pith. sign in

arxiv: 2602.10049 · v2 · pith:VEH7CUELnew · submitted 2026-02-10 · 🪐 quant-ph

Random Quantum Circuits as Seeds for Continuous Generative Models

Olli Hirviniemi , Afrad Basheer , Thomas Cope This is my paper

Pith reviewed 2026-05-22 10:41 UTC · model grok-4.3

classification 🪐 quant-ph
keywords random quantum circuitsgenerative modelsNISQ deviceshybrid quantum-classical modelstensor network contractionPauli propagationcontinuous data generationquantum machine learning
0
0 comments X

The pith

Random quantum circuits resistant to classical simulation can seed generative models for NISQ hybrid systems.

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

The paper introduces a family of random quantum circuits that resist simulation via tensor network contraction and Pauli propagation. It shows that local variables in these circuits do not concentrate, preserving the variance needed to generate diverse outputs. The central argument is that these circuits can function as a random seed inside a larger classical generative model. This construction would make large-scale quantum-classical hybrid models practical on noisy intermediate-scale quantum hardware. A reader would care if the approach opens a workable path to combining quantum randomness with classical modeling without waiting for fault-tolerant machines.

Core claim

We introduce a random circuit family and show they are robust against current classical simulation techniques, specifically tensor network contraction and Pauli propagation. We also show that local variables do not concentrate, ensuring enough variance to be able to produce a diverse set of data points. We therefore argue that using these circuits as a random seed for a larger classical generative model is a way to make large-scale quantum-classical hybrid models amenable towards NISQ devices.

What carries the argument

Random quantum circuits used as seeds for larger classical generative models, leveraging their resistance to tensor-network and Pauli-propagation simulation together with non-concentration of local variables.

If this is right

  • Hybrid models can incorporate quantum components without requiring full classical simulation of the quantum part.
  • Generative tasks gain access to non-concentrating variance that supports production of diverse continuous data points.
  • Large-scale modeling becomes feasible on current quantum hardware rather than requiring future fault-tolerant devices.
  • The seed approach separates the quantum contribution from the classical model, allowing independent scaling of each.

Where Pith is reading between the lines

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

  • The same seeding technique could be tested in other quantum machine learning settings such as sampling or optimization tasks.
  • Hardware-specific tuning of circuit depth or gate choice might further improve performance on particular NISQ platforms.
  • Direct comparison of sample diversity between the hybrid model and purely classical baselines on real devices would clarify the practical gain.

Load-bearing premise

Resistance to tensor network contraction and Pauli propagation, together with non-concentration of local variables, is sufficient to deliver practical advantages when the circuits serve as seeds in generative modeling on NISQ devices.

What would settle it

An efficient classical simulation of the circuit family via tensor networks or Pauli propagation, or a demonstration that local variables concentrate to a single value, would remove the basis for using them as effective seeds.

Figures

Figures reproduced from arXiv: 2602.10049 by Afrad Basheer, Olli Hirviniemi, Thomas Cope.

Figure 1
Figure 1. Figure 1: FIG. 1. The proposed structure for a continuous variable quantum generative model. One could also run the full circuit, rather [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. The trainable part is composed of parametrized 2-qubit gates arranged in alternating pattern. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. The generative part has [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. The average running time of Pauli propagation algorithm on a laptop with cutoff at [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

We introduce a random circuit family and show they are robust against current classical simulation techniques, specifically tensor network contraction and Pauli propagation. We also show that local variables do not concentrate, ensuring enough variance to be able to produce a diverse set of data points. We therefore argue that using these circuits as a "random seed" for a larger classical generative model is a way to make large-scale quantum-classical hybrid models amenable towards NISQ devices.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces a family of random quantum circuits claimed to be robust against tensor network contraction and Pauli propagation. It further shows that local variables do not concentrate, preserving variance for data diversity. The central argument is that these circuits can serve as random seeds within larger classical generative models, thereby rendering large-scale quantum-classical hybrid generative models feasible on NISQ hardware.

