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

This study demonstrates that the instantaneous-equilibrium approximation fails to accurately predict inertial particle clustering in non-equilibrium turbulence due to finite-time relaxation effects, and proposes a new linear relaxation model incorporating a specific scaling law that significantly reduces prediction errors compared to traditional methods.

The Gist

Imagine you are at a crowded dance floor (the turbulence). In the middle of this crowd, there are thousands of tiny, heavy dancers (the inertial particles, like water droplets in a cloud or dust in the air).

Because these heavy dancers have momentum, they can't turn instantly like the light, nimble people around them. Instead, they get flung out of the spinning whirlpools and get pushed into the straight, stretching lanes. This causes them to bunch up in specific spots, forming tight little groups or clusters.

Scientists have long known that these clusters are important because they make the heavy dancers bump into each other more often. However, most of the old rules for predicting these clusters were written for a dance floor where the music and the crowd's energy never change (a statistically stationary state).

The Problem: The "Instant" Mistake

The big question this paper asks is: What happens when the music suddenly changes?

Imagine the DJ suddenly switches from a slow, mellow beat to a high-energy, fast-paced track, and then back again.

• The Old Assumption: Scientists used to assume that the heavy dancers would instantly rearrange themselves into new groups the moment the beat changed. They thought the clustering was an "instant equilibrium."
• The Reality: The authors of this paper discovered that this assumption is wrong. Just like a heavy dancer takes a few seconds to stop spinning and start running to a new spot, the clusters of particles take time to react to the changing energy of the turbulence. They don't snap to the new shape immediately; they "relax" into it over a finite period.

The Experiment: A Dance Floor with a Rhythm

To prove this, the researchers used a supercomputer to simulate a 3D dance floor. They didn't just let the music play randomly; they programmed the energy injection to pulse up and down in a perfect rhythm (like a heartbeat).

They tested different speeds for this rhythm:

1. Fast Rhythm: The beat changed so quickly that the heavy dancers couldn't keep up at all.
2. Slow Rhythm: The beat changed slowly enough that the dancers had time to react, but not so slowly that they were perfectly in sync.

What they found:
When the rhythm was slow enough (specifically, when the time between beats was longer than the time it takes for a large swirl in the crowd to spin once), the clusters showed a phenomenon called hysteresis.

Think of hysteresis like a sticky door.

• If you push the door open (increase energy), it opens at a certain point.
• If you pull it closed (decrease energy), it doesn't close at the exact same point; it stays open a bit longer because of the "stickiness" (the inertia).
• In the simulation, for the same amount of energy in the room, the clusters were completely different depending on whether the energy was just rising or just falling.
  • When the energy was rising, the clusters were very weak (only 80% of the expected size).
  • When the energy was falling, the clusters were very strong (156% of the expected size).

This proved that you cannot just look at the current energy level to know how the particles are clustered; you have to know the history of how the energy got there.

The Solution: A New Rulebook

The researchers realized that the old "instant" rulebook was failing. So, they built a new, simpler model to fix it.

They treated the clustering process like a spring or a shock absorber on a car.

• When the road (turbulence) changes, the car doesn't instantly snap to the new height. It bounces and settles over a specific amount of time.
• They calculated exactly how long this "settling time" (relaxation time) takes. They found it depends on two things:
  1. How big the swirls in the crowd are (Large-eddy turnover time).
  2. How heavy the dancers are compared to the crowd (Stokes number).

Their new formula is: Relaxation Time = (Swirl Size) × (Heaviness)^0.40.

The Result: Much Better Predictions

They tested this new "spring" model against their computer simulations.

• The Old Model (Instant): Made huge mistakes, sometimes being off by nearly 50% for the heaviest particles. It was like guessing the car's height without accounting for the bounce.
• The New Model (Finite-Time): Drastically improved accuracy, reducing the error down to just 10%. Even when they tested it on a completely different set of conditions (a different "dance floor"), it still cut the error from 76% down to 22%.

The Takeaway

This paper tells us that in the chaotic world of non-equilibrium turbulence (where energy is constantly changing), particles don't react instantly. They have a "memory" and a reaction time. By acknowledging this delay and adding a simple "settling time" to our calculations, we can predict how particles clump together with much greater accuracy. This is crucial for understanding things like how rain forms in clouds, where the timing of droplet collisions matters immensely.

