Loading 3D assets...
...
Drag to orbit - wheel / pinch to zoom - particles are drawn about 2x larger than to scale

Nanoparticle Coagulation Lab: Brownian Collisions, Smoluchowski Kinetics and Fractal Clusters

Nanoparticles diffuse, collide and stick. Watch the cluster count fall, compare it with Smoluchowski theory, and change the liquid, the concentration or the cluster geometry.
Dimensionless time θ-
Real time-
Clusters left-
Mean cluster size-
Monomers left (MC | theory)-
Time scale τ = 2/(K₀ n₀)-
Clusters left N/N₀ vs θ: Monte Carlo (blue), Smoluchowski ODE for the chosen df (yellow: steady kernel, green: with the diffusion start-up transient), constant-kernel exact solution (grey).
Share of clusters of size k now (dots: simulation, line: theory at the same θ).
Independent numerical checks
Each line compares two separate calculations; a FAIL would show here.
The science

What happens. Particles in a liquid move by Brownian motion, collide and stick. A pair of clusters with radii ri, rj and diffusion constants Di, Dj meets at the Smoluchowski rate Kij = 4π(Di + Dj)(ri + rj). With Stokes–Einstein, D = kBT / (6πηr), equal spheres give K0 = 8kBT / (3η), independent of size.

Population balance. The number of k-clusters obeys dNk/dt = ½ Σi+j=k Kij NiNj − Nk Σj Kkj Nj. We integrate it with RK4 in the dimensionless time θ = K0 n0 t (n0 = initial number density). For a constant kernel there is an exact solution: N(t)/N0 = 1/(1 + θ/2) and Nk/N0 = (θ/2)k−1/(1 + θ/2)k+1. That is how the integrator is tested.

The particle simulation. The 3D animation is a Brownian-dynamics Monte Carlo of coalescing spheres in a periodic box (step = 0.1 monomer radii, new cluster at the mass-weighted centre, radius r = n1/df, D = 1/r). Compact droplets have df = 3; loose fractal aggregates of nanoparticles typically have df of about 1.8 to 2.1 (a typical range), and because their hydrodynamic radius grows faster they collide faster. For simplicity every touching pair sticks (stability ratio W = 1, no electrostatic barrier), and the contact radius is taken equal to the hydrodynamic radius.

Why simulation and theory differ a little. Smoluchowski’s steady-state rate ignores the start-up transient of the diffusion flux (extra collisions of neighbours that start close). The green curve adds it, Kij(t) = Kij(1 + (ri+rj)/√(π(Di+Dj)t)), and is the one the Monte Carlo is checked against; the simulated count still scatters because the box holds only 60 to 200 particles (the check tolerance is 10 percent plus 1.5/√N). Spheres are drawn about 2.2 times larger than to scale so that you can see them.

Real units. From the sliders: n0 = φ / (4πr1³/3) and τ = 2/(K0 n0) = 3η / (4 kBT n0) is the time for the cluster count to halve in the constant-kernel limit. Water-like liquids and 25 nm particles at 0.2 % volume give a few milliseconds.

Checks. (1) The RK4 integrator against the exact constant-kernel solution at θ = 2. (2) Mass conservation in the Monte Carlo (the sum of cluster sizes stays exactly the number of primary particles). (3) The Monte Carlo cluster count against the transient-corrected ODE for the chosen df. (4) The kernel in its two algebraic forms.

Not modelled. Electrostatic or steric stabilisation (DLVO), sticking probability below 1, cluster restructuring, hydrodynamic interactions, sedimentation, breakup. Real colloids are usually slower than this diffusion-limited upper bound by a stability ratio W.

The cell, stage, objective and particle meshes were modelled and baked in Blender; the stochastic dynamics and rendering run in Three.js r160.

Made with 🟠 Blender · Three.js