Whether second-moment (stress-transport) turbulence closures provide clear and systematic predictive advantages over eddy-viscosity two-equation models in complex flows featuring strong separation, rotation, streamline curvature, or three-dimensional Reynolds-stress anisotropy — and whether divergence-based formulations of stress transport such as the Perot turbulent potential model deliver systematic computational advantages (convergence rate, stiffness) — remains unresolved. Viti, Huang and Bradshaw (2007), who implemented and validated several recent stress-transport models and explicitly limited their conclusions to simple wall-bounded and shear-free flows, left the predictive-advantage question open, and the convergence and stiffness advantages claimed for divergence-based formulations in the model's own publications were not quantified in their study; later community work (the NASA Turbulence Modeling Resource, the Eisfeld–Rumsey–Togiti multi-code assessment, the NASA CFD Vision 2030 study) has addressed canonical benchmark cases and the standardization infrastructure but has not settled the systematic-advantage question for the complex-flow regime, nor quantified the divergence-formulation claims.

Background

Most engineering predictions of turbulent flow rely on computational fluid dynamics (CFD) built on the Reynolds-averaged Navier–Stokes (RANS) equations, which describe the mean flow but require a turbulence closure: a model for the unknown Reynolds-stress tensor that represents the effect of turbulent fluctuations on the mean flow. The workhorse closures are eddy-viscosity models, which invoke the Boussinesq hypothesis — an analogy with molecular viscosity that ties the Reynolds stresses linearly to the mean strain-rate tensor through a single scalar eddy viscosity. One- and two-equation models of this type (the Spalart–Allmaras one-equation model, and two-equation models such as kk–ϵ\epsilon, kk–ω\omega, and Menter's Shear Stress Transport blend of kk–ω\omega and kk–ϵ\epsilon, which compute the turbulent kinetic energy and a length-scale-determining variable) are calibrated mainly on attached, thin shear layers and remain the industrial standard. Their structural weakness is well known: a scalar eddy viscosity forces the stress tensor to align with the strain tensor and cannot represent Reynolds-stress anisotropy, the response to system rotation or streamline curvature, or secondary flows in non-circular ducts, so their accuracy degrades precisely in complex separated, rotating, curved, and three-dimensional flows.

Second-moment closures, also called Reynolds-stress models (RSM) or stress-transport models, remove the Boussinesq hypothesis: they solve individual transport equations for all six independent components of the Reynolds-stress tensor (the tensor is symmetric) plus a seventh equation for a length-scale variable. The physical mechanisms that eddy-viscosity models must fake — the return-to-isotropy and isotropization-of-production effects modeled through the pressure–strain redistribution term, and the direct production of stress anisotropy by rotation and curvature — enter these equations explicitly (classic pressure–strain models include the Launder–Reece–Rodi (LRR) and Speziale–Sarkar–Gatski (SSG) forms). Their price is computational cost, numerical stiffness, realizability constraints (the modeled stresses must remain physically admissible), and delicate near-wall treatment, for example the elliptic-blending approach of Manceau and Hanjalić that smoothly connects near-wall and weakly inhomogeneous regions. The aeronautical variant now used as a community reference is the blended SSG/LRR-ω\omega model, developed in the European FLOMANIA collaborative turbulence-model benchmarking project and documented on the NASA Turbulence Modeling Resource (TMR) website.

A related line of work reformulates stress transport in divergence form: Perot's turbulent potential model takes the divergence of the Reynolds-stress transport equations and evolves the turbulent scalar and vector potentials of the turbulent body force (the divergence of the Reynolds-stress tensor) instead of the stresses themselves. The divergence form eliminates the explicit pressure–strain redistribution term and was claimed in the model's own publications to have favorable numerical properties — Reynolds-stress-level accuracy at reduced computational cost, faster convergence, and reduced stiffness — claims that had not been quantified in controlled comparisons.

