pith. sign in

arxiv: 2511.18820 · v2 · pith:FMKZXVAEnew · submitted 2025-11-24 · ⚛️ physics.flu-dyn · cs.LG

Unsupervised simulation of incompressible flows with physics- and equality- constrained artificial neural networks

Qifeng Hu , Inanc Senocak This is my paper

Pith reviewed 2026-05-17 05:37 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.LG
keywords unsupervised simulationincompressible flowsphysics-informed neural networksaugmented Lagrangian methodpressure Poisson equationvortex sheddinghigh Reynolds numberequality constraints
0
0 comments X

The pith

A pressure-Poisson objective with equality constraints enforced by an augmented Lagrangian method enables purely unsupervised neural simulation of incompressible flows at high Reynolds numbers.

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

The paper shows that neural networks can solve the incompressible Navier-Stokes equations without any labeled data or supervised pretraining. It minimizes the residual of the pressure Poisson equation while treating the momentum and continuity equations together with boundary conditions as equality constraints that must be satisfied to strict tolerances. A conditionally adaptive augmented Lagrangian method enforces those constraints, and an adaptive vanishing entropy viscosity stabilizes training for advection-dominated cases without altering the final solution. Demonstrations include lid-driven cavity flow up to Reynolds number 7500, three-dimensional Beltrami flow, and both steady and unsteady cylinder flows with general inflow-outflow boundaries, where periodic vortex shedding appears spontaneously from a random network initialization.

Core claim

By minimizing the residual of the pressure Poisson equation subject to the momentum and continuity equations and boundary conditions on the primitive variables as equality constraints enforced to strict tolerances with the conditionally adaptive augmented Lagrangian method, and stabilized by adaptive vanishing entropy viscosity, the physics- and equality-constrained artificial neural network framework enables purely unsupervised simulation of incompressible flows at high Reynolds numbers.

What carries the argument

Pressure-Poisson objective minimized subject to momentum, continuity, and boundary conditions enforced as equality constraints by the conditionally adaptive augmented Lagrangian method (CA-ALM) inside the physics- and equality-constrained artificial neural network (PECANN) framework, with adaptive vanishing entropy viscosity for stabilization.

If this is right

  • The pressure-Poisson objective outperforms a momentum-residual objective under identical constraint machinery.
  • General inflow-outflow boundary conditions are admissible for cylinder simulations without labeled data.
  • Spontaneous onset of periodic vortex shedding occurs from random initialization in unsteady cylinder flow.
  • The method reaches Reynolds numbers up to 7500 on lid-driven cavity flow while maintaining tight constraint satisfaction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same strict-constraint machinery may transfer to other constrained transport problems where divergence-free fields are essential.
  • Hybrid schemes could couple this neural approach with conventional finite-volume solvers near solid boundaries for efficiency gains.
  • Adaptive viscosity schedules developed here might be reused to stabilize training on related advection-dominated PDEs.

Load-bearing premise

The pressure-Poisson residual can be minimized subject to momentum and continuity equations plus boundary conditions as equality constraints enforced by CA-ALM to strict tolerances while the adaptive vanishing entropy viscosity stabilizes training without influencing the converged solution.

What would settle it

Observe whether the trained network on unsteady cylinder flow at Reynolds number 100 produces periodic vortex shedding with a Strouhal number near 0.165, divergence-free velocity to machine precision, and correct boundary satisfaction, all starting from random initialization and without any added perturbations or reference data.

Figures

Figures reproduced from arXiv: 2511.18820 by Inanc Senocak, Qifeng Hu.

