Pith. sign in

REVIEW 2 major objections 6 minor 25 references

MaestroCut claims that online feedback from device drift can contract stitched variance, meet latency and reliability targets, and encrypt fragments at about 1% overhead.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

A closed-loop circuit-cutting framework that adapts partitioning, shot allocation, and estimator choice to live noise drift reports variance contraction and about 1% confidentiality overhead in simulation and emulation.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Closed-loop cutting framework is a real systems contribution, but the IND-CFA security bound does not follow from the phase-only mask actually used; deserves peer review with a mandatory security fix. the 2 major comments →

arxiv 2509.00811 v1 pith:OFIOS6VM submitted 2025-08-31 cs.CR

MAESTROCUT: Dynamic, Noise-Adaptive, and Secure Quantum Circuit Cutting on Near-Term Hardware

Samuel Punch , Krishnendu Guha This is my paper
classification cs.CR
keywords quantum circuit cuttingdrift-aware shot allocationKalman filter variance trackingtopology-aware Gaussian processestimator cascadequantum one-time padIND-CFANISQ reliability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Circuit cutting splits a quantum circuit too large for one chip into fragments, runs them, and stitches the results classically. MaestroCut is an attempt to make that loop work on real, drifting hardware by closing it: a monitor tracks per-fragment noise variance, repartitions when drift is detected, reallocates the shot budget using a topology-aware water-filling rule, picks between two reconstruction estimators on the fly, and encrypts fragments before they leave the client. The paper claims this reduces stitched variance relative to uniform and proportional shot allocation, suppresses heavy shot tails, keeps latency and reliability within service targets, and adds confidentiality at roughly 1% runtime overhead. If true, adaptive cutting would be deployable on near-term cloud hardware without sacrificing accuracy or privacy. The evidence is Tier-1 simulation and Tier-2 emulation; hardware runs are not yet complete.

Core claim

On its own terms, the paper's central claim is that the four problems blocking practical circuit cutting—sampling overhead, time-varying noise, topology-blind allocation, and fragment exposure—can be solved together rather than separately. The system tracks a per-fragment variance proxy with a Kalman filter, uses a Gaussian-process prior over the heavy-hex device graph to correlate fragments that are close to one another, and allocates shots by solving a spectral relaxation of a stitched-variance bound. The resulting allocation is water-filling: fragment i receives S u_i^(2/3) / sum_j u_j^(2/3) shots, where u_i is the tracked uncertainty. A CUSUM trigger repartitions the circuit when drift a

What carries the argument

The load-bearing object is a closed-form water-filling shot allocation. The paper upper-bounds stitched variance by lambda_max(Sigma-tilde) sum_i u_i^2 / s_i^2, where s_i are per-fragment shots, u_i = sigma-hat_i sqrt(log(N/rho)) is a tracked standard deviation times a ShotQC confidence tail factor, and Sigma-tilde is the Matérn-1/2 covariance over fragment anchors, shrunk by Kalman filter uncertainty. Minimizing that upper bound under a fixed total shot budget gives the KKT-optimal rule s_i^* = S u_i^(2/3) / sum_j u_j^(2/3). This rule converts live drift observations into a concrete, cheap-to-recompute allocation: every B shots the Kalman updates change u_i, the water-filling recomputes, an

Load-bearing premise

All the contraction and reallocation results rest on the assumption that fragment variances drift as a Gaussian random walk and correlate across fragments through an exponentially decaying spatial kernel; if real hardware drift has heavy tails, sudden breaks, or non-spatial correlations, the Kalman tracker, water-filling allocation, and CUSUM triggers may not transfer.

