{"id":"4f52027e-0a39-401e-ad79-ad34d50f319a","arxiv_id":"2509.19223","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Cryogenic alternating voltage bias scrambles the frequencies of strongly coupled two-level systems in amorphous alumina, an effect reversed by thermal cycling above 10 K.","lead":"Applying a large alternating voltage to a cold oxide capacitor makes the material's atomic 'two-level systems' jump around in frequency for hours instead of staying put. The effect reverses when the device is warmed above 10 K, and the capacitor's overall energy loss does not change, suggesting a reversible, out-of-equilibrium process inside the oxide.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Post-CABS sharp features may be charge/artifact-induced resonator shifts rather than genuine TLS avoided crossings; the density argument is circular and cannot resolve this.","rationale":"I read the paper in good faith and find the empirical observations—loss tangent invariance, reversible thermal cycling, and the disappearance of steady-state hyperbolas after CABS—to be presented with supporting data. The central claim, however, depends on interpreting the post-CABS sharp features as TLS avoided crossings. The authors themselves note that dipole moments could not be fit, so the identification rests on linecut morphology alone. This is the same weakest assumption the reader identified. I add that the density-comparison argument is not an independent check: it assumes the very interpretation (each event is a distinct TLS) and further assumes the control dipole-moment distribution to convert event counts into a density. The proposed fixed-bias fast-sweep test would directly distinguish a genuine TLS doublet from a resonator-frequency shift caused by charge noise, which is the decisive control. Given the current evidence, the CONDITIONAL verdict is appropriate; no verdict change is needed.","tokens_in":13179,"tokens_out":12344,"duration_ms":774692,"concrete_test":"Re-measure the post-CABS device at a fixed bias voltage where a transient feature was observed, acquiring S21 versus frequency with high time resolution (e.g., 100 ms per trace for 30 minutes). Fit each trace to two competing models: (A) a single Lorentzian resonator line with time-dependent center frequency (charge/capacitance noise) and (B) a coherent TLS-resonator dressed-state doublet with splitting 2g ≈ 2 MHz. If model B is required—a stable doublet with the expected splitting—the features are genuine TLS avoided crossings. If model A suffices with a common-mode frequency shift, the central claim fails. Also run a cryogenic DC-only bias control with the same peak field and duration to test whether the alternating waveform is essential.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.B identifies the sharp post-CABS features as 'avoided-crossings' based on a ~1 MHz width in linecuts [Fig. 4(b–e)], and Section IV builds the central claim—TLS frequencies jitter on minute timescales—on this identification. But the authors state that no full hyperbolas were observed and dipole moments could not be fit, so the only evidence is a single-linecut line shape. That line shape is also what would be produced by a sudden shift of the resonator frequency due to trapped charge injected by the ±100 MV/m pulses: after the Supplemental Material III preprocessing (subtracting bias-averaged and frequency-averaged backgrounds), a capacitance jump can create an apparent doublet. A DC-only control is missing, so the alternating waveform is not isolated as the cause. The density argument does not rescue the interpretation: it assumes each transient event corresponds to a TLS and assumes the control dipole-moment distribution to convert event counts into a density (95 TLS/µm³GHz), making the claim that TLS density is unchanged circular with respect to the interpretation of the features.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports experiments on a strongly coupled LC resonator with an amorphous Al2O3 parallel-plate capacitor, used to spectroscopically probe TLSs via avoided crossings as a function of DC bias. After an in-situ cryogenic alternating bias stimulation (CABS) protocol of ±10 V pulses at base temperature, the authors observe that the steady-state hyperbolic avoided crossings present in the control are replaced by sharp, transient features that do not persist from one six-minute scan to the next. Thermal cycling above 10 K restores the original spectral signatures, and the low-power loss tangent is reported as essentially unchanged across all treatments. The paper interprets the post-CABS features as avoided crossings of TLSs whose frequencies jitter on minute timescales, and estimates the post-CABS TLS density to be approximately 95 TLS/(µm³GHz) under the assumption of the control dipole-moment distribution. A speculative mechanism involving non-equilibrium strain energy and phonon bursts is proposed.","tokens_in":13493,"tokens_out":2663,"duration_ms":24130,"significance":"If the central interpretation holds, the observation that a cryogenic alternating bias reversibly destabilizes strongly coupled TLSs, while leaving the low-power loss tangent unchanged, would be an interesting contribution to the TLS dynamics literature and relevant to efforts to control TLS noise in superconducting devices. The experimental platform is well suited to this study: the small capacitor volume gives strong coupling, measurements are performed at