{"id":"135ddf60-b50b-4c3c-9da8-e265f1479678","arxiv_id":"2608.05426","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"Pairs of 88Sr atoms in an optical clock produced genuine qutrit entanglement (F=0.85(1)) and joint two-phase estimation with variance 0.82(3), below the ideal individual-qubit bound of 1.","lead":"An experiment with strontium atoms in optical tweezers created genuine three-level (qutrit) entanglement and used it to estimate two phases at once, beating the standard one-phase-per-atom method. It is the first such demonstration with individually controlled neutral atoms, a step toward higher-dimensional quantum sensors and clocks.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Postselection on atom loss is not characterized for state dependence; if loss rates differ across qutrit levels, both the fidelity certificate and the sub-threshold joint variance could be biased.","rationale":"The reader's weakest_assumption identifies exactly the load-bearing concern: the loss-postselected ensemble is used for both headline results, and state-dependent loss would bias the postselected state and the likelihood model. I agree with that assessment. The zero-dark-time limitation is explicitly acknowledged by the authors as a proof-of-principle feature and does not undermine the claim of joint multi-parameter estimation with injected phases; it is not the primary risk. The theoretical optimality proof and the scalable circuit analysis appear sound and are supported by the derivation in the Supplementary Information. The missing state-resolved loss characterization is a concrete, testable gap: the published data and code would allow a reanalysis once the loss fractions are measured. Because the reader's conditional verdict already reflects this gap, no verdict adjustment is needed; the concern does not warrant rejection, only a condition that the loss dependence be characterized.","tokens_in":22653,"tokens_out":4123,"duration_ms":40963,"concrete_test":"Measure per-state loss fractions A_0, A_1, A_2 by preparing single atoms in each qutrit level with the same circuit depth but without entangling operations, then recording survival after the same state-selective detection sequence. Use these fractions to reweight the published postselected outcome counts by p_mn → p_mn A_m A_n / Σ p_ab A_a A_b and recompute F from Eq. (S10) and the MLE variance from Eq. (S38)–(S39) with the corrected likelihood. If the corrected F drops to ≤2/3 or the corrected ν[(Δθ1)^2+(Δθ2)^2] rises to ≥1, the central claims fail; if both thresholds remain intact, the conditional verdict can be lifted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The two quantitative claims that carry the paper—F=0.85(1)>2/3 in Fig. 2 and ν[(Δθ1)^2+(Δθ2)^2]=0.82(3)<1 in Fig. 3—are computed on a loss-postselected ensemble. The main text discloses that 15–20% of atom pairs are discarded, and the Supplementary Information specifies that 14.5% of the 29,594-shot metrology dataset is discarded. However, no measurement of loss probability as a function of the internal qutrit state |0>, |1>, |2> is reported. The detection protocol classifies an atom as lost only if none of the three state-selective images detects population; an atom in |2> undergoes an extra repump-and-image cycle, so its loss probability need not equal that of |0> or |1>. State-dependent loss factors A_m transform the true no-loss outcome distribution p_mn into the observed postselected distribution proportional to p_mn A_m A_n. The fidelity formula F=(P00+P11+P22+C01+C02+C12)/3 and the phase-scan calibration of Eqs. (S36)–(S37) are built from these postselected frequencies. If A_0, A_1, A_2 differ, F is not the fidelity of the prepared state, and the likelihood used for the MLE and bootstrap is misspecified. A biased estimator can have a bootstrap variance below the unbiased Cramer-Rao bound, so νY=0.82(3)<1 would not constitute a valid violation of the ideal individual-estimation bound. The paper's independently calibrated green-curve model also implicitly assumes loss is state-independent, since it contains no loss terms. This unverified assumption is load-bearing: without state-resolved loss data, neither the entanglement certificate nor the metrological advantage is established for the actually prepared state.