What would settle it

Run the full loop on a real multi-tenant quantum cloud device for several days, recording per-fragment variance observations. Then fit two models to those observations: the paper's Gaussian-drift plus exponentially decaying spatial prior, and a heavy-tailed or breakpoint alternative. If the alternative fits substantially better and water-filling on the observed variances does not reduce stitched variance against uniform or proportional allocation, the central claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • With an online variance tracker, a circuit cut once and left unchanged no longer has to be trusted: when drift crosses a threshold, the partition can be refined and shots redistributed, so accuracy degrades gracefully instead of silently.
  • The water-filling rule tells an operator exactly how to redistribute a fixed shot budget across fragments: more shots go to fragments whose tracked uncertainty is high, and the resulting stitched-variance contraction is provable under the model.
  • The entropy-gated cascade gives a per-fragment choice between cheap classical shadows and expensive maximum-likelihood reconstruction, so the same pipeline can handle low- and high-entropy fragments without overspending.
  • Confidentiality through Pauli padding and AEAD can be added to a cutting pipeline at about 1% runtime cost, making it feasible to run sensitive workloads on semi-honest cloud providers.
  • Budget enforcement with abort and audit means a user can state a variance or leakage cap and get an immutable record if the system cannot meet it.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The water-filling step is not tied to quantum circuits: any distributed estimation task with per-worker variances that drift could use the same Kalman-plus-water-filling allocation to contract total error under a fixed sampling budget. The paper does not explore this transfer.
  • The stated security guarantee is conditional on the semi-honest provider model and on AEAD and QOTP being instantiated correctly; a malicious provider who can coherently tamper with fragments before decoy checks would need verification beyond decoys, which the paper leaves to future work.
  • The empirical alpha and beta fits in the estimator cascade are workload-dependent; on unfamiliar circuit families the cascade boundary could be miscalibrated unless the fits are re-estimated online.
  • The Tier-2 time-to-first-result misses under noisy and adversarial queueing point to the scheduler, not the estimation stack; an age-aware or size-aware dispatch policy is a natural next step that the paper itself flags.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper presents MaestroCut, a closed-loop framework for quantum circuit cutting that adapts partitioning, shot allocation, and estimator selection to runtime drift, and adds a confidentiality layer called PhasePad-OTP. The system comprises an incremental multilevel-FM partitioner with CUSUM-triggered refinement, a Kalman-filtered topology-aware shot allocator (Topo-GP) with a closed-form water-filling solution, an entropy-gated estimator cascade, and a Pauli-padding security layer. The evaluation includes Tier-1 simulation across five workloads and Tier-2 emulation with calibrated noise and queueing dynamics. The central claims are variance contraction versus uniform allocation, stable SLOs under drift, approximately 1% security overhead, and IND-CFA fragment confidentiality with a 2^{-λ} advantage bound.

Significance. If the security claim were supportable, the paper would make a useful systems contribution: it is the first framework I am aware of that co-optimizes partitioning, allocation, estimation, and confidentiality in a closed loop. The allocation mathematics (Prop. 1) is correct, the Tier-2 emulation methodology with calibrated noise and queueing is a strength, and the open-source artifact is commendable. However, the IND-CFA security bound is not established for the scheme as described, and this is a headline advertised feature. The rest of the system—drift tracking, topology-aware allocation, and the empirical evaluation—is defensible and well executed, but the paper’s overall claim of a 'cryptographically secure' cutting framework currently rests on an unsupported security result.