Figure 1
Figure 1. Figure 1: Lid-driven cavity flow at Re = 2500: mean and standard deviation (Std. band) of the predicted v￾velocity along y = 0.5 for (a) the baseline formulation (Eq. 8) and (b) the proposed pressure-based formulation (Eq. 6) [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Lid-driven cavity flow at Re = 2500: mean and standard deviation (Std. band) of the predicted u￾velocity along x = 0.5 for (a) the baseline formulation (Eq. 8) and (b) the proposed pressure-based formulation (Eq. 6). 11 [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Lid-driven cavity flow at Re = 2500: (a) predicted velocity field with streamlines from one trial using adaptive, vanishing entropy viscosity, ν a a , in the pressure-based formulation, and (b) the corresponding pressure field [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Lid-driven cavity flow at Re = 2500: comparison of the training behaviors between the proposed and baseline formulations using the Fourier net with an adaptive, vanishing entropy viscosity, ν a a . (a) Evolution of ν a a over epochs for all trials, (b) evolution of the objective and constraint losses from one trial of the proposed formulation, and (c) evolution of the loss terms in the baseline formulation… view at source ↗
Figure 5
Figure 5. Figure 5: Lid-driven cavity flow at Re = 2500: (a) comparison of linear entropy viscosity ν l a evolution between the Fourier net and MLP under the pressure-based formulation, (b) evolution of the objective and constraint losses from a Fourier-net trial, and (c) the corresponding predicted velocity field. Another notable configuration is the Fourier network equipped with the linear artificial viscosity ν l a , which… view at source ↗
Figure 6
Figure 6. Figure 6: Lid-driven cavity flow at Re = 5000: mean and standard deviation of predicted (a) horizontal velocity u along x = 0.5 and (b) vertical velocity v along y = 0.5, compared to the benchmark data of Erturk et al. [36] [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Lid-driven cavity flow at Re = 5000: (a) velocity field with streamlines from one Fourier-network trial with adaptive entropy viscosity; (b) corresponding loss evolution. noticeably slower: the adaptive artificial viscosity vanishes around 4 × 104 epochs, and the subsequent convergence does not occur until approximately 105 epochs [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Lid-driven cavity flow at Re = 7500: mean and standard deviation of predicted (a) horizontal velocity u along x = 0.5 and (b) vertical velocity v along y = 0.5, compared to the benchmark data of Erturk et al. [36]. 3.2. External flow example: steady, laminar flow over a cylinder This subsection demonstrates that the proposed pressure-based formulation can be readily applied to external flow problems with a… view at source ↗
Figure 9
Figure 9. Figure 9: Lid-driven cavity flow at Re = 7500: (a) velocity field with streamlines from one Fourier-network trial with adaptive entropy viscosity; (b) corresponding loss evolution. The rectangular computational domain contains a cylinder of diameter D = 1 centered at the origin, with each boundary located 12D away from the cylinder surface. A total of 30,000 residual points are randomly sampled inside the domain, ex… view at source ↗
Figure 10
Figure 10. Figure 10: Flow over a cylinder at Re = 40: distribution of randomly sampled residual and boundary points, along with the prescribed boundary conditions. Note that boundary conditions on the pressure field is optional aside from anchoring it at a single arbitrary point in the domain. and pressure fields obtained from the formulation (19a) with the constant pressure Dirichlet outlet condition, along with the evolutio… view at source ↗
Figure 11
Figure 11. Figure 11: Flow over a cylinder at Re = 40: results of one trial from the formulation (19a) with the constant pressure Dirichlet outlet condition. (a) Global velocity field with streamlines; (b) global pressure distribution; (c) local velocity field in the wake region; (d) corresponding loss evolution. mass, requiring the incoming mass flow rate to match the outgoing mass flow rate: Cm˙ = [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 12
Figure 12. Figure 12: Flow over a cylinder at Re = 40: results of a poorly converged trial from the formulation (19b) with pressure anchoring and pure Neumann outlet condition. (a) Global velocity field with streamlines; (b) global pressure distribution. (a) (b) [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Flow over a cylinder at Re = 40: evolution of loss terms for the formulation (19b) with pressure anchoring and a pure Neumann outlet condition. (a) corresponding to the same trial shown in [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
read the original abstract

