Pith. sign in

REVIEW 3 major objections 4 minor 55 references

Interferometric Quantum Polynomial Chaos Expansion as a Generative Model for Calorimeter Shower Simulation

T0 review · 3 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read A single re-uploading circuit with 154 trained angles, no fitted classical parameters, generates calorimeter showers matching Geant4's densities and correlations, and proves by measurement that the entangling gates produce those…

desk verdict A genuinely fully quantum generative model with a sound verification protocol, but the hardware floor claim does not survive contact with the actual shot-noise structure. read the letter →

arxiv 2608.05405 v1 pith:OVHKN2YK submitted 2026-08-05 quant-ph

Jamal Slim , Saverio Monaco , Florian Rehm , Dirk Kruecker , Frank Gaede , Kerstin Borras This is my paper · ORCID
classification quant-ph
keywords quantumpolynomialchaosexpansionre-uploadingcircuitcalorimetershowersimulationgenerativemodelentanglementdependenceenergydistancetailexpectation-valuereadout
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that one small quantum circuit can be the entire generative model for calorimeter shower images, with random uniform germs as the only input and the gate angles as the only fitted parameters. Because the germs are re-uploaded at every one of the L blocks, the circuit depth becomes the order of a polynomial chaos expansion, so expressivity grows with depth rather than with classical coefficients. Trained on Geant4 shower data, the 8-qubit, 154-angle instance reproduces per-cell marginals and cross-cell dependencies with a rank-dependence effect size D = 0.991, and its verification protocol turns the model's own settings into measurements: with the entanglers off and the shared germ wire frozen, the cells become exactly independent, while the full model's conditional laws violate the CHSH bound at |S| = 2.76. Run unchanged on a superconducting processor, the circuit retains D = 0.873 with a shot-noise contraction predicted in advance. If right, this establishes that a quantum generator can carry the full joint structure of real detector data and can certify, by measurement rather than argument, which of its gates produced the learned correlations.

What carries the argument

The QPCE interferometric ansatz of Eq. (2): an n-qubit circuit of L blocks in which the same uniform random germs (epsilon, epsilon_0) are re-uploaded at the head of every block as R_y rotations, followed by diagonal layers of trainable R_z and R_ZZ phases separated by trainable R_y walls. Re-uploading makes each measured <Z_i> an exact degree-L trigonometric polynomial in every germ, so circuit depth plays the role of the truncation order of a polynomial chaos expansion. The two designed dependence channels—the R_ZZ couplers on a hardware-native path and the shared germ wire read by every qubit—are what the verification protocol strips away to certify where the correlations come from; the energy distance (equivalently 2 $MMD^{2}$ with the distance kernel) is the single loss whose exact gradients (three state evolutions via the adjoint) drive the fit.

What would settle it

Run the entangler-strip test on hardware: set the 42 RZZ phases to zero, freeze the shared germ at its median, and execute the deployed circuit; if the maximum pairwise rank correlation departs from the 0.019–0.025 sampling floor, or if the CHSH capability computed with two measurement bases fails to exceed 2, the claim that the dependence is generated by the entangling gates and that device error cannot inflate it would be directly refuted.

Watch

Extended reading notes

Core claim

The central claim is that the QPCE ansatz—a single re-uploading circuit driven solely by uniform random germs—is the entire generative model: the observed intensities are degree-L trigonometric chaos expansions of those germs, with L the circuit depth, and the gate angles are the only fitted parameters. Trained on 2600 Geant4 showers, the deployed 8-qubit, 154-angle instance reproduces the per-cell Wasserstein distances at 2.27×$10^{-3}$ and the rank-correlation structure at mean |rho_S| = 0.566 against 0.576 in the data, with dependence effect size D = 0.991 relative to a provable independence floor. The paper's distinctive result is attribution: setting the 42 RZZ phases to zero and freezing the shared germ wire leaves the cells exactly independent (Proposition 1), releasing the wire isolates the shared-latent channel, and the trained conditional laws violate the CHSH bound at |S| = 2.76, certifying that the entangling gates—not a classical shared input—produce the dependence. Executed unchanged on a superconducting processor, the circuit reaches D = 0.873 with a shot-noise contraction predicted in advance, while Proposition 4 states a no-go: no smooth expectation-value readout of absolutely continuous latents can produce asymptotic tail dependence strictly between 0 and 1, locating the architectural change needed for tail-heavy distributions.

