Sierpiński gasket as a Martin boundary. I: Martin kernels (Q5937308)

Let \({\mathcal A}\) be an alphabet of \(N\geq 2\) letters and consider the tree \({\mathcal W}\) of words with finitely many, or possibly 0, letters. On \({\mathcal W}\) a Markov chain is constructed by specifying its one-step transition probability \(p(v,w)\) where a transition from \(v=v_1^{k_1} \dots v_m^{k_m}\) to \(w\) is only possible if (essentially) \(w\) is either of the form \(w=v_1^{k_1} \dots v_m^{k_m} \gamma\) or \(w=v_1^{k_1} \dots v_{m-1}^{k_{m-1}-1}v_mv_{m-1}^{k_m} \gamma\). Here \(v_j\) and \(\gamma\) are letters from the alphabet. The \(n\)-step transition probabilities \(p(n,v,w)\) and the Green function \(g(v,w)\) are now calculated in the usual way. The Martin kernel, formally defined as \(k(v,w)= g(v,w)/g (\emptyset,w)\) (one always has \(g(\emptyset,w)>0)\), is given by an exact formula which allows to characterize the Martin boundary explicitly. As a first step, the authors show that the Sierpiński gasket in \(\mathbb{R}^N\) is Lipschitz equivalent with the space \(\Sigma\) of infinite words (with an equivalence relation identifying certain eventually constant words). The main result of the paper is that the Martin boundary and the Sierpiński gasket in \(\mathbb{R}^N\) are homeomorphic; the homeomorphism between these spaces is explicitly constructed. From this representation of the Martin boundary one concludes that harmonic functions of the Markov chain have integral representations over the Martin boundary. Moreover, the Sierpiński gasket is the space of exits. [For part II see below].