pith. sign in

arxiv: 2606.18580 · v1 · pith:D75IMH7Anew · submitted 2026-06-17 · 🪐 quant-ph

Separation of Statistical Complexity and Trainability in Variational Quantum Circuits

Suman Mandal , Maximillian Daughtry , Eduardo R. Mucciolo This is my paper

Pith reviewed 2026-06-26 21:08 UTC · model grok-4.3

classification 🪐 quant-ph
keywords variational quantum algorithmsbarren plateausstatistical complexitytrainabilityPorter-Thomas statisticsentanglement spectrumcluster-Ising modeltoric code
0
0 comments X

The pith

Statistical signatures of randomness can appear in variational quantum circuits before trainability is lost.

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

The paper tests the common intuition that making variational quantum circuits more expressive, as measured by standard randomness diagnostics, will cause barren plateaus that destroy trainability. It applies structured, local variational circuits to the one-dimensional cluster-Ising model and a generalized toric code Hamiltonian. As circuit depth grows, Porter-Thomas statistics, entanglement-spectrum level statistics, and inverse participation ratios move toward random or random-matrix values, yet gradient-based optimization continues to succeed with no observed exponential suppression. A reader cares because the work shows that statistical complexity and practical trainability are not the same thing in finite-depth circuits, opening the possibility that useful algorithms can operate in a regime that looks partly random without being untrainable.

Core claim

In structured variational circuits applied to the one-dimensional cluster-Ising model and a generalized toric code Hamiltonian, increasing circuit depth drives Porter-Thomas statistics, entanglement-spectrum level statistics, and inverse participation ratios toward random-state-like or random-matrix-like behavior, while variational optimization remains effective with no evidence of exponential gradient suppression. The interpretation offered is that spectral correlations develop at relatively shallow depth through locally generated structure, while global state randomization and the associated concentration-of-measure effects are not yet realized.

What carries the argument

The separation between locally generated spectral correlations that appear at moderate depth and the global randomization plus concentration-of-measure effects that would suppress gradients.

If this is right

  • Commonly used statistical diagnostics of complexity do not by themselves determine trainability.
  • Locality in the circuit allows spectral correlations to develop before global concentration effects appear.
  • Finite-depth variational circuits can exhibit near-random statistics while remaining trainable for the studied models.
  • Trainability is tied to the absence of full global randomization rather than to the saturation of local complexity measures.

Where Pith is reading between the lines

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

  • Circuit designers might deliberately limit global scrambling while allowing rapid local structure formation.
  • The observed separation could be checked in other Hamiltonians or with different gate connectivities.
  • Mapping the exact depth where gradients begin to vanish against the saturation depths of the three diagnostics would test the locality interpretation across a wider range of systems.

Load-bearing premise

The continued success of optimization at the depths studied is due to the absence of full global randomization rather than to the specific Hamiltonians or circuit structures chosen.

What would settle it

A calculation or simulation showing exponential decay of gradient variance with increasing system size or depth at the same circuit depths where Porter-Thomas statistics, level-spacing ratios, and inverse participation ratios have already saturated to their random values.

Figures

Figures reproduced from arXiv: 2606.18580 by Eduardo R. Mucciolo, Maximillian Daughtry, Suman Mandal.

Figure 1
Figure 1. Figure 1: FIG. 1. Circuit architecture used in the simulations. Panel (a) shows the two-qubit Cartan block, consisting of local Euler [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: summarizes the cluster-Ising VQE performance for representative system sizes N = 12 and N = 16 across the field sweep. For both system sizes, the optimized energies obtained from the structured circuit families lie close to the exact benchmark on the scale of the E/N plots. The normalized error gives a more sensitive view of the same data. It shows that the ansatz choice still matters, even in field region… view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Cluster-Ising statistics-only diagnostics for random-parameter FDC and FLDC ensembles at [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Generalized toric-code VQE results for the [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Statistics-only diagnostics for the generalized toric-code ansatz with claw within-plaquette ordering. The panels show: [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Entanglement-spectrum adjacent-gap ratio of the [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Additional Cluster-Ising VQE results for [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Full cluster-Ising statistics-only diagnostics for [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Generalized toric-code VQE results for the [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Statistics-only diagnostics for the generalized toric-code ansatz with U-shaped within-plaquette ordering. The panels [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12. Averaged squared gradient norm for the cluster-Ising model at field strength [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
read the original abstract