major comments (2)
  1. [§IV.E, Algorithm 4, Proposition 4] The claimed IND-CFA bound (≤ 2^{-λ} + ε_AEAD(q)) does not follow from the described scheme. Algorithm 4 uses a per-fragment key K_i ∈ {0,1}^a with a = ⌈λ/2⌉ and applies only a phase mask Z^{K_i}. This is not a quantum one-time pad: the paper's own §II states QOTP requires 2n key bits for an n-qubit fragment. Averaging over K_i dephases the state in the computational basis, so an adversary can distinguish fragments whose Z-basis populations differ (e.g., |0⟩ vs |1⟩) with advantage 1. The proof sketch's hybrid step 'replace phase masks by uniform one-time pads' substitutes a different encryption than the one specified. Moreover, even if the phase mask were a full QOTP, a key of length a=λ/2 yields at best a guessing advantage of 2^{-λ/2}, not 2^{-λ}. Proposition 4 therefore does not support the fragment-confidentiality pillar of the central claim.
  2. [§V.C, Fig. 3; §VI.c] The variance-contraction validation is partly circular. The 'Theory ε(d,Q)' curve in Fig. 3 is computed from the same uncertainty proxy u_i that is used to allocate shots via Eq. (8), so the agreement between theory and empirical is to a significant degree by construction. In addition, the Tier-1 noise processes are generated from the same Matérn-1/2 kernel with Wiener drift that the Kalman/Topo-GP model assumes (Eqs. 4–5); as the paper concedes in §VI.c, 'different noise processes may affect constants.' The reported contraction ratios (Fig. 4) are therefore not yet evidence that the method transfers to hardware with heavy-tailed drift, structural breaks, or non-spatial correlations. Please add robustness experiments with misspecified priors, or temper the 'drift-resilient' claim accordingly.
minor comments (6)
  1. [§II vs §IV.E] The paper states in §II that QOTP requires 2n uniformly random key bits, but §IV.E uses a=⌈λ/2⌉ and a phase-only mask. This inconsistency should be resolved; the PhasePad-OTP scheme should not be described as inheriting QOTP guarantees.
  2. [§V.C, Fig. 3] The 'Theory ε(d,Q)' curve is not defined; no equation or algorithm is provided for ε(d,Q). Please define this quantity explicitly, including how it is computed from the model parameters.
  3. [§V.C, RQ3, Fig. 10] The cascade optimality evaluation appears to use the same fitted parameters (α, β, Bias_MLE) both to derive the decision boundary and to generate the empirical choices. Please describe the train/test split or cross-validation procedure that prevents circularity in the RQ3 result.
  4. [§IV.C, Eq. (6)] The bound uses λ_max(Σ̃), but the paper does not state how λ_max is estimated or upper-bounded in practice. Adding this detail would make the bound auditable.
  5. [Algorithm 2, line 7] The expression s^⋆_i ← max{s_min, S w_i/Z} can violate the total budget Σ s_i = S when the floor binds; the integer projection in lines 8–15 corrects it, but the formula is misleading. Suggest presenting the capped version only after projection is applied.
  6. [§IV.F] The claim 'guaranteeing ∥T_full − T_12-term∥_1 ≤ 0.03' lacks a proof or reference. Please provide a derivation or a citation for this bound.

Circularity Check

3 steps flagged

Variance contraction and cascade optimality are validated against the same equations that define the allocator/selector; the PhasePad security proof substitutes a full QOTP for the implemented phase-only mask.

specific steps
  1. fitted input called prediction [Section IV.C, Eq. (6)-(8) and Prop. 1; Section V.C Fig. 3]
    "Relaxing (6) by Lemma 1 yields the convex program min_{s_i≥s_min} Σ_i u_i^2/s_i^2 s.t. Σ_i s_i = S. ... Proposition 1 (KKT water-filling). The unique optimum ... is s⋆_i = S u_i^{2/3}/Σ_j u_j^{2/3}. ... Fig. 3: 'Variance contraction and theoretical bound. ... solid curve shows ε(d,Q) when available.'"

    The allocation is chosen to minimize the spectral-relaxed proxy Σ_i u_i^2/s_i^2, and the KKT formula is the minimizer of that proxy. Therefore the contraction of this proxy relative to uniform is a mathematical consequence of the optimizer for any positive u_i; it is not an empirical discovery about hardware noise. The 'Theory' curve is plotted against the same u_i used to allocate shots, so it reports the optimizer's own objective gap rather than independently validating that the proxy tracks true stitched variance.

  2. self definitional [Section IV.D, Eq. (9)-(10), Algorithm 3; Section V.C Fig. 10]
    "Empirical fits from Tier-1 calibration yield MSE_shad(s_i) ≈ α/s_i, MSE_MLE(s_i,H_i) ≈ β/s_i^2 + Bias_MLE(H_i)^2 ... Risk-minimising decision rule: Choose the estimator that minimises predicted MSE: E⋆_i = arg min{α/s_i, β/s_i^2 + Bias_MLE(H_i)^2}. ... Fig. 10: 'The cascade aligns with the model boundary MSE_MLE(H,s)=MSE_Shadows(s).'"

    Algorithm 3 implements exactly Eq. (10), using the same fitted α, β, and Bias_MLE from Eq. (9). Thus the RQ3 test compares the cascade's decisions to the decision rule that defines the cascade itself. Agreement between the plotted boundary and the decisions is built into the algorithm unless numerical or estimation noise corrupts the implementation; it does not provide independent evidence that the fitted MSE models are accurate.

  3. other [Section IV.E, Algorithm 4 and Proposition 4 proof sketch]
    "Algorithm 4: 'a ← ⌈λ/2⌉; ... sample K_i ∈ {0,1}^a; apply phase mask Z^{K_i} (Pauli-frame compatible)'. Proposition 4 proof sketch: 'replace headers by pseudorandom strings (AEAD) and phase masks by uniform one-time pads that commute with Pauli frames; indistinguishability reduces to key guessing.'"

