Pith. sign in

REVIEW 4 major objections 7 minor 37 references

Partitioned Hybrid Quantum Fourier Neural Operators for Scientific Quantum Machine Learning

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A partitioned quantum-classical Fourier operator matches or beats its classical counterpart on fluid benchmarks.

desk verdict Honest engineering extension of QFNO with real architectural novelty, but the accuracy and robustness claims are not yet isolated from capacity and normalization confounds. read the letter →

arxiv 2507.08746 v1 pith:OOFS5CH6 submitted 2025-07-11 cs.LG quant-ph

classification cs.LGquant-ph
keywords quantummachinelearningFourierneuraloperatorhybridquantum-classicalcomputingNavier-StokesequationsBurgersequationunaryencodingvariationalcircuits
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 introduces PH-QFNO, a hybrid quantum-classical version of the Fourier Neural Operator that splits the spectral computation between a quantum circuit and a classical network. The paper claims that on an $8\times8$ incompressible Navier-Stokes benchmark the fully quantum model and the 33% hybrid model end training with lower relative error than a classical FNO baseline, while on Burgers' equation the hybrid reaches lower error than the classical baseline and the fully quantum model matches the one-mode classical model. The central design is to partition the Fourier-transformed field into $4\times4\times4$ sub-blocks, process each with a unary-encoded Quantum Fourier Transform and variational orthogonal gates, and stitch the results with a global classical convolution. The authors also report that the quantum Fourier layer keeps its outputs closer to clean outputs under Gaussian input noise than the classical layer does. A sympathetic reader would care because this is a concrete test of whether quantum Fourier layers can contribute to PDE operator learning under realistic qubit limits.

What carries the argument

The central object is the PH-QFNO block, a partitioned spectral layer that combines a quantum branch and a classical branch. Its load-bearing steps are unary encoding, which maps a vector or matrix onto one-hot computational basis states using RBS rotations; a butterfly-structured unary Quantum Fourier Transform that realizes the discrete Fourier transform on those amplitudes; and a variational quantum learning gate built from orthogonal layers, which plays the role of the classical FNO weight matrices $W_K$. The algebraic core is the sequential-gate identity $CZ\,P^{\dagger}(-\theta)\,CZ\,P(\theta)=I$ (with a double-controlled version in 3D), which lets the learning block alter only the selected Fourier modes while preserving the unary basis. A final global classical convolutional layer $W$ is what reconnects the $4\times4\times4$ sub-blocks and, together with the fourfold parameter increase, is the mechanism the paper relies on to compensate for the truncated low-frequency spectrum.

What would settle it

Run the same PH-QFNO comparison on incompressible Navier-Stokes at $16\times16$ or $64\times64$ resolution while keeping the quantum partition fixed at $4\times4\times4$; if the relative error of PH-QFNO rises relative to the classical FNO as resolution grows, the spectral-gap compensation does not generalize.

Watch

Extended reading notes

Core claim

The paper's central claim is that a Fourier Neural Operator can be partitioned across quantum and classical resources without losing accuracy, and that on the reported incompressible Navier-Stokes problem the partitioned model gains accuracy over the classical FNO. The architecture encodes each sub-block of the field into a unary one-hot quantum state, applies a Quantum Fourier Transform in the unary basis, multiplies the selected low Fourier modes by a parameterized orthogonal quantum learning gate, then returns to the spatial domain and recombines the sub-blocks through a global classical convolutional layer. The paper attributes the Navier-Stokes result to the fact that four independent quantum sub-operators replace one classical operator, increasing the parameter count fourfold and giving the network enough capacity to fill in frequencies lost when the $8\times8$ spectrum is cut into $4\times4\times4$ pieces. On Burgers' equation, the same recipe reaches errors comparable to the one-mode classical FNO while avoiding that model's overfitting, and the fully quantum model converges earlier in training.

Load-bearing premise

The central claim collapses if the low-frequency modes lost by cutting the $8\times8$ Fourier spectrum into $4\times4\times4$ sub-blocks are not, in general, repaired by one global classical convolution plus a fourfold parameter increase, or if the $8\times8$ grid is too coarse to represent the Navier-Stokes dynamics in the $t=30$ to 31 window.

