Pith. sign in

REVIEW 4 major objections 6 minor 98 references

Diff-FlowFSI: A GPU-Optimized Differentiable CFD Platform for High-Fidelity Turbulence and FSI Simulations

T0 review · 4 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 Diff-FlowFSI is a fully differentiable, GPU-accelerated computational platform for simulating unsteady turbulent flow and fluid–structure interactions with exact end-to-end gradients.

desk verdict Genuinely new differentiable CFD platform with broad external validation, but the solid-region pressure projection is under-derived and no code is released—worth a peer review with revisions. read the letter →

arxiv 2505.23940 v1 pith:VQ5Z43YN submitted 2025-05-29 physics.flu-dyn

Xiantao Fan , Xinyang Liu , Meng Wang , Jian-Xun Wang This is my paper
classification physics.flu-dyn
keywords differentiableCFDfluid–structureinteractionimmersedboundarymethodGPUaccelerationturbulentchannelflowautomaticdifferentiationhybridneural-physicsmodelingdirectforcing
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

Diff-FlowFSI is a GPU-native, fully differentiable computational platform for simulating unsteady turbulent flow and fluid–structure interaction. The paper claims that a vectorized finite-volume solver combined with an immersed boundary method can deliver high-fidelity forward simulations while also providing exact end-to-end gradients through the entire simulation pipeline. This matters because gradient access makes inverse modeling, parameter estimation, and hybrid neural-physics learning practical for problems that traditionally require expensive adjoint or finite-difference sensitivity calculations. The platform is validated on canonical 2D and 3D benchmarks, and the reported runtime efficiency is up to two orders of magnitude better than a conventional CPU-based solver.

What carries the argument

The load-bearing mechanism is the surrogate divergence constraint of Eq. (8), which approximates the divergence of the corrected velocity inside the solid region as $\nabla\cdot u^{t+1} \approx \nabla\cdot[\epsilon(x)(u^{*} - u_c)]$, where $\epsilon$ is the volume-of-solid indicator and $u_c$ is the exact solid velocity. This constraint determines the modified pressure Poisson equation (Eq. 9) used in the immersed-boundary projection, and it reduces to the classical Poisson equation in purely fluid regions. The rest of the machinery is a vectorized finite-volume discretization on staggered grids, a regularized delta function for Eulerian–Lagrangian interpolation and force spreading, and a strong-coupling sub-iteration loop that stabilizes flexible structures with strong added-mass effects.

What would settle it

Run a grid-convergence study on a flexible-plate case with strong added-mass effects, measuring the slip velocity at the immersed boundary and the coupled displacement error against high-resolution reference data; if reducing grid spacing does not reduce the slip or if the interface error grows with structural acceleration, the surrogate divergence constraint fails.

Watch

Extended reading notes

Core claim

The authors introduce Diff-FlowFSI and claim it is a fully differentiable, GPU-accelerated computational platform for simulating unsteady turbulent flow and fluid–structure interactions with end-to-end gradient access. The method couples a staggered-grid finite-volume fluid solver with an immersed boundary treatment of static and moving solids, and a finite-element beam structure solver for flexible bodies, using strong two-way coupling at each time step. To keep the pressure projection consistent inside the solid region, the paper introduces a surrogate divergence constraint, $\nabla\cdot u^{t+1} \approx \nabla\cdot[\epsilon(x)(u^{*} - u_c)]$, which replaces the incompressibility condition inside the solid and yields a modified pressure Poisson equation. The resulting static compute graph supports reverse-mode automatic differentiation through the time loop, with implicit differentiation for inner iterative solves. Validation across vortex shedding, vortex-induced vibration, flexible plate deformation, turbulent channel flow, rough-wall flow, and periodic-hill separation supports the claims of accuracy, scalability, and gradient-based inverse modeling.

Load-bearing premise

The entire immersed-boundary coupling rests on one approximation: inside the solid region, the apparent fluid velocity is projected using a surrogate constraint based on the mismatch between the intermediate velocity and the solid's velocity, and if that approximation is inaccurate at the structure's surface, the no-slip condition and the two-way coupling degrade.

Editorial extensions

If this is right

  • Inverse problems such as inferring an unknown spring stiffness from sparse displacement measurements become single-loop gradient descents instead of repeated forward solves.
  • Hybrid neural-physics models can be trained end-to-end through the solver, giving long-horizon forecasts that stay consistent with the governing equations.
  • High-fidelity workloads like turbulent channel flow and bluff-body wakes can run on a single GPU at speeds that rival CPU clusters.
  • Design and optimization loops that need many CFD evaluations become cheaper because the same differentiable simulation provides both state and sensitivity.
  • Users can trade speed for accuracy within one workflow because the solver supports both single and double precision.

Reading between the lines

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

  • The two-orders-of-magnitude speedup compares a single GPU against a limited number of CPU cores and a non-preconditioned pressure solver against a multigrid-preconditioned one, so the figure is hardware- and configuration-dependent rather than a universal property of the method.
  • A differentiable platform of this kind makes it natural to learn subgrid and wall models by backpropagating through the entire turbulent rollout; the paper lists trainable wall models as future work rather than implementing them.
  • The same surrogate-constraint idea could extend to 3D dynamic solids and multi-GPU domain decomposition, both of which the paper flags as future work.
  • The scan-based time loop means backpropagation memory scales with the number of stored checkpoints rather than the full time horizon, which would make long-rollout training of hybrid models more practical than in unrolled solvers.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript introduces Diff-FlowFSI, a JAX-based, GPU-accelerated finite-volume solver with an immersed boundary method for incompressible flows and two-way fluid–structure interaction, and presents it as a differentiable CFD platform for turbulence and FSI simulations. The core contributions claimed are (i) a fully vectorized GPU solver, (ii) end-to-end automatic differentiation for inverse modeling and hybrid neural-CFD training, and (iii) validation across a broad range of benchmarks including 2D vortex shedding, VIV, flexible plates, 3D channel flow, rough-wall turbulence, periodic hills, and WMLES. The paper also reports large speedups over OpenFOAM and demonstrates inverse parameter recovery and hybrid learning applications. The presentation is generally clear, and the benchmark suite is extensive, but several load-bearing derivations and validations are incomplete.

