{"id":"81d2266b-07c7-46df-b91d-e2d437ee04b1","arxiv_id":"2608.02249","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Sequential quadrature readout in programmable photonic processors suffers a phase-drift penalty of 3Qτ/8, which adaptive ordering reduces to (3/8 − 1/π)Qτ.","lead":"This paper calculates how much phase drift hurts when a programmable photonic processor measures sine and cosine quadratures one after another instead of at the same time. It shows that measuring the less informative quadrature first cuts the average error by about 85 percent.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Analytic laws check out, but the paper never reports Qτ from its 35 recordings, so the small-drift regime and the 84.9% adaptive reduction are not tied to the motivating processor.","rationale":"I read the derivation in good faith and checked the key perturbation steps. Equation (55), e_{C→S} = −δτ sin²φ0 + O(δτ²), follows from the atan2 differential with x = cosφ0 and y = sin(φ0+δτ); the complementary S→C result is likewise correct. The uniform average E[sin⁴φ0] = 3/8 and the adaptive average E[min(sin⁴, cos⁴)] = 3/8 − 1/π are standard and correctly evaluated. The increment-aware information reduction I/(1+IQτ) also follows under the stated Gaussian local model. The Monte Carlo design is appropriate: exact nonlinear estimators are tested, not just the small-drift approximations, and the paper's own figures display the breakdown of the first-order laws at larger Qτ. The reader's CONDITIONAL verdict is therefore well grounded. My stress-test pass identifies one load-bearing gap that the reader also flagged: the motivating recordings are used only to advertise an 'empirical route' for estimating Qτ, but the actual estimate is omitted. Because the central quantitative statements are asymptotic in δτ, an unreported Qτ leaves the practical relevance of the 84.9% reduction and the architecture boundaries unverified. This is not a mathematical flaw and does not warrant rejection; it warrants the same conditional status, with a concrete path to resolution. I propose the direct check above: publish bQτm and verify the operating point against the exact nonlinear simulation. That single addition would settle whether the concern lands. If the estimate shows Qτ is small, the headline claims apply as stated; if not, the paper needs to re-qualify the regime or report the exact reductions at its operating point.","tokens_in":12898,"tokens_out":17213,"duration_ms":190124,"concrete_test":"Compute bQτm from the 35 recordings using Eq. (29) for the intended reconfiguration interval and for a range of separations τm, reporting the value and its uncertainty. Then run the exact nonlinear Monte Carlo of Section 9 at that operating bQτ with the paper's vsp, vt, and σp values, and compare the simulated MSE with the first-order predictions 3Q/8 and (3/8 − 1/π)Q. If the exact values lie within about 10% of the first-order laws, the central claim is applicable to the motivating system; if not, the 84.9% reduction and the crossover boundaries should be restated at the actual Qτ or explicitly qualified as asymptotic.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The first-order results — E[e²] = 3Qτ/8 (Eq. 76), the adaptive coefficient (3/8 − 1/π) (Eq. 83), and the dimensionless architecture boundaries (Eqs. 122, 124, 127) — all require Qτ to be small enough that O(Qτ²), atan2 branch crossings, and order-boundary effects are negligible. Section 3.1 defines the estimator bQτm in Eq. (29) from the 35 recordings, but no value is ever reported. Figure 5 shows the first-order law departing from exact Monte Carlo as Qτ grows, and Figure 6 shows the adaptive advantage eroding as predictor error σp increases. Without a reported bQτ, no reader can check whether the motivating eight-mode processor operates in the regime where the headline 84.9% reduction and the stated crossover values are quantitatively accurate. This is a missing empirical anchor for the central quantitative claim, not an internal inconsistency in the derivation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper analyzes the estimation error caused by sequential (temporal) acquisition of sine and cosine quadratures in programmable photonic processors, where the relative phase drifts between the two measurements. The authors define the target phase at