The slow regime of randomly biased walks on trees (Q504251): Difference between revisions

The authors investigate a random walk \((X_n)_{n\in\mathbb{N}_0}\) on a supercritical Galton-Watson tree \(\mathbb{T}\) with random transition probabilities (in a random environment) \(\omega:=(\omega(x,y))_{x\sim y, x,y\in\mathbb{T}}\), where \(x\sim y\) means that \(x\) is either a child or a parent of \(y\). Associated to the random environment \(\omega\) are random weights \(A_i(x):=\omega(x,x^{(i)})/\omega(x,x^\ast)\), \(1\leq i\leq N(x)\), where \(x^{(1)},\ldots, x^{(N(x))}\) are children of a vertex \(x\in \mathbb{T}\backslash\{\oslash\}\) (\(\oslash\) is the root of \(\mathbb{T}\)), \(N(x)\) is their number, \(x^\ast\) is the parent of \(x\). It is assumed that for different \(x\in\mathbb{T}\backslash\{\oslash\}\) the vectors \((A_1(x),\ldots, A_{N(x)}(x))\) are independent and identically distributed, also the following identities \[ \mathbb{E}\sum_{i=1}^{N(x)}A_i(x)=1, \quad \mathbb{E}\sum_{i=1}^{N(x)}A_i(x)\log A_i(x)=0, \] and that some additional integrability conditions hold, which in particular ensure that \(\sigma^2:=\mathbb{E}\sum_{i=1}^{N(x)}A_i(x)(\log A_i(x))^2\in (0,\infty)\). Recall that in the theory of branching random walks, which are essentially used in the present paper, the centered equalities define the so-called ``boundary case''. It is known [\textit{G. Faraud}, Electron. J. Probab. 16, Paper No. 6, 174--215 (2011; Zbl 1228.60115)] that under the aforementioned assumptions \((X_n)\) is null-recurrent for almost all \(\omega\). Let \(k\in\mathbb{N}\) and \(0<t_1<\ldots<t_k\leq 1\) be arbitrary. The main result of the paper states that, upon the survival of \(\mathbb{T}\), the quenched distribution of \((X_{[t_1n]},\ldots, X_{[t_kn]})\) is well approximated in the total variation distance by a product probability measure. As a consequence, the authors conclude that, conditionally on the survival, under the annealed probability, the random variables \(\sigma^2 (\log n)^{-2} |X_{[t_1n]}|,\ldots, \sigma^2 (\log n)^{-2} |X_{[t_kn]}|\) are asymptotically independent and converge in distribution to the same limit with the density \(f(x):=(2\pi x)^{-1/2}\mathbb{P}\{\eta\leq x^{-1/2}\}1_{(0,\infty)}(x)\), where \(\eta:=\sup_{s\in [0,1]}(\bar{m}(s)-m(s))\), \((m(s))_{s\in[0,1]}\) is a standard Brownian meander, and \(\bar{m}(s):=\sup_{u\in [0,s]}m(u)\). An explicit series representation of \(f\) is given.