Homogeneous shear turbulence as a second-order cone program

TL;DR

This paper formulates homogeneous shear turbulence as a second-order cone program using convex optimization, aiming to address non-realizability issues and improve turbulence modeling accuracy.

Contribution

It introduces a novel convex optimization framework for turbulence modeling, incorporating multi-point correlations and constraints like NNVP, with two models of increasing complexity.

Findings

Predicted macro length scales beyond which correlations vanish.

Qualitative agreement of anisotropy tensor with experimental data.

Steady state solutions for the second-order model obtained.

Abstract

To help resolve issues of non-realizability and restriction to homogeneity faced by analytical theories of turbulence, we explore three-dimensional homogeneous shear turbulence of incompressible Newtonian fluids via optimal control and convex optimization. The framework is composed of multi-point spatial correlations of velocity and pressure fluctuations up to the degenerate fourth order, their evolution equations and constraints. The integral of trace of the second order correlations is argued as the objective functional to be maximized. The sources of the constraints are discussed like the Cauchy-Schwarz inequality and the non-negativity of variance of products (NNVP). Two models are defined: the second-order model uses the contracted and degenerate third order correlations as control variables; the third-order model takes the degenerate fourth order correlations as control variables. Both are second-order cone programs. Computation of large-scale and huge-scale and link to big data are noted. The exponential growth rates of the asymptotic states are bounded from above by zero. The steady state of the second-order model is solved. Three finite macro length scales are predicted beyond which the two-point correlations are negligible. The predicted values of the anisotropy tensor are consistent with experimental data qualitatively (concerning the relative numerical order of the diagonal components), albeit with large quantitative differences attributed to the non-enforceability of NNVP within the model. Compared with DNS data, the predicted second order correlation functions contain flawed features of local minima too large in magnitude or present spuriously. The third-order model is expected to improve predictions, owing to its ability to include constraints from NNVP and its formulation in an enlarged control variable space. It is yet open how to solve this huge-scale problem.