{"id":"c881ef5f-dfeb-4539-bae3-aed4192ef246","arxiv_id":"2501.16691","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A tantalum-based fluxonium qubit achieves 96.2% (97.8% with a parametric amplifier) single-shot readout fidelity and 99.6% repeatability, limited by measurement-induced state mixing.","lead":"Researchers tested a high-coherence fluxonium qubit made with tantalum and measured how accurately its state can be read in a single shot. They report readout fidelities near 96-98% and show that the measurement itself can cause the qubit to leak into other states.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline QND fidelity 99.6% is an overestimate by the authors' own admission: the two-outcome repeatability protocol excludes leakage outside the computational subspace, so the abstract's uncaveated QND claim is not supported.","rationale":"The paper's most important claim is the QND fidelity of 99.6%, repeated in the abstract and conclusion and reflected in the title. The reader identified the CKP photon-number calibration as the weakest assumption; that concern affects the reported operating photon numbers and the quantitative back-action curves but not the raw assignment fidelities. The QND overestimate is more load-bearing because it directly concerns the headline number and the paper itself concedes the protocol does not account for leakage outside the computational subspace. The experiment appears honestly executed, the body text contains the caveat, and the raw data are sufficient to test the corrected number. I agree with the reader's CONDITIONAL verdict: the work is plausible and publishable with appropriate qualification, but the abstract should not present 99.6% as a leakage-inclusive QND fidelity without the caveat. Since the reader already recommended conditional acceptance, my concern does not change the verdict category.","tokens_in":11266,"tokens_out":5626,"duration_ms":55846,"concrete_test":"Reanalyze the M2 conditional IQ histograms from Fig. 2(b) with a three-outcome classifier that includes a leakage bin (the IQ region highlighted in Fig. 1(d)). Compute the leakage-inclusive repeatability F_Q^leak = [P(0|0)+P(1|1)+P(L|L)]/3 using the same M1-based conditioning, and also report the leakage fraction P(M2=L|M1=1). If the leakage-inclusive value is substantially below 99.6%, the abstract and conclusion should be revised to either quote the corrected number or explicitly state that 99.6% is a computational-subspace repeatability that overestimates true QND fidelity.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim that the readout has a QND (repeatability) fidelity of 99.6% is computed with a binary classifier on M1/M2 outcomes only. In Fig. 2(b), the authors identify events in the right histogram (M1=1) that lie outside the |g> and |e> blobs and attribute them to leakage; they then state that 'FQ slightly overestimates the true QND fidelity, as the protocol does not account for leakage outside the computational subspace.' Since QND by definition requires the measurement to preserve the computational subspace, the number 99.6% is not a leakage-inclusive QND fidelity in the usual sense. The abstract and conclusion quote it without this caveat. The raw measurement demonstration is not invalidated, but the headline claim as written is stronger than the evidence. A leakage-inclusive repeatability number, or an explicit leakage fraction, is needed to support 'QND fidelity 99.6%'. The CKP calibration issue identified by the reader is real but only relabels photon numbers, leaving the raw assignment fidelities and qualitative back-action trends intact; the QND overestimate directly changes the meaning of a headline number and is acknowledged in the body.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a characterization of dispersive readout for a tantalum-based fluxonium qubit at half-flux bias. The authors measure single-shot assignment fidelities of 96.2% without a Josephson parametric amplifier and 97.8% with a JPA, at operating photon numbers of about 112 and 126 respectively. They also perform two successive measurements and report a QND repeatability fidelity FQ = 99.6%, and they study back-action by measuring readout errors versus photon number and by observing qubit relaxation in the presence of a leaked readout tone. The overall narrative is that high-fidelity readout is possible despite a relatively small dispersive shift, and that the main