arxiv: 2604.03346 · v1 · submitted 2026-04-03 · 🪐 quant-ph

Recognition: 2 theorem links

· Lean Theorem

Learning PDEs for Portfolio Optimization with Quantum Physics-Informed Neural Networks

Letao Wang , Abdel Lisser , Sreejith Sreekumar , Zeno Toffano

Authors on Pith no claims yet

Pith reviewed 2026-05-13 19:49 UTC · model grok-4.3

classification 🪐 quant-ph

keywords PDEsportfolio optimizationquantum circuitsphysics-informed neural networkstensor rank decompositionMerton problemquantum-inspired models

0 comments

The pith

Parameterized quantum circuits using tensor rank decomposition approximate PDE solutions for portfolio optimization with higher accuracy and fewer parameters than classical networks.

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

The paper establishes that a parameterized quantum circuit can implement polynomials via tensor rank decomposition, enabling quantum physics-informed neural networks to solve partial differential equations in financial mathematics. This reduces resource complexity to polynomial when the tensor rank is moderate and provides approximation guarantees for the PDE solution. Experiments on the Merton portfolio optimization problem, which finds the optimal fraction to invest in risky versus risk-free assets, show the quantum models outperforming classical fully connected PINNs in accuracy and convergence speed while using 80 times fewer parameters. The approach also beats classical PINNs designed with the same inductive bias, indicating a quantum-induced benefit.

Core claim

We develop a parameterized quantum circuit for tensor rank decomposition of polynomials and use it to construct Quantum Physics-Informed Neural Networks and Quantum-inspired PINNs that guarantee polynomial approximations to PDE solutions, demonstrating superior performance on the Merton problem with significantly reduced parameters.

What carries the argument

Parameterized quantum circuit implementing a polynomial via tensor rank decomposition within a physics-informed neural network framework to enforce the PDE residual.

Load-bearing premise

The solution to the target PDE admits a polynomial representation with sufficiently low tensor rank for the quantum circuit to remain efficient.

What would settle it

Running the same experiment on a PDE whose solution requires high tensor rank, resulting in no accuracy gain or loss of efficiency, would falsify the practical advantage.

read the original abstract

Partial differential equations (PDEs) play a crucial role in financial mathematics, particularly in portfolio optimization, and solving them using classical numerical or neural network methods has always posed significant challenges. Here, we investigate the potential role of quantum circuits for solving PDEs. We design a parameterized quantum circuit (PQC) for implementing a polynomial based on tensor rank decomposition, reducing the quantum resource complexity from exponential to polynomial when the corresponding tensor rank is moderate. Building on this circuit, we develop a Quantum Physics-Informed Neural Network (QPINN) and a Quantum-inspired PINN, both of which guarantee the existence of an approximation of the PDE solution, and this approximation is represented as a polynomial that incorporates tensor rank decomposition. Despite using 80 times fewer parameters in experiments, our quantum models achieve higher accuracy and faster convergence than a classical fully connected PINN when solving the PDE for the Merton portfolio optimization problem, which determines the optimal investment fraction between a risky and a risk-free asset. Our quantum models further outperform a classical PINN constructed to share the same inductive bias, providing experimental evidence of quantum-induced improvement and highlighting a resource-efficient pathway toward classical and near-term quantum PDE solvers.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

The paper's tensor-rank PQC inside a QPINN for the Merton PDE claims 80x parameter savings with better accuracy, but the low-rank assumption on the known closed-form solution is unverified.

read the letter

The headline takeaway is that this work builds a parameterized quantum circuit from tensor-rank polynomial decomposition and uses it in a quantum physics-informed neural network to solve the Merton portfolio optimization PDE, reporting 80 times fewer parameters with higher accuracy and faster convergence than a standard classical PINN. They also compare against a classical PINN that shares the same inductive bias, which is a reasonable control. The circuit construction and the approximation guarantee tied to the polynomial representation are laid out clearly, and the experiments use a concrete financial example rather than a toy PDE. That combination of quantum circuit design with physics-informed training is the actual new piece here, and the empirical results give something concrete to look at. The main soft spot is exactly the one flagged in the stress test: the efficiency and approximation claims only hold if the Merton solution has moderate tensor rank under this polynomial ansatz, yet the paper supplies neither a numerical rank computation on their grid nor an a-priori bound. Since the Merton problem has a known closed-form solution, this omission leaves open the possibility that the observed gains come from a better-matched function class rather than anything quantum-specific. Details on error bars, noise models, and exact baseline implementations are also light, so the support for a genuine quantum-induced improvement stays preliminary. This is for people working on hybrid quantum-classical methods for high-dimensional PDEs or finance applications. A reader who wants to see a specific circuit construction and some numbers on the Merton problem will find material worth discussing. It deserves serious peer review because the idea is specific enough and the experiments are there to examine, even if the central efficiency claim needs tighter verification on rank and baselines.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a parameterized quantum circuit (PQC) realizing multivariate polynomials via tensor-rank decomposition to build Quantum Physics-Informed Neural Networks (QPINNs) and quantum-inspired PINNs for solving PDEs. Applied to the Merton portfolio optimization HJB equation, the quantum models are claimed to achieve higher accuracy and faster convergence than classical fully connected PINNs while using 80 times fewer parameters, with an approximation guarantee based on the polynomial tensor decomposition; this is presented as experimental evidence of quantum-induced improvement.