Significance. If the platform delivers what is claimed, it would be a valuable contribution at the intersection of CFD and scientific machine learning: a differentiable, GPU-native solver with immersed-boundary FSI would enable gradient-based inverse design and hybrid neural-physics training in settings where existing differentiable solvers are limited to simple single-physics problems. Strengths of the manuscript include its extensive external benchmark comparisons (e.g., Moser et al. DNS, Srinil et al. experiments, Breuer et al. periodic-hill DNS) with no evidence of parameter tuning to match those data, clearly quantified performance comparisons against OpenFOAM, and a modular software design that separates fluid, solid, and coupling components. The main risks are the unverified surrogate divergence constraint in the pressure projection, which is central to the FSI coupling, and a tendency to overstate the level of validation for WMLES and flexible-structure cases.

major comments (4)
  1. [§2.3.1, Eq. (7)] Equation (7) is not the standard pressure Poisson equation as written. Taking the divergence of Eq. (6) gives ∇²p^{t+1} = (ρ/Δt)∇·(u* − u^{t+1}), which is an identity containing the unknown u^{t+1}; it becomes a closed equation only after imposing a divergence constraint on u^{t+1}. The standard PPE is ∇²p^{t+1} = (ρ/Δt)∇·u* when ∇·u^{t+1}=0 in the fluid. Please correct the presentation so that Eq. (7) is either derived with the constraint made explicit or replaced by the standard form, and clarify how Eq. (9) follows.
  2. [§2.3.1, Eqs. (8)–(10)] The surrogate divergence constraint in Eq. (8) is load-bearing for the FSI/IBM pressure projection, but it is introduced without derivation and is not verified locally. Equation (9) is solved globally, so any spurious divergence source introduced in the solid/IB cells can contaminate the pressure in the adjacent fluid, and the final velocity update (10) applies the pressure correction everywhere, including the solid cells where the direct-forcing term in Eq. (5) has already imposed no-slip. The consistency of these two mechanisms near the interface is not established. No local divergence error, interface slip error, or grid-convergence study of Eq. (8) is reported. Please provide a derivation of the surrogate constraint, demonstrate its consistency with the direct-forcing IBM at the interface, and report quantitative local diagnostics. This is necessary to support the central claim of high-fidelity two-way FSI.
  3. [§3.3 and §3.4.5] Validation for two regimes that are central to the claimed capabilities is significantly thinner than for the other benchmarks. The flexible-plate cases in §3.3 report visual agreement, one inclination angle (§3.3.1), and one tip amplitude (§3.3.2), without quantitative time-series or statistical comparison to the cited references [77,78]. The WMLES results in Fig. 17 show only mean velocity profiles against the Spalding law; no turbulence statistics, resolution study, or comparison against reference WMLES/DNS data are provided, so the statement that the wall model is 'validated' is stronger than the evidence. Please add quantitative metrics or temper the validation claims.
  4. [§4.1] The inverse-modeling example recovers a spring stiffness from observations generated by the same Diff-FlowFSI solver, so it is a self-consistency test rather than an independent validation of the differentiable-programming capability. Since the paper emphasizes exact gradients, please include a direct gradient check (e.g., comparing reverse-mode AD gradients against finite differences or a hand-derived adjoint for a canonical case) and, ideally, a test with noisy or externally generated data to demonstrate robustness.
minor comments (6)
  1. [§2.3.1, Eq. (8)] The notation u_c^t and u_c^{t+1} is not defined precisely; 'exact velocity of the immersed solid' is ambiguous, especially in comparison with u_s^t. Please clarify these definitions and their relation to ϵ^t.
  2. [Algorithm 1] The convergence criterion ξ_t = ||(w^t − w^{t−1})/w^{t−1}||₂² is undefined when a component of w^{t−1} is zero, which can occur for the translational degrees of freedom in some configurations. Use an absolute increment norm or normalize by a characteristic displacement.
  3. [§2.5 and Table 1] The abstract and conclusion state that the solver supports 'dynamic solid bodies' and 3D domains, but Table 1 and §2.5 indicate that dynamic solids are implemented only in 2D. Please state this scope limitation explicitly in the abstract and conclusion to avoid overgeneralization.
  4. [Throughout] There are several typographical errors that should be fixed: 'Eularian' in §2.2, 'mtigate' and 'multipyics' in §1, 'an canonical' in §3.1, 'the the' in §2.3.2, 'outter' in Algorithm 2, 'Mass ration' in Table 2, and '¯Cc' in §3.2.2.
  5. [§3.5] The performance comparison is informative, but the paper does not state whether the reported runtimes include initialization, I/O, and Python/JIT compilation overhead, or whether the OpenFOAM cases use identical grid distributions and solver settings. Please specify these details so the speedup figures can be reproduced.
  6. [Availability] No code availability statement is provided. For a platform paper whose main contribution is a software system, releasing the code or at least stating a clear availability policy is important for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: Diff-FlowFSI's forward claims are validated against independent external benchmarks; the hybrid-learning self-citations and the Eq. (8) surrogate are auxiliary or modeling assumptions, not circular inputs.