the end of the second measurement, derive the first-order errors e_{C→S} = -δτ sin²φ0 + O(δτ²) and e_{S→C} = -δτ cos²φ0 + O(δτ²), and obtain the uniform fixed-order mean-square penalty 3Qτ/8. They then propose a phase-predicted adaptive ordering rule with uniform penalty (3/8 - 1/π)Qτ, an 84.9% MSE reduction, and an increment-aware state-space estimator whose effective Fisher information for a stale observation is I/(1+IQτ). Under an ideal balanced Poisson detection model, quadrature Fisher information equals detected photon number, yielding dimensionless architecture boundaries. Monte Carlo simulations validate the small-drift laws and quantify robustness to prediction error and nonlinear effects.","tokens_in":13148,"tokens_out":7360,"duration_ms":94583,"significance":"The analytical core is self-contained and the algebra checks: the atan2 perturbation, the uniform averages 3/8 and 3/8 - 1/π, and the effective-information formula follow from stated assumptions without fitted parameters. The paper explicitly separates ideal benchmarks (perfect phase prediction, symmetric local information) from nonlinear circular simulations, and it acknowledges detector nonidealities. The main value is a simple design rule for choosing between spatial and temporal readout, with explicit resource assumptions. The strengths—derivations grounded in standard estimation theory, Monte Carlo validation of the nonlinear estimators, and a clear statement of model scope—support the paper's usefulness if the missing empirical anchor is supplied.","major_comments":[{"comment":"The estimator bQτm defined in Eq. (29) is never evaluated. The paper motivates the receiver analysis with 35 free-running recordings from an eight-mode processor but reports no value (or even order of magnitude) for Qτ at any reconfiguration interval τ. This matters because the headline laws 3Qτ/8 and (3/8 - 1/π)Qτ, and the crossover conditions in Eqs. (122)-(127), are all linear in Qτ or IspQτ and are valid only in the small-drift regime; Fig. 5 shows the first-order law departing from exact Monte Carlo as Qτ grows. Without a reported bQτm, the reader cannot verify that the motivating processor actually operates in the regime where the 84.9% reduction and the stated crossover values are quantitatively accurate. The authors should report bQτm for at least the selected reconfiguration interval, along with a confidence interval or range, or explicitly relativize the motivating claims.","section":"Sec. 3.1 and Sec. 9"},{"comment":"The paper never states a quantitative criterion for 'sufficiently small δτ'. Equations (55), (69), (76), and (83) are derived under a first-order Taylor expansion and neglect atan2 branch crossings, but the architecture boundaries (122)-(127) use these first-order coefficients without a stated validity condition. The Monte Carlo comparison in Sec. 9 validates the fixed-order crossover at Q = 0.01, but no general threshold (e.g., Qτ below some fraction of rad²) is given. Since the boundaries are intended as design rules, the authors should add a smallness condition or an explicit statement of the range of Qτ over which the first-order coefficients are accurate to a given tolerance.","section":"Sec. 5, Sec. 6, and Sec. 8"}],"minor_comments":[{"comment":"There are typographical issues: 'Recket al.' should be 'Reck et al.', and the author names 'G¨ okhan' and 'N¨ otzel' show encoding artifacts in the full text.","section":"Throughout"},{"comment":"Equation (72) writes O(Qτ²), but the second-order corrections in Eqs. (57) and (71) are phase-dependent; the remainder notation could be clarified to indicate that the coefficient of Qτ² depends on φ0.","section":"Sec. 5"},{"comment":"The crossover values (Q⋆,fixed = 0.0100, Q⋆,IA ≈ 0.0127, Q⋆,adaptive ≈ 0.0470) are reported without error bars or a quantitative agreement metric; a table comparing the Monte Carlo crossovers with the analytical expressions (122), (124), and (127) would substantially aid the reader.","section":"Sec. 9"},{"comment":"The unwrapped phase reconstruction in Eq. (24) is sensitive to unwrapping errors, but no discussion of outlier handling or unwrap robustness is provided; a brief note on this would strengthen the definition of the estimator bQτm.","section":"Sec. 3.1"},{"comment":"The statement that simulation code and data are 'available from the authors upon reasonable request' is acceptable but a permanent repository link would improve reproducibility.