Answers are grounded in this paper and cite their sources.
In our preceding paper [1], A decomposition theorem for supermartingales (to which we shall refer from now on as the "existence paper"), we have proved that Doob's decomposition problem was solvable for a supermartingale belonging to the class (D).We shall be concerned here with the uniqueness of such a decomposition.It will be proved more precisely that, among all the possible decompositions for a potential, there is one and only one "natural" decomposition.Just as in the existence paper, we are adapting to the super- martingale case some proofs (from the paper [2]) which had been originally designed for the case of excessive functions and Markov processes.The reader is referred to the existence paper for hypotheses, terminology, and notations.The numbering of theorems and definitions also follows that of the existence paper.We begin with a proof of the general uniqueness theorem, which leads to a somewhat awkward characterization of the natural decomposition.A simpler one is given in the next section, with the help of a notion of "ac- cessibility" of stopping times.We conclude with some applications to the theory of square integrable martingales.1. DEFINITION 4. Let Y Yt} be a stochastic process, well adapted to the t family, whose sample functions a.s.are right continuous and have limits from the left (including for).Let e and r be two positive numbers (e > 0, 0 < r -< ).A sequence (T) of stopping times will be called a (Y, e)-chain over [0, r] if the following conditions are fulfilled" (i) T1 0; T(0) -< Tn+l(o) (ii) The equality Tn (0) r holds for n large enough, except for an 0 set of measure 0. (iii) The function s --Ys(w) has a.s.an oscillation smaller than e on each interval [T (), Tn+l ()[.Some remarks may be made about this definition.We note first that it is not empty.If we take indeed T 0 and, inductively, T,+(o) inf {r =< r" r _>_ T,(), Yr(w) Yr.(w) > e/2}, we get a (Y, e)-chain over [0, r].Let (S,) and (Tn) be two chains of stopping times (i.e., sequences of stopping times which satisfy conditions (i) and (ii)).We construct a new chain (R,), which will be called the refinement of (Sn) and (Tn), by ordering
No takes yet. Share an insight, caveat, or question.
P. A. Meyer (1963) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: