{"id":"03ffc6c5-cb91-4fa2-bc7b-fb85a64acb25","arxiv_id":"2412.14079","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"A metallic SNS sensor measures 1 microsecond, 8.4 GHz microwave pulses with a FWHM energy resolution finer than 0.95 zeptojoule, the best calorimetric resolution demonstrated to date.","lead":"The paper demonstrates a cryogenic sensor that measures microwave pulses carrying about 170 photons, with an energy resolution finer than 0.95 zeptojoule, a record for calorimeters. A generalist should care because this is a concrete step toward detecting individual microwave photons, which could improve quantum computing readout and dark matter searches.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (12) uses an incorrect second-order variance transform; the negative σ_S^4 E''^2 term biases the FWHM low and may inflate the 0.95 zJ signal-to-FWHM ratio.","rationale":"The reader identified the line-attenuation calibration as the weakest assumption. That is a legitimate external systematic, but the paper explicitly quotes its uncertainty and the claim would survive calibration errors up to ~0.58 dB. The more load-bearing issue is internal: Equation (12), used to convert the matched-filter-output noise into energy units, is not the correct second-order variance transformation. The printed formula yields σ_E^2 = σ_S^2 E'^2 − (σ_S^4/4)E''^2, whereas the standard delta method for a Gaussian variable gives σ_E^2 = σ_S^2 E'^2 + (σ_S^4/2)E''^2. The negative sign means the reported FWHM is biased low whenever the calibration curve E(s) is nonlinear, and the bias scales as σ_S^4. At 0.95 zJ, the signal-to-FWHM ratio of 1.17 implies σ_S/μ_S ≈ 0.36, so the fourth-order term is not obviously negligible; if E'' is large enough, the corrected σ_E could reduce the ratio below unity. Since the paper's uncertainty budget uses the same incorrect formula, the quoted error bars do not cover this bias. The test is straightforward because the matched-filter outputs and the calibration curve are both available: apply E(s) to each single-shot S_j and measure the FWHM of the resulting distribution in energy units. If the reanalysis confirms S/FWHM > 1 at 0.95 zJ, the central claim stands; otherwise the resolution bound is overstated. I therefore recommend CONDITIONAL acceptance pending this reanalysis. Credit is due for the direct single-shot distributions, the matched-filter improvement, and the Zenodo data/code deposit, which make the check feasible.","tokens_in":19309,"tokens_out":20833,"duration_ms":169651,"concrete_test":"Recompute σ_E for the 0.95 zJ pulse directly from the single-shot matched-filter outputs: apply the fitted calibration curve E(s) from Eq. (11) to each S_j, build the empirical CDF in energy units, and extract the FWHM either by an error-function fit to that CDF or from sample quantiles. Compare with the paper's Eq. (12) result and with the correct second-order delta method. Also inspect the Zenodo analysis code to verify whether Eq. (12) is implemented as printed or with the correct +σ_S^4 E''^2/2 term. If the directly computed signal-to-FWHM ratio remains above 1, the sub-zeptojoule bound survives; if not, the headline claim is not supported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central bound depends on the FWHM in energy units, which is computed from σ_E via Eq. (12). That formula simplifies to σ_E^2 = σ_S^2 E'(μ_S)^2 − (σ_S^4/4) E''(μ_S)^2. For a Gaussian signal transformed by a twice-differentiable function, the second-order delta method gives σ_E^2 = σ_S^2 E'(μ_S)^2 + (σ_S^4/2) E''(μ_S)^2. The printed formula therefore underestimates σ_E whenever the calibration curve E(s) has curvature, with the bias growing as σ_S^4. At the 0.95 zJ point, S/FWHM = μ_S/(2.355σ_S) = 1.17 implies σ_S/μ_S ≈ 0.36, so fourth-order terms are not automatically negligible. If E(s) is significantly curved over ±σ_S around the 0.95 zJ mean, the true FWHM is larger than reported, and the signal-to-FWHM ratio could drop below the claimed 1.17, potentially below unity. The uncertainty propagation in Eqs. (13)–(16) is built on this same flawed transform, so the quoted ±0.05 does not cover the bias. This is an internal, testable error rather than an external systematic; the line-attenuation calibration is a separate concern that the reader already identified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports calorimetric detection of 1-µs, 8.4 GHz microwave pulses with an SNS-based sensor, claiming a FWHM energy resolution finer than 