Significance. If the tensor-rank assumption holds and the experimental gains are robust, the work could demonstrate a resource-efficient route to PDE solvers for financial mathematics on near-term quantum hardware or via quantum-inspired classical methods, particularly for problems where high-dimensional grids challenge standard numerical techniques.

major comments (3)

[PQC construction and theoretical guarantee section] The resource-reduction claim (exponential to polynomial) and the 80x parameter advantage rest on the Merton PDE solution admitting a moderate tensor rank under the polynomial ansatz. The manuscript does not report the computed tensor rank of the known closed-form solution (logarithmic value function with constant control) on the discretization grid nor supply an a-priori bound, which is load-bearing for transferring the approximation guarantee to the concrete experiments.
[Experimental results on Merton problem] The experimental comparison to the classical fully connected PINN reports higher accuracy with far fewer parameters, yet lacks details on error bars, exact data splits, hyperparameter optimization protocol, and noise models; without these, it is unclear whether the observed improvement is attributable to the quantum structure or to differences in model capacity and training.
[Approximation guarantee discussion] The approximation guarantee is explicitly tied to the choice of tensor rank r and polynomial degree; the paper should specify how r is selected for the Merton instance and demonstrate that the guarantee remains valid under the discretization and any stochastic elements used in the numerical experiments.

minor comments (2)

[Methods] The description of the classical PINN sharing the same inductive bias as the QPINN would benefit from an explicit side-by-side architecture diagram or parameter count table.
[PQC construction] Notation for the tensor decomposition in the PQC (e.g., the precise mapping from tensor cores to circuit gates) could be illustrated with a small-scale example equation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help strengthen the manuscript. We address each major point below and will revise the paper to incorporate the suggested clarifications and additional details.

read point-by-point responses

Referee: The resource-reduction claim (exponential to polynomial) and the 80x parameter advantage rest on the Merton PDE solution admitting a moderate tensor rank under the polynomial ansatz. The manuscript does not report the computed tensor rank of the known closed-form solution (logarithmic value function with constant control) on the discretization grid nor supply an a-priori bound, which is load-bearing for transferring the approximation guarantee to the concrete experiments.

Authors: We agree that explicitly reporting the tensor rank of the closed-form solution on the experimental discretization grid would strengthen the connection between the general guarantee and our results. In the revised manuscript we will compute this rank numerically for the logarithmic value function and constant control on the grid used in experiments, and we will supply a simple a-priori bound derived from the analytic form of the solution that confirms the rank remains moderate (independent of dimension for this particular problem). revision: yes
Referee: The experimental comparison to the classical fully connected PINN reports higher accuracy with far fewer parameters, yet lacks details on error bars, exact data splits, hyperparameter optimization protocol, and noise models; without these, it is unclear whether the observed improvement is attributable to the quantum structure or to differences in model capacity and training.

Authors: We acknowledge the need for greater experimental transparency. The revised version will report error bars obtained from 10 independent runs with different random seeds, clarify that collocation points are sampled once from fixed distributions (no train/test split in the classical sense), document the hyperparameter search protocol (grid search over learning rate, network width, and polynomial degree), and state that all circuits are simulated noiselessly. We will also add a parameter-matched classical PINN baseline to isolate the effect of the tensor-rank inductive bias. revision: yes
Referee: The approximation guarantee is explicitly tied to the choice of tensor rank r and polynomial degree; the paper should specify how r is selected for the Merton instance and demonstrate that the guarantee remains valid under the discretization and any stochastic elements used in the numerical experiments.