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports an experiment with 88Sr atoms in optical tweezers implementing a two-qutrit system using the ground state and two fine-structure clock states. A controlled-Z gate and global single-qutrit rotations are used to prepare a maximally entangled two-qutrit state, with a loss-postselected fidelity F = 0.85(1) > 2/3, certifying Schmidt number three. For metrology, the authors design an unbalanced GHZ-like two-qutrit probe and a noise-robust readout circuit, and estimate two injected phases jointly via maximum likelihood. They report ν[(Δθ1)^2 + (Δθ2)^2] = 0.82(3) < 1, below the ideal individual two-level sensing threshold, and provide theoretical scaling to larger atom numbers under circuit-level noise. The Supplementary Information contains the derivation of the optimal probe, the readout optimization, the likelihood model, and the bootstrap procedure.","tokens_in":23042,"tokens_out":15005,"duration_ms":144058,"significance":"If the results hold, this is the first demonstration of genuine internal-level qutrit entanglement between neutral atoms and the first use of such states for joint multi-parameter estimation in an optical clock platform. The theoretical framework is solid: the optimal probe state is derived from the quantum Fisher information matrix, the saturation of the quantum Cramér–Rao bound is demonstrated for the chosen readout, and the scaling analysis is clearly specified. The paper also provides public data and code. The main caveat is that both quantitative claims are computed on a loss-postselected ensemble without characterizing state-dependent loss, which could bias the fidelity and the variance. This issue is addressable but currently leaves the central claims conditional.","major_comments":[{"comment":"The two central claims rest on a loss-postselected ensemble, but the dependence of loss on the qutrit level is not characterized. The main text reports discarding 15–20% of atom pairs, and the SI reports N_shot = 29,594 with 14.5% discarded. The detection protocol gives an atom in |2> an extra repump-and-image cycle, so its loss probability need not equal that of |0> or |1>. State-dependent loss factors A_m transform the true no-loss outcome distribution p_mn into the postselected distribution proportional to p_mn A_m A_n. Consequently, the fidelity F = (P00+P11+P22+C01+C02+C12)/3 and the likelihood in Eqs. (S36)–(S38) are computed from a biased distribution, and the bootstrap variance in Eq. (S39) can lie below the Cramér–Rao bound for a biased estimator. The Schmidt-number certification and the claim νY = 0.82(3) < 1 therefore require either a measurement of A_0, A_1, A_2 or an explicit error model bounding the resulting bias. Please provide state-resolved loss data or a quantitative robustness analysis.","section":"Main text (Fig. 2 and 'Joint parameter estimation') and SI ('Maximum likelihood estimator and bootstrapping')"},{"comment":"The reported figure of merit is a bootstrap variance, not a mean squared error, and the Cramér–Rao bound used as the threshold applies to unbiased estimators. Equation (S39) computes the sample covariance of the bootstrap MLE but does not report the bootstrap bias relative to the injected phases. If the likelihood model in Eq. (S38) is misspecified—for example because of state-dependent loss or the additive phase-scan model in Eq. (S37)—the estimator can be biased and its variance can fall below the bound even when the mean squared error does not. Please report the bias of the MLE at the operating point and either use mean squared error or explicitly justify unbiasedness.","section":"SI ('Maximum likelihood estimator and bootstrapping')"}],"minor_comments":[{"comment":"The symbol ν is used first as the number of repetitions and later as the factor multiplying the experimental variance; the text should explicitly state that ν = N_shot for the reported result.","section":"Main text, 'Joint parameter estimation'"},{"comment":"The horizontal axes are labeled θ1 and θ2 but the vertical axis is only labeled 'Normalized variance'; the caption should spell out the meaning of the shaded regions, the green curves, and the 'Individual est. bound' line.","section":"Fig. 3(d,e)"},{"comment":"The claim of a variance 'below the ideal individual two-level sensing threshold' should specify that this is for a two-atom pair, not for the full 12-pair array, to avoid ambiguity in the resource accounting.","section":"Abstract and main text, 'Joint parameter estimation'"},{"comment":"The standard inequality F_ρ ≤ 4Γ_ρ for mixed states is stated as an equality chain; explicitly marking it as an inequality with a citation would prevent a reader from assuming equality holds for all states.","section":"SI, Eq. (S42)"},{"comment":"The main text says 15–20% of pairs are discarded while the SI says 14.5% for the metrology dataset; please clarify that these are dataset-specific postselection fractions.","section":"SI, 'Maximum likelihood estimator and bootstrapping'"},{"comment":"The green-curve agreement is reported visually; adding a