{"id":"1f54d68e-f960-4de3-9615-efc0311eac03","arxiv_id":"2507.23314","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A block-level fidelity characterization method that corrects the success probability of structured quantum algorithms, outperforming standard ZNE in high-noise Grover search.","lead":"This paper proposes a low-cost error mitigation method for quantum algorithms made of repeated blocks: it measures the error of a single block on a shallow circuit, then corrects the full algorithm's results. The approach beats standard zero-noise extrapolation in noisy Grover search simulations and on a 127-qubit IBM processor, but rests on an unproven assumption that the block's error rate transfers to the whole algorithm.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. 16 assumes the identity-block fidelity measured from |0...0⟩ transfers to the algorithm's superposition state under repeated blocks; this is unproven and fails under non-depolarizing noise, so Eq. 17 is not established.","rationale":"The paper's central claim is that block-level fidelity extracted from shallow identity circuits can be used to correct the full structured algorithm's success probability. The reader identified the transferability of the per-block fidelity as the weakest assumption, and I agree. The paper is internally consistent in the depolarizing-noise regime: the root-extraction estimates of f (0.8689, 0.8690, 0.871) agree with the log-linear fit, and the reported Aer results do show that the identity-block data follows the assumed exponential decay. That is real but narrow evidence. It does not validate Eq. 16 because depolarizing noise is the most favorable case for state-independent fidelity; realistic noise channels (amplitude damping, coherent errors, crosstalk) introduce state dependence that the identity-block measurement, which uses only |0...0⟩, cannot capture. The hardware results contain additional inconsistencies (unmitigated 4-qubit value is given as 10.9% and 11.7% in different places; Table III has an internally inconsistent layout), and no error bars or raw shot counts are provided, so the hardware evidence cannot rescue the transferability claim. A targeted simulation with a non-depolarizing noise model would settle the concern directly. If the test passes, the method still needs the data-exclusion rule to be made principled rather than post hoc; if it fails, the method as stated overreaches. The reader's CONDITIONAL verdict is therefore appropriate, and no verdict change is needed.","tokens_in":8961,"tokens_out":4065,"duration_ms":52302,"concrete_test":"Rerun the 6-qubit Grover Aer experiment under an amplitude-damping noise model with single- and two-qubit error rates chosen so average gate fidelities match the depolarizing case (single-qubit fidelity 1−10^-4, two-qubit 1−10^-3). Extract f and c from the U†U identity block at k=1,2,3, then compute Pmit = Pr,raw/(c f^6). If Pmit deviates from the known ideal value 0.997 by more than 5 percentage points, Eq. 17 fails for non-depolarizing noise and the transferability assumption is refuted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Eq. 16, Pr,raw ≈ Pideal × c f^r, which connects the identity-block measurement to the mitigated success probability. This equation is asserted without derivation. The identity-block return probability FI(2k) is measured by initializing in |0...0⟩, applying (U†U)^k, and measuring the probability of returning to |0...0⟩. It therefore characterizes the noise channel's effect on that one computational basis state under the specific U†U sequence. The algorithm's success probability, by contrast, depends on the overlap of the noisy evolved state with the target state after r iterations of Uiter starting from H⊗n|0...0⟩. For any noise that is not global depolarizing, this per-block fidelity is state-dependent: local depolarizing, amplitude damping, and coherent errors all act differently on superposition states than on basis states, and U†U can partially cancel coherent errors that accumulate under repeated Uiter. The Aer validation uses only a depolarizing model, which is exactly the regime where state-dependence is minimized, so it cannot establish Eq. 16. A secondary but compounding problem is the post hoc exclusion of FI(2k) points that are 'indistinguishable from the uninformative baseline'; this rule lets the authors discard points that would spoil the exponential fit (e.g., the 6-qubit 0.005 row in Table IV fits f using only k=1,2 after dropping F(6)). The central claim therefore rests on an unverified transferability assumption that is likely to fail in realistic non-depolarizing noise.