single-photon powers, and the comparison across treatment stages including thermal cycling is a sensible protocol. The displayed control spectra and the loss-tangent saturation curves are valuable data. However, the main new claim rests on a feature identification that is not fully supported, and the post-CABS density estimate is partly circular. The paper would benefit from additional control measurements and a more cautious treatment of the transient-feature interpretation.","major_comments":[{"comment":"The post-CABS features are identified as TLS avoided crossings solely from linecuts showing a ~1 MHz dip/peak structure. No full hyperbolas are observed and the authors state that electric dipole moments could not be fit. The preprocessing described in SM III — subtracting bias-averaged and frequency-averaged backgrounds — can produce an apparent doublet-like feature from a sudden shift of the resonator frequency, which could be caused by trapped charge or telegraphic charge switching following the ±100 MV/m pulses. The claim that these are TLS avoided crossings needs a more direct test: e.g., demonstrating the two-branch anti-crossing signature, monitoring a single feature continuously with fast repeated scans, or showing that the feature shape and position vary with bias in the way Eq. (1) predicts. A DC-only bias control or a control with unipolar pulses is also needed to isolate the","section":"§III.B, Table I, Discussion"},{"comment":"The post-CABS density of 95 TLS/(µm³GHz) is derived by counting transient avoided crossings and assuming (i) each counted feature corresponds to one TLS and (ii) the dipole-moment distribution is the same as in the control. Since no hyperbolas were available to extract dipole moments, this assumption is not independently verified, and it is then used in the Discussion to conclude that 'the overall TLS distribution likely remains unchanged.' This is circular for the density/conservation claim. The number also carries no error bars or statistical treatment of the counting of 122 events in 500 bias points. Please provide a confidence interval, propagate the uncertainty from the assumed dipole distribution, and discuss how the result changes if a fraction of the events are not independent TLS crossings.","section":"§III.B, Table I, Discussion"},{"comment":"The loss-tangent argument is presented as evidence that the TLS population is unchanged, but the authors themselves compute in §IV that the strongly coupled TLSs visible in the control contribute tanδ0_TLS ≈ 0.12×10⁻³, an order of magnitude below the fitted low-power loss tangent of (1.88±0.07)×10⁻³. Therefore the unchanged loss tangent is largely insensitive to the strongly coupled TLSs whose dynamics are the subject of the paper. The text should explicitly acknowledge this limited sensitivity, rather than implying that the loss-tangent comparison strongly constrains the interpretation of the post-CABS features.","section":"§IV, Fig. 6"},{"comment":"The statement that TLS frequencies 'fluctuate on the order of minutes' is stronger than the data support. The experiment establishes only that the sharp features do not persist from one six-minute scan to the next, giving an upper bound on the feature lifetime of about six minutes. No continuous tracking of an individual feature was performed, so the actual fluctuation timescale is not measured. Please rephrase to say that features are transient on timescales below the scan time, and discuss what additional time-resolved measurements would be needed to extract a fluctuation timescale.","section":"Abstract, §III.B, §IV"}],"minor_comments":[{"comment":"The statement 'We don’t present any data on electric dipole moments directly after CABS because there were no TLS-induced hyperbolas found in this scan' is important and should appear in the main text, not only in the figure caption.","section":"§III.A, Fig. 3 caption"},{"comment":"The ratio of avoided crossings to bias points is reported as 0.24 post-CABS versus 0.19 pre-CABS, and the density is then stated to be 'similar' and 'within a factor of 1.3.' Given the small number of events and the absence of error bars, a statistical test (e.g., a Poisson confidence interval or a chi-square comparison) is needed before concluding the densities are consistent.","section":"§III.B"},{"comment":"The proposed phonon-burst/strain mechanism is explicitly speculative and is labeled as such, which is appropriate. However, the connection to the observed spectral changes would be clearer if the authors indicated which specific observable would distinguish the phonon-burst scenario from charge-induced frequency shifts.","section":"§IV"},{"comment":"The device was measured in two different cryostats ('control' and 'control-aged'). The possibility that the difference between the control and the aged condition includes a cryostat or wiring contribution is mentioned but not discussed quantitatively. A sentence on reproducibility of the resonator frequency and loss in a test device would help.","section":"§II.B"}],"recommendation":"major_revision","confidential_remarks":"The core experiment is interesting and the raw data appear rich, but the central post-CABS claim currently hinges on an interpretation of transient features that is not uniquely determined. The requested additional controls—especially a DC-only bias treatment and time-resolved tracking of individual features—are standard and should be within the scope of the experimental setup. I am not recommending rejection because the underlying observation (loss of steady-state hyperbolas, reversibility, unchanged loss tangent) is likely real; however, the density conservation claim needs to be decoupled from the circular assumption."