Boundary crossing probability for Brownian motion (Q2731158)

Let \(W\) be a centered standard Brownian motion in \([0,T]\), \(a\) and \(b\) two piecewise linear functions on \([0,T]\) with common nodes \((t_1,\dots, t_n)\). The probability of the event \(\{\omega: a(t)\leq W_t(\omega)\leq b(t)\}\) is given by \(E[g(W(t_1),\dots, W(t_n))]\) where the explicit form of the function \(g\) above can be derived. When \(a\) and \(b\) are general nonlinear functions in \(C^2[0, T]\), \(a(0)< 0< b(0)\), some technical lemmata lead to the following result: Let \(a_n\) and \(b_n\) be piecewise linear functions approximating \(a\) and \(b\) such that there exist differentiable functions \(\varphi_a\) and \(\varphi_b\), \(\varepsilon> 0\), satisfying \(|a''/\varphi_a|^{1/2}\), \(|b''/\varphi_b|^{1/2}\) are integrable on \([0,T]\) and: NEWLINE\[NEWLINE|a(t)- a_n(t)|\leq \varepsilon\varphi_a(t),\quad|b(t)- b_n(t)|\leq \varepsilon\varphi_b(t),NEWLINE\]NEWLINE and some other technical hypotheses, then the difference \(\Delta_n\) between the probabilities of the two events \(\{\omega: a(t)\leq W_t(\omega)\leq b(t)\}\) and \(\{\omega: a_n(t)\leq W_t(\omega)\leq b_n(t)\}\) is controlled by NEWLINE\[NEWLINE\limsup_{n\to\infty} n^2\Delta_n\leq {A(\varphi_a)+ A(\varphi_b)\over 4\sqrt{2\pi}},NEWLINE\]NEWLINE where \(A(\varphi_a)= (\int^T_0 {\sqrt{a''(u)}\over \varphi_a(u)} du)^2\|\varphi_a'\|_2\). Finally, some numerical results are done. This method improves Novikov's et al. (1999) result, the order of their approximation being \(O(\sqrt{\log n/n^3})\).