{"id":"147ed80e-d54e-4d4d-838a-6aeacfbc1485","arxiv_id":"2508.17513","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"RIDA estimates a circuit's global depolarization probability from a random half-gate identity circuit and uses it to amplify noisy expectation values.","lead":"This paper introduces RIDA, an error-mitigation method that estimates how strongly noise depolarizes a quantum circuit by running a random half of its gates in reverse, then corrects measured expectation values. Benchmarks on simulated IBM-like hardware suggest large accuracy gains over standard zero-noise extrapolation and CNOT-only depolarization methods.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"RIDA's unbiasedness proof assumes every gate error fully depolarizes the observable; under the paper's own local depolarizing model, p0 measured from an identity circuit is observable- and order-dependent, so the universal claim is not established.","rationale":"The reader's weakest assumption correctly identifies the first-order independent-error model as load-bearing. My concern is in the same area but is more fundamental: even under independent local depolarizing noise, the measured p0 from an identity circuit is not 2 sum epsilon_i because the contribution of each Pauli error to <O0> depends on whether the error, after conjugation by subsequent gates, anticommutes with O0. Thus the unbiasedness condition E[p0] = p can fail at first order, not only when errors are correlated, reversible, or large. The paper's numerical evidence is real and reasonably broad, and the offered code is a point in its favor; however, the simulations use EfficientSU2 circuits and random Pauli observables, which may be favorable to the method. The central analytic claim of universality requires either a proof that the observable-dependent weights c_i and d_i are equal in expectation under the estimation-circuit construction, or an explicit condition delimiting when p0 approximates p. Since this gap is addressable and the empirical results remain suggestive, the reader's CONDITIONAL verdict stands unchanged; the concrete test above would settle whether the concern actually lands in practice.","tokens_in":24758,"tokens_out":12035,"duration_ms":137556,"concrete_test":"For n = 4 to 7 qubits, generate target circuits from a structured ansatz with varying fractions of Z-type single-qubit rotations (e.g., layers of Rz and CNOT gates) and random Pauli observables. For each circuit, compute the exact noisy density matrix under the same local depolarizing model, fit the effective target depolarization p_eff from <O_noisy> = (1 - p_eff)<O>, and run RIDA's estimation circuit under the same noise model to obtain p0. Plot |p0 - p_eff| versus circuit depth and versus the fraction of Z-type gates. If |p0 - p_eff| is systematically nonzero and grows with depth for circuits with many Z rotations, the unbiasedness claim fails; if it stays within shot noise for all tested circuits, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"SM Section I.A defines p0 = 1 - product(1 - epsilon_i)^2 ≈ 2 sum epsilon_i, treating each gate error as fully depolarizing independent of circuit position. But under the paper's own local depolarizing noise model, p0 = 1 - <O0_epsilon> for the identity estimation circuit is not 2 sum epsilon_i: a Pauli error at gate i contributes to <O0> only if, after conjugation by the remaining gates, it anticommutes with O0. The correct first-order expression is p0 = 2 sum epsilon_i c_i with c_i in [0,1] depending on gate order and the choice of O0, while the target circuit's effective depolarization parameter is p = sum epsilon_i d_i with generally different weights d_i. Matching half the one-qubit and half the two-qubit gate counts does not force E[p0] = p; it only matches gate counts and, under the idealized formula, first moments of the error rates. The numerical success on EfficientSU2 circuits with random Pauli observables may reflect that random rotations make most errors anticommute with the measured observable, but the claimed universality and the optimality of the 50/50 selection rule are not established by the analytic argument. The same gap propagates into the analytic overhead and improvement-threshold derivations, which inherit the assumption p0 ≈ p. This is a correctness risk in the central claim, not merely a scope limitation: it bites at first order under local depolarizing noise, before considering correlated, reversible, or large errors.