{"id":"e1bbf759-0bf0-43ed-afaa-fabc7f83f34a","arxiv_id":"1908.08003","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A sine-series parameterization of pulse amplitude and phase, optimized with Nelder-Mead and a fast matrix-product approximation, generates control pulses for NMR systems up to 12 qubits and simulated lattices up to 100 qubits.","lead":"This paper proposes an optimal-control algorithm that shapes NMR radio-frequency pulses using a small set of sine functions for the amplitude and phase, and tests it on 4-, 7-, and 12-qubit NMR experiments plus larger simulations. It matters because pulse optimization is a bottleneck in quantum control, and reducing the fitted parameter count could speed up gate calibration.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Large-system 'good fidelity' rests on subsystem-average Fsub without a full-system check; the one available comparison for 12 qubits shows Fsub understates infidelity, so the 16/36/100-qubit claims are unsupported.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing concern: Fsub is used as a proxy for full-system fidelity, and this is unsupported for the large simulated lattices. The 12-qubit comparison in the paper shows Fsub < 0.007 while full unitary infidelity was 0.03, so the proxy is demonstrably optimistic. No full-system check is reported for 16, 36, or 100 qubits, despite the abstract claiming 'good fidelity' for those systems. This directly undermines the central claim's quantitative reach. Other concerns, such as the Bhole-Jones approximation error or single-qubit tomography for the 4-qubit experiment, are real but secondary: the approximation error could be tested separately, and the 4-qubit simulation already supports the algorithm's numerical performance. The Fsub issue is the most direct threat because it invalidates the evidence for the largest systems. I agree with the reader's CONDITIONAL verdict: the paper is promising and contains no demonstrable mathematical error, but the large-system claims need either full-system validation or a formal argument that Fsub bounds the global infidelity. Therefore no change to the reader's verdict is needed.","tokens_in":12749,"tokens_out":7245,"duration_ms":70615,"concrete_test":"For the 16-qubit square lattice, take the pulse optimized in Section V.D and simulate the full 16-qubit evolution using Eq. (4)–(5) with exact diagonal exponentiation of the free Hamiltonian and exact single-qubit rotations for the control term (state-vector propagation, feasible because the control term is a tensor product of single-qubit rotations). Prepare the initial state |0...0> and apply the pulse; compute the state fidelity against the ideal state with π/2 rotations on all odd qubits. Compare with Fsub < 0.01 (i.e., expected infidelity < 0.01). If the full-system infidelity exceeds 0.05, the subsystem proxy fails, and the 36- and 100-qubit fidelity claims cannot be trusted without full-system validation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that pulses control 16-, 36-, and 100-qubit systems with good fidelity is supported only by Fsub, the average infidelity over subsystem simulations defined in Eq. (14). Each subsystem simulation includes only the qubits in that subsystem and omits couplings to qubits outside it. In the square lattices of Section V.D, nearest-neighbor couplings cross subsystem boundaries (Figure 8), so the actual full-system evolution will differ from the product of subsystem evolutions. The paper provides no full-system fidelity calculation for these sizes. The only full-system check, in Section V.C for 12 qubits, found unitary infidelity 0.03 (fidelity > 0.97) for pulses with Fsub < 0.007, meaning Fsub was optimistic by about a factor of four. For larger lattices, with more inter-subsystem couplings and larger resonance-offset spread, the discrepancy could be larger, possibly turning an Fsub of 0.01–0.025 into a global infidelity of 0.1 or worse. Thus the headline claims for 16, 36, and 100 qubits are not quantitatively established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a pulse-shaping algorithm for NMR quantum control in which the amplitude and phase envelopes are parameterized as truncated sums of sine functions (Eqs. (8)-(9)), with the resulting small number of coefficients optimized by a derivative-free Nelder-Mead method. The propagator is evaluated with a fast approximation due to Bhole and Jones (Eq. (10)), and for larger systems the optimization is performed on subsystems with the average cost Fsub defined in Eq. (14). The authors report experimental implementations of the resulting pulses for 4-, 7-, and 12-qubit NMR systems and simulations for 16-, 36-, 100-, and 65536-qubit square lattices, claiming that the algorithm drastically reduces the number of fitted parameters while remaining fast and scalable and producing high-fidelity control.","tokens_in":12986,"tokens_out":7896,"duration_ms":78683,"significance":"The proposed parameterization is an interesting alternative to time-sliced GRAPE optimization and, if fully validated, would be practically useful: it naturally supplies smooth pulses, allows robustness constraints through Eq. (13), and makes the parameter count independent of pulse duration and discretization. The experimental work with real NMR systems is a genuine strength, and the 12-qubit full-system simulation check in Sec. V.C is the right kind