Strong approximations of additive functionals of a planar Brownian motion. (Q2574627)

Let \(W_1\) and \(W_2\) be two independent standard Brownian motions and \(f_1,\dots ,f_n\) measurable functions from \(\mathbb R^2\) to \(\mathbb R\). The aim of the paper is to describe the asymptotic behavior of paths of the stochastic processes \(\int _0^tf_i(W_1(s),W_2(s))\,ds\) for large times. First of all, if \(\beta \) is a standard Brownian motion, we denote by \(\ell \) the process of local times of \(\beta \) at \(0\), and \(\sigma (x)=\inf \{s>0:\beta (s)>x\}\) the first passage process of \(\beta \). Finally, we define a Lévy process \(\mathbf e(t)=\ell (\sigma (t/2))\). The main result is the following: There exist explicitly defined functionals \(C_1\), \(C_2\), \(C_3\) and \(C_4\) such that, whenever \(\nu >5/2\) and \(f_1,\dots ,f_n\) are real measurable functions satisfying \(| z| ^2(1+| \log | z| | )^{\nu }| f_i(z)| \leq K\) on \(\mathbb R^2\setminus \{0\}\) for every \(1\leq i\leq n\) and some constant \(K\), there exist, possibly on an enlarged probability space, independent standard Brownian motions \(W_1\), \(W_2\), an \(\mathbb R^n\)-valued Brownian motion \(Y=(Y_1,\dots ,Y_n)\) with covariance \(C_3(f_i,f_j)\), a process \(\widetilde {\mathbf e}\) having the same law as \(\mathbf e\) and independent of \(Y\), so that \[ \int _0^tf_i(W_1(s),W_2(s))\,ds=\frac {C_1(f_i)}{2\pi }\mathbf e(\log t)+C_2(f_i)Y_i(\widetilde {\mathbf e}(\log t))+o((\log t)^{1/2-\delta }), \] \[ | \mathbf e(\log t)-\widetilde {\mathbf e}(\log t)| =o((\log t)^{1-\delta }) \] hold for large \(t\) almost surely for some constant \(\delta >10^{-5}\). This theorem unifies the results due to \textit{G. Kallianpur} and \textit{H. Robbins} [Proc. Natl. Acad. Sci. USA 39, 525--533 (1953; Zbl 0053.10003)] and \textit{Y. Kasahara} and \textit{S. Kotani} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 49, 133--153 (1979; Zbl 0435.60080)], and yields, among other corollaries presented in the paper, a law of iterated logarithm \[ \limsup _{t\to \infty }\frac {\int _0^tf(W_1(s),W_2(s))\,ds}{\sqrt {\log t}\log \log \log t}=\frac {C_2(f)}{\sqrt 2}, \] and a central limit theorem \[ (\log t)^{1/2}\left (\frac {\int _0^tf_1(W_1(s),W_2(s))\, ds} {\int _0^tf_2(W_1(s),W_2(s))\, ds}-\frac {C_1(f_1)}{C_1(f_2)}\right )@>\text{in law}>t\to \infty > a(f_1,f_2)\frac {\mathcal N}{\sqrt {\mathbf e(1)}} \] where \(\mathcal N\) is a standard Gaussian random variable, \(\mathbf e(1)\) is exponentially distributed with mean \(1\), \(\mathcal N\) and \(\mathbf e(1)\) are independent, and \(a\) is an explicitly defined functional. Finally, in the last section, applications to winding numbers of \((W_1,W_2)\) and additive functionals of a Cauchy process are given.