Load-bearing premise

The hardware claims assume that device error beyond shot noise acts as a smooth, germ-independent, per-cell distortion that leaves rank statistics unchanged; if non-monotone, cross-cell, germ-dependent coherent error (e.g., crosstalk from untwirled fractional RZZ pulses) is present, the measured D=0.873 would not be a reliable floor on the dependence the device transports.

Editorial extensions

If this is right

  • Increasing circuit depth L raises the chaos expansion order covered by a fixed set of angles, so the model's expressivity can be tuned without adding any classical parameter.
  • The entangler-strip test provides a built-in, backend-agnostic control: every trained QPCE model can be verified, by running its own circuit at phi=0, to have zero cross-cell dependence when the entanglers are off—a certification hybrid classical-quantum models cannot supply.
  • The hardware protocol predicts the shot-noise contraction of every correlation from the trained model alone, so the error budget for a deployment can be set before any device runs.
  • Proposition 4 and its corollary imply that any smooth expectation-value generator of this family, at any depth and on any graph, has asymptotic tail dependence in {0,1}; reproducing measured intermediate tail asymmetry at finite quantile is possible, but reproducing it asymptotically requires a discrete common latent with comonotone branches.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The depth-to-order mechanism is not specific to calorimetry: any re-uploading observable in this ansatz class inherits exact finite chaos expansions, so the same 'depth is the truncation order' identity could serve other continuous-output generative tasks, with the entangler-strip test traveling with it.
  • The measured CHSH capability of 2.76 in a classically simulable model suggests that the classical/quantum distinction here is about the reachable set of conditional laws, not computational hardness; scaling to non-simulable circuits would make this a practical resource, but would also remove the exact reference that the current verification harness relies on.
  • If the shared-germ-wire design generalizes, the number of shared wires should follow the rank of the data's dependence structure (here essentially one global mode), so the model's resource count would scale with the physical collective modes of the target rather than with the number of cells.
  • The no-go theorem implies a concrete design rule for the next generation: to capture tail-heavy joint extremes, one should introduce a discrete common latent that makes the extreme-dominating branch comonotone; the paper's engineered branch example (weight 0.15, plateau at lambda_L ≈ 0.22) is a proof of concept that this rule works.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces the quantum polynomial chaos expansion (QPCE), a generative model in which a single re-uploading quantum circuit, driven only by i.i.d. uniform latent germs and a set of trained gate angles, produces continuous calorimeter cell intensities via Pauli-Z expectation values. The authors argue that re-uploading makes the circuit depth equal to the chaos expansion order, that cross-cell dependence is generated only by RZZ entanglers and one shared germ wire, and that the architecture admits built-in verification: with entanglers removed and the shared germ frozen, the cells are provably independent. They train the model on an 8-cell CLIC Geant4 dataset, report D=0.991 and mean absolute Spearman correlation 0.566 in noiseless simulation, deploy the identical circuit on the ibm fez processor with 500 germs at 1152 shots each, and further prove a no-go theorem on asymptotic tail dependence for smooth expectation-value readouts. The paper explicitly disclaims computational quantum advantage and states that the n=8 instance is exactly classically simulable.

Significance. If the central claims hold, the paper is a significant step for fully quantum generative models: it provides a model class in which the circuit is the entire generator, the learned dependence can be attributed to specific circuit elements by measurement rather than by argument, and a hardware deployment is accompanied by a pre-registered germ batch and an error budget. The manuscript is unusually careful in several respects: it validates on fresh germs against a held-out split, checks gradients against the parameter-shift rule to 1e-14, ships code and data, verifies the depth-order identity numerically, and states the classical simulability of the deployed instance without overclaiming. The main caveat, detailed below, is that the hardware floor reading of D relies on an independence assumption for shot noise that is violated by the single-setting readout protocol.

