gate through which every recovered resource metric passes. The authors explicitly call the dominant-eigenvector condition the central assumption and do not supply the spectral data needed to check it. That is not a sign of error, but it is an omitted verification for a claim that is central rather than peripheral. I considered the mixed-state stabilizer entropy issue raised in the reader's rationale: it is real, but it mainly affects raw SE comparisons, whereas the purified SE values used for the recovery claim are computed on near-pure states and can survive even if a mixed-state resource measure is not established. I also considered the in-sample selection of the H2 best circuit; that is a genuine generalization concern, but it does not attack the core claim that noise floor depends on resource order. The spectral-gap check is therefore the most direct test: if it passes, the conditional acceptance is appropriate; if it fails at the outlier points or at early path steps, the paper cannot be judged until those data points are reanalyzed. Because the reader's CONDITIONAL verdict already rests on this exact condition, my stress-test does not move the verdict.","tokens_in":964,"tokens_out":1031,"duration_ms":179365,"concrete_test":"For every noisy density matrix used in Figs. 2-10, compute the ordered spectrum lambda1 >= lambda2 >= ... >= lambda16 from the cirq/Qutip simulations and from the shadow-estimated experimental rho's, and report (i) lambda2/lambda1, (ii) the trace distance tr|rho^5/tr(rho^5) - |v1><v1||, and (iii) the fidelity |<psi_ideal|v1>|^2. If any point has lambda2^5/lambda1^5 >= 0.01 or trace distance >= 0.05, recompute that point's coherent mismatch and recovered resources by explicit projection onto v1, or exclude it. In parallel, run the identical classical-shadow pipeline on noiseless ideal states with the same shot budget; if noiseless purification already yields nonzero coherent mismatch, isolate the finite-sampling contribution before interpreting the experimental noise floor.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Every quantitative result in Secs. II C 2 and II D passes through Eq. (6): psi_purified = rho_noisy^M / tr(rho_noisy^M), with M=5. This map returns the dominant eigenvector only when lambda1 is sufficiently large relative to lambda2, and it returns the target state only when that eigenvector is close to |psi_ideal>. The paper states that this is the central assumption for purification based error mitigation protocols (Sec. II A 3) and asserts that M=5 gives pure states to a high approximation, but no eigengap, no post-purification purity, and no fidelity of the dominant eigenvector to the ideal state are reported for any simulated or experimental point. The two outlier circuit angles in Fig. 7 (theta=0.439 and 1.2) with anomalously low purity are exactly where the assumption is most likely to fail. If at those points lambda2^5/lambda1^5 is not negligible, the purified object is still a mixture, and the recovered QMI, SE, and coherent mismatch values in Figs. 7-10 no longer describe the target state. Since the path-ordering conclusion (Path 1 vs Path 2, Fig. 4) is also expressed through coherent mismatch after purification, an unverified failure of Eq. (6) at any path step would leave the ordering claim without a well-defined object. The issue is not that the authors are wrong; it is that the central assumption is load-bearing and currently unchecked.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies how magic (stabilizer Rényi entropy) and entanglement (multipartite QMI) in states generated by Pauli product formulas are affected by noise, and whether classical purification of noisy density matrices can recover these resources. The authors simulate depolarizing noise on random product-formula circuits and on two orderings of a six-unitary approximation to the H3 ground state, and run experiments on IonQ Aria using classical shadows and postselection. They report that coherent mismatch after purification tracks magic errors in low-noise regimes, that the ordering of unitaries affects the noise floor, and that a low-error circuit can be reused for a dressed H2 Hamiltonian across a dissociation curve. The central proposals are to design circuits that defer resource generation and to target states with low error rates.","tokens_in":15386,"tokens_out":5171,"duration_ms":48545,"significance":"The paper provides a plausible and timely connection between quantum resource theory and hardware noise: if the purification assumption holds, the results offer a practical heuristic for ordering Pauli exponentials and for reusing robust states. Strengths include the use of real hardware (IonQ Aria) with classical shadows and bootstrapped error bars, a concrete Hamiltonian-dressing demonstration, and an explicit statement of the purification assumption. The simulations use no fitted constants, and the correlation analyses are against independent metrics. The main value is the empirical demonstration that magic errors correlate with coherent mismatch and that resource generation order can affect recoverability in a specific example.","major_comments":[{"comment":"The purification protocol is load-bearing but its validity is not verified for any data point. Equation (6) returns the dominant eigenvector of rho_noisy only when lambda1^M dominates lambda2^M, and it describes the ideal state only if that eigenvector is close to |psi_ideal>. The text in Sec. II A 3 states this is the central assumption and asserts that M=5 yields pure states \"to a high approximation\", but no eigengap, post-purification purity, or dominant-eigenvector fidelity is reported for the simulated or experimental