Basis Reduction, and the Complexity of Branch-and-Bound

The classical branch-and-bound algorithm for the integer feasibility problem has exponential worst case complexity. We prove that it is surprisingly efficient on reformulations, in which the columns of the constraint matrix are short, and near orthogonal, i.e. a reduced basis of the generated lattice; when the entries of A (i.e. the dense part of the … Read more

Lattice-based Algorithms for Number Partitioning in the Hard Phase

The number partitioning problem (NPP) is to divide n numbers a_1,…,a_n into two disjoint subsets such that the difference between the two subset sums – the discrepancy, D, is minimized. In the balanced version of NPP (BalNPP), the subsets must have the same cardinality. With $a_j$s chosen uniformly from $[1,R]$, R > 2^n gives the … Read more

On sublattice determinants in reduced bases

We prove several inequalities on the determinants of sublattices in LLL-reduced bases. They generalize the fundamental inequalities of Lenstra, Lenstra, and Lovasz on the length of the shortest vector, and show that LLL-reduction finds not only a short vector, but also sublattices with small determinants. We also prove new inequalities on the product of the … Read more

Parallel Approximation, and Integer Programming Reformulation

We analyze two integer programming reformulations of the n-dimensional knapsack feasibility problem without assuming any structure on the weight vector $a.$ Both reformulations have a constraint matrix in which the columns form a reduced basis in the sense of Lenstra, Lenstra, and Lov\’asz. The nullspace reformulation of Aardal, Hurkens and Lenstra has n-1 variables, and … Read more

Column basis reduction and decomposable knapsack problems

We propose a very simple preconditioning method for integer programming feasibility problems: replacing the problem b’   ≤   Ax   ≤   b,   x ∈ Zn with b’   ≤   (AU)y   ≤   b,   y ∈ Zn, where U is a unimodular matrix computed via basis reduction, to make the … Read more