major comments (3)
  1. [Sec. IV F, Eq. (12); Sec. VI C/D] Proposition 3 assumes that the S-shot estimators \hat m_i and \hat m_j have errors that are independent across cells, and concludes that shot noise can only attenuate cross-cell dependence. For the deployed protocol in Algorithm 1, all n expectation values are estimated from the same S computational-basis bitstrings, so for a fixed germ the estimation errors satisfy Cov(e_i,e_j) = (1/S) E_eps[Cov_state(Z_i,Z_j)], which is generically nonzero for the entangled states this circuit prepares. The regression-dilution formula (12) is therefore not applicable, and correlated shot noise can inflate rather than only contract rank correlations when the signal and noise covariances share a sign. Consequently the predicted mean attenuation of 0.986, the deconvolved noiseless mean |rho_S|=0.5315, and the statement in Sec. VI D that device error 'cannot inflate' the measured dependence are unsupported. The hardware D=0.873 cannot be read as a lower bound on the dependence the device genuinely transports without either deriving the correct correlated-noise correction or changing the protocol to estimate each cell from independent shots.
  2. [Sec. VI D] The claim that the reported dependence statistics survive coherent error because 'rank statistics are insensitive to a smooth germ-independent bias field' is not justified. Proposition 2 covers only strictly monotone per-cell maps, which are rank-invariant; it does not cover the coherent over-rotation of fractional RZZ pulses identified in the reproduction test. A two-qubit coherent error of this kind is not a per-cell map and can change the rank structure in either direction, potentially inflating the measured dependence. Since this is precisely the channel that the reproduction test detects at chi^2/dof=2.42, the argument that D=0.873 is a floor on the transported dependence is not established. A numerical or analytical bound showing that the observed coherent error cannot increase the rank statistics would be needed to restore the floor reading.
  3. [Sec. V B] The reported confidence intervals are internally inconsistent: D=0.991 is quoted with interval [0.830,0.952], which does not contain the point estimate, and mean |rho_S|=0.566 is quoted with interval [0.552,0.558], which also excludes the point estimate. These appear to be swapped or mistyped bounds and should be corrected, since the reader cannot otherwise assess the precision of the headline simulation results.
minor comments (4)
  1. [Sec. V C and Sec. IX] The permutation test is reported as 'E=0.0137 at p=0.24' while the text says 'the model is statistically distinguishable from Geant4.' With p=0.24 the test does not detect a difference at conventional significance levels; if the intended p-value is 0.024, it should be corrected, and if the intended statement is that the model is not distinguishable, the wording should be changed.
  2. [Sec. IV C] There is a duplicated phrase: 'we report in Sec. IV C we report the measured decomposition they produce' should be reduced to a single 'we report.'
  3. [Sec. III C 5] The sentence 'The price is stated because the alternative that keeps the attribution unconditional...' is grammatically incomplete; 'the price is stated' needs a complement, e.g., 'The price is stated explicitly.'
  4. [Sec. IV C] The sentence beginning 'One bookkeeping rule is mandatory and easy to violate silently, is that...' should be rewritten; the current construction is not grammatical.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the main derivation and controls are self-contained; the flagged Proposition 3 issue is a correctness concern, not circularity.

full rationale

The paper's derivation chain is self-contained rather than circular. The depth-equals-order identity (Eq. 5) follows from the Fourier structure of re-uploading rotations and is verified numerically to machine precision; it is not fitted. Proposition 1 is proved directly from the product structure of the circuit when RZZ phases vanish, and the Verify protocol is implemented as parameter settings of the trained generator, not as an imported baseline. The CHSH capability (Sec. IV D) is an emergent measured property of the trained conditional law; no Bell term is present in the energy-distance loss, so the violation is not an artifact of the objective. Proposition 4 and Corollary 2 are proved within the paper for the whole smooth expectation-readout class and are used to identify limitations, not to assert success. The paper repeatedly disclaims classical simulability, absence of computational advantage, and the non-exportability of the gain calibration, all of which are consistent with a non-circular presentation. I also examined the skeptical attack on Proposition 3 (Sec. IV F versus Algorithm 1 step 9): the single-setting readout does make finite-shot estimators across cells correlated through the entangled state, so the claim that shot noise 'can only attenuate' and the interpretation of D=0.873 as a floor rest on an independence assumption that the protocol itself violates. That is a correctness risk in the error model, but it is not a case of a result being equivalent to its inputs by construction or by self-citation, so it does not raise the circularity score. The only mild metric-level observation is that the headline D is a normalized transform of the training MMD, but the paper also reports held-out Spearman, Wasserstein, and tail-asymmetry metrics that are independent of the loss kernel, so the central claim does not reduce to the fit.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

