{"id":"31fc146e-806b-4f05-a117-7fb3498e698d","arxiv_id":"2505.08613","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A quantum state is read out by measuring overlaps against Lorentzian basis states and fitting the coefficients classically, reducing the number of measurements for localized, continuous states.","lead":"This paper proposes a hybrid quantum-classical readout method that reconstructs a quantum state by measuring its overlaps with a small set of Lorentzian-shaped basis states and fitting the expansion classically. A smart generalist might care because it promises to read out smooth wavefunctions or spectra from a quantum computer with far fewer measurements than brute-force tomography, if the state is close to a sum of localized functions.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The fewer-measurements claim counts overlap evaluations, not measurement shots; since each overlap requires many shots (or QAE depth), the Fig. 7(b) comparison omits the proposed method's statistical error and is not apples-to-apples.","rationale":"Both the abstract and Sec. IV B frame the contribution as 'fewer measurements' or 'fewer measurement resources' than conventional readout. The demonstration of this advantage relies on comparing m_iter (number of overlap evaluations) to N (number of amplitudes). The reader's weakest assumption—that the target is LCLF—is important, but the paper's own stated scope is states 'approximately represented as a continuous function' and 'expressed as a combination of localized functions.' The optimizer failure for n=9,10 is a robustness issue that a better optimizer or better initialization might fix. The measurement-resource accounting, by contrast, is a validity issue: the central metric used for the headline comparison is not the actual measurement cost. If each overlap evaluation requires, say, 1,000 shots, then the method's total shot count for n=5 is 28,000, which is not fewer than the direct measurement's 1,000 shots. The omission of statistical error in the proposed method's Fig. 7(b) curve confirms that the plot compares ideal overlap estimates against noisy direct measurements. This is an apples-to-oranges comparison. The check I propose would settle it by fixing a total budget and recomputing both curves. This does not invalidate the approach itself; it means the burden of proof for the efficiency claim shifts to specifying a per-overlap precision budget and total measurement cost. The reader's conditional verdict (revision required) is the right call, and my concern is one concrete reason that revision is needed.","tokens_in":17024,"tokens_out":4577,"duration_ms":45764,"concrete_test":"Recompute Fig. 7(b) under a fixed total shot budget: for the proposed method, allocate B/m_iter shots to each of the m_iter overlap estimations, inject the resulting Gaussian noise into the estimated overlaps (or rerun the SWAP-test simulation with finite counts), and recompute the reconstructed-state infidelity including shot noise. Vary B from 10^3 to 10^6 and compare with the direct-measurement curve using the same B. If the proposed method's infidelity at equal B is not below the direct-measurement infidelity, the fewer-measurements claim fails. Repeat with QAE by replacing shots with total AA circuit depth to see if the quadratic speedup restores the comparison.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Sec. IV B states that a mean of 28 overlap calculations for n=5 QPE spectra is fewer than the 2^5=32 computational-basis amplitudes, 'thereby demonstrating the superiority of the proposed method.' This equates one overlap evaluation with one measurement. But an overlap evaluation (SWAP/SWITCH test) is a quantum circuit whose output must be sampled many times to estimate the inner product to a given precision. In the paper's own cost model (Sec. II E), the total quantum cost is O(c_tgt m_iter/ε^2) without QAE and O(c_tgt m_iter/ε) with QAE; there is no fixed ε per overlap specified in the numerics. The proposed method's infidelity curve in Fig. 7(b) is explicitly 'ignoring the statistical error,' while the direct-measurement baseline includes statistical error from 1,000 measurements. Thus the central quantitative claim—that the method requires fewer measurements than direct readout—is not established: even if the LCLF approximation is exact, the total measurement budget could exceed direct readout once each overlap is sampled. This is a metric problem independent of the structural LCLF assumption.