{"id":"0eeaa9ad-aeb2-4a12-a96c-f33a14fdaaab","arxiv_id":"2608.02815","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"PaQit chooses how densely to pack parallel circuit instances on a neutral-atom array to minimize energy for a given fidelity target, and the paper validates the underlying fidelity-spacing trend on real Aquila hardware.","lead":"Researchers introduce PaQit, a scheduling framework that packs multiple runs of a quantum program onto a neutral-atom machine to save energy while keeping accuracy above a user-set target. The paper models how energy, runtime, and fidelity interact on these machines and tests the approach on simulations and real QuEra Aquila hardware.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Equation (12) likely has the wrong exponent: for a detuned π/2 pulse, infidelity scales as (δ/Ω)^2 ∝ (r_b/d)^12, not (r_b/d)^6, so the η^3 fidelity law and all quantitative PaQit operating points are unsupported.","rationale":"I agree with the reader's identification of Eq. (12) as the weakest load-bearing premise, but I would sharpen it: the issue is not only that κ is uncalibrated, the exponential's exponent is likely missing a factor of two. The paper's own experiments in Figs. 6-7 only establish monotonic degradation with packing and cannot distinguish η^3 from η^6. A second, independent inconsistency is that with the paper's own β=1 power model, Eeff(f*) is monotone decreasing in η because Eeff = sn t [P0/(ηN) + α], so the claimed non-monotonic energy profile in Sec. V.D and Fig. 5 does not follow from Eqs. (14)-(15); this weakens the 'optimal operating point' narrative but actually reinforces the qualitative rule to pack as densely as fidelity allows. The central qualitative message may survive, but the quantitative operating points in Figs. 3-5 depend on the exact exponent and need to be re-derived or empirically calibrated before acceptance. This keeps the paper in CONDITIONAL status, matching the reader's verdict, so no change is recommended.","tokens_in":13745,"tokens_out":10301,"duration_ms":98614,"concrete_test":"Use the Bloqade analog emulator to simulate a 3×3 square array under the exact ramp-hold-ramp pulse in Sec. VI.D for spacings a=4,5,...,11 μm with Aquila parameters, and compute the fidelity proxy F for each configuration. Fit log(F/fmax) = -κ' (rb/d0)^p η^{p/2}, equivalently log[-log(F/fmax)] = log κ' + p log(rb/d0) + (p/2) log η, with free κ' and p. If the fitted p is close to 12, Eq. (12)'s η^3 law is refuted and Eqs. (13)-(15) with Figs. 3-5 must be regenerated; if p is close to 6, the existing form is supported. Repeat the fit on the real Aquila data in Fig. 7 to check consistency.","verdict_should_be":"UNCHANGED","load_bearing_attack":"PaQit's quantitative schedule rests on Eq. (12), f(η)=fmax exp[-κ(rb/d0)^6 η^3]. The paper justifies this by saying the cumulative perturbation scales as (rb/d)^6 and then sets κ=1 without deriving or fitting the exponential. The physical scaling is suspect. The measurable quantity is not the Rydberg energy shift V=C6/d^6 itself but its effect on the excitation probability under the global pulse. For a two-level atom driven near resonance with small detuning δ (here δ≈V from a neighbor in a Rydberg state), the π/2-pulse excitation probability has zero linear term in δ; the leading correction is O((δ/Ω)^2). Since V/Ω = (rb/d)^6 when Δ=0, the per-pair fidelity loss is O((rb/d)^12), and with d=d0 η^{-1/2}, log f ∝ -(rb/d0)^12 η^6, not -κ(rb/d0)^6 η^3. If the true exponent is 12 rather than 6, Eq. (13) overestimates the allowable packing efficiency for every target fidelity, and the runtime and energy curves in Figs. 3-5 shift substantially. The real-hardware validation confirms only monotonic degradation with packing and cannot discriminate η^3 from η^6 over the narrow spacing range probed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes PaQit, a scheduling framework for neutral-atom quantum computers that selects qubit packing density to trade fidelity against runtime and energy. The authors introduce a power model P(n)=P0+αn^β, derive serial and parallel runtime/energy bounds, define a packing efficiency η, and posit an exponential fidelity law f(η)=fmax exp[-κ(rb/d0)^6 η^3]. They invert this law to compute the packing efficiency that meets a target fidelity and then evaluate PaQit via analytical curves, Bloqade simulations, real QuEra Aquila hardware executions, and Gemini noise-model simulations. The central claim is that maximizing concurrent circuit instances, subject to a fidelity constraint set by spatial packing, minimizes both runtime and energy in static-power-dominated neutral-atom systems.","tokens_in":14014,"tokens_out":5197,"duration_ms":52852,"significance":"If the fidelity model were quantitatively correct, PaQit would provide a practical and useful scheduling policy for Rydberg platforms. The time and energy algebra in Sec. IV is straightforward and sound, and the observation that static power dominates, so parallelism amortizes the dominant energy cost, is a valuable systems-level insight. The qualitative direction of the packing-fidelity trade-off is supported by both emulation