Pith. sign in

REVIEW 5 major objections 6 minor 1 cited by

Technical report on a quantum-inspired solver for simulating compressible flows

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a tensor-network solver that would make 2D compressible-flow simulation cost scale polylogarithmically with mesh size, and reports that its current 2D iteration does not yet converge.

desk verdict Honest report, but the 2D solver's non-convergence undercuts the paper's central polylog-scaling claim. read the letter →

arxiv 2506.03833 v1 pith:WBXP7IH3 submitted 2025-06-04 physics.flu-dyn physics.comp-phquant-ph

Raghavendra Dheeraj Peddinti , Stefano Pisoni , Egor Tiunov , Alessandro Marini , Leandro Aolita This is my paper
classification physics.flu-dynphysics.comp-phquant-ph
keywords tensornetworksmatrixproductstatesquantum-inspiredalgorithmsEulerequationscompressibleflowNACA0012airfoiltransonicDMRGlinearsolver
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

The paper tries to establish that matrix product states (MPSs), a compressed representation from quantum physics, can serve as the backbone of a classical solver for the 2D steady Euler equations around an immersed airfoil, with runtime and memory that grow linearly in the number of mesh-resolution bits and polynomially in bond dimension rather than exponentially in mesh size. The proposed pipeline encodes fields as MPSs, differential operators as matrix product operators, and the airfoil geometry as an approximate MPS mask; the nonlinear steady equations are linearized and solved iteratively with a DMRG-based linear solver, and an oracle extracts coarse-grained or sampled output without expanding the MPS to a dense vector. The authors support the plausibility of low bond dimensions with compression experiments on converged Ansys Fluent fields and with a 1D Sod shock-tube benchmark. They also report that the 2D iterative solver did not converge and that its last-iterate fields do not match the Fluent reference, so the scaling claim currently rests on components whose 2D assembly remains to be stabilized.

What carries the argument

The machinery is the matrix product state / tensor train representation: a mesh function on 2^N points is written as a product of N matrices of bond dimension χ, so the full vector costs O(N $χ^{2}$) parameters. Around this sit four components: TT-cross-based MPS masks that encode the airfoil indicator function with tunable sharpness; MPO encodings of central finite-difference operators; a DMRG iterative linear-system solver applied to the linearized, artificially dissipated Euler equations; and an MPS solution oracle providing coarse-grained evaluation and $\ell^2$-norm-based pixel sampling. The complexity table lists O(N $χ^{3}$) for mask generation and O(N $χ^{4}$) for mask application and linear solves, which is what converts an exponential-in-N dense problem into a polylogarithmic one for fixed χ.

What would settle it

Run the Section II C fixed-point iteration to convergence on a 2048×1024 mesh with χ capped, and compare the final pressure and velocity fields to a converged finite-volume reference on the same geometry: the central claim is settled if the iteration converges and the fields agree within discretization error while the MPS parameter count stays polynomial in N and χ; it is falsified if the iteration keeps diverging, if the required χ grows exponentially with mesh resolution, or if the converged fields disagree with the reference.

Watch

Extended reading notes

Core claim

The central claim is that compressible-flow solutions with shock waves, such as transonic flow around a NACA 0012 airfoil, have an efficient matrix-product-state representation, and that all subroutines needed to solve the discretized Euler equations—geometry masking, finite-difference operators, linear solves, and solution extraction—can be performed inside the MPS format at cost O(N $χ^{4}$) in the worst case, where 2^N is the number of mesh points and χ is the bond dimension. On a 2048 × 1024 mesh with χ = 100, the framework produces flow fields that show shock-like features, but the iteration did not converge and the fields do not match Ansys Fluent; the paper therefore presents this as a proposal with component-level evidence rather than a working end-to-end 2D solver.

Load-bearing premise

The load-bearing premise is that the iterative scheme of Section II C converges to the physical transonic solution; without convergence, the polylog-scaling solver has no working 2D realization, and the paper itself reports that this premise failed in the current implementation.

Editorial extensions

If this is right

  • If the convergence issue is fixed, the same pipeline would simulate the 2D aerofoil case on meshes far beyond dense-vector reach, since memory would scale as N χ² rather than 2^N.
  • The MPS solution oracle would let aerodynamic design optimization query pressure or vorticity statistics without ever forming the dense field.
  • The same framework transfers to the aero-acoustic case study and to other PDEs, since the mask and oracle components are equation-agnostic.
  • On very refined meshes, setting the artificial dissipation μ = h recovers the inviscid equations, so the solver can approach the true Euler limit as resolution grows.
  • The 1D Sod shock-tube results show that a single shock front is captured by an MPS with a small fraction of the dense parameters, supporting the low-bond-dimension premise in 1D.