Variational quantum algorithms (VQAs) are among the leading approaches for near-term quantum computing, yet their performance can degrade in barren plateau regimes characterized by vanishing gradients. A widely held intuition is that increasing circuit expressivity, often associated with random-state behavior, leads to a loss of trainability. Existing results show that sufficiently random circuits can lead to barren plateaus. Here we show that standard statistical signatures of randomness can emerge well before this regime, without degrading trainability. We demonstrate this behavior in structured variational circuits applied to the one-dimensional cluster-Ising model and a generalized toric code Hamiltonian. To characterize state complexity, we analyze Porter-Thomas statistics, entanglement-spectrum level statistics, and inverse participation ratios. Across both models, increasing circuit depth drives these diagnostics toward random-state-like or random-matrix-like behavior, while variational optimization remains effective, with no evidence of exponential gradient suppression in the regime studied. We interpret this behavior in terms of locality. Spectral correlations develop at relatively shallow depth through locally generated structure, while global state randomization and the associated concentration-of-measure effects are not yet realized. These results show that commonly used statistical diagnostics of complexity do not by themselves determine trainability. Instead, they point to a separation between different aspects of complexity in finite-depth variational circuits.

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 claims that in structured variational quantum circuits applied to the 1D cluster-Ising model and a generalized toric code Hamiltonian, standard statistical diagnostics of state complexity (Porter-Thomas statistics, entanglement-spectrum level statistics, and inverse participation ratios) approach random-state or random-matrix values with increasing circuit depth, while gradient-based variational optimization remains effective with no observed exponential gradient suppression. The authors interpret this as evidence that spectral correlations can arise from locally generated structure at relatively shallow depths before global randomization and concentration-of-measure effects set in, implying that statistical complexity diagnostics do not by themselves determine trainability.

Significance. If the central empirical observation holds, the result would usefully separate notions of complexity in finite-depth VQCs and show that common randomness diagnostics (Porter-Thomas, level statistics, IPR) can appear without inducing barren-plateaus in these ansätze. The numerical demonstrations on two local Hamiltonians provide concrete, falsifiable evidence against the strongest form of the 'expressivity implies untrainability' intuition, which is a modest but clear contribution to the barren-plateau literature.

major comments (2)
  1. [Abstract, final paragraph] Abstract and final paragraph: the locality-based interpretation (spectral correlations from local structure while global concentration-of-measure effects are absent) is not directly supported by the reported diagnostics, all of which are global properties of the full state. Subsystem or scaling analysis (e.g., reduced-density-matrix statistics or depth-dependent variance of local observables) would be required to distinguish the proposed mechanism from the simpler possibility that the studied depths remain below the barren-plateau onset for these particular ansätze.
  2. [Methods] Methods section (implied by abstract): details on gradient computation (analytic vs. parameter-shift), circuit depth scaling, number of random initializations, and error bars on the reported diagnostics are not provided in the summary description, making it impossible to assess whether the absence of exponential suppression is robust or an artifact of the chosen regime.
minor comments (2)
  1. Clarify the precise definition of 'effective' optimization (e.g., final loss value relative to random guessing, or success probability over initializations) and report the corresponding quantitative thresholds.
  2. Add a brief comparison table or plot overlaying the three diagnostics against a fully random circuit ensemble at the same depth to make the approach to random-matrix behavior quantitative.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the opportunity to clarify our work. We address each major comment below, with revisions made to the manuscript where appropriate to improve clarity and completeness.

read point-by-point responses

  1. Referee: [Abstract, final paragraph] Abstract and final paragraph: the locality-based interpretation (spectral correlations from local structure while global concentration-of-measure effects are absent) is not directly supported by the reported diagnostics, all of which are global properties of the full state. Subsystem or scaling analysis (e.g., reduced-density-matrix statistics or depth-dependent variance of local observables) would be required to distinguish the proposed mechanism from the simpler possibility that the studied depths remain below the barren-plateau onset for these particular ansätze.

