{"id":"641f336e-1dc4-4891-9b25-a5472a871ea0","arxiv_id":"2504.13447","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The 0th-order AOM output is stabilized by servoing the 1st-order beam against a sampled input beam, reaching 4e-6 Hz^-1/2 relative power noise at 1e-4 Hz and 3.28e-6 Allan deviation at 500 s.","lead":"The authors stabilize laser power by keeping the 0th-order beam from an acousto-optic modulator constant and using the weak 1st-order beam as the control channel. They report a 200x reduction in relative power noise at low frequencies, down to 4e-6 per root hertz, while keeping 99% of the light available for the experiment.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The control loop never measures P0, so the stabilization rests entirely on the time-invariance of the empirically fitted coefficients in Eq. 7; the paper's own 'uncontrolled attenuation' admission and the 1–30 s Allan deviation degradation show this assumption is not fully met.","rationale":"The reader's weakest assumption is exactly the time-invariance of the calibrated linear relation in Eq. 7, and my reading of the paper confirms this as the most load-bearing concern. The control architecture is a feedforward compensation on P1 based on a model of P0, not a direct feedback on P0. Any error in the coefficients or their drift over time enters the output beam without correction. The paper itself admits 'uncontrolled attenuation' as a suspected noise source, and the measured Allan deviation shows residual instability in the 1–30 s band and a long-term rise, consistent with such parameter drift. The fitted parameters have no quoted uncertainties, and the performance comparison is made against Ptot rather than the uncontrolled P0, which slightly weakens the quantitative claim. These are addressable issues rather than fatal flaws: the experimental demonstration is real, the algebra from Eqs. 1–7 is internally consistent, and the observed reduction in relative power noise is credible. Therefore the CONDITIONAL verdict from the reader is appropriate; no change to the verdict is needed. A direct residual analysis and a second calibration would settle whether the model error is the dominant limiter.","tokens_in":7478,"tokens_out":8619,"duration_ms":79320,"concrete_test":"From the recorded 9-hour time series of Ps, P1, and P0, compute the residual E(t) = P0(t) − [kPs(t) + (m−1)P1(t) + δ] using the reported coefficients. If the Allan deviation of E(t) is substantially below that of P0(t), the model is accurate; if E(t) tracks P0(t), then the residual is model error and the controller is not directly stabilizing P0. In addition, re-fit k, m, and δ from the same data and compare with the pre-run values; a change larger than the fit uncertainty would directly demonstrate the time-invariance assumption fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the 0th-order application beam is stabilized because Eq. 7, P0 = kPs + (m−1)P1 + δ, is used as a virtual sensor. The PID regulates the combination kPs + (m−1)P1 to a setpoint, but P0 itself is never in the feedback loop. Therefore any error or drift in k, m, or δ, or any miscalibration of the photodetectors, propagates directly to P0. The paper explicitly suspects 'uncontrolled attenuation' in the discussion, and the Allan deviation shows a degradation for averaging times between 1 and 30 s and a rise from 3.28×10⁻⁶ at 500 s to 6.19×10⁻⁶ at 4.48 h, which is consistent with slow parameter drift. The fitted coefficients k = 8.5301, m = 0.208, and δ = 1.1345 mW are reported without uncertainties, so it is unknown whether the factor-of-200 improvement is limited by model error. Additionally, the improvement is quoted against the uncontrolled total power Ptot rather than the uncontrolled 0th-order beam; since P1 is only about 1% of Ptot this is a minor issue, but it means the claimed factor is not a direct measure of the controller's effect on the application beam relative to its own uncontrolled state.