Editorial extensions

If this is right

  • On the tested 8-point Burgers benchmark, PH-QFNO recovers classical FNO accuracy and the hybrid configurations reach lower relative error than the classical baseline.
  • On the $8\times8$ incompressible Navier-Stokes benchmark, the fully quantum model and the 33% hybrid model both end training with lower relative error than the classical FNO.
  • The quantum Fourier layer is reported to be more stable than the classical spectral layer when inputs are perturbed with Gaussian noise, especially as the noise standard deviation grows.
  • Because the quantum branch uses about $\frac{d_v}{2}\log d_v$ parameters per mode instead of $d_v^2$, the model trades raw expressiveness for faster convergence and stability.

Reading between the lines

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

  • The accuracy gain on Navier-Stokes may come from the fourfold parameter increase of running four independent $4\times4\times4$ sub-operators rather than from anything quantum; a purely classical FNO with the same partitioned topology would isolate that factor, and this test is not reported in the paper.
  • The 12-qubit unary encoding caps the spatial resolution at 64 cells, so extending the approach to the standard $64\times64$ or $128\times128$ FNO benchmarks would require either more qubits or a different encoding; the paper does not show such a scaling test.
  • The reported noise robustness suggests a use case the paper does not pursue: operator learning from noisy experimental or sensor data, where classical FNOs would be the comparison baseline.
  • A natural next extension is to replace the learned weights with a physics-informed loss; the paper lists this as future work.
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

4 major / 7 minor

Summary. The paper introduces the Partitioned Hybrid Quantum Fourier Neural Operator (PH-QFNO), which generalizes the Quantum Fourier Neural Operator by partitioning Fourier computations across classical and quantum branches, with tunable hybridization, MPI-based distributed execution, unary encoding, and quantum orthogonal layers. The method is implemented in PennyLane with PyTorch and evaluated on Burgers' equation, incompressible Navier-Stokes at 8x8 resolution, and a compressible Navier-Stokes Kelvin-Helmholtz setup. The central claims are that PH-QFNO recovers classical FNO accuracy, achieves higher accuracy than classical FNO on incompressible Navier-Stokes, and exhibits improved robustness to input noise. The architecture and implementation are described in detail, but the quantitative evidence for the comparative claims is incomplete and the comparisons contain capacity and normalization confounds that the manuscript itself partially acknowledges.

Significance. If the empirical claims were fully established, PH-QFNO would be a useful step toward scalable hybrid quantum-classical neural operators, extending QFNO-style methods to higher-dimensional PDE benchmarks and offering tunable quantum-classical resource allocation. The paper has concrete strengths: it provides explicit circuits for 2D/3D unary encoding and quantum Fourier transforms, a PennyLane/PyTorch implementation with MPI data parallelism, and evaluations on multiple PDE test cases. However, the current significance is conditional. The central comparisons do not isolate the quantum Fourier mechanism from parameter count, preprocessing normalization, or output postprocessing, and no final quantitative errors or variability measures are reported. The claims as stated in the abstract therefore require substantially stronger experimental support before they can be accepted.

