Pith. sign in

REVIEW 4 major objections 5 minor 112 references

A data-driven quantum Koopman method embeds moderately nonlinear dynamics into learned linear observables and runs them as shallow parallel circuits on a superconducting processor, capturing multiscale patterns while mapping a noise-to-theo

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

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

T0 review · grok-4.5

2026-07-10 19:16 UTC pith:4UR5NLJK

load-bearing objection Solid hardware-validated hybrid pipeline for moderately nonlinear continuum dynamics; the amenability boundary is empirical, not theorem-controlled, but the experiments and error bookkeeping are real. the 4 major comments →

arxiv 2607.07338 v1 pith:4UR5NLJK submitted 2026-07-08 quant-ph cs.AIphysics.flu-dyn

Quantum simulation of real-world nonlinear dynamics via Koopman method

Baoyang Zhang , Dong An , Zhaoyuan Meng , Yefei Yu , Xiaoxiao Xiao , Zhen Lu , Yue Yang This is my paper
classification quant-ph cs.AIphysics.flu-dyn
keywords quantum Koopman methodnonlinear dynamicslinear combination of Hamiltonian simulationNISQ simulationshallow quantum circuitsreaction-diffusionshallow-water equationsGulf Stream
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Quantum computers evolve states unitarily, so they cannot directly simulate the nonlinear, often dissipative dynamics that dominate chemistry, fluids, and oceans. This paper introduces the quantum Koopman method: it learns a finite set of observables from trajectory data that lift the nonlinear system into a linear evolution, then decomposes the resulting non-unitary propagator into a handful of spectral channels, each realized by a shallow circuit whose time evolution is just a single layer of phase rotations. On a superconducting chip the method is run with as many as 32 parallel 10-qubit circuits for three real-world-scale problems—a three-dimensional reaction-diffusion system, shallow-water flow on a sphere, and satellite Gulf Stream currents—recovering the dominant spatial patterns, energy spectra, and one-point statistics. Accuracy is limited first by hardware noise in the weakly nonlinear case and then by the finite Koopman subspace and the shallow circuit ansatz once scale interactions strengthen, thereby drawing a practical boundary for which nonlinear problems remain quantum-amenable on near-term devices. The claimed evolution-step speedup relative to classical dense Koopman propagation scales as roughly 2^n over n cubed when the single-layer ansatz already suffices.

Core claim

The quantum Koopman method, by jointly learning finite observables and compiling the non-unitary propagator via a diagonalized linear-combination-of-Hamiltonian-simulation decomposition into parallel shallow circuits, can simulate moderately nonlinear dynamics on present superconducting hardware (up to 32×10-qubit circuits) while capturing dominant multiscale patterns; as nonlinearity increases, the dominant error source transitions from hardware noise to the finite-dimensional Koopman representation itself, thereby delineating a practical quantum-amenable regime.

What carries the argument

The quantum Koopman method (QKM): learn Koopman observables that embed the nonlinear flow into a finite linear system, then realize the non-unitary propagator as a weighted sum of diagonal unitaries (Theorems 1–3), each executed by a topology-native circuit whose only time-evolution gates are a single layer of Rz rotations.

Load-bearing premise

That a finite set of learned observables plus a single layer of phase-rotation gates can capture enough of the spectral weight of a moderately nonlinear system that the theoretical error stays scientifically useful without needing exponentially many higher-order multi-qubit interactions.

What would settle it

Increase the retained interaction order in the time-evolution block (or enlarge the observable dimension) on a system already in the intermediate regime; if the measured training loss and long-horizon relative L2 error do not drop below the stated 10^{-3} threshold while the circuit remains shallow enough for the claimed speedup, the amenability boundary claim fails.

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

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The manuscript introduces the quantum Koopman method (QKM): nonlinear dynamics are lifted via learned Koopman observables into a finite N=2^n linear system, the non-unitary propagator e^{At} is rewritten by a diagonalized LCHS (Theorems 1–2) and approximated by parallel shallow circuits whose time-evolution block is a single-layer R_z sandwich (Theorem 3), and physical fields are prepared/decoded by a classical NN autoencoder. The pipeline is executed on the superconducting processor “Yudu” for three systems—3D Gray–Scott reaction–diffusion (6 qubits × 8 circuits), spherical shallow-water flow (10 × 32), and satellite Gulf Stream velocities (10 × 8)—reporting energy, enstrophy spectra, L2 error, and Hellinger fidelity, and arguing a transition from hardware-noise-limited to finite-Koopman/ansatz-limited accuracy that defines QKM-amenable / intermediate / prohibitive regimes with theoretical evolution speedup S ∼ O(2^n/n^3) versus classical Koopman propagation.

