Finite-Time Relaxation of Inertial Particle Clustering in Non-Equilibrium Turbulence

Taketo TOMINAGA, Ryo ONISHI · 2026

Read Finite-Time Relaxation of Inertial Particle Clustering in Non-Equilibrium Turbulence by Taketo TOMINAGA, Ryo ONISHI (2026) 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-2605-25539.txt.

Introduction

Taketo Tominaga and Ryo Onishi test whether particle clustering in shifting turbulence can be treated as if it settles at once. Their direct simulations of uniform turbulence under unsteady forcing show lag and hysteresis once the forcing period passes several large-eddy turnover times. For the heaviest particles, clustering at one dissipation rate takes two far apart values. A finite-time relaxation model then cuts the worst prediction error from 49 percent to 10 percent, with a similar gain on a fresh validation case.

Catalog record

CREATORTaketo TOMINAGA, Ryo ONISHI
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33765
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.25539

Browse the full library · About the Tower of Babel

Full text

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

Finite-Time Relaxation of Inertial Particle Clustering in Non-Equilibrium Turbulence Taketo TOMINAGA, Ryo ONISHI Abstract Inertial particles in turbulence form clusters, which strongly affect particle collisions and transport properties. Clustering models based on statistically stationary turbulence implicitly assume the instantaneous-equilibrium approximation when applied to time-varying non-equilibrium turbulence. However, the validity of this approximation remains unclear. In this study, the temporal response of inertial particle clustering in non-equilibrium turbulence was investigated using direct numerical simulation of homogeneous isotropic turbulence with unsteady forcing. Periodic responses of the flow and clustering intensity were evaluated by varying the forcing period. The flow showed non-equilibrium scaling for all forcing periods. The relationship between instantaneous energy dissipation rate and clustering intensity showed hysteresis exceeding statistically stationary fluctuations when the forcing period exceeded several large-eddy turnover times. For the particles with the largest inertia, clustering intensity took values of 0.80 and 1.56 times the reference value at the same instantaneous energy dissipation rate. This demonstrates that the instantaneous-equilibrium approximation is not appropriate under such conditions. A linear relaxation model was constructed from transient responses, in which clustering intensity approaches the instantaneous-equilibrium value with a finite relaxation time. The relaxation time scaling was identified as τ g = 1.0 ​ T e ​ ( t ) ​ St ​ ( t ) 0.40 \tau_{g}=1.0T_{\mathrm{e}}(t)\,\mathrm{St}(t)^{0.40} , where T e ​ ( t ) T_{\mathrm{e}}(t) and St ⁡ ( t ) \mathrm{St}(t) are the instantaneous large-eddy turnover time and Stokes number. The model reduced the maximum relative error from 49% to 10% for the particles with the largest inertia and from 76% to 22% in an independent validation case. These results demonstrate that finite-time relaxation improves prediction accuracy for clustering intensity in non-equilibrium turbulence. I Introduction Inertial particles in turbulence appear in a wide range of geophysical and engineering applications, such as cloud turbulence, aerosol transport, spray combustion, and solid-particle transport. Inertial particles are not uniformly distributed in turbulence and can form highly non-uniform spatial distributions, known as preferential concentration 8 ; 15 . Such non-uniform particle distributions are referred to as clustering and strongly affect particle collisions and transport properties 16 ; 19 . In cloud microphysics, the enhancement of cloud droplet collisions by turbulence has been widely studied as one mechanism for the rapid formation of warm rain 2 ; 4 ; 10 . In particular, inertial particle clustering increases the collision kernel by increasing the radial distribution function (RDF) g(r=R) at the contact distance R=2r_{\mathrm{p}} , where r_{\mathrm{p}} is the particle radius 19 ; 10 . To quantify this effect, g(r=R) and the collision kernel have been modeled based on direct numerical simulation (DNS) and theoretical analyses 14 ; 19 ; 1 ; 18 ; 11 . In addition, the effect of turbulence on cloud droplet growth has been investigated by incorporating these collision kernels into the stochastic collision–coalescence equation 20 ; 10 ; 11 . These models are based on clustering statistics in statistically stationary turbulent flows. Therefore, when they are applied to non-equilibrium turbulent flows in which the turbulent state varies in time, it is implicitly assumed that g(r=R) instantaneously changes to the equilibrium value corresponding to the turbulent state at that time. However, the validity of this assumption remains unclear. Indeed, Zapata et al. 21 studied