Setting
There are ''m'' resources that are assumed to be ''homogeneous'' and ''divisible''. There are ''n'' agents, each of whom has a personal function that attributes a numeric value to each "bundle" (combination of resources). The valuations are assumed to beAlgorithm
PAM works in the following way. * Calculate the max-product allocation; denote it by z. * For each agent ''i'': ** Calculate the max-product allocation when ''i'' is not present''.'' ** Let ''fi'' = (the product of the other agents in z) / (the max-product of the other agents when ''i'' is not present). ** Give to agent ''i'' a fraction ''fi'' of each resource he gets in z.Properties
PAM has the following properties. * It is a truthful mechanism - each agent's utility is maximized by revealing his/her true valuations. * For each agent ''i'', the utility of ''i'' is at least 1/''e'' ≈ 0.368 of his/her utility in the max-product allocation. * When the agents have additive linear valuations, the allocation is envy-free.PA vs VCG
The PA mechanism, which does not use payments, is analogous to the VCG mechanism, which uses monetary payments. VCG starts by selecting the ''max-sum'' allocation, and then for each agent ''i'' it calculates the max-sum allocation when ''i'' is not present, and pays ''i'' the ''difference'' (max-sum when ''i'' is present)-(max-sum when ''i'' is not present). Since the agents are quasilinear, the utility of ''i'' is reduced by an ''additive'' factor. In contrast, PA does not use monetary payments, and the agents' utilities are reduced by a ''multiplicative'' factor, by taking away some of their resources.Optimality
It is not known whether the fraction of 0.368 is optimal. However, there is provably no truthful mechanism that can guarantee to each agent more than 0.5 of the max-product utility.Extensions
The PAM has been used as a subroutine in a truthful cardinal mechanism for one-sided matching.References
{{reflist Fair division protocols Mechanism design