Reading between the lines

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

  • The reported non-convergence is consistent with the segregated pressure-velocity update and the mask's no-slip treatment at a boundary that should satisfy a slip condition; a natural test is to replace the mask with a normal-velocity-zeroing projection and rerun the same linear solver.
  • Low bond dimension of converged Fluent steady states does not imply low bond dimension along the iteration trajectory; the operator (Ô_adv + Ô_diss) may mix scales transiently even when the fixed point is compressible, so a practical solver may need adaptive bond-dimension control during iterations.
  • If a stable 2D formulation is found, the same complexity argument suggests quantum hardware could later accelerate the linear-solve subroutine, making the framework a bridge between quantum-inspired classical and fault-tolerant quantum CFD.
  • The compression evidence for 2D shock fields suggests a standalone use as a lossy compression format for CFD output and for Monte-Carlo-style sampling in optimization loops, independent of whether the iterative solver converges.
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

5 major / 6 minor

Summary. The manuscript proposes a tensor-network (MPS/MPO) solver for the two-dimensional steady Euler equations around a NACA 0012 airfoil, and claims in the abstract that the solver scales polylogarithmically with mesh size in both runtime and memory. The main ingredients are a TT-cross MPS mask for the immersed geometry, a DMRG-based linear solver, a segregated fixed-point iteration with artificial dissipation, a Chorin-type projection step for continuity, and an MPS solution oracle for coarse-grained evaluation and pixel sampling. The paper reports a 1D Sod shock-tube benchmark that matches the analytical solution, a compression study of a Fluent steady-state solution (Fig. 4), and an asymptotic complexity table for individual subroutines. It also explicitly states, in Section III, that the 2D solver did not converge and that its final fields do not match the Ansys Fluent solution.

Significance. If the central claim were substantiated, a classical tensor-network solver with polylogarithmic scaling in mesh size for transonic 2D Euler flows would be a notable contribution to CFD and quantum-inspired computing. The manuscript has two genuine positive elements: the 1D Sod benchmark in Appendix A demonstrates that a shock profile can be represented and solved accurately in MPS form with substantial compression, and Fig. 4 provides evidence that the Fluent steady-state solution is compressible to low bond dimension. However, those results address necessary conditions, not the sufficiency of the proposed iterative scheme. The paper's own admission of non-convergence in Section III, combined with the absence of any iteration-count or bond-dimension-growth analysis, means that the abstract's scaling claim is not supported by an end-to-end working 2D solver. The significance of the work is therefore conditional on future convergence and stability analysis that the current manuscript does not provide.