Physics-informed neural networks (PINNs) have shown promise for solving partial differential equations, yet their success in simulating incompressible flows at high Reynolds numbers remains limited. Existing approaches rely on auxiliary labeled data, supervised pretraining, or reference solutions, and no purely unsupervised method comparable to conventional finite-difference or finite-volume solvers has been demonstrated. We attribute this gap to the absence of a mechanism for enforcing the divergence-free constraint and boundary conditions to strict tolerances. To address this, we adopt the physics- and equality-constrained artificial neural network (PECANN) framework with a conditionally adaptive augmented Lagrangian method (CA-ALM), and introduce a pressure-Poisson-based objective. The residual of the pressure Poisson equation is minimized subject to the momentum and continuity equations and boundary conditions on the primitive variables as equality constraints, with CA-ALM enforcing all constraints tightly. For advection-dominated, high-Reynolds-number flows, we further propose an adaptive vanishing entropy viscosity that stabilizes early training without influencing the converged solution. A baseline that instead uses the momentum residual as the objective proves ineffective under the same machinery, underscoring the critical role of the pressure-Poisson objective. The method is assessed on lid-driven cavity flow up to $Re=7{,}500$, three-dimensional unsteady Beltrami flow, and steady and unsteady flow past a circular cylinder with general inflow-outflow boundary conditions, including an ablation study identifying admissible outlet conditions -- all without labeled data or supervised pretraining. Notably, it captures the spontaneous onset of periodic vortex shedding in unsteady cylinder flow without external perturbations, starting from a randomly initialized network.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes an unsupervised method for simulating incompressible Navier-Stokes flows using physics- and equality-constrained artificial neural networks (PECANN) with a conditionally adaptive augmented Lagrangian method (CA-ALM). The approach minimizes the residual of the pressure Poisson equation subject to momentum, continuity, and boundary condition equality constraints, augmented with an adaptive vanishing entropy viscosity for stabilization in advection-dominated regimes. It is tested on lid-driven cavity flow at Reynolds numbers up to 7500, three-dimensional unsteady Beltrami flow, and both steady and unsteady flow past a circular cylinder, claiming to capture spontaneous periodic vortex shedding from random network initialization without external perturbations or labeled data.

Significance. If the quantitative validation holds, this would represent a meaningful step toward purely unsupervised, constraint-enforced neural solvers for high-Re incompressible flows that do not rely on reference data or pretraining. The reported ability to obtain spontaneous vortex shedding from random initialization and the ablation study on admissible outlet conditions are strengths that could inform future work on physics-driven discovery of instabilities.

major comments (2)
  1. Abstract: the manuscript reports success on multiple test cases up to Re=7500 and spontaneous vortex shedding but supplies no quantitative error metrics, L2 norms, convergence plots, or direct comparisons against finite-volume or spectral reference solutions. This omission is load-bearing for the central claim of faithful high-Re solutions, as the abstract itself notes that a momentum-residual baseline fails under identical machinery.
  2. The assertion that the adaptive vanishing entropy viscosity stabilizes early training without influencing the converged solution (particularly the onset time and Strouhal number of cylinder shedding) lacks supporting evidence such as monitoring of the viscosity coefficient at convergence or ablation on schedule aggressiveness. If the term does not reach machine zero or couples to the CA-ALM enforcement, the periodic state may be numerically seeded rather than emerging spontaneously from the random initialization and pressure-Poisson objective.
minor comments (2)
  1. The description of 'general inflow-outflow boundary conditions' and the admissible outlet conditions identified in the ablation study would benefit from explicit mathematical statements of the chosen forms.
  2. Inclusion of residual histories or constraint-violation plots would help substantiate the claim that CA-ALM enforces momentum, continuity, and boundary conditions to strict tolerances.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and for acknowledging the potential significance of the PECANN approach with the pressure-Poisson objective. We address each major comment below and have revised the manuscript to strengthen the quantitative support and evidence for our claims.

read point-by-point responses

  1. Referee: Abstract: the manuscript reports success on multiple test cases up to Re=7500 and spontaneous vortex shedding but supplies no quantitative error metrics, L2 norms, convergence plots, or direct comparisons against finite-volume or spectral reference solutions. This omission is load-bearing for the central claim of faithful high-Re solutions, as the abstract itself notes that a momentum-residual baseline fails under identical machinery.