Whether the extra physics of stress-transport closures actually translates into systematically better predictions, as opposed to case-by-case improvements, is a long-standing question in turbulence modeling. Historical comparisons were confounded by implementation differences: variants of nominally the same model published in different journals or at different times produced different results in different codes, and grid convergence — agreement of the computed solution under systematic mesh refinement — was rarely demonstrated, so model error could not be cleanly separated from numerical error. Community responses include the European turbulence-research networks ERCOFTAC (the European Research Community on Flow, Turbulence and Combustion) and QNET-CFD, and, later, the NASA Turbulence Modeling Resource website maintained under the Turbulence Model Benchmarking Working Group of the American Institute of Aeronautics and Astronautics (AIAA), which documents model equations and variants, supplies verification and validation cases with grids and grid-converged reference solutions, and establishes naming conventions. Attributing an advantage to the closure class rather than to implementation differences or numerical error involves grid-converged solutions for both tiers in a common numerical framework and quantitative comparison against experimental or high-fidelity reference data across the flow classes that instantiate the complexity mechanisms; isolating the effect of the equation form on convergence and stiffness involves matched comparisons holding code, discretization, grids, and cases fixed.

In this context, Viti, Huang and Bradshaw implemented several recently proposed stress-transport models (including divergence-based formulations such as the Perot turbulent potential model and models in the elliptic-blending second-moment lineage) and validated them against experiment on deliberately simple wall-bounded and shear-free flows. They concluded that the study provided no basis for judging performance in complex flows. The study and the surrounding literature leave four issues unresolved: (1) whether stress-transport models give clear and systematic predictive advantages over eddy-viscosity two-equation models in flows with strong separation, rotation, curvature, or three-dimensional anisotropy; (2) whether divergence-based formulations deliver the computational advantages claimed in the model's own publications, such as faster convergence or reduced stiffness, which their validation study did not quantify; (3) the non-reproducibility of one candidate model whose published variants and available code/memoranda were mutually inconsistent; and (4) the absence of a standardized community infrastructure — agreed benchmark cases with reference numerical solutions, explicit discretization prescriptions, and mechanisms for transferring models from authors to implementers. The present problem record concerns the scientific core, issues (1) and (2): the systematic comparative assessment of the stress-transport tier against the two-equation tier, in both accuracy and computational cost. Issues (3) and (4) are community-infrastructure matters.

Problem Statement

Determine whether stress-transport (second-moment) turbulence closures provide clear and systematic predictive advantages over eddy-viscosity two-equation models in complex turbulent flows featuring strong separation, rotation, streamline curvature, or three-dimensional Reynolds-stress anisotropy, and whether divergence-based formulations of stress transport (such as the Perot turbulent potential model) provide systematic computational advantages — for example faster convergence rates or reduced equation stiffness — over conventional forms. The two sub-questions serve one objective: establishing, for the turbulence-closure hierarchy as used in engineering CFD, whether the additional physical content and the particular mathematical form of stress-transport closures translate into reproducible gains in accuracy or cost. The assessment addresses stress-transport closures and eddy-viscosity two-equation models as closure classes — the source study itself implemented several second-moment closures, including a divergence-based one, against simpler eddy-viscosity models — so a verdict on the systematic-advantage question cannot rest on a single solver-selected pair of models. An advantage counts as clear and systematic only when the stress-transport tier outperforms the two-equation tier, against the reference data, in a flow class for each of the four listed mechanisms — strong separation, rotation, streamline curvature, and three-dimensional Reynolds-stress anisotropy — on the accuracy metric designated in that case's reference data; advantages confined to isolated classes or metrics are case-by-case improvements. For the divergence-formulation sub-question, the divergence-based form counts as systematically advantageous only when it converges faster and/or is measurably less stiff than the matched conventional form in every matched case studied, where convergence and stiffness are measured by quantities such as iteration count or solution time to a fixed tolerance for convergence, and allowable time step or linear-system condition number for stiffness; advantages confined to isolated matched cases are case-by-case improvements.

