pith. machine review for the scientific record. sign in

arxiv: 2605.08872 · v1 · submitted 2026-05-09 · ⚛️ physics.flu-dyn

Recognition: 2 theorem links

· Lean Theorem

Data-driven Symbolic Closure for Turbulence Modeling in the Lattice Boltzmann Framework

Wanru Deng, Yihan Zhang, Yuanjun Dai, Yujie Fu

Pith reviewed 2026-05-12 01:20 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords turbulence modelingLattice Boltzmann Methodsymbolic regressionsubgrid-scale closuredata-driven modelingTaylor-Green vortexlid-driven cavity
0
0 comments X

The pith

A symbolic expression discovered from DNS data of two flows serves as a subgrid closure in Lattice Boltzmann turbulence modeling.

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

The paper seeks to replace conventional algebraic subgrid-scale models in Lattice Boltzmann simulations with an explicit analytical formula learned directly from high-fidelity data. It applies symbolic optimization to direct numerical simulation results for Taylor-Green vortex and lid-driven cavity flows while embedding physical constraints such as dimensional consistency and tensor invariants. The resulting expression depends nonlinearly on strain-rate and rotation-rate quantities. If the approach holds, it would reduce the need for case-by-case tuning and empirical wall corrections in turbulent flow computations.

Core claim

By feeding DNS data from Taylor-Green Vortex and Lid-Driven Cavity flows into Physical Symbolic Optimization, the authors obtain an explicit subgrid-scale stress closure for the Lattice Boltzmann Method. Virtual dimensional analysis and nonlinear tensor invariants are enforced during the search, producing a highly nonlinear function of strain-rate and rotation-rate invariants. In numerical tests this closure reproduces kinetic-energy dissipation peaks more faithfully than the Smagorinsky model, resolves secondary corner vortices, and transfers without modification to turbulent channel flow at Re_tau = 180.

What carries the argument

The explicit symbolic closure expression obtained through constrained symbolic optimization; it replaces the algebraic subgrid-scale term inside the Lattice Boltzmann collision operator.

If this is right

  • The model reproduces kinetic energy dissipation rate peaks more accurately than the standard Smagorinsky closure in the tested configurations.
  • Secondary corner vortices appear more clearly resolved in lid-driven cavity simulations.
  • The closure transfers directly to wall-bounded channel flow at Re_tau = 180 with no added damping functions.
  • Nonlinear dependence on both strain-rate and rotation-rate invariants emerges automatically from the data-driven search.

Where Pith is reading between the lines

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

  • The same constrained symbolic search could be repeated on datasets from other numerical schemes to produce compatible closures for finite-volume or finite-element solvers.
  • The explicit nonlinear form supplies a concrete candidate for analytic improvement of existing subgrid models in traditional computational fluid dynamics.
  • Further zero-shot tests on flows with separation or transition would clarify the range over which the discovered expression remains valid.

Load-bearing premise

That an expression fitted only to Taylor-Green Vortex and lid-driven cavity data encodes turbulent physics general enough to apply unchanged to wall-bounded channel flow.

What would settle it

Applying the same symbolic closure to direct numerical simulation data of a third turbulent regime, such as decaying homogeneous isotropic turbulence, and verifying whether the predicted dissipation rate spectrum deviates systematically from the reference solution.

Figures

Figures reproduced from arXiv: 2605.08872 by Wanru Deng, Yihan Zhang, Yuanjun Dai, Yujie Fu.

