Thin points for Brownian motion (Q1594509)

Let \(\{W_t\}_t\) be Brownian motion in \(\mathbb{R}^n\), \(n\geq 2\), and denote by \(\mu_T(A)= \int^T_0\mathbf{1}_A(W_t) dt\) the occupation measure. For any \(T\), \(a> 0\) the set of thin times is \[ \text{THIN}_a:= \Biggl\{t\in (0,T): \liminf_{h\to 0} {\mu_T(B_h(W_t))\over h^2/|\log h|}= a\Biggr\} \] and \(x\in\mathbb{R}^n\) is a thin point if \(x= W_t\) for some thin time \(t\). The main result of the paper is that a.s. for every \(a>0\) the Hausdorff dimension \[ \dim_{\text{H}}(\text{THIN}_a)= 1-a^{-1},\tag{\(*\)} \] while the packing dimension \(\dim_{\text{P}}(\text{THIN}_{1+ a})= 1\). As a consequence of this result one has for any analytic subset \(E\subset (0,T)\) that a.s. \[ \inf_{t\in E} \liminf_{h\to 0} {\mu_T(B_h(W_t))\over h^2/|\log h|}= {1\over \dim_{\text{P}}(E)}. \] The proof of the upper bound in \((*)\) follows from an application of the sharp asymptotics for the two-sided exit problems for Brownian motion obtained by \textit{J.-C. Gruet} and \textit{Z. Shi} [J. Theor. Probab. 9, No. 2, 429-445 (1996; Zbl 0870.60072)]. The lower bound requires random fractal techniques which are reminiscent of earlier methods used by the present authors for Brownian thick points in [Ann. Probab. 28, No. 1, 1-35 (2000)]. The general idea is again to construct a discrete `limsup' random fractal which is in some sense close to a subset of \(\text{THIN}_a\) and which enjoys some `quasi-locality' property (i.e., independence of subsets). The technique is developed in some generality in an independent section of the paper. The central results of this section are: (1) a zero-one-type law for the probability that the random fractal hits an analytic set \(E\), depending on the relation of \(\dim_{\text{P}}(E)\) to a certain index of the fractal; (2) a concrete construction of random fractals with certain hitting properties. The proof of \((*)\) uses the inclusions: \(\bigcup_{c\leq a}\text{Qscape}_c\subset \bigcup_{c\leq a} \text{THIN}_c\subset \bigcup_{c\leq a} \text{BiFast}_c\) for the sets of quick escape times \(\text{Qscape}_c\) and bilaterally fast times \(\text{BiFast}_c\). These sets are, similar to \(\text{THIN}_c\), given by a \(\liminf\) relation but with the occupation measure being replaced by some two-sided exit time from \(B_h(W_t)\), resp. the length of the time interval containing a.s. all visits to the ball \(B_h(W_t)\). The lower bound in \((*)\) finally follows by inscribing some `limsup' random fractal into the Qscape-sets and estimating its dimension.