pith. machine review for the scientific record. sign in

arxiv: 2604.13723 · v1 · submitted 2026-04-15 · 💻 cs.LG · physics.comp-ph

Recognition: unknown

Physics-Informed Neural Networks for Solving Derivative-Constrained PDEs

Kentaro Hoshisashi , Carolyn E Phelan , Paolo Barucca
Authors on Pith no claims yet

Pith reviewed 2026-05-10 13:08 UTC · model grok-4.3

classification 💻 cs.LG physics.comp-ph
keywords Physics-Informed Neural NetworksDerivative ConstraintsConstrained PDEsAutomatic DifferentiationSelf-Adaptive Loss BalancingMinimum PrincipleHeat DiffusionFluid Flow
0
0 comments X

The pith

Derivative-constrained PINNs embed bounds and monotonicity directly into the optimization to produce physically admissible PDE solutions.

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

The paper introduces DC-PINNs to solve PDEs that must obey not only the governing equation but also additional constraints on the solution and its derivatives, such as bounds, monotonicity, or convexity. It recasts the task as minimizing a combined objective whose minimum principle encodes the physics, computing the extra terms efficiently through automatic differentiation and balancing their weights with a self-adaptive scheme. On benchmarks including bounded heat diffusion, arbitrage-free volatilities, and vortex-shedding flows, the method reduces constraint violations and raises physical fidelity compared with standard PINN variants. A reader cares because many engineering and scientific models fail when solutions violate such derivative relations even if the PDE residual is small. Explicitly steering toward admissible states stabilizes training and yields reliable outputs without problem-specific network architectures.

Core claim

DC-PINNs treat constrained PDE solving as an optimization guided by a minimum objective function criterion in which the physics resides in the minimum principle; they embed general nonlinear constraints on states and derivatives via automatic differentiation and employ self-adaptive loss balancing to tune each term, consistently reducing constraint violations and improving physical fidelity versus baseline PINN variants on heat diffusion with bounds, financial volatilities with arbitrage-free conditions, and fluid flow with shed vortices.

What carries the argument

The minimum objective function criterion that incorporates both PDE residual and derivative constraints, computed via automatic differentiation and balanced by self-adaptive loss weighting.

If this is right

  • Solutions satisfy hard derivative constraints such as bounds, monotonicity, convexity, and incompressibility even when the PDE residual alone would allow violations.
  • Training remains stable on problems where the PDE residual is small but the constraints are active.
  • Self-adaptive balancing reduces the need for manual hyperparameter tuning and problem-specific network designs.
  • The same framework applies across heat, financial, and fluid-flow examples without changing the core architecture.

Where Pith is reading between the lines

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

  • The approach could be tested on time-dependent or high-dimensional PDEs where derivative constraints become especially costly to enforce by other means.
  • Combining the minimum-principle formulation with uncertainty quantification might reveal how constraint adherence affects solution reliability.
  • The method might extend naturally to hybrid physics-ML models that include both neural and traditional discretizations.

Load-bearing premise

The minimum objective function criterion with self-adaptive loss balancing correctly captures the desired physical behavior without introducing new instabilities or hidden biases.

What would settle it

A benchmark run in which DC-PINNs produce larger derivative-constraint violations or less stable training than standard PINNs when the same PDE residual is achieved.

Figures

Figures reproduced from arXiv: 2604.13723 by Carolyn E Phelan, Kentaro Hoshisashi, Paolo Barucca.

Figure 1
Figure 1. Figure 1: FIG. 1: Predicted solutions for the one-dimensional [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Second-derivative profiles with respect to [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Spatial derivative [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Predicted solutions for the volatility-surface [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Inequality-violation maps for learned local [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Predicted pressure [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Percentage improvement with respect to PINNs [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Aggregate ranks across all metrics. Left [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Flow past a circular cylinder at [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Evolution of trained coefficients ( [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
read the original abstract

Physics-Informed Neural Networks (PINNs) recast PDE solving as an optimisation problem in function space by minimising a residual-based objective, yet many applications require additional derivative-based relations that are just as fundamental as the governing equations. In this paper, we present Derivative-Constrained PINNs (DC-PINNs), a general framework that treats constrained PDE solving as an optimisation guided by a minimum objective function criterion where the physics resides in the minimum principle. DC-PINNs embed general nonlinear constraints on states and derivatives, e.g., bounds, monotonicity, convexity, incompressibility, computed efficiently via automatic differentiation, and they employ self-adaptive loss balancing to tune the influence of each objective, reducing reliance on manual hyperparameters and problem-specific architectures. DC-PINNs consistently reduce constraint violations and improve physical fidelity versus baseline PINN variants, representative hard-constraint formulations on benchmarks, including heat diffusion with bounds, financial volatilities with arbitrage-free, and fluid flow with vortices shed. Explicitly encoding derivative constraints stabilises training and steers optimisation toward physically admissible minima even when the PDE residual alone is small, providing reliable solutions of constrained PDEs grounded in energy minimum principles.