full rationale

The paper's central forward-simulation claims are self-contained and benchmarked against external data: Section 3.1 compares cylinder St, CL,rms, CD to independent numerical/experimental studies; Section 3.2 compares VIV amplitudes to Chen et al. and Bao et al.; Section 3.3 compares plate reconfiguration to experiments; Section 3.4 compares channel-flow statistics to Moser et al., Kim et al., Abe et al., and periodic-hill DNS to Breuer et al.; Section 3.4.2 compares rough-wall profiles to Santiago et al. No free solver parameter was fitted to make these curves match; the IBM and wall-model constants are standard literature values. The surrogate divergence constraint in Eq. (8) is a modeling approximation introduced to close the pressure projection in the immersed solid region, and it feeds the modified Poisson equation Eq. (9); it is not claimed to be derived from first principles, so its accuracy is a numerical-fidelity concern rather than a circular reduction. The WMLES section (Sec. 3.4.5) compares the simulated u+ profile to the Spalding law used to build the equilibrium wall model (Eq. 27, Fig. 17); this is a self-consistency check of the closure implementation, not an external validation, and it is not a load-bearing claim of first-principles prediction. The hybrid-learning demonstrations in Sec. 4 cite and adapt the authors' earlier works ([32],[34]); these are auxiliary capability demonstrations, while the central solver claims rest on the external benchmarks in Sec. 3. Accordingly, no circular step is identified.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The platform's central claim rests on standard CFD formulations (finite volume, projection, IBM) and on several modeling approximations introduced by the authors, most notably the surrogate divergence constraint for immersed solids. No new physical entities are postulated.

free parameters (2)
  • Smagorinsky coefficient = not reported
    Constant Smagorinsky SGS model is used for LES/WMLES (Section 3.4); the coefficient is a model constant chosen by hand from the literature and its value is not stated, making LES results not fully reproducible.
  • Wall-model regularization epsilon = 1e-6
    Used in Eq (28) to prevent division by zero in the wall-model eddy viscosity; hand-chosen numerical regularization.
assumptions (7)
  • domain assumption Incompressible Navier-Stokes equations with constant density and viscosity are the fluid model.
    Used throughout; restricts platform to incompressible flows.
  • domain assumption Euler-Bernoulli beam theory governs flexible structures, with only bending deformations and no axial/shear effects.
    Section 2.2, Eq (2); limits flexible structure model to slender 2D beams.
  • standard math Chorin projection method splits momentum advancement and pressure correction.
    Section 2.3.1, Eqs (4)-(10); standard fractional-step method.
  • ad hoc to paper Direct-forcing IBM with surrogate divergence constraint approximates incompressibility in solid cells.
    Eqs (8)-(9); approximate constraint introduced to adapt projection to immersed solids; not formally justified.
  • domain assumption Strong coupling sub-iterations converge at tolerance 1e-5 within 100 iterations for all FSI cases.
    Algorithm 1; convergence is asserted, not demonstrated, and the paper reports only 2-3 iterations for the flexible plate cases.
  • domain assumption Equilibrium wall model based on Spalding law with kappa=0.4 and B=5.
    Section 3.4.5, Eq (27); standard wall model with constants from prior literature.
  • domain assumption Constant-coefficient Smagorinsky SGS closure is used for LES/WMLES.
    Section 3.4; standard SGS model, but its coefficient value is not reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diff-FlowFSI: A GPU-Optimized Differentiable CFD Platform for High-Fidelity Turbulence and FSI Simulations." pith.science (2026). https://pith.science/paper/VQ5Z43YN

@misc{pith2026250523940,
  author       = {Pith},
  title        = {Pith review of: Diff-FlowFSI: A GPU-Optimized Differentiable CFD Platform for High-Fidelity Turbulence and FSI Simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VQ5Z43YN}},
  note         = {Machine review of arXiv:2505.23940}
}
read the original abstract

Turbulent flows and fluid-structure interactions (FSI) are ubiquitous in scientific and engineering applications, but their accurate and efficient simulation remains a major challenge due to strong nonlinearities, multiscale interactions, and high computational demands. Traditional CFD solvers, though effective, struggle with scalability and adaptability for tasks such as inverse modeling, optimization, and data assimilation. Recent advances in machine learning (ML) have inspired hybrid modeling approaches that integrate neural networks with physics-based solvers to enhance generalization and capture unresolved dynamics. However, realizing this integration requires solvers that are not only physically accurate but also differentiable and GPU-efficient. In this work, we introduce Diff-FlowFSI, a GPU-accelerated, fully differentiable CFD platform designed for high-fidelity turbulence and FSI simulations. Implemented in JAX, Diff-FlowFSI features a vectorized finite volume solver combined with the immersed boundary method to handle complex geometries and fluid-structure coupling. The platform enables GPU-enabled fast forward simulations, supports automatic differentiation for gradient-based inverse problems, and integrates seamlessly with deep learning components for hybrid neural-CFD modeling. We validate Diff-FlowFSI across a series of benchmark turbulence and FSI problems, demonstrating its capability to accelerate scientific computing at the intersection of physics and machine learning.

Figures

Figures reproduced from arXiv: 2505.23940 by the authors.

