Overview

The Wiener process, also known as standard Brownian motion, is a fundamental continuous-time stochastic process in probability theory. It models the random motion of particles suspended in a fluid (physical Brownian motion) and serves as the canonical building block for stochastic calculus and diffusion processes. A Wiener process {W_t, t ≥ 0} is defined by starting at zero (W_0 = 0), having almost surely continuous sample paths, independent increments, and normally distributed increments such that W_t − W_s ∼ N(0, t−s) for 0 ≤ s < t. Its properties include infinite total variation and non‑differentiability almost everywhere, making it a classic example of a non‑smooth but continuous process.

1 Definition and Basic Properties

1.1 Mathematical definition

A Wiener process is a real-valued stochastic process {W_t : t ≥ 0} defined on a probability space (Ω, ℱ, ℙ) satisfying:

* W_0 = 0 almost surely. * For any 0 ≤ s < t, the increment W_t – W_s is normally distributed with mean 0 and variance t – s. * For any 0 ≤ s < t, the increment W_t – W_s is independent of the σ‑algebra generated by {W_u : u ≤ s}. * The sample paths t ↦ W_t(ω) are continuous for almost every ω.

1.2 Characteristic properties

1.2.1 Gaussian increments

All increments are Gaussian random variables. More generally, any finite linear combination of increments is jointly normally distributed, making the Wiener process a Gaussian process.

1.2.2 Independence of increments

For any partition 0 = t_0 < t_1 < … < t_n, the increments W_{t_k} – W_{t_{k-1}} are mutually independent. This property extends to any collection of non‑overlapping time intervals.

1.2.3 Continuous sample paths

The process admits a modification for which all sample paths are continuous functions of time. This continuity is part of the standard definition and is essential for applications in stochastic calculus.

1.3 Markov property and martingale property

The Wiener process is a Markov process: for any t ≥ 0 and any Borel set B, ℙ(W_{t+s} ∈ Bℱ_t) = ℙ(W_{t+s} ∈ BW_t) almost surely, where ℱ_t is the natural filtration. It is also a martingale: for s < t, 𝔼[W_tℱ_s] = W_s. Moreover, the process W_t² – t is a martingale.

2 Constructions of the Wiener Process

2.1 Donsker’s theorem (random walk scaling)

Donsker’s theorem states that a scaled simple symmetric random walk converges in distribution to the Wiener process. More precisely, define S_n = Σ_{i=1}^n X_i with ℙ(X_i = ±1) = ½, and define the piecewise‑linear interpolation S^{(n)}(t) = (1/√n) S_{⌊nt⌋} + (nt–⌊nt⌋) X_{⌊nt⌋+1}/√n. Then S^{(n)} converges weakly in the space of continuous functions to a Wiener process. This provides a fundamental connection between discrete and continuous stochastic models.

2.2 Lévy–Ciesielski construction (Haar wavelet basis)

This constructive method uses the Haar orthonormal basis of L²[0,1]. Let (φ_n) be a sequence of functions derived from the Haar wavelets, and let (ξ_n) be independent standard normal variables. Then define W_t = Σ_{n=0}^∞ ξ_n ∫_0^t φ_n(s) ds. The series converges uniformly on [0,1] almost surely, yielding a continuous process with the desired covariance structure.

2.3 Kolmogorov extension theorem and consistency

The Kolmogorov extension theorem guarantees the existence of a stochastic process with prescribed finite‑dimensional distributions. For the Wiener process, specifying that increments are independent Gaussian with variance t–s defines a consistent family of distributions on ℝ^{[0,∞)}. The theorem provides a measure on the product space, though additional steps are required to obtain continuous sample paths (often via a modification or by restricting to a suitable subspace).

3 Path Properties

3.1 Non‑differentiability

With probability 1, a Wiener path is nowhere differentiable. This follows from the law of the iterated logarithm: lim sup_{t→0}W_t/ √(2t log log(1/t)) = 1 almost surely, implying that the difference quotient does not converge for any t. Consequently, the sample path is of unbounded total variation on every interval.

3.2 Quadratic variation

The quadratic variation of the Wiener process over [0,T] is T. That is, for any sequence of partitions with mesh tending to zero, Σ_{k} (W_{t_k} – W_{t_{k-1}})^2 converges in probability to T. This property is fundamental for defining the Itô integral: because the process has finite quadratic variation, it can serve as an integrator.

3.3 Local time and the reflection principle

The reflection principle states that for a Wiener process and a barrier a &gt; 0, the process reflected after hitting a has the same distribution as the original process up to the hitting time τ_a. Formally, ℙ(τ_a ≤ t, W_t ≤ x) = ℙ(W_t ≥ 2a – x) for x ≤ a. This yields the distribution of the maximum M_t = max_{0 ≤ s ≤ t} W_s: ℙ(M_t ≥ a) = 2ℙ(W_t ≥ a) = 2(1 – Φ(a/√t)), where Φ is the standard normal CDF. The local time L(t,a) measures the amount of time the Wiener process spends near level a, and satisfies Tanaka’s formula.

3.3.1 Skorokhod embedding

The Skorokhod embedding theorem states that any centred probability distribution with finite second moment can be represented as the distribution of W_τ for some stopping time τ. This provides a powerful link between the Wiener process and general random walks, and is used in the study of limit theorems and stochastic integrals.

4.1 Brownian bridge

A Brownian bridge is a Wiener process conditioned to end at 0 at time T. It can be constructed as B_t = W_t – (t/T) W_T for t ∈ [0,T]. Its covariance is min(s,t) – st/T, and it is a Gaussian process with almost sure continuity. The Brownian bridge appears in Kolmogorov–Smirnov statistics and nonparametric hypothesis testing.

