{"id":"f575d5e6-7799-4c26-a07e-9a6303fa495b","arxiv_id":"2412.05482","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Low-power internal quality factor fluctuations in superconducting resonators are attributed to time-varying TLS loss tangent, with relative standard deviation of 13%.","lead":"Superconducting microwave resonators, key parts of quantum computers, show large slow swings in energy loss at low power, varying by about 13 percent over 12 to 16 hours. The new experiments trace these swings to tiny two-level defects in the materials, which could destabilize qubit performance over time.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The attribution to TLS loss-tangent fluctuations assumes fitted Qi is unbiased; intra-sweep resonance-frequency noise (documented in Appendix C) could mimic all three supporting indicators and is never controlled for.","rationale":"The reader's weakest assumption concerns contamination of Fδ0_TLS via the saturation model and QHP stability. That is a legitimate concern, but the paper partially addresses it by showing QHP is stable and HP fluctuations are uncorrelated with LP/MP. My concern is a different contamination path: the S21 fitting itself may convert resonance-frequency noise into apparent Qi fluctuations when the frequency drifts within a sweep. This is particularly relevant because the paper explicitly documents large low-frequency fr noise at low power (Appendix C), and because the proposed artifact would reproduce all three supporting indicators: power/temperature dependence, LP-MP correlation, and stable Qc. The paper's other controls (JPA on/off, two-resonator coherence, Qc stability) do not rule it out. This does not overturn the paper, but it strengthens the conditionality of the loss-tangent attribution, so the existing CONDITIONAL verdict remains appropriate. Agreement with the reader is partial because we identify different weak points, though both concern the integrity of the inferred TLS loss-tangent signal.","tokens_in":13051,"tokens_out":6131,"duration_ms":74953,"concrete_test":"Repeat the low-power 12-hour measurement on the same resonator with at least two VNA configurations: (1) the original sweep settings and (2) a sweep shortened by roughly 10x using faster step dwell / wider IF bandwidth while preserving SNR; optionally (3) a fixed-frequency single-tone or two-tone phase/amplitude tracking measurement calibrated to Qi. If σQi/Qi remains near 13% across all configurations, intra-sweep frequency noise is not the cause. If it drops significantly, the reported loss-tangent fluctuations are partly artifact. An analytical cross-check: simulate S21 sweeps from Eq. (1) using the measured S_fr spectrum with a fluctuating center frequency over the actual sweep duration, and compare the apparent Qi distribution with the observed one. The sweep time should be reported to make this test possible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim relies on the inference that each fitted Qi value is an unbiased instantaneous loss measurement. The S21 traces are fit to the static Lorentzian in Eq. (1), but Appendix C (Fig. A.3) reports substantial low-frequency resonance-frequency noise that grows at low power. If the resonance center shifts by a significant fraction of the linewidth during a single VNA frequency sweep, the circle fit will see a time-averaged, broadened line and report a biased Qi that fluctuates with the frequency noise. Such an artifact would (i) decrease with power and temperature because the measured S_fr decreases, (ii) correlate between LP and MP because the same TLS environment produces both frequency and apparent loss fluctuations, and (iii) leave Qc stable, as observed. The paper convincingly rules out amplifier noise and Qc instability, but never checks the sweep-time dependence or otherwise separates true loss fluctuations from frequency-noise contamination. Since the frequency noise itself is TLS-induced, the broad statement that TLS fluctuates is not threatened, but the specific and load-bearing claim that the Qi fluctuations are variations of the TLS loss tangent - and hence the quantitative Fδ0_TLS extraction and the qubit T1 hypothesis - would be overstated if a material fraction of σQi comes from this measurement artifact.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports large temporal fluctuations in the internal quality factor Qi of superconducting Nb-on-Si coplanar waveguide resonators at low power and low temperature, with relative standard deviation sigma_Qi/Qi of about 13% over 12 to 16 hours. The authors attribute these fluctuations to variations in the TLS loss tangent, supported by the decrease of fluctuations at higher power and temperature, the correlation structure in interleaved low- and medium-power measurements, and the stability of the coupling quality factor. They further track the effective TLS loss tangent F delta0_TLS, characterize its log-normal distribution and its dependence on averaging time, and hypothesize that such loss-tangent fluctuations contribute to background T1 fluctuations in superconducting qubits.","tokens_in":13348,"tokens_out":7721,"duration_ms":81894,"significance":"If the attribution to TLS loss-tangent fluctuations holds, the results are significant for resonator-based loss characterization and for