Let be the Snell envelope of integrable adapted rewards , and use the Doob decomposition in discrete time , with and nondecreasing. Definewith . Since is -measurable, is a stopping time.
Before the first increase, , so is a martingale. If , its next compensator increment is positive and the Snell envelope recursion forces ; if , this is the terminal condition. The optional stopping theorem gives , attaining the supermartingale majorant bound. This rule may wait through earlier ties between reward and continuation; it need not be the first time .
Articles by others on the same topic
There are currently no matching articles.