Title:Negative Weights Make Unique Games Harder

Abstract:In the second paper of this series, the author introduces GUGP($\rho$) model and defines the counterpart of UGC on GUGP($\rho$), which is a generalization of UGP where the weights of edges are allowed to be negative, and the magnitude of the total weight of all satisfied negative edges is no more than $\rho$ times the total weight of all positive edges. The author shows a gap amplification theorem on $(c,s)$-approximation NP-hardness of GUGP($\rho$),which states that the $(c,s)$-approximation NP-hardness of UGP implies the $((\rho+1)c-\rho,(\rho+1)s-\rho)$-approximation NP-hardness of GUGP($\rho$) for any $\rho>0$ The theorem confirms that the region of $(c,s)$-approximation NP-hardness of GUGP($\rho$) is significantly larger than the known region of $(c,s)$-approximation NP-hardness of UGP. As a corollary, the author shows that UGC on GUGP($\rho$) holds true, if Min UGP is NP-hard to approximated within any constant.