Two-point Turbulence Closures in Physical Space

Noah Zambrano 1, Karthik Duraisamy 1 · 2025

Read Two-point Turbulence Closures in Physical Space by Noah Zambrano 1, Karthik Duraisamy 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-00669.txt.

Introduction

Noah Zambrano and Karthik Duraisamy write a turbulence closure that lives in physical space rather than Fourier space. Two-point closures track how flow statistics at pairs of points evolve, but they usually need spectral transforms that fail near sharp jumps. Their approach evolves the longitudinal correlation function directly, in discrete physical form, for uniform isotropic turbulence as the proving ground.

Linear terms keep the near-exact form prized from rapid distortion theory, while nonlinear moments close along the lines of the classic EDQNM model. New features appear: a matrix exponential inside the evolution equation and triple integrals from the nonlocal pressure term. Nonlocal length-scale information enters on its own. Checks against direct simulations and classical forecasts succeed, and extensions to uneven, directional flow are sketched for flows where spectral methods falter.

Catalog record

CREATORNoah Zambrano 1, Karthik Duraisamy 1
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE98 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-00669

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Two-point Turbulence Closures in Physical Space Noah Zambrano 1, Karthik Duraisamy 1 Abstract This work presents a predictive two-point statistical closure framework for turbulence formulated in physical space. A closure model for ensemble-averaged, incompressible homogeneous isotropic turbulence (HIT) is developed as a starting point to demonstrate the viability of the approach in more general flows. The evolution equation for the longitudinal correlation function is derived in a discrete form, circumventing the need for a Fourier transformation. The formulation preserves the near-exact representation of the linear terms, a defining feature of rapid distortion theory. The closure of the nonlinear higher-order moments follows the phenomenological principles of the Eddy-Damped Quasi-Normal Markovian (EDQNM) model of Orszag 1970 . Several key differences emerge from the physical-space treatment, including the need to evaluate a matrix exponential in the evolution equation and the appearance of triple integrals arising from the non-local nature of the pressure–Poisson equation. This framework naturally incorporates non-local length-scale information into the evolution of turbulence statistics. Verification of the physical-space two-point closure is performed by comparison with direct numerical simulations of statistically stationary forced HIT and with classical EDQNM predictions for decaying HIT. Finally, extensions to inhomogeneous and anisotropic turbulence are discussed, emphasizing advantages in applications where spectral methods are ill-conditioned, such as compressible flows with discontinuities. † † affiliation: 1 Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI 48103 1 Introduction In turbulence modeling, the Navier–Stokes equations serve as the foundation for deriving evolution equations of coarse-grained statistical quantities, which in turn provide closure for the coarse-grained equations themselves. The coarse-graining procedure, introduced to mitigate computational expense, depends on the choice of averaging or filtering operator—such as ensemble, temporal, or spatial averaging—selected according to the modeling framework. Application of this operator yields statistical moments that may be formulated as single- or multi-point, higher-order quantities. Owing to the prevalence of spatial derivatives in the governing equations, multi-point statistics offer a natural description of turbulent interactions. Nevertheless, most practical turbulence models employ single-point closures to reduce cost and analytical complexity, necessitating additional modeling assumptions to represent intrinsically multi-point phenomena. Single-point closures often fail to accurately account for nonlocal interactions, anisotropy, and energy transfer across scales with sufficient fidelity. Two-point closures, such as the Eddy Damped Quasi‐Normal Markovian (EDQNM) ( Orszag 1970 ) , Test‐Field Models (TFM) ( Kraichnan 1971 ) , Direct Interaction Approximation (DIA) ( Kraichnan 1959 ) , and related approaches, address many of these limitations, albeit demonstrated in canonical, and often homogeneous turbulence. These offer explicit representations of the two-point correlation functions which directly model scale‐to‐scale energy transfer. Two-point methods also allow representation of structural and wave effects – e.g. rotation and stratification – that are essentially inaccessible to simpler closures ( Sagaut & Cambon 2018 ) . Another advantage of two-point closures is the reduced reliance on