Significance. If the claims hold, the work is a meaningful hardware-validated step for moderately nonlinear dynamics on NISQ devices: it unifies data-driven Koopman lifting with LCHS-derived, topology-native parallel circuits rather than heuristic VQAs; Theorems 1–3 are stated with SI proofs that match the circuit design; experiments expand prior nonlinear/fluid demos to up to 32 parallel 10-qubit circuits and include real observational data; SI provides noiseless baselines, ablation on h and R_zz, and an explicit error budget. Code is deposited. These strengths make the paper of interest even if the “practical boundary” interpretation must be qualified.

major comments (4)
  1. Abstract, Discussion, and Fig. 5: the central claim that the experiments “identify a practical boundary for quantum-amenable nonlinear dynamics” (noise-limited → representation-limited) and partition systems into QKM-amenable/intermediate/prohibitive regimes rests on ε_th ≲ ε^*=10^{-3}, operationalized by training loss ℓ_train (Tab. I; SI §8C). Theorems 1–3 rigorously bound only spectral sampling (ε_spec) and single-layer R_z residual (ε_ansatz via high-order Pauli-Z coefficients). Projection error ε_proj from truncating the infinite-dimensional Koopman generator to the learned N-dimensional observables is uncontrolled; the Discussion and SI §8 explicitly leave rigorous bounds on ε_proj open. Without a controlled separation of ε_proj from autoencoder capacity, the regime partition and the claim of reach beyond analytical linearizations remain empirical. Please either (i) provide a quanti
  2. Methods “Complexity analysis” and Tab. I: the advertised speedup S ∼ O(2^n/n^3) (and S_evo = 2^n/(h n)) is relative to classical dense/sparse propagation of the same finite-N Koopman system, not to classical simulation of the original nonlinear PDE/DNS. Experimental S in Tab. I is 0.14, 0.22, and 0.88—all below unity—so no end-to-end wall-clock advantage is demonstrated. The manuscript should state the baseline explicitly in the abstract/results, report classical Koopman wall-clock cost for the same N,h, and avoid language that could be read as advantage over classical nonlinear solvers. Clarify also that S_total includes M shots and that practical advantage requires N ≳ O(1/ε_meas^2).
  3. Results (spherical fluid, Gulf Stream) and SI §9: for the two more nonlinear cases, noiseless emulation L2 errors are already comparable to hardware (sphere ≲0.06 vs hardware ~0.1; ocean ~0.21 matching hardware), so the “transition to theory-limited” is largely the autoencoder+ansatz residual. A load-bearing comparison is missing: classical NN/Koopman (or the same encoder–decoder with classical matrix exp(At)) at identical N and training budget. Without it, one cannot attribute multiscale fidelity to the quantum LCHS circuits rather than to the learned reduced model. Please add this classical baseline for at least one intermediate case and discuss what the quantum step uniquely contributes at current n.
  4. Theorem 1 and SI §1A: the LCHS form requires L ⪯ 0, obtained by a global shift u(t)=e^{bt}c(t). For chaotic/unstable geophysical flows this rescaling changes the observable magnitudes and the Cauchy–Lorentz weights; the manuscript does not report the chosen b, the resulting ||L||_2 used in the Theorem 2 bound, or sensitivity of ε_spec and training loss to b. Please document the shift for each benchmark and verify that the reported spectral-sampling bound remains meaningful after rescaling.
minor comments (5)
  1. Fig. 1 and Fig. 2 captions/labels contain garbled text (“Λinear”, “efiolution”, “hared operator”, “Yudu” layout labels). Clean for production.
  2. Tab. I: define Sevo vs S consistently in the caption; “theoretical evolution speedup” vs “quantum speedup” is easy to conflate with end-to-end runtime.
  3. SI §3 Table S1: “Moderate (physical space)” for QKM vs Carleman/KvN is qualitative; a one-sentence quantitative criterion (e.g., spectral radius or Reynolds/reaction number range) would help.
  4. Hellinger fidelity is reported as a hardware-vs-ideal quantum distribution metric; briefly note in Methods that it does not measure physical-field accuracy (that is ε_L2), to avoid conflation.
  5. References: several arXiv preprints of related quantum-fluid work are cited; ensure final versions are updated where available and that the authors’ prior data-driven Koopman preprint is clearly distinguished from the present hardware contribution.