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a hybrid quantum-classical protocol for reading out a quantum state that is well approximated by a linear combination of discrete Lorentzian functions (LCLF). The target state's overlaps with shifted Lorentzian basis states are estimated on a quantum processor via SWITCH or SWAP tests, then a classical optimizer adjusts the Lorentzian centers, decay rates, and expansion coefficients by solving a generalized eigenvalue problem (phase-sensitive readout) or a linear least-squares problem (amplitude-squared readout). The authors present quantum circuits, including an amplitude-estimation enhancement, and give an asymptotic cost model in Sec. II E. Numerical demonstrations reconstruct a grid-based wavefunction for n=5..10 qubits and a synthetic QPE-based X-ray absorption spectrum for n=5, and the paper claims the method requires fewer measurements than conventional direct state readout because the number of overlap evaluations is small and approximately independent of n.","tokens_in":17333,"tokens_out":4082,"duration_ms":42394,"significance":"If the resource claim were established, the protocol would be a useful tool for readout of structured quantum states, particularly in first-quantized simulation and spectral-function settings. The core algebraic formulation is sound: the rank-one GEVP for fixed centers and decay rates, the least-squares fit for squared amplitudes, and the analytical overlap formulas in Appendix C are concrete and checkable. The paper also provides explicit circuits for SWITCH/SWAP tests and QAE, and it carefully separates quantum and classical costs in Sec. II E. These are genuine strengths. However, the headline quantitative claim—that the method outperforms direct measurement in total measurement cost—is not supported by the numerics as presented, because the comparison mixes overlap evaluations with measurement shots and omits statistical error for the proposed method. The paper also inherits an approximation error floor from the LCLF ansatz and reports occasional optimization failures at n=9,10, which need to be quantified before the scalability claim is accepted.","major_comments":[{"comment":"The central claim of requiring fewer measurements is not established because the comparison counts overlap evaluations, not measurement shots, and because the proposed method's infidelity is plotted without statistical error. In Fig. 7(b), the black broken line for the LCLF method explicitly ignores statistical error, while the direct-measurement baseline includes shot noise from 1,000 measurements. In Sec. IV B, the statement that a mean of 28 overlap calculations for n=5 is fewer than 2^5=32 computational-basis amplitudes equates one overlap evaluation with one measurement. But each SWITCH/SWAP overlap estimate is itself a random variable requiring O(1/epsilon^2) shots (or circuit depth O(1/epsilon) with QAE), as the paper's own cost model in Sec. II E states. The comparison should fix a target total infidelity, include statistical error for the proposed method, and compare total numbers of shots (or total circuit depth) for both approaches. As written, even if the LCLF approximation were exact, the total measurement budget for the proposed method could exceed direct readout once each overlap is sampled to the required precision.","section":"Sec. IV B and Fig. 7(b)"},{"comment":"The scalability claim that infidelity is independent of n is demonstrated only for the idealized case of known optimal centers and decay rates, not for the full optimization used to find those parameters. The text states that for n=9 and n=10 the final infidelity remained above 0.01 because the optimal peak positions were not found in 1 out of 10 trials. This means the readout protocol has a non-vanishing failure probability at the largest sizes tested, yet the paper's summary claim does not mention this caveat. The authors should either report the full distribution of infidelities including failed trials, show the success probability as a function of n, or clearly condition the scalability claim on convergence of the classical optimizer.","section":"Sec. IV A and Fig. 7(e)"},{"comment":"The metric m_iter, the number of unique overlap evaluations, is not by itself a quantum resource metric. The paper argues that m_iter is approximately independent of n, but the actual quantum cost is c_tgt * m_iter / epsilon^2 (or /epsilon with QAE), with no specification of the precision epsilon per overlap used in the numerics. The apparent flatness of Fig. 7(f) therefore does not, by itself, imply that the method scales favorably relative to direct readout; the total cost includes both the number of overlap evaluations and the shot budget per evaluation, and the latter depends on the required accuracy of the LCLF fit, which in turn depends on the approximation error floor. The authors should present a total-resource comparison for a fixed target infidelity, including the per-overlap shot count and the contribution of the LCLF approximation error.","section":"Sec. II E and Fig. 7(f)"}],"minor_comments":[{"comment":"The displayed stationary condition contains an error: the final term 'edℓ′ = 0' is not part of the equation and should be removed; also the summation index ℓ' is used inconsistently in the same line.","section":"Eq. (8)"},{"comment":"The index structure of the derivative formula is unclear; it should specify which indices are summed and which are fixed, to avoid confusion between the derivative index ℓ and the summation