{"id":"7fad3fc0-ebdc-4823-8ec8-6761bae36f9f","arxiv_id":"2507.10771","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A delta-sweep convergence check plus power-law extrapolation of Pauli-term counts gives practitioners a practical way to decide whether Pauli path simulation can be trusted and how much it will cost.","lead":"Pauli path simulators approximate quantum circuits by dropping small correction terms; this paper works out when that dropping is safe and how to forecast the cost. It offers a protocol that sweeps the truncation threshold to detect apparent convergence, and a power-law model for predicting memory and runtime from cheap test runs.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Resource extrapolation is validated only in near-power-law regimes; the paper itself documents a small-angle, non-Clifford RZZ circuit family where the coefficient distribution deviates visibly from Eq. (11), yet no extrapolation test is reported there.","rationale":"The reader's weakest-assumption statement identifies the power-law form holding under eta-spikes and small-angle distortions as the key risk; my concern is the same, sharpened to the fact that the paper's own Appendix E.2 exhibits a circuit family where that assumption is known to fail, and Section 3.3 provides no out-of-sample check in that family. The extrapolation protocol is the paper's main proposed tool, so the missing validation is load-bearing. This does not overturn the reader's CONDITIONAL verdict; it reinforces the condition: accept the practical protocol only if the authors add a validation in the Appendix E.2 regime or clearly delimit the protocol's scope to near-power-law circuits. I agree rather than partially agree because the reader's weakest assumption and my concern are essentially the same failure mode, just with a concrete locus and test attached. I am not proposing REJECT because the paper is transparent about the heuristic status of its assumptions, the code is released, and the reported in-domain validations are genuine evidence. The right move is to keep CONDITIONAL and require the additional test before treating the extrapolation as generally reliable.","tokens_in":21455,"tokens_out":5529,"duration_ms":70811,"concrete_test":"Run the Appendix E.2 circuit (θZZ = -π/36, θX = π/12, T = 30) through the Section 3.3 protocol: start at δ0 = 0.005, decrease by r = 1/√2 for five coarse values, fit log(Nmax) versus log(1/δ), extrapolate Nmax to δ = 1e-5, and compare with a full PPS run at δ = 1e-5. If the relative error exceeds ~20%, the central extrapolation claim fails in exactly the small-angle regime the authors flag as deviating from the power-law model; if it stays within that bound, the concern is empirically retired for this documented failure mode.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central link is from the power-law density Eq. (11) to the extrapolation formulas Eqs. (17)-(19). Deriving Eq. (17) requires the unproven PPS hypothesis of Appendix D and replaces the finite Pauli sum by a Riemann integral over a smooth power-law density. Appendix E.2 then documents a regime, with θZZ = -π/36 and θX = π/12, where the coefficient distribution deviates strongly from Eq. (11) and only the tail beyond xmin = lδ remains power-law; Fig. 16's inset shows a single coefficient contributing roughly 47% of the norm-squared. In that regime Eq. (13) is not controlled, and Eq. (19)'s neglect of the term 2 log(||O_{k*}(δ1)||/||O_{k*}(δ2)||) can bias the log-linear slope. The two reported extrapolation validations, Fig. 5 and the 2D Ising example in Section 3.3, are cases with comparatively mild deviations; no validation is reported for the Appendix E.2 circuit family. Because the paper's practical promise is that coarse-δ runs predict Nmax and runtime at finer δ, an untested failure mode inside a documented regime is a load-bearing gap rather than a cosmetic caveat.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a practical protocol for estimating the memory and runtime requirements of Pauli Path Simulations (PPS) at fine truncation thresholds δ from brief coarse-δ runs, together with a convergence diagnostic for deciding whether PPS can be trusted. The central modeling claim is that, for a broad class of utility-scale circuits, the absolute Pauli coefficients of the evolved observable follow a truncated power law ρ(t)=m δ^m/(2|t|^{m+1}) (Eq. 11), leading to a log-linear scaling of the maximum number of Pauli terms Nmax with log(1/δ) (Eqs. 17-19). The authors validate this extrapolation on the kicked Ising circuit with random single-qubit angles (Fig. 5, within 6%) and on a 2D Ising example (predicted ≈2.1 billion vs. reported ≈2.5 billion). They then introduce an 