0.95 zJ at 20 mK. The authors characterize the noise equivalent power in bolometric mode, then collect 1000 single-shot traces per pulse energy between 0.95 and 3.8 zJ, apply a matched filter built from a double-exponential template, fit empirical CDFs to extract signal means and standard deviations, and convert the signal spread into energy units using an inverse-Lorentzian calibration curve. The central claim is that the signal-to-FWHM ratio at 0.95 zJ is 1.17 ± 0.05, implying sub-zeptojoule resolution; interpolation of the data gives an estimated resolution of 0.83 ± 0.04 zJ.","tokens_in":19658,"tokens_out":16018,"duration_ms":130155,"significance":"If the central claim survives scrutiny, this is a substantial advance: it would be the first direct calorimetric measurement with sub-zeptojoule FWHM resolution, roughly an order of magnitude better than the previous best calorimeter (17.6 zJ for a Ti TES). The paper also ships reproducible data and code on Zenodo, uses matched filtering to improve the signal-to-noise ratio, and provides a transparent accounting of statistical uncertainties. The main reservation is the variance-transformation error discussed below, which directly affects the quoted signal-to-FWHM ratio and therefore the headline resolution.","major_comments":[{"comment":"Equation (12) is an incorrect second-order variance transform. Expanding the printed expression gives σ_E^2 = σ_S^2 [E'(μ_S)]^2 − (σ_S^4/4)[E''(μ_S)]^2. For a Gaussian signal transformed by E(s), the correct second-order delta method gives σ_E^2 = σ_S^2 [E'(μ_S)]^2 + (σ_S^4/2)[E''(μ_S)]^2 (plus higher-order terms). The printed formula even yields a negative variance for a simple quadratic transformation Y = X^2 with X ∼ N(0,σ^2), so the sign error is unambiguous. Because the FWHM W_E in Fig. 4b and the interpolated energy resolution in Extended Data Fig. 3 are computed directly from σ_E via Eq. (12), the negative σ_S^4 term biases σ_E low and inflates the signal-to-FWHM ratio. At the 0.95 zJ point, σ_S/μ_S ≈ 0.36, so σ_S^4 is not negligible, and the inverse-Lorentzian calibration curve E(s) has nonzero curvature. The authors must recompute σ_E, W_E, and the signal-to-FWHM ratio with the correct transform and determine whether the value 1.17 remains above unity. The uncertainty propagation in Eqs. (13)–(16) inherits the same error and should be revised accordingly.","section":"Methods, Eq. (12)"},{"comment":"The absolute energy scale of every pulse, including the central 0.95 zJ point, rests on the total input line attenuation of 119.24 ± 0.1 dB, calibrated in a separate thermal cycle using a different bolometer. The statement 'We consider the change in line attenuation between the thermal cycles of the cryostat to be negligible' is an assumption, not a measurement. Since a change of only 0.1 dB shifts the pulse energy by about 2.3%, and a larger drift would directly affect the claimed bound, the authors should provide support for this assumption, for example a stability check with a known source in the same cycle or a sensitivity analysis of the conclusion to the attenuation value.","section":"Methods, total input line attenuation"}],"minor_comments":[{"comment":"The abstract says 'corresponding to 170 photons at 8.4 GHz' while the Results section quotes '171 ± 4' for the 0.95 zJ pulse and '150 ± 7' for the interpolated 0.83 zJ value; the numbers should be harmonized.","section":"Abstract and Results"},{"comment":"The presentation of Eq. (6) is difficult to parse because the integrand appears as a fraction with '4' over 'NEP^2(f)'. Please clarify whether the intended formula is ∫ 4/NEP^2(f) df or ∫ 4 NEP^2(f) df, and add a brief dimensional check.","section":"Methods, Eq. (6)"},{"comment":"The figure caption states that error bars denote one-standard-deviation confidence intervals, but it is not stated whether the 0.1 dB line-attenuation uncertainty is included in these intervals; please state this explicitly.","section":"Fig. 4b"},{"comment":"The template K(t) is extracted from averaged 3.8 zJ pulses, and the text asserts that the nonlinearity 'does not significantly affect' the filtering for the pulse energies considered, but no quantitative comparison is given; a short analysis of how the filtered SNR changes when the template is derived at a lower energy would strengthen the presentation.","section":"Methods, matched-filter template"},{"comment":"The sentence 'we find that the noise in the output signal is primarily arising from the amplification chain' is