Technical Summary

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

Problem Statement
Inertial particles in turbulent flows exhibit preferential concentration, forming clusters that significantly influence collision rates and transport properties. Current models for particle clustering, particularly in cloud microphysics, are largely derived from statistically stationary turbulence. When applied to time-varying, non-equilibrium turbulence, these models implicitly assume an "instantaneous-equilibrium approximation," positing that the radial distribution function (RDF), $g(r)$, adjusts immediately to the instantaneous turbulent state (e.g., energy dissipation rate). However, the validity of this assumption in non-equilibrium conditions remains unclear. Previous studies have suggested that particle clustering may respond to turbulent state changes with a time lag, exhibiting hysteresis, but the intensity of this hysteresis and the associated relaxation times have not been quantitatively characterized.

Methodology
The authors employed Direct Numerical Simulation (DNS) of homogeneous isotropic turbulence using the Lagrangian Cloud Simulator in Julia (LCS.jl). The study utilized two primary forcing strategies to generate non-equilibrium turbulence:

1. Periodic Unsteady Forcing: The energy injection rate, $P(t)$, was switched alternately between strong ($P_s$) and weak ($P_w$) phases with varying periods ($T$). This allowed for the evaluation of periodic responses and the extraction of hysteresis loops via phase averaging over 100 cycles.
2. Step-Once Unsteady Forcing: The energy injection rate was switched once from a strong steady state to a weak steady state. This transient response was used to construct and validate a linear relaxation model.

The simulations covered a range of Taylor Reynolds numbers ($Re_\lambda \approx 84\text{--}216$) and Stokes numbers ($St_s = 0.25\text{--}4.0$). The clustering intensity was quantified using the RDF at the contact distance, $g \equiv g(r=2r_p)$.

Key Results

• Non-Equilibrium Scaling: The turbulent flow consistently exhibited non-equilibrium dissipation scaling ($C_\epsilon \propto Re_\lambda^{-1}$) across all forcing periods, confirming the generation of a non-equilibrium turbulent state.
• Hysteresis in Clustering: A clear hysteresis loop was observed in the relationship between the instantaneous energy dissipation rate ($\epsilon$) and clustering intensity ($g^*$) when the forcing period ($T^*$) exceeded several large-eddy turnover times ($T_e$).
  • For $T^* = 12.0$ and $St_s = 4.0$, the clustering intensity took two distinct values (0.80 and 1.56 times the reference) for the same instantaneous $\epsilon$, depending on the flow history.
  • This hysteresis exceeded the natural fluctuations observed in statistically stationary turbulence, demonstrating that the instantaneous-equilibrium approximation fails under these conditions.
  • The direction and shape of the hysteresis loops depended on the Stokes number, reflecting the non-monotonic relationship between equilibrium clustering intensity and $\epsilon$.
• Finite-Time Relaxation: The study identified that clustering intensity relaxes toward the instantaneous-equilibrium value with a finite time scale. A linear relaxation model was constructed:
  $$\frac{d g_{noneq}}{dt} = \frac{g_{eq}(t) - g_{noneq}(t)}{\tau_g}$$
  where the relaxation time $\tau_g$ scales as $\tau_g = 1.0 T_e(t) St(t)^{0.40}$.

Model Validation and Performance
The proposed linear relaxation model (noneq-model) was validated against independent DNS data (Case S2) and compared against the instantaneous-equilibrium model (eq-model).

• For particles with the largest inertia ($St_s = 4.0$), the maximum relative error of the noneq-model was reduced from 49% (eq-model) to 10% in the training case (S1) and from 76% to 22% in the independent validation case (S2).
• The improvement in prediction accuracy was most significant for high Stokes numbers, where the relaxation time is longest. For lower Stokes numbers, the instantaneous-equilibrium approximation remained relatively accurate due to shorter relaxation times.

Significance
This study demonstrates that the instantaneous-equilibrium approximation is inappropriate for describing inertial particle clustering in non-equilibrium turbulence when the forcing period exceeds several large-eddy turnover times. By characterizing the clustering response as a finite-time relaxation process, the authors provide a quantitative basis for improving clustering models in time-varying turbulent flows. The identified scaling law for relaxation time allows for the construction of more accurate predictive models for particle-laden flows in non-equilibrium conditions, such as those found in atmospheric cloud microphysics and engineering applications.