Higher moments for random multiplicative measures (Q887518)

Let \(\{ \mu_k\}\) be a sequence of random measures determined by random weights on an \(m\)-ary tree (a tree where each vertex has \(m\) children). Denote by \(W_{\mathbb{I}}\) the weight at a particular vertex \(\mathbb{I}\) of the tree. For ease of visualization, each vertex in the tree can be identified with an \(m\)-ary subinterval of \([0,1]\), so \(\mu_k\) can be thought of as a measure on \([0,1]\). Let \(\mu_k([i/m^{\ell},(i+1)/m^{\ell}])= W_{\mathbb{I}1} \dots W_{\mathbb{I}\ell}\), where \(\mathbb{I}1,\dots, \mathbb{I}\ell\) are the vertices from the root of the tree to the vertex representing this interval. Call this interval a cylinder set \(c_{\mathbb{I}}\). Such a collection of measures has been called a multiplicative cascade and it is known that \(\mu_k(A)\) is a martingale. A natural question is: how does this martingale converge as \(k \rightarrow \infty\)? This paper addresses the question of when does the martingale converge in \(L^{q}\) for \(q>1\). If the weights are self-similar, that is, the weights of the children of any particular vertex always have the same distribution independent of their ancestor, there is a long known (see [\textit{J.-P. Kahane}, C. R. Acad. Sci., Paris, Sér. A 278, 621--623 (1974; Zbl 0275.60062)] or [\textit{Q.-S. Liu}, Stochastic Processes Appl. 86, No. 2, 263--286 (2000; Zbl 1028.60087)]) necessary and sufficient condition, namely, that \[ \mathbb{E} \left( \sum_{i=1}^{m} W_i^{q}\right)^{1/q} < 1. \] For non-self-similar cascades, \textit{J. Barral} and \textit{B. B. Mandelbrot} [Proc. Symp. Pure Math. 72, 3--16 (2004; Zbl 1088.28004)] established the same necessary and sufficient condition as the current author does, but they only proved it for \(1<q<2\). In the author's words: ``The analysis required seems significantly more awkward when \(q>2\) and in particular when \(q\) is not an integer.'' A significant source of this awkwardness is how to handle what are the leftover terms in the product due to the fractional part of \(q\) (see, for example, Proposition 3.1.). The result of this paper is that the martingale \(\mu_k(A)\) converge in \(L^{q}\) if and only if, for a given Borel probability measure \(\mu\) on the boundary of the tree, one has \[ \limsup_{k \rightarrow \infty} \left( \sum_{|\mathbb{I}| = k} \mathbb{E} \chi_{c_{\mathbb{I}}}\mu(c_{\mathbb{I}})^{q})\right)^{1/k} < 1. \]