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes RIDA (Random Inverse Depolarizing Approximation), an error mitigation method built on the global depolarizing channel model. For a target circuit with observable O, RIDA constructs a 'depolarization estimation circuit' by randomly selecting half of the target's one-qubit gates and half of its two-qubit gates, followed by the inverses of the selected gates, so that the estimation circuit implements the identity and has a known noiseless expectation <O_0> = 1. Measurement of this circuit gives p_0 = 1 - <O_0_epsilon>, an estimate of the depolarization probability, and the target expectation value is corrected via <O> approximately <O_epsilon>/(1-p_0). The authors report numerical simulations on 4- to 7-qubit EfficientSU2 circuits under an IBM-Kingston-inspired local depolarizing noise model and a twirled coherent noise model, and claim that RIDA outperforms exponential ZNE + TREX and CNOT-only depolarization + quadratic ZNE across all tested error multipliers, shot numbers, and observable expectation values, with 16- to 47-fold reductions in shot/gate overhead relative to benchmarks and an analytical overhead that is cubic/quintic better than the benchmarks. The Supplemental Material provides proofs of the optimality of the 50/50 gate-selection rule, an equivalence between RIDA and TREX for measurement error, analytic overhead scalings, and extended numerical results.","tokens_in":25079,"tokens_out":45412,"duration_ms":419690,"significance":"If the claims were fully established, RIDA would be a significant practical tool: the method is simple, requires no noise characterization, uses a reusable estimation circuit, and the numerical evidence is extensive and favorable. The authors make the code publicly available and state precise, falsifiable predictions (depolarization-probability accuracy, RMSE as a function of shot number, improvement thresholds), which are genuine strengths. However, the load-bearing theoretical claims — universality over observables, unbiasedness and optimality of the 50/50 selection, and optimal shot-overhead scaling — are derived under an idealized per-gate global depolarizing model and are not reconciled with the local depolarizing noise model actually simulated. In particular, the estimation-circuit observable is left unspecified for target observables that are not Z-type Pauli strings, which affects the validity of Eq. (4) for a large fraction of the paper's own numerical tests. These are central-claim issues, not presentation points.","major_comments":[{"comment":"The estimation circuit implements the identity on |0...0>, so the noiseless expectation <O_0> used in Eqs. (3)-(4) equals 1 only if O_0 is a Pauli string in the +1 stabilizer of |0...0>, i.e., a product of Z operators. For a weight-1 X or Y Pauli string, <O_0> = 0; under the depolarizing model Eq. (2) the noisy expectation is then identically zero, Eq. (4) gives p_0 = 1, and the correction factor 1/(1-p_0) is undefined. The manuscript never specifies how O_0 is realized for such target observables (for example, a measurement-basis rotation on the estimation circuit, or a different O_0 whose estimate is argued to be observable-independent), and this is in direct tension with the claim that RIDA estimation circuits 'involve the same measurements as the target circuit.' The numerical tests include random weight-1 Pauli strings (SM II.A), so as written either the implemented procedure differs from the described one or the method fails on two thirds of the test observables. The universality claim therefore requires a resolution, and the TREX-equivalence argument (SM I.C, Eqs. (29)-(38)), which presupposes +/-1 measurement outcomes on the identity circuit, must be reconciled with that resolution as well.","section":"Method, Eq. (4); SM I.C; SM II.A"},{"comment":"The derivation p_0 = 1 - product(1-epsilon_i)^2 approximately 2*sum(epsilon_i) treats every gate error as contributing weight 1 to the measured depolarization probability, independent of the gate's location and of the measured observable. This is self-consistent only under a per-gate global depolarizing model. Under the paper's own simulation noise model — 'local, uncorrelated depolarizing noise channels' (SM II.B) — the first-order contribution of gate i to the measured p_0 is 2*epsilon_i*c_i, where c_i depends on whether the Heisenberg-evolved observable has non-identity support on gate i's qubits at that circuit position; the target circuit's effective polarization is correspondingly p = sum(epsilon_i*d_i) with generally different weights d_i. Equations (27)-(28) equalize only the unweighted numbers of one- and two-qubit gates, so they do not imply E[p_0] = E[p]; this requires a relation between the c_i and d_i that is not shown and is false for low-weight observables. Consequently, the unbiasedness of RIDA and the optimality of the 50/50 selection rule are not established under the noise model used in the numerics, and the same assumption propagates into the overhead (SM III.C-E) and improvement-threshold (SM III.B) derivations. The authors should re-derive E[p_0] under local depolarizing noise, e.g., using the scrambling of the propagated observable in random circuits, or explicitly state the global-depolarizing-model assumption and delimit its domain of validity.","section":"SM I.A, Eqs. (1)-(2) and (22)-(28)"},{"comment":"The sampling-overhead analysis and the claim that RIDA's shot overhead 'coincides