The generalized basis reduction algorithm
Mathematics of Operations Research
Solving a System of Linear Diophantine Equations with Lower and Upper Bounds on the Variables
Mathematics of Operations Research
Decomposition of Integer Programs and of Generating Sets
ESA '97 Proceedings of the 5th Annual European Symposium on Algorithms
A Class of Hard Small 0-1 Programs
INFORMS Journal on Computing
Combining Problem Structure with Basis Reduction to Solve a Class of Hard Integer Programs
Mathematics of Operations Research
Hard Equality Constrained Integer Knapsacks
Mathematics of Operations Research
Lattice based extended formulations for integer linear equality systems
Mathematical Programming: Series A and B
Journal of Global Optimization
Column basis reduction and decomposable knapsack problems
Discrete Optimization
Hi-index | 0.00 |
Lattice-based reformulation techniques have been used successfully both theoretically and computationally. One such reformulation is obtained from the lattice kerℤ(A)={x∈ℤn|Ax=0}. Some of the hard instances in the literature that have been successfully tackled by lattice-based techniques, such as market split and certain classes of knapsack instances, have randomly generated input A. These instances have been posed to stimulate algorithmic research. Since the considered instances are very hard even in low dimension, less experience is available for larger instances. Recently we have studied larger instances and observed that the LLL-reduced basis of kerℤ(A) has a specific sparse structure. In particular, this translates into a map in which some of the original variables get a "rich" translation into a new variable space, whereas some variables are only substituted in the new space. If an original variable is important in the sense of branching or cutting planes, this variable should be translated in a non-trivial way. In this paper we partially explain the obtained structure of the LLL-reduced basis in the case that the input matrix A consists of one row a. Since the input is randomly generated our analysis will be probabilistic. The key ingredient is a bound on the probability that the LLL algorithm will interchange two subsequent basis vectors. It is worth noticing that computational experiments indicate that the results of this analysis seem to apply in the same way also in the general case that A consists of multiple rows. Our analysis has yet to be extended to this general case. Along with our analysis we also present some computational indications that illustrate that the probabilistic analysis conforms well with the practical behavior.