understanding long-time noise in superconducting qubits. The paper's strengths include a well-controlled measurement campaign with multiple diagnostics: an explicit JPA on/off comparison (Appendix D), interleaved two-resonator coherence checks (Appendix E), Qc stability analysis (Appendix I), and reporting of fitting uncertainties. The inclusion of data from several devices, chips, and cooldowns, together with comparison to literature values, makes the universality claim credible. The central quantitative claim, however, depends on the fitted Qi being an unbiased instantaneous loss measurement, and that point is not yet fully established.","major_comments":[{"comment":"The attribution of the observed Qi fluctuations to TLS loss-tangent variations rests on the assumption that each fitted Qi value is an unbiased instantaneous loss measurement. However, Appendix C and Fig. A.3 document substantial low-frequency resonance-frequency noise S_fr that increases at low power, and the paper never checks the impact of this frequency noise on the static Lorentzian fit of Eq. (1). If the resonance frequency shifts by a significant fraction of the linewidth during a single VNA sweep, the circle fit will see a time-averaged, broadened line and report a biased Qi that fluctuates with the frequency noise. Such an artifact would (i) decrease at higher power and temperature because S_fr decreases, (ii) produce correlations between LP and MP fluctuations, and (iii) leave Qc stable, as observed in Fig. A.9. The paper convincingly rules out amplifier noise and Qc instability, but it does not report the sweep duration, compute the expected broadening from the measured S_fr, or examine the cross-correlation between Qi and fr fluctuations. For these reasons, the quantitative extraction of F delta0_TLS in Fig. 3 and the central claim of TLS loss-tangent fluctuations are not yet fully supported. I ask the authors to add at least one of the following: (a) report the VNA frequency span, number of points, IF bandwidth, and sweep time, and estimate the in-sweep frequency jitter from the measured S_fr; (b) compute the signal coherence or cross-spectral density between Qi and fr and show that the two are not strongly correlated at relevant timescales; or (c) repeat a subset of measurements with substantially different sweep durations and demonstrate that the inferred sigma_Qi is independent of sweep time.","section":"Appendix C / Eq. (1)"}],"minor_comments":[{"comment":"The paper states sigma_Qi/Qi = 13% but does not provide an uncertainty estimate for this relative standard deviation, which is a key quantitative result.","section":"Main text / Fig. 1"},{"comment":"The fits to the linear model sigma_Qi proportional to Qi in Fig. 4 and Fig. A.8 are shown without reporting the fitted slopes, confidence intervals, or goodness-of-fit; reporting these would help the reader judge the claimed universality.","section":"Fig. 4 / Appendix H"},{"comment":"The data availability statement limits access to 'available from the corresponding author upon reasonable request'; making the raw time traces and fit parameters publicly available would strengthen reproducibility.","section":"Data Availability Statement"}],"recommendation":"major_revision","confidential_remarks":"This is a carefully executed experimental study with a clear central claim and several thoughtful controls. The blocking scientific issue is the possible contamination of Qi fluctuations by resonance-frequency noise during the VNA sweep, which is not controlled for. I am not asking for new physics, only for a targeted diagnostic such as a sweep-time comparison or a Qi-fr cross-correlation analysis. If the authors can provide that evidence, the paper would meet the standard for publication. The data-availability policy is a secondary concern given the journal's policies."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Leif, quick take: this is a genuinely useful experimental paper and the main observation—13% relative Qi fluctuations over 12–16 hours at low power, ubiquitous across chips and cooldowns—deserves attention. The controls are good: JPA on/off, two-resonator interleaving, Qc stability, fitting uncertainties. The power/temperature dependence of the fluctuation amplitude, the LP-MP correlation, and the log-normal Fδ0_TLS distribution are all presented carefully and make a credible case that TLS are involved.\n\nThe new content is real. Prior work (Ref. 29, Béjanin et al.) saw slow dissipation fluctuations, but this paper adds the systematic power/temperature scaling, the interleaved correlation structure, the measurement-time convergence for the mean and width, and the σQi ∝ Qi scaling across many devices. That is a step forward.\n\nThe soft spot I keep coming back to is the one the stress-test flagged: the paper never separates true loss-tangent fluctuations from a measurement artifact. You fit each S21 sweep to a static Lorentzian, but Appendix C shows substantial resonance-frequency noise that grows at low power. If fr jitters by a sizable fraction of the linewidth within a single sweep, the circle fit sees a time-averaged line and reports a biased, fluctuating Qi. That artifact would also (i) decrease with power and temperature, (ii) correlate between LP and MP because the same TLS drive both noises, and (iii) leave Qc stable, exactly as observed. The paper rules out amplifier and common-mode noise, but not this.