Brownian motion works as a continuous-time stochastic process with continuous paths, independent increments and Gaussian changes whose variance grows in proportion to elapsed time. Also called a Wiener process in its mathematical form, Brownian motion is the canonical limit of many small random steps and the central model behind diffusion, stochastic calculus, heat-flow connections, financial mathematics, filtering and random dynamical systems. Its paths are continuous yet almost surely nowhere classically differentiable, making Brownian motion the point where ordinary calculus stops being enough and stochastic calculus begins.
Brownian motion explains how smooth-looking large-scale diffusion can emerge from microscopic randomness. Variance grows linearly with time while typical displacement grows with the square root of time. Scaled random walks converge toward Brownian motion under functional central-limit results, so many different small-step mechanisms can share the same large-scale stochastic limit.
The difficult idea is that continuity does not imply smoothness. Brownian paths have no jumps, but zooming in never reveals a stable tangent. Their quadratic variation is nonzero, which changes the chain rule and creates the Ito correction in stochastic calculus. That roughness is not a nuisance added to the model; it is the defining geometry that makes Brownian motion useful.
1. Brownian Motion Is a Process, Not One Normal Distribution
A Brownian motion Wₜ assigns a random value to every nonnegative time. Each fixed-time value is normal, but the object of interest is the complete random path and its joint behaviour through time.
2. Standard Brownian Motion Starts at Zero
W₀=0 almost surely. Shifted or drifted versions can start elsewhere, but zero is the canonical origin.
3. Increments Are Gaussian
For s<t, Wₜ−Wₛ is normally distributed with mean zero and variance t−s. Longer intervals permit larger typical changes.
4. Disjoint Increments Are Independent
Changes over non-overlapping intervals are independent in standard Brownian motion. This gives the process its strong Markov and martingale structure.
5. Paths Are Continuous but Rough
There are no jumps, yet Brownian paths are almost surely nowhere classically differentiable. Difference quotients fluctuate rather than settling to a tangent.
6. Variance Grows Linearly With Time
Var(Wₜ)=t, so standard deviation grows as √t. This square-root scaling is the signature of ordinary diffusion.
7. Covariance Is min(s,t)
Earlier randomness is shared by Wₛ and Wₜ; the increment after the earlier time is independent. That produces Cov(Wₛ,Wₜ)=min(s,t).
8. Random Walks Converge Toward Brownian Motion
Many independent mean-zero finite-variance steps, centred and rescaled in space and time, converge in path distribution toward Brownian motion. This functional CLT connects discrete randomness to continuous diffusion.
9. Self-Similarity Preserves Statistical Shape Across Scales
For c>0, the process W_{ct} has the same finite-dimensional distribution as √c Wₜ. Zooming time changes amplitude by the square-root rule.
10. Brownian Motion Is Markov
Given the present value Wₜ, future increments are independent of the earlier path. See How Markov Chains Work for the discrete-state mechanism behind Markov memory compression.
11. Brownian Motion Is a Martingale
Under its natural information filtration, the conditional expectation of a future Brownian value equals the current value. Standard Brownian motion has no predictable drift.
12. Quadratic Variation Equals Elapsed Time
Sum squared increments across a refining partition. For Brownian motion the sum converges to t over [0,t]. Smooth paths would have zero quadratic variation.
13. Nonzero Quadratic Variation Changes Calculus
Because (dW)² behaves like dt at leading order, second derivatives survive in Ito’s formula. Ordinary first-order chain-rule intuition is incomplete.
14. Ito Integrals Use Non-Anticipating Integrands
Stochastic integration defines integrals against Brownian motion using information available up to each time. The resulting integral is itself random and has an isometry linking its second moment to an ordinary time integral.
15. Ito’s Formula Is the Stochastic Chain Rule
For a smooth function f(t,Wₜ), Ito’s formula contains ordinary time and first-derivative terms plus a one-half second-derivative correction caused by quadratic variation.
16. The Heat Equation Is Connected to Brownian Expectations
Solutions to diffusion equations can be represented through expectations over Brownian paths under appropriate conditions. Probability and partial differential equations become two descriptions of the same diffusion mechanism.
17. First-Passage Times Ask When a Boundary Is Hit
When does Brownian motion first reach a level? First-passage distributions support barrier problems, reliability, sequential analysis and finance.
18. Reflection Principles Convert Path Events Into Distribution Calculations
Symmetry of Brownian increments allows certain maximum and hitting probabilities to be computed by reflecting paths after boundary crossings.