with the optimal result for an unbiased estimator' treat p as a known constant. In the method, p is replaced by p_0, which is itself measured from estimation circuits with finite shot counts (10^7 total shots distributed over 50 circuits in the numerics, SM II.A). By the delta method, Var(<O_epsilon>/(1-p_0)) is approximately Var(<O_epsilon>)/(1-p)^2 + <O_epsilon>^2*Var(p_0)/(1-p)^4; the second term, omitted from Eqs. (101)-(103), is comparable to the first when the estimation and target shot budgets are equal (as in SM II.A). The optimal-scaling claim therefore holds only in the limit of negligible estimation noise; the analysis should include this term and quantify the estimation shot budget required to reach the quoted optimum.","section":"SM III.C, Eqs. (101)-(103)"}],"minor_comments":[{"comment":"The title as displayed contains a typo ('Depolarizi ng'); please ensure the submitted file is clean.","section":"Title"},{"comment":"Please clarify the shot accounting in the comparisons: the text says estimation circuits use 10^7 total shots and target circuits use 'the same number of total shots,' but it is not explicit whether a method's total budget includes its estimation/calibration circuits; this matters for the claim of lower baseline shot overhead.","section":"SM II.A"},{"comment":"The TREX-equivalence proof is written for weight-1 Pauli strings with +/-1 outcomes; the extension to higher-weight strings and to linear combinations follows by linearity but should be stated explicitly.","section":"SM I.C"},{"comment":"The caption of Fig. 2(e) refers to a 'fixed shot number' without giving its value; please state the shot number in the caption.","section":"Fig. 2"},{"comment":"The exponential ZNE implementation relies on heuristic fallbacks (Eqs. (73)-(76)) when the three-point data are non-monotonic; because these fallbacks affect the benchmark exactly in the high-error regime where RIDA's reported advantage is largest, the authors should justify or cite them.","section":"SM II.F"}],"recommendation":"major_revision","confidential_remarks":"The strength of the paper is its extensive numerics and public code, but the relationship between the described method and the implemented estimation observable should be verified against the repository during revision; if the code measures a different observable (e.g., a Z-parity) than the target observable, that must be stated. The novelty relative to Urbanek et al. (2021) and Choi et al. (2025) is adequately acknowledged."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a paper where the numerics are more convincing than the theory. RIDA is genuinely simple: take half the one- and two-qubit gates of the target, form an identity circuit from those gates and their inverses, estimate a depolarization probability p0 from a known noiseless expectation, and divide the noisy target expectation by (1-p0). The random half-gate inverse construction is new relative to Urbanek's CNOT-only estimation, and the terminal-qubit handling plus the TREX-equivalence argument are useful details. The benchmarks are broad—qubit counts, shots, error multipliers, coherent and incoherent noise—and the code is public. I take the empirical claim seriously: for EfficientSU2 circuits with random Pauli observables under these noise models, RIDA outperforms the benchmarks.\n\nThe soft spot is the analytic claim that E[p0]=p and that the 50/50 gate-selection rule is optimal. The proof in SM Sec. I treats every gate error as independently and fully depolarizing everything, giving p0 = 1 - product(1-eps_i)^2. But the simulations use local depolarizing noise. Under local depolarizing noise, the first-order contribution of a gate error to <O0> depends on whether the error Pauli gets conjugated into an anticommuting channel before measurement. That weight c_i can be anywhere from 0 to 1 depending on gate order and on the observable O0. The target circuit's effective p has different weights d_i. Matching half the gate counts does not force E[p0]=p. The paper also never specifies O0 for an arbitrary target Pauli string: a Z-type identity-circuit measurement will not have the same depolarization footprint as an X or Y measurement on the target. This is a load-bearing gap, not a scope nitpick. It bites at first order under the paper's own noise model.\n\nWhere does that leave RIDA? The numerics still suggest it is a useful, low-overhead heuristic for the circuit families and noise models tested. But the \"universal\" framing and the cubic/quintic overhead superiority claims go beyond what is proven; the overhead derivations inherit the p0≈p assumption, so they are not guaranteed bounds. Citations look fine, there is no fitting to target data, and the public code is a real plus.