Figure 1
Figure 1. Figure 1: FIG. 1: Schematic illustration of data-driven symbolic closure for turbulence modeling [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Comparison of flow structures between LBM-DNS, LBM-Smagorinsky, and [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Temporal evolution of normalized total kinetic energy [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Velocity profiles for LDC flow at [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Comparison of fluctuation velocity contours and streamlines for the LDC case. [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Generalization test on turbulent channel flow at [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
read the original abstract

Turbulence modeling within the Lattice Boltzmann Method (LBM) framework has long relied on traditional algebraic sub-grid scale (SGS) models, which often suffer from over-dissipation and lack of spatial selectivity near solid boundaries. In this work, we utilize Physical Symbolic Optimization (Phi-SO) to discover explicit analytical closures from high-fidelity DNS datasets of Taylor-Green Vortex (TGV) and Lid-Driven Cavity (LDC) flows. Central to our methodology is the integration of virtual dimensional analysis and non-linear tensor invariants, a strategy that enforces physical scaling laws directly within the symbolic search process. The resulting model exhibits a highly non-linear dependency on both strain-rate and rotation-rate invariants. Numerical validations confirm that this symbolic closure outperforms the standard Smagorinsky approach in capturing kinetic energy dissipation rate peaks and resolving delicate secondary corner vortices. Furthermore, the model exhibits robust zero-shot generalization to wall-bounded turbulent channel flow (Re_tau = 180) without the aid of any supplemental wall-damping corrections. This work highlights the potential of symbolic regression to uncover robust, interpretable physical laws for the next generation of intelligent computational fluid dynamics solvers.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a data-driven method using Physical Symbolic Optimization (Phi-SO) to derive explicit symbolic sub-grid-scale (SGS) closures for turbulence modeling within the Lattice Boltzmann framework. The approach trains on DNS data from Taylor-Green Vortex (TGV) and lid-driven cavity (LDC) flows, incorporating virtual dimensional analysis and non-linear tensor invariants of strain-rate and rotation-rate tensors. It claims that the resulting highly non-linear closure outperforms the standard Smagorinsky model in capturing kinetic energy dissipation peaks and secondary vortices in the training flows, and exhibits robust zero-shot generalization to wall-bounded turbulent channel flow at Re_tau=180 without any wall-damping functions or retuning.

Significance. If the discovered symbolic expression proves to be both accurate and sufficiently universal, the work would demonstrate a viable path toward interpretable, physics-constrained algebraic closures that avoid ad-hoc damping corrections common in traditional SGS models. This could improve predictive fidelity in LBM-based simulations of complex geometries while maintaining computational efficiency, provided the zero-shot transfer is substantiated beyond the specific test case.

major comments (3)
  1. [Abstract and §4] Abstract and §4 (numerical validations): The central claim of outperformance and 'robust zero-shot generalization' to channel flow is asserted without providing the explicit symbolic expression discovered by Phi-SO, quantitative error metrics (e.g., integrated dissipation rate errors or vortex strength comparisons), or details on how post-hoc coefficient choices were avoided. This prevents independent verification of whether the data support the stated superiority over Smagorinsky.
  2. [§3] §3 (symbolic regression methodology): The integration of virtual dimensional analysis and non-linear invariants is described, but no a-priori analysis of the resulting expression's limiting behavior (near-wall asymptotics, high-Re scaling, or behavior under sustained shear production) is supplied. Without this, the assumption that invariants learned from TGV (homogeneous decaying) and LDC (transitional wall-bounded) encode Re- and geometry-independent physics remains untested, directly bearing on the zero-shot channel-flow claim.
  3. [§4] §4 (channel flow results): The reported numerical validations for Re_tau=180 channel flow constitute the sole evidence for generalization, yet the manuscript provides no comparison of SGS stress tensor components or energy spectra against DNS, nor any sensitivity study to the training dataset composition. This leaves open the possibility that the expression overfits patterns specific to TGV/LDC rather than capturing universal physics.
minor comments (2)
  1. [Abstract] The abstract refers to 'highly non-linear dependency' but does not quantify the degree of non-linearity or compare it to existing invariant-based models (e.g., those of Lund & Novikov).
  2. [Results figures] Figure captions and legends in the results section should explicitly state the grid resolution, filter width, and exact definition of the dissipation rate used for comparisons.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have helped us strengthen the clarity and verifiability of the manuscript. We address each major comment below and have incorporated revisions to provide the requested details and analyses.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (numerical validations): The central claim of outperformance and 'robust zero-shot generalization' to channel flow is asserted without providing the explicit symbolic expression discovered by Phi-SO, quantitative error metrics (e.g., integrated dissipation rate errors or vortex strength comparisons), or details on how post-hoc coefficient choices were avoided. This prevents independent verification of whether the data support the stated superiority over Smagorinsky.

