Least Inference Navier-Stokes (LINS): A Parameter-Free Turbulence Closure Model

Traditional Large Eddy Simulation (LES) closures, such as the Smagorinsky model, rely on empirical coefficients to reconcile the energy transfer between resolved and subgrid scales. These coefficients often require ad-hoc tuning or explicit wall-damping functions to handle anisotropic regimes. In this work, we propose the Least Inference Navier-Stokes (LINS) model, a subgrid closure derived entirely from information-theoretic principles. By treating the computational grid as an information filter rather than a discretization error, we hypothesize that the turbulent viscosity arises from the minimum loss of information required to maintain numerical stability at the Nyquist limit. We validate the LINS model across three canonical flow regimes without external calibration. First, in 3D Homogeneous Isotropic Turbulence, the model correctly recovers the Kolmogorov k −5/3 inertial subrange and the physical dissipation tail. Second, in turbulent channel flow, LINS naturally recovers the Law of the Wall (u + ∝ ln y +) and preserves near-wall anisotropic streaks, exhibiting intrinsic wall-damping behavior without the need for Van Driest functions. Finally, in high-Reynolds number vortex shedding, the model demonstrates localized activation within shear layers while preserving wake coherence. These results suggest that the missing physics in turbulence modeling can be recovered directly from the information geometry of the grid itself, offering a robust, parameter-free path for high-fidelity fluid simulations.