the Brownian bridge, sheet, and time-inversion symmetry

Brownian motion is richer than a single process: it generates a family of close relatives and a group of symmetries that are themselves powerful tools. Three central ones are the Brownian bridge (Brownian motion conditioned to return to a fixed endpoint), the Brownian sheet (a two-parameter Gaussian field generalising Brownian motion to a 'random surface'), and the scaling and time-inversion symmetries that make Brownian motion a self-similar, time-reversible object.

The Brownian bridge from 0 to 0 on [0, 1] is the process X_t = B_t - t B_1, equivalently Brownian motion conditioned on B_1 = 0; it is a centred Gaussian process with covariance Cov(X_s, X_t) = s(1 - t) for s <= t, pinned at both ends. It is the scaling limit of the empirical process (Donsker), so the Kolmogorov-Smirnov statistic sup |X_t| has the bridge's law, making it central to goodness-of-fit. The scaling (self-similarity) symmetry: for any c > 0, the process (1/sqrt(c)) B_(ct) is again a standard Brownian motion — distance scales like the square root of time, the source of every sqrt(t) in the theory. The time-inversion symmetry: the process defined by X_0 = 0 and X_t = t B_(1/t) for t > 0 is again a standard Brownian motion. Time inversion swaps the behaviour near 0 with the behaviour near infinity, which is why the law of the iterated logarithm has matching statements at t -> 0 and t -> infinity, and why local roughness and global growth are mirror images. Time reversal (B_(1 - t) - B_1 versus B) and the Markov-property symmetries round out the group. The Brownian sheet W(s, t) is the centred Gaussian field on [0, infinity)^2 with covariance E[W(s,t) W(s', t')] = min(s, s') min(t, t'); it is Brownian motion in each coordinate separately and is the canonical two-parameter (space-time) Gaussian noise integrator, the model for additive random surfaces and the driver of stochastic PDE.

Why it matters: the bridge gives exact distributions for conditioned diffusions, interpolation, and Donsker-type weak-convergence statistics; the scaling symmetry is the structural reason Brownian motion is the universal scaling limit (Donsker's invariance principle) of any finite-variance random walk; time inversion is a free theorem-generator, instantly transferring asymptotic results between the two ends of time. The honest caveats: a Brownian bridge is NOT a Markov process in the naive sense — conditioning on the endpoint introduces a pull toward it (its SDE is dX = -(X / (1 - t)) dt + dB), so it has an explosive drift as t -> 1; the Brownian sheet, despite the name, does NOT have independent increments over rectangles in any simple way and its sample-path geometry (e.g. its zero set) is genuinely two-dimensional and more delicate than the one-parameter case. Self-similarity is a statement in LAW, not a pathwise identity.