4.2 Geometric Brownian motion

Geometric Brownian motion is defined by S_t = S_0 exp((μ – σ²/2)t + σ W_t). It is a basic model for asset prices in mathematical finance. The process is always positive and its logarithm follows a Brownian motion with drift.

4.3 Ornstein–Uhlenbeck process

The Ornstein–Uhlenbeck process is a mean‑reverting diffusion given by the stochastic differential equation dX_t = θ(μ – X_t) dt + σ dW_t. It is stationary and Gaussian, with exponential autocorrelation.

4.3.1 Definition as a transformation of Wiener process

The Ornstein–Uhlenbeck process can be expressed as a time‑changed and scaled Wiener process: X_t = μ + σ e^{-θ t} W_{e^{2θ t} / (2θ)} (for a suitable initial condition). Alternatively, it is the unique solution of the Langevin equation.

5 Stochastic Calculus with the Wiener Process

5.1 Itô integral

The Itô integral extends the Riemann–Stieltjes integral to integrands driven by the Wiener process. For a predictable process f(t,ω) such that 𝔼[∫_0^T f² ds] &lt; ∞, the Itô integral ∫_0^T f dW is defined as the limit in probability of sums Σ f(t_{k}) (W_{t_{k+1}} – W_{t_k}). The integral is a martingale and satisfies the Itô isometry: 𝔼[(∫ f dW)²] = 𝔼[∫ f² ds].

5.1.1 Definition and properties

The construction proceeds through step functions and then extends by completeness. Key properties include: linearity, the martingale property (if f is adapted and integrable), zero expectation, and the Itô isometry. The integral can also be defined for a wider class of integrands via localisation.

5.2 Itô’s lemma

Itô’s lemma is the chain rule of stochastic calculus. For a twice‑differentiable function f(t,x) and a Wiener process W_t, it states:

df(t, W_t) = f_t dt + f_x dW_t + (1/2) f_{xx} dt.

More generally, for an Itô process dX_t = μ dt + σ dW_t, Itô’s lemma gives:

df(t, X_t) = (f_t + μ f_x + (1/2) σ² f_{xx}) dt + σ f_x dW_t.

This formula is central to solving stochastic differential equations and to mathematical finance.

5.3 Stratonovich integral

The Stratonovich integral is an alternative definition that follows the midpoint rule: ∫ f ∘ dW = lim Σ f((t_k + t_{k+1})/2) (W_{t_{k+1}} – W_{t_k}). It satisfies the ordinary chain rule (no extra drift term) and is often used in physics and engineering because it preserves the standard rules of calculus. The Stratonovich integral is related to the Itô integral through a correction term: ∫ f ∘ dW = ∫ f dW + (1/2) ∫ (∂f/∂t) dt (when f is independent of time, the correction is half the first variation in the sense of Wong–Zakai).

6 Applications

6.1 Finance: Black–Scholes model

In the Black–Scholes model, the price of a non‑dividend paying stock evolves as geometric Brownian motion: dS_t = μ S_t dt + σ S_t dW_t. The Wiener process models the unpredictable component of returns. Using Itô’s lemma and a hedging argument, the Black–Scholes partial differential equation is derived, leading to an explicit formula for European option prices.

6.2 Physics: diffusion and Langevin equation

The Wiener process models the random displacement of a Brownian particle. The Langevin equation m dV_t = –γ V_t dt + √(2γ k_B T) dW_t describes the velocity of a particle in a fluid, where the noise term represents random collisions. The position process then follows an integrated Ornstein–Uhlenbeck process, which converges to Wiener scaling in the overdamped limit.

6.3 Filtering and control theory

In filtering theory, the Wiener process models the system noise and measurement noise in continuous‑time state‑space models. The Kalman–Bucy filter, which provides optimal recursive estimates, relies on the solution of stochastic differential equations driven by Wiener processes. In stochastic control, the Wiener process is used to model uncertainties, and the optimal control problem is solved via dynamic programming or the Hamilton–Jacobi–Bellman equation.

7 Extensions

7.1 Multidimensional Wiener process

A d‑dimensional Wiener process is a vector W_t = (W_t^{(1)}, …, W_t^{(d)}) where each component is an independent one‑dimensional Wiener process. Its increments are multivariate normal with covariance matrix (t–s) I_d. This process serves as the noise source for multidimensional stochastic differential equations and for subjects such as partial differential equations and stochastic geometry.

7.2 Fractional Brownian motion

Fractional Brownian motion (fBm) is a generalization of the Wiener process with Hurst parameter H ∈ (0,1). Its covariance is given by 𝔼[B_t^H B_s^H] = (t^{2H} + s^{2H} –t–s^{2H})/2. For H=½, fBm reduces to standard Wiener motion; for H>½, increments are positively correlated (persistence); for H<½, they are negatively correlated (anti‑persistence). FBm has continuous but non‑differentiable paths; however, it is not a semimartingale (except H=½) and requires a different stochastic calculus.

7.3 Wiener process with drift

A Wiener process with drift is defined by X_t = μ t + σ W_t, where μ is the drift parameter and σ &gt; 0 is the diffusion coefficient. The process is still a Gaussian process with independent increments, but each increment has mean μ(t–s) instead of zero. This process is used to model random motion with a systematic component, such as a particle in a constant force field or stock prices under a probability measure other than the risk‑neutral one.