Polydisperse collision kernels in droplet-laden turbulence with implications for rain formation

Through direct numerical simulations of polydisperse droplets in turbulence, this study reveals how polydispersity differentially affects collision rates across droplet sizes, proposes improved models for the collision kernel and radial distribution function to address existing bidisperse errors, and demonstrates that turbulent intermittency accelerates droplet growth to help overcome the rain formation bottleneck.

The Gist

Imagine you are watching a massive, chaotic dance party inside a cloud. The dancers are tiny water droplets, and the music is the wind (turbulence). For rain to start, these dancers need to bump into each other, hold hands (coalesce), and grow big enough to fall to the ground.

But there's a problem. Scientists call this the "Bottleneck."

There is a specific size range for these dancers (about the width of a human hair) where they are too heavy to be carried easily by the wind, but too light to fall fast enough to bump into others. It's like trying to get two people to meet in a crowded room where they are both moving too slowly to find each other. For a long time, scientists couldn't explain how rain starts so quickly when the math says it should take hours or days.

This paper, written by researchers at ETH Zürich, uses super-computers to simulate this dance party and figure out how the "bottleneck" gets broken. Here is the breakdown of their discovery in simple terms:

1. The Two Ways Droplets Meet

The researchers looked at two main ways droplets collide:

• The "Clumping" Effect (Preferential Concentration): Imagine the wind is swirling. Heavy droplets get flung out of the swirls and land in the calm spots between them. If many droplets land in the same calm spot, they clump together like kids at a water fountain. This makes collisions more likely.
• The "Slingshot" Effect: Sometimes, the wind is so chaotic that it flings droplets in opposite directions at high speeds. They cross paths like cars on a highway, crashing into each other with high speed.

2. The Problem with "Different Sizes"

In real clouds, droplets aren't all the same size. Some are tiny, some are medium. The researchers asked: Does having a mix of sizes help or hurt the collision process?

They found a surprising split in behavior:

• For the tiny dancers (Small Stokes Numbers): Having different sizes helps. It's like a game of tag where the fast runners (small droplets) and slow walkers (medium droplets) are in the same room. Because they move differently, they cross paths more often. The "mix" increases the chance of a bump.
• For the bigger dancers (Large Stokes Numbers): Having different sizes hurts. If the dancers are too heavy, they stop following the wind's rhythm. The heavy ones go one way, the medium ones go another, and they stop clumping together. The "mix" actually makes them avoid each other.

3. Fixing the "Map"

Scientists use "maps" (mathematical models) to predict how often droplets collide. The researchers found that the old maps were wrong.

• The Old Map: Assumed that if you mixed sizes, the droplets would still clump together nicely.
• The New Map: Shows that when sizes are different, the droplets actually scatter and stop clumping as much as we thought. The old map was overestimating how often they would meet.

The team created a new, simpler formula (a new map) that accounts for this scattering. It's like realizing that in a mixed crowd, people don't stick together in tight groups; they spread out more. This new formula works much better for predicting rain formation.

4. The "Lucky Droplet" Theory

The most exciting part of the paper is about Turbulent Intermittency.
Imagine the dance floor isn't evenly windy. Instead, there are tiny, super-intense pockets of chaos (like a sudden gust of wind in one corner of the room) while the rest of the room is calm.

The researchers simulated these "pockets of chaos." They found that in these intense spots:

• Droplets grow much faster.
• A few "lucky" droplets get caught in these super-windy pockets, get slingshotted around, and grow huge very quickly.
• This explains the "Bottleneck": Even if the average wind is too weak to make rain, these rare, intense pockets allow a few lucky droplets to break through the barrier and start the rain process.

5. Gravity's Role

Finally, they added gravity back into the simulation (since real clouds have gravity). They found that for the smallest droplets in the "bottleneck," the wind (turbulence) is actually more important than gravity. Gravity only becomes the main driver once the droplets get a bit bigger.

The Big Picture

This paper tells us that rain formation is less about a steady, slow process and more about chaos and luck.

• Chaos: The wind creates pockets of extreme turbulence that act like accelerators.
• Luck: A few droplets get caught in these pockets and grow fast enough to become rain.

By understanding that mixing different droplet sizes changes how they clump, and by recognizing the power of these "lucky" turbulent pockets, scientists can now build better weather models. This means we might eventually get much more accurate predictions about when and where it will rain!

Technical Summary

1. Problem Statement

The formation of rain in warm clouds (ice-free) faces a well-known "bottleneck" problem: droplets with radii between 15 µm and 40 µm grow too slowly via condensation or gravitational settling to initiate precipitation within the typical 30-minute lifespan of a cloud.

• The Challenge: Turbulence is hypothesized to accelerate this growth by enhancing collision rates, but accurate parameterization of the collision kernel ($\Gamma$) for polydisperse (varying sizes) droplets in turbulence remains elusive.
• The Gap: Existing models (e.g., Onishi, Wang, Zhou) often rely on low Reynolds number simulations or simplified assumptions (like "frozen" flow fields) that fail to capture the complex interplay between preferential concentration (clustering), differential sampling (size-dependent velocity differences), and the sling effect (caustics) at high Reynolds numbers relevant to atmospheric clouds.

2. Methodology

The authors employed Direct Numerical Simulations (DNS) to resolve the full turbulence spectrum and particle dynamics without subgrid modeling.

