We show that the Learning with Errors (LWE) problem is classically at least as hard as standard worst-case lattice problems, even with polynomial modulus. Previously this was only known under quantum reductions.Our techniques capture the tradeoff between the dimension and the modulus of LWE instances, leading to a much better understanding of the landscape of the problem. The proof is inspired by techniques from several recent cryptographic constructions, most notably fully homomorphic encryption schemes. * Stanford University, zvika@stanford.edu. Supported by a Simons Postdoctoral Fellowship and DARPA.Our focus in this paper is on the latter problem, learning with errors. In this problem our goal is to distinguish with some non-negligible advantage between the following two distributions:where s is chosen uniformly from Z n q and so are the a i ∈ Z n q , u i are chosen uniformly from Z q , and the "noise" e i ∈ Z is sampled from some distribution supported on small numbers, typically a (discrete) Gaussian distribution with standard deviation αq for α = o(1).The LWE problem has proved to be amazingly versatile, serving as the basis for a multitude of cryptographic constructions: secure public-key encryption under both chosen-plaintext [Reg05, PVW08, LP11] and chosen-ciphertext [PW08, Pei09, MP12] attacks, oblivious transfer [PVW08], identity-based encryption [GPV08, CHKP10, ABB10a, ABB10b], various forms of leakage-resilient cryptography (e.g., [AGV09, ACPS09, GKPV10]), fully homomorphic encryption [BV11, BGV12, Bra12] (following the seminal work of Gentry [Gen09]), and much more. It was also used to show hardness of learning problems [KS06].Contrary to the SIS problem, however, the hardness of LWE is not sufficiently well understood. The main hardness reduction for LWE [Reg05] is similar to the one for SIS mentioned above, except that it is quantum. This means that the existence of an efficient algorithm for LWE, even a classical (i.e., non-quantum) one, only implies the existence of an efficient quantum algorithm for lattice problems. This state of affairs is quite unsatisfactory: even though one might conjecture that efficient quantum algorithms for lattice problems do not exist, our understanding of quantum algorithms is still at its infancy. It is therefore highly desirable to come up with a classical hardness reduction for LWE.Progress in this direction was made by [Pei09] (with some simplifications in the followup by Lyubashevsky and Micciancio [LM09]). The main result there is that LWE with exponential modulus is as hard as some standard lattice problems using a classical reduction. As that hardness result crucially relies on the exponential modulus, the open question remained as to whether LWE is hard for smaller moduli, in particular polynomial moduli. In addition to being an interesting question in its own right, this question is of special importance since many cryptographic applications, as well as the learning theory result of Klivans and Sherstov [KS06], are instantiated in this setting. Some addit...

Abstract. Most lattice-based cryptographic schemes are built upon the assumed hardness of the Short Integer Solution (SIS) and Learning With Errors (LWE) problems. Their efficiencies can be drastically improved by switching the hardness assumptions to the more compact Ring-SIS and Ring-LWE problems. However, this change of hardness assumptions comes along with a possible security weakening: SIS and LWE are known to be at least as hard as standard (worst-case) problems on euclidean lattices, whereas Ring-SIS and Ring-LWE are only known to be as hard as their restrictions to special classes of ideal lattices, corresponding to ideals of some polynomial rings. In this work, we define the Module-SIS and Module-LWE problems, which bridge SIS with Ring-SIS, and LWE with Ring-LWE, respectively. We prove that these average-case problems are at least as hard as standard lattice problems restricted to module lattices (which themselves bridge arbitrary and ideal lattices). As these new problems enlarge the toolbox of the lattice-based cryptographer, they could prove useful for designing new schemes. Importantly, the worst-case to average-case reductions for the module problems are (qualitatively) sharp, in the sense that there exist converse reductions. This property is not known to hold in the context of Ring-SIS/Ring-LWE: Ideal lattice problems could reveal easy without impacting the hardness of Ring-SIS/Ring-LWE.