particle clustering in non-equilibrium turbulence that spontaneously emerges under statistically stationary counter-rotating-column forcing. They reported that particle clustering responds to the turbulent state with a time lag and exhibits hysteresis. Nevertheless, the hysteresis of clustering intensity and its relaxation time have not been quantitatively characterized. In this study, the temporal response of inertial particle clustering in non-equilibrium turbulence is investigated using DNS of homogeneous isotropic turbulence with unsteady forcing, in which the energy injection rate varies in time. First, the response to periodic unsteady forcing is analyzed, and hysteresis in the relationship between the energy dissipation rate and clustering intensity is evaluated. Furthermore, a linear relaxation model is constructed, in which clustering intensity approaches the instantaneous-equilibrium value with finite-time relaxation. The relaxation time scaling included in the model is identified from transient-response DNS data obtained by switching the energy injection rate only once. The predictive performance of the obtained model is validated using independent DNS data. This study provides a basis for a clustering model applicable to time-varying turbulent flows by describing inertial particle clustering in non-equilibrium turbulence as a finite-time relaxation process. II Methods The DNS in this study was performed using the Lagrangian Cloud Simulator in Julia (LCS.jl) 17 . LCS.jl is a multi-platform simulation code based on the Lagrangian Cloud Simulator (LCS) by Onishi et al. 10 and shows high scalability for large-scale particle-tracking simulations on GPU environments. II.1 Flow Phase The carrier fluid follows the incompressible Navier–Stokes equations: \nabla\cdot\bm{U}=0, (1) \frac{\partial\bm{U}}{\partial t}+(\bm{U}\cdot\nabla)\bm{U}=-\frac{1}{\rho}\nabla p+\nu\nabla^{2}\bm{U}+\bm{F}(\bm{x},t). (2) Here, \bm{U} is the fluid velocity, \rho is the density, p is the pressure, and \bm{F} is the forcing term. Turbulence was maintained by the forcing term \bm{F} , which injects energy into large scales with wavenumber |\bm{k}|<2.5 . Reduced Communication Forcing (RCF), which is a modified Large Scale Forcing method, was used for turbulent forcing 12 . RCF enables energy injection only into large scales while reducing communication and computational costs by applying a box mean filter before transformation into wavenumber space, rather than transforming the entire flow field into wavenumber space. In RCF, the energy injection rate P(t)\equiv\frac{1}{V}\int_{V}\bm{U}\cdot\bm{F}\,\mathrm{d}V (3) is controlled to a prescribed value by adjusting the amplitude of the forcing term \bm{F} . Here, V is the volume of the computational domain, and \mathrm{d}V is the volume element. The governing equations were discretized on a uniform Cartesian grid, and the variables were stored on a MAC arrangement 5 . For spatial discretization, the conservative fourth-order central-difference scheme of Morinishi et al. 9 was applied to the convective term, a fourth-order central-difference scheme was applied to the viscous term, and a second-order central-difference scheme was applied to the pressure term. Time integration was performed using a two-stage second-order Runge–Kutta method. Pressure–velocity coupling was handled using the HSMAC method with Red–Black coloring 6 . Iterations were continued until the root-mean-square (RMS) of the velocity divergence became smaller than \delta/\Delta . Here, \Delta is the grid spacing, and \delta was set to 10^{-3} following Onishi et al. 10 . For stability, the time step \Delta t was determined to satisfy the convective and viscous CFL constraints: \mathrm{CFL}_{\mathrm{conv}}=\frac{\max(\|\bm{U}\|)\Delta t}{\Delta}\leq C_{1},\quad\mathrm{CFL}_{\mathrm{vis}}=\frac{\nu\Delta t}{\Delta^{2}}\leq C_{2}. (4) Here, C_{1}=C_{2}=0.3 was used. The computational domain was a cube with side length 2\pi L_{0} , and periodic boundary conditions were applied in all directions. The grid spacing was \Delta=2\pi L_{0}/N . Here, N is the number of grid points in each direction. The bulk Reynolds number is defined as \mathrm{Re}=U_{0}L_{0}/\nu . Here, U_{0} is the characteristic velocity, and L_{0} is the characteristic length. Three-dimensional domain decomposition was used for parallelization across processes. The number of divisions was selected so that each subdomain was as close to a cube as possible, and the communication cost was reduced by minimizing the surface-area-to-volume ratio. Onishi et al. 12 showed that these numerical methods maintain sufficient accuracy even compared with spectral methods. II.2 Particle Phase The particle motion is governed by the following system of ordinary differential equations: \frac{\mathrm{d}\bm{X}}{\mathrm{d}t}=\bm{V}, (5) \frac{\mathrm{d}\bm{V}}{\mathrm{d}t}=-\frac{f}{\tau_{\mathrm{p}}}\{\bm{V}-\bm{U}(\bm{X},t)\}.

Read the complete text