quantitative goodness-of-fit or residual plot would strengthen the claim that the two-phase-offset model captures the data.","section":"Fig. 3(c)"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong experimental demonstration with a solid theory supplement. The main concern is the uncharacterized state-dependent loss; if the authors can provide state-resolved loss measurements or bound the resulting bias, the results would likely be acceptable. I do not see a fundamental flaw in the derivations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth a serious referee. The experiment does something new that the field will care about: genuine qutrit entanglement between individually controlled neutral atoms in their internal energy levels, followed by a proof-of-principle joint two-phase estimation that beats the ideal individual two-level bound. The reported numbers, F = 0.85(1) versus the 2/3 Schmidt-number threshold and nuY = 0.82(3) versus the individual bound of 1, are consistent with each other and with the independently calibrated error model in the SI. The optimal probe derivation from the quantum Fisher information is straightforward but correct, and the scalable circuit analysis is a useful addition for future work. Data and code are available, which is good practice.\n\nThe main soft spot is the loss postselection, and the stress-test note puts its finger on the right mechanism: the |2> state requires an extra repump-and-image cycle, so its loss probability need not equal that of |0> and |1>. The paper discloses the 15-20% discard but gives no state-resolved loss data. That means the certified fidelity and the metrological variance are properties of the postselected ensemble, not directly of the state produced by the circuit. I do not think this is fatal. The margin above 2/3 is large enough that even a few percent of state-dependent loss would not overturn the entanglement verdict, and the phase-scan calibration of the likelihood is performed on the same postselected data, so the MLE is at least self-consistent for that ensemble. But the revision should supply the loss-versus-state characterization and, ideally, a worst-case analysis of how much the fidelity and variance could shift under the observed loss imbalance.\n\nThe second caveat is that the metrology uses injected phases with zero dark time, no systematic effects, and postselection. The authors state this limitation explicitly in the conclusion, so it is not hidden. It does mean the paper establishes a capability, not a clock, and the claim should be read that way.\n\nThe citation pattern is appropriate; the self-citations are to the group's own prior control and apparatus papers, which is expected here. Theory and experiment are internally consistent, and the limitations are stated in the text.\n\nMy recommendation: send this to peer review. It deserves referee time despite the caveats. Ask the authors to add state-resolved loss data and to discuss the effect of postselection on the metrological advantage; that should be enough to move this from conditional to solid.","headline":"A genuine first: neutral-atom internal-level qutrit entanglement with a credible joint two-phase estimation result, but the uncharacterized loss postselection keeps the central claims conditional.","tokens_in":23626,"tokens_out":2665,"would_cite":true,"duration_ms":28694,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P40","81P45","81P50"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper reports an experimental demonstration of genuine three-level (qutrit) entanglement between pairs of strontium-88 atoms in an optical-clock array, together with a joint two-phase estimation protocol that achieves a variance…","keywords":["qutrit entanglement","multiparameter quantum metrology","optical clock","strontium-88","Rydberg entanglement","triple-magic optical tweezers","quantum Fisher information","joint phase estimation"],"falsifier":"Measure the loss probability per experimental cycle separately for atoms prepared in $|0\\rangle$, $|1\\rangle$, and $|2\\rangle$; if these probabilities differ by more than the statistical uncertainty, re-analyse the postselected data without assuming state-independent loss and check whether $F>2/3$ and $\\nu[(\\Delta\\theta_1)^2+(\\Delta\\theta_2)^2]<1$ survive.","tokens_in":22428,"feed_emoji":"⏱️","tokens_out":7803,"duration_ms":59488,"temperature":0.7,"pith_summary":"The paper claims to have created genuine three-level (qutrit) entanglement between pairs of strontium-88 atoms inside an optical-clock platform, and to have used that entanglement to estimate two optical-clock phases simultaneously with a precision that beats what perfect two-level (qubit) probes could achieve. The authors report a loss-postselected state fidelity $F = 0.85(1)$, which exceeds the $2/3$ threshold that proves the state cannot be explained by qubit-level entanglement, and a joint estimation cost $\\nu[(\\Delta\\theta_1)^2+(\\Delta\\theta_2)^2] = 0.82(3)$, below the ideal qubit bound of $1$. If these results hold, they establish the internal energy levels of neutral atoms as a workable high-dimensional resource for both entanglement-based quantum information and multi-parameter metrology, and they identify an optimal, deliberately unbalanced entangled state that yields a fixed $\\sim0.48\\times$ advantage over individual qubit estimation regardless of atom number. A sympathetic reader should care because this is the first demonstration of genuine high-dimensional internal-state entanglement between individually controlled neutral atoms, and it turns an actual clock platform into a multi-parameter sensor.","feed_headline":"Joint multi-parameter sensing beats qubit limit in an atomic clock","feed_subtitle":"Strontium pairs entangle three levels and measure two phases with variance 0.82, below the ideal qubit bound.","key_machinery":"The machinery that carries the argument is a triple-magic all-optical qutrit in $^{88}$Sr: the ground state and two fine-structure clock states are all trapped at a common magic wavelength by angle-tuning a magnetic field, giving coherence among three levels, while global Raman and three-photon pulses provide arbitrary single-qutrit rotations $R_{01}$ and $R_{12}$, a Rydberg-excitation controlled-$Z$ gate entangles atom pairs, and a three-step imaging sequence reads out each of the three levels while flagging lost atoms. The metrological key is an unbalanced GHZ-like probe state $\\sqrt{p_0}|0\\rangle^{\\otimes N}+\\sqrt{p_1}|1\\rangle^{\\otimes N}+\\sqrt{p_2}|2\\rangle^{\\otimes N}$ with populations $p_0=p_2=(10-\\sqrt{10})/18$ and $p_1=(\\sqrt{10}-1)/9$, which is proven optimal for the two commuting generators, plus a readout sequence that saturates the quantum Cram\\'er\\,Rao bound locally while remaining robust to the experimentally asymmetric errors of the $R_{01}$ and $R_{12}$ gates. The fidelity certification uses the Schmidt-number bound: any state whose Schmidt number is at most two has overlap at most $2/3$ with a maximally entangled two-qutrit state, so the measured $F=0.85(1)$ witnesses genuine qutrit entanglement.","core_discovery":"The central experimental discovery is that a pair of $^{88}$Sr atoms, each encoded in the three internal levels $|0\\rangle={}^1S_0$, $|1\\rangle={}^3P_0$ and $|2\\rangle={}^3P_2(m_J=0)$, can be entangled into a state with fidelity $F = 0.85(1)$ to the maximally entangled two-qutrit state $(|00\\rangle+|11\\rangle+|22\\rangle)/\\sqrt{3}$, certifying genuine Schmidt-number-3 entanglement. The central theoretical discovery is that for estimating two commuting phases generated by $g_1=(|0\\rangle\\langle0|-|1\\rangle\\langle1|)/\\sqrt2$ and $g_2=(|1\\rangle\\langle1|-|2\\rangle\\langle2|)/\\sqrt2$, the optimal $N$-qutrit probe is an unbalanced GHZ-like state with populations $p_0=p_2=(10-\\sqrt{10})/18$, $p_1=(\\sqrt{10}-1)/9$, achieving joint variance $Y_{\\rm joint}=(11+2\\sqrt{10})/(9N^2)\\approx0.4812/N^2$, about half the ideal individual two-qubit bound $4/N^2$. The authors verify the two-qutrit case with the noise-robust readout $V_{\\rm ro}=R_{01}(v_3,\\phi_3)R_{12}(v_2,\\phi_2)CZ R_{12}(v_1,\\phi_1)$ and maximum-likelihood estimation on a postselected 29,594-shot dataset, obtaining $\\nu[(\\Delta\\theta_1)^2+(\\Delta\\theta_2)^2]=0.82(3)<1$. They further show numerically that the joint strategy remains favourable for large atom numbers under per-gate depolarizing and dephasing noise, provided gate infidelities are at the $10^{-4}$ level or better.","pith_inferences":["Our extension: because the optimal probe concentrates probability on the two outer states $|0\\rangle$ and $|2\\rangle$, the protocol is naturally suited to clock configurations where the $|1\\rangle$ level is short-lived or noisier; one could test whether engineering the levels to put the noisiest transition in the middle improves robustness.","Our extension: the same covariance-based optimality argument should generalise to $d$-level atoms estimating $d-1$ adjacent phase differences; a natural next test is whether the $\\approx2\\times$ resource advantage persists for estimation of three or more non-commuting phases, where the quantum Cram\\'er\\,Rao bound is not always saturable.","Our extension: the 15\\,--\\,20\\% loss postselection could be turned into a feature if loss is partially state-dependent, since the discarded fraction itself would carry