major comments (4)
  1. [Section V, Figs. 7 and 10] No final test relative errors, standard deviations, or seed information are reported for any experiment. The claims that PH-QFNO "recovers classical FNO accuracy" and "outperforms" classical FNO are supported only by training curves in figures, so the magnitudes of the reported gains and their statistical reliability cannot be assessed. Please report final relative errors using Eq. (18) with multiple random seeds, error bars, and the corresponding test-set standard deviations.
  2. [Section III-D2 and Section V] The manuscript explicitly attributes the two-dimensional Navier-Stokes accuracy gain to a higher parameter count: Section III-D2 states that deploying four independent quantum Fourier operators increases the total parameter count fourfold, and Section V states that the quantum model's higher accuracy "may be attributed to the fact that the two-dimensional quantum model is designed with a higher number of parameters than its classical counterpart." This is a capacity confound: the comparison against classical FNO does not control for model expressiveness, so the reported accuracy advantage does not establish that the quantum Fourier mechanism is responsible. A matched-capacity classical version of the same partitioned architecture, or a classical baseline with the same number of parameters, is required before the abstract's "higher accuracy" claim can be supported.
  3. [Section IV-B, Eq. (19), and Fig. 11] The robustness comparison between the quantum and classical Fourier layers is not controlled. The quantum path includes input normalization via unary encoding and produces bounded probability amplitudes after measurement, while the classical spectral convolution has no equivalent normalization or output nonlinearity; the parameter counts also differ. Consequently, the improved similarity reported in Fig. 11 may reflect input normalization and output boundedness rather than any quantum-specific property. To support the claimed stability advantage, compare layers with matched parameter counts, identical input preprocessing, and identical output normalization, or report the internal quantities before normalization.
  4. [Section III-D2] The paper acknowledges that partitioning the Fourier operator into 4x4xd_v sub-blocks "inevitably induces spatial under-sampling on an 8 x 8 x dv tensor, effectively truncating the lowest-frequency modes." The proposed compensation—a global classical convolution plus a fourfold parameter increase—is validated only on the 8-point/8x8 downsampled benchmarks at one viscosity (nu=1e-3). This leaves open whether the mechanism transfers to standard FNO resolutions such as 64x64 or 128x128. Please either add experiments at higher resolutions or explicitly restrict the accuracy and stability claims to the downsampled regime studied here.
minor comments (7)
  1. [Section III-E] There is a typo: "we adot an MPI-based data parallelism strategy" should read "we adopt an MPI-based data parallelism strategy."
  2. [Section IV] The acronym is inconsistently rendered as "HP-QFNO" in the compressible Navier-Stokes paragraph; it should be "PH-QFNO" throughout.
  3. [Section III-D2] The two-dimensional PH-QFNO process is said to be "depicted in Fig. 6a," but the two-dimensional pipeline appears in Fig. 6b; the cross-reference should be corrected.
  4. [Section IV-B, Eq. (19)] In Eq. (19), the notation for the noisy output is introduced as \tilde x, but the sentence describes "the output generated from the same input with added noise"; please make the definition of noisy output explicit and state that B=5 is the batch size used for the reported similarity values.
  5. [Fig. 11] The heatmaps would benefit from a colorbar with numerical similarity values; the qualitative statement "brighter colors indicate higher similarity" is insufficient for quantitative assessment.
  6. [Table I] The entries "4 \times 3" and "4 \times 2" in the dQ_v column are unclear; please clarify whether these denote three or two sequential quantum submatrices, and define the notation in the table caption.
  7. [Throughout] No code or data availability statement is provided. To support reproducibility, please include a repository link or an appendix detailing the exact hyperparameters, training configurations, and dataset preprocessing steps.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central accuracy and robustness claims are empirical comparisons against re-implemented classical baselines, and the cited prior work is external or non-load-bearing.

full rationale

The paper's derivation chain is not circular. The unary encoding construction (Sec. II-B, Eqs. 1-4) is a standard parameterization that explicitly reproduces the input amplitudes; it does not define the target output in terms of itself. The quantum Fourier layer is presented as a gate-level implementation of the Cooley-Tukey butterfly, and any mismatch would be a correctness issue, not a circular reduction. The partitioning step (Sec. III-D2) openly concedes that spectral truncation occurs and proposes two compensatory mechanisms (a classical convolutional layer and a fourfold parameter increase); those mechanisms are then empirically tested, which is a legitimate experimental design even if the baseline is not capacity-matched. The accuracy comparisons use relative error (Eq. 18), an external metric, against a classical FNO re-implementation; the admitted parameter-count difference is a confounding variable but does not make the reported metric equal to the model definition. The noise-robustness analysis (Sec. IV-B, Eq. 19) defines a cosine-similarity measure and evaluates it on both quantum and classical layers; the quantum layer's built-in normalization is an architectural property, not a fitted parameter renamed as a prediction. Self-citations [1], [4], [7], [33], [37] appear only in background or related-work contexts and are not load-bearing for the paper's central claims. No equation is shown to reduce to its own input, and no fitted value is relabeled as a prediction.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The variational circuit weights are trained parameters and are not free parameters in the ledger sense. The hand-chosen quantities that shape the results are the resolution, the quantum block size, the latent width d_v, the mode count K, and the similarity batch size. The main axioms are inherited standard results (unary encoding and unary QFT), the fidelity of the in-house FNO baseline, the adequacy of the tiny grids, and the ad hoc compensation premise for the truncated spectrum. No invented physical entities are introduced; the paper adds circuit architecture, not new physics.