\n\nI want to be clear: the broad claim that TLS affect these fluctuations is not threatened, because the frequency noise itself is TLS-induced. But the specific and load-bearing claim—that the Qi fluctuations are variations of the TLS loss tangent, with a quantitative Fδ0_TLS distribution and a link to qubit T1 background fluctuations—would be overstated if a material fraction of σQi is contamination. The fix is straightforward: vary the VNA sweep time or analyze the correlation between the Qi noise and the fr noise. I'm surprised they didn't do that.\n\nNo public data is a minor issue for a Ph.D.-level experimental paper; readers can ask the corresponding author.\n\nVerdict: the paper deserves a serious referee. Send it to review, and ask the referee to request the sweep-time control and a quantitative comparison of the Qi and fr noise amplitudes.","headline":"Solid, well-controlled study of long-timescale Qi fluctuations in superconducting resonators, but the attribution to TLS loss-tangent fluctuations leaves an unaddressed frequency-noise contamination channel.","tokens_in":13883,"tokens_out":3482,"would_cite":true,"duration_ms":37273,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Low-power quality factors of superconducting microwave resonators fluctuate by 13% over 12–16 hours, and the paper traces these fluctuations to time-varying two-level-system loss.","keywords":["two-level systems","loss tangent","internal quality factor","superconducting microwave resonators","TLS saturation","quality factor fluctuations","qubit coherence","low-frequency noise"],"falsifier":"Measure low-power $Q_i$ for 16 hours on two identical resonators, one with its TLS-rich surface layer altered by a different surface treatment, while continuously tracking a high-power reference: if the 13% fluctuations are TLS loss-tangent variations, the fractional fluctuation amplitude should scale with the extracted $F\\delta^{0}_{\\mathrm{TLS}}$ and nearly vanish when the TLS contribution is suppressed, whereas a fridge or amplifier drift would survive unchanged.","tokens_in":12907,"feed_emoji":"📉","tokens_out":15112,"duration_ms":134290,"temperature":0.7,"pith_summary":"Superconducting microwave resonators, which are used as stand-ins for qubits when benchmarking materials loss, show slow, large wandering of their internal quality factor at low probe power: over 12–16 hours the relative standard deviation is about 13%, and single excursions reach 37% from the mean. The paper argues that this wandering is not measurement noise or coupling drift but real variation in the loss tangent set by two-level systems (TLS) in the device materials. The evidence is that fluctuations shrink as power and temperature increase, exactly the behaviour of saturable TLS, and that interleaved low- and medium-power measurements fluctuate together while high-power measurements do not. If true, standard single-sweep extractions of the TLS loss tangent capture only a snapshot of a distribution, and the same fluctuating loss could contribute to the time-varying relaxation times seen in superconducting qubits.","feed_headline":"Superconducting resonator loss wanders 13% over 12–16 h","feed_subtitle":"Drift tracks two-level-system loss, vanishing with power and temperature, so single snapshots miss the spread.","key_machinery":"The load-bearing object is the effective TLS loss tangent $F\\delta^{0}_{\\mathrm{TLS}}$, the filling-factor-weighted intrinsic TLS loss tangent. The paper tracks its time dependence through the TLS saturation model $1/Q_i = F\\delta^{0}_{\\mathrm{TLS}}\\tanh(\\hbar\\omega_r/2k_BT)(1+\\langle n\\rangle/n_c)^{-\\beta} + 1/Q_{\\mathrm{PI}}$ and, in the two-plateau limit, the simple subtraction $F\\delta^{0}_{\\mathrm{TLS}}\\simeq 1/Q_{\\mathrm{LP}} - 1/Q_{\\mathrm{HP}}$. A second supporting mechanism is the admittance model with a uniform TLS density of states and fluctuating couplings, which predicts the observed linear relation $\\sigma_{Q_i}\\propto Q_i$. The diagnostic toolkit is low-frequency noise spectroscopy (spectral densities $S_{Q_i}$) and signal coherence between interleaved traces, which distinguish genuine TLS-driven fluctuations from amplifier noise or common-mode drift.","core_discovery":"On the paper's own terms, the finding is that the internal quality factor $Q_i$ of distributed-element superconducting microwave resonators is not a fixed number at low power: it wanders with a relative standard deviation of about 13% over 12–16 hours, and the wandering is caused by temporal fluctuations of the effective TLS loss tangent $F\\delta^{0}_{\\mathrm{TLS}}$. This is established by showing that the fluctuation amplitude falls by up to four orders of magnitude as input power is raised through the TLS saturation region, and by roughly an order of magnitude as temperature is raised toward the quasiparticle-dominated regime. In 16-hour interleaved runs, the low-power and medium-power traces are strongly coherent while neither correlates with the high-power trace, and the high-power quality factor stays roughly fixed while the low-power value swings between $3.3\\times10^5$ and $1.0\\times10^6$. The effective TLS loss tangent extracted as $F\\delta^{0}_{\\mathrm{TLS}}\\simeq 