of validation. However, the large-system claims currently rest on Fsub, which is the optimization objective and is computed without inter-subsystem couplings; the one available calibration point indicates that Fsub is optimistic. The large-system claims in the abstract and conclusion are therefore not quantitatively established as written.","major_comments":[{"comment":"The success metric for the 16-, 36-, and 100-qubit simulations is Fsub, the average over subsystem infidelities. This quantity is exactly the cost function minimized by the optimizer, so reporting Fsub < 0.012 states that the fit converged; it does not independently quantify global control fidelity. Because each subsystem simulation omits couplings across subsystem boundaries in the nearest-neighbor lattices of Fig. 8, the full-system evolution can differ from the product of the subsystem evolutions. The only available calibration is the 12-qubit case in Sec. V.C, where Fsub < 0.007 corresponds to a full-system infidelity of about 0.03, already a factor of roughly four larger. To support the headline claim of 'good fidelity' for 16, 36, and 100 qubits, the authors should compute full-system fidelities for the 16- and 36-qubit cases using Eqs. (4)-(5), and provide a quantitative bound or an independent validation for the 100-qubit case.","section":"Section V.D, Eq. (14)"},{"comment":"The reported experimental pseudo-pure state fidelity of 0.9993 is obtained by comparing the theoretical state with the tensor product of four individually tomographed single-qubit states. This procedure discards all correlations between qubits and is not a valid estimate of the fidelity of the actual four-qubit state to the target |1111>: correlated errors can be invisible to the tensor-product quantity, and the tensor-product state can have a higher overlap with the product target than the real state. The full four-qubit density matrix should be reconstructed, or a lower bound on the true fidelity should be obtained, before 0.9993 is cited as an experimental state-preparation fidelity.","section":"Section V.A, Fig. 3(d)"},{"comment":"The 7-qubit experimental validation is a single-qubit spectrum check on qubit 7 after the labelled-PPS preparation; it does not characterize the full 7-qubit state. The numerical full-circuit fidelity greater than 0.99 is computed with the same evolution model used in the optimization, so it is not an independent experimental validation. The text should state this limitation explicitly, or present a fuller tomographic check, so that the experimental support for the 7-qubit claim is not overstated.","section":"Section V.B, Fig. 6(c)"}],"minor_comments":[{"comment":"The manuscript contains several grammatical and typographical errors, including 'Most quantum processors requires' in the abstract, 'qbits' in Sec. V.D, and 'Psuedo-pure' in the caption of Fig. 3; a careful language edit is needed.","section":"Abstract and throughout"},{"comment":"The approximation from Bhole and Jones is used without stating its validity conditions or giving an error bound for the parameter ranges considered; the paper itself notes in Sec. V.D that the error grows with the resonance-offset spread, so this limitation should be quantified or tested.","section":"Section IV, Eq. (10)"},{"comment":"The text says the 16-qubit system was divided into 7 groups of 4 qubits, which implies overlapping groups rather than a partition; the authors should clarify whether the groups overlap and how Eq. (14) weights qubits that appear in multiple groups.","section":"Section V.D, Fig. 8"},{"comment":"The 12-qubit full-system fidelity is reported only as 'greater than 0.97'; giving the computed value and the number of tested pulses would make the Fsub-to-fidelity gap easier to assess.","section":"Section V.C"},{"comment":"The claim that the algorithm converges faster than GRAPE is not supported by any quantitative comparison; a table with wall-clock times and final fidelities for the same system and computer would be needed.","section":"Section IV"},{"comment":"No data availability statement or code repository is provided, which limits reproducibility of the optimized pulse shapes and of the numerical results.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The core algorithmic idea is reasonable and the experimental NMR work is valuable, but the large-system claims in the abstract and conclusion are currently supported only by an optimization objective. If the authors add full-system checks for 16 and 36 qubits, quantify the Fsub-to-fidelity gap, and temper the 100-qubit claim, the paper could be acceptable. The relation to existing chopped-random-basis methods (e.g., CRAB) should also be clarified."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nRead the Peterson et al. paper. Short version: it is a reasonable engineering paper with real NMR experiments, but the headline claim about controlling 100 qubits with 'good fidelity' is not supported, because that number is just the optimization cost Fsub, not a full-system fidelity.\n\nWhat's actually new: they parameterize amplitude and phase as short sine series, use a gradient-free optimizer, and rely on the Bhole-Jones approximation to speed up matrix exponentials. The truncated-sine idea is essentially CRAB (ref [13]) and analytic-function pulse design (ref [10]); the authors cite those, so the novelty is modest, mostly the combination and the subsystem averaging. The experiments are the strongest part: 4, 7, and 12-qubit setups with real spectra, and for 12 qubits they do compute the full unitary and report fidelity >0.97, which is a genuine check.