Figure 1
Figure 1. Overview of the features and workflow of the differentiable CFD platform (Diff-FlowFSI). [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Sketch of the immersed solid with sharp interface, where crosses represent cell centers. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. (a) Mass-damper-spring system for rigid solids, allowing for an arbitrary solid shape. (b) Dis [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (18 more)
Figure 4
Figure 4. Figure 4: (a) Computational domain; (b) pressure coefficients for static cylinder at [PITH_FULL_IMAGE:figures/full_fig_p023_4.png]
Figure 5
Figure 5. Figure 5: Validation of unsteady vortex shedding from a stationary cylinder over the Reynolds number range [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]
Figure 6
Figure 6. Figure 6: Streamlines for vortex shedding from a static cylinder: (a) [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Validation results for the 1-DOF VIV of a single cylinder: (a) cross-flow vibration amplitudes [PITH_FULL_IMAGE:figures/full_fig_p026_7.png]
Figure 8
Figure 8. Figure 8: Validation results for the 2-DOF VIV of a single cylinder: (a) cross-flow vibration amplitudes [PITH_FULL_IMAGE:figures/full_fig_p027_8.png]
Figure 9
Figure 9. Figure 9: Vorticity modes for the 2-DOF VIV of a single cylinder. [PITH_FULL_IMAGE:figures/full_fig_p027_9.png]
Figure 10
Figure 10. Figure 10: Validation of static reconfiguration of a vertical flexible plate: (a) computational domain and [PITH_FULL_IMAGE:figures/full_fig_p029_10.png]
Figure 11
Figure 11. Figure 11: Validation of oscillatory response of a flexible plate mounted on a cylinder: (a) computational [PITH_FULL_IMAGE:figures/full_fig_p030_11.png]
Figure 12
Figure 12. Figure 12: Validation results for 3D turbulent channel flow at [PITH_FULL_IMAGE:figures/full_fig_p032_12.png]
Figure 13
Figure 13. Figure 13: The vortical structures for 3D turbulence based on Q criteria at different flow through time, [PITH_FULL_IMAGE:figures/full_fig_p033_13.png]
Figure 14
Figure 14. Figure 14: Validation results for turbulent flow over cubic rough walls: (a) instantaneous 3D vortical struc [PITH_FULL_IMAGE:figures/full_fig_p035_14.png]
Figure 15
Figure 15. Figure 15: Validation results for separated turbulent flow over periodic hills: (a) instantaneous 3D velocity [PITH_FULL_IMAGE:figures/full_fig_p036_15.png]
Figure 16
Figure 16. Figure 16: Validation results for turbulent vortex shedding behind a 3D finite-length cylinder at [PITH_FULL_IMAGE:figures/full_fig_p037_16.png]
Figure 17
Figure 17. Figure 17: Average profiles of streamwise velocity along wall-normal directions for [PITH_FULL_IMAGE:figures/full_fig_p039_17.png]
Figure 18
Figure 18. Figure 18: Running time of simulating turbulent channel flow at [PITH_FULL_IMAGE:figures/full_fig_p041_18.png]
Figure 19
Figure 19. Figure 19: Inverse modeling using AD in Diff-FlowFSI: prediction of unknown spring stiffness and recon [PITH_FULL_IMAGE:figures/full_fig_p043_19.png]
Figure 20
Figure 20. Figure 20: Hybrid neural Diff-FlowFSI solver for predicting VIV dynamics of rigid and flexible structures: [PITH_FULL_IMAGE:figures/full_fig_p044_20.png]
Figure 21
Figure 21. Figure 21: Hybrid neural solver architecture design for turbulence: (a) schematic of deep fusion architecture [PITH_FULL_IMAGE:figures/full_fig_p046_21.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

98 extracted references · 68 canonical work pages

  1. [1]

    W. L. Oberkampf, T. G. Trucano, Verification and validation in computational fluid dynamics, Progress in aerospace sciences 38 (3) (2002) 209–272

    2002

  2. [2]

    M. M. Bhatti, M. Marin, A. Zeeshan, S. I. Abdelsalam, Recent trends in computational fluid dynamics, Frontiers in Physics 8 (2020) 593111

    2020

  3. [3]

    M. H. Zawawi, A. Saleha, A. Salwa, N. Hassan, N. M. Zahari, M. Z. Ramli, Z. C. Muda, A review: Fundamentals of computational fluid dynamics (cfd), in: AIP conference proceedings, Vol. 2030, AIP Publishing, 2018

    2018

  4. [4]

    M. Ma, J. Lu, G. Tryggvason, Using statistical learning to close two-fluid multiphase flow equations for a simple bubbly system, Physics of Fluids 27 (9) (2015)

    2015

  5. [5]

    H. R. Fairbanks, L. Jofre, G. Geraci, G. Iaccarino, A. Doostan, Bi-fidelity approxima- tion for uncertainty quantification and sensitivity analysis of irradiated particle-laden turbulence, Journal of Computational Physics 402 (2020) 108996

    2020

  6. [6]

    An AMReX-based Compressible Reacting Flow Solver for High-speed Reacting Flows relevant to Hypersonic Propulsion

    S. Sharma, R. Bielawski, O. Gibson, S. Zhang, V. Sharma, A. H. Rauch, J. Singh, S. Abisleiman, M. Ullman, S. Barwey, et al., An amrex-based compressible reacting flow solver for high-speed reacting flows relevant to hypersonic propulsion, arXiv preprint arXiv:2412.00900 (2024)

    work page Pith review arXiv 2024

  7. [7]