metrological information; one could test a loss-tolerant estimator that includes the loss channel in the likelihood."],"forward_implications":["Joint estimation with the optimal qutrit state beats splitting an array into two qubit subensembles by a constant factor $0.4812$ for any atom number in the noiseless case.","The demonstrated variance $0.82(3)<1$ shows that even with experimental imperfections and loss postselection, the joint qutrit strategy outperforms ideal individual two-level sensing for pairs of atoms.","Under circuit-level depolarizing and dephasing noise, the advantage of joint estimation persists for state-of-the-art numbers of entangled atoms (dozens) when gate infidelities are at the $10^{-4}$ level, guiding scalable future demonstrations.","The same triple-magic, Rydberg-entangling, and state-resolved readout building blocks provide the toolkit for higher-dimensional quantum information processing in neutral atoms, including qudit error correction and simulation."],"supporting_citations":[{"why":"Supplies the triple-magic optical-tweezer trapping of the three $^{88}$Sr levels and the basic qutrit coherence that the experiment builds on.","marker":"[37]"},{"why":"Provides the Rydberg controlled-$Z$ entangling operation used to create the entangled qutrit pairs.","marker":"[8]"},{"why":"Provides the universal global $R_{01}$ gates and state-resolved readout used for qutrit control and measurement.","marker":"[51]"},{"why":"Establishes the theoretical joint multi-phase estimation advantage over individual estimation that the probe states are designed to reach.","marker":"[46]"},{"why":"Gives the entanglement-certification framework used to interpret $F>2/3$ as Schmidt number three.","marker":"[45]"},{"why":"Provides the Schmidt-number bound stating that fidelity above $2/3$ to a maximally entangled two-qutrit state implies genuine qutrit entanglement.","marker":"[53]"},{"why":"Supplies the ideal individual two-level GHZ benchmark ($4/N^2$) and the 29,594-shot bootstrap dataset methodology.","marker":"[16]"}],"fun_headline_variants":["Entangled qutrits beat qubit limit in optical clock","Joint phase estimation on qutrits outperforms qubit bound","Optical clock qutrits achieve joint sensing below qubit limit","Two-qutrit entanglement enables joint clock metrology","Strontium qutrits beat qubit sensing limit in clocks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Atom loss during the experiment is independent of the internal state of each atom, so that discarding pairs with lost atoms does not bias the certified fidelity or the joint-estimation variance.","fun_headline_variants_meta":{"raw":{"variants":["Entangled qutrits beat qubit limit in optical clock","Joint phase estimation on qutrits outperforms qubit bound","Optical clock qutrits achieve joint sensing below qubit limit","Two-qutrit entanglement enables joint clock metrology","Strontium qutrits beat qubit sensing limit in clocks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0005,"raw_usage":{"total_tokens":2574,"prompt_tokens":1200,"completion_tokens":1374,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":816,"completion_tokens_details":{"reasoning_tokens":1288}},"tokens_in":816,"tokens_out":1374,"duration_ms":9942,"temperature":1.0,"reasoning_tokens":1288,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T13:23:47.707084+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the loss probability per experimental cycle separately for atoms prepared in $|0\\rangle$, $|1\\rangle$, and $|2\\rangle$; if these probabilities differ by more than the statistical uncertainty, re-analyse the postselected data without assuming state-independent loss and check whether $F>2/3$ and $\\nu[(\\Delta\\theta_1)^2+(\\Delta\\theta_2)^2]<1$ survive.","supporting_citations":[{"cited_title":"Ammenwerth, H","cited_arxiv_id":null,"evidence_quote":"Supplies the triple-magic optical-tweezer trapping of the three $^{88}$Sr levels and the basic qutrit coherence that the experiment builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the universal global $R_{01}$ gates and state-resolved readout used for qutrit control and measurement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the theoretical joint multi-phase estimation advantage over individual estimation that the probe states are designed to reach."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Schmidt-number bound stating that fidelity above $2/3$ to a maximally entangled two-qutrit state implies genuine qutrit entanglement."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the ideal individual two-level GHZ benchmark ($4/N^2$) and the 29,594-shot bootstrap dataset methodology."}],"review_version":1}