No new physical entities are postulated. The shared germ wire is an architectural latent variable, not a new force, particle, or dimension. The ledger lists the trained gate angles, the data-derived readout constants, and the hand-chosen architecture hyperparameters as the model's free parameters.

free parameters (3)
  • gate angles (beta, phi, theta, a, w) = 154 angles
    All trainable parameters, optimized with L-BFGS against the energy distance on a frozen germ batch (Sec. III F, Table II). These are the model parameters, not hidden constants.
  • readout constants (m_i, s_i) = 16 numbers: per-cell midpoint and half-range of training intensities
    Computed once from training data (Algorithm 1, line 2) and never revisited. Data-derived affine unit conversions; they affect the marginals but not rank-dependent metrics.
  • architecture hyperparameters (L, path topology, shared germ wire, shot count) = L=6, |E|=7 path, one shared germ, S=1152
    Hand-chosen design choices. The paper does not claim a validation-driven search, but the fit depends on them.
assumptions (6)
  • standard math An observable of a circuit that encodes a variable through r rotations is a trigonometric polynomial with at most r harmonics in that variable (Fourier analysis of re-uploading circuits).
    Invoked in Sec. III C 1 to prove depth equals chaos order (Eq. 5). Prior work: Schuld et al. 2021.
  • standard math Energy distance equals 2 MMD^2 for the distance kernel k(x,y)=||x||+||y||-||x-y||.
    Used in Sec. III F to define the loss. Prior work: Sejdinovic et al. 2013.
  • standard math CHSH inequality |S|<=2 holds for every classical local-response generative model with shared randomness.
    Used in Sec. IV D for the certification claim.
  • domain assumption The 8-cell CLIC calorimeter dataset is a representative, stationary sample of Geant4 showers, and the 65/35 split gives unbiased held-out evaluation.
    Sec. V A. If the dataset were not representative, the reported D and correlation matches would not transfer.
  • domain assumption The hardware residuals beyond shot noise are dominated by a smooth, germ-independent per-cell bias field (or by monotone per-cell gain/readout) that leaves rank statistics invariant.
    Sec. VI D. This underlies the claim that D=0.873 is a lower bound on transported dependence. Non-monotone cross-cell crosstalk would invalidate this reading.
  • standard math For the tail no-go theorem, the smooth maps attain their minima at non-degenerate critical points.
    Proposition 4 proof sketch uses Morse balls around interior critical points. The abstract's 'every smooth generator' omits this condition; boundary minima are not covered.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Interferometric Quantum Polynomial Chaos Expansion as a Generative Model for Calorimeter Shower Simulation." pith.science (2026). https://pith.science/paper/OVHKN2YK

@misc{pith2026260805405,
  author       = {Pith},
  title        = {Pith review of: Interferometric Quantum Polynomial Chaos Expansion as a Generative Model for Calorimeter Shower Simulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OVHKN2YK}},
  note         = {Machine review of arXiv:2608.05405}
}
read the original abstract

We present the quantum polynomial chaos expansion, a generative algorithm in which a single circuit is the entire model, and we use it to learn calorimeter images. In a classical chaos expansion the randomness is the input and the coefficients are fitted. Here the randomness is still the only input, entering the circuit as rotation angles and re-uploaded at every block, so that each measured observable is a chaos expansion of the latent variables whose order equals the circuit depth, and what is fitted are the gate angles themselves. Expressivity therefore grows with depth rather than with classical coefficients, correlations between outputs arise only from entangling gates, and a single latent wire read by all qubits carries the collective mode of the data. Nothing fitted stands between the circuit and the sample, so switching the entanglers off is a setting of the model itself and provably yields independent outputs, and attribution of the learned correlations to individual gates becomes a measurement. Choosing between two measurement bases shot by shot sharpens attribution into certification, and the trained model violates the Bell bound obeyed by every classical generative model with local response, whatever its size. We train the model on Geant4 shower data, execute the identical circuit on a superconducting processor with its accuracy loss predicted in advance, prove a no-go theorem for the tail dependence of every smooth generator read out through expectation values, and identify the circuit primitive that removes this limit.