and real hardware data. However, the central quantitative fidelity law, Eq. (12), is asserted without a derivation and is not fitted to data; moreover, the physical scaling exponent is likely wrong. Consequently, the numerical operating points, runtime/energy curves, and the claimed quantitative predictive power are not established. The work is a reasonable framework proposal but requires a corrected, validated fidelity model before its quantitative claims can be accepted.","major_comments":[{"comment":"Equation (12) is asserted without derivation from the Rydberg Hamiltonian, and the stated justification is physically incomplete. For an atom driven near resonance, the leading correction to the π/2-pulse excitation probability from a small detuning δ is second order in δ/Ω, not first order. Since δ/Ω=(rb/d)^6, the per-pair infidelity should scale as (rb/d)^12, giving log f ∝ -(rb/d0)^12 η^6 rather than -(rb/d0)^6 η^3. Because Eq. (13) inverts Eq. (12) to produce PaQit's packing efficiencies, and Eqs. (14)-(15) and Figs. 3-5 depend on that inversion, the quantitative schedule is currently unsupported. The authors need either a careful derivation from the Rydberg dynamics or a quantitative fit to experimental fidelity data that can discriminate the exponent.","section":"Sec. V.B, Eq. (12)"},{"comment":"The constants κ=1.0 and fmax=1.0 are set by hand, and no quantitative comparison is made between Eq. (12) and the measured fidelity curves in Figs. 6 and 7. The real-hardware single-atom fidelity proxy is around 0.9, not 1.0, so fmax=1.0 is not hardware-grounded; the lack of a fit or uncertainty analysis for κ and fmax means the validation supports only the qualitative direction of the trade-off, not the claimed quantitative agreement. Since PaQit outputs numerical operating points, the paper needs a calibrated fidelity model with confidence intervals or at least a sensitivity analysis.","section":"Sec. VI.B and VI.C"},{"comment":"Figures 3-5 are direct plots of the assumed fidelity model with κ=1 and fmax=1, so they are consequences of the postulate rather than independent validation. The text in Sec. VII.A presents these trends as if they were confirmed by the experiments, but the experiments confirm only monotonic degradation with denser packing. A quantitative overlay of the model on the Figs. 6 and 7 data, or a statement that the analytical curves are illustrative only, is necessary to avoid circular support for the model.","section":"Sec. VII.A, Figs. 3-5"}],"minor_comments":[{"comment":"The introduction states that PaQit is available at a GitHub URL, while Sec. VI.D says the source code will be released upon publication; please make the availability statement consistent.","section":"Sec. VI.D"},{"comment":"There is a typo: \"on real neutral-atom hardware,, we\" should read \"on real neutral-atom hardware, we\".","section":"Sec. VI.D"},{"comment":"Equation (13) requires ln(f*/fmax) < 0, which excludes f* = 1 when fmax = 1; the domain of validity of the inversion should be stated explicitly.","section":"Sec. V.C, Eq. (13)"},{"comment":"The power-model parameters α and β are taken from non-peer-reviewed web sources; please either cite a peer-reviewed source or state clearly that α and β are illustrative estimates with unknown uncertainty.","section":"Sec. VI.A"},{"comment":"The fidelity proxy F = 1 - (1/p_ideal)(1/N)Σ|ρ_i - p_ideal| can exceed 1 or become negative for large deviations; please state its range and explain why it is an appropriate proxy for the qualitative packing trend.","section":"Sec. VI.B"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within scope for the venue and the systems-level framing is timely, but the central quantitative model needs to be reworked or refitted. The skeptical concern about the exponent in Eq. (12) appears well-founded on physical grounds; even if the exponent were corrected, the authors must show a quantitative fit to the hardware data rather than relying on qualitative agreement. The revision should be substantial, but the qualitative framework and the validation methodology are salvageable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nThe one-sentence take: the qualitative rule PaQit proposes—pack concurrent circuit copies as densely as the fidelity budget allows, because static power dominates—is sensible and useful, but the quantitative model behind it is asserted, and the fidelity exponent in Eq. (12) is likely wrong by a factor of two. The operating points in Figs. 3-5 are therefore not supported as they stand.\n\nCredit where it's due. The paper cleanly derives serial and parallel runtime/energy bounds (Sec. IV), and the insight that the packing-fidelity trade-off can be turned into a scheduler knob is genuinely useful. The real Aquila data in Fig. 7 is the strongest part: it shows that spacing, not raw atom count, drives the fidelity loss, and that sufficiently spaced atoms can be added almost for free. That is a concrete, reproducible observation. The simulator is simple and the paper is clearly written.