major comments (5)
  1. [Section III, Fig. 5] The central claim of the abstract is not supported by the reported implementation. Section III states that 'the MPS solver failed to converge' and presents the 'last stable iteration' whose pressure and velocity fields 'do not match those of Ansys Fluent.' Since the fixed-point iteration is the outer loop that produces the solution, a non-converged iteration cannot be used as evidence for a solver that 'scales polylogarithmically with the mesh size.' This is a load-bearing gap, not a presentation issue.
  2. [Section II C, Eq. (5), and Table I] The complexity analysis in Table I itemizes per-subroutine costs but contains no term for the number of outer fixed-point iterations M(N). The total runtime is at least M(N) times the per-iteration linear-solver cost O(N chi^4), and the manuscript provides no bound, empirical measurement, or even a qualitative statement about how M(N) grows with N. Without this, the claimed polylogarithmic scaling for the full solver is unestablished.
  3. [Section II C, Projection step] The projection of the flux components rho u and rho v onto their divergence-free manifold via a DMRG-based Poisson solver is taken from the incompressible Chorin projection method of Ref. [1]. The manuscript does not justify the validity of this operation for the compressible Euler equations, where continuity is a hyperbolic transport equation rather than a divergence-free constraint on the flux. This may be a source of the observed non-convergence, and the claim requires either a theoretical justification or a numerical demonstration that the projected iterates solve the correct discrete Euler system.
  4. [Section II B and Section IV] The mask implementation enforces all velocity components to be zero at the object boundary, while for inviscid Euler only the normal component should vanish. The authors acknowledge this in Section II B and list it in the Outlook as a likely cause of instability. Because boundary treatment directly changes the physical solution the iteration is seeking, this is not a minor numerical detail; it undermines the fidelity of the 2D result and must be addressed before the solver can be claimed to simulate the stated transonic airfoil problem.
  5. [Section II E and Section IV Outlook] The claim that the MPS representation keeps bond dimensions low is not sufficient to establish efficiency. Section IV states that 'it is in general unknown if chi can be kept under control' during iteration. Since both runtime and memory scale polynomially in chi, any growth of chi with mesh resolution could destroy the polylogarithmic scaling. The paper provides no bound on chi(N) for the iterative solver, only a compression study of the converged Fluent solution, which does not characterize the intermediate states generated by the fixed-point iteration.
minor comments (6)
  1. [Abstract and Section I A] The abstract claims 'polylogarithmically with the mesh size' while Section I A states runtime and memory 'scale linearly in N' for a mesh of 2^N points; these statements are consistent only if the relationship between N and mesh size is stated explicitly. The manuscript should use one convention throughout to avoid confusion.
  2. [Introduction] The company name in the challenge is written as 'BWM' but appears to refer to BMW; please correct the typo.
  3. [Section I A] The text refers to 'the 3D turbulence case study that is presented in App. A', but Appendix A actually presents the 1D Sod shock-tube case. The cross-reference is incorrect.
  4. [Appendix A 1, Fig. 7] The horizontal axes of the three panels are labeled with values 0 to 30000 but no axis title is given. These appear to be spatial grid indices for a 2^15-point domain, not physical length or time; please add axis labels.
  5. [Section II C and Section III] The artificial dissipation coefficient mu is a key free parameter, but the manuscript does not report the value used in the 2D simulations or in the 1D Sod benchmark. Reporting mu and the resulting effective Reynolds number would improve reproducibility.
  6. [Section III] The DMRG linear-system solver is described as 'approximately solved via iterative approaches' but no tolerances, sweep counts, or stopping criteria are given. These details would be needed to reproduce the results.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the scaling claim rests on standard MPS complexity results and external benchmarks, not on any fitted value or self-referential reduction.

full rationale

The paper does not exhibit a load-bearing circular reduction. The central scaling claim in Table I is assembled from standard tensor-network complexity results (TT-cross [13], TT compression [23,24], DMRG linear-system solver [9,17]), none of which are fitted to the paper's own outputs. The 1D Sod benchmark in Appendix A.1 is checked against the analytical solution (Fig. 7), and the 2D case is compared with an independent Ansys Fluent baseline (Fig. 5), so the solver's output is not being validated by its own inputs. The self-citations to the authors' prior work [1] supply the MPS mask construction and the DMRG-based Poisson solver, but these are described subroutines used inside the non-converging steady-state iteration; they are not the claimed prediction, and no equation in the paper reduces a predicted quantity to a fitted parameter by construction. The paper also openly reports that the 2D iteration did not converge (Sec. III and Outlook), which is a correctness and robustness limitation rather than a circularity. Overall, the derivation chain is self-contained against external checks, and the only self-referential elements are minor and non-load-bearing.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on several hand-chosen parameters and unproven modeling assumptions. The most consequential is the convergence of the fixed-point iteration, which the paper itself reports failed. No new physical entities are introduced.

free parameters (4)
  • Artificial dissipation coefficient µ = Not reported; theory suggests µ = h (grid size)
    Introduced in Sec II C to stabilize central differences at shocks; controls shock smearing and is not determined by a first-principles criterion.
  • Mask sharpness parameter α = alpha = 10^4 in the 2D airfoil case (Sec III)
    Defines the smooth mask m(x) in Eq. 2; larger alpha gives a sharper geometry but affects MPS bond dimension and is chosen by hand.
  • Maximum MPS bond dimension χ = chi = 100 for the 21-bit 2D simulation; chi = 96 for the 25-bit mask (Sec III)
    Truncation threshold controls accuracy versus cost; the solver's complexity scales polynomially in chi and the paper assumes a small chi persists during iteration without proving it.
  • Initial flow field guesses for the fixed-point iteration = Not specified
    The iteration in Eq. 5 depends on ρ(0), (ρu)(0), (ρv)(0), and (E+p)(0); no values are given and convergence was not reached.
assumptions (5)
  • domain assumption The inviscid 2D Euler equations in steady conservative form (Eq. 4) adequately model transonic flow around NACA0012.
    Standard CFD modeling assumption invoked in Sec II C; the case study defines the target physics.
  • domain assumption MPS/MPO representations with central finite differences and TT-cross masks provide accurate compressed representations of flows and geometries.
    Used throughout Sec II A-B; the 2D low-bond-dimension claim is supported only by compressing an Ansys Fluent solution (Fig 4), not by the solver's own output.
  • ad hoc to paper The segregated fixed-point iteration of the linearized Euler system converges to the steady state.
    Sec II C, Eq. 5; contradicted by Sec III, which reports the 2D solver did not converge. This is the load-bearing assumption that failed.
  • ad hoc to paper Constant-coefficient artificial dissipation stabilizes the central-difference shock computation.
    Sec II C and II E; the paper motivates it via equivalence to upwinding, but the 2D run did not converge, so this stabilization is insufficient as implemented.
  • ad hoc to paper Projection of fluxes onto the divergence-free manifold (Chorin projection from [1]) is valid for enforcing continuity in compressible Euler equations.
    Sec II C; Chorin projection is an incompressible-flow technique and its use for compressible steady-state continuity is not derived in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Technical report on a quantum-inspired solver for simulating compressible flows." pith.science (2026). https://pith.science/paper/WBXP7IH3