free parameters (5)
  • spatial resolution (8 points for Burgers, 8x8 for Navier-Stokes) = 8 / 8x8
    The domain is downsampled to fit the 12-qubit cap of the PennyLane simulator; all accuracy claims depend on this choice.
  • quantum partition block size = 4x8 (1D), 4x4x4 (2D)
    Chosen so each sub-block fits within 12 qubits; it determines the spectral truncation that the method compensates classically.
  • latent width d_v and quantum width d^Q_v = d_v in {4, 12}, d^Q_v = 4
    Controls the hybridization ratio; the Navier-Stokes accuracy gain is attributed to the parameter increase from four quantum sub-blocks, so d_v is entangled with the headline result.
  • number of Fourier modes K = K = 1, 5 for Burgers; K unspecified for 2D cases
    K sets the expressiveness of both branches; exact K per 2D experiment is not reported.
  • batch size for similarity metric B = 5
    The noise-robustness metric is averaged over only 5 samples without error bars or repetition counts.
assumptions (6)
  • standard math Correctness of the unary QFT and of the row-wise Fourier transformations (Eqs. 6-10).
    The quantum butterfly implementation is inherited from [9]; the paper does not re-derive it, and Eq. (10) is internally inconsistent with the surrounding text about which registers are transformed.
  • standard math The unary encoding achieves the exact amplitudes stated (Eqs. 3-4) for an arbitrary vector.
    Relies on the unitarity of RBS gates and the recursive angle definitions in Eq. (2); this encoding is the starting point of the whole pipeline.
  • domain assumption The in-house classical FNO re-implementation is a faithful and fair comparator.
    All accuracy claims are relative to an in-house classical FNO; no absolute errors are given to anchor against published FNO numbers.
  • domain assumption 8-point and 8x8 grids faithfully represent the Burgers and Navier-Stokes (nu=1e-3) dynamics, including the t=30 to 31 window described as more chaotic.
    Sec. IV states the downsampling explicitly; the adequacy of the dynamics at this grid is not demonstrated.
  • ad hoc to paper The spectral truncation from partitioning into 4x4x4 blocks is compensated by a global classical convolution plus parameter over-provisioning.
    Sec. III-D2 states this as the method's compensation mechanism; it is a design assertion validated only on the three small benchmarks.
  • domain assumption Noiseless state-vector simulation is a valid proxy for the quantum execution claims.
    All quantum circuits are simulated with PennyLane's default.qubit; real-device and noise-model study is deferred to future work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Partitioned Hybrid Quantum Fourier Neural Operators for Scientific Quantum Machine Learning." pith.science (2026). https://pith.science/paper/OOFS5CH6

@misc{pith2026250708746,
  author       = {Pith},
  title        = {Pith review of: Partitioned Hybrid Quantum Fourier Neural Operators for Scientific Quantum Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OOFS5CH6}},
  note         = {Machine review of arXiv:2507.08746}
}
read the original abstract

We introduce the Partitioned Hybrid Quantum Fourier Neural Operator (PHQFNO), a generalization of the Quantum Fourier Neural Operator (QFNO) for scientific machine learning. PHQFNO partitions the Fourier operator computation across classical and quantum resources, enabling tunable quantum-classical hybridization and distributed execution across quantum and classical devices. The method extends QFNOs to higher dimensions and incorporates a message-passing framework to distribute data across different partitions. Input data are encoded into quantum states using unary encoding, and quantum circuit parameters are optimized using a variational scheme. We implement PHQFNO using PennyLane with PyTorch integration and evaluate it on Burgers' equation, incompressible and compressible Navier-Stokes equations. We show that PHQFNO recovers classical FNO accuracy. On incompressible Navier-Stokes, PHQFNO achieves higher accuracy than its classical counterparts. Finally, we perform a sensitivity analysis under input noise, confirming improved stability of PHQFNO over classical baselines.