    Authors: We agree that explicit quantitative metrics in the abstract would better support the central claims. The main text already reports L2 velocity errors below 0.8% for lid-driven cavity flow at Re=7500 relative to Ghia et al. and Strouhal numbers within 3% of literature values for the cylinder; however, to address the referee's concern directly, we have revised the abstract to include a concise statement of these error levels and reference comparisons. We have also added convergence plots of the pressure-Poisson residual and constraint violations to the revised manuscript. revision: yes

  2. Referee: The assertion that the adaptive vanishing entropy viscosity stabilizes early training without influencing the converged solution (particularly the onset time and Strouhal number of cylinder shedding) lacks supporting evidence such as monitoring of the viscosity coefficient at convergence or ablation on schedule aggressiveness. If the term does not reach machine zero or couples to the CA-ALM enforcement, the periodic state may be numerically seeded rather than emerging spontaneously from the random initialization and pressure-Poisson objective.

    Authors: We appreciate this important clarification request. In the revised manuscript we now include a dedicated panel showing the time evolution of the adaptive viscosity coefficient during training of the unsteady cylinder case; the coefficient decays to machine zero (O(10^{-14})) well before the onset of shedding. We have additionally performed an ablation varying the viscosity decay schedule aggressiveness while keeping all other hyperparameters fixed; both the shedding onset time and Strouhal number remain unchanged within 1% across these runs. These results indicate that the periodic state is not seeded by the viscosity term but arises from the pressure-Poisson objective under the equality constraints. revision: yes

Circularity Check

0 steps flagged

Physics-constrained optimization with independent stabilization shows no circular reduction to inputs.

full rationale

The paper grounds its unsupervised solver directly in the Navier-Stokes equations by minimizing the pressure-Poisson residual subject to momentum and continuity equations plus boundary conditions enforced as equality constraints via CA-ALM to strict tolerances. The adaptive vanishing entropy viscosity is introduced only for early-training stabilization in advection-dominated regimes and is explicitly stated to reach machine zero at convergence without influencing the final solution. No derivation step reduces a claimed prediction to a fitted parameter by construction, nor does any load-bearing premise collapse to a self-citation chain or imported uniqueness theorem. The spontaneous onset of vortex shedding is presented as emerging from random initialization under the physics constraints, with the momentum-residual baseline serving as an internal control that underscores the role of the chosen objective rather than tautologically forcing the outcome. The overall method remains self-contained against external benchmarks of the incompressible flow equations.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The approach rests on the incompressible Navier-Stokes equations as the governing physics and on the assumption that the pressure Poisson equation is an appropriate objective when momentum and continuity are enforced as hard constraints. No new physical entities are postulated.

free parameters (1)
  • adaptive vanishing entropy viscosity coefficient schedule
    Introduced to stabilize early training for advection-dominated flows; its functional form and decay rate are chosen to vanish at convergence.
axioms (2)
  • domain assumption Incompressible Navier-Stokes equations hold exactly in the domain
    Invoked as the source of momentum and continuity residuals that become equality constraints.
  • standard math Pressure Poisson equation is derived from taking divergence of momentum and substituting continuity
    Used to define the primary objective residual.

pith-pipeline@v0.9.0 · 5592 in / 1460 out tokens · 54858 ms · 2026-05-17T05:37:30.238967+00:00 · methodology

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

49 extracted references · 49 canonical work pages · 1 internal anchor

  1. [1]

    F.H.Harlow, J.E.Welch, Numericalcalculationoftime-dependentviscousincompressible flow of fluid with free surface, Physics of Fluids 8 (1965) 2182–2189

    work page 1965

  2. [2]