states. The two outlier points in Fig. 7 (theta=0.439 and 1.2) have anomalously low purity and are exactly where this assumption is most likely to fail. If at those points the purified object is still a mixture, the recovered QMI, SE, and coherent mismatch in Figs. 7-10 no longer refer to the target state, and the ordering conclusion expressed through coherent mismatch loses its well-defined object. Please report the spectral gap and the fidelity of the purified state to the ideal state for each data point, or restrict the conclusions to the region where this is verified.","section":"Sec. II A 3, Eq. (6)"},{"comment":"The claim that operator ordering affects the noise floor is based on two specific orderings of a single H3 instance. The path U2 in Table II was selected because it \"diverged most strongly\" in resource generation while keeping reasonable Trotter error; this is a selection on the outcome, and no systematic sampling over random orderings or over different target states is provided. Without such sampling (or an analytic argument), the general statements in the Introduction (\"product formulas... have a noise floor based on when they generate entanglement and magic\") and in the Conclusions are not established beyond the chosen example. Please test the ordering effect on a random set of orderings and report how often the \"defer resource generation\" rule holds, and whether the effect size is stable.","section":"Sec. II C 2, Fig. 4"},{"comment":"The advantage of tailoring the H2 dissociation curve to the \"best\" circuit is demonstrated in-sample. The circuit angle theta=0.401 is selected as the one with the lowest ratio of postselected data in the same experimental dataset that is then used to compute the dressed dissociation curve and compare with the original VQE circuit. This selection bias could inflate the reported advantage. Please provide an out-of-sample evaluation (for example, selecting the angle from a calibration run and evaluating on a separate run) or report the variability of the conclusion across multiple repetitions of the selection procedure.","section":"Sec. II D 1, Fig. 8"}],"minor_comments":[{"comment":"The definition of multipartite QMI is garbled: the sum over k_i is not defined clearly, the entropy arguments are written as X_{k1}, X_{k1}, ..., X_{kn}, and the overall formula appears to differ from the standard Watanabe formula in Ref. [35]. Please rewrite with explicit index notation or reproduce the definition from Ref. [35].","section":"Eq. (3)"},{"comment":"\"2-Stabilizer Renyi entropy\" should be \"2-Rényi stabilizer entropy\"; the definition in Eq. (4) uses alpha, but only M2 is used.","section":"Table I"},{"comment":"The circuit text contains typos: \"e^{-i0.0.401YX}\" and \"e^{-i0.0.2007YZXZ}\" should have single decimals (for example, e^{-i0.401 YX}).","section":"Sec. IV B 2"},{"comment":"\"Bootstrapped the classical shadows 250 using the postselected data\" should read \"bootstrapped the classical shadows 250 times using the postselected data\".","section":"Sec. IV B 1"},{"comment":"The Pearson correlations in Fig. 3 b) are averaged over stratified subsets; it would help to report the number of circuits and the standard deviation on the correlation values explicitly in the text, since the figure alone does not show the spread across subsets.","section":"Sec. II C 1"},{"comment":"Reference [27] is incomplete (\"Nature 622, 481 (2023)\" with no author list); please supply the full author list.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is an interesting empirical study with a clear potential use for circuit compilation and Hamiltonian dressing. The referee concerns about the unverified purification assumption and the single pair of orderings are central; the authors may be able to address them with additional analysis of existing data, which is why I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a useful empirical paper, not a theorem paper. It demonstrates on 4-qubit simulations and IonQ Aria that classical purification to order M=5 partly restores stabilizer entropy and QMI from noisy Pauli product states, and that the order in which magic and entanglement are generated appears to set the noise floor. The hardware experiments are the strongest part: the raw-vs-purified curves in Figs. 7 and 10 are coherent, and the H2 Hamiltonian dressing trick—reusing one low-error circuit for the whole dissociation curve and cutting measured terms from 150 to 19—is a genuinely practical observation.\n\nWhat is actually new: the empirical correlation between magic of the ideal state and the coherent mismatch after purification in the low-noise regime, and the ordering-dependent noise floor. The individual tools (purification, stabilizer entropy, classical shadows) are established; the contribution is an extension and a set of observations, not a new framework. That is fine, but it affects how much weight the claims should carry.\n\nThe soft spots are mostly about load-bearing assumptions. Eq. (6) with M=5 is the central map. The paper says the dominant eigenvector must approximate the target state and asserts that M=5 gives pure states to high approximation, but it never reports eigengaps, post-purification purity, or fidelity of the purified state for any simulated or experimental point. The two outlier angles in Fig. 7 (theta=0.439 and 1.2) are exactly where the assumption could fail. Without those checks, the recovered QMI, SE, and coherent mismatch numbers are not guaranteed to describe the ideal state. This is the main correctness risk. The ordering conclusion also rests on only two hand-picked H3 paths; that is enough to motivate the claim, not enough to establish a general principle. The best-circuit selection in the H2 experiment is in-sample. No code or data are released, so the experimental results cannot be independently reconstructed. The mixed-state version of the stabilizer entropy is used without defining how Eq. (5) is extended to noisy density matrices. These are fixable in revision—report spectral gaps and purities, sample more orderings, release code and data—but they need to be addressed.