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope and the analytical derivations appear sound. The main obstacle is the missing reported value of Qτ from the motivating 35 recordings; if the authors can supply that value (or clearly reposition the claims as hypothetical), the paper would be suitable for publication. I would not reject on the basis of the derivations themselves."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This one is a solid theory paper with an empirical anchor problem. The central derivation—the first-order atan2 drift formulas, the adaptive ordering coefficient (3/8 − 1/π), and the Fisher-information reduction I/(1+IQτ)—is correct and, as far as I can tell, not in the prior literature. The Monte Carlo checks line up with the analytic laws, and the dimensionless architecture boundaries are a useful addition. That part deserves a serious referee.\n\nThe soft spot is the missing bQτ. Section 3.1 defines an estimator from the 35 recordings, but the paper never reports the value. So the headline 84.9% reduction and the crossover boundaries are not tied to the actual processor that motivated the study. Figure 5 shows the first-order law breaks down as Qτ grows; without knowing whether the real device sits in the small-drift regime, the practical relevance is an open question. This is a fixable gap, not a fatal one, but it is a real gap. Also, the code and data are only on request; a public artifact would help.\n\nI was a bit more charitable than the stress-test note: the missing Qτ doesn't undermine the algebra, and the paper explicitly states the limits of its local approximations in Sec. 8.5. But the empirical claim is the one place where the reader has to take the authors at their word, and the paper makes it easy to check—just report the number.\n\nWho is this for? People designing temporal receivers for programmable photonic processors, especially those doing phase tracking with sequential quadrature measurements. It gives them design rules they can use. It won't reshape the field, but it's honest progress.\n\nI'd send it to review. The theory is solid, the siting is appropriate, and the missing Qτ is addressable in a revision.","headline":"Solid analytical paper with a missing empirical anchor: the central Qτ estimate is never reported, so the practical claims hang on an unstated number.","tokens_in":13644,"tokens_out":2762,"would_cite":true,"duration_ms":75987,"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":"This paper derives the phase-drift penalty of sequential sine/cosine readout in programmable photonic processors: fixed ordering costs $3Q_\\tau/8$ in mean-square error, and a phase-predicted ordering rule reduces that to…","keywords":["programmable photonic processor","phase drift","phase instability","quadrature readout","phase estimation","adaptive measurement","Fisher information","feedback stabilization"],"falsifier":"Take simultaneous sine and cosine records from the same processor, estimate $Q_\\tau$ at a chosen reconfiguration interval via Eq. (29), then form sequential estimates with both fixed orders and with the adaptive order using a phase predictor whose error is small compared with the ordering boundary; if the ratio of adaptive to fixed mean-square error does not approach $(3/8-1/\\pi)/(3/8)\\approx0.151$ as $Q_\\tau\\to 0$, the first-order drift law is wrong.","tokens_in":12722,"feed_emoji":"🔬","tokens_out":11339,"duration_ms":119906,"temperature":0.7,"pith_summary":"This paper asks how much phase-estimation accuracy is lost when a programmable photonic processor reads out its sine and cosine quadratures sequentially instead of simultaneously, and when a temporal readout is actually worth using. Working from the four-port configuration of an eight-mode processor, the authors derive that, to first order in the phase increment $\\delta_\\tau$, the reconstruction error is $-\\delta_\\tau\\sin^2\\phi_0$ for cosine-then-sine order and $-\\delta_\\tau\\cos^2\\phi_0$ for the reverse, giving a uniform mean-square drift error of $3Q_\\tau/8$, where $Q_\\tau=\\mathrm{Var}(\\delta_\\tau)$ is the phase-increment variance. They then show that if a predictor of the phase is available, measuring the locally less informative quadrature first lowers the uniform