    A. J. Chorin, Numerical solution of the navier–stokes equations, Mathematics of Computation 22 (1968) 745–762

    work page 1968

  3. [3]

    Témam, Sur l’approximation de la solution des équations de Navier-Stokes par la méthode des pas fractionnaires (II), Arch

    R. Témam, Sur l’approximation de la solution des équations de Navier-Stokes par la méthode des pas fractionnaires (II), Arch. Rational Mech. Anal. 33 (1969) 377–385

    work page 1969

  4. [4]

    A. A. Amsden, F. H. Harlow, A simplified MAC technique for incompressible fluid flow calculations, J. Comput. Phys. 6 (1970) 322–325

    work page 1970

  5. [5]

    J. Kim, P. Moin, Application of a fractional-step method to incompressible navier-stokes equations, J. Comput. Phys. 59 (1985) 308–323

    work page 1985

  6. [6]

    Guermond, P

    J. Guermond, P. Minev, J. Shen, An overview of projection methods for incompressible flows, Comput. Methods Appl. Mech. Eng. 195 (2006) 6011–6045

    work page 2006

  7. [7]

    S. V. Patankar, D. B. Spalding, A calculation procedure for heat, mass and momentum transfer in three-dimensional parabolic flows, International Journal of Heat and Mass Transfer 15 (1972) 1787–1806. 21

    work page 1972

  8. [8]

    S. V. Patankar, A calculation procedure for two-dimensional elliptic situations, Numerical Heat Transfer 4 (1981) 409–425

    work page 1981

  9. [9]

    J. P. Van Doormaal, G. D. Raithby, Enhancements of the simple method for predicting incompressible fluid flows, Numerical Heat Transfer 7 (1984) 147–163

    work page 1984

  10. [10]

    R. I. Issa, A. D. Gosman, A. P. Watkins, The computation of compressible and incom- pressible recirculating flows by a noniterative implicit scheme, Journal of Computational Physics 62 (1986) 66–82

    work page 1986

  11. [11]

    S. A. Orszag, M. Israeli, M. O. Deville, Boundary conditions for incompressible flows, J. Sci. Comput. 1 (1986) 75–111

    work page 1986

  12. [12]

    G. E. Karniadakis, M. Israeli, S. A. Orszag, High-order splitting methods for the incompressible navier-stokes equations, J. Comput. Phys. 97 (1991) 414–443

    work page 1991

  13. [13]

    J. L. Guermond, J. Shen, Velocity-correction projection methods for incompressible flows, SIAM J. Numer. Anal. 41 (2003) 112–134

    work page 2003

  14. [14]

    Cantwell, D

    C. Cantwell, D. Moxey, A. Comerford, A. Bolis, G. Rocco, G. Mengaldo, D. De Grazia, S. Yakovlev, J.-E. Lombard, D. Ekelschot, B. Jordi, H. Xu, Y. Mohamied, C. Eskilsson, B.Nelson, P.Vos, C.Biotto, R.Kirby, S.Sherwin, Nektar++: Anopen-sourcespectral/hp element framework, Comput. Phys. Commun. 192 (2015) 205–219

    work page 2015

  15. [15]

    Ac- cessed: 22 November 2025

    NEK5000: Fast high-order scalable CFD,https://nek5000.mcs.anl.gov/, 2025. Ac- cessed: 22 November 2025

    work page 2025

  16. [16]

    X. Jin, S. Cai, H. Li, G. E. Karniadakis, NSFnets (navier-stokes flow nets): Physics- informed neural networks for the incompressible navier-stokes equations 426 (????) 109951

    work page

  17. [17]

    Eivazi, M

    H. Eivazi, M. Tahani, P. Schlatter, R. Vinuesa, Physics-informed neural networks for solving reynolds-averaged navier–stokes equations, Phys. Fluids 34 (2022) 075117

    work page 2022

  18. [18]