Figures

Figures reproduced from arXiv: 2608.05405 by the authors.

Figure 1
Figure 1. FIG. 1: QPCE drawn as it executes. The upper row is the entire deployed model. Germs enter the interferometer, [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: QPCE interferometer, drawn for [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Deployed coupler graph and the one-to-one map [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4: Marginal intensity distributions for the eight calorimeter cells. Geant4 reference (black points, [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Total deposited energy [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Spearman rank-correlation structure. Geant4 reference, deployed model in noiseless simulation, and [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Tail behaviour, the regime where model families separate (Sec. [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 40 canonical work pages

  1. [1]

    Germ re-uploading sets the chaos order The decisive structural element is that the germ rota- tion openseveryblock rather than only the first. By the Fourier structure of variational circuits [41], an observ- able of a circuit that encodes a germ throughrrotations is a trigonometric polynomial with at mostrharmonics in that germ. Encoded once, every⟨Z i⟩i...

    work page

  2. [2]

    Noise amplitudes Using a trainable encoding scale isriskyunder any dependence-sensitive objective. For instance, shrinking a private noise amplitude increases output correlation monotonically, so a correlation-only loss would drive it to zero and produce near-perfect correlations by collaps- ing the generative diversity. In the private-only family we ther...

    work page

  3. [3]

    RZZ (ϕ) is a continuous entangler withϕ= 0 givingex- actseparability, conditional on the shared germ, and cor- relation growing smoothly with|ϕ|

    Continuous couplers rather than CNOTs A CNOT is maximally entangling and has no dial, so a model family built on it contains no controlled null. RZZ (ϕ) is a continuous entangler withϕ= 0 givingex- actseparability, conditional on the shared germ, and cor- relation growing smoothly with|ϕ|. This is not merely convenient for optimization, it is actually wha...

    work page

  4. [4]

    An IQP circuit is a sin- gle block of gates diagonal in one basis conjugated by fixed mixing layers,U=H ⊗nD H⊗n, equivalently a cir- cuit whose gates all commute

    Circuit class and simulability The circuit class of the ansatz determines which com- plexity results do and do not pertain to it, so we state it exactly and verify it directly. An IQP circuit is a sin- gle block of gates diagonal in one basis conjugated by fixed mixing layers,U=H ⊗nD H⊗n, equivalently a cir- cuit whose gates all commute. The ansatz of Eq....

    work page

  5. [5]

    This is a global factor with a front/back sign flip, physi- cally the shower-start depth

    One shared germ wire The dataset’s rank-correlation matrix has one eigen- value carrying 64.1% of the dependence, with eigenvector (−0.33,−0.40,−0.42,−0.38,−0.16,+0.28,+0.40,+0.38). This is a global factor with a front/back sign flip, physi- cally the shower-start depth. A rank-one reconstruction from that mode alone reproduces all 28 off-diagonal correla...

  6. [6]

    Agostinelliet al., Nucl

    S. Agostinelliet al., Nucl. Instrum. Methods Phys. Res. A506, 250 (2003)

    2003

  7. [7]

    Allisonet al., IEEE Trans

    J. Allisonet al., IEEE Trans. Nucl. Sci.53, 270 (2006)

    2006

  8. [8]

    ATLAS Collaboration, Eur. Phys. J. C70, 823 (2010)

    work page 2010

Show all 55 references
  1. [9]

    ATLAS Collaboration, CERN-LHCC-2020-015 (2020)

    2020

  2. [10]

    Paganini, L

    M. Paganini, L. de Oliveira, and B. Nachman, Phys. Rev. D97, 014021 (2018)

    2018

  3. [11]

    Erdmannet al., Comput

    M. Erdmannet al., Comput. Softw. Big Sci.2, 4 (2018)

    2018

  4. [12]

    Paganini, L

    M. Paganini, L. de Oliveira, and B. Nachman, Phys. Rev. Lett.120, 042003 (2018)

    2018

  5. [13]

    Krause and D

    C. Krause and D. Shih, arXiv:2106.05285 (2021)

    2021 arXiv

  6. [14]

    Krause and D

    C. Krause and D. Shih, Phys. Rev. D107, 113003 (2023)

    2023

  7. [15]

