Information-Based Symbolic Regression of Local Closures for RANS Transition Modeling

Local-correlation transition models enable laminar–turbulent transition prediction in Reynolds-averaged Navier–Stokes simulations by replacing non-local boundary-layer quantities with local surrogate variables. This work proposes an information-based symbolic-regression framework to construct compact and interpretable surrogates for the Pohlhausen pressure-gradient parameter, λθ, and the momentum-thickness Reynolds number, Reθ. A Falkner–Skan database is used to identify informative local dimensionless features through the IT- π methodology and to derive explicit expressions for λθL and ReθL. The resulting closures reconstruct the corresponding integral quantities substantially more accurately than the baseline expressions. Candidate closures are selected according to reconstruction accuracy, algebraic complexity, and wall-normal activation behavior. They are then implemented in a one-equation intermittency model and assessed using the ERCOFTAC T3 transitional flat-plate series. The results are case dependent, with improvements for several bypass and pressuregradient cases and deterioration for two low-turbulence-intensity cases. Although recalibration of the remaining transition-model correlations may be required, the proposed closures provide a more interpretable representation of the local boundary-layer state and offer a promising route toward compact, data-informed, and physically transparent local-correlation transition models.