Authors: We will add an explicit subsection describing the practical selection of r: r is chosen as the smallest integer such that the tensor-rank polynomial approximates the known closed-form solution to within a prescribed tolerance on a dense validation grid. We will also bound the additional discretization error separately and show that the overall guarantee continues to hold. Training uses deterministic full-batch gradient descent, so no stochasticity affects the loss; this will be stated clearly. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper's core claims rest on experimental comparisons showing quantum models outperforming classical PINNs (including one with matched inductive bias) on the Merton PDE, using 80x fewer parameters. The PQC construction for tensor-rank polynomial representation and the stated approximation guarantee follow directly from the explicit architectural choice and standard approximation properties of polynomials; these are not derived by reducing to the target result itself. No self-citations, fitted inputs renamed as predictions, or uniqueness theorems are invoked in a load-bearing way that collapses the central result to its inputs. The moderate-rank assumption is an empirical precondition verified by the reported success of the experiments rather than a hidden definitional loop.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the target PDE solution can be represented as a moderate-rank tensor polynomial and that this representation can be realized by a parameterized quantum circuit whose parameters are learned via physics-informed training.

free parameters (1)

tensor rank r
Chosen moderate to keep quantum resource scaling polynomial; value not specified in abstract but directly controls circuit depth and parameter count.

axioms (1)

domain assumption The PDE solution admits a polynomial approximation whose tensor rank remains moderate.
Invoked to guarantee existence of the approximation and to reduce quantum complexity from exponential to polynomial.

pith-pipeline@v0.9.0 · 5515 in / 1391 out tokens · 35892 ms · 2026-05-13T19:49:57.048326+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel (J uniquely calibrated reciprocal cost) unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

tensor-decomposed polynomial p(x) = Σ λ_r ∏ p_r,j(x_j) with R moderate; QPINN hypothesis space contains all such polynomials of fixed degree and bounded rank (Theorem 3, Corollary 4)
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean embed_strictMono_of_one_lt (orbit embedding monotonicity) unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Merton HJB solution v(t,x) = exp(-k(T-t)) x^γ/γ admits rank-1 tensor decomposition (separable univariate factors)

What do these tags mean?

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

Reference graph

Works this paper leans on

63 extracted references · 63 canonical work pages · 1 internal anchor

[1]

The Review of Economics and Statistics51, 247– 57 (1969) https://doi.org/10.2307/1926560

Merton, R.: Lifetime portfolio selection under uncertainty: The continuous-time case. The Review of Economics and Statistics51, 247– 57 (1969) https://doi.org/10.2307/1926560

work page doi:10.2307/1926560 1969
[2]

Mathematics13, 2664 (2025) https://doi.org/10.3390/math13162664

Gonzalo, V., Wahl, M., Zagst, R.: Dynamic portfolio optimization using information from a crisis indica- tor. Mathematics13, 2664 (2025) https://doi.org/10.3390/math13162664

work page doi:10.3390/math13162664 2025
[3]

Mathematics13(3) (2025) https: //doi.org/10.3390/math13030535

Di Persio, L., Fraccarolo, N.: Exploring optimisation strategies under jump-diffusion dynamics. Mathematics13(3) (2025) https: //doi.org/10.3390/math13030535

work page doi:10.3390/math13030535 2025
[4]

Journal of Indus- trial and Management Optimization18(6), 3831–3845 (2022) https://doi.org/10.3934/ jimo.2021137

Wang, X., Hu, L.: A new method to solve the hamilton-jacobi-bellman equation for a stochastic portfolio optimization model with boundary memory. Journal of Indus- trial and Management Optimization18(6), 3831–3845 (2022) https://doi.org/10.3934/ jimo.2021137

work page 2022
[5]

International Journal of Computer Mathematics93, 1– 10 (2014) https://doi.org/10.1080/00207160

Kilianov´ a, S., Trnovska, M.: Robust portfo- lio optimization via solution to the hamil- ton–jacobi–bellman equation. International Journal of Computer Mathematics93, 1– 10 (2014) https://doi.org/10.1080/00207160. 2013.871542

work page doi:10.1080/00207160 2014
[6]

Gollier, C.: The Economics of Risk and Time, pp. 43–54. MIT Press, Cambridge, MA (2001). https://doi.org/10.7551/mitpress/ 2622.001.0001

work page doi:10.7551/mitpress/ 2001
[7]

Artificial Intelligence Review58(323) (2025) https://doi.org/10

Luo, K., Zhao, J., Wang, Y., Li, J., Wen, J., Liang, J., Soekmadji, H., Liao, S.: Physics- informed neural networks for pde problems: a comprehensive review. Artificial Intelligence Review58(323) (2025) https://doi.org/10. 1007/s10462-025-11322-7

work page 2025
[8]