Circularity Check

2 steps flagged

No load-bearing circular reduction in the theorems or hardware claims; mild fitted-proxy use of training loss for the amenability partition plus non-load-bearing self-citation of the authors’ prior data-driven Koopman preprint.

specific steps
  1. fitted input called prediction [SI §8C (Eqs. S49–S50) + Results ‘Pathway to quantum utility’ + Tab. I + Fig. 5]
    "we estimate the relative theoretical and optimization error as ε_th :=relative L2-ε_th ≈ L_train ≈ ℓ_train ... With measured training losses ℓ_train ranging from 10^{-3} to 2×10^{-2} (see Tab. I), the 3D reaction-diffusion benchmark falls within the QKM-amenable regime, whereas the spherical shallow-water and Gulf Stream benchmarks lie in the QKM-intermediate regime."

    The operational criterion ε_th ≲ 10^{-3} that partitions systems into QKM-amenable/intermediate/prohibitive (and thereby ‘identifies a practical boundary’) is defined to be the training loss of the model fitted to trajectories of those same systems. The regime labels are therefore the fit quality by construction, not an independent prediction or theorem-derived quantity (Theorems 1–3 bound only ε_spec and ε_ansatz; ε_proj remains open).

  2. self citation load bearing [Introduction (paragraph on existing approaches) + Ref. [41]]
    "Although a recent Koopman-inspired approach [41] has explored embedding physical priors into variational circuits, a unified framework that can simultaneously access moderately nonlinear regimes and compile into hardware-feasible circuits remains absent. ... [41] B. Zhang, Z. Lu, Y. Zhao, and Y. Yang, Data-driven quantum Koopman method for simulating nonlinear dynamics, preprint arXiv:2507.21890 (2025)."

    The data-driven quantum Koopman premise is introduced via citation to the authors’ own prior preprint; however the citation is not load-bearing for Theorems 1–3, the LCHS circuit construction, the hardware experiments, or the regime analysis, which go beyond the prior work. Mild only.

full rationale

Theorems 1–3 (diagonalized LCHS, spectral sampling bound, single-layer Rz residual via Walsh/Pauli-Z coefficients) are self-contained mathematical derivations with explicit proofs in SI §1 and do not reduce to fitted constants or self-citations. The end-to-end pipeline is explicitly data-driven (joint training of encoder/decoder/circuit parameters on trajectories, Alg. 1), with train/val/test splits and a 2024 held-out Gulf Stream window providing external checks; hardware L2/Hellinger metrics and noiseless-vs-hardware comparisons are independent of the training objective. The only mild circularity is operationalizing the QKM-amenable/intermediate partition (ε_th ≲ 10^{-3}) directly by the fitted training loss ℓ_train (Tab. I, Fig. 5, SI §8C), so the regime labels are by construction the fit quality rather than an independent first-principles prediction. Self-citation of the authors’ prior arXiv:2507.21890 is present but not load-bearing for the new theorems, hardware results, or regime analysis. Projection-error openness is an acknowledged assumption, not a circular step. Overall the derivation chain is not forced by definition or self-citation.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 2 invented entities

The central claim rests on standard Koopman/LCHS math plus several domain and paper-specific modeling choices: finite learnable invariant subspaces, L ⪯ 0 after rescaling, spectral compressibility into low-order Pauli-Z terms for the shallow ansatz, structured physical initial conditions for poly(n) state prep, and operational hyperparameters (n,h,R,r,ε*) that define the amenability boundary. No new physical particles; the invented objects are the QKM pipeline and the three-regime taxonomy.