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a laser power stabilization method for the 0th-order beam of an acousto-optic modulator (AOM). The authors derive a linear relation, Eq. (7), P0(t) = kPs(t) + (m−1)P1(t) + δ, linking the 0th-order beam power P0 to the monitored sampling beam power Ps and the 1st-order beam power P1, with coefficients k, m, and δ obtained by fitting. A digital PID controller adjusts the 1st-order beam so that the combination kPs + (m−1)P1 stays at a setpoint, thereby stabilizing P0 without measuring it directly. The paper reports a 9-hour continuous run in which the 1st-order beam follows the total power fluctuations, a relative power noise reduction by a factor of 200 (to 4×10⁻⁶ Hz⁻¹/² at 10⁻⁴ Hz) compared with the uncontrolled total power, and an Allan deviation of 3.28×10⁻⁶ at 500 s averaging time, with 99% of the AOM output available as the application beam.","tokens_in":7771,"tokens_out":5427,"duration_ms":50948,"significance":"If the method proves robust, it offers a practical alternative to conventional in-loop laser power stabilization: the application beam need not be tapped for feedback, so 99% of the AOM output remains usable, and no beam-splitter ratio optimization is required. The paper provides a clear physical derivation, an experimental demonstration over 9 hours, and an honest discussion of residual noise sources. The idea of using a virtual sensor built from the AOM conservation law is genuinely different from the usual in-loop and split-ratio methods. However, the central claim currently rests on a single run, on empirically fitted coefficients reported without uncertainties, and on a control law that never measures P0; the significance is therefore conditional on additional robustness evidence.","major_comments":[{"comment":"The control law in Eq. (7) is the core of the method, but the coefficients k = 8.5301, m = 0.208, and δ = 1.1345 mW are reported without uncertainties or a statement of how often the calibration was repeated. Because P0 is not measured by the feedback loop, any stationary error or slow drift in these coefficients, or in the photodetector calibrations, appears directly as an error in the reconstructed P0. The paper's own discussion attributes residual noise to 'uncontrolled attenuation' and therefore acknowledges that δ is not truly constant. Please add a sensitivity analysis showing how errors in k, m, and δ propagate to P0, report calibration uncertainties, and demonstrate that the fitted coefficients remain stable over the 9-hour run.","section":"Eq. (7) and the fitted coefficients after Fig. 4"},{"comment":"The headline factor of 200 and the Allan deviation are computed from a single 9-hour trace. No error bars, confidence intervals, or repeated measurements are given, so run-to-run variability cannot be assessed. In addition, the Allan deviation rises from 3.28×10⁻⁶ at 500 s to 6.19×10⁻⁶ at 4.48 h, and there is a notable degradation between 1 and 30 s averaging times; the paper should discuss whether these features are consistent with the unmonitored model drift in Eq. (7) and with the 'uncontrolled attenuation' limitation stated in the discussion.","section":"Figs. 6 and 7"},{"comment":"The description of the control loop lacks the details needed to reproduce the experiment: the PID gains, update rate, actuator calibration procedure, setpoint computation, and any anti-windup or saturation handling are not specified. The statement that RPN can only be evaluated up to about 30 Hz because of the PID execution rate suggests a limited loop bandwidth, but no loop transfer function or stability margin is shown. Please provide a quantitative description of the controller, including sample rates and gains, or a block-diagram transfer-function model.","section":"Control algorithm, Fig. 1(b) and the text following Fig. 5"},{"comment":"The reported factor-of-200 improvement compares the controlled P0 with the uncontrolled total power Ptot, not with the uncontrolled P0. Since P1 is about 1% of Ptot, this baseline is probably a good proxy, but the paper should state this explicitly and, ideally, measure the uncontrolled 0th-order beam to confirm that the improvement is not partly an artifact of comparing against a quantity that includes the actuator's own action. This point is load-bearing for the quantitative claim of a factor of 200.","section":"Figs. 6 and 7, performance baseline"}],"minor_comments":[{"comment":"The notation f(Ps,P1) = kPs + δ + mP1 in the schematic should be reconciled with the numbering and symbols used in Eqs. (4)–(6), where δ is a composite constant