\n\nBottom line: a serious referee should engage with this. It is an honest, mostly well-cited empirical study with real hardware data. I would not cite it in its current form, but with the spectral-gap and ordering checks added, it becomes a useful reference for anyone designing ansatze around noise-resilient state preparation.","headline":"Solid four-qubit empirical study showing purification can recover magic and entanglement from noisy Pauli product states, but the ordering claim and the central M=5 purification assumption need verification before the conclusions generalize.","tokens_in":15917,"tokens_out":2397,"would_cite":false,"duration_ms":24147,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81P45"],"pacs":["03.67.Lx","03.67.Mn"],"model":"deepseek-v4-flash","headline":"The paper connects the noise floor of a purified noisy quantum state to the magic inside it and shows that circuits generating magic and entanglement early carry a higher coherent mismatch throughout.","keywords":["magic states","stabilizer Rényi entropy","quantum mutual information","Pauli product formulas","purification-based error mitigation","noise floor","classical shadows","Hamiltonian dressing"],"falsifier":"From the same classical-shadow density matrices, compute the dominant eigenvector's fidelity to the ideal state and the gap between the two largest eigenvalues: any data point where that fidelity is not close to one, or where purity after five purification steps stays well below one, would show that the recovered resource numbers describe the noisy state's principal mode rather than the target, and the noise-floor ordering conclusions would need to be revisited.","tokens_in":14902,"feed_emoji":"⚛️","tokens_out":10442,"duration_ms":89404,"temperature":0.7,"pith_summary":"This paper tries to show that low-fidelity noisy quantum states can still be useful: as long as the dominant eigenvector of the noisy density matrix approximates the intended state, classical purification recovers estimates of magic, entanglement, and ground-state energies from states produced by Pauli product formulas. Its central finding is that the post-purification noise floor is tied to the magic in the state and, more concretely, to when during circuit construction that magic and entanglement are generated. Two circuits that approximate the same target state to similar chemical accuracy can have very different noise floors: the one that generates resources early has a larger coherent mismatch at every step. In simulation and on ion-trap hardware, correlations prove more robust to noise than magic, and experimentally the authors use a single low-error circuit, with Hamiltonian dressing, to compute an entire dissociation curve while cutting the number of measured Hamiltonian terms.","feed_headline":"Magic early in a circuit raises its noise floor","feed_subtitle":"Purifying noisy states recovers magic and entanglement; deferring their creation lowers the noise floor.","key_machinery":"The machinery is a small set of quantum information metrics applied to states built from Pauli exponentials $e^{-i\\theta P}$. Magic is quantified by the 2-stabilizer R\\'enyi entropy $M_2 = -\\log_2\\left(\\frac{1}{2^n}\\sum_P \\langle\\psi|P|\\psi\\rangle^4\\right) - S$, correlations by the multipartite quantum mutual information (QMI), and noise and recoverability by purity, the purified state $\\rho^M_{\\rm noisy}/\\mathrm{tr}(\\rho^M_{\\rm noisy})$ that extracts the dominant eigenvector, and the coherent mismatch $c = 1 - |\\langle\\psi_{\\rm ideal}|\\psi_{\\rm purified}\\rangle|^2$. The purification step converts a low-overlap noisy state into a near-pure state whose resources and energies can be compared with theory, and the coherent mismatch is the quantity that ties the noise floor to the order of resource generation.","core_discovery":"The paper establishes a concrete relationship between the noise floor of a purified noisy state and the quantum resources inside it. For Pauli product states under depolarizing noise, the coherent mismatch after purification tracks the stabilizer entropy (magic) of the state in the low-noise regime, and at stronger noise it tracks errors in magic and correlations rather than state overlap. Given two unitaries that approximate the same linear H3 ground state with energy errors well below chemical accuracy, the path that generates magic and entanglement early has a higher coherent mismatch throughout, even at steps where the other path has more magic and more correlation; operator ordering therefore sets how recoverable a state is. On ion-trap hardware, density matrices reconstructed from postselected single-qubit classical shadows allow the same purification to recover magic, QMI, and chemically accurate energies from raw states whose overlap with the target is low and decreasing. The authors further show that dressing the Hamiltonian makes one low-error circuit compute an entire H2 dissociation curve, reducing measurement overhead.","pith_inferences":["If the ordering sensitivity is a general feature of Pauli product formulas rather than a quirk of these two circuits, then Trotterization and adaptive ansatz searches could be augmented with a resource-deferral penalty: among sequences with comparable