Approximating Quadratic 0-1 Programming via SOCP

Sanjiv Kapoor,Hemanshu Kaul

DOI: https://doi.org/10.48550/arXiv.1312.7042

2013-12-26

Computational Complexity

Abstract:We consider the problem of approximating Quadratic O-1 Integer Programs with bounded number of constraints and non-negative constraint matrix entries, which we term as PIQP. We describe and analyze a randomized algorithm based on a program with hyperbolic constraints (a Second-Order Cone Programming -SOCP- formulation) that achieves an approximation ratio of $O(a_{max} \frac{n}{\beta(n)})$, where $a_{max}$ is the maximum size of an entry in the constraint matrix and $\beta(n) \leq \min_i{W_i} $, where $W_i$ are the constant terms that define the constraint inequalities. We note that by appropriately choosing $\beta(n)$ the randomized algorithm, when combined with other algorithms that achieve good approximations for smaller values of $ W_i$, allows better algorithms for the complete range of $W_i$. This, together with a greedy algorithm, provides a $O^*(a_{max} n^{1/2} )$ factor approximation, where $O^*$ hides logarithmic terms. Our solution is achieved by a randomization of the optimal solution to the relaxed version of the hyperbolic program. We show that this solution provides the approximation bounds using concentration bounds provided by Chernoff-Hoeffding and Kim-Vu.