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an error-mitigation method for quantum algorithms with a repetitive block structure, specifically Grover search. The core idea is to estimate the per-block fidelity f and a state-preparation/measurement constant c by measuring return probabilities of the identity block (U_iter^dag U_iter)^k from the all-zero state, fitting an exponential decay model, and then dividing the raw success probability of the full algorithm by c f^r (Eqs. 16-17). The method is validated on IBM Aer simulations of 6-qubit Grover search under depolarizing noise and on IBM Eagle r3 hardware for 3- and 4-qubit Grover search, with reported success probabilities close to theoretical values where standard ZNE fails.","tokens_in":9298,"tokens_out":6829,"duration_ms":79674,"significance":"If the central assumption holds, the method is a low-overhead alternative to ZNE for structured algorithms and the reported gains are substantial: in the 6-qubit Aer simulation the proposed method reaches 96.86% versus 68.37% for ZNE, and in the 4-qubit hardware run it reports about 90.7% versus about 11.7% for ZNE. The paper also provides a useful concrete demonstration that per-block fidelity extracted from identity circuits can be exponentially consistent under a depolarizing model, with root-extracted f values of 0.8689, 0.8690, and 0.871 for k=1,2,3 in the 6-qubit simulation. No code or data release is mentioned, which limits independent verification of the reported numbers.","major_comments":[{"comment":"The load-bearing relation Pr,raw ≈ Pideal × c f^r is asserted without derivation. The identity-block return probability FI(2k) is measured on the initial computational basis state |0...0> under (U_iter^dag U_iter)^k, whereas Pr,raw is the probability of finding the target state after r forward applications of U_iter starting from H^⊗n|0...0>. These are different observables, and equating their decay factors assumes a state-independent per-block noise channel. This holds for global depolarizing noise but fails for local depolarizing, amplitude damping, and coherent errors, where the noise channel acts differently on superposition states and where U_iter^dag U_iter can partially cancel coherent errors. Since Eq. (17) divides by f^r, any mismatch is multiplicatively magnified. The Aer validation uses only a per-gate depolarizing model, which is precisely the regime where the assumption is least stressed; the hardware runs do not independently verify the relation. I recommend adding a derivation under explicit noise assumptions and validating the transferability on non-depolarizing noise models (e.g., amplitude damping, coherent rotation errors) or on randomized circuits with known ideal probabilities.","section":"Sec. III, Eqs. (16)-(17)"},{"comment":"The rule to discard FI(2k) points that are 'indistinguishable from the uninformative baseline' is not quantitatively defined: no threshold, statistical test, or error bar is specified. In the high-noise 6-qubit case (error rate 0.005), FI(6) is excluded, leaving only two k values and reducing the fit to the two-point formula Eq. (12). This means the claim of a 'highly consistent exponential decay' is not validated in exactly the high-noise regime where the method is claimed to outperform ZNE. Because the exclusion is applied after inspecting the data, it can also artificially enforce the exponential model. Please specify a prespecified criterion for baseline exclusion and report the fits with and without the excluded points.","section":"Sec. III, paragraph after Eq. (11); Sec. IV-C, Table IV"},{"comment":"As I read Table IV, the reported Pmit values for the 5-qubit error-rate-0.005 row (0.93) and the 6-qubit error-rate-0.005 row (0.299) are not reproduced by Eq. (17) using the reported f and the c implied by F(2) and F(4). For the 5-qubit row, Eq. (17) gives a value above 1 before clipping, not 0.93; for the 6-qubit row, with f=0.572 and c≈F(2)/f^2≈0.80, the denominator c f^6 is roughly 0.03, which would give Pmit far above 1, not 0.299. If my column parsing is incorrect, the table headers should be revised to make the grouping unambiguous; if the reported Pmit was obtained by a different formula, that formula should be given explicitly. This arithmetic discrepancy undermines confidence in the quantitative claims and should be resolved with per-run data.","section":"Table IV and Eq. (17)"}],"minor_comments":[{"comment":"'in contract' should be 'in contrast'.","section":"Fig. 2 caption"},{"comment":"The placeholder 'Fig.X' and the inconsistent