    Authors: We agree that the explicit symbolic expression and quantitative metrics are essential for independent verification. In the revised manuscript, we now present the full symbolic expression discovered by Phi-SO in §4. We have also added quantitative error metrics, including relative errors in the integrated kinetic energy dissipation rate and comparisons of secondary vortex strengths for the TGV and LDC cases. The Phi-SO procedure embeds physical constraints and optimization directly in the symbolic search, eliminating the need for post-hoc coefficient tuning beyond the invariant basis. revision: yes

  2. Referee: [§3] §3 (symbolic regression methodology): The integration of virtual dimensional analysis and non-linear invariants is described, but no a-priori analysis of the resulting expression's limiting behavior (near-wall asymptotics, high-Re scaling, or behavior under sustained shear production) is supplied. Without this, the assumption that invariants learned from TGV (homogeneous decaying) and LDC (transitional wall-bounded) encode Re- and geometry-independent physics remains untested, directly bearing on the zero-shot channel-flow claim.

    Authors: The non-linear invariants were selected precisely because they are coordinate-invariant and dimensionally consistent by construction via virtual dimensional analysis. We acknowledge the benefit of explicit limiting-case analysis. The revised §3 now includes an a-priori examination demonstrating that the discovered expression recovers the expected near-wall asymptotic scaling of the SGS stress (∼ y^3) and reduces to a Smagorinsky-like form in the high-Re inertial range under sustained shear, thereby supporting the physical generality underlying the zero-shot transfer. revision: yes

  3. Referee: [§4] §4 (channel flow results): The reported numerical validations for Re_tau=180 channel flow constitute the sole evidence for generalization, yet the manuscript provides no comparison of SGS stress tensor components or energy spectra against DNS, nor any sensitivity study to the training dataset composition. This leaves open the possibility that the expression overfits patterns specific to TGV/LDC rather than capturing universal physics.

    Authors: We have expanded §4 to include direct comparisons of the SGS stress tensor components (particularly the off-diagonal shear component) and the turbulent kinetic energy spectra against the DNS reference for the Re_tau=180 channel. In addition, a sensitivity study was performed by retraining Phi-SO on TGV-only and LDC-only subsets; the leading non-linear terms remain stable across these ablations, indicating that the expression captures shared physics rather than dataset-specific artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity in data-driven symbolic regression pipeline

full rationale

The paper explicitly employs Physical Symbolic Optimization (Phi-SO) to discover a closure by fitting to external DNS data from Taylor-Green Vortex and lid-driven cavity flows, then performs numerical validation and zero-shot testing on separate turbulent channel flow DNS at Re_tau=180. This constitutes a standard train/test split on independent external benchmarks rather than any first-principles derivation that reduces to its own inputs by construction. No self-definitional equivalences, fitted quantities renamed as predictions, load-bearing self-citations, or smuggled ansatzes are present in the described methodology. The central claims rest on out-of-sample numerical evidence against held-out DNS, satisfying the criteria for a self-contained, non-circular result.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that the symbolic search, constrained by dimensional analysis and tensor invariants, yields a physically meaningful and generalizable SGS model from limited DNS data of two specific flows.