limitations are state-mixing and leakage outside the computational subspace.","tokens_in":11521,"tokens_out":2636,"duration_ms":27086,"significance":"If the reported numbers are substantiated, the paper is a useful experimental data point for fluxonium readout: it shows that a conventional junction-array super-inductor can support single-shot fidelities above 96% without a parametric amplifier and above 97% with one, with a fast (260 ns) amplifier-assisted readout. The systematic comparison of JPA-off and JPA-on behavior, the study of integration time versus photon number, and the back-action experiments are valuable for the fluxonium community. The paper also deserves credit for explicitly identifying leakage and state-mixing as limiting factors, and for reporting measurement efficiency and noise temperature. The main caveats are that the headline QND fidelity is an overestimate by the authors' own admission, and that the photon-number calibration is transferred from a different operating point; these need to be resolved for the quantitative claims to be fully supported.","major_comments":[{"comment":"The abstract and conclusion quote 'QND (repeatability) fidelity FQ = 99.6%' without qualification, but the body states that 'FQ slightly overestimates the true QND fidelity, as the protocol does not account for leakage outside the computational subspace.' Since QND by definition requires that a measurement preserve the computational subspace, the quoted number is not a leakage-inclusive QND fidelity. This is load-bearing because the QND claim is a headline result. The authors should report a leakage-inclusive repeatability number, or at least an explicit leakage fraction extracted from the same M1/M2 data, and should adjust the abstract and conclusion wording accordingly.","section":"§II.A, Eq. (2) and Fig. 2(b)"},{"comment":"The CKP calibration is performed at integer flux bias with the qubit at 4.85 GHz (Supplementary §II), but all readout experiments are at half-flux bias where the qubit frequency is 328.12 MHz (Table I). The paper assumes the same room-temperature-power to cavity-photon conversion at the operating point without an in-situ calibration. If the conversion differs, the quoted operating photon numbers n ≈ 112 and n ≈ 126 and the photon-number trends in Fig. 3 would shift. The raw assignment fidelities are not invalidated, but the central quantitative readout model and the back-action analysis as a function of n depend on this calibration. The authors should calibrate at the operating point or provide a quantitative argument that the conversion is flux-insensitive.","section":"Supplementary §II (CKP calibration) and §II.A (photon numbers)"},{"comment":"No statistical or systematic uncertainties are reported for the assignment fidelities, the QND fidelity, or the extracted measurement efficiency and noise temperature. With 10,000 repetitions the statistical error on a 97% fidelity is nontrivial at the quoted precision (about ±0.3% at one standard deviation for a binomial proportion), and there are additional systematic effects from threshold choice, residual excited-state population after cooling, and photon-number calibration. The 'best fidelity' claims should be accompanied by error bars or at least by a statement of the dominant systematic uncertainty.","section":"§II.A and Table I"}],"minor_comments":[{"comment":"Typo in the abstract: 'we extract a QND fidelity if 99.6%' should read 'of 99.6%'.","section":"Abstract and conclusion"},{"comment":"The sentence 'the calculated QND fidelity (FQ) for M2 is the same as the heraled assignment fidelity (F) for M2' contains a typo ('heraled'); consider clarifying the distinction between QND repeatability and heralded assignment fidelity, since the latter is a different quantity.","section":"§II.A"},{"comment":"Minor typo: 'The data is plotted or a few different fractional amplitudes' should read 'for a few different fractional amplitudes'.","section":"§II.B"},{"comment":"The caption states that the dashed circle indicates one standard deviation of the IQ distribution, but it would be helpful to state whether the circle is centered on the mean and whether the standard deviation is computed for the |g> or |e> blob; the current figure may be ambiguous to readers.","section":"Fig. 4"},{"comment":"The SNR definition in Eq. (1), SNR = |<Ig> - <Ie>|/(σg + σe), is unusual; more standard definitions use the quadrature sum in the denominator. Since