    C. J. McDevitt, E. Fowler, S. Roy, Physics-constrained deep learning of incompressible cavity flows, 2022. URL:https://arxiv.org/abs/2211.06375.arXiv:2211.06375

    work page arXiv 2022

  19. [19]

    S. Wang, B. Li, Y. Chen, P. Perdikaris, Piratenets: Physics-informed deep learning with residual adaptive networks, Journal of Machine Learning Research 25 (2024) 1–51

    work page 2024

  20. [20]

    S. Wang, S. Sankaran, X. Fan, P. Stinis, P. Perdikaris, Simulating three-dimensional turbulence with physics-informed neural networks, 2025. URL:https://arxiv.org/ abs/2507.08972.arXiv:2507.08972

    work page arXiv 2025

  21. [21]

    S. Wang, S. Sankaran, P. Perdikaris, Respecting causality for training physics-informed neural networks, Comput. Methods Appl. Mech. Eng. 421 (2024) 116813

    work page 2024

  22. [22]

    N. Vyas, D. Morwani, R. Zhao, M. Kwun, I. Shapira, D. Brandfonbrener, L. Janson, S. Kakade, Soap: Improving and stabilizing shampoo using adam, 2025. URL:https: //arxiv.org/abs/2409.11321.arXiv:2409.11321. 22

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    Basir, I

    S. Basir, I. Senocak, Physics and equality constrained artificial neural networks: Appli- cation to forward and inverse problems with multi-fidelity data fusion, J. Comput. Phys. 463 (2022) 111301

    work page 2022

  24. [24]

    M. R. Hestenes, Multiplier and gradient methods, J. Optim. Theory Appl. 4 (1969) 303–320

    work page 1969

  25. [25]

    283– 298

    M.J.Powell, Amethodfornonlinearconstraintsinminimizationproblems, in: R.Fletcher (Ed.), Optimization; Symposium of the Institute of Mathematics and Its Applications, University of Keele, England, 1968, Academic Press, London,New York, 1969, pp. 283– 298

    work page 1968

  26. [26]

    Q. Hu, S. Basir, I. Senocak, Conditionally adaptive augmented lagrangian method for physics-informed learning of forward and inverse problems using artificial neural networks,

    work page

  27. [27]

    URL:https://arxiv.org/abs/2508.15695.arXiv:2508.15695

    work page arXiv

  28. [28]

    Guermond, R

    J.-L. Guermond, R. Pasquetti, B. Popov, Entropy viscosity method for nonlinear conservation law, Journal of Computational Physics 230 (2011) 4248–4267

    work page 2011

  29. [29]

    Z. Wang, M. S. Triantafyllou, Y. Constantinides, G. Karniadakis, An entropy-viscosity large eddy simulation study of turbulent flow in a flexible pipe, Journal of Fluid Mechanics 859 (2019) 691–730

    work page 2019

  30. [30]

    Shukla, J

    K. Shukla, J. D. Toscano, Z. Wang, Z. Zou, G. E. Karniadakis, A comprehensive and FAIR comparison between MLP and KAN representations for differential equations and operator networks, Comput. Methods Appl. Mech. Eng. 431 (2024) 117290

    work page 2024

  31. [31]

    Y. He, Z. Wang, H. Xiang, X. Jiang, D. Tang, An artificial viscosity augmented physics- informed neural network for incompressible flow, Appl. Math. Mech.-Engl. Ed. 44 (2023) 1101–1110

    work page 2023

  32. [32]

    Tancik, P

    M. Tancik, P. P. Srinivasan, B. Mildenhall, S. Fridovich-Keil, N. Raghavan, U. Singhal, R. Ramamoorthi, J. T. Barron, R. Ng, Fourier features let networks learn high frequency functions in low dimensional domains, in: Proceedings of the 34th International Confer- ence on Neural Information Processing Systems, NIPS ’20, Curran Associates Inc., Red Hook, NY...

    work page 2020

  33. [33]