and m multiplies P1.","section":"Fig. 1 caption"},{"comment":"The term 'acoustic optic modulator' should be 'acousto-optic modulator'; the latter is the standard spelling in the field.","section":"Throughout"},{"comment":"The attenuation terms δ1, δ2, δ3, and δNPBS are introduced in the figure but not all are defined in the caption; please label each term directly in the figure or caption to make the derivation in Eqs. (1)–(4) easier to follow.","section":"Fig. 2"},{"comment":"The statement 'This proves the relationship in Eq. 6 is correct' is too strong for a linear fit over a limited operating range; a phrase such as 'is consistent with' would be more appropriate.","section":"Text after Fig. 4"},{"comment":"The frequency markers for the factor-200, factor-20, and factor-5 improvements are not visible in the printed figure; please add vertical reference lines or a table of the values at the quoted frequencies.","section":"Fig. 6"},{"comment":"The photodetector calibration procedure and the accuracy of the power measurements are not described; one sentence on how the detectors' responsivities and ADC scaling were calibrated would improve reproducibility.","section":"Experimental setup"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's citation practice is unremarkable and there are no self-citation concerns. The central risk is not circularity but fragility: the virtual-sensor model has unquantified coefficients and the control loop never observes P0. I would require a sensitivity analysis and repeated measurements before publication. The idea is sufficiently distinct from Refs. [21] and [22] to be publishable if the robustness issues are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read the Li et al. paper. The core idea is genuinely new as far as the cited literature goes: they stabilize the 0th-order AOM beam, which holds 99% of the power, by servoing the weak 1st-order beam through a calibrated conservation relation P0 = kPs + (m−1)P1 + δ. Since P0 is never measured in the loop, the control quality depends completely on the accuracy and time-invariance of that model. That's the crucial design choice, and the paper is upfront about it.\n\nWhat's done well: the derivation from power balance is clean and the experimental validation of Eq. (6) via two linear plots is reasonable. The 9-hour run with the 1st-order beam visibly tracking the fluctuations of the total power is a nice qualitative demonstration. The temperature dependence measurement (0.018 mW/°C) is a useful detail. And the authors explicitly flag \"uncontrolled attenuation\" as a residual noise source, which is honest.\n\nThe soft spots: The coefficients k, m, δ come from linear fits with no uncertainties, and no repetition statistics are given for the RPN or Allan deviation. This is a single run, so we can't judge run-to-run variability. The improvement factor is quoted against the uncontrolled total power, not against the uncontrolled 0th-order beam itself. Since P1 is only 1% of the total, this is a reasonable proxy, but it's not the strict controller-effectiveness comparison. The bigger issue is exactly what the stress-test note says: because P0 is not fed back, any drift in δ, m, or the detector calibrations enters directly as output drift. The Allan deviation's bump at 1–30 s and its rise from 3.28e-6 at 500 s to 6.19e-6 at 4.48 h are consistent with such model drift. So the headline numbers are plausible but should be treated as a demonstration, not a metrological claim.\n\nFor a reader, the paper is most useful as a proof-of-concept of the virtual-sensor approach. Anyone working on AOM-based power stabilization for clocks or interferometers will want to know about it, and the low 1% power overhead is a real practical advantage. It's not a breakthrough in understanding, but it's a solid 'new architecture' paper.