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

Raissi, M., Perdikaris, P., Karniadakis, G.E.: Physics-informed neural networks: A deep 16 learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computa- tional Physics378, 686–707 (2019) https:// doi.org/10.1016/j.jcp.2018.10.045

work page doi:10.1016/j.jcp.2018.10.045 2019
[9]

SIAM Journal on Computing26(5), 1484–1509 (1997) https: //doi.org/10.1137/s0097539795293172

Shor, P.W.: Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer. SIAM Journal on Computing26(5), 1484–1509 (1997) https: //doi.org/10.1137/s0097539795293172

work page doi:10.1137/s0097539795293172 1997
[10]

Aaijet al.[LHCb Collaboration]

Harrow, A.W., Hassidim, A., Lloyd, S.: Quantum algorithm for linear systems of equations. Physical Review Letters103(15) (2009) https://doi.org/10.1103/physrevlett. 103.150502

work page doi:10.1103/physrevlett 2009
[11]

Commu- nications in Mathematical Physics356(3), 1057–1081 (2017) https://doi.org/10.1007/ s00220-017-3002-y

Berry, D.W., Childs, A.M., Ostrander, A., Wang, G.: Quantum algorithm for lin- ear differential equations with exponentially improved dependence on precision. Commu- nications in Mathematical Physics356(3), 1057–1081 (2017) https://doi.org/10.1007/ s00220-017-3002-y

work page 2017
[12]

Physi- cal Review A111(6) (2025) https: //doi.org/10.1103/physreva.111.062404

Hunout, J., Laizet, S., Iannucci, L.: Variational quantum algorithm based on lagrange polynomial encoding to solve differential equations. Physi- cal Review A111(6) (2025) https: //doi.org/10.1103/physreva.111.062404

work page doi:10.1103/physreva.111.062404 2025
[13]

In: Samuel J

Brassard, G., Høyer, P., Mosca, M., Tapp, A.: Quantum amplitude amplification and estimation. In: Samuel J. Lomonaco, J., Brandt, H.E. (eds.) Quantum Computation and Information. Contemporary Mathemat- ics, vol. 305, pp. 53–74. American Mathe- matical Society, Providence, RI (2002). https: //doi.org/10.1090/conm/305/05215

work page doi:10.1090/conm/305/05215 2002
[14]

Kacewicz, B.: Almost optimal solution of initial-value problems by randomized and quantum algorithms. J. Complex.22(5), 676– 690 (2006)

work page 2006
[15]

npj Quan- tum Information6, 61 (2020) https://doi

Gaitan, F.: Finding flows of a navier–stokes fluid through quantum computing. npj Quan- tum Information6, 61 (2020) https://doi. org/10.1038/s41534-020-00291-0

work page doi:10.1038/s41534-020-00291-0 2020
[16]

Quantum Information Process- ing21(1), 30 (2022) https://doi.org/10.1007/ s11128-021-03391-8

Oz, F., Vuppala, R.K.S.S., Kara, K., Gaitan, F.: Solving burgers’ equation with quantum computing. Quantum Information Process- ing21(1), 30 (2022) https://doi.org/10.1007/ s11128-021-03391-8

work page 2022
[17]

Scientific Reports 13, 7767 (2023) https://doi.org/10.1038/ s41598-023-34966-3

Oz, F., San, O., Kara, K.: An efficient quantum partial differential equation solver with chebyshev points. Scientific Reports 13, 7767 (2023) https://doi.org/10.1038/ s41598-023-34966-3

work page 2023
[18]

Physical Review Letters100(16) (2008) https://doi

Giovannetti, V., Lloyd, S., Maccone, L.: Quantum random access memory. Physical Review Letters100(16) (2008) https://doi. org/10.1103/physrevlett.100.160501

work page doi:10.1103/physrevlett.100.160501 2008
[19]

IEEE Transac- tions on Quantum Engineering1, 1–13 (2020) https://doi.org/10.1109/tqe.2020.2965803

Matteo, O.D., Gheorghiu, V., Mosca, M.: Fault-tolerant resource estimation of quan- tum random-access memories. IEEE Transac- tions on Quantum Engineering1, 1–13 (2020) https://doi.org/10.1109/tqe.2020.2965803

work page doi:10.1109/tqe.2020.2965803 2020
[20]