indices.","section":"Eq. (13)"},{"comment":"The caption should explicitly state which curves include statistical error and which do not, and the proposed method should be shown with error bars or shaded confidence intervals in all panels, not only in the direct-measurement baseline.","section":"Sec. IV A, Fig. 7 caption"},{"comment":"The sentence 'In these cases, the optimal peak positions were not found in only 1 out of 10 trials' is ambiguous: it could mean one failed trial in each of the two sizes, or one failure total, or that the final infidelity stayed above 0.01 in only those failed trials. Please reword to state exactly how many of the ten trials converged for each n.","section":"Sec. IV A, paragraph on n=9 and n=10"},{"comment":"Please define precisely what counts as one 'quantum overlap calculation' in m_iter, including whether cached evaluations are reused, because this quantity is the basis for the claimed superiority over direct measurement.","section":"Sec. IV B and Fig. 8(c)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a reasonable incremental contribution building on the authors' earlier encoding work, and the core linear-algebra formulation is sound. The main problem is the unsupported resource comparison, which is fixable by redoing the numerical comparison in terms of total measurement shots or total circuit depth with statistical error included. I recommend major revision rather than rejection, provided the authors address the comparison metric and the optimization-failure rate at larger n."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real technique for reading out quantum states that are linear combinations of a few localized Lorentzians, and the circuits and fitting algebra are presented cleanly. But the headline efficiency claim—fewer measurements than direct readout—is not supported by the current numerics, because they compare overlap evaluations to measurement shots and omit the proposed method's statistical error in the key plot.\n\nWhat's new: prior work from the same group [30,31] encoded states as LCLFs; here they invert that direction and estimate overlaps to fit the expansion. The SWITCH/SWAP test circuits, the GEVP for the coefficients, and the least-squares version for squared amplitudes are all straightforward and the derivations check out. The analytical overlap formulas in Appendix C are useful. The QPE-LF application is a natural match—spectra are sums of Lorentzians, so the readout is tailored to the structure.\n\nSoft spots: the resource comparison is apples-to-oranges. Fig. 7(b) plots infidelity ignoring statistical error for the proposed method, but the direct-measurement baseline includes shot noise. And Sec. IV B counts one overlap evaluation as one measurement; a SWAP/SWITCH test requires many shots (or QAE depth) to estimate the inner product to a fixed epsilon. The paper's own cost model has O(1/epsilon^2) per overlap, but they never fix epsilon in the numerics. So the \"mean 28 vs 32\" claim doesn't demonstrate superiority in measurement budget. Also the cost model ignores the fact that the SWITCH test requires a controlled target unitary—that's not always cheap. The optimization fails in 1 of 10 trials for n=9,10, which is honest but means the method as presented isn't robust at larger sizes. And no comparison against compressed sensing or classical shadows, which are the obvious benchmarks for this task. The structural assumption (target well approximated by small LCLF span) is clearly stated, so no circularity—it's just a conditional advantage.\n\nVerdict: the machine is well built for a specific niche. The central quantitative claim needs rework: either provide a fixed-epsilon budget for each overlap and compare total shots, or drop the 'fewer measurements' claim and frame it as 'fewer overlap evaluations, each requiring additional sampling.' That is an honest framing. I'd send to peer review; the technique deserves a serious referee, but it needs major revision on the resource analysis.","headline":"A real overlap-based readout technique for localized states, but the 'fewer measurements' claim compares overlap evaluations to shots and omits the proposed method's statistical error—needs a fair resource comparison.","tokens_in":17850,"tokens_out":2112,"would_cite":false,"duration_ms":21330,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":["03.67.