Information-Preserving SGS model based on the local inter-scale equilibrium hypothesis

TL;DR

This paper introduces an information-theoretic data-driven SGS model for turbulence simulation, maximizing mutual information to preserve local inter-scale equilibrium, improving interpretability and generalization.

Contribution

It proposes a novel SGS model based on mutual information maximization, addressing extrapolation issues and enhancing physical interpretability in turbulence modeling.

Findings

A priori tests align parameters with empirical values.

A posteriori tests show accuracy comparable to existing models.

Model enhances interpretability without empirical parameter tuning.

Abstract

Large eddy simulation has been widely used to simulate turbulence at balanced computational cost and accuracy. Many Subgrid-Scale (SGS) models have been proposed over the years, where data-driven and machine learning-aided approaches set the recent trend. To address the problem of extrapolation in these models, we propose a new data-driven SGS model based on an information-theoretic picture of turbulence. To this end, we estimate the model parameters by maximizing mutual information, which correspond to the scale-by-scale local equilibrium hypothesis in developed turbulence or "information preservation." An a priori test confirmed that the estimated parameters are in good agreement with the previously reported empirical values. Furthermore, a posteriori tests on periodic box turbulence and channel turbulence exhibited accuracy comparable to the existing models. These results suggest the utility of the information-theoretic picture of turbulence for constructing more generic SGS models without the need for empirically prescribed model parameters, while enhancing physical interpretability beyond black-box approaches.