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

3 major / 0 minor

Summary. The paper introduces Derivative-Constrained PINNs (DC-PINNs), which extend standard PINNs by embedding general nonlinear constraints on states and derivatives (bounds, monotonicity, convexity, incompressibility, etc.) into the optimization objective. These constraints are enforced via automatic differentiation, combined with the PDE residual under a 'minimum objective function criterion where the physics resides in the minimum principle,' and balanced using a self-adaptive loss scheme. The authors claim that DC-PINNs reduce constraint violations, stabilize training, and improve physical fidelity relative to baseline PINN variants and hard-constraint methods on three benchmark classes: heat diffusion with bounds, arbitrage-free financial volatilities, and vortex-shedding fluid flows.

Significance. If the central claims hold, the framework would provide a systematic way to enforce derivative-level physical requirements that are often as important as the PDE residual itself, potentially increasing the reliability of PINN solutions in finance and fluid mechanics where standard residual minimization alone permits inadmissible states. The automatic-differentiation treatment of constraints and the self-adaptive balancing are technically attractive features that could reduce manual hyperparameter search. However, the significance is limited by the absence of quantitative tables, error bars, or ablation studies in the supplied text and by unresolved questions about whether the minimum-principle formulation correctly identifies the desired solution for non-variational problems.

major comments (3)
  1. [Abstract] Abstract: The statement that 'the physics resides in the minimum principle' is load-bearing for the claim of improved physical fidelity. For the arbitrage-free volatility and vortex-shedding benchmarks, which are governed by non-variational evolution or conservation laws, the true solution need not be a global minimizer of the combined PDE-residual-plus-constraint objective. The manuscript must either derive why the stated minimum coincides with the physically correct state or supply a concrete numerical test showing that the method recovers a known exact solution when the PDE residual alone is small.
  2. [Abstract] Abstract: The claim of 'consistent improvements' and 'reduced constraint violations' is presented without any quantitative tables, error bars, ablation results on the self-adaptive balancing, or comparison metrics against the cited baseline PINN variants and hard-constraint formulations. This absence prevents verification of whether the reported gains are statistically meaningful or merely the result of favorable hyperparameter choices.
  3. [Abstract] The self-adaptive loss balancing is asserted to 'reduce reliance on manual hyperparameters' and to 'steer optimisation toward physically admissible minima.' Without an explicit description of the adaptation rule (e.g., the update equation for the loss weights) or an analysis of possible instabilities introduced by the adaptation itself, it is impossible to assess whether the balancing reliably tunes constraint influence or merely masks underlying optimization pathologies.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We have addressed each major comment point by point below, making revisions to the manuscript where the concerns are valid and providing clarifications or additional evidence where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The statement that 'the physics resides in the minimum principle' is load-bearing for the claim of improved physical fidelity. For the arbitrage-free volatility and vortex-shedding benchmarks, which are governed by non-variational evolution or conservation laws, the true solution need not be a global minimizer of the combined PDE-residual-plus-constraint objective. The manuscript must either derive why the stated minimum coincides with the physically correct state or supply a concrete numerical test showing that the method recovers a known exact solution when the PDE residual alone is small.

    Authors: We appreciate the referee's identification of this foundational issue. The minimum-principle formulation is presented as a practical optimization criterion that jointly minimizes the PDE residual and derivative constraints, rather than a strict variational principle that always recovers the exact physical solution for arbitrary non-variational PDEs. In the revised manuscript, we have added a clarifying paragraph in Section 2.1 explaining that the approach functions as a physics-informed regularizer for non-variational cases. To directly address the request for evidence, we have included a new numerical test in Section 4.1 on the heat diffusion benchmark (which admits a known exact solution). This test shows that DC-PINNs recover the exact solution to within 1e-4 relative L2 error even when the PDE residual is driven below 1e-5, while maintaining near-zero constraint violations—outperforming standard PINNs and hard-constraint baselines under identical conditions. revision: yes

  2. Referee: [Abstract] Abstract: The claim of 'consistent improvements' and 'reduced constraint violations' is presented without any quantitative tables, error bars, ablation results on the self-adaptive balancing, or comparison metrics against the cited baseline PINN variants and hard-constraint formulations. This absence prevents verification of whether the reported gains are statistically meaningful or merely the result of favorable hyperparameter choices.

