3:00 PM - 3:15 PM
[AAS05-06] Local Clustering of Inertial Cloud Droplets in Non-Equilibrium Turbulence
Keywords:Clustering of Inertial Particles, Non-Equilibrium Turbulence, Cloud Microphysics, Supercomputer, Multiphase Flow
It has been pointed out that cloud turbulence may accelerate the onset of rainfall by promoting the collisional growth of cloud droplets. The mechanism of this accelerated collision effect still requires further investigation. For example, the inhomogeneous distribution (clustering) of small inertial particles such as cloud droplets, which plays a significant role in the collision process, needs to be clarified and modeled.
Actual cloud turbulence exists in a non-equilibrium state where the energy input and dissipation are imbalanced locally. However, previous studies have conducted numerical simulations assuming an equilibrium state. Focusing on the impact of this non-equilibrium state, this study aims to clarify the clustering mechanism of small inertial particles in a non-equilibrium homogeneous isotropic turbulence using large-scale numerical simulations.
Lagrangian Cloud Simulator (LCS) (Onishi et al., 2015) was used for simulations. LCS solves the flow field using a Eulerian approach while tracking particles using a Lagrangian approach, allowing for detailed calculations of individual particle motion. The attached figure shows a snapshot of the spatial distribution of 327680 particles in homogeneous isotropic turbulence computed on a 64^3 grid.
The flow field is discretized from the incompressible Navier-Stokes equations using an energy-conserving fourth-order central difference method (Morinishi et al., 1998) and advanced in time explicitly using a two-stage second-order Runge-Kutta method. The HSMAC method is employed for pressure-velocity coupling. Energy is injected into the large-scale flow field using Reduced Communication Forcing (RCF) (Onishi et al., 2011). This forcing method downsamples the flow field, enabling selective energy injection into the large-scale structures while significantly reducing communication overhead.
Particles are advected by the fluid forces, but their feedback on the fluid field is neglected due to their small size. The hydrodynamic interactions between particles are computed using the Binary-based Superposition Method (BiSM) (Onishi et al., 2013). BiSM approximates the influence of multiple particles as a superposition of two-particle interactions, allowing for fast and accurate computations when particle number density is low, as is the case in atmospheric clouds.
The conventional LCS was written in FORTRAN90 and was limited to CPU-based calculations. In this study, LCS was rewritten in Julia to support large-scale GPU-based computations. This modification enabled highly efficient large-scale simulations using multiple GPUs. Specifically, the largest simulations performed in this study used 64 GPUs to compute the motion on a 2048^3 grid and 10^9 small inertial particles.
To investigate particle clustering in non-equilibrium turbulence, a stepwise variation in energy injection was introduced to realize a non-equilibrium state in the large-scale flow field. Particles with various inertia values were introduced into the generated non-equilibrium turbulent flows, and the radial distribution function at contact (g(r=R); R is the collision radius), a near-field radial distribution function, was computed.
The simulation results show discrepancies between the temporal response of the radial distribution function and the energy dissipation rate when external forcing is varied in a stepwise manner. Furthermore, the radial distribution functions in non-equilibrium turbulent flows differ from those obtained under conditions of temporal-mean energy dissipation rate and temporal-mean Reynolds number. This finding suggests that a collision frequency model incorporating the non-equilibrium nature of turbulence would be required for more reliable cloud modeling.
Actual cloud turbulence exists in a non-equilibrium state where the energy input and dissipation are imbalanced locally. However, previous studies have conducted numerical simulations assuming an equilibrium state. Focusing on the impact of this non-equilibrium state, this study aims to clarify the clustering mechanism of small inertial particles in a non-equilibrium homogeneous isotropic turbulence using large-scale numerical simulations.
Lagrangian Cloud Simulator (LCS) (Onishi et al., 2015) was used for simulations. LCS solves the flow field using a Eulerian approach while tracking particles using a Lagrangian approach, allowing for detailed calculations of individual particle motion. The attached figure shows a snapshot of the spatial distribution of 327680 particles in homogeneous isotropic turbulence computed on a 64^3 grid.
The flow field is discretized from the incompressible Navier-Stokes equations using an energy-conserving fourth-order central difference method (Morinishi et al., 1998) and advanced in time explicitly using a two-stage second-order Runge-Kutta method. The HSMAC method is employed for pressure-velocity coupling. Energy is injected into the large-scale flow field using Reduced Communication Forcing (RCF) (Onishi et al., 2011). This forcing method downsamples the flow field, enabling selective energy injection into the large-scale structures while significantly reducing communication overhead.
Particles are advected by the fluid forces, but their feedback on the fluid field is neglected due to their small size. The hydrodynamic interactions between particles are computed using the Binary-based Superposition Method (BiSM) (Onishi et al., 2013). BiSM approximates the influence of multiple particles as a superposition of two-particle interactions, allowing for fast and accurate computations when particle number density is low, as is the case in atmospheric clouds.
The conventional LCS was written in FORTRAN90 and was limited to CPU-based calculations. In this study, LCS was rewritten in Julia to support large-scale GPU-based computations. This modification enabled highly efficient large-scale simulations using multiple GPUs. Specifically, the largest simulations performed in this study used 64 GPUs to compute the motion on a 2048^3 grid and 10^9 small inertial particles.
To investigate particle clustering in non-equilibrium turbulence, a stepwise variation in energy injection was introduced to realize a non-equilibrium state in the large-scale flow field. Particles with various inertia values were introduced into the generated non-equilibrium turbulent flows, and the radial distribution function at contact (g(r=R); R is the collision radius), a near-field radial distribution function, was computed.
The simulation results show discrepancies between the temporal response of the radial distribution function and the energy dissipation rate when external forcing is varied in a stepwise manner. Furthermore, the radial distribution functions in non-equilibrium turbulent flows differ from those obtained under conditions of temporal-mean energy dissipation rate and temporal-mean Reynolds number. This finding suggests that a collision frequency model incorporating the non-equilibrium nature of turbulence would be required for more reliable cloud modeling.