important for the interpretation but is supported only qualitatively; consider showing the measured noise PSD relative to the expected amplifier noise floor in the main text or Extended Data.","section":"Results and analysis"}],"recommendation":"major_revision","confidential_remarks":"The reader's report recommended accept, but the sign error in Eq. (12) is a serious technical flaw that was missed. I would like the editor to ensure the authors address it head-on. The dataset and code on Zenodo should allow a quick re-analysis; if the corrected signal-to-FWHM ratio at 0.95 zJ remains above unity, the paper is a strong candidate for acceptance. The line-attenuation stability assumption is a second concern that should be documented or bounded."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline result—a calorimetric energy resolution finer than 0.95 zJ for microwave pulses, roughly an order of magnitude better than the previous 17.6 zJ record—is a real achievement. The paper backs it with direct single-shot distributions (1000 traces per input energy), a careful input-line attenuation calibration, and a matched-filter analysis that is honestly described, including the empirically fitted double-exponential template whose microscopic origin remains unknown. The promise of data and code on Zenodo is good practice. I read this as the first direct calorimetric demonstration below one zeptojoule, not just an extrapolation from steady-state NEP.\n\nBut there is a load-bearing math error. Equation (12) is not the second-order delta method; for a Gaussian variable it should be σ_E² ≈ σ_S² [E′(μ)]² + (σ_S⁴/2)[E″(μ)]². The printed formula gives a negative (σ_S⁴/4)[E″]² term after simplification, so it underestimates the energy variance whenever the calibration curve has curvature. At the 0.95 zJ operating point σ_S/μ_S ≈ 0.36, so fourth-order terms are not negligible. The quoted signal-to-FWHM of 1.17 ± 0.05 is derived through this transform, as is the uncertainty propagation, so the bias is not covered by the error bars. The authors need to redo the variance conversion with the correct formula and check whether the ratio at 0.95 zJ stays above unity. If it drops below, the 'finer than 0.95 zJ' claim fails and the interpolated 0.83 zJ is also off. This is testable with the deposited data.\n\nA secondary concern: the total line attenuation (119.24 ± 0.1 dB) is calibrated in a separate cooldown with a different bolometer. That is a standard and reasonable approach, but it is an unverified assumption that the attenuation is identical in the main run. The paper could say a bit more about the sensitivity of the result to this assumption.\n\nOn balance: the experimental work is solid and the paper is transparent. The variance-transform error is internal, not a fundamental flaw in the measurement. Send it to peer review and require the corrected analysis. If the corrected numbers still put the resolution below a zeptojoule, this is a high-impact result. If not, the conclusion weakens but the methods and data are still a useful contribution to the field.","headline":"A real experimental milestone, but Eq. (12) uses the wrong variance transform and the central sub-zeptojoule claim needs reanalysis before it stands.","tokens_in":20248,"tokens_out":5553,"would_cite":true,"duration_ms":45719,"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 calorimeter resolves microwave pulses under one zeptojoule","keywords":["calorimetry","zeptojoule","energy resolution","SNS Josephson junctions","matched filtering","microwave photon detection","cryogenic bolometer","noise equivalent power"],"falsifier":"Measure the same pulse sequence with an independent, traceable calibration of the power delivered to the chip, and check whether the signal-to-FWHM ratio at the nominal $0.95$ zJ input remains above unity; if the true on-chip energy is even $0.1$ dB lower than assumed, the claimed bound fails. The known Poisson variance of coherent input states could be used in the same test to check that the extracted energy scale matches the calibrated scale.","tokens_in":19087,"feed_emoji":"🌡️","tokens_out":9885,"duration_ms":74123,"temperature":0.7,"pith_summary":"The paper claims that an SNS (superconductor–normal-conductor–superconductor) thermal sensor, operated as a calorimeter rather than a bolometer, can resolve the energy of individual 1-microsecond, 8.4 GHz microwave pulses down to a full-width-at-half-maximum (FWHM) resolution finer than $0.95$ zeptojoule, about $170$ photons at that