Significance. If the hardness and non-concentration claims hold at NISQ-accessible scales, the approach could provide a concrete route to hybrid models that exploit quantum randomness without requiring deep quantum circuits for the full generative task. The emphasis on resistance to specific classical methods and explicit variance preservation is a positive feature, though the practical advantage remains conditional on depth and size scaling.

major comments (2)
  1. [Abstract and hardness analysis] The robustness claims against tensor network contraction and Pauli propagation are presented without any reported circuit depth, qubit number, or total gate count (see abstract and the hardness analysis section). This omission directly affects the central NISQ-amenability claim, because simulation hardness in random circuit families is known to be depth-dependent; without these parameters it is impossible to verify that the demonstrated hardness occurs inside the coherence and fidelity limits of current NISQ devices.
  2. [Variance and diversity section] The assertion that local variables do not concentrate is used to guarantee sufficient variance for generative diversity, yet no quantitative measures (e.g., variance values, histograms, or scaling with depth) or explicit comparison to concentrating circuits are supplied. This weakens the link between the circuit property and the utility as a seed for continuous generative models.
minor comments (2)
  1. [Methods] The precise gate set, connectivity, and sampling procedure for the random circuit family should be stated explicitly, preferably with a pseudocode or circuit diagram, to allow reproduction.
  2. [Generative model integration] A short discussion of how the classical generative model is trained or conditioned on the quantum seed outputs would clarify the hybrid workflow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation of our results on random quantum circuits for hybrid generative models. We address each major comment below and have revised the manuscript to incorporate the suggested improvements.

read point-by-point responses

  1. Referee: [Abstract and hardness analysis] The robustness claims against tensor network contraction and Pauli propagation are presented without any reported circuit depth, qubit number, or total gate count (see abstract and the hardness analysis section). This omission directly affects the central NISQ-amenability claim, because simulation hardness in random circuit families is known to be depth-dependent; without these parameters it is impossible to verify that the demonstrated hardness occurs inside the coherence and fidelity limits of current NISQ devices.

    Authors: We agree that the abstract and hardness analysis section would be strengthened by explicitly reporting the circuit parameters. In the revised manuscript we have updated the abstract to state the qubit numbers (up to 20) and depths (linear in qubit number) used throughout the hardness analysis. We have also inserted a concise summary paragraph and parameter table at the opening of the hardness analysis section that lists qubit count, depth, and total gate count for the families considered. These scales are chosen to lie within the coherence times and gate fidelities of present-day NISQ hardware, thereby supporting the NISQ-amenability argument. revision: yes

  2. Referee: [Variance and diversity section] The assertion that local variables do not concentrate is used to guarantee sufficient variance for generative diversity, yet no quantitative measures (e.g., variance values, histograms, or scaling with depth) or explicit comparison to concentrating circuits are supplied. This weakens the link between the circuit property and the utility as a seed for continuous generative models.

    Authors: We acknowledge that the original variance section relied primarily on analytical non-concentration arguments and qualitative illustrations. In the revision we have added explicit numerical variance values for local Pauli observables as a function of depth, together with histograms of the observed distributions. We have further included a direct comparison against a concentrating reference circuit family (fixed-depth Haar-random circuits) to quantify the variance preservation. These additions are now placed in the revised variance and diversity section and directly tie the non-concentration property to the utility of the circuits as seeds for continuous generative models. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on direct analysis of circuit properties

full rationale

The paper introduces a random circuit family and reports analytical or numerical demonstrations of robustness to tensor network contraction and Pauli propagation, along with non-concentration of local variables. These results are presented as independent properties of the defined circuits rather than outputs of any fitting procedure or self-referential definition. The subsequent argument that the circuits can serve as seeds for NISQ-amenable hybrid generative models is framed as a suggestion following from those properties, not as a derived quantity forced by the inputs. No equations, uniqueness theorems, or ansatzes are shown to reduce to prior self-citations or fitted parameters by construction. The derivation chain remains self-contained against external benchmarks of circuit simulation hardness.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central proposal rests on the unverified assertion that the introduced circuit family possesses the stated robustness and variance properties; no free parameters or invented entities are evident from the abstract.