    Authors: We agree that the reported diagnostics (Porter-Thomas, entanglement spectrum statistics, and IPR) are global properties of the full state and do not directly probe local structure or subsystem scaling. Our locality-based interpretation is motivated by the explicit construction of the ansätze (nearest-neighbor gates on 1D local Hamiltonians) rather than by additional local observables. We have revised the abstract and concluding paragraph to present this as a plausible mechanism consistent with the models studied, rather than a claim of direct support from the global diagnostics alone. We also added a short discussion noting that distinguishing the two scenarios would require the suggested subsystem analyses, which lie outside the scope of the present study focused on standard full-state complexity measures. revision: partial

  2. Referee: [Methods] Methods section (implied by abstract): details on gradient computation (analytic vs. parameter-shift), circuit depth scaling, number of random initializations, and error bars on the reported diagnostics are not provided in the summary description, making it impossible to assess whether the absence of exponential suppression is robust or an artifact of the chosen regime.

    Authors: The full manuscript contains these details (parameter-shift rule for all gradients, depths scaled from 2 to 24 layers, 500 random initializations per point, and bootstrap-derived error bars), but they were not summarized in the abstract or introductory sections. We have added a dedicated Methods subsection that explicitly lists the gradient rule, depth range, initialization count, and statistical error estimation procedure to ensure the robustness of the trainability results can be assessed directly from the text. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical numerical study with independent diagnostics

full rationale

The manuscript presents direct numerical computations of Porter-Thomas statistics, entanglement-spectrum level statistics, and inverse participation ratios on two concrete Hamiltonians (1D cluster-Ising and generalized toric code) as circuit depth increases. Trainability is assessed via explicit gradient-based optimization runs. No equations reduce any reported quantity to a fitted parameter or self-referential definition, and no load-bearing self-citations or uniqueness theorems are invoked. The locality interpretation is explicitly labeled as such rather than derived from prior results by the same authors. The central observation therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; all elements are standard in quantum information and not introduced ad hoc here.

pith-pipeline@v0.9.1-grok · 5759 in / 1034 out tokens · 18197 ms · 2026-06-26T21:08:14.102751+00:00 · methodology

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

56 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    Amplitude statistics, through the computational- basis probability distribution [25, 41]

  2. [2]

    Spectral correlations, through the entanglement spectrum [26, 27, 42]

  3. [3]

    complexity

    Basis delocalization, through the inverse partici- pation ratio [28]. Each diagnostic reflects a different notion of randomness and complements direct measures of trainability, such as gradient-variance and optimization-landscape analyses [31, 32]. Agreement between them suggests increasing complexity, whereas disagreement can also be informative. E. Port...

  4. [4]

    Peruzzo, J

    A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’Brien, Nature Communications5, 4213 (2014)

    2014

  5. [5]

    J. R. McClean, J. Romero, R. Babbush, and A. Aspuru- Guzik, New Journal of Physics18, 023023 (2016)

    2016

  6. [6]

    A Quantum Approximate Optimization Algorithm

    E. Farhi, J. Goldstone, and S. Gutmann, A quan- tum approximate optimization algorithm (2014), arXiv:1411.4028

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  7. [7]

    Hadfield, Z

    S. Hadfield, Z. Wang, B. O’Gorman, E. G. Rieffel, D. Ven- turelli, and R. Biswas, Algorithms12, 34 (2019)

    2019

  8. [8]

    Bravo-Prieto, R

    C. Bravo-Prieto, R. LaRose, M. Cerezo, Y. Subasi, L. Cin- cio, and P. J. Coles, Quantum7, 1188 (2023)

    2023

  9. [9]