table numbering (Table II vs. Table III in the hardware section) should be fixed before publication.","section":"Sec. IV-C and captions"},{"comment":"'Aersimulator' should be written as 'Aer simulator' or 'IBM Aer simulator' consistently.","section":"Throughout"},{"comment":"The text states that ZNE for the 4-qubit hardware run reaches 12.7%, but Table III reports Pzne=0.117 (11.7%); these numbers should be reconciled.","section":"Sec. IV-B"},{"comment":"The two equations are redundant as written; consider stating that FI(2k)=c f^{2k} and F(k)=c f^k once, and explaining why c is assumed to be 1 in the Aer simulations while it is fitted on hardware.","section":"Sec. III, Eqs. (10)-(11)"}],"recommendation":"major_revision","confidential_remarks":"The central idea is timely and the reported gains are striking, but the paper's main formal step, Eq. (16), is an unproven transferability assumption, and the numerical validation contains apparent inconsistencies in Table IV. If the authors can supply a derivation or a clear noise-model justification for Eq. (16), add non-depolarizing validation, and correct the arithmetic, the paper could become a solid contribution. I would also encourage the authors to release the simulation and hardware-analysis scripts and per-run data, since the reproducibility of the quantitative claims is otherwise difficult to assess."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core idea here is real and worth taking seriously. Rather than extrapolating the full circuit, the authors measure the fidelity of a single repeated block by constructing U^dag U and fitting an exponential decay, then divide that block fidelity out of the raw success probability. That combination is not in the cited ZNE, layerwise Richardson, or identity-insertion papers. The Aer simulation shows a nice internal consistency: the root-extracted block fidelities from FI(2), FI(4), and FI(6) agree to three decimal places (0.8689, 0.8690, 0.871), and the method beats ZNE in the high-noise regime. That is a useful result for structured iterative algorithms like Grover.\n\nThe soft spots are concentrated in one load-bearing place: Eq. (16), Pr,raw ~ Pideal * c * f^r. This is asserted, not derived, and it assumes the fidelity measured by returning to |0...0> under (U^dag U)^k transfers to the algorithm's superposition state after repeated Uiter blocks. That is plausible only under global depolarizing noise. Local depolarizing, amplitude damping, and coherent errors are state-dependent, and U^dag U can partially cancel coherent errors that accumulate under Uiter. The Aer validation uses only depolarizing noise, which is exactly the regime where the assumption is least tested. The stress-test note is right: this transferability is unproven and likely to fail in realistic noise.\n\nThere are also smaller problems. The post hoc exclusion of FI(2k) points that are 'indistinguishable from the uninformative baseline' is never quantified; it is a curve-fitting escape hatch. The hardware tables are internally inconsistent: the 4-qubit unmitigated probability appears as 10.9% in the text, 11.7% in the table and figure caption, and ZNE is reported as both 12.7% and 11.7% in the same discussion. No code, data, or error bars are provided, which is disappointing for a paper whose whole argument is empirical. The '20% higher success probability' in the abstract is not actually shown in the tables.\n\nStill, the central mechanism is clever, the exponential-decay consistency is real, and the method is clearly better than conventional ZNE in the high-noise depolarizing cases. The paper is not confused or incoherent; it is a promising technique with an under-supported theoretical bridge. Who is this for? People working on error mitigation for structured circuits, especially iterative algorithms. They would get a clear, lightweight recipe and a cautionary tale about extrapolating from depolarizing simulations.