On first range times of linear diffusions (Q2576792)

In an earlier paper [Stochastics Stochastics Rep. 43, No. 1--2, 93--115 (1993; Zbl 0808.60069)] the second author showed that the running maximum \(S_t = \sup_{s\leq t}W_t\) and minimum \(I_t := -\inf_{s\leq t}W_t\) of a standard Brownian motion has the following property: the distribution that \(S_t + I_t\) exceeds an independent, uniformly distributed threshold \(U\in [0,a]\) is the same as that of Brownian motion \(W_t\) stopped upon exiting the interval \((U, a-U)\). In the present paper the authors show that one may replace \((U,a-U)\) by \((X,X')\) where \(X,X'\) are random variables whose joint distribution is the same as the joint distribution of \((UV,(1-U)V)\) where \(U\) is uniform on \([0,1]\) and \(V\) is positive, independent of \(U\) and otherwise arbitrary. Other characterizations of this class, in terms of the joint distribution function, are also given. This problem is related to stopping Brownian motion, or, more generally, a linear diffusion process \(Z_t\), at a (randomized) first range time. By this the authors mean the right-continuous inverse \(\theta(r)\) of the process \(R_t = S_t + I_t\) (\(S_t, I_t\) are now defined for \(Z_t\) rather than \(W_t\)). It is shown that for \((X,X')\) in the above mentioned class one has the following identity in law: \(\theta(X) \overset{d}{=} H(X,X')\) where \(H(a,b)\) is the first exit time of \(Z_t\) from the interval \((-a,b)\). Moreover, \((Z_s, s\leq \theta(X)) \overset{d}{=} (Z_s, s\leq H(X,X'))\). Finally, Section 4 contains the following Ray-Knight theorem for Brownian motion \((W_t)\): let \(\theta(r)\) be as above (defined for \(W_t\)), let \(T\) be independent, exponentially distributed with parameter \(\alpha\) and denote by \(L(t,y)\) the local time of \(W_t\) at level \(y\). Then \(P(L(\theta(T),0)\in d\xi) = f(\alpha,\xi)\,dx\), where \(f(\alpha,\xi)\) is explicitly determined in terms of Bessel functions.