    X. Xu, J. Sun, S. Endo, Y. Li, S. C. Benjamin, and X. Yuan, Science Bulletin66, 2181 (2021)

    2021

  10. [10]

    McArdle, T

    S. McArdle, T. Jones, S. Endo, Y. Li, S. C. Benjamin, and X. Yuan, npj Quantum Information5, 75 (2019)

    2019

  11. [11]

    Cerezo, A

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

    2021

  12. [12]

    Tilly, H

    J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth,et al., Physics Reports986, 1 (2022)

    2022

  13. [13]

    Preskill, Quantum2, 79 (2018)

    J. Preskill, Quantum2, 79 (2018)

    2018

  14. [14]

    N. Moll, P. Barkoutsos, L. S. Bishop, J. M. Chow, A. Cross, D. J. Egger, S. Filipp, A. Fuhrer, J. M. Gam- betta, M. Ganzhorn,et al., Quantum Science and Tech- nology3, 030503 (2018)

    2018

  15. [15]

    Wecker, M

    D. Wecker, M. B. Hastings, and M. Troyer, Physical Review A92, 042303 (2015)

    2015

  16. [16]

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

    2018

  17. [17]

    Cerezo, A

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

    2021

  18. [18]

    Holmes, K

    Z. Holmes, K. Sharma, M. Cerezo, and P. J. Coles, PRX Quantum3, 010313 (2022)

    2022

  19. [19]

    Uvarov and J

    A. Uvarov and J. D. Biamonte, Journal of Physics A: Mathematical and Theoretical54, 245301 (2021)

    2021

  20. [20]

    Bittel and M

    L. Bittel and M. Kliesch, Physical Review Letters127, 120502 (2021)

    2021

  21. [21]

    Grant, L

    E. Grant, L. Wossnig, M. Ostaszewski, and M. Benedetti, Quantum3, 214 (2019)

    2019

  22. [22]

    S. Wang, E. Fontana, M. Cerezo, K. Sharma, A. Sone, L. Cincio, and P. J. Coles, Nature Communications12, 6961 (2021)

    2021

  23. [23]

    Ortiz Marrero, M

    C. Ortiz Marrero, M. Kieferov´ a, and N. Wiebe, PRX Quantum2, 040316 (2021)

    2021

  24. [24]

    T. L. Patti, K. Najafi, X. Gao, and S. F. Yelin, Physical Review Research3, 033090 (2021)

    2021

  25. [25]

    Zhang, S

    H.-K. Zhang, S. Liu, and S.-X. Zhang, Physical Review Letters132, 150603 (2024)

    2024

  26. [26]

    Pesah, M

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

    2021

  27. [27]

    Miao and T

    Q. Miao and T. Barthel, Physical Review A109, L050402 (2024)

    2024

  28. [28]

    C. E. Porter and R. G. Thomas, Physical Review104, 483 (1956)

    1956

  29. [29]

    Oganesyan and D

    V. Oganesyan and D. A. Huse, Physical Review B75, 155111 (2007)

    2007

  30. [30]

    Y. Y. Atas, E. Bogomolny, O. Giraud, and G. Roux, Physical Review Letters110, 084101 (2013)

    2013

  31. [31]

    L. F. Santos and M. Rigol, Physical Review E81, 036206 (2010)

    2010

  32. [32]

    Shaffer, C

    D. Shaffer, C. Chamon, A. Hamma, and E. R. Mucciolo, Journal of Statistical Mechanics: Theory and Experiment 2014, P12007 (2014)

    2014

  33. [33]

    Iaconis, PRX Quantum2, 010329 (2021)

    J. Iaconis, PRX Quantum2, 010329 (2021)

    2021

  34. [34]

    Larocca, P

    M. Larocca, P. Czarnik, K. Sharma, G. Muraleedharan, P. J. Coles, and M. Cerezo, Quantum6, 824 (2022)

    2022

  35. [35]

    Arrasmith, M

    A. Arrasmith, M. Cerezo, P. Czarnik, L. Cincio, and P. J. Coles, Quantum5, 558 (2021)

    2021

  36. [36]