frequency. This matters because no calorimeter has previously demonstrated energy resolution in the single-zeptojoule range; the best earlier calorimetric result cited is $17.6$ zJ. The claim is obtained directly from single-shot time traces processed with a matched filter, not merely inferred from steady-state noise-equivalent-power measurements. If correct, the same sensor architecture with a graphene absorber is a route toward real-time calorimetric detection of single microwave photons near 10 GHz.","feed_headline":"A calorimeter resolves microwave pulses under one zeptojoule","feed_subtitle":"Direct 1-µs, 8.4 GHz pulse measurements reach a FWHM resolution of 0.95 zJ, about 170 photons.","key_machinery":"The carrying object is the SNS radiation sensor: a micron-long AuPd normal-metal nanowire that absorbs the microwave pulse and acts as the thermal mass, capacitively shunted to a series array of SNS Josephson junctions whose inductance changes with electron temperature, forming an LC oscillator whose resonance frequency shifts with absorbed power. The argument is carried by the matched filter, which weights each single-shot trace's Fourier components by the inverse noise power spectral density and a template of the pulse response, improving the signal-to-noise ratio by more than 30 percent relative to simple averaging, and by the nonlinear calibration curve that converts arbitrary filtered-signal units into energy units under the Lorentzian reflection model.","core_discovery":"On the paper's own terms, the central discovery is that an SNS junction thermometer coupled to a metallic nanowire absorber measures $1$-$\\mu$s, $8.4$ GHz microwave pulses with a FWHM energy resolution finer than $(0.95 \\pm 0.02)$ zJ $= (5.9 \\pm 0.12)$ meV, corresponding to about $171 \\pm 4$ photons. The authors record 1000 un-averaged traces per pulse energy, apply a matched filter whose template is a double-exponential fit to averaged high-energy pulses, fit the empirical cumulative distribution functions to error functions, and convert the fitted signal spread into energy units through a Lorentzian calibration model. At the lowest applied pulse energy, $0.95$ zJ, the signal-to-FWHM ratio is $1.17 \\pm 0.05$, so the resolution must be finer than that input energy. Interpolating the converted standard deviations gives an estimated resolution of $(0.83 \\pm 0.04)$ zJ, consistent with the $1.03$ zJ estimate obtained from the independently measured noise equivalent power.","pith_inferences":["If the input-line attenuation calibration were in error by more than its stated $0.1$ dB, the $0.95$ zJ bound would shift proportionally; an independent, traceable power calibration would settle this.","Because the input pulses are coherent states with known Poisson photon statistics, the measured width-versus-energy curve could serve as an internal cross-check of the energy calibration, a check the paper does not perform.","The interpolated $0.83$ zJ estimate depends on the nonlinear Lorentzian model; a direct measurement at a pulse energy where the signal-to-FWHM ratio is near unity would confirm or refute that interpolation.","The same calorimetric readout with matched filtering could serve as an energy-resolving discriminator of superconducting qubit states, extending the authors' earlier thermal-detector readout work."],"forward_implications":["If the claim holds, this is the first direct calorimetric measurement with sub-zeptojoule FWHM energy resolution, roughly an order of magnitude below the previous best calorimeter's $17.6$ zJ.","The matched-filter processing transfers directly to other slow thermal detectors, improving their single-shot energy resolution without any change in hardware.","Because the dominant noise is in the amplification chain, adding a quantum-limited parametric amplifier at the millikelvin stage should further improve the same device's resolution.","Combining this readout with a lower-heat-capacity graphene absorber, as the authors argue, gives a concrete route toward real-time calorimetric detection of single microwave photons in the 10 GHz range.","The demonstrated dynamic range is only a few zeptojoules, but the authors point out that a second probe tone or a frequency comb could extend it."],"supporting_citations":[{"why":"Supplies the SNS nanobolometer device and its record noise equivalent power, the platform this work operates in calorimetric mode, and the basis for the 0.75 zJ predicted resolution.","marker":"[18]"},{"why":"Reports the