'apparent convergence' protocol based on sweeping δ and checking whether successive estimates agree, and apply it to IBM's kicked Ising experiments, finding both convergent and non-convergent regimes. The BlueQubit SDK implementing the method is released publicly.","tokens_in":21762,"tokens_out":3062,"duration_ms":37654,"significance":"If the central claims hold, this is a genuinely useful practical contribution: it turns resource estimation for PPS from guesswork into an inexpensive extrapolation procedure and provides a principled way to think about PPS convergence in the absence of rigorous error bounds. The paper's strengths include out-of-sample validation on two distinct circuits, a concrete and falsifiable prediction for the 2D Ising example, a released software implementation, and an honest discussion of regimes where the power-law model fails. However, the analytic derivation rests on the unproven 'PPS hypothesis' of Appendix D, and the convergence diagnostic is inherently a self-consistency check rather than a comparison against ground truth. The practical value of the method is therefore conditional on the validity of the power-law assumption for the user's circuit, and the paper would be substantially strengthened by testing the extrapolation in the very regimes it identifies as deviating from the power law.","major_comments":[{"comment":"The central resource-extrapolation claim is validated only on circuits whose coefficient distributions are close to the power-law form of Eq. (11). Appendix E.2 explicitly documents a circuit family (θZZ = -π/36, θX = π/12) with strong, persistent deviations from the power law, and the text notes that the 2D Ising example from Section 3.3 resembles this distorted distribution. Yet no extrapolation test against direct Nmax or runtime values is reported for this family. Since the practical promise of Section 3.3 is that coarse-δ runs predict fine-δ resources across the circuits a user might encounter, an untested failure mode inside a documented regime is load-bearing. Please add an out-of-sample extrapolation test for the Appendix E.2 family, or provide a criterion that tells the user when Eqs. (13)-(17) are reliable.","section":"§3.3 and Appendix E.2"},{"comment":"The approximation in Eq. (19) drops the term 2 log(||O_{k*}(δ1)||/||O_{k*}(δ2)||) based on the observation in Fig. 4 that the norm varies little with δ. This justification is circuit-dependent: Fig. 16's inset shows a single coefficient contributing roughly 47% of the norm-squared, and in such cases the norm-ratio term can be large enough to bias the log-linear slope used to extract m*. The paper should either quantify the size of this term for the validated examples or provide a practical test for when the dropped term is negligible.","section":"§3.2, Eq. (19) and Fig. 16"},{"comment":"The analytic support for the power-law density and for Eq. (17) rests on the 'PPS hypothesis': that commuting and anticommuting Pauli sets share the same coefficient distribution, and that paired coefficients in P^anti and σP^anti are independent. These assumptions are stated but not proven, and no numerical test of them is reported. Since the paper presents the power-law model as a theoretical framework rather than a purely empirical fit, the authors should either validate the two parts of the hypothesis on the circuits studied, or explicitly state that Eq. (17) is an empirical model whose analytic derivation is conditional.","section":"Appendix D"},{"comment":"The 'apparent convergence' protocol is a self-consistency check that never compares against an external ground truth, and the paper itself notes that the estimates are non-monotonic in δ and that local minima can falsely indicate convergence. This is not by itself a flaw, but the paper uses the diagnostic to classify problems into trustworthy and untrustworthy categories (e.g., Fig. 8), which is a stronger claim than 'these runs were self-consistent'. The manuscript should state more clearly that apparent convergence is neither necessary nor sufficient for accuracy, and should discuss what external evidence (e.g., known exact results, hardware data, or cross-method comparison) would strengthen a verdict of convergence.","section":"§4.1 and §4.2"}],"minor_comments":[{"comment":"The abstract and the opening line of Section 1 contain the typo 'In this this paper'.","section":"Abstract and §1"},{"comment":"The notation P_{k+1} = P_k ∪ σ_{k+1}P^anti_k could be clarified: as written it suggests a union of sets of Pauli operators, but the coefficients also change according to Eq. (9). A brief sentence on this would help readers.","section":"§2"},{"comment":"The symbol ≲ is described as 'an upper bound up to a constant factor'; since Eq. (17) already contains a specific prefactor, the statement would be clearer if the constant were identified explicitly.","section":"§3.2, Eq. (20)"},{"comment":"In the sentence describing the η-spike at k=2036, the angle 'approximately 55.13°' appears inconsistent with the earlier restriction θ_j ∈ [-π/4, π/4] from Eq. (3); please clarify whether this is an angle before Clifford recompilation or a different convention.","section":"Appendix E.