Trotter error, prefer the one that generates magic and entanglement latest; this is a testable extension the paper gestures toward but does not esta","The robustness gap between correlations and magic suggests a design heuristic for near-term algorithms: lean on correlation-heavy quantities and keep magic-hungry subroutines as short or as late as possible, since magic is the resource that hardware noise corrupts first.","Because the dominant-eigenvector assumption is checkable from the same shadow density matrices, one could verify the eigengap and the dominant eigenvector's fidelity to the ideal state at every data point; where that check fails, the recovered resource values should be read as properties of the noisy state's principal mode rather than of the intended computation.","The Hamiltonian dressing demonstration implies that the economic advantage of finding one robust circuit grows with the number of Hamiltonians to be processed, since the same expectation values are reused; this amortization argument is only implicit in the paper."],"forward_implications":["Circuit compilers and adaptive ansatz builders can lower the noise floor by ordering Pauli exponentials to defer magic and entanglement generation; the paper's H3 comparison shows the same six exponentials ordered differently change the coherent mismatch at every step.","Purification-based error mitigation can deliver chemically accurate energies from raw states whose overlap with the target is low and still falling, so overlap alone is an inadequate guide to state quality in noisy settings.","In the low-noise regime, the noise floor is set mainly by the state's magic, so algorithms requiring substantial magic should expect a higher floor and should budget more error mitigation for magic-bearing subroutines.","Simple hardware signals such as the postselection survival ratio correlate with purity and coherent mismatch, making them practical proxies for identifying the most recoverable circuits and, via Hamiltonian dressing, reusing one good state across many Hamiltonians."],"supporting_citations":[{"why":"Defines the coherent mismatch and states the dominant-eigenvector assumption that underpins purification-based error mitigation.","marker":"[36]"},{"why":"Introduces the stabilizer Rényi entropies used here to quantify magic.","marker":"[25]"},{"why":"Provides the classical shadows protocol used to reconstruct the experimental density matrices.","marker":"[44]"},{"why":"Extends classical shadows to postselected data, which all experimental metric estimates rely on.","marker":"[45]"},{"why":"Introduces density-matrix power purification, the operation used to extract the dominant eigenvector in Eq. 6.","marker":"[29]"},{"why":"Supplies the virtual-distillation context for purification-based recovery that motivates extracting resources from noisy states.","marker":"[28]"},{"why":"Gives the ILC method that builds the H3 unitary whose operator orderings are compared for noise-floor differences.","marker":"[39]"},{"why":"Gives the qubit coupled cluster method used to construct the Be-atom circuit path.","marker":"[41]"}],"fun_headline_variants":["Defer magic to lower the noise floor","Noise floor rises when magic is early","Purification recovers magic from noisy states","Correlations outlast magic under noise","Order of magic creation sets noise floor"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the dominant eigenvector of each noisy density matrix still being the ideal target state; if hardware noise rotates that eigenvector away, the purified magic, correlations, and coherent mismatch stop describing the state the circuit was meant to create.","fun_headline_variants_meta":{"raw":{"variants":["Defer magic to lower the noise floor","Noise floor rises when magic is early","Purification recovers magic from noisy states","Correlations outlast magic under noise","Order of magic creation sets noise floor"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1421,"prompt_tokens":907,"completion_tokens":514,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":450}},"tokens_in":523,"tokens_out":514,"duration_ms":5109,"temperature":1.0,"reasoning_tokens":450,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:22:32.145576+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"From the same classical-shadow density matrices, compute the dominant eigenvector's fidelity to the ideal state and the gap between the two largest eigenvalues: any data point where that fidelity is not close to one, or where purity after five purification steps stays well below one, would show that the recovered resource numbers describe the noisy state's principal mode rather than the target, and the noise-floor ordering conclusions would need to be revisited.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the stabilizer Rényi entropies used here to quantify magic."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the classical shadows protocol used to reconstruct the experimental density matrices."},{"cited_title":"Aliverti-Piuri, K","cited_arxiv_id":null,"evidence_quote":"Introduces density-matrix power purification, the operation used to extract the dominant eigenvector in Eq. 6."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the virtual-distillation context for purification-based recovery that motivates extracting resources from noisy states."},{"cited_title":"Leone, S","cited_arxiv_id":null,"evidence_quote":"Gives the ILC method that builds the H3 unitary whose operator orderings are compared for noise-floor differences."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the qubit coupled cluster method used to construct the Be-atom circuit path."}],"review_version":1}