    The implemented encryption is only a phase mask Z^{K_i} with |K_i|=⌈λ/2⌉, which is not a quantum one-time pad. The paper itself states in Section II that hiding an n-qubit fragment requires 2n uniformly random key bits. The proof's hybrid step replaces the phase mask with a uniform one-time pad over the full Pauli group, i.e., with a stronger primitive than the scheme actually uses. Consequently the claimed 2^{-λ}+ε_AEAD(q) bound is derived from a different algorithm than the one specified, so the IND-CFA security claim does not follow for Algorithm 4.

full rationale

The paper's core allocation is a valid minimizer of its stated proxy, so it is not circular in the sense of a false algebraic derivation. However, the paper presents variance contraction as evidence for the method while the 'theory' curve is the optimizer's own objective ratio: the KKT water-filling formula is the minimizer of Σ u_i^2/s_i^2, so contraction of that proxy is by construction. The estimator-cascade optimality evaluation is a direct tautology: the cascade and the 'predicted H–s boundary' are the same fitted decision rule. The PhasePad security bound is a load-bearing substitution gap rather than a self-citation: the proof replaces the implemented phase-only mask with a full QOTP, which the scheme does not use. This affects the headline confidentiality claim but is not exactly a circular derivation, so I record it under 'other'. Self-citations are not the source of the problem; the issues are internal to the paper's definitions and proof choices. Overall, two central empirical validations reduce by construction and the security proof relies on an un-implemented stronger primitive, warranting a score of 6.

Axiom & Free-Parameter Ledger

7 free parameters · 9 axioms · 0 invented entities

The central claims depend on fitted MSE coefficients and bias models for the cascade, unreported hyperparameters for the GP and Kalman trackers, hand-set CUSUM thresholds and partition weights, and a security parameterization (ceil(lambda/2)-bit phase mask) inconsistent with the claimed 2^{-lambda} bound and with the 2n-bit QOTP requirement quoted from the literature. The mathematical core (spectral relaxation, water-filling) is standard, so the ledger is dominated by domain assumptions about noise structure and by fitted engineering constants.

free parameters (7)
  • Cascade MSE coefficients alpha, beta = workload-dependent, fitted from Tier-1 calibration
    Eq. (9): MSE_shad approx alpha/s_i, MSE_MLE approx beta/s_i^2 + Bias^2. The 'MSE-optimal' decision rule is defined by these fits, so the RQ3 optimality claim inherits them.
  • Bias_MLE(H_i) model = measured from 1% pilot entropy
    Shifts the cascade crossover point s_x(H_i) in Eq. (10); asserted to be small at high entropy.
  • Matérn kernel parameters sigma_k^2, ell = not reported
    Eq. (5) defines Sigma_pq = sigma_k^2 exp(-d(p,q)/ell); shapes the topology prior used in the relaxed variance bound and the allocation.
  • Kalman process and observation noise q_i, r_i = q_i calibrated from historical deltas; r_i unspecified
    Eq. (4); controls how fast the tracker follows drift and how much it trusts ShotQC observations.
  • CUSUM slack kappa and threshold h = chosen for target ARL0; values not reported
    Eq. (2); sets the repartition trigger frequency for each calibrated metric.
  • Partition policy weights alpha, beta, gamma = hand-chosen, sum to 1
    Eq. (1); trades cut count, e-bits, and predicted queue delay in the partition objective.
  • Decoy fraction eta and verifier threshold eps_ver = eta = 0.02; eps_ver not stated
    PhasePad decoy budget and abort threshold; drives the detectable-tampering claims of Prop. 5.
axioms (9)
  • standard math Rayleigh quotient bound: y^T Sigma y <= lambda_max(Sigma) ||y||^2 for positive semidefinite Sigma
    Lemma 1, used to convert the stitched-variance bound (Eq. 6) into the separable program (7).
  • standard math KKT stationarity and strict convexity give the unique water-filling optimum s_i proportional to u_i^{2/3}
    Proposition 1, Eq. (8).
  • domain assumption Per-fragment variance follows a random walk: sigma^2_t = sigma^2_{t-1} plus Gaussian noise
    Eq. (4); the Kalman tracker's model. Admitted in Section VI.c to be an assumption.
  • domain assumption Cross-fragment variance correlation is Matérn-1/2 in heavy-hex graph distance
    Eq. (5); without this, Topo-GP has no topology signal and the 'topology-aware' claim loses content.
  • ad hoc to paper Estimator MSE has power-law forms MSE_shad approx alpha/s and MSE_MLE approx beta/s^2 + Bias^2
    Eq. (9), asserted from 'empirical fits'; the cascade optimality claim depends on these functional forms.
  • domain assumption ShotQC tail factor sqrt(log(N/rho)) is a valid per-fragment variance proxy u_i
    Eq. (6); imported from ShotQC [5] without derivation here.
  • domain assumption AES-GCM-SIV is standard-secure and its AEAD distinguishing advantage bounds the hybrid argument
    Proposition 4 proof; standard crypto assumption, but unproven in this paper.
  • domain assumption Semi-honest provider for confidentiality, trusted client and post-processor, no collusion
    Section III threat model; weaker than a malicious-provider model and explicitly scoped.
  • ad hoc to paper A phase mask with a = ceil(lambda/2) bits yields 2^{-lambda} indistinguishability
    Proposition 4 relies on it; inconsistent with the paper's own 2n-bit QOTP requirement and with the described mask length.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MAESTROCUT: Dynamic, Noise-Adaptive, and Secure Quantum Circuit Cutting on Near-Term Hardware." pith.science (2026). https://pith.science/paper/OFIOS6VM

