Lattices that admit logarithmic worst-case to average-case connection factors

We exhibit an average-case problem that is as hard as finding γ(n)-approximate shortest nonzero vectors in certain n-dimensional lattices in the worst case, for γ(n) = O(√log n). The previously best known factor for any non-trivial class of lattices was γ(n) = Õ(n). Our results apply to families of lattices having special algebraic structure. Specifically, we consider lattices that correspond to ideals in the ring of integers of an algebraic number field. The worst-case problem we rely on is to find approximate shortest vectors in these lattices, under an appropriate form of preprocessing of the number field. For the connection factors γ(n) we achieve, the corresponding decision problems on ideal lattices are not known to be NP-hard; in fact, they are in P. However, the search approximation problems still appear to be very hard. Indeed, ideal lattices are well-studied objects in computational number theory, and the best known algorithms for them seem to perform no better than the best known algorithms for general lattices. To obtain the best possible connection factor, we instantiate our constructions with infinite families of number fields having constant root discriminant. Such families are known to exist and are computable, though no efficient construction is yet known. Our work motivates the search for such constructions. Even constructions of number fields having root discriminant up to O(n2/3-ε) would yield connection factors better than Õ(n). As an additional contribution, we give reductions between various worst-case problems on ideal lattices, showing for example that the shortest vector problem is no harder than the closest vector problem. These results are analogous to previously-known reductions for general lattices.