New algorithms break mathematical barriers for lattice problems
New randomized algorithms improve the speed and space needed for discrete Gaussian sampling and the Shortest Vector Problem.
New classical randomized algorithms break the barrier. These algorithms work for discrete Gaussian sampling (DGS) and the Shortest Vector Problem (SVP).
The algorithms use Gaussian difference dynamics (GDD). This is a framework for tracking expected counts and pairwise clustering of vectors through repeated differencing.
For shifted DGS on arbitrary lattices with width and fixed , the time is . The space used is .
For SVP on arbitrary lattices, the algorithm returns a shortest nonzero vector with high probability. It takes time and space.
The GDD framework uses a closed recursion for the first two counting moments. These moments control quantities relative to a Gaussian reference.
Transitions occur on thin spherical shells. At these shells, Gaussian rejection weights can be renormalized at subexponential cost.
Key facts
- The algorithms break the barrier for DGS and SVP.
- The SVP algorithm takes time and space.
- The GDD framework tracks expected counts and pairwise clustering through repeated differencing.