free parameters (1)
  • coefficients inside the discovered symbolic expression
    Symbolic regression optimizes numerical coefficients to match the DNS dissipation and stress data of TGV and LDC flows.
axioms (2)
  • domain assumption Virtual dimensional analysis and non-linear tensor invariants enforce physical consistency during the symbolic search
    The method integrates these constraints directly into Phi-SO to restrict the search space.
  • domain assumption The discovered expression constitutes a valid sub-grid-scale closure for the LBM collision operator
    Assumed on the basis of fitting to the training DNS and subsequent numerical tests.

pith-pipeline@v0.9.0 · 5508 in / 1491 out tokens · 54881 ms · 2026-05-12T01:20:28.008419+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

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

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages

  1. [1]

    Kummerl\"ander, Adrian and Bingert, Tim and Bukreev, Fedor and Czelusniak, Luiz Eduardo and Dapelo, Davide and Hafen, Nicolas and Heinzelmann, Marc and Ito, Shota and Jeberger, Julius and Kusumaatmaja, Halim and Marquardt, Jan E. and Rennick, Michael and Pertzel, Tim and Prinz, Frantiek and Sadric, Martin and Schecher, Maximilian and Simonis, Stephan and ...

    work page doi:10.5281/zenodo.10684609 2024

  2. [2]

    Recovery of the Navier-Stokes equations using a lattice-gas Boltzmann method , author =. Phys. Rev. A , volume =. 1992 , month =. doi:10.1103/PhysRevA.45.R5339 , url =

    work page doi:10.1103/physreva.45.r5339 1992

  3. [3]

    Y. H. Qian and D. D'Humières and P. Lallemand , title =. 1992 , month =. doi:10.1209/0295-5075/17/6/001 , url =

    work page doi:10.1209/0295-5075/17/6/001 1992

  4. [4]

    and Meiron, Daniel I

    Brachet, Marc E. and Meiron, Daniel I. and Orszag, Steven A. and Nickel, B. G. and Morf, Rudolf H. and Frisch, Uriel , year=. Small-scale structure of the Taylor–Green vortex , volume=. doi:10.1017/S0022112083001159 , journal=

    work page doi:10.1017/s0022112083001159

  5. [5]

    and Koseff, Jeffrey R

    Prasad, Ajay K. and Koseff, Jeffrey R. , title =. Physics of Fluids , year =

    work page

  6. [6]

    , title =

    Tenachi, Wassim and Ibata, Rodrigo and Diakogiannis, Foivos I. , title =. 2023 , month =. doi:10.3847/1538-4357/ad014c , url =

    work page doi:10.3847/1538-4357/ad014c 2023

  7. [7]

    2006 , note =

    LES of turbulent square jet flow using an MRT lattice Boltzmann model , journal =. 2006 , note =. doi:https://doi.org/10.1016/j.compfluid.2005.04.009 , url =

    work page doi:10.1016/j.compfluid.2005.04.009 2006

  8. [8]

    Journal of Turbulence , volume =

    Zhenhua Xia and Yipeng Shi and Yu Chen and Moran Wang and Shiyi Chen , title =. Journal of Turbulence , volume =. 2015 , publisher =

    work page 2015

  9. [9]

    Smagorinsky , title =

    J. Smagorinsky , title =. Monthly Weather Review , volume =. 1963 , pages=

    work page 1963

  10. [10]

    2021 , issn =

    Large-Eddy Simulation of smooth and rough channel flows using a one-dimensional stochastic wall model , journal =. 2021 , issn =. doi:https://doi.org/10.1016/j.compfluid.2021.105135 , url =

    work page doi:10.1016/j.compfluid.2021.105135 2021

  11. [11]

    Physics of Fluids , volume =

    Ryu, Sungmin and Iaccarino, Gianluca , title =. Physics of Fluids , volume =. 2014 , month =

    work page 2014

  12. [12]

    Physics of Fluids , volume =

    Bou-Zeid, Elie and Meneveau, Charles and Parlange, Marc , title =. Physics of Fluids , volume =. 2005 , month =

    work page 2005

  13. [13]

    and Ducros, F

    Nicoud, F. and Ducros, F. , title =. Flow, Turbulence and Combustion , year =

    work page

  14. [14]