@misc{pith2026250900811,
  author       = {Pith},
  title        = {Pith review of: MAESTROCUT: Dynamic, Noise-Adaptive, and Secure Quantum Circuit Cutting on Near-Term Hardware},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OFIOS6VM}},
  note         = {Machine review of arXiv:2509.00811}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present MaestroCut, a closed-loop framework for quantum circuit cutting that adapts partitioning and shot allocation to device drift and workload variation. MaestroCut tracks a variance proxy in real time, triggers re-cutting when accuracy degrades, and routes shots using topology-aware priors. An online estimator cascade (MLE, Bayesian, GP-assisted) selects the lowest-error reconstruction within a fixed budget. Tier-1 simulations show consistent variance contraction and reduced mean-squared error versus uniform and proportional baselines. Tier-2 emulation with realistic queueing and noise demonstrates stable latency targets, high reliability, and ~1% software overhead under stress scenarios. These results indicate that adaptive circuit cutting can provide accuracy and efficiency improvements with minimal operational cost on near-term hardware.

Figures

Figures reproduced from arXiv: 2509.00811 by Krishnendu Guha, Samuel Punch.

Figure 1
Figure 1. Figure 1: System pipeline for MAESTROCUT: policy-aware compilation, circuit cutting and phase padding, adaptive shot allocation, monitored execution with leakage-bound enforcement, and immutable audit logging. B. Threat Model & Security Layer We assume a trusted client C and post-processor P, and an execution provider S that is semi-honest for confidentiality (may inspect traffic and metadata) and potentially malici… view at source ↗
Figure 2
Figure 2. Figure 2: Threat model over the MAESTROCUT pipeline. Red boxes denote adversaries; dashed arrows indicate attack surfaces; inline labels show primary mitigations. Participants and trust boundaries.: • Client (C). Generates and rotates Kmast per fragment batch; performs PhasePad encryption and post-decrypt Local￾Shadow checks. • Provider (S). Executes Execute(F˜ i , si); returns cipher￾text outcomes with calibration … view at source ↗
Figure 3
Figure 3. Figure 3: Variance contraction and theoretical bound. Dashed line marks the 0.6 target; solid curve shows ε(d, Q) when available. Topology-aware allocation improves over uniform. qaoa30 uccsd24 tfim20 clifford24 phase16 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 C o ntractio n # (V ar/V arunif) lower is better qaoa30 uccsd24 tfim20 clifford24 phase16 0 10 20 30 40 50 60 70 T ail p95 sh ots # qaoa30 uccsd24 tfim20 clifford2… view at source ↗
Figure 4
Figure 4. Figure 4: Allocator ablation. Variance contraction, p95 tail shots, and final MSE across workloads. Topo-GP contracts variance and suppresses tails without sacrificing accuracy. 10 20 30 40 50 Shots per fragment (last step) 0.0 0.2 0.4 0.6 0.8 1.0 Cumulative Probability Shot Tail Behaviour (ECDF) qaoa30 uccsd24 tfim20 clifford24 phase16 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 7
Figure 7. Figure 7: Closed-loop timeline. Measurement proxy, allocation, and innovation with repartition triggers (vertical ticks). within=0.26 between=0.23 between/within=0.90 ( ) (Pre) within=0.25 between=0.24 between/within=0.95 ( ) =+0.05 (Post) = post pre 0.0 0.2 0.4 0.6 0.8 1.0 Covariance 0.4 0.2 0.0 0.2 0.4 Covariance Topology Covariance (pre vs post) qaoa30 (seed 0) [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 5