this definition is used to extract efficiency and noise temperature, please clarify why σg + σe is appropriate or supply the standard definition.","section":"Supplementary §III"},{"comment":"The parenthetical 'and R Vijaya)' in the author list appears to be a formatting artifact; the correct name appears to be 'R Vijay' from the corresponding-author email and references.","section":"Author list metadata"}],"recommendation":"major_revision","confidential_remarks":"The paper is experimentally solid in its raw readout characterization, and the main concerns are not about fabrication or data integrity. The QND claim is the most risky part because the authors themselves note the overestimate; the abstract currently overstates what the data support. The CKP calibration transfer is a legitimate concern but likely fixable with an in-situ calibration or a sensitivity analysis. I would not reject the paper, but the authors should address the QND metric and the calibration issue before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuine experimental benchmark paper, not a breakthrough, and it is mostly honest. The best fidelity numbers (96.2% without JPA, 97.8% with, 99.6% repeatability) are new for a tantalum fluxonium and are useful to the field. The strongest part is the careful decomposition of readout error into SNR and mixing terms, and the demonstration that high photon number causes IQ blob merging and leakage beyond the computational subspace—a useful data point for fluxonium readout.\n\nThe main soft spot is one you already spotted: the 99.6% QND fidelity is a binary repeatability number that excludes leakage, and the authors say so in the body but not in the abstract or conclusion. That is a real mismatch, but not a fatal one—the protocol is standard for repeatability benchmarking, and a leakage-inclusive number can be derived from the same data. I would ask them to quote the leakage fraction explicitly and soften the abstract.\n\nThe CKP calibration at integer flux rather than the half-flux operating point is a legitimate concern, but it only relabels the photon-number axis. The raw assignment fidelities, the error decomposition, and the qualitative back-action trends do not depend on that calibration. It matters for quantitative claims like 'n≈112' and the efficiency extraction, so they should either move the calibration or add an in-situ check.\n\nOther minor issues: no error bars on any fidelity numbers (with 10k shots, statistical error is tiny, but systematic uncertainty from histogram fitting and state prep is not quantified), and the back-action/relaxation experiment is presented as qualitative evidence—fine, but the 'indication of leakage' is inferred from saturation levels rather than measured spectroscopically.\n\nCitation pattern looks fine; they cite the relevant fluxonium and readout literature, including the recent large-photon-number fluxonium readout and transmon ionization work. No self-citation issues. The methods are standard, so novelty is moderate but real.\n\nVerdict: worth a serious referee. Send it to review, and ask for a leakage-inclusive QND number or leakage fraction, uncertainties on the main fidelities, and a word on the CKP calibration at the operating bias. The raw measurements are credible and the paper is useful for people working on fluxonium processors.","headline":"Solid experimental readout benchmark for tantalum fluxonium; the 99.6% QND headline is slightly oversold but the body admits the caveat.","tokens_in":12040,"tokens_out":1666,"would_cite":true,"duration_ms":16359,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A tantalum-based fluxonium qubit is read out single-shot with 96.2% assignment fidelity (97.8% with a Josephson parametric amplifier) and a QND repeatability of 99.6%.","keywords":["fluxonium qubit","dispersive readout","QND measurement","Josephson parametric amplifier","measurement-induced state transitions","tantalum qubit","single-shot fidelity","readout back-action"],"falsifier":"Perform the CKP ac-Stark calibration at the half-flux operating point and, in the same run, measure the population beyond $\\{|g\\rangle, |e\\rangle\\}$ with a third readout discriminator that distinguishes leaked levels; if the inferred photon numbers shift significantly or the measured leakage at optimal power is negligible compared with the total assignment error, the back-action and error-budget claims would need