free parameters (6)
  • Observable dimension N = 2^n and qubit count n
    Chosen per benchmark (n=6 or 10) to match grid size under n = O(log D); directly sets expressivity and claimed speedup.
  • Number of spectral channels h
    Hyperparameter balancing Theorem 2 error vs cost; set to 8 or 32 with scaling h = O(n).
  • State-prep depth factors R, r
    Chosen so Rr = O(n); trade expressibility against gate error (Tab. I).
  • Shot count M
    6144 or 10240 shots; controls measurement error bound.
  • Operational accuracy threshold ε* = 10^{-3}
    Hand-set criterion (via training loss) that partitions QKM-amenable vs intermediate regimes in Fig. 5.
  • NN encoder/decoder and circuit rotation parameters
    Fitted end-to-end on trajectory data; define the learned Koopman embedding and V(k), Λ(k).
axioms (5)
  • domain assumption Koopman generator admits a useful finite-N projection learned from data that preserves dominant dynamics.
    Methods ‘Koopman operator theory’; authors note optimal dimension remains open.
  • standard math Hermitian part L of A can be made ⪯ 0 by rescaling so diagonalized LCHS (Theorem 1) applies.
    Theorem 1 and SI §1A; standard LCHS hypothesis after shift.
  • domain assumption Physical initial fields are structured (smooth/symmetric) so poly(n)-depth PQCs plus NN encoder suffice for state preparation.
    Results ‘Quantum-circuit realization’ and SI §4 oracle argument.
  • ad hoc to paper Target diagonal Hamiltonians concentrate spectral weight in low-order Pauli-Z strings so single-layer Rz error is small (Theorem 3).
    Defines QKM-amenable regime; ablation shows Rzz reduces train loss, confirming residual high-order content.
  • standard math Uniform h-point Cauchy–Lorentz quadrature of the LCHS integral converges as in Theorem 2 bound.
    Theorem 2 / SI §1B.
invented entities (2)
  • Quantum Koopman method (QKM) pipeline independent evidence
    purpose: Jointly trained encoder, parallel LCHS spectral circuits, and decoder for non-unitary nonlinear evolution on NISQ hardware.
    Core contribution; builds on prior Koopman/LCHS pieces but packages a specific hardware-native ansatz.
  • QKM-amenable / intermediate / prohibitive regimes no independent evidence
    purpose: Taxonomy of nonlinear systems by spectral compressibility under shallow Rz ansatz and ε_th threshold.
    Operational partition from training loss and Fig. 5; not an independent physical law.

pith-pipeline@v1.1.0-grok45 · 41332 in / 3738 out tokens · 52221 ms · 2026-07-10T19:16:16.755830+00:00 · methodology

0 comments
read the original abstract

Nonlinear dynamics is ubiquitous in nature, ranging from chemical pattern formation to ocean circulation, yet its simulation on quantum computers is fundamentally limited by the unitary nature of quantum evolution. We propose the quantum Koopman method, a data-driven framework that embeds nonlinear dynamics into a learned linear representation and implements the resulting evolution using shallow quantum circuits. This method learns Koopman observables from trajectory data, projects the lifted dynamics onto a finite-dimensional subspace, and decomposes the corresponding non-unitary propagator into parallel spectral channels. We utilize the Koopman method on a superconducting processor to simulate three distinct nonlinear systems, comprising reaction-diffusion dynamics, fluid motion on a sphere, and satellite-derived observations of Gulf Stream currents, employing up to 32 parallel circuits of 10 qubits. These quantum simulations capture the dominant multiscale patterns and statistical signatures of the underlying dynamics, and reveal a transition from performance limited by hardware noise in weakly nonlinear systems to performance limited by finite-dimensional Koopman representations as nonlinear scale interactions increase. This transition identifies a practical boundary for quantum-amenable nonlinear dynamics, establishing a hardware-validated route for simulating moderately nonlinear dynamics on near-term quantum hardware.

Figures

Figures reproduced from arXiv: 2607.07338 by Baoyang Zhang, Dong An, Xiaoxiao Xiao, Yefei Yu, Yue Yang, Zhaoyuan Meng, Zhen Lu.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

Review history (2 revisions) →

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

112 extracted references · 112 canonical work pages · 4 internal anchors

  1. [1]

    An , author J.-P

    author author D. An , author J.-P. \ Liu ,\ and\ author L. Lin ,\ title title black Linear combination of Hamiltonian Simulation for Nonunitary Dynamics with Optimal State Preparation Cost ,\ @noop journal journal Phys. Rev. Lett. \ volume 131 ,\ pages 150603 ( year 2023 ) NoStop

    work page 2023

  2. [2]

    author author R. M. \ Wilcox ,\ title title black Exponential operators and parameter differentiation in quantum physics ,\ @noop journal journal J. Math. Phys. \ volume 8 ,\ pages 962 ( year 1967 ) NoStop

    work page 1967

  3. [3]