Figure 5. Figure 5: Shot-tail distributions. Final-step distributions with p95 indicators; Topo-GP keeps tails within budget. 20 40 60 Allocated si qaoa30 (½=0.88) uccsd24 (½=0.88) 0 1 2 Kalman ¹i tfim20 (½=0.88) 0 1 2 Kalman ¹i 20 40 60 Allocated si clifford24 (½=0.88) 0 1 2 Kalman ¹i phase16 (½=0.88) Variance Shots Coupling (last step) [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 9
Figure 9. Figure 9: Embedding and allocation before/after. Spectral communities and hotspot redistribution under repartition (shared scale). RQ3: Estimator Cascade Optimality. 0.0 0.2 0.4 0.6 0.8 1.0 Entropy Hi 20 25 30 35 40 45 S h ots si Cascade Decision Boundary qaoa30 (pooled seeds & steps) Shadows MLE ¢MSE=0 boundary ± margin [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 13
Figure 13. Figure 13: Runtime & memory. (a) Runtime per batch (s); (b) peak memory (GB). Where n is fixed, we report per-workload means with 95% CIs; otherwise scatter with a trendline. Contraction 0.6 Tail p95 60 shots Overhead 1% qaoa30 uccsd24 tfim20 clifford24 phase16 Tier-1 Targets Dashboard 0 1 [PITH_FULL_IMAGE:figures/full_fig_p010_13.png] view at source ↗
Figure 10
Figure 10. Figure 10: Decision boundary H vs. s. The cascade aligns with the model boundary MSEMLE(H, s) = MSEShadows(s), with transient outliers. qaoa30 uccsd24 tfim20 clifford24 phase16 100 80 60 40 20 0 Reduction in Final MSE vs Uniform (%) Accuracy contribution (per-seed spread + 95% CI) Proportional Topo-GP [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 14
Figure 14. Figure 14: Tier-1 targets dashboard. Pass/fail for contraction (≤ 0.6), tail shots, and overhead (≤ 1%). D. Tier 2 — Emulation (Calibrated Noise & Queue Dynamics) Primarily addresses RQ2 and RQ4; also stress-tests end￾to-end performance targets. We emulate calibrated device noise and queue dynamics under four scenarios (Baseline, Noisy, Bursty, Adversarial). Unless stated otherwise, we report medians with bootstrap … view at source ↗
Figure 11
Figure 11. Figure 11: Accuracy contribution. Reduction in final MSE relative to Uniform; Topo-GP provides the main gains and the cascade adds further improvements on low-entropy fragments. RQ4: Security & Overheads (PhasePad-OTP). 0 1 2 3 4 5 PhasePad overhead (%) 95 96 97 98 99 100 Relative throughput (%) (100% = 0% overhead) Overhead vs throughput (normalized per seed) qaoa30 uccsd24 tfim20 clifford24 phase16 1% target [PIT… view at source ↗
Figure 18
Figure 18. Figure 18: Overhead vs. throughput. Relative throughput (normalised per scenario) vs. PhasePad overhead; vertical line at 1%. Target summary.: The dashboard in [PITH_FULL_IMAGE:figures/full_fig_p011_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Tier-2 targets dashboard. Pass/fail for jitter, TTFR, reliability caps, and 1% overhead. Tier-2 discussion.: Jitter remains comfortably below the 150 ms SLO across all scenarios, indicating that the control loop and queueing strategy avoid burst amplification in short￾term latencies. In contrast, TTFR meets the 220 ms target in Baseline and Bursty but misses under Noisy (marginal overshoot consistent with… view at source ↗
Figure 17
Figure 17. Figure 17: Tier-2 throughput. Raw and success QPS (median ± 95% CI). Labels show absolute QPS and ∆ vs. Baseline. Overhead (%) 95 96 97 98 99 100 Relative throughput (%) 1.0% Baseline Overhead (%) Relative throughput (%) 1.0% Noisy 0 1 2 3 4 5 Overhead (%) 95 96 97 98 99 100 Relative throughput (%) 1.0% Bursty 0 1 2 3 4 5 Overhead (%) Relative throughput (%) 1.0% Adversarial Overhead vs throughput (normalized per sc… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

