We present a simulation scheme for simulating Brownian bridges on complete and connected Lie groups. We show how this simulation scheme leads to absolute continuity of the Brownian bridge measure with respect to the guided process measure. This result generalizes the Euclidean result of Delyon and Hu to Lie groups. We present numerical results of the guided process in the Lie group \SO(3). In particular, we apply importance sampling to estimate the metric on \SO(3) using an iterative maximum likelihood method.
Let X be the solution of (?). The SDE (?) yields a strong solution on [0,T) and satisfies \lim_{t\uparrow T} Y_t = v almost surely. Moreover, the conditional expectation of X given X_T=v is
for every \mathcal{F}_t-measurable non-negative function f on G, t<T, where \varphi_t is given in (?).
The solution of SDE (?) satisfies \lim_{t \rightarrow T} Y_t =v almost surely.
Let \{D_n\}_{n=1}^{\infty} be an exhaustion of G, that is, the sequence consists of open, relatively compact subsets of M such that \bar{D}_n \subseteq D_{n+1} and G = \bigcup_{n=1}^{\infty} D_n. Furthermore, let \tau_{D_n} denote the first exit time of Y from D_n, then from (?) we have that the sequence \bigl(\mathbb{E} [1_{\{t < \tau_{D_n}\}} r_v^2(Y_t)]\bigr)_{n=1} ^{\infty} is non-decreasing and bounded, hence from the monotone convergence theorem, it has a limit which is bounded by the right-hand side of (?). Applying Jensen’s inequality to the left-hand side of (?)
Since obviously \mathbb{E} [ r_v(Y_T)] = r_v(Y_T)\mathbb{Q}(r_v(Y_T) \neq 0), by Fatou’s lemma \mathbb{E}[r_v(Y_T)] \leq \liminf_{t\rightarrow T} \mathbb{E}[r(Y_t)] = 0, we conclude that r(Y_t) \rightarrow 0, \mathbb{Q}-almost surely.
Let 0 < t_1 <t_2< \dots < t_N <T and h be a continuous bounded function function on G^N. With \psi_t as in (?), then
The proof is similar to that of [delyon_simulation_2006, Lemma 7]. Let (U,\phi) be a normal chart centered at v \in G. First, since the cut locus of any complete connected manifold has (volume) measure zero, we can integrate indifferently in any normal chart. For any t \in (t_N,T) we have
where d\operatorname{Vol}(z)= \sqrt{\det(A(z))}dz denotes the volume measure on G, dz the Lebesgue measure, and A the metric tensor. Moreover,
and of course \Phi_1(t,z) = p(t,e,z). Using the normal chart and applying the change of variable x=(T-t)^{1/2}y we get
The conclusion follows from Bayes’ formula.
With \varphi_t as defined above then \varphi_t \overset{L_1}{\rightarrow} \varphi_T.
Note that for each t \in [0,T) we have \mathbb{E}^{\mathbb{Q}}[\varphi_t]<\infty as well as \varphi_t \overset{}{\rightarrow} \varphi_T almost surely by Lemma 1. The result then follows from the uniform integrability of \left\{\varphi_t \colon t \in [0,T)\right\}, which can be found in Appendix C.2 in [thompson2015submanifold].
The text on this page is from arXiv:2106.03431v1 by Mathias Højgaard Jensen, Sarang Joshi and Stefan Sommer, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2106/2106.03431.jsonl.