    O'Donnell ,\ @noop title Analysis of boolean functions \ ( publisher Cambridge University Press ,\ year 2014 ) NoStop

    author author R. O'Donnell ,\ @noop title Analysis of boolean functions \ ( publisher Cambridge University Press ,\ year 2014 ) NoStop

    work page 2014

  4. [4]

    Data-driven quantum Koopman method for simulating nonlinear dynamics

    author author B. Zhang , author Z. Lu , author Y. Zhao ,\ and\ author Y. Yang ,\ title title black Data-driven quantum Koopman method for simulating nonlinear dynamics ,\ @noop journal journal preprint arXiv:2507.21890 \ ( year 2025 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Ronneberger , author P

    author author O. Ronneberger , author P. Fischer ,\ and\ author T. Brox ,\ title title black U-net: convolutional networks for biomedical image segmentation ,\ in\ @noop booktitle Med. Image Comput. Comput.-Assist. Interv. (MICCAI) \ ( organization Springer ,\ year 2015 )\ pp.\ pages 234--241 NoStop

    work page 2015

  6. [6]

    Li , author J

    author author H. Li , author J. Xie , author C. Zhang , author Y. Zhang ,\ and\ author Y. Zhao ,\ title title black A transformer-based convolutional method to model inverse cascade in forced two-dimensional turbulence ,\ @noop journal journal J. Comput. Phys. \ volume 520 ,\ pages 113475 ( year 2025 ) NoStop

    work page 2025

  7. [7]

    Xie , author W

    author author E. Xie , author W. Wang , author Z. Yu , author A. Anandkumar , author J. M. \ Alvarez ,\ and\ author P. Luo ,\ title title black SegFormer : simple and efficient design for semantic segmentation with transformers ,\ @noop journal journal Adv. Neural Inf. Process. Syst. \ volume 34 ,\ pages 12077 ( year 2021 ) NoStop

    work page 2021

  8. [8]

    author author BAQIS ,\ @noop title black Quafu superconducting quantum computing ,\ howpublished https://quafu-sqc.baqis.ac.cn https://quafu-sqc.baqis.ac.cn/home ( year 2024 ) NoStop

    work page 2024

  9. [9]

    \ Liu , author H

    author author J.-P. \ Liu , author H. . \ Kolden , author H. K. \ Krovi , author N. F. \ Loureiro , author K. Trivisa ,\ and\ author A. M. \ Childs ,\ title title black Efficient quantum algorithm for dissipative nonlinear differential equations ,\ @noop journal journal Proc. Natl. Acad. Sci. U. S. A. \ volume 118 ,\ pages e2026805118 ( year 2021 ) NoStop

    work page 2021

  10. [10]

    Quantum algorithms for general nonlinear dynamics based on the Carleman embedding

    author author D. Jennings , author K. Korzekwa , author M. Lostaglio , author A. T. \ Sornborger , author Y. Subasi ,\ and\ author G. Wang ,\ title title black Quantum algorithms for general nonlinear dynamics based on the Carleman embedding ,\ @noop journal journal arXiv preprint arXiv:2509.07155 \ ( year 2025 ) NoStop

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    Joseph ,\ title title black Koopman–von Neumann approach to quantum simulation of nonlinear classical dynamics ,\ @noop journal journal Phys

    author author I. Joseph ,\ title title black Koopman–von Neumann approach to quantum simulation of nonlinear classical dynamics ,\ @noop journal journal Phys. Rev. Res. \ volume 2 ,\ pages 043102 ( year 2020 ) NoStop

    work page 2020

  12. [12]

    Novikau \ and\ author I

    author author I. Novikau \ and\ author I. Joseph ,\ title title black Quantum algorithm for the advection-diffusion equation and the Koopman-von Neumann approach to nonlinear dynamical systems ,\ @noop journal journal Comput. Phys. Commun. \ volume 309 ,\ pages 109498 ( year 2025 ) NoStop

    work page 2025

  13. [13]

    Gray \ and\ author S

    author author P. Gray \ and\ author S. Scott ,\ title title black Autocatalytic reactions in the isothermal, continuous stirred tank reactor: isolas and other forms of multistability ,\ @noop journal journal Chem. Eng. Sci. \ volume 38 ,\ pages 29 ( year 1983 ) NoStop

    work page 1983

  14. [14]