    Mikuni and B

    V. Mikuni and B. Nachman, Phys. Rev. D106, 092009 (2022)

    2022

  8. [16]

    Mikuni and B

    V. Mikuni and B. Nachman, arXiv:2308.03847 (2023)

    2023

  9. [17]

    Lindskog, A

    F. Lindskog, A. McNeil, and U. Schmock, Kendall’s tau for elliptical distributions, inCredit Risk: Measure- ment, Evaluation and Management(Physica-Verlag, Hei- delberg, 2003), p. 149

    2003

  10. [18]

    R. F. Werner, Quantum states with Einstein-Podolsky- Rosen correlations admitting a hidden-variable model, Phys. Rev. A40, 4277 (1989)

    1989

  11. [19]

    Horodecki, P

    R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Quantum entanglement, Rev. Mod. Phys. 81, 865 (2009)

    2009

  12. [20]

    Coyle, D

    B. Coyle, D. Mills, V. Danos, and E. Kashefi, The Born supremacy: quantum advantage and training of an Ising Born machine, npj Quantum Inf.6, 60 (2020)

    2020

  13. [21]

    Quetzalcoatl Toledo-Marinet al., Conditioned quantum-assisted deep generative surrogate for particle- calorimeter interactions, npj Quantum Inf

    J. Quetzalcoatl Toledo-Marinet al., Conditioned quantum-assisted deep generative surrogate for particle- calorimeter interactions, npj Quantum Inf. (2025), arXiv:2410.22870

    2025 arXiv

  14. [22]

    F. Rehm, S. Vallecorsa, M. Grossi, K. Borras, and D. Kr¨ ucker, A full quantum generative adversarial net- work model for high energy physics simulations, Quan- tum Sci. Technol.9, 015009 (2024)

    2024

  15. [23]

    F. J. Schreiber, J. Eisert, and J. J. Meyer, Classical surrogates for quantum learning models, Phys. Rev. Lett. 131, 100803 (2023)

    2023

  16. [24]

    Huang, M

    H.-Y. Huang, M. Broughton, M. Mohseni, R. Babbush, S. Boixo, H. Neven, and J. R. McClean, Power of data in quantum machine learning, Nat. Commun.12, 2631 (2021)

    2021

  17. [25]

    Hoqueet al., CaloQV AE: Simulating high-energy particle-calorimeter interactions using hybrid quantum- classical generative models, Eur

    S. Hoqueet al., CaloQV AE: Simulating high-energy particle-calorimeter interactions using hybrid quantum- classical generative models, Eur. Phys. J. C84(2024)

    2024

  18. [26]

    S. Y. Chang, S. Herbert, S. Vallecorsa, E. F. Combarro, and R. Duncan, Dual-parameterized quantum circuit GAN model in high energy physics, EPJ Web Conf.251, 03050 (2021)

    2021

  19. [27]

    Liu and L

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

    2018

  20. [28]

    Benedettiet al., npj Quantum Inf.5, 45 (2019)

    M. Benedettiet al., npj Quantum Inf.5, 45 (2019)

    2019

  21. [29]

    Zoufal, A

    C. Zoufal, A. Lucchi, and S. Woerner, npj Quantum Inf. 5, 103 (2019)

    2019

  22. [30]

    Lloyd and C

    S. Lloyd and C. Weedbrook, Phys. Rev. Lett.121, 040502 (2018)

    2018

  23. [31]

    Schuld and N

    M. Schuld and N. Killoran, PRX Quantum3, 030101 (2022)

    2022

  24. [32]

    J. M. K¨ ubler, S. Buchholz, and B. Sch¨ olkopf, inAdvances in Neural Information Processing Systems 34(Curran Associates, 2021), p. 12661

    2021

  25. [33]

    Bowles, S

    J. Bowles, S. Ahmed, and M. Schuld, arXiv:2403.07059 (2024)

    2024 arXiv

  26. [34]

    Joe,Dependence Modeling with Copulas(Chapman & Hall/CRC, Boca Raton, 2014)

    H. Joe,Dependence Modeling with Copulas(Chapman & Hall/CRC, Boca Raton, 2014)

    2014

  27. [35]

    Bedford and R

    T. Bedford and R. M. Cooke, Ann. Statist.30, 1031 (2002)

    2002

  28. [36]

