Gradient-Aware AI Agents for Autonomous Turbulence Model Discovery via Differentiable CFD
Organizations: Sibley School of Mechanical and Aerospace Engineering, Cornell University, Ithaca, 14853, NY, USA
Abstract
Turbulence modeling relies on human choices of closure form, physical inputs, and tunable parameters. Training closures a posteriori is efficient only with gradients through long chaotic rollouts, which few solvers provide. Most a posteriori approaches therefore resort to gradient-free optimization, and nearly all fix the closure form in advance, learning only its parameters. In this work, we introduce a gradient-aware agentic framework that poses closure discovery as a bilevel problem: a multi-agent large language model (LLM) system searches over closure forms, while a GPU-native differentiable computational fluid dynamics (CFD) solver supplies gradients, making differentiable a posteriori learning an action of the autonomous AI agent. For each proposal, the agents construct a differentiable training program by embedding the analytical, neural, or hybrid closure in the solver, implementing a differentiable objective, and validating the gradient path. They then decide whether to evaluate the closure, compute its parameter sensitivities, or train it a posteriori, select the optimization settings, and revise subsequent proposals using the simulation outcomes and returned gradients. Truncated-horizon gradients make this training usable under chaotic dynamics for closures with tens of thousands of parameters. Demonstrations include dissipation-law and subgrid-scale closure discovery in two-dimensional turbulence and wall-model discovery for large-eddy simulation of three-dimensional channel flow. With gradient-based training, the agent recovers reference coefficients that forward evaluation alone does not reach within the same budget, and autonomously discovers and trains analytical and neural closures that improve agreement with the targeted reference statistics. These results demonstrate the feasibility of end-to-end autonomous turbulence model development in time-resolved flows.
Figures & tables
| Symbolic regression [ Schmelzer et al., 2020 , Wu et al., 2025 , Weatheritt and Sandberg, 2016 , Zhao et al., 2020 , Waschkowski et al., 2022 , Li et al., 2024 , Jakhar et al., 2026 , Ma et al., 2026 , Fu et al., 2026 ] | RL closure discovery [ Novati et al., 2021 , Bae and Koumoutsakos, 2022 , Kurz et al., 2023 , Beck and Kurz, 2023 , Zhou and Bae, 2024 , von Bassewitz et al., 2025 ] | Differentiable/adjoint learning [ Um et al., 2020 , List et al., 2022 , Shankar et al., 2025 , Frezat et al., 2022 , Kim et al., 2026 , Freitas et al., 2025 , Fan et al., 2026d , Zhang et al., 2026 , Sirignano et al., 2020 , Sirignano and MacArt, 2023 , Kakka and MacArt, 2026 , Dehtyriov et al., 2026 , Saverio et al., 2026 ] | Yang et al. [ Yang et al., 2025 ] | AutoTurb [ Zhang et al., 2025 ] | PiT-PO [ Wang et al., 2026a ] | AI CFD Scientist [ Somasekharan et al., 2026 ] | PhysMiner [ Chen et al., 2026a ] | This work | |
| Open-space closure proposal a | ✓ b | – | – | ✓ | ✓ b | ✓ b | ✓ | ✓ | ✓ |
| Neural/hybrid closures | – | ✓ c | ✓ | ✓ d | – | – | – | – | ✓ |
| A posteriori parameter training | ✓ e | ✓ | ✓ | – | ✓ e | – f | ✓ e | – | ✓ |
| Evaluation in 3D turbulence g | ✓ | ✓ | ✓ | ✓ | – | – | – | ✓ | ✓ |
| Gradient feedback to proposer h | – | – | – | – | – | – | – | – | ✓ |
| Automated code validation | – | – | – | – | ✓ i | – | ✓ | – | ✓ |
| Case 1: dissipation law | Case 2: SGS closure | Case 3: wall model | |
| Problem | |||
| Base model | Forced 2D turbulence on a grid; linear drag | Forced 2D turbulence on a grid; linear drag , no SGS closure | Channel WMLES on a grid; equilibrium wall model, Vreman SGS |
| Reference | Same flow with drag and hyperviscosity | Same physics resolved on a grid | DNS at [ Moser et al., 1999 ] |
| Objective | Log-spectral error | Log-spectral error | Mean-velocity error in wall units |
| Closure interface | Body force on the resolved field | Eddy viscosity | Wall-stress magnitude |
| Agent may modify | Dissipation form, inputs, and parameters; forcing parameters | Form, inputs, and parameters of | Form, inputs, and parameters of |
Appendix figures & tables1 asset
Supplementary material from the paper’s appendix.
Appendix
Computational Fluid Dynamics
Computational fluid dynamics numerically models fluid motion, turbulence, heat and mass transport, and fluid–structure interactions.
See all 308 papers