Quantum machine learning.Nature, 549(7671):195–202, Sep 2017

Biamonte, J., Wittek, P., Pan- cotti, N., Rebentrost, P., Wiebe, N., Lloyd, S.: Quantum machine learn- ing. Nature549(7671), 195–202 (2017) https://doi.org/10.1038/nature23474

work page doi:10.1038/nature23474 2017
[21]

Quantum Science and Technology4(4), 043001 (2019) https://doi.org/10.1088/2058-9565/ab4eb5

Benedetti, M., Lloyd, E., Sack, S., Fioren- tini, M.: Parameterized quantum circuits as machine learning models. Quantum Science and Technology4(4), 043001 (2019) https: //doi.org/10.1088/2058-9565/ab4eb5

work page doi:10.1088/2058-9565/ab4eb5 2019
[22]

Cerezoet al., Variational quantum al- gorithms, Nat

Cerezo, M., Arrasmith, A., Babbush, R., Ben- jamin, S.C., Endo, S., Fujii, K., McClean, J.R., Mitarai, K., Yuan, X., Cincio, L., Coles, P.J.: Variational quantum algorithms. Nature Reviews Physics3(9), 625–644 (2021) https: //doi.org/10.1038/s42254-021-00348-9

work page doi:10.1038/s42254-021-00348-9 2021
[23]

Kyriienko, O., Paine, A.E., Elfving, V.E.: Solving nonlinear differential equations with differentiable quantum circuits. Phys. Rev. A 103, 052416 (2021) https://doi.org/10.1103/ PhysRevA.103.052416

work page 2021
[24]

Frontiers in Applied Mathematics and Statistics8(2022) 17

Markidis, S.: On physics-informed neural net- works for quantum computers. Frontiers in Applied Mathematics and Statistics8(2022) 17

work page 2022
[25]

Entropy26(8) (2024) https://doi.org/10.3390/e26080649

Trahan, C., Loveland, M., Dent, S.: Quantum physics-informed neu- ral networks. Entropy26(8) (2024) https://doi.org/10.3390/e26080649

work page doi:10.3390/e26080649 2024
[26]

In: Workshop Pro- ceedings of the 53rd International Confer- ence on Parallel Processing

Hegde, P.R., Markidis, S.: Quantum physics informed neural networks. In: Workshop Pro- ceedings of the 53rd International Confer- ence on Parallel Processing. ICPP Workshops ’24, pp. 114–115. Association for Comput- ing Machinery, New York, NY, USA (2024). https://doi.org/10.1145/3677333.3678272

work page doi:10.1145/3677333.3678272 2024
[27]

Scientific Reports15(1), 18823 (2025) https://doi.org/ 10.1038/s41598-025-02959-z

Berger, S., Hosters, N., M¨ oller, M.: Trainable embedding quantum physics informed neural networks for solving nonlinear pdes. Scientific Reports15(1), 18823 (2025) https://doi.org/ 10.1038/s41598-025-02959-z

work page doi:10.1038/s41598-025-02959-z 2025
[28]

Panichi, G., Corli, S., Prati, E.: Quantum physics-informed neural networks for multi- variable partial differential equations. Phys. Rev. Appl.25, 014001 (2026) https://doi. org/10.1103/6nh4-yh2y

work page doi:10.1103/6nh4-yh2y 2026
[29]

Machine Learning: Science and Technology6(4), 045053 (2025) https://doi.org/10.1088/2632-2153/ae1c91

Farea, A., Khan, S., Serdar Celebi, M.: Qcpinn: quantum-classical physics- informed neural networks for solving pdes. Machine Learning: Science and Technology6(4), 045053 (2025) https://doi.org/10.1088/2632-2153/ae1c91

work page doi:10.1088/2632-2153/ae1c91 2025
[30]

black hole

Chen, Z., Shaviner, G.G., Chandravamsi, H., Pisnoy, S., Frankel, S.H., Pereg, U.: Quan- tum physics-informed neural networks for maxwell’s equations: circuit design, “black hole” barren plateaus mitigation, and gpu acceleration. Quantum Machine Intelligence 8(1), 21 (2026) https://doi.org/10.1007/ s42484-026-00365-w

work page 2026
[31]

Lantigua, S., Giraldi, G., Portugal, R.: Classical-quantum hybrid architecture for physics-informed neural networks. Phys. Rev. A, (2026) https://doi.org/10.1103/fdd3-qz1s

work page doi:10.1103/fdd3-qz1s 2026
[32]

Schuld, M., Sweke, R., Meyer, J.J.: Effect of data encoding on the expressive power of variational quantum-machine-learning mod- els. Phys. Rev. A103, 032430 (2021) https: //doi.org/10.1103/PhysRevA.103.032430

work page doi:10.1103/physreva.103.032430 2021
[33]

