Black-Box Randomized Reductions in Algorithmic Mechanism Design

We give the first black-box reduction from arbitrary approximation algorithms to truthful approximation mechanisms for a non-trivial class of multi-parameter problems. Specifically, we prove that every packing problem that admits an FPTAS also admits a truthful-in-expectation randomized mechanism that is an FPTAS. Our reduction makes novel use of smoothed analysis, by employing small perturbations as a tool in algorithmic mechanism design. We develop a “duality'' between linear perturbations of the objective function of an optimization problem and of its feasible set, and use the “primal'' and “dual'' viewpoints to prove the running time bound and the truthfulness guarantee, respectively, for our mechanism.