On subsets of \(L^ p\) and p-stable processes (Q582690): Difference between revisions

Let \(\Lambda\) be an \(L_ p\)-space. Let \((X_ t)_{t\in \Lambda}\) be a p-stable process indexed by \(\Lambda\) such that for some fixed \(r<p\) the mapping \(t\to X_ t\) is an isometric embedding of \(\Lambda\) into \(L_ r\). The author studies the subsets \(T\subset \Lambda\) such that the process \((X_ t)_{t\in T}\) is almost surely bounded. He shows that sample boundedness occurs as the result of two distinct phenomena. The first phenomenon, now well understood [thanks to the author's penetrating work in Acta Math. 159, 99-149 (1987) and in Ann. Probab. 16, No.4, 1584-1595 (1988)] is the existence of a majorizing measure (which plays a role analogous to the Haar measure in the stationary and Gaussian case) adapted to the situation. The other phenomenon, more delicate, involves the supremum of a family of positive random variables in which all aspects of cancellation seem to have disappeared. This second phenomenon cannot be described purely in terms of a metric. More precisely, the main result is a decomposition of any sample bounded p-stable process \((p<2)\) into two parts each of which falls into one of the above two categories, i.e. their sample boundedness is a result of one of the above two phenomena. Several examples are given showing that the subsets exhibited by this decomposition into two parts can be very complicated.