• Flow Conditions: Simulations of homogeneous, isotropic turbulence at three Taylor-scale Reynolds numbers: $Re_\lambda = 55, 179,$ and $418$. The highest value ($418$) represents the upper limit of current computational capabilities for fully resolved DNS.
• Particle Dynamics:
  • Modeled as inertial point particles with a density ratio $\rho_p/\rho_f = 800$ (water in air).
  • Stokes numbers ($St$) ranged from 0.02 to 2.0, covering the critical bottleneck range.
  • Gravity: Neglected in the primary analysis to isolate turbulent mechanisms. Reintroduced in specific sections to test model robustness.
• Collision Statistics:
  • Used a ghost-collision approach (particles pass through each other) to count collisions efficiently.
  • Computed three key metrics for bidisperse pairs $(i, j)$:
    1. Collision Kernel ($\Gamma_{ij}$): Rate of collisions normalized by number density.
    2. Radial Relative Velocity ($\langle w_r \rangle$): Mean relative velocity at contact.
    3. Radial Distribution Function ($g_{ij}$): Measure of spatial clustering at contact.
• Coalescence Simulations: A subset of simulations included actual coalescence (merging) to study the evolution of the Droplet Size Distribution (DSD) under varying local dissipation rates.

3. Key Contributions

A. Comprehensive High-$Re_\lambda$ Data

The study provides the most extensive map of bidisperse collision kernels to date, covering $St \in [0.02, 2]$ at $Re_\lambda = 418$. This fills a critical gap between low-$Re$ theoretical models and atmospheric conditions.

B. Identification of Model Failures

By benchmarking against the Onishi framework (which combines the Zhou model for clustering and Wang model for velocity), the authors identified systematic errors:

• Overestimation of Clustering: The Zhou model (Eq. 8) significantly overestimates the spatial overlap (correlation coefficient $\rho_{ij}$) between droplets of different sizes, particularly when Stokes numbers diverge.
• Velocity Underestimation: The Wang model underestimates relative velocities for heavier particles where the sling effect dominates.

C. New Parameterization for Bidisperse Clustering

The authors propose a new parameterization for the bidisperse correlation coefficient $\rho_{ij}$.

• Insight: They found that the standard ratio $\phi = \max(St_i/St_j, St_j/St_i)$ is insufficient. Instead, they introduced a dispersion parameter $\theta_{ij}$:
  $$\theta_{ij} = \frac{|St_i - St_j|}{St_i + St_j}$$
• Result: The correlation coefficient decays much faster with $\theta$ than previously thought. They fitted a new exponential function (Eq. 11) that accurately captures the rapid decorrelation of clusters as size differences increase.

D. Novel Collision Kernel Model

Based on the new clustering model, they derived a compact, linear parameterization for the bidisperse collision kernel:
$$\tilde{\Gamma}_{ij} = \theta_{ij}\Gamma_{0i} + (1 - \theta_{ij})\Gamma_{ii}$$
Where:

• $\Gamma_{ii}$ is the monodisperse kernel (inertial effects).
• $\Gamma_{0i}$ is the kernel between a tracer-like droplet and a droplet of size $i$ (differential sampling effects).
• Significance: This model successfully reproduces DNS data with errors generally $<10\%$ and remains qualitatively valid even when gravitational settling is included.

4. Key Results

• Regime Transition: The effect of polydispersity on collision rates is non-monotonic and depends on the Stokes number:
  • Low $St$($< 0.4$): Polydispersity enhances collisions. The increase in relative velocity due to differential sampling outweighs the loss of clustering.
  • High $St$($> 0.4$): Polydispersity attenuates collisions. The rapid decorrelation of spatial clusters (loss of preferential concentration) dominates, reducing the collision rate compared to monodisperse cases.
• Reynolds Number Sensitivity: While clustering shows some sensitivity to $Re_\lambda$, the functional form of the bidisperse correlation coefficient (using $\theta$) collapses well across all simulated Reynolds numbers, suggesting the proposed model is robust.
• Turbulent Intermittency & "Lucky Droplets":
  • Simulations of coalescing droplets showed that in regions of high local energy dissipation ($\varepsilon$), droplet growth is markedly accelerated.
  • A small fraction of "lucky" droplets undergo rapid, correlated collisions, leading to a broadening of the DSD with long tails.
  • This supports the hypothesis that turbulent intermittency (rare, intense events) helps overcome the bottleneck barrier, rather than mean-field turbulence statistics.
• Gravity's Role: While gravity reduces clustering for monodisperse droplets, it enhances collisions for polydisperse pairs via differential settling. However, the proposed parameterization remains predictive even with gravity, as the underlying physics of the linear blend between inertial and differential sampling effects holds.

5. Significance and Implications

• Weather and Climate Modeling: The study provides a physically grounded, high-fidelity parameterization for the collision kernel that can be implemented in Large-Eddy Simulations (LES) and global climate models. This is crucial for improving the prediction of warm rain initiation.
• Resolution of the Bottleneck: The results strongly support the theory that turbulent intermittency is the key mechanism allowing droplets to bypass the 15–40 µm growth barrier. It suggests that rain initiation is controlled by the "fat tails" of probability distributions (rare intense events) rather than mean statistics.
• Modeling Paradigm Shift: The work challenges the reliance on "frozen flow" approximations in previous models, demonstrating that dynamic flow fields significantly alter clustering statistics. It also establishes that differential sampling is a dominant mechanism for small droplets, not just gravitational settling.

In summary, this paper bridges the gap between theoretical turbulence physics and cloud microphysics by providing high-resolution DNS data, correcting existing models for bidisperse interactions, and offering a robust new parameterization that highlights the critical role of turbulent intermittency in rain formation.