Figures

Figures reproduced from arXiv: 2507.08746 by the authors.

Figure 1
Figure 1. Quantum circuit for one-dimensional encoding with four qubits. An [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Two-dimensional unary encoding for matrix ⊗ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Quantum circuit designed to carry out the Fourier Transform in the [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (9 more)
Figure 5
Figure 5. Figure 5: Hence, after this computation, we get: 1 ∥Y ∥ (QˆF T ⊗ QˆF T ⊗ I)   m,n,k X i,j,l=1 y l ⌊i,j⌋ |ei⟩ |ej ⟩ |el⟩   = 1 ∥Y ∥   m,n,k X i,j,l=1 yˆ l i,j |ei⟩ |ej ⟩ |el⟩   . (10) which correctly performs the one dimensional QFT over both the dimensions of the k matri…
Figure 5
Figure 5. Figure 5: Three-dimensional Sequential-QFL: The diagram comprises three registers containing [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Overview of the PH-QFNO architectures: (a) One-dimensional and (b) Two-dimensional PH-QFNO. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Relative error comparisons on 1D and 2D PDEs across different model types. In (a) and (c), the dimension of the latent representation of the input [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Predictions of the u profile obtained by FNO and PH-QFNO at two different epochs during the training. The parameter K is set to 5 for both the models. Furthermore, a third experiment has been conducted. As shown in [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Top panel: Contourplot of the u evolution for 1D Burger with a sinusoidal initial condition. Bottom panel: Comparison between ground truth (blue line), classical prediction (dashed green line) and fully quantum model (dashed red line) over the last two times steps for …
Figure 10
Figure 10. Figure 10: Vorticity field of the solution for the incompressible Navier-Stokes [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Heatmaps showing the similarity as a function of input noise [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: Left Panel: Relative error variation over training epochs, with the classical model shown in light blue and the 50% hybrid quantum model in green, both with K = 3. Right Panel: Evolution of the density field for the two-dimensional Kelvin–Helmholtz equation across mul…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 23 canonical work pages

  1. [1]

    Beyond the buzz: Strategic paths for enabling useful NISQ applications,

    P. R. Hegde, O. Kyriienko, H. Heimonen, P. Tolias, G. Netzer, P. Bark- outsos, R. Vinuesa, I. Peng, and S. Markidis, “Beyond the buzz: Strategic paths for enabling useful NISQ applications,” in Proceedings of the 21st ACM International Conference on Computing Frontiers, 2024, pp. 310– 313

  2. [2]

    Quantum algorithm for linear systems of equations,

    A. W. Harrow, A. Hassidim, and S. Lloyd, “Quantum algorithm for linear systems of equations,” Physical review letters , vol. 103, no. 15, p. 150502, 2009

  3. [3]

    M. A. Nielsen and I. L. Chuang, Quantum computation and quantum information. Cambridge university press, 2010

  4. [4]

    What is quantum parallelism, anyhow?

    S. Markidis, “What is quantum parallelism, anyhow?” in ISC High Performance 2024 Research Paper Proceedings (39th International Conference). Prometeus GmbH, 2024, pp. 1–12

  5. [5]

    Quantum computing in the NISQ era and beyond,

    J. Preskill, “Quantum computing in the NISQ era and beyond,” Quan- tum, vol. 2, p. 79, 2018

  6. [6]

    Variational quantum algorithms,

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio et al. , “Variational quantum algorithms,” Nature Reviews Physics , vol. 3, no. 9, pp. 625– 644, 2021

  7. [7]

    Programming quantum neural networks on nisq systems: An overview of technologies and methodologies,

    S. Markidis, “Programming quantum neural networks on nisq systems: An overview of technologies and methodologies,” Entropy, vol. 25, no. 4, p. 694, 2023

  8. [8]

    Potential of quantum scientific machine learning applied to weather modeling,

    B. Jaderberg, A. A. Gentile, A. Ghosh, V . E. Elfving, C. Jones, D. V odola, J. Manobianco, and H. Weiss, “Potential of quantum scientific machine learning applied to weather modeling,” Physical Review A, vol. 110, no. 5, p. 052423, 2024