19. Brownian Bridges Condition the Endpoint
A Brownian bridge behaves like Brownian motion constrained to return to a specified endpoint. It appears in empirical-process theory, goodness-of-fit tests and conditional path simulation.
20. Drift Creates Brownian Motion With Direction
Xₜ=μt+σWₜ adds deterministic drift and volatility scale. Mean moves linearly while random dispersion grows with √t.
21. Geometric Brownian Motion Keeps Values Positive
A multiplicative stochastic differential equation produces lognormal marginal distributions and positive paths. It is historically central in mathematical finance, though real returns can violate its constant-volatility and light-tail assumptions.
22. Brownian Motion Models Diffusion, Not Every Random Path
Jump processes, long-memory processes and anomalous diffusion require other stochastic models. Brownian motion is a canonical baseline, not a universal law of randomness.
23. Physical Brownian Motion Motivated the Mathematics
Microscopic molecular collisions produce irregular motion of suspended particles. The mathematical Wiener process idealises the large-scale stochastic limit rather than tracking every molecular collision.
24. Diffusion Coefficients Change the Scale
Physical diffusion often uses variance proportional to 2Dt in one dimension, where D is a diffusion coefficient. Standard Brownian motion corresponds to a particular normalisation.
25. Multidimensional Brownian Motion Has Vector Increments
Independent coordinate Brownian motions create isotropic motion; covariance structures can create correlated components. Geometry then matters for hitting sets and stochastic differential equations.
26. Brownian Motion Drives Stochastic Differential Equations
An SDE such as dXₜ=b(Xₜ,t)dt+σ(Xₜ,t)dWₜ combines systematic drift with random diffusion. Existence, uniqueness and numerical approximation depend on properties of b and σ.
27. Euler–Maruyama Approximates SDE Paths
Discretise time, approximate drift over each step and add Gaussian increments scaled by √Δt. Numerical stochastic simulation has both time-discretisation error and Monte Carlo error.
28. Brownian Motion Appears in Filtering
Continuous-time state-space models use Brownian noise to represent uncertain system evolution and measurement processes. Filtering updates hidden-state distributions as noisy observations arrive.
29. Finance Uses Risk-Neutral Brownian Models Carefully
Derivative pricing can transform drift under a change of probability measure so discounted asset prices become martingales under model conditions. This is a pricing construction, not a claim that real-world expected returns vanish.
30. Girsanov’s Theorem Changes Drift by Changing Measure
Under suitable conditions, a change of probability measure can transform Brownian motion with one drift structure into Brownian motion under another measure. The theorem formalises how probability weights alter path likelihoods.
31. Brownian Motion Has Infinite Total Variation
Its path wiggles accumulate without a finite ordinary variation budget. This blocks classical Riemann–Stieltjes integration against Brownian paths in the usual bounded-variation framework.
32. The Law of the Iterated Logarithm Describes Extreme Long-Run Fluctuation
Brownian motion grows on a scale slightly larger than √t along extreme subsequences. Fine path results show that diffusion has rich structure beyond its fixed-time normal distributions.
33. What Brownian Motion Preserves
It preserves continuous random fluctuation, Gaussian increment structure, square-root scaling, Markov memorylessness and a universal diffusion limit.
34. What Brownian Motion Discards
It discards jumps, long-range memory, finite-speed microscopic mechanics and many forms of heavy-tailed or anomalous movement.
35. Hostile Test: Smooth Calculus Applied to a Brownian Path
Treating dW like an ordinary infinitesimal and dropping second-order terms produces the wrong chain rule because Brownian quadratic variation survives.
36. Hostile Test: Geometric Brownian Motion Treated as a Complete Market Model
Constant volatility, continuous paths and lognormal prices can miss volatility clustering, jumps and heavy tails. Mathematical tractability is not empirical completeness.
37. Practical Workflow
Define the stochastic target, decide whether continuous Gaussian diffusion is defensible, specify drift and diffusion scale, identify path-dependent quantities, choose analytic or numerical methods, validate against increment and tail behaviour, and separate model error from Monte Carlo discretisation error.
38. Canonical Boundary
Stochastic Processes owns the wider family. This article owns Brownian motion and Wiener-process mechanics. Monte Carlo Simulation owns sampled computation broadly.
Final Thought
Brownian motion is the mathematical discovery that a path can be continuous everywhere and smooth nowhere—and that this roughness can still obey extraordinarily precise probability laws.