{"id":"813024e4-35e5-48c6-b765-455210cdf87c","arxiv_id":"2412.16293","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A SPAM-calibration measurement can correct linear-inversion quantum process tomography via the gauge-fixed formula Ghat = Ehat^{-1/2} G0 Ehat^{-1/2}, reducing SPAM-induced bias.","lead":"This paper gives a simple add-on to standard quantum process tomography that removes bias caused by imperfect state preparation and measurement, using about twice as much data. It is a practical recipe that experimental quantum computing groups can apply without the heavy machinery of gate set tomography.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. 3 only puts the estimate on the gauge orbit of G; the claimed fidelity improvement over standard QPT is not proven and may depend on SPAM asymmetry and the user's gauge parameter p.","rationale":"The reader's CONDITIONAL verdict is appropriate. I considered the stationarity/invertibility assumption identified by the reader; it is a real experimental condition but is standard for self-calibration protocols and not the weakest link in the paper's own argument. The more load-bearing gap is that Eq. (3)'s algebraic content—forcing the estimate onto the gauge orbit—guarantees eigenvalue immunity but not fidelity improvement. Since the abstract and simulations advertise accuracy/fidelity, this missing implication is the part of the central claim most exposed. The proposed random-model sweep is a direct, low-cost check. If it fails, the paper should either restrict its claim to gauge-invariant quantities or provide a principled rule for choosing p; if it passes, the claim is strengthened. This keeps the verdict at CONDITIONAL.","tokens_in":8699,"tokens_out":29860,"duration_ms":263276,"concrete_test":"Randomly sample single-qubit SPAM superoperators A and B with varying asymmetry (depolarizing and coherent errors of different strengths and relative magnitudes), simulate standard QPT and better QPT with p=0, 0.5, 1 for a fixed Xπ/2 gate with identical shot counts, and compute the error in estimated process fidelity relative to the true lab-frame G. If any realistic model (in particular |A| >> |B| or |B| >> |A| with p=1/2) yields larger fidelity error for better QPT than for standard QPT, the advertised 'more accurate' claim requires qualification; if an exhaustive sweep finds none, the empirical claim is supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"With true P = MGS and calibration I = MS, writing A = M0^{-1}M and B = S S0^{-1} gives Ghat0 = AGB and Ehat = AB. Substituting into Eq. (3) (or the p-weighted version) yields Ghat(p) = E^{p-1} A G B E^{-p} = (E^p B^{-1}) G (B E^{-p}) = C_p G C_p^{-1}. So the correction provably lands Ghat on the gauge orbit of G, which makes eigenvalues exactly SPAM-immune. But it does not by itself make any gauge-dependent figure of merit—in particular the process fidelity used in Figure 1—closer to its lab-frame value. The distance along the orbit is set by C_p, which depends on the actual (unobserved) state-preparation error B and on the user's chosen p. The paper's claim that SPAM-corrected QPT is 'more accurate' is therefore an empirical statement about three simulated SPAM models, not a consequence of Eq. (3). A wrong prior for p, or a strongly asymmetric SPAM error under the default p=1/2, could in principle move the estimate along the orbit away from the true gate, and the paper's footnote 1 already admits unphysical fidelity values for large coherent SPAM. The paper provides no bound or criterion for when Eq. (3) outperforms Eq. (2) for fidelity.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a modification of standard linear-inversion QPT that adds a SPAM-calibration experiment measuring the Gram matrix I = MS. From the two datasets, I = MS and P = MGS, the authors construct an estimate of the SPAM error superoperator E = M0^{-1} M S S0^{-1} and correct the naive process estimate via Ghat = Ehat^{-1/2} G0 Ehat^{-1/2} (or a generalization with a gauge parameter p). They claim this yields more accurate and less biased gate estimates, show simulations on three single-qubit SPAM models, and extend the idea to overcomplete data and to maximum-likelihood estimation. The core algebraic identity is correct: the corrected estimate lies on the gauge orbit of the true process, so gauge-invariant quantities such as process eigenvalues are exactly immune to SPAM error.","tokens_in":9032,"tokens_out":6550,"duration_ms":60067,"significance":"If the practical claims hold, the protocol gives a substantial accuracy improvement over standard QPT at the cost of only one additional calibration experiment, and the paper's closed-form correction is simple enough for immediate adoption. The simulations are clearly described and show consistent improvement on three SPAM models, and the overcomplete extension is a useful contribution. The paper also correctly identifies the gauge freedom and explains why gauge-invariant quantities are robust. However, the mathematical guarantee is limited to gauge-invariant properties, and the paper's stronger claim that the corrected estimate is 'more accurate' in terms of gate fidelity is not established by the algebra and is only empirically supported by the simulations.","major_comments":[{"comment":"The paper's headline claim that