Nature549(7671), 242–246 (2017) https://doi.org/10.1038/nature23879

Kandala, A., Mezzacapo, A., Temme, K., Takita, M., Brink, M., Chow, J.M., Gambetta, J.M.: Hardware-efficient variational quantum eigensolver for small molecules and quantum mag- nets. Nature549(7671), 242–246 (2017) https://doi.org/10.1038/nature23879

work page doi:10.1038/nature23879 2017
[34]

https://arxiv

Chang, S.Y., Cerezo, M.: A Primer on Quan- tum Machine Learning (2025). https://arxiv. org/abs/2511.15969

work page arXiv 2025
[35]

Quantum singular value transformation and beyond: exponential improvements for quantum matrix arithmetics

Gily´ en, A., Su, Y., Low, G.H., Wiebe, N.: Quantum singular value transformation and beyond: exponential improvements for quan- tum matrix arithmetics. In: Proceedings of the 51st Annual ACM SIGACT Sympo- sium on Theory of Computing. STOC 2019, pp. 193–204. Association for Computing Machinery, New York, NY, USA (2019). https://doi.org/10.1145/3313276.331...

work page doi:10.1145/3313276.3316366 2019
[36]

In: The Thirty-eighth Annual Conference on Neu- ral Information Processing Systems (2024)

Yu, Z., Chen, Q., Jiao, Y., Li, Y., Lu, X., Wang, X., Yang, J.Z.: Non- asymptotic approximation error bounds of parameterized quantum circuits. In: The Thirty-eighth Annual Conference on Neu- ral Information Processing Systems (2024). https://openreview.net/forum?id=XCkII8nCt3

work page 2024
[37]

SIAM Review 51(3), 455–500 (2009) https://doi.org/10

Kolda, T.G., Bader, B.W.: Tensor decom- positions and applications. SIAM Review 51(3), 455–500 (2009) https://doi.org/10. 1137/07070111X

work page 2009
[38]

Undergraduate Texts in Mathematics

Logan, J.D.: Applied Partial Differential Equations, 3rd edn. Undergraduate Texts in Mathematics. Springer, Cham (2015). https: //doi.org/10.1007/978-3-319-12493-3

work page doi:10.1007/978-3-319-12493-3 2015
[39]

Computers & Mathematics with Applications132, 48– 62 (2023) https://doi.org/10.1016/j.camwa

Tang, S., Feng, X., Wu, W., Xu, H.: Physics- informed neural networks combined with polynomial interpolation to solve nonlinear partial differential equations. Computers & Mathematics with Applications132, 48– 62 (2023) https://doi.org/10.1016/j.camwa. 2022.12.008

work page doi:10.1016/j.camwa 2023
[40]

In: Feldman, V., Rakhlin, A., Shamir, O

Cohen, N., Sharir, O., Shashua, A.: On the expressive power of deep learning: A 18 tensor analysis. In: Feldman, V., Rakhlin, A., Shamir, O. (eds.) Proceedings of the 29th Conference on Learning Theory (COLT 2016). Proceedings of Machine Learning Research, vol. 49, pp. 698–728. PMLR, New York, NY, USA (2016). https://proceedings.mlr.press/v49/cohen16.html

work page 2016
[41]

IEEE Transac- tions on Pattern Analysis and Machine Intel- ligence, 1–1 (2021) https://doi.org/10.1109/ tpami.2021.3058891

Chrysos, G.G., Moschoglou, S., Bouritsas, G., Deng, J., Panagakis, Y., Zafeiriou, S.P.: Deep polynomial neural networks. IEEE Transac- tions on Pattern Analysis and Machine Intel- ligence, 1–1 (2021) https://doi.org/10.1109/ tpami.2021.3058891

work page arXiv 2021
[42]

In: Antonacopoulos, A., Chaudhuri, S., Chel- lappa, R., Liu, C.-L., Bhattacharya, S., Pal, U

Vemuri, S.K., B¨ uchner, T., Niebling, J., Den- zler, J.: Functional tensor decompositions for physics-informed neural networks. In: Antonacopoulos, A., Chaudhuri, S., Chel- lappa, R., Liu, C.-L., Bhattacharya, S., Pal, U. (eds.) Pattern Recognition, pp. 32–46. Springer, Cham (2025)

work page 2025
[43]

Cambridge University Press, New York, USA (1981)

Powell, M.J.D.: Approximation Theory and Methods. Cambridge University Press, New York, USA (1981)

work page 1981
[44]