axioms (2)
  • domain assumption The introduced random circuit family is robust against tensor network contraction and Pauli propagation.
    Presented as a demonstrated result in the abstract without supporting details or methods.
  • domain assumption Local variables in these circuits do not concentrate, providing sufficient variance for diverse data generation.
    Stated as shown but without quantitative evidence or definitions in the provided abstract.

pith-pipeline@v0.9.0 · 5589 in / 1306 out tokens · 57827 ms · 2026-05-22T10:41:34.537234+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    P. W. Shor, SIAM Journal on Computing26, 1484 (1997)

    work page 1997

  2. [2]

    Preskill, Quantum2, 79 (2018)

    J. Preskill, Quantum2, 79 (2018)

    work page 2018

  3. [3]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles, Nature Reviews Physics3, 625–644 (2021)

    work page 2021

  4. [4]

    Introduction to variational quantum algorithms,

    M. St¸ ech ly, “Introduction to variational quantum algorithms,” (2024), arXiv:2402.15879 [quant-ph]

    work page arXiv 2024

  5. [5]

    Jerbi, L

    S. Jerbi, L. J. Fiderer, H. P. Nautrup, J. M. K¨ ubler, H. J. Briegel, and V. Dunjko, Nature Communications14(2023)

    work page 2023

  6. [6]

    J. M. McClean, S. Boixo, V. N. Smelyanskiy, R. Babbush, and H. Neven, Nature Communications9, 4812 (2018)

    work page 2018

  7. [7]

    H. Qi, L. Wang, H. Zhu, A. Gani, and C. Gong, Quantum Information Processing22, 435 (2023)

    work page 2023

  8. [8]

    Arrasmith, Z

    A. Arrasmith, Z. Holmes, M. Cerezo, and P. J. Coles, Quantum Science and Technology7, 045015 (2022)

    work page 2022

  9. [9]

    Larocca, S

    M. Larocca, S. Thanasilp, S. Wang, K. Sharma, J. Biamonte, P. J. Coles, L. Cincio, J. R. McClean, Z. Holmes, and M. Cerezo, “A review of barren plateaus in variational quantum computing,” (2024), arXiv:2405.00781 [quant-ph]

    work page arXiv 2024

  10. [10]

    Cerezo, A

    M. Cerezo, A. Sone, T. Volkoff, L. Cincio, and P. J. Coles, Nature Communications12, 1791 (2021)

    work page 2021

  11. [11]

    Khatri, R

    S. Khatri, R. LaRose, A. Poremba, L. Cincio, A. T. Sornborger, and P. J. Coles, Quantum3, 140 (2019)

    work page 2019

  12. [12]

    Zhao and X.-S

    C. Zhao and X.-S. Gao, Quantum5, 466 (2021)

    work page 2021

  13. [13]

    Liu, L.-W

    Z. Liu, L.-W. Yu, L.-M. Duan, and D.-L. Deng, Physical Review Letters129, 270501 (2022)

    work page 2022

  14. [14]

    Basheer, Y

    A. Basheer, Y. Feng, C. Ferrie, and S. Li, Proceedings of the AAAI Conference on Artificial Intelligence37, 6770–6778 (2023)

    work page 2023

  15. [15]

    Letcher, S

    A. Letcher, S. Woerner, and C. Zoufal, Quantum8(2024)

    work page 2024

  16. [16]

    Suzuki and M

    Y. Suzuki and M. Li, Phys. Rev. A110, 012409 (2024)

    work page 2024

  17. [17]

    Leone, S

    L. Leone, S. F. Oliviero, L. Cincio, and M. Cerezo, Quantum8, 1395 (2024)

    work page 2024

  18. [18]

    Pesah, M

    A. Pesah, M. Cerezo, S. Wang, T. Volkoff, A. T. Sornborger, and P. J. Coles, Physical Review X11, 041011 (2021)

    work page 2021

  19. [19]