\n\nThe soft spots are in the metrics. The 4-qubit PPS fidelity of 0.9993 is computed from a tensor product of single-qubit tomographic states; that ignores correlations and can be optimistic. For the 16, 36, and 100-qubit lattices, the only reported metric is Fsub, the average of subsystem infidelities. The stress-test note is right: the 12-qubit comparison shows Fsub=0.007 while the full unitary infidelity is 0.03, so Fsub can be optimistic by about a factor of four. With cross-boundary couplings in the lattices, there's no reason to expect that factor to stay small. So the large-qubit claims are not quantitatively established. Also, no code or data is released, and the Bhole-Jones approximation error is not quantified.\n\nNone of this is a load-bearing mathematical error. The method is plausible and the small experiments are real. The paper deserves peer review, but it needs major revisions: benchmark against CRAB and GRAPE, give full-system fidelities or a formal argument for the subsystem proxy, quantify the approximation error, and release code/data. Without those, the scalability claim should be scaled back.\n\nWho's it for: people doing NMR pulse engineering or quantum optimal control in similar settings. I wouldn't cite it for the method's novelty, but maybe as a data point on subsystem optimization.\n\nRecommendation: send to a serious referee, but be prepared for major revision.\n\nCheers.","headline":"A useful NMR pulse-shaping trick wrapped in overclaimed scalability: the small experiments are real, the 100-qubit fidelity claim is just the optimization cost.","tokens_in":13572,"tokens_out":2570,"would_cite":false,"duration_ms":23841,"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":"The paper argues that fitting a small set of sinusoidal amplitude and phase modulations can produce high-fidelity control pulses for quantum systems, demonstrating this on 4-, 7-, and 12-qubit NMR experiments and in simulations up to 100…","keywords":["quantum control","shaped pulses","pulse optimization","NMR quantum computing","sine-series parameterization","high-fidelity gates","scalable control"],"falsifier":"For a 16-qubit lattice pulse, compute the complete $2^{16}$-dimensional evolution with an independent time-step integrator using the optimized amplitude and phase shapes, and compare the resulting fidelity with the reported $F_{sub}$; if the full fidelity falls far below $1-F_{sub}$ (for instance below 0.99 when $F_{sub}$ is 0.01), the subsystem proxy is the weak point.","tokens_in":12477,"feed_emoji":"🧲","tokens_out":11269,"duration_ms":103890,"temperature":0.7,"pith_summary":"This paper argues that the hard part of quantum control—finding a high-fidelity amplitude- and phase-modulated pulse—can be made drastically cheaper by parametrizing the pulse as a small sum of sinusoids rather than as hundreds of independent time-step values. The authors present an algorithm that fits only tens of coefficients, uses a fast time-slicing approximation so the parameter count does not grow with pulse duration, and verifies it experimentally on 4-, 7-, and 12-qubit NMR systems. For idealized 16-, 36-, and 100-qubit spin lattices they report optimized pulses with good average subsystem fidelity in under an hour to a few hours, suggesting the approach scales beyond current NMR processors. A sympathetic reader would take away that pulse design can be framed as a low-dimensional smooth-function fitting problem.","feed_headline":"A small set of sines steers up to 100 qubits","feed_subtitle":"NMR tests on 4, 7 and 12 qubits show high-fidelity shaped pulses; larger lattices follow in simulation.","key_machinery":"The carrying object is the truncated sine-series envelope: $\\Omega(t)=\\sum_{k=1}^{s_A}a_k\\sin(b_k t+c_k)$ and $\\varphi(t)=\\sum_{k=1}^{s_P}d_k\\sin(f_k t+g_k)$. These few coefficients define a smooth pulse whose time-sliced propagators can be evaluated with an exponentiation-free approximation that precomputes the non-diagonal pieces once and updates only diagonal exponentials at each time step. That combination makes each function evaluation linear in the number of time slices rather than in the number of optimization variables, which is what allows a 24- to 78-parameter search to be rerun many times quickly. For systems too large to store the full quantum state, the paper additionally averages the fidelity over selected subsystems.","core_discovery":"The central claim is that for a fixed quantum-control task, the optimal radio-frequency pulse can be represented by a short Fourier-like series for the amplitude $\\Omega(t)$ and the phase $\\varphi(t)$, so the optimization searches over the series coefficients rather than over every time slice. Doing so cuts the number of fitted parameters from the hundreds used by standard time-discretized gradient ascent to 24–78, which accelerates convergence and naturally yields smooth pulses. The paper supports the claim by preparing pseudo-pure states and implementing rotations in real NMR experiments on 4, 7, and 12 qubits, and by computing optimized pulses for model square lattices of 16, 36, and 100 spins, with a memory and time estimate for 65,536. For the 12-qubit system