    Embrechts, A

    P. Embrechts, A. McNeil, and D. Straumann, inRisk Management: Value at Risk and Beyond, edited by M. A. H. Dempster (Cambridge University Press, 2002), p. 176

    2002

  29. [37]

    Wiener, Am

    N. Wiener, Am. J. Math.60, 897 (1938)

    1938

  30. [38]

    Xiu and G

    D. Xiu and G. E. Karniadakis, SIAM J. Sci. Comput.24, 619 (2002)

    2002

  31. [39]

    Xiu,Numerical Methods for Stochastic Computations (Princeton University Press, 2010)

    D. Xiu,Numerical Methods for Stochastic Computations (Princeton University Press, 2010)

    2010

  32. [40]

    Mitaraiet al., Phys

    K. Mitaraiet al., Phys. Rev. A98, 032309 (2018)

    2018

  33. [41]

    Schuldet al., Phys

    M. Schuldet al., Phys. Rev. A99, 032331 (2019)

    2019

  34. [42]

    Grettonet al., J

    A. Grettonet al., J. Mach. Learn. Res.13, 723 (2012)

    2012

  35. [43]

    Romero and A

    J. Romero and A. Aspuru-Guzik, Variational quan- tum generators: Generative adversarial quantum ma- chine learning for continuous distributions, Adv. Quan- tum Technol.4, 2000003 (2021)

    2021

  36. [44]

    Barthe, M

    A. Barthe, M. Grossi, S. Vallecorsa, J. Tura, and V. Dun- jko, Parameterized quantum circuits as universal gen- erative models for continuous multivariate distributions, arXiv:2402.09848 (2024)

    2024 arXiv

  37. [45]

    Aftab, C

    J. Aftab, C. Schwab, H. Yang, and J. Zech, Quan- tum circuit encodings of polynomial chaos expansions, arXiv:2506.01811 (2025)

    2025

  38. [46]

    Schuld, R

    M. Schuld, R. Sweke, and J. J. Meyer, Effect of data encoding on the expressive power of variational quan- tum machine-learning models, Phys. Rev. A103, 032430 (2021)

    2021

  39. [47]

    G. J. Sz´ ekely and M. L. Rizzo, Energy statistics: A class of statistics based on distances, J. Stat. Plan. Inference 143, 1249 (2013)

    2013

  40. [48]

    Baringhaus and C

    L. Baringhaus and C. Franz, On a new multivariate two- sample test, J. Multivar. Anal.88, 190 (2004)

    2004

  41. [49]

    Sejdinovic, B

    D. Sejdinovic, B. Sriperumbudur, A. Gretton, and K. Fukumizu, Equivalence of distance-based and RKHS- based statistics in hypothesis testing, Ann. Statist.41, 2263 (2013)

    2013

  42. [50]

    Jones and J

    T. Jones and J. Gacon, Efficient calculation of gradi- ents in classical simulations of variational quantum algo- rithms, arXiv:2009.02823 (2020)

    2020 arXiv

  43. [51]

    Monaco, J

    S. Monaco, J. Slim, K. Borras, and D. Kr¨ ucker, desyqml/clic: Initial public release, version v1.0.0, Zen- odo (2025), DOIhttps://doi.org/10.5281/zenodo. 16027525

    2025 doi

  44. [52]

    Jurcevicet al., Quantum Sci

    P. Jurcevicet al., Quantum Sci. Technol.6, 025020 (2021)

    2021

  45. [53]

    IBM Quantum, Qiskit Runtime Primitives docu- mentation,https://docs.quantum.ibm.com/api/ qiskit-ibm-runtime(2024)

    2024

  46. [54]

    IBM Quantum,https://quantum.ibm.com/services/ resources(2024)

    2024

  47. [55]

    Slim,QPCE: fully quantum interferometric polyno- mial chaos expansion, GitHub repository (2026),https: //github.com/jamalslim/qpce_code

    J. Slim,QPCE: fully quantum interferometric polyno- mial chaos expansion, GitHub repository (2026),https: //github.com/jamalslim/qpce_code. 20 Appendix A: V alidation of the Numerical Pipeline Every result rests on a batched statevector engine, on histogram error bars, and on ...

    2026

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.