empirical constants; in some formulations, closures can predict or constrain what would otherwise be ad hoc parameters in single-point or lower-order models ( Bos & Bertoglio 2006 ) . Two-point closures also naturally reduce to rapid distortion theory in the rapid distortion/linear limit. This limiting behavior can be exploited in problems where the turbulence does not have time to interact with itself and the turbulence dynamics are significantly simplified, circumventing the closure problem in some special cases. Collectively, these advantages make two-point closures an appealing framework for flows where non‐locality, anisotropy, and details of energy transfer are important. Historically, most two-point turbulence closures have been formulated in spectral space, where the governing equations for two-point velocity correlations simplify considerably due to statistical homogeneity. Classical models such as EDQNM and DIA operate on the energy spectrum and directly capture triadic interactions across wavenumbers ( Orszag 1970 ; Lesieur & Schertzer 1978 ) . Spectral representations make the scale-to-scale transfer terms transparent and computationally efficient to evaluate. However, spectral methods tend to be limited to special canonical cases because of the mathematical limitations of the Fourier transform, and the vast majority of spectral models are formulated for incompressible flows. Very few works in the literature focus on variable-density cases, most notably the variable-density models by Besnard et al. 1996 and Clark & Spitz 1995 , and less on compressible flows, with the main example being the modeling work for compressible HIT by Bertoglio et al. 2001 . In a strict sense, application of Fourier transforms to problems with discontinuities, such as compressible flows with shocks, is ill-posed. Canuto et al. 2007 discusses the application of spectral methods to such problems, but almost all applications are spectral collocation methods which only transform the spatial derivatives to spectral space and not the dynamical variables. Without transformation of the variables, many of the spectral-based closure ideas originally developed for higher-order turbulence moments cannot be applied to this method. Therefore, this provides motivation for development of a two-point framework that enables translation of closure ideas to more complex discontinuous problems. A natural approach to this problem is to work exclusively in physical space. Two-point closures are less common in physical space; yet they offer several advantages: they can be extended to inhomogeneous flows, more naturally preserve the locality of boundary conditions, and often provide more direct connections to experimentally measurable two-point statistics. Notable early developments include the work of Besnard et al. 1996 , who formulated a two-point closure in physical space for variable-density turbulence but ultimately utilized a Fourier transformation, and Cambon & Rubinstein 2006 , who advocated for real-space correlation closures as a complement to spectral models. More recently, scale-space transport equations based on the generalized Kármán–Howarth-Monin equations ( Von Kármán & Howarth 1938 ; Monin 1959 ) have been derived explicitly in physical space, allowing one to track energy transfer simultaneously across spatial locations and separations. Most of these methods use the scale-space energy density function of Hamba 2015 . The work of Arun et al. 2021 provides an example of how the scale-space transform can be used to compute two-point correlations in inhomogeneous compressible flows. However, these studies generally do not use the scale-space energy density function framework to derive closures of the moment equations; they are used as analysis/diagnostic tools for DNS data. To the authors’ knowledge, there has not been development of a two-point physical space framework that can be systematically used to convert the relevant unclosed moment PDEs into ODEs and apply the closure ideas originally developed for spectral methods. The purpose of this work is to begin the development of a framework that can be used for prediction. This work will consider the closure of incompressible HIT as a starting point to demonstrate the viability of the physical space two-point formulation and how it may be used in a predictive setting. HIT has been explored by many, including progress on two-point closures in physical space using the Kármán-Howarth ( Von Kármán & Howarth 1938 ) equations. The early work of Domaradzki & Mellor 1984 and Oberlack & Peters 1993 explored closure of the Kármán-Howarth equations with an eddy-viscosity model. Further work on modeling higher-order structure functions by Thiesset et al. 2013 and Djenidi & Antonia 2022 , to name a few, have also relied on eddy-viscosity type closures. All these methods rely on the Kármán-Howarth equations, which cannot be

Read the complete text