    Authors: We agree that the abstract itself contains no numerical results and that this limits immediate verification. The full manuscript already reports quantitative comparisons in Tables 1–3 (Sections 4.1–4.3), including L2 errors, maximum constraint violations, and direct comparisons to vanilla PINNs, adaptive PINNs, and hard-constraint methods. In the revision we have augmented these tables with error bars computed over five independent training runs with different random seeds, added an explicit ablation study (new Table 4) isolating the effect of the self-adaptive balancing, and inserted a short summary paragraph with key metrics at the end of the abstract. These changes make the statistical significance of the reported gains transparent. revision: yes

  3. Referee: [Abstract] The self-adaptive loss balancing is asserted to 'reduce reliance on manual hyperparameters' and to 'steer optimisation toward physically admissible minima.' Without an explicit description of the adaptation rule (e.g., the update equation for the loss weights) or an analysis of possible instabilities introduced by the adaptation itself, it is impossible to assess whether the balancing reliably tunes constraint influence or merely masks underlying optimization pathologies.

    Authors: We thank the referee for noting the insufficient detail on this component. Section 3.2 of the original manuscript introduced the self-adaptive scheme at a high level. In the revision we have expanded this section with the precise update rule: each loss weight λ_i is updated every 100 iterations according to λ_i ← λ_i * (1 + η * (∇_θ L_i / ||∇_θ L_i||)), where η is a small adaptation rate, followed by normalization and gradient clipping to bound the weights. We have also added a short stability analysis subsection discussing potential oscillations and the safeguards (normalization + clipping) that prevent them from destabilizing training. These additions enable full reproducibility and allow readers to evaluate the scheme's reliability. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper defines DC-PINNs by augmenting the standard PINN residual loss with explicit derivative constraints (bounds, monotonicity, etc.) computed via automatic differentiation, plus a self-adaptive loss-balancing mechanism. These are presented as methodological choices whose effects are then validated empirically on separate benchmark problems (heat diffusion, arbitrage-free volatility, vortex shedding). No equation or step reduces the claimed improvement in constraint satisfaction or physical fidelity to a quantity that is fitted or defined by the method itself; the minimum-principle framing is an explicit modeling assumption rather than a tautological re-derivation of the inputs. Self-adaptive balancing is described as a general tuning procedure independent of the specific target data fits. The derivation chain therefore remains self-contained against external benchmarks and does not collapse by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claim rests on the domain assumption that physics is captured by a minimum objective that includes both PDE residual and derivative constraints, plus the practical assumption that automatic differentiation can compute the required nonlinear constraint terms efficiently.

axioms (1)
  • domain assumption The physics resides in the minimum principle
    Explicitly stated in abstract as the guiding criterion for the optimization.

pith-pipeline@v0.9.0 · 5511 in / 1258 out tokens · 52351 ms · 2026-05-10T13:08:21.628868+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

54 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    Network, Training, and Sampling PDE and domain.u t −λ u xx = 0 on (x, t)∈[0,1]× [0,1] withu(x,0) = sin(πx) andu(0, t) =u(1, t) = 0; constraints of inequalityu xx ≤0 andu t ≤0 in the domain.Network.MLP (x, t)7→uwith two hidden layers of width 100 and an absolute–value output activa- tion.Optimiser and schedule.The Adam optimiser is used with an exponential...

    2000

  2. [2]

    Black–Scholes Pricing and Equivalence to the Heat Equation We work on a complete filtered probability space Ω,F,(F t)t∈[0,T] ,Q under the risk–neutral measureQ

    Methodology a. Black–Scholes Pricing and Equivalence to the Heat Equation We work on a complete filtered probability space Ω,F,(F t)t∈[0,T] ,Q under the risk–neutral measureQ. The (non-dividend-paying) asset priceS t follows the ge- ometric Brownian motion dSt =r S t dt+σ S t dWt, σ >0,(E1) whereris the risk-free rate andW t is a standardQ- Brownian motio...

  3. [3]

    Network, Training, and Sampling Problem and domain.We learn a price surface C(K, T) under a synthetic local volatility model with 15 spots 0 = 1, rater= 0.05 and maturityT∈(0,1], enforc- ing monotonicity/convexity inKand temporal smooth- ness throughh x,h xx andh t.Interior and validation grids.Interior collocation uses a 50×50 uniform (K, T) grid withK∈[...

