Skorokhod's embedding theorem

Lore & Background

The theorem exists in two forms. The first embedding theorem states that for any real-valued random variable X with expected value 0 and finite variance, there exists a stopping time τ (with respect to the natural filtration of a canonical Wiener process W) such that Wτ has the same distribution as X, with E[τ] = E[X²] and E[τ²] ≤ 4E[X⁴]. The second embedding theorem concerns a sequence of independent and identically distributed random variables X₁, X₂, ... each with expected value 0 and finite variance, and their partial sums Sₙ = X₁ + ... + Xₙ. It asserts the existence of a sequence of stopping times τ₁ ≤ τ₂ ≤ ... such that the W_{τₙ} have the same joint distributions as the partial sums Sₙ, and the increments τ₁, τ₂ − τ₁, τ₃ − τ₂, ... are independent and identically distributed, with E[τₙ − τ_{n−1}] = E[X₁²] and E[(τₙ − τ_{n−1})²] ≤ 4E[X₁⁴].

Reader's Guide

Skorokhod's embedding theorem is a foundational tool in probability theory, linking arbitrary distributions and random walks to Brownian motion. Its first form provides a way to represent any mean-zero, finite-variance random variable as the position of a Wiener process at a stopping time, with moment constraints on the stopping time. The second form extends this to sequences of i.i.d. random variables, embedding the entire partial-sum process into a single Wiener process via a sequence of stopping times with independent and identically distributed increments. This allows the study of random walks through the lens of Brownian motion, facilitating limit theorems and other analyses. The theorem is named for A. V. Skorokhod, a Ukrainian mathematician, and is referenced in standard texts such as Billingsley's 'Probability and Measure'.

Did You Know?

• Skorokhod's embedding theorem is either or both of two theorems.
• The first theorem requires the random variable to have expected value 0 and finite variance.

The Central Idea

Skorokhod's embedding theorem occupies a distinctive niche at the intersection of probability theory and stochastic processes. At its heart, the result provides a powerful bridge: it shows that a broad class of random variables, or collections thereof, can be reinterpreted as values of a canonical Wiener process (Brownian motion) sampled at carefully chosen stopping times. Rather than treating a given random variable as an isolated object, the theorem embeds it into the continuous, pathwise structure of Brownian motion. This reframing is not merely cosmetic; it allows probabilists to import the rich machinery of martingale theory, optional stopping, and path properties of the Wiener process into settings that originally involved only discrete random variables. The theorem is stated in two forms, and a given application may invoke either one or both. Both results carry the name of the Ukrainian mathematician A. V. Skorokhod, whose work on stochastic processes made this connection between discrete randomness and continuous-time diffusion both natural and rigorous.

The Single-Variable Result

The first embedding theorem addresses the simplest nontrivial case. Suppose X is a real-valued random variable whose expected value is zero and whose variance is finite. The theorem asserts the existence of a stopping time τ, adapted to the natural filtration of a canonical real-valued Wiener process W, such that the random variable W evaluated at τ carries exactly the same probability distribution as X. Beyond mere distributional matching, the result supplies two quantitative moment constraints on the stopping time. The first is an exact identity: the expected value of τ equals the second moment of X. The second is an upper bound: the expected value of τ squared is no greater than four times the fourth moment of X. These moment conditions are what make the embedding analytically useful, because they guarantee that the stopping time is not merely a formal device but one with controlled growth, tying the size of the waiting time directly to the spread of the original distribution.

The Sequential Generalization

The second embedding theorem extends the single-variable picture to an entire sequence. Let X₁, X₂, … be independent and identically distributed random variables, each with zero mean and finite variance, and define the partial sums Sₙ as the running total through the n-th term. The theorem guarantees the existence of a non-decreasing sequence of stopping times τ₁ ≤ τ₂ ≤ … such that the vector of Wiener-process values at those times reproduces the joint distributions of the partial sums. Crucially, the increments between successive stopping times—τ₁, τ₂ − τ₁, τ₃ − τ₂, and so on—are themselves independent and identically distributed. Each increment has expected value equal to the variance of a single summand, and its second moment is bounded above by four times the fourth moment of that summand. This independence of the inter-arrival increments mirrors the independence of the original random variables, making the embedding a faithful structural analogue rather than a mere distributional coincidence.

Attribution and Mathematical Standing

Although the theorems bear the name of A. V. Skorokhod, a Ukrainian mathematician whose broader contributions to stochastic processes are extensive, the embedding results have found a permanent home in standard probability literature. Patrick Billingsley's widely used text Probability and Measure (John Wiley & Sons, 1995) presents both results as Theorems 37.6 and 37.7, situating them within a chapter on Brownian motion and martingale methods. The fact that these theorems are numbered consecutively in a foundational reference underscores their role as a paired toolkit: the first handles a single random variable, the second handles an entire i.i.d. sequence, and together they give probabilists a flexible way to translate questions about sums of random variables into questions about Brownian motion observed at random times. This translation is particularly valuable because it converts discrete, algebraic problems into continuous, pathwise ones where the full apparatus of stochastic calculus becomes available.