Diffusion: The Science of Particle Movement and Stochastic Processes

At its most fundamental level, diffusion is the net movement of entities—such as atoms, ions, molecules, or energy—typically from a region of higher concentration to a region of lower concentration. This process is driven by a gradient in chemical potential or Gibbs free energy (the maximum amount of non-expansion work that can be extracted from a closed system). While most diffusion occurs "downhill," certain phenomena like spinodal decomposition allow for "uphill" diffusion, where substances move toward areas of higher concentration.

Because diffusion relies on the inherent randomness of the moving entities, it is classified as a stochastic process. This randomness makes diffusion a powerful mathematical tool used not only in physics and chemistry but also in finance, marketing, information theory, and neural networks to model real-life unpredictable scenarios.

Some particles are dissolved in a glass of water. At first, the particles are all near one top corner of the glass. If the particles randomly move around ("diffuse") in the water, they eventually become distributed randomly and uniformly from an area of high concentration to an area of low, and organized (diffusion continues, but with no net flux).

Key Facts

Driving Force: Primarily driven by concentration gradients and chemical potential.
Nature: A stochastic (random) process based on the random walk of particles.
Scope: Occurs in gases, liquids, and solids, as well as in abstract systems like social networks and data science.
Key Laws: Governed largely by Fick's laws, which relate diffusion flux to concentration gradients.
Modern Application: Used in AI via diffusion-based generative models to create new data samples.

The Mechanics of Diffusion

Atomistic Perspective and Brownian Motion

From an atomistic point of view, diffusion is the result of a random walk. In molecular diffusion, particles in a gas, liquid, or solid are self-propelled by their own kinetic energy. This was famously observed in 1827 by Robert Brown, who noticed that small particles suspended in a liquid exhibited a rapid, irregular motion now known as Brownian movement. The theoretical foundation for this motion and the atomistic nature of diffusion was later developed by Albert Einstein.

Diffusion from a microscopic and b macroscopic point of view. Initially, there are solute molecules on the left side of a barrier (purple line) and none on the right. The barrier is removed, and the solute diffuses to fill the whole container. Top: A single molecule moves around randomly. Middle: With more molecules, there is a statistical trend that the solute fills the container more and more uniformly. Bottom: With an enormous number of solute molecules, all randomness is gone: The solute appears to move smoothly and deterministically from high-concentration areas to low-concentration areas. There is no microscopic force pushing molecules rightward, but there appears to be one in the bottom panel. This apparent force is called an entropic force.

Macroscopic vs. Microscopic Views

While a single molecule moves randomly without a specific destination, the collective behavior of billions of molecules creates a statistical trend. This results in a smooth, deterministic flow from high-concentration areas to low-concentration areas. This apparent force, emerging from randomness, is referred to as an entropic force.

Three-dimensional rendering of diffusion of purple dye in water.

Diffusion Across Disciplines

Physics and Chemistry

In physics, diffusion is studied through various lenses, from the mean free path (the average distance a particle travels between collisions) introduced by Rudolf Clausius, to the Boltzmann equation. In solids, diffusion occurs through local defects such as vacancies or interstitial atoms, a concept elaborated by Yakov Frenkel in 1926.

Diffusion furnaces used for thermal oxidation

Biology and Medicine

Diffusion is critical for biological survival. A primary example is dialysis, where a semipermeable membrane allows the diffusion of solutes to clean the blood.

Schematic of semipermeable membrane during hemodialysis, where blood is red, dialysing fluid is blue, and the membrane is yellow.

Artificial Intelligence

In machine learning, diffusion models (or score-based generative models) use principles from non-equilibrium thermodynamics. These models learn a diffusion process for a dataset to generate new, similar elements by sampling from complex probability distributions.

Mathematical Models of Diffusion

Fick's Laws

Adolf Fick proposed the fundamental laws of diffusion in 1855, drawing analogies to heat conduction (Fourier's law) and electric current (Ohm's law). Fick's First Law states that the diffusion flux is proportional to the negative gradient of spatial concentration. Fick's Second Law predicts how diffusion causes the concentration to change over time.

Diffusion in Specialized Media

Porous Media: Diffusion is influenced by porosity and advective molar flux, often described by the advection-dispersion equation.
Membranes: The Teorell formula describes the diffusion of ions through membranes, accounting for both concentration gradients and electrical forces.
Semiconductors: Light can generate carriers (electrons and holes) that diffuse toward opposite ends of a semiconductor; electrons typically have a higher diffusion constant than holes.

As carriers are generated (green:electrons and purple:holes) due to light shining at the center of an intrinsic semiconductor, they diffuse towards two ends. Electrons have higher diffusion constant than holes leading to fewer excess electrons at the center as compared to holes.

Surface and Solid State Diffusion

In solids, particles may oscillate near equilibrium positions before making "jumps" to the nearest free space. This jump model is essential for understanding how materials behave at the atomic level.

Diffusion in the monolayer: oscillations near temporary equilibrium positions and jumps to the nearest free places.

Random collisions of particles in a gas.

Summary of Diffusion Types

Common Types of Diffusion and Their Characteristics
Type	Description	Example/Context
Molecular	Movement from dense to less dense areas	Dye spreading in water
Atomic	Movement of atoms within a solid lattice	Zinc penetrating copper
Facilitated	Protein-assisted movement across membranes	Biological cell transport
Thermodiffusion	Diffusion caused by temperature gradients	Gas mixtures in varying heat
Electronic	Movement of electrons creating current	Semiconductor devices
Trans-cultural	Spread of cultural traits	Sociology/Geography

Frequently Asked Questions

What is the difference between diffusion and bulk flow?

Diffusion is the random, stochastic movement of individual particles driven by a concentration gradient. Bulk flow (or advection) is the mass movement of a fluid and its dissolved solutes together, usually driven by a pressure gradient.

Can diffusion happen against a concentration gradient?

Yes. While standard diffusion moves from high to low concentration, "uphill" diffusion can occur in specific processes such as spinodal decomposition, driven by changes in chemical potential.

How does temperature affect the diffusion coefficient in gases?

According to kinetic theory, the diffusion coefficient in gases generally increases as temperature rises and decreases as pressure increases.

What is a random walk in the context of diffusion?

A random walk is a mathematical path consisting of a succession of random steps. In diffusion, it describes the trajectory of a single particle as it collides randomly with others, eventually leading to the overall spreading of the substance.

How are diffusion models used in AI?

Diffusion models in AI learn to reverse a process of adding noise to data. By learning to "denoise" an image or signal, the model can generate entirely new, high-quality data that follows the distribution of the original training set.