Dover Publications, Mineola, New York (2001)

Boyd, J.P.: Chebyshev and Fourier Spec- tral Methods, 2nd edn. Dover Publications, Mineola, New York (2001)

work page 2001
[45]

Finance Research Letters69, 106185 (2024) https://doi.org/10

Kopeliovich, Y., Pokojovy, M.: Portfolio opti- mization with feedback strategies based on artificial neural networks. Finance Research Letters69, 106185 (2024) https://doi.org/10. 1016/j.frl.2024.106185

work page arXiv 2024
[46]

Mitarai, M

Mitarai, K., Negoro, M., Kitagawa, M., Fujii, K.: Quantum circuit learning. Physi- cal Review A98(3) (2018) https://doi.org/ 10.1103/physreva.98.032309

work page doi:10.1103/physreva.98.032309 2018
[47]

Encyclopaedia of Math- ematical Sciences, vol

Nikol’skii, S.M.: Analysis III: Spaces of Differ- entiable Functions. Encyclopaedia of Math- ematical Sciences, vol. 91. Springer, Berlin, Heidelberg (2013). https://doi.org/10.1007/ 978-3-642-36247-3

work page 2013
[48]

Quantum Computing in the NISQ era and beyond.Quantum, 2:79, August 2018

Preskill, J.: Quantum computing in the nisq era and beyond. Quantum2, 79 (2018) https: //doi.org/10.22331/q-2018-08-06-79

work page internal anchor Pith review doi:10.22331/q-2018-08-06-79 2018
[49]

Li, M., Guo, F.-Q., Jin, Z., Yan, L.-L., Liang, E.-J., Su, S.-L.: Multiple-qubit controlled unitary quantum gate for rydberg atoms using shortcut to adiabaticity and optimized geometric quantum operations. Phys. Rev. A 103, 062607 (2021) https://doi.org/10.1103/ PhysRevA.103.062607

work page 2021
[50]

Nature Communications9(1) (2018) https:// doi.org/10.1038/s41467-018-07090-4

McClean, J.R., Boixo, S., Smelyanskiy, V.N., Babbush, R., Neven, H.: Barren plateaus in quantum neural network training landscapes. Nature Communications9(1) (2018) https:// doi.org/10.1038/s41467-018-07090-4

work page doi:10.1038/s41467-018-07090-4 2018
[51]

https://arxiv.org/abs/2112.00811

Cotler, J., Huang, H.-Y., McClean, J.R.: Revisiting dequantization and quan- tum advantage in learning tasks (2021). https://arxiv.org/abs/2112.00811

work page arXiv 2021
[52]

1038/s41467-025-63099-6

Cerezo, M., Larocca, M., Garc´ ıa-Mart´ ın, D., Diaz, N.L., Braccia, P., Fontana, E., Rudolph, M.S., Bermejo, P., Ijaz, A., Thanasilp, S., Anschuetz, E.R., Holmes, Z.: Does provable absence of barren plateaus imply classical simulability? Nature Com- munications16(1) (2025) https://doi.org/10. 1038/s41467-025-63099-6

work page 2025
[53]

In: Proceedings of the 52nd Annual ACM SIGACT Symposium on Theory of Computing

Chia, N.-H., Gily´ en, A., Li, T., Lin, H.- H., Tang, E., Wang, C.: Sampling-based sublinear low-rank matrix arithmetic framework for dequantizing quantum machine learning. In: Proceedings of the 52nd Annual ACM SIGACT Symposium on Theory of Computing. STOC 2020, pp. 387–400. Association for Computing Machinery, New York, NY, USA (2020). https://doi.org/1...

work page doi:10.1145/3357713.3384314 2020
[54]

In: International Conference on Learning Representations (ICLR) 2020 (2020).https://arxiv.org/abs/1904.00962

You, Y., Li, J., Reddi, S., Hseu, J., Kumar, S., Bhojanapalli, S., Song, X., Demmel, J., Keutzer, K., Hsieh, C.-J.: Large batch opti- mization for deep learning: Training bert in 76 minutes. In: International Conference on Learning Representations (ICLR) 2020 (2020).https://arxiv.org/abs/1904.00962

work page arXiv 2020
[55]

Schreiber, F.J., Eisert, J., Meyer, J.J.: Clas- sical surrogates for quantum learning models. Phys. Rev. Lett.131, 100803 (2023) https: //doi.org/10.1103/PhysRevLett.131.100803 19

work page doi:10.1103/physrevlett.131.100803 2023
[56]

Journal of Mathematics and Physics6, 164–189 (1927)

Hitchcock, F.L.: The expression of a tensor or a polyadic as a sum of products. Journal of Mathematics and Physics6, 164–189 (1927)

work page 1927
[57]

Physi- cal Review A52(5), 3457–3467 (1995) https: //doi.org/10.1103/physreva.52.3457

Barenco, A., Bennett, C.H., Cleve, R., DiVin- cenzo, D.P., Margolus, N., Shor, P., Sleator, T., Smolin, J.A., Weinfurter, H.: Elemen- tary gates for quantum computation. Physi- cal Review A52(5), 3457–3467 (1995) https: //doi.org/10.1103/physreva.52.3457

work page doi:10.1103/physreva.52.3457 1995
[58]

Physical Review A106(4) (2022) https:// doi.org/10.1103/physreva.106.042602

Silva, A.J., Park, D.K.: Linear-depth quan- tum circuits for multiqubit controlled gates. Physical Review A106(4) (2022) https:// doi.org/10.1103/physreva.106.042602

work page doi:10.1103/physreva.106.042602 2022
[59]

Quantum Info

M¨ ott¨ onen, M., Vartiainen, J.J., Bergholm, V., Salomaa, M.M.: Transformation of quan- tum states using uniformly controlled rota- tions. Quantum Info. Comput.5(6), 467–473 (2005) 20 Appendix A Tensor-decomposed polynomial derivation We first recall the notion oftensor rank decomposition[37]. Given indexn m ∈I m,∀m∈[D] withDas an integer, letA ∈F I1×I2×...

work page 2005
[60]

ifLis even/odd thenp(x) must be an even/odd function), 3.∀x∈[−1,1],|p(x)| ≤1

deg(p(x))≤L, 2.p(x) has parityLmod 2 (i.e. ifLis even/odd thenp(x) must be an even/odd function), 3.∀x∈[−1,1],|p(x)| ≤1. The circuit diagram ofU θ(x) is shown in Figure 2. We have several different methods to estimate ⟨+|Uθ(x)|+⟩to obtainp(x). In this work, we consider only the Hadamard test method, whose circuit diagram is shown in Figure C1. Fig. C1Hada...

work page
[61]

ProofLetp(x) = 1 2 podd(x) + 1 2 peven(x) wherep even (x) =p(x) +p(−x) andp odd (x) =p(x)−p(−x)

deg(p(x))≤L, 2.∀x∈[−1,1],|p(x)| ≤ 1 2. ProofLetp(x) = 1 2 podd(x) + 1 2 peven(x) wherep even (x) =p(x) +p(−x) andp odd (x) =p(x)−p(−x). We have deg(peven (x))≤L,deg(p odd(x))≤L−1 whenLis even, and deg(p even (x))≤L−1,deg(p odd(x))≤LwhenL is odd. We also know that|p odd(x)| ≤1 and|p even(x)| ≤1 since|p(x)| ≤ 1

work page
[62]

□ In fact,∀x∈[−1,1], the condition|p(x)| ≤ 1 2 is stronger than necessary to guarantee the existence of a quantum model

Hence from the Lemma 5, we know there existθ odd ∈R L,θ even ∈R L+1 (Lis even) orθ odd ∈R L+1,θ even ∈R L (Lis odd) such that podd(x) =⟨+|U θodd(x)|+⟩ and peven(x) =⟨+|U θeven(x)|+⟩ then we can build|0⟩⟨0| ⊗U θodd +|1⟩⟨1| ⊗U θeven such that ⟨+|⊗2(|0⟩⟨0| ⊗U θodd +|1⟩⟨1| ⊗U θeven)(x)|+⟩⊗2 = 1 2 ⟨+|Uθodd(x)|+⟩+ 1 2 ⟨+|Uθeven(x)|+⟩= 1 2 podd(x) + 1 2 peven(x)...

work page
[63]

Then we can do the Hadamard test to estimatep(x) =⟨+| ⊗2(|0⟩⟨0| ⊗U θ1(x) +|1⟩⟨1| ⊗U θ2(x))|+⟩⊗2 and analyze the circuit complexity. Proposition 1.For any real polynomialp(x)∈R[x]that satisfiesdeg(p(x))≤Land∀x∈ [−1,1],|p(x)| ≤ 1 2, there exists a quantum modelQthat consists of a PQCW p(x)and an observableZ (0) such that fQ(x) :=⟨0|W † p (x)Z (0)Wp(x)|0⟩=p(...

work page