    Vreman, A. W. , title =. Physics of Fluids , volume =. 2004 , month =

    work page 2004

  15. [15]

    Physics of Fluids , volume =

    Nicoud, Franck and Toda, Hubert Baya and Cabrit, Olivier and Bose, Sanjeeb and Lee, Jungil , title =. Physics of Fluids , volume =. 2011 , month =

    work page 2011

  16. [16]

    Turbulence Modeling in the Age of Data

    Duraisamy, Karthik and Iaccarino, Gianluca and Xiao, Heng. Turbulence Modeling in the Age of Data. Annual Review of Fluid Mechanics. 2019. doi:https://doi.org/10.1146/annurev-fluid-010518-040547

    work page doi:10.1146/annurev-fluid-010518-040547 2019

  17. [17]

    and San, O

    Maulik, R. and San, O. and Rasheed, A. and Vedula, P. , year=. Subgrid modelling for two-dimensional turbulence using neural networks , volume=. doi:10.1017/jfm.2018.770 , journal=

    work page doi:10.1017/jfm.2018.770 2018

  18. [18]

    and Hassan, Yassin A

    Taghizadeh, Salar and Witherden, Freddie D. and Hassan, Yassin A. and Girimaji, Sharath S. , title =. Physics of Fluids , volume =. 2021 , month =

    work page 2021

  19. [19]

    Super-resolution reconstruction of turbulent flows with machine learning , volume=

    Fukami, Kai and Fukagata, Koji and Taira, Kunihiko , year=. Super-resolution reconstruction of turbulent flows with machine learning , volume=. doi:10.1017/jfm.2019.238 , journal=

    work page doi:10.1017/jfm.2019.238 2019

  20. [20]

    Generalizable physics-constrained modeling using learning and inference assisted by feature-space engineering , author =. Phys. Rev. Fluids , volume =. 2021 , month =

    work page 2021

  21. [21]

    Ensemble Kalman method for learning turbulence models from indirect observation data , volume=

    Zhang, Xin-Lei and Xiao, Heng and Luo, Xiaodong and He, Guowei , year=. Ensemble Kalman method for learning turbulence models from indirect observation data , volume=. doi:10.1017/jfm.2022.744 , journal=

    work page doi:10.1017/jfm.2022.744 2022

  22. [22]

    Brunton and Joshua L

    Steven L. Brunton and Joshua L. Proctor and J. Nathan Kutz , title =. Proceedings of the National Academy of Sciences , volume =

    work page

  23. [23]

    Physics of Fluids , volume =

    Liao, Hui and Sun, Xuxiang and Liu, Yilang and Zhang, Weiwei , title =. Physics of Fluids , volume =. 2025 , month =

    work page 2025

  24. [24]

    and Fang, Y

    Reissmann, M. and Fang, Y. and Ooi, A. S. H. and Sandberg, R. D. , title =. Physics of Fluids , volume =. 2025 , month =

    work page 2025

  25. [25]

    and Dwight, R.P

    Schmelzer, M. and Dwight, R.P. and Cinnella, P. , title =. Flow Turbulence Combust , year =

    work page

  26. [26]

    S. B. Pope , title =

    work page

  27. [27]

    J. H. Ferziger and M. Peric and Robert L. Street , title =

    work page

  28. [28]

    Moin and K

    P. Moin and K. Mahesh , title =. Annual Review of Fluid Mechanics. 1998 , volume =

    work page 1998

  29. [29]

    P. R. Spalart , title =. International Journal of Heat and Fluid Flow , year =

    work page

  30. [30]

    Yang , title =

    Z. Yang , title =. Chinese Journal of Aeronautics , year =

    work page

  31. [31]

    Kim and P

    J. Kim and P. Moin and R. Moser , title =. Journal of Fluid Mechanics , volume=

    work page