The model (Witten–Sander 1981). A seed sits at the origin. Random walkers start on a sphere just outside the cluster and move by fixed-length steps in uniformly random directions (Brownian motion). A walker that comes within one particle diameter of the cluster sticks with probability p (otherwise it keeps wandering); one that strays too far is discarded and relaunched.
Why it branches. Tips screen the fjords: a walker is far more likely to touch an outer tip than to diffuse into a narrow gap (the Laplace field of the diffusing concentration is largest at tips) — the same instability as in electrodeposition, mineral dendrites, dielectric breakdown and viscous fingering.
Fractal dimension. A mass-radius law N ∝ RgD defines D, with Rg² = (1/N)Σ|r−rcm|². The panel on the right fits D from the log–log history of this very cluster. Literature values for off-lattice DLA: D ≈ 1.71 in 2D and D ≈ 2.5 in 3D (Meakin 1983; Tolman & Meakin 1989). A hard-sphere, space-filling ball would be D = 3, a line D = 1. Fits on clusters of only a few thousand particles carry a few-percent uncertainty and drift with size.
Honest simplifications. Far from the cluster the walker takes longer jumps equal to its distance to the cluster's bounding sphere (walk-on-spheres, a speed-up that does not change where walkers land); a walker that hits while not sticking stays put and tries again. The number of walkers only changes how fast the cluster grows, not its shape. Particle size, step and time are in arbitrary units; no hydrodynamics, no particle charge, no re-dissolution. Sticking probability below 1 slowly crosses over to thicker, more compact growth at large sizes.
Independent checks (run on load, seeded):
| running… |