revision.","tokens_in":11087,"feed_emoji":"⚛️","tokens_out":8958,"duration_ms":79691,"temperature":0.7,"pith_summary":"This paper reports single-shot dispersive readout of a tantalum-based fluxonium qubit, a superconducting circuit with a highly anharmonic spectrum that is a candidate building block for quantum processors. The authors demonstrate an assignment fidelity of $96.2\\%$ with about $\\bar{n}\\sim 112$ cavity photons and a $2.82\\,\\mu$s integration time, improving to $97.8\\%$ with a Josephson parametric amplifier at $\\bar{n}\\sim 126$ and $260$ ns. A two-measurement protocol gives a QND repeatability fidelity of $99.6\\%$, meaning the same answer is returned on immediate repetition almost every time. The paper also shows that increasing readout power beyond the optimum causes the qubit to leak outside the computational subspace, so fidelity is ultimately limited by state-mixing rather than by signal-to-noise. If these results hold, fast high-fidelity measurement is available for fluxonium-based error correction and reset protocols.","feed_headline":"Tantalum fluxonium read out at 97.8 percent fidelity","feed_subtitle":"Two-pulse check repeats the same answer 99.6 percent of the time; high readout photons leak the qubit out of its computational states.","key_machinery":"The central object is a fluxonium qubit, a superconducting circuit in which a small Josephson junction is shunted by a large inductance, producing a highly anharmonic spectrum; here the capacitor is tantalum and the junctions are aluminium. Readout is dispersive: the qubit is transversely coupled to a 3D copper cavity at $7.167$ GHz, shifting the cavity frequency by $\\chi_{ge}\\approx 1.2$ MHz depending on the qubit state, and a reflected readout pulse is demodulated into IQ quadrature blobs that are thresholded for single-shot assignment. Assignment fidelity is decomposed with double-Gaussian fits: the overlap of the dominant Gaussians gives the SNR error, while the weight in the secondary Gaussians captures state-preparation and state-mixing errors. The CKP (chi-kappa-power) calibration converts room-temperature readout power into cavity photon number via the ac-Stark shift, and QND character is tested with two identical readout pulses $M_1$ and $M_2$ separated by $200$ ns. Back-action is probed with a leaked readout tone of variable amplitude and duration followed by a final ensemble measurement.","core_discovery":"At half-flux bias, where the qubit frequency is $328.12$ MHz, the paper reports that a single-shot dispersive readout of a tantalum/aluminium fluxonium qubit reaches an assignment fidelity of $96.2\\%$ with an average of about $112$ cavity photons and $2.82$ $\\mu$s integration, and $97.8\\%$ with a Josephson parametric amplifier at about $126$ photons and $260$ ns. A two-measurement protocol yields a QND repeatability fidelity $F_Q = 99.6\\%$, with $\\bar{P}(0|0)=0.995$ and $\\bar{P}(1|1)=0.997$. The authors attribute the remaining error mainly to state preparation and state mixing, not to signal-to-noise; with the JPA the SNR error is only about $0.01\\%$. At photon numbers beyond the optimum, the IQ blobs merge and the qubit leaks outside the computational subspace, and a leaked-tone relaxation experiment shows accelerated decay and saturation above the expected equilibrium, evidence of measurement-induced transitions analogous to transmon ionization.","pith_inferences":["A leakage-aware three-outcome readout that classifies $|h\\rangle$ and $|i\\rangle$ as a separate error channel would turn the observed IQ-blob merging into a quantitative leakage rate and a corrected QND fidelity.","Performing the CKP calibration in situ at the half-flux operating point would test whether the photon-number dependence of the back-action curves is quantitatively accurate; the raw single-shot fidelities would stand either way.","Because the optimal photon number is where SNR gain is balanced by the onset of measurement-induced transitions, designs that push the higher fluxonium levels further from the cavity frequency should tolerate larger photon numbers and faster readout.","The same double-Gaussian error decomposition could be applied to other fluxonium readout demonstrations to compare state-mixing rates directly rather than headline fidelities."],"forward_implications":["A tantalum fluxonium with a conventional Josephson-junction-array superinductor can be read out in a single shot at $96.2\\%$ without an amplifier and $97.8\\%$ with