-a"],"model":"deepseek-v4-flash","headline":"This paper claims that an unknown quantum state that is close to a sum of a few localized peaks can be reconstructed from overlap measurements whose number does not grow with the number of qubits.","keywords":["quantum state readout","overlap-based fitting","Lorentzian function states","quantum phase estimation","X-ray absorption spectra","quantum amplitude estimation","SWAP test","state tomography"],"falsifier":"Run the protocol on a target state deliberately far from the Lorentzian ansatz, such as random signs on all computational-basis amplitudes or a dense comb of closely spaced peaks, and count the SWAP/SWITCH overlap evaluations needed to reach an infidelity below 0.01. If that count grows with $N=2^n$, or if the optimizer fails in a constant fraction of trials as $n$ increases, the claimed qubit-independent readout cost does not hold.","tokens_in":16840,"feed_emoji":"⚛️","tokens_out":12448,"duration_ms":113897,"temperature":0.7,"pith_summary":"This paper claims that a quantum state shaped like a few localized peaks can be read out by measuring overlaps between the unknown state and shifted bell-shaped Lorentzian basis states, then fitting those overlap values on a classical computer. The reconstruction is a linear combination of Lorentzian states, and the fitted peak positions, widths, and coefficients serve as the extracted features. The numerical results show that for a Gaussian-like target, the reconstruction infidelity stays near $7.1\\times10^{-3}$ for $n=5$ through $10$ qubits, so the required measurement count does not grow with system size, unlike direct readout. For quantum-phase-estimation spectra, reconstructing the squared amplitudes needed on average 28 overlap evaluations versus 32 computational-basis amplitudes for $n=5$, meaning fewer measurements than direct basis-by-basis readout. This matters because complete state readout is a bottleneck in quantum simulations of continuous functions such as wavefunctions and spectra.","feed_headline":"Peaked quantum states read out at qubit-independent cost","feed_subtitle":"Fitting a quantum state to a few bell-shaped peaks needs fewer measurements than reading every basis amplitude.","key_machinery":"The load-bearing object is the shifted discrete Lorentzian function state $|L;a,k_c\\rangle$: a normalized bell-shaped wavefunction with power-law tails, decay rate $a$, and center $k_c$, prepared from a discrete Slater state by a quantum Fourier transform and a translation operator. The method represents the unknown target as a linear combination of $n_{\\rm loc}$ such states and turns state readout into overlap-based fitting: SWITCH tests supply complex overlaps when phases are needed, SWAP tests supply squared overlaps when only probabilities are needed, and a classical generalized-eigenvalue solve extracts the optimal coefficients from the measured overlap matrix. The analytical formula for the Lorentzian overlap matrix keeps classical cost low, and in the quantum-phase-estimation application two inverse quantum Fourier transforms cancel, shortening the circuit that feeds the fit.","core_discovery":"The central claim is that an unknown $n$-qubit target state $|\\psi_{\\rm tgt}\\rangle$ that is well approximated by a linear combination of $n_{\\rm loc}$ shifted discrete Lorentzian function states can be read out by maximizing the fidelity $F=|\\langle\\psi_{\\rm tgt}|\\psi_{\\rm LCLF}\\rangle|^2$. For fixed centers and widths, the optimal coefficients are the largest-eigenvalue eigenvector of the generalized eigenvalue problem $G\\mathbf{d}=\\kappa S\\mathbf{d}$, where $G$ is built from measured overlaps between the target and each Lorentzian state and $S$ is the analytically known Lorentzian overlap matrix. The peak centers and decay rates are then optimized classically, with gradients evaluated by finite differences and additional overlap measurements. The paper demonstrates two variants: a phase-sensitive readout using the SWITCH test, and a probability-only readout using the SWAP test that suffices for spectral functions. The numerical evidence is that reconstruction infidelity is independent of the number of qubits, and that for quantum-phase-estimation spectra the number of overlap evaluations needed is smaller than the number of computational-basis amplitudes.","pith_inferences":["Editorial inference: The same overlap-fitting scheme should transfer to other localized basis families, such as Gaussians or wavelets, making it a general primitive for reading out smooth quantum states rather than a Lorentzian-specific trick.","Editorial inference: The qubit-independence claim is conditional on the optimizer finding the right peaks; the paper's own 1-in-10 failure to converge for $n=9$ and $n=10$ suggests the practical scaling cliff may appear earlier than the asymptotic statement implies.","Editorial inference: A natural stress test is to apply the method to two very close Lorentzian peaks; the overlap matrix becomes nearly singular, and the number of measurements or the classical conditioning may degrade even though the state is still a small linear combination.","Editorial inference: Combining this readout with adaptive initial centers, for example from mean-field solutions, could turn the measurement-count advantage into an end-to-end protocol for iterative quantum simulations, since the solver's previous step supplies the next guess."],"forward_implications":["For states that are sums of a few localized peaks, full state readout requires a number of overlap evaluations that does not grow with the number of qubits; the paper reports infidelity near $7\\times10^{-3}$ for $n=5$ through $10$ with 1,000 measurements.","For quantum-phase-estimation spectra, reading out squared amplitudes costs on average 28 overlap evaluations for $n=5$, fewer than the 32 computational-basis amplitudes direct readout would require.","The fitting procedure gives two readout modes: phase-sensitive reconstruction through the SWITCH test, and probability-only reconstruction through the SWAP test, the latter being sufficient for spectral quantities such as X-ray absorption spectra.","Because peak centers are optimized by Metropolis sampling, the number of quantum overlap evaluations grows roughly in proportion to the Hamming distance between the initial and true peak positions, so good initial guesses directly reduce measurement cost.","Quantum amplitude estimation can be applied to the SWAP and SWITCH tests, replacing the Monte Carlo $O(1/\\varepsilon^2)$ shot-count scaling with a Heisenberg-limited $O(1/\\varepsilon)$ circuit-depth scaling."],"supporting_citations":[{"why":"It supplies the discrete Lorentzian function states that the readout uses as its expansion basis.","marker":"[30]"},{"why":"It supplies the maximal-fidelity fitting and generalized eigenvalue formulation used to set the expansion coefficients.","marker":"[31]"},{"why":"It supplies the SWITCH and SWAP overlap-measurement tests used to evaluate inner products between the target and the basis states.","marker":"[37]"},{"why":"It supplies the discrete Slater function state and its shallow preparation circuit, from which the shifted Lorentzian states are built.","marker":"[36]"},{"why":"It supplies quantum amplitude estimation, which the paper uses to reduce the measurement cost of each overlap.","marker":"[42]"},{"why":"It supplies the quantum-phase-estimation route to optical absorption spectra that motivates the probability-only readout variant.","marker":"[33]"},{"why":"It supplies the QPE input state whose Lorentzian broadening makes the spectral readout problem naturally fit the method.","marker":"[55]"}],"fun_headline_variants":["Overlap-based readout needs fewer measurements","Quantum states read out via Lorentzian overlap","Peak fitting cuts quantum readout measurements","Overlap feature extraction trims quantum sampling"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the quantum state being read out is almost a sum of a few bell-shaped peaks whose positions and widths the classical optimizer can find; if the state has many peaks, broad nonlocal structure, or random sign fluctuations, the measurement savings and accuracy claims no longer hold.","fun_headline_variants_meta":{"raw":{"variants":["Overlap-based readout needs fewer measurements","Quantum states read out via Lorentzian overlap","Peak fitting cuts quantum readout measurements","Overlap feature extraction trims quantum sampling"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000261,"raw_usage":{"total_tokens":1573,"prompt_tokens":902,"completion_tokens":671,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":518,"completion_tokens_details":{"reasoning_tokens":616}},"tokens_in":518,"tokens_out":671,"duration_ms":7117,"temperature":1.0,"reasoning_tokens":616,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:50:07.353333+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the protocol on a target state deliberately far from the Lorentzian ansatz, such as random signs on all computational-basis amplitudes or a dense comb of closely spaced peaks, and count the SWAP/SWITCH overlap evaluations needed to reach an infidelity below 0.01. If that count grows with $N=2^n$, or if the optimizer fails in a constant fraction of trials as $n$ increases, the claimed qubit-independent readout cost does not hold.","supporting_citations":[{"cited_title":"Brassard and P","cited_arxiv_id":null,"evidence_quote":"It supplies quantum amplitude estimation, which the paper uses to reduce the measurement cost of each overlap."},{"cited_title":"Kosugi, S","cited_arxiv_id":null,"evidence_quote":"It supplies the discrete Lorentzian function states that the readout uses as its expansion basis."},{"cited_title":"Tensor decomposition technique for qubit encoding of maximal-fidelity Lorentzian orbitals in real-space quantum chemistry","cited_arxiv_id":"2501.07211","evidence_quote":"It supplies the maximal-fidelity fitting and generalized eigenvalue formulation used to set the expansion coefficients."},{"cited_title":"Klco and M","cited_arxiv_id":null,"evidence_quote":"It supplies the discrete Slater function state and its shallow preparation circuit, from which the shifted Lorentzian states are built."},{"cited_title":"Kosugi and Y.-i","cited_arxiv_id":null,"evidence_quote":"It supplies the quantum-phase-estimation route to optical absorption spectra that motivates the probability-only readout variant."},{"cited_title":"Kosugi and Y.-i","cited_arxiv_id":null,"evidence_quote":"It supplies the QPE input state whose Lorentzian broadening makes the spectral readout problem naturally fit the method."}],"review_version":1}