    @noop title black The code is available at https://github.com/YYgroup/QKM https://github.com/YYgroup/QKM. Stop

    work page

  15. [15]

    Galewsky , author R

    author author J. Galewsky , author R. K. \ Scott ,\ and\ author L. M. \ Polvani ,\ title title black An initial-value problem for testing numerical models of the global shallow-water equations ,\ @noop journal journal Tellus Ser. A-Dyn. Meteorol. Oceanogr. \ volume 56 ,\ pages 429 ( year 2004 ) NoStop

    work page 2004

  16. [16]

    author author K. J. \ Burns , author G. M. \ Vasil , author J. S. \ Oishi , author D. Lecoanet ,\ and\ author B. P. \ Brown ,\ title title black Dedalus: a flexible framework for numerical simulations with spectral methods ,\ @noop journal journal Phys. Rev. Res. \ volume 2 ,\ pages 023068 ( year 2020 ) NoStop

    work page 2020

  17. [17]

    author author E.U. Copernicus Marine Service (CMEMS) ,\ @noop title black Global ocean gridded L4 sea surface heights and derived variables reprocessed 1993 ongoing ,\ howpublished https://doi.org/10.48670/moi-00148 https://doi.org/10.48670/moi-00148 ( year 2024 ) NoStop

    work page doi:10.48670/moi-00148 1993

  18. [18]

    Hersbach , author B

    author author H. Hersbach , author B. Bell , author P. Berrisford , author S. Hirahara , author A. Hor \'a nyi , author J. Mu \ n oz-Sabater , author J. Nicolas , author C. Peubey , author R. Radu , author D. Schepers , et al. ,\ title title black The ERA5 global reanalysis ,\ @noop journal journal Q. J. R. Meteorol. Soc. \ volume 146 ,\ pages 1999 ( year...

    work page 1999

  19. [19]

    \ Zhang , author T

    author author X.-M. \ Zhang , author T. Li ,\ and\ author X. Yuan ,\ title title black Quantum state preparation with optimal circuit depth: implementations and applications ,\ @noop journal journal Phys. Rev. Lett. \ volume 129 ,\ pages 230504 ( year 2022 ) NoStop

    work page 2022

  20. [20]

    Sun , author G

    author author X. Sun , author G. Tian , author S. Yang , author P. Yuan ,\ and\ author S. Zhang ,\ title title black Asymptotically optimal circuit depth for quantum state preparation and general unitary synthesis ,\ @noop journal journal IEEE Trans. Comput.-Aided Des. Integr. Circuits Syst. \ volume 42 ,\ pages 3301 ( year 2023 ) NoStop

    work page 2023

  21. [22]

    Meng , author L

    author author Z. Meng , author L. Chen , author J.-P. \ Liu ,\ and\ author G. He ,\ title title black Toward end-to-end quantum simulation of rapidly distorted turbulence ,\ @noop journal journal J. Comput. Phys. \ volume 558 ,\ pages 114888 ( year 2026 ) NoStop

    work page 2026

  22. [23]

    Sim , author P

    author author S. Sim , author P. D. \ Johnson ,\ and\ author A. Aspuru-Guzik ,\ title title black Expressibility and entangling capability of parameterized quantum circuits for hybrid quantum-classical algorithms ,\ @noop journal journal Adv. Quantum Technol. \ volume 2 ,\ pages 1900070 ( year 2019 ) NoStop

    work page 2019

  23. [24]

    \.Z yczkowski \ and\ author H.-J

    author author K. \.Z yczkowski \ and\ author H.-J. \ Sommers ,\ title title black Average fidelity between random quantum states ,\ @noop journal journal Phys. Rev. A \ volume 71 ,\ pages 032313 ( year 2005 ) NoStop

    work page 2005

  24. [25]

    Goodfellow , author Y

    author author I. Goodfellow , author Y. Bengio , author A. Courville ,\ and\ author Y. Bengio ,\ @noop title Deep learning ,\ Vol. volume 1 \ ( publisher MIT press Cambridge ,\ year 2016 ) NoStop

    work page 2016

  25. [26]

    2015 , journal =

    Read the fine print , author =. 2015 , journal =

    work page 2015

  26. [27]

    Quantum Mach

    Future of quantum computing , author=. Quantum Mach. Intell. , volume=

    work page

  27. [28]