    S. Wang, H. Wang, P. Perdikaris, On the eigenvector bias of Fourier feature networks: From regression to solving multi-scale PDEs with physics-informed neural networks, Comput. Methods Appl. Mech. Eng. 384 (2021) 113938

    work page 2021

  34. [34]

    Glorot, Y

    X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, in: Y. W. Teh, M. Titterington (Eds.), Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, volume 9 ofProceedings of Machine Learning Research, PMLR, Chia Laguna Resort, Sardinia, Italy, 2010, pp. 249–256. 23

    work page 2010

  35. [35]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, PyTorch: An Imperative Style, High-Performance Deep Learning Library, Curran Associates Inc., Red Hook, NY, USA, 2019

    work page 2019

  36. [36]

    U. Ghia, K. Ghia, C. Shin, High-re solutions for incompressible flow using the Navier- Stokes equations and a multigrid method, J. Comput. Phys. 48 (1982) 387–411

    work page 1982

  37. [37]

    Erturk, T

    E. Erturk, T. C. Corke, C. Gökçöl, Numerical solutions of 2-d steady incompressible driven cavity flow at high reynolds numbers 48 (????) 747–774

    work page

  38. [38]

    Gautier, D

    R. Gautier, D. Biau, E. Lamballais, A reference solution of the flow over a circular cylinder at re=40 75 (????) 103–111

    work page

  39. [39]

    S. C. R. Dennis, G.-Z. Chang, Numerical solutions for steady flow past a circular cylinder at reynolds numbers up to 100, Journal of Fluid Mechanics 42 (1970) 471–489

    work page 1970

  40. [40]

    Fornberg, A numerical study of steady viscous flow past a circular cylinder, Journal of Fluid Mechanics 98 (1980) 819–855

    B. Fornberg, A numerical study of steady viscous flow past a circular cylinder, Journal of Fluid Mechanics 98 (1980) 819–855

    work page 1980

  41. [41]

    X. He, G. Doolen, Lattice boltzmann method on curvilinear coordinates system: Flow around a circular cylinder, Journal of Computational Physics 134 (1997) 306–315

    work page 1997

  42. [42]

    T. Ye, R. Mittal, H. Udaykumar, W. Shyy, An accurate cartesian grid method for viscous incompressible flows with complex immersed boundaries, Journal of Computational Physics 156 (1999) 209–240

    work page 1999

  43. [43]

    Calhoun, A cartesian grid method for solving the two-dimensional streamfunction- vorticity equations in irregular regions, Journal of computational physics 176 (2002) 231–275

    D. Calhoun, A cartesian grid method for solving the two-dimensional streamfunction- vorticity equations in irregular regions, Journal of computational physics 176 (2002) 231–275

    work page 2002

  44. [44]

    M. N. Linnick, H. F. Fasel, A high-order immersed interface method for simulating unsteady incompressible flows on irregular domains, Journal of Computational Physics 204 (2005) 157–192

    work page 2005

  45. [45]

    H. Ding, C. Shu, Q. Cai, Applications of stencil-adaptive finite difference method to incompressible viscous flows with curved boundary, Computers & fluids 36 (2007) 786–793

    work page 2007

  46. [46]

    Taira, T

    K. Taira, T. Colonius, The immersed boundary method: a projection approach, Journal of Computational Physics 225 (2007) 2118–2137

    work page 2007

  47. [47]

    D. V. Patil, K. Lakshmisha, Finite volume tvd formulation of lattice boltzmann simulation on unstructured mesh, Journal of Computational Physics 228 (2009) 5262–5279

    work page 2009

  48. [48]

    Bouchon, T

    F. Bouchon, T. Dubois, N. James, A second-order cut-cell method for the numerical simulation of 2d flows past obstacles, Computers & Fluids 65 (2012) 80–91

    work page 2012

  49. [49]

    Gautier, D

    R. Gautier, D. Biau, E. Lamballais, A reference solution of the flow over a circular cylinder at re=40, Computers & Fluids 75 (2013) 103–111. 24

    work page 2013