\n\nMy recommendation: yes, send it to peer review. A competent referee can push for the missing error bars and a head-to-head controlled-vs-uncontrolled P0 measurement, and the open-loop sensing deserves a more careful long-term stability analysis. The idea deserves archival publication, but with revisions.","headline":"Clever virtual-sensor AOM stabilizer that leaves 99% of the light for the application beam; the single-run data and open-loop model drift are real caveats, but the idea deserves peer review.","tokens_in":8313,"tokens_out":3861,"would_cite":false,"duration_ms":34943,"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":"Laser power can be stabilized by controlling only the unused diffraction order through an AOM conservation law.","keywords":["laser power stabilization","acousto-optic modulator","conservation law","relative power noise","Allan deviation","PID control","0th-order diffraction beam","long-term stability"],"falsifier":"Re-measure $k$, $m$, and $\\delta$ by the same linear fits before and after a long control run with the loop disabled, using an independent photodetector on $P_0$; if any fitted coefficient shifts by more than the reported stability level (about 0.01 mW on a 34 mW beam), the assumed time-invariant conservation law is falsified.","tokens_in":7271,"feed_emoji":"🎯","tokens_out":9519,"duration_ms":78329,"temperature":0.7,"pith_summary":"An acousto-optic modulator (AOM) splits a laser into a 0th-order beam and a 1st-order beam, and this paper claims the 0th-order beam can be stabilized without ever measuring it. The argument rests on a conservation law: the total diffracted power equals a constant times the sampled input power, plus a fixed attenuation, plus a temperature term proportional to the 1st-order power. The control loop measures only the sampling beam and the 1st-order beam, then adjusts the 1st-order beam so that the combination equal to the 0th-order power stays constant. If the claim is right, long-term laser power stabilization becomes possible while leaving 99% of the light in the application beam, with a reported 200-fold reduction of relative power noise at $10^{-4}$ Hz and an Allan deviation of $3.28\\times10^{-6}$ at 500 s. That matters for atomic clocks, laser interferometers, and gyroscopes, where power noise couples into frequency or phase error.","feed_headline":"AOM conservation law cuts laser power noise 200-fold at low frequencies","feed_subtitle":"Only 1% of the light drives the control; 99% stays available as a stable application beam.","key_machinery":"The load-bearing object is the AOM conservation law as expressed in Eq. (7): $P_0(t)=kP_s(t)+(m-1)P_1(t)+\\delta$. This identity turns stabilization of the unmeasured 0th-order beam into keeping a weighted sum of two measured beams fixed. The $mP_1$ term accounts for the temperature-sensitive transmission of the AOM crystal, which the paper finds proportional to the 1st-order beam power at low diffraction efficiency; the constant $\\delta$ absorbs fixed attenuations in the optical path. A digital PID controller takes the sampling-beam power $P_s$ and the 1st-order power $P_1$ as inputs and adjusts the AOM driver voltage to keep the combination constant, which is why only 1% of the total light is needed for control.","core_discovery":"The central claim is that the relation $P_0(t)=kP_s(t)+(m-1)P_1(t)+\\delta$ is a valid, time-resolved power-balance equation for the AOM, and that enforcing it with a digital PID controller stabilizes the 0th-order application beam $P_0$. The constants $k=8.5301$, $m=0.208$, and $\\delta=1.1345$ mW are obtained by linear fits; because $m<1$, the $P_1$ term is negative, so holding $kP_s+(m-1)P_1+\\delta$ constant is equivalent to holding $P_0$ constant. In a 9-hour run the 1st-order beam tracked the total-power fluctuations and $P_0$ stayed at about 0.01 mW on a 34.15 mW baseline, with 99% of the light in the application beam. The relative power noise reached $4\\times10^{-6}$ Hz$^{-1/2}$ at $10^{-4}$ Hz, a factor-200 reduction over the uncontrolled total power, and the Allan deviation reached $3.28\\times10^{-6}$ at 500 s and $6.19\\times10^{-6}$ at 4.48 hours.","pith_inferences":["A natural extension not tested in the paper: if the AOM's temperature term is modeled with a faster thermal state rather than as proportional to $P_1$, the control bandwidth could extend beyond the current PID limit of about 30 Hz and shrink the Allan-deviation bump between 1 and 30 s.","The same conservation-balance idea should transfer to any power-splitting element with one sacrificial output, such as an electro-optic modulator or a waveguide coupler, where total power is conserved.","Because the control loop never measures $P_0$, periodic