one, so high-coherence fluxonium does not need a granular-aluminium superinductor for good readout.","With the JPA, the $260$ ns integration time is about two orders of magnitude shorter than the measured coherence times, making the readout fast enough for iterative quantum error correction.","Since the SNR error is only $0.01\\%$ with the JPA, near-term improvements in readout fidelity must come from better state preparation and from suppressing measurement-induced mixing, not from more amplification.","Pushing the readout photon number above the optimum makes the IQ blobs merge and leaks the qubit out of the computational subspace; keeping the photon number near $\\bar{n}\\sim 126$ preserves QND behavior at $99.6\\%$.","The observed high-power leakage behavior parallels transmon ionization, so theoretical models of measurement-induced transitions in fluxonium are needed to predict and avoid the loss of QND-ness."],"supporting_citations":[{"why":"Provides the chi-kappa-power (CKP) calibration that converts room-temperature readout drive power into cavity photon number, used for all quoted $\\bar{n}$ values.","marker":"[31]"},{"why":"Prior demonstration of QND dispersive readout of a fluxonium at large photon numbers using a granular aluminium superinductor; the baseline this tantalum junction-array device is compared against.","marker":"[3]"},{"why":"Introduced the high-coherence fluxonium qubit platform whose tantalum-based variant is measured here.","marker":"[18]"},{"why":"Showed millisecond coherence in a superconducting qubit, the coherence context that makes this readout fast relative to $T_1$ and $T_2$.","marker":"[19]"},{"why":"Supplies the two-photon sideband cooling protocol used to initialize the qubit from its roughly 35% thermal excited population.","marker":"[27]"},{"why":"Identifies measurement-induced state transitions beyond the rotating-wave approximation, the mechanism invoked to explain high-power back-action.","marker":"[12]"},{"why":"Benchmarking framework for repeated measurements; cited for why the binary-threshold QND fidelity overestimates true QND-ness when leakage is present.","marker":"[32]"},{"why":"Documents measurement-induced transmon ionization, the analogous leakage phenomenon the authors compare their fluxonium back-action to.","marker":"[34]"}],"fun_headline_variants":["Fluxonium QND readout: 97.8% fidelity, 99.6% repeatability","Tantalum fluxonium readout hits 97.8% with parametric amp","High-fidelity QND readout in tantalum fluxonium","Measurement back-action checked in fluxonium QND readout","97.8% readout fidelity in fluxonium with JPA"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conversion from room-temperature readout power to cavity photon number is calibrated at integer flux bias with the qubit at $4.85$ GHz, but all readout experiments run at half-flux bias with the qubit at $328.12$ MHz, and the paper assumes the conversion is unchanged; if it differs, the quoted photon numbers and back-action trends shift, though the raw assignment fidelities do not.","fun_headline_variants_meta":{"raw":{"variants":["Fluxonium QND readout: 97.8% fidelity, 99.6% repeatability","Tantalum fluxonium readout hits 97.8% with parametric amp","High-fidelity QND readout in tantalum fluxonium","Measurement back-action checked in fluxonium QND readout","97.8% readout fidelity in fluxonium with JPA"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000349,"raw_usage":{"total_tokens":1930,"prompt_tokens":989,"completion_tokens":941,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":605,"completion_tokens_details":{"reasoning_tokens":841}},"tokens_in":605,"tokens_out":941,"duration_ms":8322,"temperature":1.0,"reasoning_tokens":841,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T11:22:53.361805+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform the CKP ac-Stark calibration at the half-flux operating point and, in the same run, measure the population beyond $\\{|g\\rangle, |e\\rangle\\}$ with a third readout discriminator that distinguishes leaked levels; if the inferred photon numbers shift significantly or the measured leakage at optimal power is negligible compared with the total assignment error, the back-action and error-budget claims would need revision.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the two-photon sideband cooling protocol used to initialize the qubit from its roughly 35% thermal excited population."}],"review_version":1}