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a practical guide with useful heuristics and honest caveats, and the two out-of-sample validations are encouraging. The main gap is that the central extrapolation method is not tested in the very regime the paper identifies as deviating from the power-law model; this is fixable within the manuscript's scope by adding such a test or a clear applicability criterion. The paper would also benefit from a sharper separation between the empirical claims and the conditional analytic derivation in Appendix D. The topic fits the journal well, and I would be supportive after the requested additions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: this is a practical paper that delivers a real method. For someone running PPS/SPD on utility-scale circuits, the delta-sweep protocol gives a cheap way to predict Nmax and runtime at fine delta from coarse runs, and the 'apparent convergence' diagnostic gives a way to decide whether PPS can be trusted without an external ground truth. The two counterintuitive findings--smaller delta not always improving accuracy, deeper circuits sometimes being easier--come out of numerical evidence, not speculation.\n\nWhat's genuinely new: the log-log linear extrapolation of Nmax versus 1/delta, validated out-of-sample on the kicked Ising circuit (within 6%) and on the 2D Ising circuit from Begusic-Chan (estimated ~2.1B vs reported ~2.5B). The BlueQubit SDK is released and there's a working code snippet in Appendix A. The paper is also honest about what it does not know: the convergence check is a self-consistency test, not an error bound, and they say so.\n\nWhere the soft spots are: the analytic model in Appendix D depends on an unproven 'PPS hypothesis'--equal distributions for commuting and anticommuting sets plus independence of paired coefficients. The paper labels it a hypothesis, but Eq. (17) is built on it. Second, m-estimation is delicate; a 5% error in m gives roughly 50% error in Nmax via Eq. (17), though the bootstrap extrapolation in Section 3.3 avoids direct m-estimation by regressing on curve spacing. The stress-test note's worry about the small-angle RZZ family (Appendix E.2) is partially answered: the 2D Ising test is explicitly a distorted-distribution case resembling E.2 and it still gives a ballpark estimate. What's genuinely missing is an extrapolation test on the exact E.2 circuit family (theta_ZZ = -pi/36, theta_X = pi/12) with the true Nmax for comparison. That's the clean way to close the gap, and its absence is a real omission, not a cosmetic caveat.\n\nWho this is for: anyone benchmarking PPS or deciding whether PPS is worth running on a specific circuit. It deserves a serious referee. I'd ask for the E.2 extrapolation test, seed/instance variability, and some discussion of when the self-consistency check can be fooled, but the core method and the honest validation are solid.","headline":"A practical, honestly hedged resource-estimation and convergence-check protocol for PPS, with real out-of-sample validations; the main gap is an untested extrapolation on the paper's own documented small-angle distortion family.","tokens_in":22269,"tokens_out":3990,"would_cite":true,"duration_ms":41934,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.-a","03.67.Lx"],"model":"deepseek-v4-flash","headline":"Pauli Path Simulations obey a power law in their coefficient counts, so memory needs can be extrapolated from short coarse runs.","keywords":["Pauli path simulation","Sparse Pauli dynamics","Heisenberg picture","power-law distribution","resource estimation","truncation threshold","classical simulability","quantum simulation"],"falsifier":"Take a circuit whose coefficient distribution is visibly distorted, as in Fig. 15 (small correlated angles, non-Clifford RZZ), run the coarse-to-fine $\\delta$ sweep, and compare the extrapolated $N_{\\max}$ at $\\delta=5\\times 10^{-5}$ with the value from a full simulation. If the log-log plot of $N_{\\max}$ versus $1/\\delta$ is not approximately linear, or if the prediction error is much larger than the few percent seen in the clean power-law case, the central extrapolation claim fails for that family. A second falsifier: if sweeping $\\delta$ over a range of $\\varepsilon_{\\mathrm{tol}}$ and ratios $r$ never yields $\\ell$ successive values within tolerance on a circuit that the paper's protocol labels as apparently convergent, the convergence diagnostic is not robust.","tokens_in":21227,"feed_emoji":"⚛️","tokens_out":9425,"duration_ms":112638,"temperature":0.7,"pith_summary":"Pauli Path Simulation (PPS) is a classical method that works in the Heisenberg picture: it expands the evolved observable as a sum of Pauli strings and keeps only coefficients above a threshold $\\delta$. The paper's central claim is that for a broad class of utility-scale circuits the distribution of the kept coefficients settles into a truncated power law, $\\rho(t)=m\\delta^m/(2|t|^{m+1})$, and this makes the maximum number of tracked Pauli terms grow roughly as $\\delta^{-m}$. As a result, $\\log N_{\\max}$ is approximately linear in $\\log(1/\\delta)$, so memory and runtime at fine $\\delta$ can be extrapolated from a few short coarse-$\\delta$ runs. The paper also proposes a practical diagnostic: sweep $\\delta$ and check whether expectation values appear to converge; this splits problems into those where PPS can be trusted as a verification tool and those where it only gives a fluctuating Monte-Carlo-like estimate. Applied to the 127-qubit kicked Ising circuits, the protocol shows both regimes and exposes two counterintuitive facts: smaller $\\delta$ does not always improve accuracy, and deeper circuits can be easier to simulate than shallower ones.","feed_headline":"Power-law scaling makes Pauli simulator costs predictable","feed_subtitle":"Short coarse runs extrapolate to fine truncations; a δ-sweep tells when the method can be trusted.","key_machinery":"The load-bearing object is the distribution $\\rho(t)$ of the absolute values of the Pauli coefficients of the evolved observable, together with the truncation threshold $\\delta$. The key identity is the truncated power law $\\rho(t)=m\\delta^m/(2|t|^{m+1})$, whose slowly varying exponent $m$ controls how fast the number of Pauli terms grows as $\\delta$ shrinks. The analytic support for the power law (Appendix D) invokes what the paper calls the PPS hypothesis: commuting and anticommuting Pauli sets share the same coefficient distribution, and the paired coefficients $c_P$ and $c_{\\sigma P}$ are independent. The argument then converts the density into a resource estimate through a Riemann-sum approximation of the observable's second moment, leading to Eq. (17) and the log-linear relation Eq. (19). On the convergence side, the mechanism is a $\\delta$-sweep protocol: compute expectation values at $\\delta_n = r^n \\delta_0$ and declare apparent convergence when $\\ell$ successive values agree within $\\varepsilon_{\\mathrm{tol}}$.","core_discovery":"The discovery is that the apparently chaotic branching-and-merging of Pauli terms in PPS is governed by a simple statistical regularity. After enough gates, the absolute values of the coefficients of the evolved observable follow a truncated power law, $\\rho(t)=m\\delta^m/(2|t|^{m+1})$ for $|t|>\\delta$ (Eq. 11), with an exponent $m$ that drifts slowly through the circuit. Feeding this density into a Riemann-sum estimate of the second moment yields $N_{\\max} \\approx \\frac{2-m_*}{m_*} \\frac{\\lVert O_{k_*} \\rVert^2}{\\delta^{m_*}}(1-\\delta^{2-m_*})$ (Eq. 17), and therefore $\\log N_{\\max}$ grows approximately linearly in $\\log(1/\\delta)$ (Eq. 19). The paper claims this is what makes resource extrapolation work: a handful of coarse-$\\delta$ runs, with $\\delta$ reduced by a constant ratio, give growth curves whose regular vertical spacing predicts $N_{\\max}$ at much smaller $\\delta$ to within a few percent. On the reliability side, the paper claims that sweeping $\\delta$ and testing for apparent convergence separates PPS problems into two classes, and that some problems whose PPS results match the published hardware experiment do not actually pass the convergence test, meaning PPS there is a Monte-Carlo-like estimate rather than a verified prediction.","pith_inferences":["One could adapt the same coarse-to-fine extrapolation to other truncation-based classical methods, such as tensor-network bond dimension or sparse-state truncation, by treating the truncation parameter as $\\delta$.","If the power-law regularity holds broadly, the PPS hypothesis could be tested directly: measure the commuting/anticommuting coefficient distributions and the $c_P/c_{\\sigma P}$ correlation on small circuits and compare with the independence assumption, giving a rigorous validity condition for the extrapolation.","The observed link between non-Cliffordness and convergence difficulty raises the possibility that the exponent $m$ or the frequency of $\\eta$-spikes tracks a magic monotone, turning resource estimation into a function of the circuit's magic content.","One testable extension: for a fixed architecture, map the depth at which a circuit crosses from non-convergent to apparently convergent, which would give a practical classical-simulability frontier for that architecture."],"forward_implications":["Memory and runtime for a fine-$\\delta$ PPS run can be estimated from a few minutes of coarse-$\\delta$ runs, without committing to the expensive simulation.","Practitioners can decide before a run whether the extrapolated $N_{\\max}$ fits their available memory and time budget.","The $\\delta$-sweep gives a no-ground-truth test of PPS reliability, flagging problems where PPS apparently converges and problems where it is only a Monte-Carlo-like estimate.","Because smaller $\\delta$ does not monotonically improve accuracy, the cheapest $\\delta$ that gives the desired tolerance is the right choice, not the smallest feasible one.","Convergence difficulty does not transfer across circuit depths: a deeper version of the same circuit can converge with fewer resources, so each circuit must be probed independently."],"supporting_citations":[{"why":"introduced the PPS reproduction of kicked Ising expectation values that this paper re-examines with its convergence protocol.","marker":"[BC23]"},{"why":"supplies the original fast and converged PPS algorithm and the bit-array representation of Pauli sums that the resource model builds on.","marker":"[BGC23]"},{"why":"provides the sparse Pauli dynamics method and the 2D Ising test case used to validate the Nmax extrapolation.","marker":"[BcvacC25]"},{"why":"supplies the hardware expectation values from the 127-qubit kicked Ising experiment that the convergence study compares against.","marker":"[K+23]"},{"why":"provides the Clifford recompilation that restricts gate angles to [-pi/4, pi/4], a prerequisite for the coefficient dynamics analyzed.","marker":"[QWE19]"},{"why":"gives the Pauli propagation framework and magic-resource context that the convergence discussion uses for framing difficulties.","marker":"[RJT+25]"},{"why":"connects magic in the Heisenberg picture to PPS difficulty, supporting the explanation of why some fixed-angle circuits fail to converge.","marker":"[DKT24]"}],"fun_headline_variants":["Pauli simulator costs extrapolate via power-law coefficient decay","δ-sweep classifies Pauli simulators into convergent and guesswork","Truncated power law predicts Pauli simulator resource needs","δ-sweep separates reliable Pauli predictions from estimates","Pauli coefficient power law enables resource cost prediction"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The protocol's predictions rest on the assumption that a single truncated power law with a slowly varying exponent $m$ describes the coefficient distribution of the circuit being simulated; when $\\eta$-spikes or small-angle rotations distort that distribution (Appendices E and F), the extrapolated $N_{\\max}$ and the apparent-convergence verdict can be wrong.","fun_headline_variants_meta":{"raw":{"variants":["Pauli simulator costs extrapolate via power-law coefficient decay","δ-sweep classifies Pauli simulators into convergent and guesswork","Truncated power law predicts Pauli simulator resource needs","δ-sweep separates reliable Pauli predictions from estimates","Pauli coefficient power law enables resource cost prediction"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001355,"raw_usage":{"total_tokens":5596,"prompt_tokens":1136,"completion_tokens":4460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":752,"completion_tokens_details":{"reasoning_tokens":4377}},"tokens_in":752,"tokens_out":4460,"duration_ms":36712,"temperature":1.0,"reasoning_tokens":4377,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:26:17.062460+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a circuit whose coefficient distribution is visibly distorted, as in Fig. 15 (small correlated angles, non-Clifford RZZ), run the coarse-to-fine $\\delta$ sweep, and compare the extrapolated $N_{\\max}$ at $\\delta=5\\times 10^{-5}$ with the value from a full simulation. If the log-log plot of $N_{\\max}$ versus $1/\\delta$ is not approximately linear, or if the prediction error is much larger than the few percent seen in the clean power-law case, the central extrapolation claim fails for that family. A second falsifier: if sweeping $\\delta$ over a range of $\\varepsilon_{\\mathrm{tol}}$ and ratios $r$ never yields $\\ell$ successive values within tolerance on a circuit that the paper's protocol labels as apparently convergent, the convergence diagnostic is not robust.","supporting_citations":[],"review_version":1}