    Park and N

    C.-Y. Park and N. Killoran, arXiv preprint arXiv:2302.08529 (2023)

    work page arXiv 2023

  20. [20]

    Y. Wang, B. Qi, C. Ferrie, and D. Dong, Phys. Rev. Appl.22, 054005 (2024)

    work page 2024

  21. [21]

    Zhang, L

    K. Zhang, L. Liu, M.-H. Hsieh, and D. Tao, Advances in Neural Information Processing Systems35, 18612 (2022)

    work page 2022

  22. [22]

    Cerezo, M

    M. Cerezo, M. Larocca, D. Garc´ ıa-Mart´ ın, N. L. Diaz, P. Braccia, E. Fontana, M. S. Rudolph, P. Bermejo, A. Ijaz, S. Thanasilp, E. R. Anschuetz, and Z. Holmes, Nature Communications16, 7907 (2025)

    work page 2025

  23. [23]

    Cheng, J

    S. Cheng, J. Chen, and L. Wang, Entropy20, 583 (2018)

    work page 2018

  24. [24]

    Liu and L

    J.-G. Liu and L. Wang, Phys. Rev. A98, 062324 (2018)

    work page 2018

  25. [25]

    Benedetti, D

    M. Benedetti, D. Garcia-Pintos, O. Perdomo, V. Leyton-Ortega, Y. Nam, and A. Perdomo-Ortiz, npj Quantum Information 5, 45 (2019)

    work page 2019

  26. [26]

    M. H. Amin, E. Andriyash, J. Rolfe, B. Kulchytskyy, and R. Melko, Phys. Rev. X8, 021050 (2018)

    work page 2018

  27. [27]

    Kieferov´ a and N

    M. Kieferov´ a and N. Wiebe, Phys. Rev. A96, 062327 (2017)

    work page 2017

  28. [28]

    Benedetti, J

    M. Benedetti, J. Realpe-G´ omez, R. Biswas, and A. Perdomo-Ortiz, Phys. Rev. X7, 041052 (2017)

    work page 2017

  29. [29]

    Ragone, B

    M. Ragone, B. N. Bakalov, F. Sauvage, A. F. Kemper, C. O. Marrero, M. Larocca, and M. Cerezo, Nature Communications (2024)

    work page 2024

  30. [30]

    Fontana, D

    E. Fontana, D. Herman, S. Chakrabarti, N. Kumar, R. Yalovetzky, J. Heredge, S. H. Sureshbabu, and M. Pistoia, Nature Communications (2024)

    work page 2024

  31. [31]

    Huang, R

    H. Huang, R. Kueng, and J. Preskill, Nature Physics16, 1050 (2020)

    work page 2020

  32. [32]

    M. J. Bremner, A. Montanaro, and D. J. Shepherd, Quantum1, 8 (2017)

    work page 2017

  33. [33]

    Lerch, R

    S. Lerch, R. Puig, M. S. Rudolph, A. Angrisani, T. Jones, M. Cerezo, S. Thanasilp, and Z. Holmes, arXiv:2411.19896 (2024), arXiv:2411.19896 [quant-ph]

    work page arXiv 2024

  34. [34]

    J. C. Bridgeman and C. T. Chubb, Journal of Physics A: Mathematical and Theoretical50, 223001 (2017)

    work page 2017

  35. [35]

    I. L. Markov and Y. Shi, SIAM Journal on Computing38, 963 (2008), https://doi.org/10.1137/050644756

    work page doi:10.1137/050644756 2008

  36. [36]

    C. Wang, T. Liu, P. Cui, and K. Xu, inCombinatorial Optimization and Applications, edited by W. Wang, X. Zhu, and D.-Z. Du (Springer Berlin Heidelberg, Berlin, Heidelberg, 2011) pp. 491–499

    work page 2011

  37. [37]

    S. Wang, E. Fontana, M. Cerezo, K. Sharma, A. Sone, L. Cincio, and P. J. Coles, Nature Communications12(2021), 10.1038/s41467-021-27045-6

    work page doi:10.1038/s41467-021-27045-6 2021