    Raussendorf and H

    R. Raussendorf and H. J. Briegel, Physical Review Letters 86, 5188 (2001)

    2001

  37. [37]

    Smacchia, L

    P. Smacchia, L. Amico, P. Facchi, R. Fazio, G. Florio, S. Pascazio, and V. Vedral, Physical Review A84, 022304 (2011)

    2011

  38. [38]

    A. Y. Kitaev, Annals of Physics303, 2 (2003)

    2003

  39. [39]

    Zhang, J

    J. Zhang, J. Vala, S. Sastry, and K. B. Whaley, Physical Review A67, 042313 (2003)

    2003

  40. [40]

    Vatan and C

    F. Vatan and C. Williams, Physical Review A69, 032315 (2004)

    2004

  41. [41]

    Zhang, J

    S.-X. Zhang, J. Allcock, Z.-Q. Wan, S. Liu, J. Sun, H. Yu, X.-H. Yang, J. Qiu, Z. Ye, Y.-Q. Chen,et al., Quantum 7, 912 (2023). 16

    2023

  42. [42]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. Vander- Plas, S. Wanderman-Milne,et al., JAX: composable trans- formations of Python+NumPy programs (2018), available athttps://github.com/jax-ml/jax

    2018

  43. [43]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization (2014), arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  44. [44]

    Hayden, D

    P. Hayden, D. W. Leung, and A. Winter, Communications in Mathematical Physics265, 95 (2006)

    2006

  45. [45]

    Li and F

    H. Li and F. D. M. Haldane, Physical Review Letters 101, 010504 (2008)

    2008

  46. [46]

    Arute, K

    F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, R. Barends, R. Biswas, S. Boixo, F. G. Brandao, D. A. Buell,et al., Nature574, 505 (2019)

    2019

  47. [47]

    Ragone, B

    M. Ragone, B. N. Bakalov, F. Sauvage, A. F. Kemper, C. Ortiz Marrero, M. Larocca, and M. Cerezo, Nature Communications15, 7172 (2024)

    2024

  48. [48]

    Chamon, A

    C. Chamon, A. Hamma, and E. R. Mucciolo, Physical Review Letters112, 240501 (2014)

    2014

  49. [49]

    Z.-C. Yang, C. Chamon, A. Hamma, and E. R. Mucciolo, Physical Review Letters115, 267206 (2015)

    2015

  50. [50]

    Z.-C. Yang, A. Hamma, S. M. Giampaolo, E. R. Mucciolo, and C. Chamon, Physical Review B96, 020408 (2017)

    2017

  51. [51]

    Deshpande, M

    A. Deshpande, M. Hinsche, K. Najafi, K. Sharma, R. Sweke, and C. Zoufal, Dynamic parameterized quan- tum circuits: expressive and barren-plateau free (2024), arXiv:2411.05760

    work page arXiv 2024

  52. [52]

    Characterizing and Benchmarking Dynamic Quantum Circuits

    S. Shirgure, E. K¨ okc¨ u, A. Mitra, W. A. de Jong, C. Iancu, and S. Niu, Characterizing and benchmarking dynamic quantum circuits (2026), arXiv:2604.03360

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  53. [53]

    K. J. Sung, J. Yao, M. P. Harrigan, N. C. Rubin, Z. Jiang, L. Lin, R. Babbush, and J. R. McClean, Quantum Science and Technology5, 044008 (2020)

    2020

  54. [54]

    Volkoff and P

    T. Volkoff and P. J. Coles, Quantum Science and Tech- nology6, 025008 (2021)

    2021

  55. [55]

    X. Liu, G. Liu, H.-K. Zhang, J. Huang, and X. Wang, IEEE Transactions on Quantum Engineering5, 1 (2024)

    2024

  56. [56]

    Zlokapa, B

    A. Zlokapa, B. Villalonga, S. Boixo, and D. A. Lidar, npj Quantum Information9, 36 (2023)

    2023