    J. N. Fuhg, G. Anantha Padmanabha, N. Bouklas, B. Bahmani, W. Sun, N. N. Vlassis, M. Flaschel, P. Carrara, L. De Lorenzis, A review on data-driven constitutive laws for solids, Archives of Computational Methods in Engineering (2024) 1–43

    2024

  8. [8]

    X. Fan, M. Ge, W. Tan, Q. Li, Impacts of coexisting buildings and trees on the perfor- mance of rooftop wind turbines: An idealized numerical study, Renewable Energy 177 (2021) 164–180. 49

    2021

Show all 98 references
  1. [9]

    Xiao, J.-L

    H. Xiao, J.-L. Wu, J.-X. Wang, R. Sun, C. Roy, Quantifying and reducing model-form uncertainties in reynolds-averaged navier–stokes simulations: A data-driven, physics- informed bayesian approach, Journal of Computational Physics 324 (2016) 115–136

    2016

  2. [10]

    Duraisamy, G

    K. Duraisamy, G. Iaccarino, H. Xiao, Turbulence modeling in the age of data, Annual review of fluid mechanics 51 (1) (2019) 357–377

    2019

  3. [11]

    Jofre, A

    L. Jofre, A. Doostan, Rapid aerodynamic shape optimization under uncertainty using a stochastic gradient approach, Structural and Multidisciplinary Optimization 65 (7) (2022) 196

    2022

  4. [12]

    S. L. Brunton, B. R. Noack, P. Koumoutsakos, Machine learning for fluid mechanics, Annual review of fluid mechanics 52 (1) (2020) 477–508

    2020

  5. [13]

    P. Du, X. Zhu, J.-X. Wang, Deep learning-based surrogate model for three-dimensional patient-specific computational fluid dynamics, Physics of Fluids 34 (8) (2022) 081906

    2022

  6. [14]

    X. Han, H. Gao, T. Pfaff, J.-X. Wang, L. Liu, Predicting physics in mesh-reduced space with temporal attention, in: International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=XctLdNfCmP

    2022

  7. [15]

    P. Du, M. H. Parikh, X. Fan, X.-Y. Liu, J.-X. Wang, Conditional neural field latent dif- fusion model for generating spatiotemporal turbulence, Nature Communications (2024)

    2024

  8. [16]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics- informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440

    2021

  9. [17]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational physics 378 (2019) 686–707. 50

    2019

  10. [18]

    L. Sun, H. Gao, S. Pan, J.-X. Wang, Surrogate modeling for fluid flows based on physics- constrained deep learning without simulation data, Computer Methods in Applied Me- chanics and Engineering 361 (2020) 112732

    2020

  11. [19]

    Arzani, J.-X

    A. Arzani, J.-X. Wang, R. M. D’Souza, Uncovering near-wall blood flow from sparse data with physics-informed neural networks, Physics of Fluids 33 (7) (2021) 071905. doi:10.1063/5.0055600

    2021 doi

  12. [20]

    Y. Chen, L. Lu, G. E. Karniadakis, L. Dal Negro, Physics-informed neural networks for inverse problems in nano-optics and metamaterials, Optics express 28 (8) (2020) 11618–11633

    2020

  13. [21]

    H. Gao, L. Sun, J.-X. Wang, PhyGeoNet: physics-informed geometry-adaptive con- volutional neural networks for solving parameterized steady-state PDEs on irregular domain, Journal of Computational Physics 428 (2021) 110079

    2021

  14. [22]

    R. Li, J. Zhou, J.-X. Wang, T. Luo, Physics-informed bayesian neural networks for solving phonon boltzmann transport equation in forward and inverse problems with sparse and noisy data, ASME Journal of Heat and Mass Transfer 147 (3) (2025) 032501

    2025

  15. [23]

    Z. Chen, Y. Liu, H. Sun, Physics-informed learning of governing equations from scarce data, Nature communications 12 (1) (2021) 6136

    2021

  16. [24]

    Krishnapriyan, A

    A. Krishnapriyan, A. Gholami, S. Zhe, R. Kirby, M. W. Mahoney, Characterizing possi- ble failure modes in physics-informed neural networks, Advances in Neural Information Processing Systems 34 (2021) 26548–26560

    2021

  17. [25]

    S. Wang, X. Yu, P. Perdikaris, When and why pinns fail to train: A neural tangent kernel perspective, Journal of Computational Physics 449 (2022) 110768

    2022

  18. [26]

    S. Wang, Y. Teng, P. Perdikaris, Understanding and mitigating gradient flow pathologies in physics-informed neural networks, SIAM Journal on Scientific Computing 43 (5) (2021) A3055–A3081. 51

    2021

  19. [27]

    Brenner, J

    M. Brenner, J. Eldredge, J. Freund, Perspective on machine learning for advancing fluid mechanics, Physical Review Fluids 4 (10) (2019) 100501

    2019

  20. [28]

    Wang, J.-L

    J.-X. Wang, J.-L. Wu, H. Xiao, Physics-informed machine learning approach for recon- structing reynolds stress modeling discrepancies based on DNS data, Physical Review Fluids 2 (3) (2017) 034603

    2017

  21. [29]

    Tompson, K

    J. Tompson, K. Schlachter, P. Sprechmann, K. Perlin, Accelerating eulerian fluid sim- ulation with convolutional networks, in: International conference on machine learning, PMLR, 2017, pp. 3424–3433

    2017

  22. [30]

