Skip to main content
1 of 2
Pierre PC
  • 3.7k
  • 10
  • 24

(I'm changing your notations a bit, because I want to call $B$ a Brownian bridge.)

Fix $0<s\leq1$, and let $W$ be a standard Brownian motion indexed by $[0,1]$. I call the (law of the) random variable $u\in[0,s]\mapsto B_u-\frac usB_s$ a Brownian bridge of size $s$. It is but the Brownian motion modified to be $0$ at $s$, by subtracting a linear function.

I claim that the "superposition" of a Brownian bridge $B$ of size $s$ and a normal variable $N$ of variance $1$ is a Brownian motion on $[0,1]$: here, superposition means the process $$u\mapsto B_u+uN\text.$$ One can prove this fact by noting that such a process is Gaussian, and has the desired finite dimensional covariances.

Now for what you're asking: the Brownian motion $W$ can be decomposed in for independent variables (see image below):

  • two bridges $B$ and $B'$ of sizes $s$ and $1-s$;
  • two Gaussian variables $N$ and $N'$ of variances $s$ and $1-s$.

(To prove this, I would go the other way around and show that given such random independent variables, the "superposition" is Gaussian and has the required covariance.)

Bridge decomposition

Once this is done, we see that $$\mathcal{G}_t = \sigma(N,N',B)$$ and so $$ \mathbb{E}[W_t-W_s|\mathcal{G}_s] = \mathbb{E}\left[B'_{t_s} + \frac{t-s}{1-s}N'\middle|\mathcal{G}_s\right] = \frac{t-s}{1-s}N' = \frac{t-s}{1-s}(B_{1-s}-B_s)\text. $$

Pierre PC
  • 3.7k
  • 10
  • 24