Reference request for conditional and unconditional covariance of n-times integrated Brownian motion

74 Views Asked by At

I'm working through an old Diaconis paper on Bayesian numerical analysis, and am currently calculating the details behind his brief comments on using $n$-times integrated Brownian motion as a function prior for quadratures.

In short, that means working out the covariance, and then the expectation and covariance conditional on a later value of the same $n$-integrated motion, i.e. the equivalent to a Brownian bridge.

I've been working these out from scratch, mostly by heavy use of recurrence relations, and so far I've tentatively got the statistics for $n$-integrated Brownian motion $B_t^{(n)},$ $n\geq0,$ as being

$$\begin{aligned}\textrm{Cov}(B_s^{(n)}, B_t^{(n)}) &= \mathbb{E}(B_s^{(n)} B_t^{(n)}) \\&= \frac{1} {(2n+1)!} \sum_{k=0}^n (-1)^{n-k} \binom{2n+1}{k} \min(s,t)^{2n+1-k} \max(s,t)^k,\end{aligned}$$ $$\mathrm{Var}(B_t^{(n)}) = \frac{1} {(2n+1)!} \binom{2n}{n} t^{2n+1},$$ $$\mathbb{E}(B_s^{(n)} | B_t^{(n)}) = \binom{2n}{n}^{-1} \sum_{k=0}^n (-1)^{n-k} \binom{2n+1}{k} \left(\frac{s}{t}\right)^{2n+1-k} B_t^{(n)},$$ $$\textrm{e.g. } \mathbb{E}(B_s^{(1)} | B_t^{(1)}) = \frac{1} {2} \left(\frac{s}{t}\right)^2 \left(3-\frac{s}{t}\right) B_t^{(1)},$$ $$\mathbb{E}(B_r^{(n)} B_s^{(n)} | B_t^{(n)})\textrm{ pending}.$$

Feedback on whether I'm correct so far would be nice, but what I'd really like are pointers to good references on the subject, preferably books. Thanks in advance.