SPAM-corrected QPT is 'more accurate' is not a consequence of Eq. (3). The derivation shows only that the corrected estimate lies on the gauge orbit of the true process G; specifically, Ghat(p) = C_p G C_p^{-1} with C_p = E^p B^{-1} and B = S S0^{-1}. Consequently, any gauge-dependent figure of merit, such as the process fidelity used in Figure 1, changes by an amount controlled by the unobserved state-preparation error B and the user's choice of p. The paper provides no bound or criterion for when the corrected fidelity estimate is closer to the truth than the uncorrected estimate. The fidelity improvement is an empirical observation for the three simulated SPAM models, and the paper should state this limitation explicitly or supply an analysis of the conditions under which improvement is guaranteed.","section":"Section 3, Eq. (3) and abstract"},{"comment":"The admission that 'the estimated gate fidelity can be greater than 1 for large coherent SPAM errors' directly contradicts the introduction's claim that the corrected estimate is 'truthfully higher-fidelity.' As presented, Eq. (3) can return unphysical estimates. The paper should either restrict the fidelity claim to sufficiently small SPAM errors or incorporate the numerical gauge constraint mentioned in the footnote into the protocol itself. Without this, the central 'easy better' message is overstated.","section":"Footnote 1"},{"comment":"The derivation assumes that the same SPAM operations M and S appear in both the calibration experiment (I) and the process experiment (P). This assumption is not stated as a formal condition. If SPAM drifts between the two data acquisitions, Ehat no longer describes the SPAM affecting the process data, and Eq. (3) can in principle move the estimate to a point on the gauge orbit that is farther from the truth. The paper only discusses a related failure in Section 5 for non-Markovian errors that enlarge the active state space; drift is a more general and equally realistic limitation that should be acknowledged.","section":"Section 3 and Section 5"},{"comment":"The corrected estimate requires computing Ehat^{-1/2}, which presumes that Ehat is invertible and has a well-defined square root. The paper does not discuss conditions on E (e.g., no zero eigenvalues, or for real square roots, no negative real eigenvalues), nor does it specify which square-root branch is intended. For SPAM error superoperators that are not diagonalizable or not positive, Ehat^{-1/2} may be undefined or complex. This is a technical gap in the closed-form correction that should be addressed, even if the simulations happen to avoid it.","section":"Section 3, Eq. (3)"}],"minor_comments":[{"comment":"The phrase 'relatively näive' contains a typo; it should be 'relatively naive.'","section":"Section 6"},{"comment":"The notation E^{1/2} is used without specifying that it is a matrix square root. For clarity, state that E is a d^2 x d^2 matrix in the Liouville representation and that a principal square root is intended when E is Hermitian positive semidefinite.","section":"Section 3, Eq. (3)"},{"comment":"The caption describes the shot count and circuit count in the main text, but including the circuit counts (12 for standard QPT, 24 for SPAM-corrected) directly in the caption would improve readability.","section":"Figure 1 caption"},{"comment":"The 'suggestive approximation' after Eq. (11) is presented without stating that it is not used in the simulations and is only heuristic; please label it as such.","section":"Section 5, Eq. (11)"}],"recommendation":"major_revision","confidential_remarks":"The paper has a simple, elegant idea that is likely to be useful to the QPT community, and the simulations support the practical benefit for the tested SPAM models. However, the abstract and introduction overstate the theoretical status of the accuracy claim, and the internal inconsistency with footnote 1 needs to be resolved. The requested changes—qualifying the fidelity claim, stating the SPAM stationarity assumption, and addressing the square-root conditions—are all within the scope of a revision and do not require new conceptual machinery."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a well-written, accessible recipe for removing SPAM bias from linear-inversion QPT by adding one calibration experiment. The math is correct under the stated assumptions, and the eigenvalue immunity is a genuine formal result. But the advertised \"more accurate\" claim is not proven—it's an empirical observation from three simulations—and the core trick is essentially the same gauge-freedom idea the authors already used in their earlier closed-form self-consistent tomography (Ref. [8]). Worth taking seriously, but the framing needs care.\n\nThe paper's main contribution is practical clarity. It spells out the correction Ghat = E^{-1/2} G0 E^{-1/2} (and its p-weighted generalization) in a way that any QPT user can apply, with a clear derivation. The overcomplete pseudoinverse version in Section 5 is a useful extension. The simulations in Figure 1 cover three distinct SPAM-error models and show consistent improvement in estimated fidelity, and Figure 2 correctly demonstrates that the process eigenvalues are exactly immune to SPAM error—that part is a straightforward consequence of the gauge-orbit argument, and it's a nice point to make explicitly.