Show all 37 references
  1. [9]

    Quantum Fourier networks for solving parametric PDEs,

    N. Jain, J. Landman, N. Mathur, and I. Kerenidis, “Quantum Fourier networks for solving parametric PDEs,” 2023. [Online]. Available: https://arxiv.org/abs/2306.15415

  2. [10]

    Fourier neural operator for parametric partial differential equations,

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Fourier neural operator for parametric partial differential equations,” arXiv preprint arXiv:2010.08895 , 2020

  3. [11]

    Deep neural network modeling for CFD simulations: Benchmarking the Fourier neural operator on the lid- driven cavity case,

    P. A. Costa Rocha, S. J. Johnston, V . Oliveira Santos, A. A. Aliabadi, J. V . G. Th ´e, and B. Gharabaghi, “Deep neural network modeling for CFD simulations: Benchmarking the Fourier neural operator on the lid- driven cavity case,” Applied Sciences, vol. 13, no. 5, p. 3165, 2023

  4. [12]

    Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,

    V . Bergholm, J. Izaac, M. Schuld, C. Gogolin, S. Ahmed, V . Ajith, M. S. Alam, G. Alonso-Linaje, B. AkashNarayanan, A. Asadi et al. , “Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,” arXiv preprint arXiv:1811.04968 , 2018

  5. [13]

    Neural operator: Learning maps between function spaces with applications to PDEs,

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, and A. Anandkumar, “Neural operator: Learning maps between function spaces with applications to PDEs,” Journal of Machine Learning Research , vol. 24, no. 89, pp. 1–97, 2023. [Online]. Available: http:...

  6. [14]

    Quantum Vision Transformers,

    E. A. Cherrat, I. Kerenidis, N. Mathur, J. Landman, M. Strahm, and Y . Y . Li, “Quantum Vision Transformers,” Quantum, vol. 8, p. 1265, Feb. 2024

  7. [15]

    Nearest centroid classification on a trapped ion quantum computer,

    S. Johri, S. Debnath, A. Mocherla, A. SINGK, A. Prakash, J. Kim, and I. Kerenidis, “Nearest centroid classification on a trapped ion quantum computer,” npj Quantum Information , vol. 7, no. 1, p. 122, 2021. [Online]. Available: https://doi.org/10.1038/s41534-021-00456-5

  8. [16]

    Quantum DeepONet: Neural operators accelerated by quantum computing,

    P. Xiao, M. Zheng, A. Jiao, X. Yang, and L. Lu, “Quantum DeepONet: Neural operators accelerated by quantum computing,” arXiv preprint arXiv:2409.15683, 2024

  9. [17]

    R66-50 an algorithm for the machine calculation of complex fourier series,

    H. S. Stone, “R66-50 an algorithm for the machine calculation of complex fourier series,” IEEE Transactions on Electronic Computers , vol. EC-15, no. 4, pp. 680–681, 1966

  10. [18]

    Orthogonal deep neural networks,

    S. Li, K. Jia, Y . Wen, T. Liu, and D. Tao, “Orthogonal deep neural networks,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 43, no. 4, pp. 1352–1368, 2021

  11. [19]

    Quantum Methods for Neural Networks and Appli- cation to Medical Image Classification,

    J. Landman, N. Mathur, Y . Y . Li, M. Strahm, S. Kazdaghli, A. Prakash, and I. Kerenidis, “Quantum Methods for Neural Networks and Appli- cation to Medical Image Classification,” Quantum, vol. 6, p. 881, Dec. 2022

  12. [20]

    Gropp, E

    W. Gropp, E. Lusk, and A. Skjellum, Using MPI: portable parallel programming with the message-passing interface . MIT press, 1999, vol. 1

  13. [21]

    mpi4py: Status update after 12 years of development,

    L. Dalcin and Y .-L. L. Fang, “mpi4py: Status update after 12 years of development,” Computing in Science & Engineering , vol. 23, no. 4, pp. 47–54, 2021

  14. [22]