    Akiba, Takaki and Morii, Youhi and Maruta, Kaoru , year =. Sci. Rep. , volume =

    work page

  28. [29]

    Quantum dynamics simulation of the advection-diffusion equation , author=. Phys. Rev. Res. , volume=. 2025 , publisher=

    work page 2025

  29. [30]

    Linear combination of

    An, Dong and Liu, Jin-Peng and Lin, Lin , year =. Linear combination of. Phys. Rev. Lett. , volume =

    work page

  30. [31]

    Quantum variational algorithms are swamped with traps , author=. Nat. Commun. , volume=. 2022 , publisher=

    work page 2022

  31. [32]

    Improving vision transformers by revisiting high-frequency components , author =. Eur. Conf. Comput. Vis. (ECCV) , pages =. 2022 , organization =

    work page 2022

  32. [33]

    Elementary gates for quantum computation , author=. Phys. Rev. A , volume=. 1995 , publisher=

    work page 1995

  33. [34]

    Long-term evolution of ocean eddy activity in a warming world , author=. Nat. Clim. Chang. , volume=. 2022 , publisher=

    work page 2022

  34. [35]

    Compact quantum algorithms for time-dependent differential equations , author=. Phys. Rev. Res. , volume=. 2025 , publisher=

    work page 2025

  35. [36]

    Hybrid quantum algorithms for flow problems , author=. Proc. Natl. Acad. Sci. U. S. A. , volume=. 2023 , publisher=

    work page 2023

  36. [37]

    Machine learning for the physics of climate , author=. Nat. Rev. Phys. , volume=. 2025 , publisher=

    work page 2025

  37. [38]

    Nature , volume=

    Quantum machine learning , author=. Nature , volume=

    work page

  38. [39]

    Quantum algorithm for solving the advection equation using

    Brearley, Peter and Laizet, Sylvain , year =. Quantum algorithm for solving the advection equation using. Phys. Rev. A , volume =

    work page

  39. [40]

    and Budi

    Brunton, Steven L. and Budi. Modern. 2022 , journal =

    work page 2022

  40. [41]

    Quantum algorithm for the advection--diffusion equation simulated with the lattice

    Budinski, Ljubomir , journal=. Quantum algorithm for the advection--diffusion equation simulated with the lattice. 2021 , publisher=

    work page 2021

  41. [42]

    Application de la th

    Carleman, Torsten , journal =. Application de la th

    work page

  42. [43]

    Challenges and opportunities in quantum machine learning , author=. Nat. Comput. Sci. , volume=

    work page

  43. [44]

    Chen, Shiyi and Doolen, Gary D , journal=. Lattice

    work page

  44. [45]

    Enabling large-scale and high-precision fluid simulations on near-term quantum computers , author=. Comput. Methods Appl. Mech. Eng. , volume=. 2024 , publisher=

    work page 2024

  45. [46]

    Symbolic analysis of

    Chen, Min and Cheng, Jinglei and Li, Pingzhi and Wang, Haoran and Chen, Tianlong and Liu, Junyu , journal=. Symbolic analysis of. 2026 , publisher=

    work page 2026

  46. [47]

    Quantum algorithm for systems of linear equations with exponentially improved dependence on precision , author=. SIAM J. Comput. , volume=. 2017 , publisher=

    work page 2017

  47. [48]

    Global ocean gridded

    work page

  48. [49]

    The code is available at

    work page

  49. [50]

    Approximation by superpositions of a sigmoidal function , author=. Math. Control Signals Syst. , volume=. 1989 , publisher=

    work page 1989

  50. [51]

    2022 , journal =

    Practical quantum advantage in quantum simulation , author =. 2022 , journal =

    work page 2022

  51. [52]

    PRX Quantum , volume=

    Quantum computing for high-energy physics: state of the art and challenges , author=. PRX Quantum , volume=

    work page

  52. [53]

    1982 , journal =

    Simulating physics with computers , author =. 1982 , journal =

    work page 1982

  53. [54]

    2022 , journal =

    Towards quantum computing of turbulence , author =. 2022 , journal =

    work page 2022

  54. [55]

    Neural Netw

    On the approximate realization of continuous mappings by neural networks , author=. Neural Netw. , volume=. 1989 , publisher=

    work page 1989

  55. [56]