25 extracted references · 16 canonical work pages · 5 internal anchors

  1. [1]

    Fast quantum circuit cutting with randomized measurements,

    A. Lowe et al. , “Fast quantum circuit cutting with randomized measurements,” Quantum, vol. 7, p. 934, Mar. 2023, ISSN : 2521-327X. DOI: 10.22331/q-2023- 03-02-934. [Online]. Available: http://dx.doi.org/10. 22331/q-2023-03-02-934

    doi:10.22331/q-2023- 2023

  2. [2]

    D. Chen, B. Baheri, V . Chaudhary, Q. Guan, N. Xie, and S. Xu, Approximate quantum circuit cutting , 2022. arXiv: 2212.01270 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2212.01270

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  3. [3]

    Optimal joint cutting of two-qubit rotation gates,

    C. Ufrecht et al. , “Optimal joint cutting of two-qubit rotation gates,” Physical Review A, vol. 109, no. 5, 2024, ISSN : 2469-9934. DOI: 10.1103/physreva.109.052440. [Online]. Available: http://dx.doi.org/10.1103/PhysRevA. 109.052440

    work page doi:10.1103/physreva.109.052440 2024

  4. [4]

    Ancilla-driven blind quantum computation for clients with different quantum capabilities,

    Q. Dai, J. Quan, X. Lou, and Q. Li, “Ancilla-driven blind quantum computation for clients with different quantum capabilities,” 2023

    work page 2023

  5. [5]

    P.-H. Chen, D. -W. Chiou, and J. -H. R. Jiang, Enhanced quantum circuit cutting framework for sampling over- head reduction, 2024. arXiv: 2412.17704 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2412.17704

    work page arXiv 2024

  6. [6]

    Fragqc: An efficient quantum error reduction technique using quantum circuit fragmentation,

    S. Basu, A. Das, A. Saha, A. Chakrabarti, and S. Sur- Kolay, “Fragqc: An efficient quantum error reduction technique using quantum circuit fragmentation,” Journal of Systems and Software , vol. 214, p. 112 085, Aug. 2024, ISSN : 0164-1212. DOI: 10.1016/j.jss.2024.112085. [Online]. Available: http://dx.doi.org/10.1016/j.jss.2024. 112085

    arXiv 2024

  7. [7]

    Private quantum computation: An introduction to blind quantum computing and related protocols,

    J. F. Fitzsimons, “Private quantum computation: An introduction to blind quantum computing and related protocols,” npj Quantum Information, vol. 3, no. 1, p. 23,

    work page

  8. [8]

    C. Lu, R. Telang, A. Aysu, and K. Basu, Quantum leak: Timing side-channel attacks on cloud-based quantum services, 2024. arXiv: 2401.01521 [quant-ph]

    Pith/arXiv arXiv 2024

  9. [9]

    Simulat- ing large quantum circuits on a small quantum computer,

    T. Peng, A. W. Harrow, M. Ozols, and X. Wu, “Simulat- ing large quantum circuits on a small quantum computer,” Physical Review Letters , vol. 125, no. 15, Oct. 2020, ISSN : 1079-7114. DOI: 10.1103/physrevlett.125.150504. [Online]. Available: http : / / dx . doi . org / 10 . 1103 / PhysRevLett.125.150504

    doi:10.1103/physrevlett.125.150504 2020

  10. [10]

    Cutqc: Using small quantum computers for large quantum circuit evaluations,

    W. Tang, T. Tomesh, M. Suchara, J. Larson, and M. Martonosi, “Cutqc: Using small quantum computers for large quantum circuit evaluations,” in Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Sys- tems, ser. ASPLOS ’21, ACM, Apr. 2021, pp. 473–486. DOI: 10.1145/3445814.3446758. [Online]. ...

    arXiv 2021

  11. [11]