first-order penalty to $(3/8-1/\\pi)Q_\\tau$, an 84.9 percent reduction. The same ordering principle emerges from a state-space treatment in which the Fisher information of a stale measurement falls from $I$ to $I/(1+IQ_\\tau)$. These laws matter because they turn a processor's measured phase statistics into a dimensionless crossover condition, expressed in the product of spatial Fisher information and phase-increment variance, that tells an operator when sequential readout beats simultaneous readout.","feed_headline":"Adaptive readout cuts phase-drift error by 84.9%","feed_subtitle":"It also derives the stability window where sequential readout beats simultaneous in photonic processors.","key_machinery":"The load-bearing object is the differential of the two-argument arctangent, $d\\theta=(x\\,dy-y\\,dx)/(x^2+y^2)$, applied to a pair in which one quadrature is sampled before the drift and the other after. This identity turns a delayed quadrature sample into an explicit first-order phase error whose drift coefficient weights the locally insensitive quadrature: $-\\sin^2\\phi_0$ for cosine-first order and $-\\cos^2\\phi_0$ for sine-first order. A second mechanism is the phase-predicted ordering rule of Eq. (78), which compares $\\sin^2\\phi_p$ with $\\cos^2\\phi_p$ (or, with unequal noise, the full Fisher informations $I_C(\\phi_p)$ and $I_S(\\phi_p)$) and measures the less informative quadrature first; symmetry reduces the uniform average of the minimized coefficient to $3/8-1/\\pi$. The third mechanism is the increment-aware marginalization of Eq. (95), which represents a stale observation as a fresh observation plus an extra Gaussian term of variance $Q_\\tau$, so its effective information is $I/(1+IQ_\\tau)$; this is what makes the state-space estimator stop trusting an old sample once the intervening phase uncertainty dominates. Finally, the ideal balanced Poisson calculation shows each quadrature carries Fisher information equal to its detected signal-photon number, which gives the physical resource scale for the architecture boundaries.","core_discovery":"The paper's central claim is that sequential quadrature readout has a definite, computable phase-drift penalty, and that this penalty can be reduced by ordering the two measurements according to a phase predictor. With the estimate defined at the completion of the second measurement, the exact atan2 reconstruction of the mixed-time pair obeys, to first order in the phase increment, $e_{C\\to S}=-\\delta_\\tau\\sin^2\\phi_0+O(\\delta_\\tau^2)$ and $e_{S\\to C}=-\\delta_\\tau\\cos^2\\phi_0+O(\\delta_\\tau^2)$. Averaging over a uniformly distributed initial phase gives a fixed-order drift mean-square error of $3Q_\\tau/8$. When a predictor $\\phi_p$ for the initial phase is available, the rule \"measure the quadrature with the smaller local Fisher information first\" gives $\\mathbb{E}[\\min(\\sin^4\\phi_0,\\cos^4\\phi_0)]=3/8-1/\\pi$, an 84.9 percent reduction in the drift-induced mean-square error. The same ordering principle is recovered from a local state-space model: marginalizing the unknown phase increment reduces a stale observation's Fisher information from $I$ to $I/(1+IQ_\\tau)$, so delaying a highly informative quadrature is more costly than delaying a weak one. In the ideal balanced Poisson benchmark, each quadrature's Fisher information equals its detected signal-photon number, which converts the whole comparison into dimensionless architecture boundaries in the plane of total spatial information $I_{\\mathrm{sp}}$ and phase-increment variance $Q_\\tau$.","pith_inferences":["The same \"measure the locally less informative quantity first\" principle should transfer to any tracking problem where two observations of a slowly drifting parameter have phase-dependent Fisher information; the sine-cosine quadrature pair is the symmetric special case.","For a practical controller, the useful quantity is not the perfect-prediction coefficient $3/8-1/\\pi$ but the coefficient as a function of predictor error $\\sigma_p$; the paper's $\\sigma_p=0.1$ rad result suggests the gain degrades smoothly, so an operator could calibrate the operating point from the measured $\\sigma_p$.","The empirical estimator $\\hat{Q}_{\\tau_m}$ built from synchronized simultaneous records gives a direct way to populate the architecture boundaries on a real