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper reports something new: after applying a strong alternating bias at cryogenic temperatures to an alumina capacitor, the familiar steady-state TLS hyperbolas disappear and are replaced by sharp features that look like avoided crossings with a width around 1 MHz, jittering on minute timescales. Thermal cycling above 10 K brings the original spectrum back, and the low-power loss tangent stays flat. The loss-tangent result is clean, and the authors are appropriately cautious about not fitting dipole moments after treatment. Prior alternating-bias work was at 353 K, so the cryogenic protocol is a real extension.\n\nThe soft spot is the secondary claim that the TLS density is unchanged. They count 122 transient features, assume they are independent TLSs, assume the same dipole moment distribution as the control, and get 95 TLS/µm³GHz. They then take that as evidence that the overall distribution is unchanged. That is circular. Also, no DC-only control is shown, so we cannot rule out that the sharp features are resonator frequency shifts from trapped charge or other field noise rather than genuine avoided crossings. The linecuts alone do not settle that.\n\nThe stress-test note about artifacts is worth taking seriously, but it is not damning. The thermal reversibility and the fact that features reappear across many bias points make a pure artifact less likely, though not impossible. Still, the interpretation as TLS frequency jitter needs a control experiment: apply a steady DC bias of the same magnitude and duration, and see if similar transient features appear. That would separate alternating-field effects from plain bias stress.\n\nThe speculative strain/phonon mechanism is labeled as a proposal, so that is fine. There are also minor issues: the control-aged density is consistent in the table and text, but the reader flagged it; I did not find an actual discrepancy. The machine-learning fitting details are in the supplement and look adequate.\n\nThis paper deserves a serious referee. It is not ready as-is because of the circular density argument and missing control, but the central observation is genuinely interesting and the loss-tangent invariance is a solid data point. If I were the editor, I would send it to review with a request for a DC-only control and explicit error bars on the post-CABS density. A serious reviewer could then decide whether the transient features are really TLS dynamics.","headline":"A credible new observation of transient TLS jitter after cryogenic alternating bias, but the density claim is circular and a missing DC-only control leaves the artifact question open.","tokens_in":14030,"tokens_out":3138,"would_cite":false,"duration_ms":42991,"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 strong alternating electric field applied at millikelvin temperatures makes two-level systems in amorphous alumina jump between frequencies on a timescale of minutes, and reheating above 10 K restores their original stable spectra.","keywords":["two-level systems","amorphous alumina","cryogenic alternating bias","avoided crossings","TLS spectral diffusion","loss tangent","thermal cycling","strong coupling"],"falsifier":"Fix the bias voltage after CABS and repeatedly measure the resonator transmission at that fixed point. If the sharp features disappear or become static when the voltage is not being swept, they are likely artifacts of the sweep or of charge rearrangements induced by the bias line rather than evidence of fluctuating TLS frequencies; if the features persist and change at a fixed bias, the transient-avoided-crossing interpretation is supported. Likewise, shortening the scan to well under a minute should reveal whether the apparent jitter is actually a telegraphic switch of the TLS energy between","tokens_in":13023,"feed_emoji":"⚡","tokens_out":4492,"duration_ms":42102,"temperature":0.7,"pith_summary":"Two-level systems (TLSs)—tunneling defects that riddle amorphous oxides and drain energy from superconducting qubits—are usually assumed to sit in fixed configurations described by the standard tunneling model. This paper claims that a cryogenic alternating bias treatment (30 hours of ±10 V pulses creating ±100 MV/m in the oxide) drives those defects out of equilibrium: the steady-state avoided-crossing hyperbolas vanish from the resonator spectrum and are replaced by sharp avoided crossings whose frequencies jitter on minute timescales, an upper bound set by the six-minute scan. The effect is reversible: warming the sample above 10 K and cooling back restores the original hyperbolas. At the same time the low-power loss tangent stays essentially unchanged, meaning the average defect ensemble that causes loss is untouched even though the spectroscopically visible defects are scrambled. If correct, this is direct evidence that a large cryogenic alternating field injects non-equilibrium energy (proposed to come from strain-relaxation phonon bursts) that destabilizes TLS frequencies without adding loss.","feed_headline":"Voltage pulses make oxide defects jitter for minutes","feed_subtitle":"After cryogenic alternating bias, two-level systems lose stable spectra until warmed above 