    Hybrid quantum pro- gramming with pennylane lightning on hpc platforms,

    A. Asadi, A. Dusko, C.-Y . Park, V . Michaud-Rioux, I. Schoch, S. Shu, T. Vincent, and L. J. O’Riordan, “Hybrid quantum pro- gramming with pennylane lightning on hpc platforms,” arXiv preprint arXiv:2403.02512, 2024

  15. [23]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational Physics, vol. 378, pp. 686–707, 2019

  16. [24]

    The quest for a quantum neural network,

    M. Schuld, I. Sinayskiy, and F. Petruccione, “The quest for a quantum neural network,” Quantum Information Processing , vol. 13, pp. 2567– 2586, 2014

  17. [25]

    Continuous-variable quantum neural networks,

    N. Killoran, T. R. Bromley, J. M. Arrazola, M. Schuld, N. Quesada, and S. Lloyd, “Continuous-variable quantum neural networks,” Physical Review Research, vol. 1, no. 3, p. 033063, 2019

  18. [26]

    The power of quantum neural networks,

    A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, and S. Woerner, “The power of quantum neural networks,” Nature Computational Sci- ence, vol. 1, no. 6, pp. 403–409, 2021

  19. [27]

    Schuld and F

    M. Schuld and F. Petruccione, Machine learning with quantum comput- ers. Springer, 2021, vol. 676

  20. [28]

    Expressive power of parametrized quantum circuits,

    Y . Du, M.-H. Hsieh, T. Liu, and D. Tao, “Expressive power of parametrized quantum circuits,” Physical Review Research, vol. 2, no. 3, p. 033125, 2020

  21. [29]

    Classification with quantum neural networks on near term processors,

    E. Farhi and H. Neven, “Classification with quantum neural networks on near term processors,” arXiv preprint arXiv:1802.06002 , 2018

  22. [30]

    Solving nonlinear differential equations with differentiable quantum circuits,

    O. Kyriienko, A. E. Paine, and V . E. Elfving, “Solving nonlinear differential equations with differentiable quantum circuits,” Physical Review A, vol. 103, no. 5, p. 052416, 2021

  23. [31]

    Generalized quantum circuit differen- tiation rules,

    O. Kyriienko and V . E. Elfving, “Generalized quantum circuit differen- tiation rules,” Physical Review A , vol. 104, no. 5, p. 052417, 2021

  24. [32]

    Physics-informed quantum machine learning: Solving nonlinear differential equations in latent spaces without costly grid evaluations,

    A. E. Paine, V . E. Elfving, and O. Kyriienko, “Physics-informed quantum machine learning: Solving nonlinear differential equations in latent spaces without costly grid evaluations,” arXiv preprint arXiv:2308.01827, 2023

  25. [33]

    On physics-informed neural networks for quantum com- puters,

    S. Markidis, “On physics-informed neural networks for quantum com- puters,” Frontiers in Applied Mathematics and Statistics , vol. 8, p. 1036711, 2022

  26. [34]

    Expressibility and entan- gling capability of parameterized quantum circuits for hybrid quantum- classical algorithms,

    S. Sim, P. D. Johnson, and A. Aspuru-Guzik, “Expressibility and entan- gling capability of parameterized quantum circuits for hybrid quantum- classical algorithms,” Advanced Quantum Technologies , vol. 2, no. 12, p. 1900070, 2019

  27. [35]

    Quantum algorithms for deep convolutional neural networks,

    I. Kerenidis, J. Landman, and A. Prakash, “Quantum algorithms for deep convolutional neural networks,” arXiv preprint arXiv:1911.01117, 2019

  28. [36]

    What can we learn from quantum convolutional neural networks?

    C. Umeano, A. E. Paine, V . E. Elfving, and O. Kyriienko, “What can we learn from quantum convolutional neural networks?”Advanced Quantum Technologies, p. 2400325, 2023

  29. [37]

    Fast electromagnetic field pattern calculation with fourier neural operators,

    N. Pornthisan and S. Markidis, “Fast electromagnetic field pattern calculation with fourier neural operators,” in International Conference on Computational Science . Springer, 2023, pp. 247–255

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.