References

  1. Viti V., Huang G., Bradshaw P., "Numerical study of stress-transport turbulence models: Implementation and validation issues", Computers & Fluids 36(8):1373–1383, 2007. DOI: 10.1016/j.compfluid.2006.11.006, https://doi.org/10.1016/j.compfluid.2006.11.006 (source of the question)
  2. Viti V., Huang G., Bradshaw P., "Implementation and Validation of Recent Stress-Transport Turbulence Models: the Need for Standardized Procedures", AIAA Paper 2005-112, 2005. DOI: 10.2514/6.2005-112, https://doi.org/10.2514/6.2005-112
  3. Perot B., "Turbulence modeling using body force potentials", Physics of Fluids 11(9):2645–2656, 1999. DOI: 10.1063/1.870126, https://doi.org/10.1063/1.870126
  4. Perot B., Are S., Wang C., "Application of the Turbulent Potential Model to Complex Flows", Proceedings of the 5th International Symposium on Turbulence Modeling and Measurements (ETMM5), 2002, pages 117–126. DOI: 10.1016/B978-008044114-6/50010-7, https://doi.org/10.1016/B978-008044114-6/50010-7
  5. Pecnik R., Sanz W., "Application of the Turbulent Potential Model to Heat Transfer Predictions on a Turbine Guide Vane", Journal of Turbomachinery 129(3):628–635, 2007. DOI: 10.1115/1.2720494, https://doi.org/10.1115/1.2720494
  6. Rumsey C. L., "Consistency, Verification, and Validation of Turbulence Models for Reynolds-Averaged Navier-Stokes Applications", 3rd European Conference for Aerospace Sciences (EUCASS), 2009. NTRS accession 20090026518, https://ntrs.nasa.gov/citations/20090026518
  7. Rumsey C. L., "Recent Developments on the Turbulence Modeling Resource Website (Invited)", AIAA Paper 2015-2927, 2015. DOI: 10.2514/6.2015-2927, https://doi.org/10.2514/6.2015-2927 (full text also available as NTRS 20160006057, https://ntrs.nasa.gov/citations/20160006057)
  8. Eisfeld B., Rumsey C., Togiti V., "Verification and Validation of a Second-Moment-Closure Model", AIAA Journal 54(5):1524–1541, 2016. DOI: 10.2514/1.J054718, https://doi.org/10.2514/1.J054718 (the substantially identical conference version, "Second-Moment RANS Model Verification and Validation Using the Turbulence Modeling Resource Website", AIAA Aviation 2015, is available as NTRS 20160006056, https://ntrs.nasa.gov/citations/20160006056)
  9. Cécora R.-D., Radespiel R., Eisfeld B., Probst A., "Differential Reynolds-Stress Modeling for Aeronautics", AIAA Journal 53(3):739–755, 2015. DOI: 10.2514/1.J053250, https://doi.org/10.2514/1.J053250
  10. Manceau R., "Recent progress in the development of the Elliptic Blending Reynolds-stress model", International Journal of Heat and Fluid Flow 51:195–220, 2015. DOI: 10.1016/j.ijheatfluidflow.2014.09.002, https://doi.org/10.1016/j.ijheatfluidflow.2014.09.002
  11. Slotnick J., Khodadoust A., Alonso J., Darmofal D., Gropp W., Lurie E., Mavriplis D., "CFD Vision 2030 Study: A Path to Revolutionary Computational Aerosciences", NASA/CR-2014-218178, 2014. NTRS accession 20140003093, https://ntrs.nasa.gov/citations/20140003093
  12. Duraisamy K., Iaccarino G., Xiao H., "Turbulence Modeling in the Age of Data", Annual Review of Fluid Mechanics 51:357–377, 2019. DOI: 10.1146/annurev-fluid-010518-040547, https://doi.org/10.1146/annurev-fluid-010518-040547
  13. Hu X., Huang G., Durbin P., Yang X., "Viscous-Layer Compressibility Correction for Two-Equation Reynolds-Averaged Navier–Stokes Models", AIAA Journal 63(3):993–1004, 2025. DOI: 10.2514/1.J064541, https://doi.org/10.2514/1.J064541