Roadmap to Diffusion Processes

Picture a single grain of pollen suspended in a drop of water. It is never still: it jitters, drifts, doubles back, tracing a path so erratic that no magnification ever smooths it out. That ceaseless wandering — continuous in time, yet jagged at every scale — is the picture to keep in mind. A diffusion process is the mathematics that makes it precise.

Such a path demands new tools because of a quiet paradox: the motion never jumps, yet it is so rough that it has no velocity at any instant. The ordinary calculus of smooth curves has nothing to grip. You cannot say where the particle will be a moment from now — but you can describe the statistics of where it tends to go, and those statistics are governed by two competing forces. One is drift, a systematic pull that may depend on where the particle sits; the other is the random noise that scatters it regardless. Reading a diffusion means feeling how these two balance at every point.

Here is the surprise: although any single trajectory is unpredictable, the long-run behavior is often beautifully orderly, settling into a fixed distribution that forgets where it started. That bridge between small-scale wildness and large-scale calm is why the same few ideas describe a stock price, a diffusing molecule, a firing neuron, and a gene drifting through a population. The concepts below are the vocabulary for making these intuitions exact. Take them in order, or jump to whichever one stands between you and what you are trying to understand — the path is rigorous, but it was laid down for a human being to walk.