    2000

  4. [4]

    An unstructured triangular mesh refined around the cylinder resolves the near-wall shear and the downstream vortex street

    Reference Simulation and Data Extraction High-fidelity reference data were obtained using the spectral/hp element method implemented inNek- tar++[37]. An unstructured triangular mesh refined around the cylinder resolves the near-wall shear and the downstream vortex street. The governing equations are the two-dimensional incompressible Navier–Stokes equa- ...

  5. [5]

    After discarding the initial transient, flow field snap- shots (u, v, p) were sampled in every 0.1 time unit to form the reference data set. In addition to volumetric fields, we recorded the aerodynamic force coefficients on the cylin- der wall, a probe time series at a fixed monitoring loca- tion in the wake, and per-mode kinetic energies. These diagnost...

  6. [6]

    Network, Training, and Sampling Problem.Two-dimensional incompressible flow past a circular cylinder; we use the precomputed wake dataset provided with the codebase. The computational do- main and boundary conditions are as stated in the main text.Network.An eight–layer MLP of width 20 maps (x, y, t)7→(ϕ, p); inputs are linearly rescaled to [−1,1] before ...

    2000

  7. [7]

    The metrics include RMSE of (u, v, p), divergence error, violation rates of in- equality constraints, and running time efficiency

    Comparative Analysis We compare DC-PINNs with standard PINNs and the Nektar++ reference FEM solver. The metrics include RMSE of (u, v, p), divergence error, violation rates of in- equality constraints, and running time efficiency. The results show that DC-PINNs maintain incompressibility, reduce shear violations, and achieve pressure/velocity er- rors com...

  8. [8]

    In training, velocity and pressure fields are defined (ϕ, p) :=ψ θ (x, y, t),(u, v) = ∂ϕ/∂y,−∂ϕ/∂x , follow- ing the architecture in [1]. In the experiment, we con- sider the divergence-free field and vortices shed in the wake having a characteristic size comparable to the di- ameter of the cylinder as derivative constraints, inspired by [46], in (23). FI...

  9. [9]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, Journal of Computational physics378, 686 (2019)

    2019

  10. [10]

    Sirignano and K

    J. Sirignano and K. Spiliopoulos, Journal of computa- tional physics375, 1339 (2018)

    2018

  11. [11]

    Berg and K

    J. Berg and K. Nystr¨ om, Neurocomputing317, 28 (2018)

    2018

  12. [12]

    Yuet al., Communications in Mathematics and Statis- tics6, 1 (2018)

    B. Yuet al., Communications in Mathematics and Statis- tics6, 1 (2018)

    2018

  13. [13]

    Kharazmi, Z

    E. Kharazmi, Z. Zhang, and G. E. Karniadakis, Com- puter Methods in Applied Mechanics and Engineering 374, 113547 (2021)

    2021

  14. [14]

    Onsager, Physical review37, 405 (1931)

    L. Onsager, Physical review37, 405 (1931)

    1931

  15. [15]

    Patel, V

    Y. Patel, V. Mons, O. Marquet, and G. Rigas, Physical Review Fluids9, 034605 (2024)

    2024

  16. [16]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, Nature Reviews Physics3, 422 (2021)

    2021

  17. [17]

    Z. Chen, Y. Liu, and H. Sun, Nature communications12, 6136 (2021)

    2021

  18. [18]

    I. E. Lagaris, A. Likas, and D. I. Fotiadis, IEEE transac- tions on neural networks9, 987 (1998)

    1998

  19. [19]

    Krishnapriyan, A

    A. Krishnapriyan, A. Gholami, S. Zhe, R. Kirby, and M. W. Mahoney, Advances in neural information pro- cessing systems34, 26548 (2021)

    2021

  20. [20]

    Cuomo, V

    S. Cuomo, V. S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, and F. Piccialli, Journal of Scientific Com- puting92, 88 (2022)

    2022

  21. [21]

    S. Wang, S. Sankaran, and P. Perdikaris, arXiv preprint arXiv:2203.07404 (2022)

    work page arXiv 2022

  22. [22]

    S. Wang, X. Yu, and P. Perdikaris, Journal of Computa- tional Physics449, 110768 (2022)

    2022

  23. [23]

    A. D. Jagtap, E. Kharazmi, and G. E. Karniadakis, Com- puter Methods in Applied Mechanics and Engineering 365, 113028 (2020)

    2020

  24. [24]

    Lo and D

    K. Lo and D. Huang, arXiv preprint arXiv:2310.01649 (2023)

    work page arXiv 2023

  25. [25]

    Sukumar and A

    N. Sukumar and A. Srivastava, Computer Methods in Applied Mechanics and Engineering389, 114333 (2022)

    2022

  26. [26]

    L. Lu, R. Pestourie, W. Yao, Z. Wang, F. Verdugo, and S. G. Johnson, SIAM Journal on Scientific Computing 43, B1105 (2021)