    J. Wu, H. Xiao, R. Sun, Q. Wang, Reynolds-averaged navier–stokes equations with explicit data-driven reynolds stress closure can be ill-conditioned, Journal of Fluid Me- chanics 869 (2019) 553–586

    2019

  23. [31]

    Taghizadeh, F

    S. Taghizadeh, F. D. Witherden, S. S. Girimaji, Turbulence closure modeling with data- driven techniques: physical compatibility and consistency considerations, New Journal of Physics 22 (9) (2020) 093023

    2020

  24. [32]

    Fan, J.-X

    X. Fan, J.-X. Wang, Differentiable hybrid neural modeling for fluid-structure interaction, arXiv preprint arXiv:2303.12971 (2023)

    2023 arXiv

  25. [33]

    X.-Y. Liu, M. Zhu, L. Lu, H. Sun, J.-X. Wang, Multi-resolution partial differential equations preserved learning framework for spatiotemporal dynamics, Communications Physics 7 (1) (2024) 31

    2024

  26. [34]

    X. Fan, D. Akhare, J.-X. Wang, Neural differentiable modeling with diffusion-based super-resolution for two-dimensional spatiotemporal turbulence, Computer Methods in Applied Mechanics and Engineering 433 (2025) 117478

    2025

  27. [35]

    Kochkov, J

    D. Kochkov, J. A. Smith, A. Alieva, Q. Wang, M. P. Brenner, S. Hoyer, Ma- chine learning–accelerated computational fluid dynamics, Proceedings of the National Academy of Sciences 118 (21) (2021) e2101784118. 52

    2021

  28. [36]

    Akhare, T

    D. Akhare, T. Luo, J.-X. Wang, Diffhybrid-uq: Uncertainty quantification for differen- tiable hybrid neural modeling, arXiv preprint arXiv:2401.00161 (2023)

    2023 arXiv

  29. [37]

    Akhare, T

    D. Akhare, T. Luo, J.-X. Wang, Physics-integrated neural differentiable (PiNDiff) model for composites manufacturing, Computer Methods in Applied Mechanics and Engineer- ing 406 (2023) 115902

    2023

  30. [38]

    Akhare, Z

    D. Akhare, Z. Chen, R. Gulotty, T. Luo, J.-X. Wang, Probabilistic physics-integrated neural differentiable modeling for isothermal chemical vapor infiltration process, npj Computational Materials 10 (1) (2024) 120

    2024

  31. [39]

    List, L.-W

    B. List, L.-W. Chen, N. Thuerey, Learned turbulence modelling with differentiable fluid solvers: physics-based loss functions and optimisation horizons, Journal of Fluid Mechanics 949 (2022) A25

    2022

  32. [40]

    F. D. A. Belbute-Peres, T. Economon, Z. Kolter, Combining differentiable pde solvers and graph neural networks for fluid flow prediction, in: international conference on machine learning, PMLR, 2020, pp. 2402–2411

    2020

  33. [41]

    Shankar, D

    V. Shankar, D. Chakraborty, V. Viswanathan, R. Maulik, Differentiable turbulence: Closure as a partial differential equation constrained optimization, Physical Review Fluids 10 (2) (2025) 024605

    2025

  34. [42]

    Shang, J

    W. Shang, J. Zhou, J. Panda, Z. Xu, Y. Liu, P. Du, J.-X. Wang, T. Luo, Jax-bte: A gpu-accelerated differentiable solver for phonon boltzmann transport equations, arXiv preprint arXiv:2503.23657 (2025)

    2025 arXiv

  35. [43]

    C. H. Bischof, P. D. Hovland, B. Norris, On the implementation of automatic differen- tiation tools, Higher-Order and Symbolic Computation 21 (2008) 311–331

    2008

  36. [44]

    McNamara, A

    A. McNamara, A. Treuille, Z. Popovi´ c, J. Stam, Fluid control using the adjoint method, ACM Transactions On Graphics (TOG) 23 (3) (2004) 449–456. 53

    2004

  37. [45]

    Hinterberger, M

    C. Hinterberger, M. Olesen, Automatic geometry optimization of exhaust systems based on sensitivities computed by a continuous adjoint cfd method in openfoam, Tech. rep., SAE Technical Paper (2010)

    2010

  38. [46]

    Y. Shi, C. A. Mader, S. He, G. L. Halila, J. R. Martins, Natural laminar-flow airfoil optimization design using a discrete adjoint approach, AIAA Journal 58 (11) (2020) 4702–4722

    2020

  39. [47]

    C. A. Mader, J. R. Martins, J. J. Alonso, E. Van Der Weide, Adjoint: An approach for the rapid development of discrete adjoint solvers, AIAA journal 46 (4) (2008) 863–873

    2008

  40. [48]

    P. Holl, V. Koltun, K. Um, N. Thuerey, phiflow: A differentiable pde solving framework for deep learning via physical simulations, in: NeurIPS workshop, Vol. 2, 2020

    2020

  41. [49]

    N. A. A. Deniz A. Bezgin, Aaron B. Buhendwa, Jax-fluids: A fully-differentiable high- order computational fluid dynamics solver for compressible two-phase flows, arXiv preprint arXiv (2022)

    2022

  42. [50]

    Boustani, M

    J. Boustani, M. F. Barad, C. C. Kiris, C. Brehm, An immersed boundary fluid–structure interaction method for thin, highly compliant shell structures, Journal of Computational Physics 438 (2021) 110369

    2021

  43. [51]