\n\nRecommendation: send it to peer review, but the referee should demand a derivation or, at minimum, a numerical test of Eq. (16) under non-depolarizing noise (amplitude damping, coherent Z errors). The authors should also release the simulation code and raw data. With that, the paper could be solid; without it, the headline claims are not yet earned.","headline":"Block-fidelity mitigation is a genuinely new idea and the Aer numbers hang together, but Eq. (16) is asserted without derivation and the evidence is mostly depolarizing simulation, so the headline claims go beyond what is supported.","tokens_in":884,"tokens_out":975,"would_cite":false,"duration_ms":28683,"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":"Structured quantum algorithms can be error-mitigated by fitting the fidelity of the repeated block and dividing out the accumulated decay.","keywords":["quantum error mitigation","zero-noise extrapolation","structured quantum algorithms","Grover's algorithm","exponential decay model","block fidelity","success probability reconstruction","return probability"],"falsifier":"Simulate a 4-qubit Grover search under amplitude-damping or coherent-overrotation noise, extract $f$ from the $U^{\\dagger}U$ return probabilities, apply $P_{r,\\mathrm{raw}}/(c f^r)$, and compare with the exactly known ideal success probability of 96.1% for the noiseless circuit. If the corrected value disagrees by more than the statistical error bars, the identity-block transfer assumption is false.","tokens_in":8726,"feed_emoji":"🎯","tokens_out":8783,"duration_ms":87441,"temperature":0.7,"pith_summary":"This paper is trying to establish that a quantum algorithm built from one core operational block repeated $r$ times can be error-mitigated far more cheaply than by noise-scaling the whole circuit. The proposed route is to measure the shallow identity block $U_{\\mathrm{iter}}^{\\dagger}U_{\\mathrm{iter}}$ for several repetition counts, fit the return probability $F_I(2k)=c f^{2k}$, and then reconstruct the noiseless success probability as $P_{\\mathrm{mit}}\\approx P_{r,\\mathrm{raw}}/(c f^r)$. The payoff would be a mitigation method that keeps working when standard Zero-Noise Extrapolation breaks down because its high-noise data points collapse to random guessing. In the paper's tests, the method recovers 96.9% success probability on a simulated 6-qubit Grover search where ZNE reaches 68.4%, and 90.7% on a real 4-qubit hardware run where ZNE reaches only 11.7%.","feed_headline":"Per-block fidelity fit restores Grover success to 96.9%","feed_subtitle":"A shallow identity-block measurement extracts per-repetition error, so mitigation survives when ZNE data hits random guessing.","key_machinery":"The load-bearing object is the identity block $U_I=U_{\\mathrm{iter}}^{\\dagger}U_{\\mathrm{iter}}$, assembled from one repetition of the algorithm's core block followed by its inverse. Measuring the probability of returning to the initial state after $k$ such identity blocks gives $F_I(2k)=c f^{2k}$; fitting an exponential to those shallow measurements yields the per-repetition fidelity $f$ and the state-preparation-and-measurement factor $c$. That single parameter $f$ is the mechanism that carries the argument, because it converts the full circuit's deep error accumulation into a per-block decay that can be measured almost independently of the algorithm's target state.","core_discovery":"The central claim is Equation (17): for a structured circuit whose core block is repeated $r$ times, the raw success probability follows $P_{r,\\mathrm{raw}}\\approx P_{\\mathrm{ideal}}\\times c f^r$, with $f$ the fidelity retained by one application of the core block and $c$ the initial-state (state-preparation and measurement) fidelity. The paper proposes to extract $f$ and $c$ from shallow identity-block circuits $U_I=(U_{\\mathrm{iter}}^{\\dagger}U_{\\mathrm{iter}})^k$, whose return probability to the all-zero state is fit by $F_I(2k)=c f^{2k}$, and then to divide the raw success probability by $c f^r$. It reports that the fit is highly consistent in simulations and that this reconstruction approaches the theoretical success probability even where standard ZNE fails.","pith_inferences":["Editorial inference: if the per-block fidelity transfer holds, the same construction should apply to other repeating-block algorithms such as QAOA layers or Trotterized time evolution, where the same identity-block return-probability fit could be run.","Editorial inference: the ratio $F_I(4)/F_I(2)=f^2$ means two shallow measurements already determine $f$; the paper's three-point log-linear fit is a consistency check, and a direct test would compare both estimators on hardware.","Editorial inference: the transfer assumption is most vulnerable under non-depolarizing noise, since the identity-block return probability samples the noise channel on computational-basis states rather than on the intermediate superpositions that carry the algorithm's amplitude amplification."],"forward_implications":["For any algorithm with a clear repeated block, error mitigation overhead becomes proportional to the block depth rather than the full circuit depth.","In low-noise conditions the corrected success probability approaches the theoretical value and beats standard