recalibration of $k$, $m$, and $\\delta$ or an occasional direct check of $P_0$ would protect against the coefficient drift the paper lists as a residual noise source."],"forward_implications":["Long-term power stability can be maintained while 99% of the total AOM output remains available for the application, because the control beam consumes only about 1%.","Relative power noise near $10^{-4}$ Hz drops by about a factor of 200, to $4\\times10^{-6}$ Hz$^{-1/2}$, in an AOM-based setup.","The method extends to many hours: Allan deviation grows from $3.28\\times10^{-6}$ at 500 s to $6.19\\times10^{-6}$ at 4.48 hours.","Because the application beam is never measured, the control loop avoids beam-splitter splitting-ratio drift, a limitation the paper identifies in existing schemes."],"supporting_citations":[{"why":"Documents the high-split-ratio beam-splitter limitation on long-term application-beam stability and available power that the new method avoids.","marker":"[21]"},{"why":"Supplies the long-term Allan-deviation benchmark (2e-6 at 10^4 s) achieved with an AOM in a small temperature-controlled setup.","marker":"[22]"},{"why":"Shows the AOM crystal temperature rises with driver output, the effect modeled by the mP1 term in the conservation law.","marker":"[23]"},{"why":"Uses thermal control of an AOM for stability, supporting the temperature-control design of the setup.","marker":"[24]"},{"why":"Demonstrates thermal-induced changes in an acousto-optic frequency shifter, supporting the residual temperature-sensitivity discussion.","marker":"[25]"}],"fun_headline_variants":["AOM conservation law cuts laser noise 200-fold at low freq","Laser power noise reduced 200x via AOM conservation law","AOM conservation: stabilize 0th order, keep 99% light","Use AOM conservation to slash laser power noise 200x","Conservation law in AOM: 200x quieter laser beam"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the calibrated linear relation $P_0(t)=kP_s(t)+(m-1)P_1(t)+\\delta$ with $k=8.5301$, $m=0.208$, and $\\delta=1.1345$ mW remains valid and time-invariant throughout the control run, even though the loop never measures $P_0$; if any coefficient drifts, holding the combination constant no longer holds $P_0$ constant.","fun_headline_variants_meta":{"raw":{"variants":["AOM conservation law cuts laser noise 200-fold at low freq","Laser power noise reduced 200x via AOM conservation law","AOM conservation: stabilize 0th order, keep 99% light","Use AOM conservation to slash laser power noise 200x","Conservation law in AOM: 200x quieter laser beam"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000257,"raw_usage":{"total_tokens":1604,"prompt_tokens":994,"completion_tokens":610,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":610,"completion_tokens_details":{"reasoning_tokens":518}},"tokens_in":610,"tokens_out":610,"duration_ms":5415,"temperature":1.0,"reasoning_tokens":518,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:08:22.369921+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-measure $k$, $m$, and $\\delta$ by the same linear fits before and after a long control run with the loop disabled, using an independent photodetector on $P_0$; if any fitted coefficient shifts by more than the reported stability level (about 0.01 mW on a 34 mW beam), the assumed time-invariant conservation law is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Documents the high-split-ratio beam-splitter limitation on long-term application-beam stability and available power that the new method avoids."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the long-term Allan-deviation benchmark (2e-6 at 10^4 s) achieved with an AOM in a small temperature-controlled setup."},{"cited_title":"Tricot, D","cited_arxiv_id":null,"evidence_quote":"Shows the AOM crystal temperature rises with driver output, the effect modeled by the mP1 term in the conservation law."},{"cited_title":"Kobayashi, Y","cited_arxiv_id":null,"evidence_quote":"Uses thermal control of an AOM for stability, supporting the temperature-control design of the setup."},{"cited_title":"Zhang, Y","cited_arxiv_id":null,"evidence_quote":"Demonstrates thermal-induced changes in an acousto-optic frequency shifter, supporting the residual temperature-sensitivity discussion."}],"review_version":1}