10 K—with no added loss.","key_machinery":"The experimental platform is a lumped-element LC resonator whose bias-bridge capacitors concentrate the rf field in a 5.6 µm³ amorphous alumina volume, producing strong coupling (g/2π ≈ 1.3 MHz) between individual TLSs and the oscillator. TLS energies are tuned by a DC bias through the standard tunneling-model relation ε = (Δ0² + (Δ − 2p·E_g)²)^1/2, which maps each TLS to a hyperbola in transmission versus voltage. The central observation is the disappearance of these hyperbolas after CABS and the appearance of transient avoided crossings; the six-minute scan time acts as an upper bound on the TLS frequency jitter. Thermal cycling above 10 K is the reversible switch that restores the steady-","core_discovery":"The paper's central claim is that after Cryogenic Alternating Bias Stimulation (CABS), strongly coupled TLSs in a 49 nm amorphous Al2O3 parallel-plate capacitor no longer obey the standard tunneling-model expectation of stable hyperbolic avoided crossings in a transmission-versus-bias scan. Instead the post-CABS spectrum shows randomly distributed, sharp avoided crossings with coupling ~1 MHz that last less than five minutes, so each six-minute scan catches a different configuration. The authors count 122 such crossings over 500 bias points and estimate the TLS density at ~95 TLS/(µm3·GHz), statistically indistinguishable from the pre-CABS density. Thermal cycling to 10 K or room temperature","pith_inferences":["Inference: the sharp distinction—loss tangent unchanged while visible TLSs scramble—suggests CABS acts only on strongly coupled, spectroscopically visible TLSs or on their local elastic environments, not on the broad ensemble that dominates low-power loss. This could be tested by measuring qubit T1 and frequency noise on a CABS-treated junction.","Inference: tracking a single post-CABS avoided crossing with a faster, repeated probe (scan time of seconds rather than six minutes) would convert the upper bound into an actual spectral-diffusion rate and show whether the frequency motion is continuous drift or telegraphic switching between discrete configurations.","Inference: the 10 K reset threshold is a natural place to look for the energy scale of the mechanism; repeating the CABS exposure at different base temperatures or measuring the time needed for hyperbolas to reappear after warming could map the barrier distribution that holds the non-equilibrium state."],"forward_implications":["Post-CABS TLS spectra cannot be described by the static standard tunneling model; any complete model must include TLS frequency dynamics on minute timescales.","CABS does not degrade the resonator: the intrinsic low-power loss tangent remains ~1.9×10−3, so alternating-bias treatment can disrupt TLS configurations without adding loss.","A single thermal cycle above 10 K fully restores frequency-stable TLS spectra, so the CABS-induced state is reversible rather than a permanent structural change.","The density of strongly coupled TLSs is approximately the same before and after CABS (~95 vs 75–93 TLS/(µm³·GHz)), meaning the effect is a redistribution of TLS frequencies rather than creation or annihilation of defects.","The estimated TLS density from counting transient avoided crossings matches the control within a factor of 1.3, supporting the interpretation that the same ensemble is present but moving."],"fun_headline_variants":["Cryogenic voltage pulses send oxide defects into minutes-long jitter","Transient TLS spectra persist for minutes after cryogenic bias","Cryogenic bias induces reversible, lossless minute-scale TLS fluctuations","Voltage pulses make TLSs fluctuate for minutes, reversible at 10 K"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the sharp, time-varying features seen after CABS are genuine avoided crossings of the resonator with two-level systems whose frequencies are shifting, rather than artifacts of the voltage sweep, drifting baselines, or electric-field noise from trapped charges created by the ±100 MV/m pulses.","fun_headline_variants_meta":{"raw":{"variants":["Cryogenic voltage pulses send oxide defects into minutes-long jitter","Transient TLS spectra persist for minutes after cryogenic bias","Cryogenic bias induces reversible, lossless minute-scale TLS fluctuations","Voltage pulses make TLSs fluctuate for minutes, reversible at 10 K"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000543,"raw_usage":{"total_tokens":2450,"prompt_tokens":767,"completion_tokens":1683,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":511,"completion_tokens_details":{"reasoning_tokens":1617}},"tokens_in":511,"tokens_out":1683,"duration_ms":10139,"temperature":1.0,"reasoning_tokens":1617,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T15:24:16.892027+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix the bias voltage after CABS and repeatedly measure the resonator transmission at that fixed point. If the sharp features disappear or become static when the voltage is not being swept, they are likely artifacts of the sweep or of charge rearrangements induced by the bias line rather than evidence of fluctuating TLS frequencies; if the features persist and change at a fixed bias, the transient-avoided-crossing interpretation is supported. Likewise, shortening the scan to well under a minute should reveal whether the apparent jitter is actually a telegraphic switch of the TLS energy between","supporting_citations":[],"review_version":1}