@misc{pith2026250603833,
  author       = {Pith},
  title        = {Pith review of: Technical report on a quantum-inspired solver for simulating compressible flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WBXP7IH3}},
  note         = {Machine review of arXiv:2506.03833}
}
read the original abstract

This document presents a quantum-inspired solver for 2D Euler equations, accepted at the final phase of the Airbus-BWM Group Quantum Computing Challenge (ABQCC) 2024. We tackle the case study of Quantum Solvers for Predictive Aeroacoustic and Aerodynamic modeling tasks. We propose a tensor network based solver that scales polylogarithmically with the mesh size, in both runtime and memory. This provides a promising avenue for tackling the curse of dimensionality that plagues the direct numerical simulations in the field of computational fluid dynamics.

Figures

Figures reproduced from arXiv: 2506.03833 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Multiscale passive scalar turbulence in a compressed subspace via tensor trains

    physics.flu-dyn 2026-07 conditional novelty 5.0 of 10

    A hybrid tensor train that keeps coarse scales exact and compresses only fine scales reproduces intermittent small-scale statistics of a 2D passive scalar better than standard TT, Galerkin, or wavelet compression at e...

Reference graph

Works this paper leans on

31 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [1]

    Technical report on a quantum-inspired solver for simulating compressible flows

    1D Euler equations 10 ∗ egor.tiunov@tii.ae arXiv:2506.03833v1 [physics.flu-dyn] 4 Jun 2025 2 I. INTRODUCTION While quantum computers promise a drastic advantage over classical ones, fault-tolerant machines are yet to be realized. Because of this, in our recent work [ 1], we developed a quantum-inspired framework, based on tensor networks, for solving the ...

    work page Pith review arXiv 2025

  2. [2]

    R. D. Peddinti, S. Pisoni, A. Marini, P. Lott, H. Argentieri, E. Tiunov, and L. Aolita, Communications Physics 7 (2024)

    work page 2024

  3. [3]

    Schollw¨ ock, Annals of Physics326, 96 (2011)

    U. Schollw¨ ock, Annals of Physics326, 96 (2011)

    2011

  4. [4]

    Or´ us, Annals of Physics349, 117 (2014)

    R. Or´ us, Annals of Physics349, 117 (2014)

    2014

  5. [5]

    Verstraete, V

    F. Verstraete, V. Murg, and J. I. Cirac, Advances in Physics 57, 143 (2008). 9

    work page 2008

  6. [6]

    Shinaoka, M

    H. Shinaoka, M. Wallerberger, Y. Murakami, K. Nogaki, R. Sakurai, P. Werner, and A. Kauch, Physical Review X 13, 021015 (2023)

    work page 2023

  7. [7]

    Kiffner and D

    M. Kiffner and D. Jaksch, arXiv:2303.03010 [Physics.flu-dyn] (2023), -

    arXiv 2023

  8. [8]

    Ye and N

    E. Ye and N. F. G. Loureiro, Phys. Rev. E 106, 035208 (2022)

    work page 2022

Show all 31 references
  1. [9]

    Latorre, abs/quant-ph/0510031 (2005)

    J. Latorre, abs/quant-ph/0510031 (2005)

    2005 arXiv

  2. [10]

    Gourianov, M

    N. Gourianov, M. Lubasch, S. Dolgov, Q. Y. van den Berg, H. Babaee, P. Givi, M. Kiffner, and D. Jaksch, Nature Computational Science 2, 30 (2022)