\n\nWho is this for? People working on practical depolarizing-model error mitigation will want to read it, and the paper deserves a serious referee. I would send it to peer review, but the referee should insist on a precise definition of the estimation observable and a first-order analysis under local depolarizing noise—or, failing that, a revised claim that RIDA is a well-tested heuristic rather than a universal unbiased estimator.","headline":"RIDA is a simple, well-benchmarked error-mitigation heuristic whose universality claim rests on a first-order full-depolarization assumption that the paper's own local-noise model contradicts.","tokens_in":25586,"tokens_out":4801,"would_cite":false,"duration_ms":55980,"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":"The paper introduces RIDA, which estimates a circuit's global depolarization probability with a random inverse subcircuit and divides the noisy expectation value by one minus that probability.","keywords":["quantum error mitigation","depolarizing noise model","random inverse circuits","expectation value estimation","zero-noise extrapolation","readout error mitigation","NISQ","randomized benchmarking"],"falsifier":"Construct a two-qubit target whose ideal expectation is known, and engineer two gate errors that partially cancel inside the estimation circuit, for example an over-rotation on a selected gate followed by an under-rotation of equal magnitude on its inverse. Compare the RIDA estimate $\\hat p$ with the true depolarization probability obtained by fitting the noisy and ideal expectation values at high shot count; a disagreement larger than shot noise would show that the independence-and-irreversibility assumption is violated and the $1/(1-\\hat p)$ correction is biased.","tokens_in":24569,"feed_emoji":"⚛️","tokens_out":8770,"duration_ms":85339,"temperature":0.7,"pith_summary":"This paper introduces RIDA, an error mitigation method that estimates the total depolarization probability of a target quantum circuit without knowing its ideal output. The estimate comes from a single randomly built identity circuit: half of the target's one- and two-qubit gates, balanced in type, followed by their inverses, so the noise-free expectation value is known to be one. The noisy expectation of that estimation circuit directly gives the depolarization probability, and dividing the target's noisy expectation value by one minus that probability yields an approximate error-free value. Across simulated incoherent and coherent noise, on circuits of four to seven qubits over a range of error rates and shot counts, the paper reports that RIDA has lower root-mean-square error than benchmark combinations of exponential zero-noise extrapolation with readout twirling and CNOT-only depolarization with quadratic extrapolation.","feed_headline":"Random half-circuit trick strips noise from quantum results","feed_subtitle":"Run half a circuit plus its inverse, read off the depolarization rate, and divide it out of noisy expectation values.","key_machinery":"The load-bearing object is the RIDA estimation circuit: from the target circuit, randomly select exactly half of the one-qubit gates and half of the two-qubit gates, excluding gates on terminal qubits that cannot affect the measured expectation value, then append the inverse of that selected half. The whole estimation circuit is the identity in the absence of noise, so its error-free expectation value is exactly $1$; measuring it under noise gives $p_0 \\approx 1 - \\langle O_0^{\\mathrm{noisy}}\\rangle$, which under the global depolarizing approximation estimates the target's $p$. Because each gate has a 50% chance of appearing once in each half, the expected composition matches the target, and fixing the count at half minimizes estimator variance. The corrected expectation is $\\langle O \\rangle = \\langle O_{\\mathrm{noisy}}\\rangle / (1-p_0)$.","core_discovery":"RIDA's central claim is that the global depolarization probability of a target circuit can be extracted from a random identity circuit with the same gate composition, and that this estimate is accurate enough to correct expectation values by the depolarizing-model formula $\\langle O \\rangle = \\langle O_{\\mathrm{noisy}}\\rangle/(1-p)$. The paper argues that selecting exactly half of the one-qubit gates and half of the two-qubit gates minimizes the variance of the estimate and makes the average estimation-circuit depolarization probability equal to the target's under a first-order independent-error model. The method is universal in the sense that it applies to any expectation-value-estimating circuit, requires no knowledge of individual error rates, and intrinsically handles measurement error; with twirled readout its measurement-error mitigation is shown to be mathematically equivalent to TREX. In numerical tests it outperforms exponential ZNE plus TREX and CNOT-only depolarization plus quadratic ZNE across all considered circuit sizes, error multipliers, shot numbers, and both incoherent and coherent error models.","pith_inferences":["A complementary-half test could expose the method's key assumption: generating a second estimation circuit from the unselected half of the