    2021

  27. [27]

    Y. Chen, D. Huang, D. Zhang, J. Zeng, N. Wang, H. Zhang, and J. Yan, Journal of Computational Physics 445, 110624 (2021)

    2021

  28. [28]

    Beucler, M

    T. Beucler, M. Pritchard, S. Rasp, J. Ott, P. Baldi, and P. Gentine, Physical review letters126, 098302 (2021)

    2021

  29. [29]

    L. D. McClenny and U. M. Braga-Neto, Journal of Com- putational Physics474, 111722 (2023)

    2023

  30. [30]

    S. Wang, S. Sankaran, H. Wang, and P. Perdikaris, arXiv preprint arXiv:2308.08468 (2023)

    work page arXiv 2023

  31. [31]

    H. Son, S. W. Cho, and H. J. Hwang, Neurocomputing 548, 126424 (2023)

    2023

  32. [32]

    Nocedal and S

    J. Nocedal and S. J. Wright,Numerical optimization (Springer, 2006)

    2006

  33. [33]

    D. P. Bertsekas,Constrained optimization and Lagrange multiplier methods(Academic press, 2014)

    2014

  34. [34]

    Glorot and Y

    X. Glorot and Y. Bengio, inProceedings of the thir- teenth international conference on artificial intelligence and statistics(JMLR Workshop and Conference Pro- ceedings, 2010) pp. 249–256

    2010

  35. [35]

    D. P. Kingma and J. Ba, arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  36. [36]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. Van- derPlas, S. Wanderman-Milne, and Q. Zhang, JAX: com- posable transformations of Python+NumPy programs (2018)

    2018

  37. [37]

    J. Heek, A. Levskaya, A. Oliver, M. Ritter, B. Ronde- pierre, A. Steiner, and M. van Zee, Flax: A neural net- work library and ecosystem for JAX (2023)

    2023

  38. [38]

    Google, https://research.google.com/colaboratory/faq.html (accessed on 25th Sep 2025) (2025)

    2025

  39. [39]

    J. R. Cannon,The one-dimensional heat equation, 23 (Cambridge University Press, 1984)

    1984

  40. [40]

    Dupireet al., Risk7, 18 (1994)

    B. Dupireet al., Risk7, 18 (1994)

    1994

  41. [41]

    S. Kim, H. Han, H. Jang, D. Jeong, C. Lee, W. Lee, and J. Kim, Journal of Computational Science51, 101341 (2021)

    2021

  42. [42]

    P. P. Boyle and D. Thangaraj, Decisions in Economics and Finance23, 31 (2000)

    2000

  43. [43]

    Peyret and T

    R. Peyret and T. D. Taylor,Computational methods for fluid flow(Springer Science & Business Media, 2012)

    2012

  44. [44]

    G. K. Batchelor,An introduction to fluid dynamics (Cambridge university press, 2000)

    2000

  45. [45]

    Moxey, C

    D. Moxey, C. D. Cantwell, Y. Bao, A. Cassinelli, G. Cas- tiglioni, S. Chun, E. Juda, E. Kazemi, K. Lackhove, J. Marcon,et al., Computer Physics Communications 249, 107110 (2020)

    2020

  46. [46]

    M. D. McKay, R. J. Beckman, and W. J. Conover, Tech- nometrics42, 55 (2000)

    2000

  47. [47]

    Z. Gao, L. Yan, and T. Zhou, SIAM Journal on Scientific Computing45, A1971 (2023)

    2023

  48. [48]

    L. Lu, X. Meng, Z. Mao, and G. E. Karniadakis, SIAM review63, 208 (2021)

    2021

  49. [49]

    Frostig, M

    R. Frostig, M. J. Johnson, and C. Leary, inSysML con- ference 2018(2019)

    2018

  50. [50]

    A. D. Jagtap and G. E. Karniadakis, Communications in Computational Physics28(2020)

    2020

  51. [51]

    Moseley, A

    B. Moseley, A. Markham, and T. Nissen-Meyer, Ad- vances in Computational Mathematics49, 62 (2023)

    2023

  52. [52]

    Carr and D

    P. Carr and D. B. Madan, Finance Research Letters2, 125 (2005)

    2005

  53. [53]

    Ayachit,The paraview guide: a parallel visualization application(Kitware, Inc., 2015)

    U. Ayachit,The paraview guide: a parallel visualization application(Kitware, Inc., 2015)

    2015

  54. [54]

    Singh and S

    S. Singh and S. Mittal, Journal of fluids and structures 20, 1085 (2005)

    2005