\n\nThe main soft spot is the headline claim about accuracy. As your stress-test note shows, the correction places the estimate on the gauge orbit of the true process, which fixes the gauge-invariant content but not the gauge-variant fidelity. Whether the corrected fidelity is closer to the true value depends on the actual SPAM asymmetry and the user's choice of p. The paper's own footnote 1 admits that large coherent SPAM can produce unphysical fidelities. So \"more accurate\" is a statement about the simulated models, not a theorem. That's a legitimate limitation, but not a fatal one—the recipe delivers exactly what it says in the gauge-invariant sense, and the simulations do show the advertised behavior in realistic regimes.\n\nA second soft spot is that the stationarity assumption (same M and S for calibration and gate data) is only mentioned in passing. If SPAM drifts, the correction is simply not applicable, and the paper should say so more prominently. Also, the novelty relative to Ref. [8] should be stated explicitly; the closed-form tomography paper already used the same gauge correction idea. That doesn't make this paper valueless—it's a simpler and more accessible presentation—but the authors should be clear about what's new.\n\nMinor: no code or data are provided, which would help others apply the recipe.\n\nWho is this for? Experimentalists and labs that use QPT and want a low-cost fix. It's a practical methods paper, not a deep theoretical advance. The algebra is sound and the recipe is likely useful, so it deserves peer review, but the final version should either temper the accuracy claim or add a condition under which it holds.","headline":"A clean, practical recipe for SPAM-corrected QPT with a correct gauge-invariant core; the headline fidelity claim is only empirically supported, and the novelty over the authors' earlier work should be stated.","tokens_in":9543,"tokens_out":3451,"would_cite":true,"duration_ms":30036,"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":"SPAM-calibrated QPT removes most bias with only double the data.","keywords":["quantum process tomography","SPAM error","gauge freedom","linear inversion","maximum likelihood estimation","gate eigenvalues","SPAM calibration"],"falsifier":"Run the protocol on a known gate with SPAM that drifts between the calibration and process circuits, then check whether the corrected gate eigenvalues remain accurate; if they drift with the delay time, the claimed SPAM immunity fails.","tokens_in":8498,"feed_emoji":"⚛️","tokens_out":5617,"duration_ms":45368,"temperature":0.7,"pith_summary":"Quantum process tomography (QPT) normally assumes the states used for preparation and the measurements used to probe a gate are exactly known, which is never true and biases the resulting gate estimate. This paper shows that running one extra \"process tomography on nothing\" experiment, measuring the SPAM operations on their own, gives enough information to correct that bias. The correction is explicit and simple: estimate the SPAM error superoperator from the calibration data and conjugate the standard linear-inversion estimate by its square root. The result is a gate estimate whose gauge-invariant properties, such as the eigenvalues of the process matrix, are exactly immune to SPAM error, at the cost of only twice as much data.","feed_headline":"Add one calibration run to fix SPAM bias in QPT","feed_subtitle":"A single extra calibration experiment corrects the SPAM bias in standard quantum process tomography.","key_machinery":"The central object is the SPAM error superoperator $E = M_0^{-1} I S_0^{-1}$, estimated from calibration data as $\\hat{E} = M_0^{-1} \\hat{I} S_0^{-1}$. It captures how the true state preparations and measurement effects deviate from the a priori matrices $M_0$ and $S_0$. The load-bearing identity is the correction $\\hat{G} = \\hat{E}^{-1/2} \\hat{G}_0 \\hat{E}^{-1/2}$, which follows from choosing the factorization $\\hat{E} = \\alpha \\beta$ with $\\alpha = \\beta = \\hat{E}^{1/2}$ so that the SPAM error is split equally between states and effects; for overcomplete data the same logic uses Moore-Penrose pseudoinverses and rank-$d^2$ truncation of $\\hat{I}$.","core_discovery":"The paper's central claim is that standard linear-inversion QPT can be made SPAM-robust by measuring the SPAM calibration matrix $I = MS$, estimating the SPAM error superoperator $E = M_0^{-1} I S_0^{-1}$, and replacing the naive estimate $\\hat{G}_0 = M_0^{-1} \\hat{P} S_0^{-1}$ with $\\hat{G} = \\hat{E}^{-1/2} \\hat{G}_0 \\hat{E}^{-1/2}$. This symmetric square-root split assigns half the observed SPAM error to state preparation and half to measurement, which is the gauge choice that respects the a priori SPAM model as much as possible. The same procedure extends to overcomplete data by using pseudoinverses and truncating the calibration matrix to rank $d^2$, and to principled estimators like maximum likelihood by treating the SPAM operations as nuisance parameters. The paper demonstrates the correction on simulated single-qubit X$\\pi/2$ gates with three SPAM error models, showing that the corrected estimates of gate fidelity are more accurate than standard QPT in every case, and that gate eigenvalues are completely unaffected by SPAM error.","pith_inferences":["A