graphene-based SNS bolometer with a predicted 0.05 zJ resolution, the stated path toward single-photon calorimetric detection.","marker":"[19]"},{"why":"Reports the titanium transition-edge sensor with 17.6 zJ resolution, the previous best calorimetric result that this work claims to surpass by an order of magnitude.","marker":"[17]"},{"why":"Provides the cryogenic calibration procedure used to determine the 119.24 ± 0.1 dB total input-line attenuation from which all on-chip pulse energies are computed.","marker":"[44]"},{"why":"Gives the thermal-calorimeter theory, including the noise-equivalent-power to energy-resolution relation and the Lorentzian reflection model used for calibration.","marker":"[9]"},{"why":"Reports earlier zeptojoule pulse detection with similar SNS devices and documents the double-exponential thermal response adopted as the matched-filter template.","marker":"[42]"},{"why":"Supplies the Poissonian photon-number statistics of the coherent input pulses used in the uncertainty analysis of the energy resolution.","marker":"[43]"}],"fun_headline_variants":["Calorimeter reaches 0.95 zeptojoule energy resolution","Microwave pulses measured directly at sub-zeptojoule scale","Zeptojoule calorimetry detects 170-photon microwave pulses","SNS sensor resolves 8.4 GHz pulses at 0.95 zJ energy","Direct calorimetry of microwave pulses reaches sub-zeptojoule resolution"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the separately calibrated input-line attenuation of $119.24 \\pm 0.1$ dB is unchanged during the main experiment, because every on-chip pulse energy, including the $0.95$ zJ input, is derived from that calibration; a secondary premise is that the nonlinear model used to convert signal spread into energy units is correct, which matters more for the interpolated $0.83$ zJ estimate than for the direct bound.","fun_headline_variants_meta":{"raw":{"variants":["Calorimeter reaches 0.95 zeptojoule energy resolution","Microwave pulses measured directly at sub-zeptojoule scale","Zeptojoule calorimetry detects 170-photon microwave pulses","SNS sensor resolves 8.4 GHz pulses at 0.95 zJ energy","Direct calorimetry of microwave pulses reaches sub-zeptojoule resolution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000661,"raw_usage":{"total_tokens":3121,"prompt_tokens":1141,"completion_tokens":1980,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":757,"completion_tokens_details":{"reasoning_tokens":1884}},"tokens_in":757,"tokens_out":1980,"duration_ms":14373,"temperature":1.0,"reasoning_tokens":1884,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:31:05.989258+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the same pulse sequence with an independent, traceable calibration of the power delivered to the chip, and check whether the signal-to-FWHM ratio at the nominal $0.95$ zJ input remains above unity; if the true on-chip energy is even $0.1$ dB lower than assumed, the claimed bound fails. The known Poisson variance of coherent input states could be used in the same test to check that the extracted energy scale matches the calibrated scale.","supporting_citations":[{"cited_title":"Nature 586(7827), 47–51 (2020) https://doi.org/10","cited_arxiv_id":null,"evidence_quote":"Reports the graphene-based SNS bolometer with a predicted 0.05 zJ resolution, the stated path toward single-photon calorimetric detection."},{"cited_title":"Applied Physics Letters 101(5), 052601 (2012) https://doi.org/10.1063/1.4739839","cited_arxiv_id":null,"evidence_quote":"Reports the titanium transition-edge sensor with 17.6 zJ resolution, the previous best calorimetric result that this work claims to surpass by an order of magnitude."},{"cited_title":"Review of Scientific Instruments 94(5), 054710 (2023) https: 23 //doi.org/10.1063/5.0143761","cited_arxiv_id":null,"evidence_quote":"Provides the cryogenic calibration procedure used to determine the 119.24 ± 0.1 dB total input-line attenuation from which all on-chip pulse energies are computed."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Reports earlier zeptojoule pulse detection with similar SNS devices and documents the double-exponential thermal response adopted as the matched-filter template."},{"cited_title":"Correlation measurement of propagating microwave photons at millikelvin","cited_arxiv_id":"2407.05147","evidence_quote":"Supplies the Poissonian photon-number statistics of the coherent input pulses used in the uncertainty analysis of the energy resolution."}],"review_version":1}