\n\nThe soft spots are serious. Eq. (12) is load-bearing and it's just asserted: 'the cumulative perturbation scales with (rb/d)^6' and then an exponential appears. For the detuned π/2 pulse they actually run, the per-pair error is second order in the detuning, so the exponent should be 12, not 6, and the packing dependence should be η^6, not η^3. I checked this against the stress-test note and the argument holds up. κ=1 and fmax=1 are hand-set, so the curves aren't calibrated. The power model leans on non-peer-reviewed blog posts. The paper claims a non-monotonic energy profile, but with its own β=1 linear power model and P0 dominating, E_eff is monotonically increasing with fidelity—there's no interior optimum. And the code availability statement contradicts itself: the abstract says the code is available on GitHub, while Sec. VI.D says it will be released upon publication.\n\nThese are all addressable. A revision should derive or fit the fidelity law, calibrate κ on the Aquila data, correct the monotonicity claim, and clean up the code statement. If the exponent does change, the quantitative conclusions shift substantially, but the qualitative framework and the hardware validation survive.\n\nThis paper is for systems people working on neutral-atom scheduling and energy efficiency. They'd get value from the qualitative analysis and the experimental trends. I wouldn't cite the current quantitative model in my own work, but I'd track a revised version.\n\nI'd send it to a serious referee—an editor shouldn't desk-reject it—but it needs major revision. I'd recommend major revision, with an eye toward accepting after the fidelity model is recalibrated.","headline":"A useful qualitative packing rule that collapses on an uncalibrated and likely mis-exponented fidelity model; the hardware data is the real contribution.","tokens_in":14551,"tokens_out":5351,"would_cite":false,"duration_ms":47947,"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":"On neutral-atom quantum computers, the energy- and runtime-optimal schedule is to pack identical circuit shots as densely as the target fidelity allows.","keywords":["neutral atom quantum computing","energy efficiency","qubit packing","quantum resource scheduling","fidelity optimization","Rydberg blockade","runtime optimization","quantum hardware systems"],"falsifier":"Measure the state-preparation fidelity of a fixed atom grid on real hardware across a sweep of nearest-neighbor spacings $d$ and plot $-\\ln(f/f_{\\max})$ against $d^{-6}$; the exponential model predicts a straight line through the origin, so a clearly nonlinear or intercept-shifted plot would falsify the packing optimization.","tokens_in":13470,"feed_emoji":"⚛️","tokens_out":8903,"duration_ms":77666,"temperature":0.7,"pith_summary":"Neutral-atom quantum computers consume a large fixed baseline of power just to keep trap lasers and control electronics running, regardless of how many qubits are active. This paper seizes on that static-power-dominated regime: run many copies of the same circuit in the same atom array, packed as close as interaction physics permits, so the baseline power is paid once for many computations. The paper derives a quantitative relation between packing density and fidelity, showing that residual Rydberg interactions between packed copies degrade fidelity exponentially as packing efficiency cubed. It then turns a user-chosen fidelity target into a concrete packing density and, from there, into runtime and energy. If the model holds, schedulers can systematically trade a controlled amount of fidelity for large energy and time savings, or the reverse.","feed_headline":"Dense qubit packing cuts energy and runtime on neutral-atom hardware","feed_subtitle":"Running many circuit copies in one array amortizes baseline power; fidelity sets the packing limit.","key_machinery":"The central object is the exponential fidelity model $f(\\eta)=f_{\\max}\\exp[-\\kappa(r_b/d_0)^6\\eta^3]$. Packing efficiency $\\eta$ is the fraction of array qubits actively used; the blockade radius $r_b$ is the distance inside which two Rydberg atoms suppress each other's excitation. The model converts the $1/r^6$ van der Waals interaction between packed copies into a single analytic curve, and inverting it yields the maximum packing efficiency for a target fidelity. That number feeds the runtime formula $T_{\\mathrm{eff}}=sn/(\\eta N)\\,t$ and the energy formula $E_{\\mathrm{eff}}=T_{\\mathrm{eff}}\\,(P_0+\\alpha(\\eta N)^\\beta)$, reducing the entire scheduling problem to a one-dimensional choice of $\\eta$.","core_discovery":"The paper's central claim is that on neutral-atom computers, the energy- and runtime-optimal way to run a circuit is to execute as many identical shots as possible in parallel in the same atom array, and the only limit is the fidelity one is willing to accept. The quantitative core is Eq. (12): with packing efficiency $\\eta$, fidelity is $f(\\eta)=f_{\\max}\\exp[-\\kappa(r_b/d_0)^6\\eta^3]$, where $r_b$ is the Rydberg blockade radius and $d_0$ is the full-occupancy lattice spacing. Because the interaction between neighboring circuit instances falls as the sixth power of their separation, dense packing eventually costs accuracy, but the static-power-dominated energy budget makes extra