HOWTO: asynchronous PFC-MRDAC: optimization in distributed constraint problems /spl plusmn/Adopt

Each agent has private problems. Private concerns can often be formulated in a general framework such as constraint satisfaction (where everything is modeled by either variables, values, or constraints). Often agents need to find agreement with others for the allocation of final resources. The constraint satisfaction (CSPs) is only a special case of optimization. In this paper it is shown how a very general technique for distributed CSPs, replica-based multiply asynchronous search (R-MAS) (comprising ABT, ABTR, AAS, DMAC, DMAC-ABT), can be extended and applied to optimization problems in distributed Weighted CSPs (WCSPs). Centralized WCSPs can be seen as MAX-CSPs where several constraints can link the same variables. PFC-MRDAC is a good approach to MAX-CSPs. How to asynchronize and adapt it to distributed WCSPs? This article describes how asynchronous consistency maintenance can be introduced in Adopt and in asynchronous branch&bound. The main new ideas proposed in this article are that: (a) a concept called weighted consistency nogood (WCN) allows to maintain consistency in DisWC-SPs, (b) leading to an asynchronous equivalent of PFC-MRDAC. (c) The feedback that B&B or Adopt needs about low bounds can be extracted from such WCNs.