    2022

  3. [11]

    V. A. Kazeev and B. N. Khoromskij, SIAM Journal on Matrix Analysis and Applications 33, 742 (2012)

    2012

  4. [12]

    I. V. Oseledets, Constructive Approximation 37, 1–18 (2012)

    2012

  5. [13]

    J. J. Garc ´ ıa-Ripoll, Quantum5, 431 (2021)

    2021

  6. [14]

    Oseledets and E

    I. Oseledets and E. Tyrtyshnikov, Linear Algebra and its Applications 432, 70 (2010)

    2010

  7. [15]

    Z. Li, N. B. Kovachki, C. Choy, B. Li, J. Kossaifi, S. P. Otta, M. A. Nabian, M. Stadler, C. Hundt, K. Azizzadenesheli, and A. Anandkumar, in Thirty-seventh Conference on Neural Information Processing Systems(2023)

    2023

  8. [16]

    C.-W. Shu, Essentially non-oscillatory and weighted essentially non-oscillatory schemes for hyperbolic conservation laws, in Advanced Numerical Approximation of Nonlinear Hyperbolic Equations: Lectures given at the 2nd Session of the Centro Internazionale Matematico Estivo (C....

    1997

  9. [17]

    Jameson, W

    A. Jameson, W. Schmidt, and E. Turkel, Numerical solution of the euler equations by finite volume methods using runge kutta time stepping schemes (1981)

    1981

  10. [18]

    I. V. Oseledets and S. V. Dolgov, SIAM Journal on Scientific Computing 34, A2718 (2012)

    2012

  11. [19]

    Sirignano and K

    J. Sirignano and K. Spiliopoulos, Journal of Computational Physics 375, 1339 (2018)

    2018

  12. [20]

    A. J. Ferris and G. Vidal, Phys. Rev. B 85, 165146 (2012)

    2012

  13. [21]

    Z.-Y. Han, J. Wang, H. Fan, L. Wang, and P. Zhang, Physical Review X 8, 031012 (2018)

    2018

  14. [22]

    Swanson, R

    R. Swanson, R. Radespiel, E. Turkel, R. Swanson, R. Radespiel, and E. Turkel, in 13th Computational Fluid Dy- namics Conference (1997) p. 1945

    1997

  15. [23]

    Lindsey, arXiv preprint arXiv:2311.12554 (2023)

    M. Lindsey, arXiv preprint arXiv:2311.12554 (2023)

    2023 arXiv

  16. [24]

    Al Daas, G

    H. Al Daas, G. Ballard, P. Cazeaux, E. Hallman, A. Miedlar, M. Pasha, T. W. Reid, and A. K. Saibaba, SIAM Journal on Scientific Computing 45, 10.1137/21m1451191 (2023)

    2023 doi

  17. [25]

    Z. Sun, J. Huang, C. Xiao, and C. Yang, arXiv preprint arXiv:2410.04385 (2024)

    2024

  18. [26]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, and A. Anandkumar, Fourier neural operator for parametric partial differential equations (2021), arXiv:2010.08895 [cs.LG]

    2021 arXiv

  19. [27]

    Farge, Annual Review of Fluid Mechanics 24, 395 (1992)

    M. Farge, Annual Review of Fluid Mechanics 24, 395 (1992)

    1992

  20. [28]

    D. P. Truong, M. I. Ortega, I. Boureima, G. Manzini, K. Ø. Rasmussen, and B. S. Alexandrov, Tensor networks for solving realistic time-independent boltzmann neutron transport equation (2023), arXiv:2309.03347 [math.NA]

    2023 arXiv

  21. [29]

    Lubasch, P

    M. Lubasch, P. Moinier, and D. Jaksch, Journal of Computational Physics 372, 587 (2018)

    2018

  22. [30]

    Kornev, S

    E. Kornev, S. Dolgov, K. Pinto, M. Pflitsch, M. Perelshtein, and A. Melnikov, arXiv preprint arXiv:2305.10784 (2023). 10 Appendix A: Additional results for 1D case In this appendix we provide additional evidence to support the validity of our approach. Here, we discuss how a s...

    2023 arXiv

  23. [31]

    They are described by the one-dimensional Euler equations, i.e., ⃗ u= u

    1D Euler equations We study the Sod shock tube problem, a canonical test case for CFD solvers. They are described by the one-dimensional Euler equations, i.e., ⃗ u= u. The initial conditions of the system are shown in Eq. A1 where the subscripts L and R refer to the left and r...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.