1/Q_{\\mathrm{LP}} - 1/Q_{\\mathrm{HP}}$ follows a log-normal distribution with mean $(9.0\\pm2.2)\\times10^{-7}$, and its fluctuations explain the spread in $Q_i$. Across many resonators, chips, and cooldowns the fluctuation size scales linearly with $Q_i$, giving $\\sigma_{Q_i}/Q_i\\simeq13\\%$ at low power and $0.5\\%$ at high power.","pith_inferences":["If the fluctuating-coupling model is correct, surface or interface treatments that lower the TLS density should reduce the fractional fluctuation amplitude $\\sigma_{Q_i}/Q_i$, not merely raise the mean $Q_i$; reporting both would make a sharper materials benchmark.","The log-normal distribution of $F\\delta^{0}_{\\mathrm{TLS}}$ and its gradual symmetrization with averaging time imply that qubit $T_1$ statistics depend on sampling rate: faster sampling should expose a skewed tail of low-$T_1$ events that long averages smooth away.","Because the $Q_i$ noise spectra are roughly $1/f$, the underlying fluctuators are slow; correlating $Q_i$ wandering with independent charge-noise or TLS-population measurements on the same chip could identify the physical fluctuators.","The interleaved low/medium/high-power coherence test could be reused as a general diagnostic to separate TLS-dominated loss from other loss channels in any resonator, including qubit readout resonators."],"forward_implications":["Because a single low-power sweep can miss the instantaneous TLS loss tangent by tens of percent, one-shot extractions should be replaced or supplemented by repeated measurements over at least a few hours to recover the mean.","Time-averaging a long record captures the mean TLS loss tangent but not the width of its distribution; that width is itself reproducible information about the material environment.","The effect is generic across resonators, chips, and cooldowns with $\\sigma_{Q_i}/Q_i\\simeq13\\%$ at low power, so any low-power resonator benchmark that ignores the fluctuations is incomplete.","If the same loss-tangent wandering occurs in superconducting qubits, then background $T_1$ fluctuations need not require individual near-resonant TLS; a time-fluctuating TLS ensemble would produce similar statistics.","The proportionality $\\sigma_{Q_i}\\propto Q_i$ with a power-dependent constant (13% low, 0.5% high) gives a quantitative expectation for quality-factor noise at a chosen operating point."],"supporting_citations":[{"why":"Supplies earlier long-time dissipation measurements on similar resonators whose data appear in the sigma-Qi comparison, showing the fluctuations are not unique to this setup.","marker":"Ref. 29"},{"why":"Contributes independent quality-factor fluctuation data points that fall on the same linear trend as the authors' measurements.","marker":"Ref. 33"},{"why":"Provides the admittance model with a uniform TLS density of states and fluctuating couplings used to motivate the linear relation and the 13% fluctuation estimate.","marker":"Ref. 28"},{"why":"Supplies the TLS saturation model of Eq. (2) used to extract the effective TLS loss tangent and to select the power and temperature operating points.","marker":"Ref. 30"},{"why":"Supplies the temperature-dependent TLS saturation and quasiparticle loss model used to choose the low, medium, and high temperature points.","marker":"Ref. 12"},{"why":"Documents substantial temporal fluctuations of qubit relaxation time T1, providing the qubit context for the paper's hypothesis about TLS loss-tangent fluctuations.","marker":"Ref. 21"}],"fun_headline_variants":["Resonator Q ripples 13% from TLS loss drift","TLS loss tangent fluctuates 13% over 12-16h","Superconducting loss wanders as TLS fluctuates","Q-factor drift: TLS loss varies 13% in 12h"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole attribution rests on the TLS saturation model of Eq. (2) and on the assumption that the high-power quality factor $Q_{\\mathrm{HP}}$ sits on a stable, power-independent plateau; if that reference drifts with time or the saturation model misdescribes the power dependence, the inferred TLS loss-tangent fluctuations would be contaminated by other loss channels.","fun_headline_variants_meta":{"raw":{"variants":["Resonator Q ripples 13% from TLS loss drift","TLS loss tangent fluctuates 13% over 12-16h","Superconducting loss wanders as TLS fluctuates","Q-factor drift: TLS loss varies 13% in 12h"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000419,"raw_usage":{"total_tokens":2218,"prompt_tokens":1070,"completion_tokens":1148,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":686,"completion_tokens_details":{"reasoning_tokens":1073}},"tokens_in":686,"tokens_out":1148,"duration_ms":8989,"temperature":1.0,"reasoning_tokens":1073,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:40:42.772775+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure low-power $Q_i$ for 16 hours on two identical resonators, one with its TLS-rich surface layer altered by a different surface treatment, while continuously tracking a high-power reference: if the 13% fluctuations are TLS loss-tangent variations, the fractional fluctuation amplitude should scale with the extracted $F\\delta^{0}_{\\mathrm{TLS}}$ and nearly vanish when the TLS contribution is suppressed, whereas a fridge or amplifier drift would survive unchanged.","supporting_citations":[],"review_version":1}