    Ramaswamy, M

    B. Ramaswamy, M. Kawahara, Arbitrary lagrangian–eulerianc finite element method for unsteady, convective, incompressible viscous free surface fluid flow, International Journal for Numerical Methods in Fluids 7 (10) (1987) 1053–1075

    1987

  44. [52]

    Malecha, L

    Z. Malecha, L. Miros law, T. Tomczak, Z. Koza, M. Matyka, W. Tarnawski, D. Szczerba, et al., Gpu-based simulation of 3d blood flow in abdominal aorta using openfoam, Archives of Mechanics 63 (2) (2011) 137–161

    2011

  45. [53]

    Rathnayake, S

    T. Rathnayake, S. Jayasena, M. Narayana, Openfoam on gpus using amgx, in: Pro- ceedings of the 25th High Performance Computing Symposium, 2017, pp. 1–12. 54

    2017

  46. [54]

    C. R. Harris, K. J. Millman, S. J. Van Der Walt, R. Gommers, P. Virtanen, D. Cour- napeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, et al., Array programming with numpy, Nature 585 (7825) (2020) 357–362

    2020

  47. [55]

    C. S. Peskin, The immersed boundary method, Acta numerica 11 (2002) 479–517

    2002

  48. [56]

    O. A. Bauchau, J. I. Craig, Euler-Bernoulli beam theory, Springer, 2009, pp. 173–221

    2009

  49. [57]

    C. S. Peskin, Flow patterns around heart valves: a numerical method, Journal of com- putational physics 10 (2) (1972) 252–271

    1972

  50. [58]

    Uhlmann, An immersed boundary method with direct forcing for the simulation of particulate flows, Journal of computational physics 209 (2) (2005) 448–476

    M. Uhlmann, An immersed boundary method with direct forcing for the simulation of particulate flows, Journal of computational physics 209 (2) (2005) 448–476

    2005

  51. [59]

    O. C. Zienkiewicz, R. L. Taylor, The finite element method for solid and structural mechanics, Elsevier, 2005

    2005

  52. [60]

    M. Liu, D. G. Gorman, Formulation of rayleigh damping and its extensions, Computers & structures 57 (2) (1995) 277–285

    1995

  53. [61]

    C. Ji, A. Munjiza, J. Williams, A novel iterative direct-forcing immersed boundary method and its finite volume applications, Journal of Computational Physics 231 (4) (2012) 1797–1821

    2012

  54. [62]

    A. G. Baydin, B. A. Pearlmutter, A. A. Radul, J. M. Siskind, Automatic differentiation in machine learning: a survey, Journal of Marchine Learning Research 18 (2018) 1–43

    2018

  55. [63]

    Akhare, P

    D. Akhare, P. Du, T. Luo, J.-X. Wang, Implicit neural differential model for spatiotem- poral dynamics, arXiv preprint arXiv:2504.02260 (2025)

    2025 arXiv

  56. [64]

    Blondel, V

    M. Blondel, V. Roulet, The elements of differentiable programming arXiv:2403.14606v1 [cs.LG] 21 mar 2024 (2024). 55

    2024 arXiv

  57. [65]

    Blondel, Q

    M. Blondel, Q. Berthet, M. Cuturi, R. Frostig, S. Hoyer, F. Llinares-L´ opez, F. Pe- dregosa, J.-P. Vert, Efficient and modular implicit differentiation, arXiv preprint arXiv:2105.15183 (2021)

    2021 arXiv

  58. [66]

    R. D. Henderson, Nonlinear dynamics and pattern formation in turbulent wake transi- tion, Journal of fluid mechanics 352 (1997) 65–112

    1997

  59. [67]

    C. v. Wieselsberger, Neuere feststellungen under die gesetze des flussigkeits und luftwiderstandes, Phys. z. 22 (1921) 321–328

    1921

  60. [68]

    Posdziech, R

    O. Posdziech, R. Grundmann, A systematic approach to the numerical calculation of fundamental quantities of the two-dimensional flow over a circular cylinder, Journal of fluids and structures 23 (3) (2007) 479–499

    2007

  61. [69]

    J. Park, K. Kwon, H. Choi, Numerical solutions of flow past a circular cylinder at reynolds numbers up to 160, KSME international Journal 12 (1998) 1200–1205

    1998

  62. [70]

    C. H. Williamson, Oblique and parallel modes of vortex shedding in the wake of a circular cylinder at low reynolds numbers, Journal of Fluid Mechanics 206 (1989) 579– 627

    1989

  63. [71]

    W. Chen, C. Ji, J. Williams, D. Xu, L. Yang, Y. Cui, Vortex-induced vibrations of three tandem cylinders in laminar cross-flow: Vibration response and galloping mechanism, Journal of Fluids and Structures 78 (2018) 215–238

    2018

  64. [72]

    Y. Bao, C. Huang, D. Zhou, J. Tu, Z. Han, Two-degree-of-freedom flow-induced vibra- tions on isolated and tandem cylinders with varying natural frequency ratios, Journal of Fluids and Structures 35 (2012) 50–75

    2012

  65. [73]

    Gsell, J

    S. Gsell, J. Favier, Direct-forcing immersed-boundary method: A simple correction preventing boundary slip error, Journal of Computational Physics 435 (2021) 110265

    2021

  66. [74]

    R. D. Gabbai, H. Benaroya, An overview of modeling and experiments of vortex-induced vibration of circular cylinders, Journal of sound and vibration 282 (3-5) (2005) 575–616. 56

    2005

  67. [75]