processor: compute $I_{\\mathrm{sp}}$ from the calibrated detection model, estimate $Q_\\tau$ at the intended switching interval, and check which side of the boundary the operating point lies on.","One could test the increment-aware estimator's advantage outside the small-drift regime by comparing the circular posterior of Eq. (104) with fixed-order atan2 at larger $Q_\\tau$; the paper's Monte Carlo already shows the first-order laws depart from exact results there, but the circular estimator's behavior at those drifts is presented for a specific noise model rather than as a closed-form law."],"forward_implications":["For information gain $g=1.6$, temporal readout is favored over simultaneous readout only below the dimensionless boundaries $I_{\\mathrm{sp}}Q_\\tau=1.00$ (fixed order), $3.75$ (symmetric increment-aware), and $6.61$ (ideal adaptive order).","The adaptive advantage is robust to moderate predictor error: at predictor error $\\sigma_p=0.1$ rad the reduction is still 83.2 percent, and it degrades to the fixed-order value only as the predictor becomes uninformative.","In the noisy receiver simulation, the increment-aware estimator pushes the spatial-to-temporal crossover from $Q\\approx0.0100$ to $Q\\approx0.0127$, and adaptive ordering pushes it to $Q\\approx0.0470$, meaning each refinement extends the range over which sequential readout is preferable.","A stale quadrature can never carry more than $1/Q_\\tau$ of information about the end-time phase, so even a perfect measurement of the initial phase cannot beat the uncertainty of the intervening process increment.","The architecture rules depend only on the product $I_{\\mathrm{sp}}Q_\\tau$ and the gain factor $g$, not on any particular delay-scaling law, so they can be applied at whatever reconfiguration interval an operator chooses."],"supporting_citations":[{"why":"Supplies the four-port sine/cosine configuration and the synchronized phase-instability recordings that motivate the receiver analysis and provide the empirical estimator for $Q_\\tau$.","marker":"[9]"},{"why":"Provides the rectangular interferometer mesh design used by the eight-mode processor, fixing the optical transformation that produces the complementary sine and cosine power pairs.","marker":"[2]"},{"why":"Defines the universal interferometer mesh architecture whose tunable phase settings are subject to the drifts studied here.","marker":"[1]"}],"fun_headline_variants":["Adaptive readout cuts phase-drift error by 84.9%","Quadrature ordering reduces drift penalty by 85%","Stability window for sequential vs simultaneous readout","Photonic processors: adaptive quadrature order lowers drift error","Sequential readout: less drift with smart quadrature ordering"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the phase drifts only slightly during the short gap between the two quadrature measurements, so the first-order error formulas and the choice of arctangent branch are valid, and that a predictor of the current phase is accurate enough to choose the right measurement order.","fun_headline_variants_meta":{"raw":{"variants":["Adaptive readout cuts phase-drift error by 84.9%","Quadrature ordering reduces drift penalty by 85%","Stability window for sequential vs simultaneous readout","Photonic processors: adaptive quadrature order lowers drift error","Sequential readout: less drift with smart quadrature ordering"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000572,"raw_usage":{"total_tokens":2848,"prompt_tokens":1235,"completion_tokens":1613,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":851,"completion_tokens_details":{"reasoning_tokens":1529}},"tokens_in":851,"tokens_out":1613,"duration_ms":14667,"temperature":1.0,"reasoning_tokens":1529,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:58:16.382271+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take simultaneous sine and cosine records from the same processor, estimate $Q_\\tau$ at a chosen reconfiguration interval via Eq. (29), then form sequential estimates with both fixed orders and with the adaptive order using a phase predictor whose error is small compared with the ordering boundary; if the ratio of adaptive to fixed mean-square error does not approach $(3/8-1/\\pi)/(3/8)\\approx0.151$ as $Q_\\tau\\to 0$, the first-order drift law is wrong.","supporting_citations":[],"review_version":2}