Iterating Brownian motions, ad libitum (Q2248931)

The \(n\)-fold iterated Brownian motion \(W_n\) is defined by \[ W_n(t) = B_n(B_{n-1}(\dots (B_2(B_1(t))) \dots)), \] where \(B_1,B_2,\dots\) are independent two-sided standard Brownian motions, such that \(B_n(0)=0\) for all \(n \geq 1\). \textit{K. Burdzy} [Prog. Probab. 33, 67--87 (1992; Zbl 0789.60060)] has previously studied the sample-path properties of this process in the case \(n=2\). The present paper explores the asymptotics of the process \(W_n\) as \(n \rightarrow \infty\). As the authors explain, the process \(W_n\) is not a semimartingale unless \(n=1\) and increasing \(n\) roughens the sample paths of the process (\(W_n\) can be shown to have finite \(2^n\)-variation). Thus, it is not obvious that \(W_n\) has a limit as \(n \rightarrow \infty\) and, indeed, the question turns out to be delicate. The first main result of the paper shows that any finite-dimensional distribution of \(W_n\) converges weakly to an exchangeable distribution. However, weak convergence of the process \(W_n\) in any reasonable function space seems impossible and, in fact, it is shown in the paper that the laws of \(W_n\), \(n \geq 1\), are not tight in the space of continuous paths, equipped with the locally uniform topology. The proof of the convergence of the finite-dimensional distributions of \(W_n\) is based on a nice argument that treats \((W_n(x_1),\dots,W_n(x_p))\), for distinct nonzero real numbers \(x_1,\dots,x_p\), as a Markov chain parameterized by \(n\), which is shown to be positive Harris recurrent. The second main result of the paper concerns the occupation measure \(\mu_n\) of \(W_n\), defined via \[ \int_{\mathbb R} f\, d \mu_n = \int_0^1 f(W_n(t))\, dt \] for any Borel function \(f: {\mathbb R} \rightarrow [0,\infty)\). The authors show that the random measure \(\mu_n\) converges, as \(n \rightarrow \infty\), to a random measure \(\mu_\infty\), which is absolutely continuous with respect to the Lebesgue measure almost surely. This result parallels the principle, suggested by Simeon Berman, that very rough functions should have smooth local times.