ZNE.","In high-noise conditions, where ZNE's extrapolation points fall to the random-guessing level, the block-fidelity correction still yields substantially higher success probability.","Because $f$ is extracted once from short circuits, the same fit can be reused for different numbers of repetitions $r$, not only for the one $r$ that was run.","The fitted $c$ also accounts for state-preparation and measurement errors, so the method does not need a separate SPAM calibration circuit."],"supporting_citations":[{"why":"Supplies the zero-noise extrapolation with identity insertions that the method's block-level identity construction extends.","marker":"[7]"},{"why":"Defines digital zero-noise extrapolation via unitary folding, the baseline that the paper compares against.","marker":"[8]"},{"why":"Provides Grover's search algorithm, the structured repeated-block algorithm used for validation.","marker":"[13]"},{"why":"Supplies the unitary-folding and exponential decay picture that motivates the per-block fidelity model.","marker":"[14]"},{"why":"Supplies the log-linear and exponential extrapolation techniques used to fit the block fidelity parameters.","marker":"[16]"},{"why":"Prior circuit-structure-aware mitigation approach (layerwise Richardson extrapolation) that this method builds on by exploiting repeated blocks.","marker":"[11]"}],"fun_headline_variants":["Per-block fidelity fit restores Grover to 96.9%","Repetitive structure beats ZNE under high noise","Shallow identity blocks predict full-circuit success","Divide and conquer quantum errors: per-block mitigation","ZNE fails, block extrapolation wins on 127-qubit Grover"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the per-block error rate measured by running the block forward and backward from the all-zero state is the same error rate that eats into the algorithm's success probability on every repetition, regardless of what the algorithm's qubits are actually doing.","fun_headline_variants_meta":{"raw":{"variants":["Per-block fidelity fit restores Grover to 96.9%","Repetitive structure beats ZNE under high noise","Shallow identity blocks predict full-circuit success","Divide and conquer quantum errors: per-block mitigation","ZNE fails, block extrapolation wins on 127-qubit Grover"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000253,"raw_usage":{"total_tokens":1606,"prompt_tokens":1029,"completion_tokens":577,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":645,"completion_tokens_details":{"reasoning_tokens":495}},"tokens_in":645,"tokens_out":577,"duration_ms":7757,"temperature":1.0,"reasoning_tokens":495,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:50:58.739756+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a 4-qubit Grover search under amplitude-damping or coherent-overrotation noise, extract $f$ from the $U^{\\dagger}U$ return probabilities, apply $P_{r,\\mathrm{raw}}/(c f^r)$, and compare with the exactly known ideal success probability of 96.1% for the noiseless circuit. If the corrected value disagrees by more than the statistical error bars, the identity-block transfer assumption is false.","supporting_citations":[{"cited_title":"Zero-noise extrapolation for quantum-gate error mit- igation with identity insertions,","cited_arxiv_id":null,"evidence_quote":"Supplies the zero-noise extrapolation with identity insertions that the method's block-level identity construction extends."},{"cited_title":"Digital zero noise extrapolation for quan- tum error mitigation,","cited_arxiv_id":null,"evidence_quote":"Defines digital zero-noise extrapolation via unitary folding, the baseline that the paper compares against."},{"cited_title":"A fast quantum mechanical algorithm for database search,","cited_arxiv_id":null,"evidence_quote":"Provides Grover's search algorithm, the structured repeated-block algorithm used for validation."},{"cited_title":"Measuring the capabilities of quan- 8 tum computers,","cited_arxiv_id":null,"evidence_quote":"Supplies the unitary-folding and exponential decay picture that motivates the per-block fidelity model."},{"cited_title":"Sidi, Practical extrapolation methods: Theory and ap- plications, vol","cited_arxiv_id":null,"evidence_quote":"Supplies the log-linear and exponential extrapolation techniques used to fit the block fidelity parameters."},{"cited_title":"Quantum error mitigation by layerwise richardson extrapolation,","cited_arxiv_id":null,"evidence_quote":"Prior circuit-structure-aware mitigation approach (layerwise Richardson extrapolation) that this method builds on by exploiting repeated blocks."}],"review_version":1}