parallelism otherwise nearly free. Inverting the fidelity formula gives the largest $\\eta$ allowed by a fidelity target, which directly determines effective runtime and energy. The paper validates the qualitative relationship on simulated and real hardware, where fidelity degrades as spacing shrinks and levels off once spacing exceeds roughly ten micrometers.","pith_inferences":["Beyond the paper: the same packing rule should extend to partitioning the array into zones that run different circuits, each with its own spacing threshold, although the paper only analyzes batched identical shots.","Beyond the paper: if near-term hardware has significant per-shot loading, measurement, or reset overhead, the energy benefit of dense packing shrinks; adding a serial overhead term to $E_{\\mathrm{eff}}$ would make the model more predictive.","Beyond the paper: the observed spacing threshold of about ten micrometers offers a direct calibration path—estimating $\\kappa$ from a log-linear fit of fidelity versus $d^{-6}$ would upgrade the model from qualitative to quantitative, since the paper sets $\\kappa=1$ without fitting.","Beyond the paper: the fidelity-versus-spacing curve could be measured routinely on larger arrays during device calibration, turning packing efficiency into a continuously tunable knob for energy-aware scheduling in production workloads."],"forward_implications":["In the static-power-dominated regime, dense packing of identical shots reduces both wall-clock time and total energy, so the energy-optimal schedule and the runtime-optimal schedule coincide.","Packing efficiency falls only as $(-\\ln f^*)^{1/3}$, but runtime and energy rise steeply as fidelity targets tighten, making low-fidelity requirements cheap and very strict requirements expensive.","On weakly interacting hardware, small $r_b/d_0$, near-maximal packing costs almost nothing; on strongly interacting hardware, the same fidelity target forces a much sparser layout and much higher energy.","Spacing, not raw atom count, is the dominant scheduling variable: once atoms are far enough apart to avoid crosstalk, adding more atoms to the array causes little additional fidelity loss.","The scheduler operates purely at the software and control-layout level, requiring no hardware modifications and recomputing packing density when fidelity targets or calibration parameters change."],"supporting_citations":[{"why":"Supplies the analog platform's device parameters, blockade-radius estimate, and the real-hardware fidelity measurements used for validation.","marker":"[14]"},{"why":"Provides the digital neutral-atom platform's noise model and native gate specifications used for the circuit-structure validation.","marker":"[15]"},{"why":"Provides the power-scaling figures (about 7 kW at 256 qubits and 10 kW at 10,000 qubits) that ground the static-power-dominated regime.","marker":"[5]"},{"why":"Corroborates the near-flat power-scaling estimates used to justify the power model in Eq. (3).","marker":"[6]"}],"fun_headline_variants":["Neutral-atom efficiency hinges on qubit packing and fidelity trade-off","Pack qubits densely, save energy, but fidelity sets the limit","PaQit: balance energy, runtime, and fidelity with optimal qubit packing","Dense qubit packing: near-free parallelism, but fidelity limits it"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that fidelity falls exponentially as $\\kappa(r_b/d)^6$, with the proportionality constant $\\kappa$ set to 1; the hardware data confirm only the direction of the effect, not this specific functional form.","fun_headline_variants_meta":{"raw":{"variants":["Neutral-atom efficiency hinges on qubit packing and fidelity trade-off","Pack qubits densely, save energy, but fidelity sets the limit","PaQit: balance energy, runtime, and fidelity with optimal qubit packing","Dense qubit packing: near-free parallelism, but fidelity limits it"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001042,"raw_usage":{"total_tokens":4373,"prompt_tokens":928,"completion_tokens":3445,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":3365}},"tokens_in":544,"tokens_out":3445,"duration_ms":24711,"temperature":1.0,"reasoning_tokens":3365,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:59:06.933000+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the state-preparation fidelity of a fixed atom grid on real hardware across a sweep of nearest-neighbor spacings $d$ and plot $-\\ln(f/f_{\\max})$ against $d^{-6}$; the exponential model predicts a straight line through the origin, so a clearly nonlinear or intercept-shifted plot would falsify the packing optimization.","supporting_citations":[{"cited_title":"The dual-pronged energy-saving potential of quantum com- puters,","cited_arxiv_id":null,"evidence_quote":"Provides the power-scaling figures (about 7 kW at 256 qubits and 10 kW at 10,000 qubits) that ground the static-power-dominated regime."},{"cited_title":"How quantum computing could reshape energy use in high-performance computing,","cited_arxiv_id":null,"evidence_quote":"Corroborates the near-flat power-scaling estimates used to justify the power model in Eq. (3)."}],"review_version":1}