gates should give the same $p$ up to shot noise; disagreement would signal correlated or reversible errors that the first-order model ignores.","Because the estimation circuit's identity structure does not depend on rotation angles, RIDA may serve as a cheap per-layer noise probe for parameterized circuits, giving a depolarization estimate that follows the circuit's gate skeleton rather than its specific parameters; the paper's reuse of one estimation circuit for a class of circuits points in this direction but does not prove it on general","If the TREX equivalence carries over to non-depolarizing measurement noise, RIDA could replace dedicated readout calibration inside existing ZNE pipelines, lowering overhead by one calibration stage; the paper proves the equivalence only under the depolarizing model with twirled readout."],"forward_implications":["One estimation circuit, reused across a class of similar target circuits, provides both gate-error and measurement-error mitigation without a separate readout calibration step.","RIDA's sampling overhead matches the optimal unbiased-estimator scaling; under the paper's assumptions this is a cubic improvement over exponential ZNE and a quintic improvement over CNOT-only depolarization plus quadratic ZNE.","In the high-error regime where other methods break down, RIDA remains the only considered method with practically useful RMSE, and the analytic scaling indicates it needs far fewer shots than exponential ZNE to beat the unmitigated result.","Because the corrected error is low across the full range of error-free expectation values, RIDA is suited to estimating arbitrary Pauli-string observables, not only near-extremal ones."],"supporting_citations":[{"why":"Supplies the global depolarizing channel model that RIDA inverts.","marker":"[25]"},{"why":"Supplies the depolarizing-model correction formula and the CNOT-only depolarization method used as the principal baseline.","marker":"[29]"},{"why":"Supplies the TREX readout error mitigation method to which RIDA's measurement-error mitigation is shown equivalent.","marker":"[21]"},{"why":"Supplies the zero-noise extrapolation framework used as the exponential ZNE baseline.","marker":"[17]"},{"why":"Supplies related depolarization-probability estimation circuits with known error-free expectation values that RIDA generalizes.","marker":"[32]"},{"why":"Supplies the identity-circuit-plus-inverse construction used in randomized benchmarking that RIDA borrows for its estimation circuits.","marker":"[33]"},{"why":"Contains the statistical proofs that half-gate selection is optimal and that RIDA's measurement-error mitigation is equivalent to TREX.","marker":"[38]"}],"fun_headline_variants":["Random half-circuit pairs erase quantum noise","Depolarization from random halves fixes quantum results","Universal error mitigation beats ZNE benchmarks","RIDA: simple random trick outperforms noise fixes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The estimate is unbiased only if gate errors are independent, irreversible, and small enough that the estimation circuit's depolarization probability is approximated by twice the sum of the individual gate error rates; under correlated, partially reversible, or large errors, the half-and-half selection is no longer guaranteed to match the target circuit's depolarization probability.","fun_headline_variants_meta":{"raw":{"variants":["Random half-circuit pairs erase quantum noise","Depolarization from random halves fixes quantum results","Universal error mitigation beats ZNE benchmarks","RIDA: simple random trick outperforms noise fixes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000336,"raw_usage":{"total_tokens":1794,"prompt_tokens":812,"completion_tokens":982,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":428,"completion_tokens_details":{"reasoning_tokens":926}},"tokens_in":428,"tokens_out":982,"duration_ms":10254,"temperature":1.0,"reasoning_tokens":926,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:04:17.336984+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a two-qubit target whose ideal expectation is known, and engineer two gate errors that partially cancel inside the estimation circuit, for example an over-rotation on a selected gate followed by an under-rotation of equal magnitude on its inverse. Compare the RIDA estimate $\\hat p$ with the true depolarization probability obtained by fitting the noisy and ideal expectation values at high shot count; a disagreement larger than shot noise would show that the independence-and-irreversibility assumption is violated and the $1/(1-\\hat p)$ correction is biased.","supporting_citations":[{"cited_title":"The error-free expectation value for the target circuit then immediately follows from the relationship Eq","cited_arxiv_id":null,"evidence_quote":"Contains the statistical proofs that half-gate selection is optimal and that RIDA's measurement-error mitigation is equivalent to TREX."}],"review_version":2}