the optimized pulse reaches full-system simulated fidelity above 0.97; for the larger lattices the reported metric is the average fidelity over small subsystems.","pith_inferences":["A natural next test is to warm-start the optimization with coefficients from a nearby task or Hamiltonian; the paper does not explore this, but the gradient-free search would make such reuse especially cheap.","If the subsystem proxy is validated against full-system fidelity, the method's memory advantage suggests pulse design can scale to very large spin lattices only if the model Hamiltonian is accurate, making model error the new bottleneck.","The same sine-series parametrization could be extended to hardware-specific constraints such as bandwidth limits or slew rates by restricting the frequency and magnitude ranges of the fitted coefficients, a feature the paper leaves implicit."],"forward_implications":["For systems described by the NMR Hamiltonian, a smooth pulse optimized on 24–78 parameters can replace hundreds of time-step variables: the paper reports $F_{sub}<0.007$ for 12 qubits and full-system simulated fidelity above 0.97.","Pulse duration and discretization can be changed without changing the number of fitted parameters, and the paper exploits this by refining the time step during optimization while keeping the cost linear in the number of time slices.","Because the optimized pulse is a sum of smooth sines, it meets spectrometer constraints on amplitude changes, and the same 63-parameter shape works for model 16-, 36-, and 100-qubit lattices when multiple rotating frames are used.","The paper states that the approach generalizes to other quantum-technology platforms that use shaped electromagnetic pulses, not only NMR."],"supporting_citations":[{"why":"Supplies GRAPE, the standard time-discretized gradient-ascent method this algorithm is compared against.","marker":"[4]"},{"why":"Supplies the fast, exponentiation-free approximation for each time-slice propagator that makes the parameter search inexpensive.","marker":"[20]"},{"why":"Provides the derivative-free numerical optimization method used to fit the sine-series coefficients.","marker":"[18]"},{"why":"Earlier 12-qubit NMR control work that motivates the largest experimental demonstration and the comparison point for 12-qubit control.","marker":"[16]"},{"why":"Shows how to divide an NMR system into subgroups, the strategy the algorithm adopts for larger systems.","marker":"[17]"},{"why":"Gives the NMR quantum-processing framework, including pseudo-pure states and refocusing, used in the experiments.","marker":"[2]"}],"fun_headline_variants":["Sparse Fourier pulses steer 100 qubits","Fewer pulse parameters, 100 qubits controlled","Compact pulse series scales quantum control to 100 qubits","Fourier pulse shapes cut parameters, control 100 qubits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The scalability conclusion rests on the assumption that the average fidelity measured on small subgroups reflects the performance of the whole system; for the 16-, 36-, and 100-qubit simulations the paper never simulates the full system, so significant couplings across subgroup boundaries could break the proxy.","fun_headline_variants_meta":{"raw":{"variants":["Sparse Fourier pulses steer 100 qubits","Fewer pulse parameters, 100 qubits controlled","Compact pulse series scales quantum control to 100 qubits","Fourier pulse shapes cut parameters, control 100 qubits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000734,"raw_usage":{"total_tokens":3246,"prompt_tokens":874,"completion_tokens":2372,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":2306}},"tokens_in":490,"tokens_out":2372,"duration_ms":18643,"temperature":1.0,"reasoning_tokens":2306,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:52:02.430394+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a 16-qubit lattice pulse, compute the complete $2^{16}$-dimensional evolution with an independent time-step integrator using the optimized amplitude and phase shapes, and compare the resulting fidelity with the reported $F_{sub}$; if the full fidelity falls far below $1-F_{sub}$ (for instance below 0.99 when $F_{sub}$ is 0.01), the subsystem proxy is the weak point.","supporting_citations":[{"cited_title":"Khaneja, T","cited_arxiv_id":null,"evidence_quote":"Supplies GRAPE, the standard time-discretized gradient-ascent method this algorithm is compared against."},{"cited_title":"Bhole and J","cited_arxiv_id":null,"evidence_quote":"Supplies the fast, exponentiation-free approximation for each time-slice propagator that makes the parameter search inexpensive."},{"cited_title":"Wright, Numerical Optimization, Second edition, Springer-Verlag New York (2006)","cited_arxiv_id":null,"evidence_quote":"Provides the derivative-free numerical optimization method used to fit the sine-series coefficients."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier 12-qubit NMR control work that motivates the largest experimental demonstration and the comparison point for 12-qubit control."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows how to divide an NMR system into subgroups, the strategy the algorithm adopts for larger systems."},{"cited_title":"Oliveira, R","cited_arxiv_id":null,"evidence_quote":"Gives the NMR quantum-processing framework, including pseudo-pure states and refocusing, used in the experiments."}],"review_version":1}