direct testable consequence not pursued in the paper: the size of the SPAM correction can be used to flag unstable experiments, because if the calibration data and process data do not share the same SPAM, the corrected eigenvalues will drift with the time between the two data sets.","The gauge choice $p=0.5$ is a prior, not a measurement; users who have independent estimates of which of state preparation or measurement is noisier can set $p$ accordingly, and the paper's simulations suggest the residual error penalty for a wrong guess is small for small SPAM error.","The method could be combined with randomized benchmarking or other reference-free protocols to provide a low-cost SPAM-robust fidelity estimate, though the paper does not develop this connection."],"forward_implications":["Anyone already performing standard QPT can obtain a more accurate gate estimate simply by adding one calibration experiment and applying Eq. (3), with no new hardware or complex analysis.","Gauge-invariant quantities extracted from the process estimate, such as the eigenvalues of the transfer matrix, become exactly immune to SPAM error, so they can be used as reliable diagnostic figures even when SPAM is unknown.","For overcomplete data sets, the correction is still available through pseudoinverses and low-rank truncation, making the method applicable to typical experimental data with more circuits than the minimal informationally complete set.","The same calibration strategy can be incorporated into maximum-likelihood estimation, either independently, sequentially, or jointly, so statistically optimal estimates can also be freed from SPAM bias."],"supporting_citations":[{"why":"Defines the standard QPT linear-inversion model $P = M_0 G S_0$ that this paper corrects.","marker":"[1]"},{"why":"Introduces self-consistent process tomography, establishing the SPAM-bias problem and the calibration strategy this work simplifies.","marker":"[6]"},{"why":"Provides the gate-set-tomography framing that motivates treating SPAM as unknown and learning it from calibration data.","marker":"[9]"},{"why":"Formulates the gauge freedom in gate sets that makes the division of SPAM error between states and effects unobservable.","marker":"[11]"},{"why":"Articulates the gauge-invariant information content of quantum channels, supporting the use of process eigenvalues as SPAM-immune quantities.","marker":"[13]"},{"why":"Supplies the maximum-likelihood principle the paper proposes to use for statistically principled SPAM-corrected estimation.","marker":"[16]"}],"fun_headline_variants":["One calibration run fixes SPAM bias in QPT","Easy SPAM-robust QPT via symmetric correction","Calibrate once, split SPAM error, get better QPT","Gauge fix: square-root SPAM correction improves QPT","Single extra measurement yields unbiased quantum process tomography"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The correction breaks down if the state preparation and measurement operations change between the calibration experiment and the process experiment, or if the estimated SPAM error superoperator is singular so its square root is not well defined.","fun_headline_variants_meta":{"raw":{"variants":["One calibration run fixes SPAM bias in QPT","Easy SPAM-robust QPT via symmetric correction","Calibrate once, split SPAM error, get better QPT","Gauge fix: square-root SPAM correction improves QPT","Single extra measurement yields unbiased quantum process tomography"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000476,"raw_usage":{"total_tokens":2325,"prompt_tokens":872,"completion_tokens":1453,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":488,"completion_tokens_details":{"reasoning_tokens":1372}},"tokens_in":488,"tokens_out":1453,"duration_ms":10396,"temperature":1.0,"reasoning_tokens":1372,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:44:09.603802+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the protocol on a known gate with SPAM that drifts between the calibration and process circuits, then check whether the corrected gate eigenvalues remain accurate; if they drift with the delay time, the claimed SPAM immunity fails.","supporting_citations":[{"cited_title":"Prescription for experimental determination of the dynamics of a quantum black box","cited_arxiv_id":null,"evidence_quote":"Defines the standard QPT linear-inversion model $P = M_0 G S_0$ that this paper corrects."},{"cited_title":"Self-consistentquantumprocesstomography","cited_arxiv_id":null,"evidence_quote":"Introduces self-consistent process tomography, establishing the SPAM-bias problem and the calibration strategy this work simplifies."},{"cited_title":"Gate set tomography","cited_arxiv_id":null,"evidence_quote":"Formulates the gauge freedom in gate sets that makes the division of SPAM error between states and effects unobservable."},{"cited_title":"Gauge invariant information concerning quantum channels","cited_arxiv_id":null,"evidence_quote":"Articulates the gauge-invariant information content of quantum channels, supporting the use of process eigenvalues as SPAM-immune quantities."},{"cited_title":"Quantum-state estimation","cited_arxiv_id":null,"evidence_quote":"Supplies the maximum-likelihood principle the paper proposes to use for statistically principled SPAM-corrected estimation."}],"review_version":1}