    Srinil, H

    N. Srinil, H. Zanganeh, A. Day, Two-degree-of-freedom viv of circular cylinder with variable natural frequency ratio: Experimental and numerical investigations, Ocean Engineering 73 (2013) 179–194

    2013

  68. [76]

    Prasanth, S

    T. Prasanth, S. Mittal, Vortex-induced vibrations of a circular cylinder at low reynolds numbers, Journal of Fluid Mechanics 594 (2008) 463–491

    2008

  69. [77]

    Zhang, G

    X. Zhang, G. He, X. Zhang, Fluid–structure interactions of single and dual wall-mounted 2d flexible filaments in a laminar boundary layer, Journal of Fluids and Structures 92 (2020) 102787

    2020

  70. [78]

    Pfister, O

    J.-L. Pfister, O. Marquet, Fluid–structure stability analyses and nonlinear dynamics of flexible splitter plates interacting with a circular cylinder flow, Journal of Fluid Mechanics 896 (2020) A24

    2020

  71. [79]

    R. D. Moser, J. Kim, N. N. Mansour, Direct numerical simulation of turbulent channel flow up to re τ = 590, Physics of fluids 11 (4) (1999) 943–945

    1999

  72. [80]

    J. Kim, P. Moin, R. Moser, Turbulence statistics in fully developed channel flow at low reynolds number, Journal of fluid mechanics 177 (1987) 133–166

    1987

  73. [81]

    H. Abe, H. Kawamura, Y. Matsuo, Direct numerical simulation of a fully developed turbulent channel flow with respect to the reynolds number dependence, J. Fluids Eng. 123 (2) (2001) 382–393

    2001

  74. [82]

    M. M. Rai, P. Moin, Direct simulations of turbulent flow using finite-difference schemes, Journal of computational physics 96 (1) (1991) 15–53

    1991

  75. [83]

    R. B. Dean, Reynolds number dependence of skin friction and other bulk flow variables in two-dimensional rectangular duct flow (1978)

    1978

  76. [84]

    Steen, W

    P. Steen, W. Brutsaert, Saph and schoder and the friction law of blasius, Annual Review of Fluid Mechanics 49 (2017) 575–582. 57

    2017

  77. [85]

    J. L. Santiago, A. Martilli, F. Mart ´ ın, Cfd simulation of airflow over a regular array of cubes. part i: Three-dimensional simulation of the flow and validation with wind-tunnel measurements, Boundary-layer meteorology 122 (2007) 609–634

    2007

  78. [86]

    S.-G. Cai, A. Ouahsine, J. Favier, Y. Hoarau, Moving immersed boundary method, International Journal for Numerical Methods in Fluids 85 (5) (2017) 288–323

    2017

  79. [87]

    K. Zhou, S. Balachandar, An analysis of the spatio-temporal resolution of the immersed boundary method with direct forcing, Journal of Computational Physics 424 (2021) 109862

    2021

  80. [88]

    Kim, M.-C

    Y. Kim, M.-C. Lai, Simulating the dynamics of inextensible vesicles by the penalty immersed boundary method, Journal of Computational Physics 229 (12) (2010) 4840– 4853

    2010

  81. [89]

    Breuer, N

    M. Breuer, N. Peller, C. Rapp, M. Manhart, Flow over periodic hills–numerical and experimental study in a wide range of reynolds numbers, Computers & Fluids 38 (2) (2009) 433–457

    2009

  82. [90]

    Douglas, et al., Strouhal number for vortex-induced vibration excitation of long slender structures (2022)

    A. Douglas, et al., Strouhal number for vortex-induced vibration excitation of long slender structures (2022)

    2022

  83. [91]

    Scotti, C

    A. Scotti, C. Meneveau, D. K. Lilly, Generalized smagorinsky model for anisotropic grids, Physics of Fluids A: Fluid Dynamics 5 (9) (1993) 2306–2308

    1993

  84. [92]

    A single formula for the law of the wall, Journal of Applied Mechanics 28 (3) (1961) 455–458

    1961

  85. [93]

    H. J. Bae, A. Lozano-Dur´ an, Effect of wall boundary conditions on a wall-modeled large-eddy simulation in a finite-difference framework, Fluids 6 (3) (2021) 112

    2021

  86. [94]

    Liefvendahl, T

    M. Liefvendahl, T. Mukha, S. Rezaeiravesh, Formulation of a wall model for LES in a collocated finite-volume framework, Department of Information Technology, Uppsala University, 2017. 58

    2017

  87. [95]

    Larsson, S

    J. Larsson, S. Kawai, J. Bodart, I. Bermejo-Moreno, Large eddy simulation with mod- eled wall-stress: recent progress and future directions, Mechanical Engineering Reviews 3 (1) (2016) 15–00418

    2016

  88. [96]

    X. I. Yang, G. I. Park, P. Moin, Log-layer mismatch and modeling of the fluctuating wall stress in wall-modeled large-eddy simulations, Physical review fluids 2 (10) (2017) 104601

    2017

  89. [97]

    Y. M. Lee, J. H. Lee, J. Lee, Artificial neural network-based wall-modeled large-eddy simulations of turbulent channel and separated boundary layer flows, Aerospace Science and Technology 132 (2023) 108014

    2023

  90. [98]

    Zhuang, D

    J. Zhuang, D. Kochkov, Y. Bar-Sinai, M. P. Brenner, S. Hoyer, Learned discretizations for passive scalar advection in a two-dimensional turbulent flow, Physical Review Fluids 6 (6) (2021) 064605. 59

    2021

Pith tools

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