    Bechtold, J

    M. Bechtold, J. Barzen, F. Leymann, and A. Mandl, Circuit cutting with non-maximally entangled states ,

    work page

  12. [12]

    QRCC: Evaluating Large Quantum Circuits on Small Quantum Computers through Integrated Qubit Reuse and Circuit Cutting

    A. Pawar et al., Qrcc: Evaluating large quantum circuits on small quantum computers through integrated qubit reuse and circuit cutting , 2025. DOI: https : / / doi . org / 10 . 1145 / 3622781 . 3674179. arXiv: 2312 . 10298 [quant-ph]. [Online]. Available: https://arxiv.org/ abs/2312.10298

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    Efficient quantum circuit cutting by neglecting basis elements,

    D. T. Chen et al., “Efficient quantum circuit cutting by neglecting basis elements,” 2023

    work page 2023

  14. [14]

    Quantum circuit cutting for classical shadows,

    D. T. S. Chen, Z. H. Saleem, and M. A. Perlin, “Quantum circuit cutting for classical shadows,” ACM Transactions on Quantum Computing , vol. 5, no. 2, pp. 1–21, Jun. 2024, ISSN : 2643-6817. DOI: 10.1145/3665335. [Online]. Available: http://dx.doi.org/10.1145/3665335

    doi:10.1145/3665335 2024

  15. [15]

    Z. Li, M. Guo, M. Barad, W. Tang, E. Z. Zhang, and Y . Huang, A case for quantum circuit cutting for nisq applications: Impact of topology, determinism, and sparsity, 2024. arXiv: 2412.17929 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2412.17929

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [16]

    Ambainis, M

    A. Ambainis, M. Mosca, A. Tapp, and R. de Wolf, Private quantum channels and the cost of randomizing quantum information, 2000

    work page 2000

  17. [17]

    Universal blind quantum computation,

    A. Broadbent, J. Fitzsimons, and E. Kashefi, “Universal blind quantum computation,” in FOCS, 2009

    work page 2009

  18. [18]

    Quantum circuit cutting with maximum-likelihood tomography,

    M. A. Perlin, Z. H. Saleem, M. Suchara, and J. C. Os- born, “Quantum circuit cutting with maximum-likelihood tomography,” npj Quantum Information , vol. 7, no. 1, 2021, ISSN : 2056-6387. DOI: 10 . 1038 / s41534 - 021 - 00390-6. [Online]. Available: http://dx.doi.org/10.1038/ s41534-021-00390-6

    work page 2021

  19. [19]

    F. Burt, K. -C. Chen, and K. K. Leung, A multilevel framework for partitioning quantum circuits , 2025. arXiv: 2503.19082 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2503.19082

    arXiv 2025

  20. [20]

    Adap- tive mitigation of time-varying quantum noise,

    S. Dasgupta, T. S. Humble, and A. Danageozian, “Adap- tive mitigation of time-varying quantum noise,” in 2023 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 01, 2023, pp. 99–110. DOI: 10.1109/QCE57702.2023.00020

    arXiv 2023

  21. [21]

    CaliScalpel: In-Situ and Fine-Grained Qubit Calibration Integrated with Surface Code Quantum Error Correction

    X. Fang et al., Caliscalpel: In-situ and fine-grained qubit calibration integrated with surface code quantum error correction, 2024. arXiv: 2412 . 02036 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2412.02036

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  22. [22]

    Piveteau and D

    C. Piveteau and D. Sutter, Circuit knitting with classical communication , 2022. arXiv: 2205 . 00016 [quant-ph]

    work page 2022

  23. [23]

    Predicting many properties of a quantum system from very few measurements,

    H.-Y . Huang, R. Kueng, and J. Preskill, “Predicting many properties of a quantum system from very few measurements,” Nature Physics, 2020

    work page 2020

  24. [2017]

    1038 / s41534 - 017 - 0025 - 3

    DOI: 10 . 1038 / s41534 - 017 - 0025 - 3. [Online]. Available: https://doi.org/10.1038/s41534-017-0025-3

    doi:10.1038/s41534-017-0025-3

  25. [2023]

    Circuit Cutting with Non-Maximally Entangled States

    arXiv: 2306 . 12084 [quant-ph]. [Online]. Available: https://arxiv.org/abs/2306.12084

    work page internal anchor Pith review Pith/arXiv arXiv

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.