    Finding flows of a

    Gaitan, Frank , journal=. Finding flows of a. 2020 , publisher=

    work page 2020

  56. [57]

    Tellus Ser

    An initial-value problem for testing numerical models of the global shallow-water equations , author=. Tellus Ser. A-Dyn. Meteorol. Oceanogr. , volume=. 2004 , publisher=

    work page 2004

  57. [58]

    2020 , journal =

    Quantum speedup for aeroscience and engineering , author =. 2020 , journal =

    work page 2020

  58. [59]

    2025 , journal =

    Quantum. 2025 , journal =

    work page 2025

  59. [60]

    A quantum-inspired approach to exploit turbulence structures , author=. Nat. Comput. Sci. , volume=. 2022 , publisher=

    work page 2022

  60. [61]

    Autocatalytic reactions in the isothermal, continuous stirred tank reactor: isolas and other forms of multistability , author =. Chem. Eng. Sci. , volume =. 1983 , publisher =

    work page 1983

  61. [62]

    Quantum algorithm for linear systems of equations , author=. Phys. Rev. Lett. , volume=. 2009 , publisher=

    work page 2009

  62. [63]

    2023 , journal =

    Disentangling hype from practicality: on realistically achieving quantum advantage , author =. 2023 , journal =

    work page 2023

  63. [64]

    Neural Netw

    Multilayer feedforward networks are universal approximators , author=. Neural Netw. , volume=. 1989 , publisher=

    work page 1989

  64. [65]

    Variational quantum algorithms for computational fluid dynamics , author=. AIAA J. , volume=. 2023 , publisher=

    work page 2023

  65. [66]

    Jennings, David and Korzekwa, Kamil and Lostaglio, Matteo and Wang, Guoming , year =. Quantum

    work page

  66. [67]

    Time complexity analysis of quantum algorithms via linear representations for nonlinear ordinary and partial differential equations , author =. J. Comput. Phys. , volume =

    work page

  67. [68]

    Quantum simulation of partial differential equations via

    Jin, Shi and Liu, Nana and Yu, Yue , journal=. Quantum simulation of partial differential equations via

    work page

  68. [69]

    Joseph, Ilon , journal =

    work page

  69. [70]

    Koopman, Bernard O , journal =

    work page

  70. [71]

    Barren plateaus in variational quantum computing , author=. Nat. Rev. Phys. , pages=. 2025 , publisher=

    work page 2025

  71. [72]

    A multiple-circuit approach to quantum resource reduction with application to the quantum lattice

    Lee, Melody and Song, Zhixin and Kocherla, Sriharsha and Adams, Austin and Alexeev, Alexander and Bryngelson, Spencer H , journal=. A multiple-circuit approach to quantum resource reduction with application to the quantum lattice. 2026 , publisher=

    work page 2026

  72. [73]

    Learning spatiotemporal dynamics with a pretrained generative model , author=. Nat. Mach. Intell. , volume=. 2024 , publisher=

    work page 2024

  73. [74]

    Fourier neural operator for parametric partial differential equations , author=. Int. Conf. Learn. Represent. (ICLR) , year=

    work page

  74. [75]

    2004 , journal =

    On the homotopy analysis method for nonlinear problems , author =. 2004 , journal =

    work page 2004

  75. [76]

    Data-driven learning for the

    Lin, Yen Ting and Tian, Yifeng and Livescu, Daniel and Anghel, Marian , journal=. Data-driven learning for the. 2021 , publisher=

    work page 2021

  76. [77]

    Efficient quantum algorithm for dissipative nonlinear differential equations , author =. Proc. Natl. Acad. Sci. U. S. A. , volume =. 2021 , publisher =

    work page 2021

  77. [78]

    2017 , journal =

    Analysis of operator splitting errors for near-limit flame simulations , author =. 2017 , journal =

    work page 2017

  78. [79]

    Quantum computing of reacting flows via

    Lu, Zhen and Yang, Yue , year =. Quantum computing of reacting flows via. Proc. Combust. Inst. , volume =

    work page

  79. [80]

    Variational quantum algorithms for nonlinear problems , author=. Phys. Rev. A , volume=. 2020 , publisher=

    work page 2020

  80. [81]

    Deep learning for universal linear embeddings of nonlinear dynamics , author=. Nat. Commun. , volume=. 2018 , publisher=

    work page 2018

Showing first 80 references.