A Polynomial Algorithm for Some Relaxed Subset Sum Problems with Real Numbers

TL;DR

This paper introduces a polynomial-time algorithm for certain relaxed subset sum problems with real numbers by transforming the problem into a geometric optimization over polytopes and intersections of balls.

Contribution

It formulates the subset sum problem as a distance maximization over polytopes and intersections of balls, providing a novel geometric characterization and an efficient solution method.

Findings

The algorithm can solve specific relaxed subset sum problems with real numbers.

The geometric approach characterizes optimal solutions via a parameterized family of polytopes.

The method identifies solutions by analyzing the evolution of polytopes relative to the intersection of balls.

Abstract

In this paper we study the subset sum problem with real numbers. Starting from the given problem, we formulate a quadratic maximization problem over a polytope, P, which is eventually written as a distance maximization to a fixed point over the polytope. Next, starting from the obtained polytope, we construct an intersection of balls which includes the polytope and show that in case the subset sum problem has a solution we can find it by maximizing the distance to the fixed point over the intersection of balls. That is, we show that the points which maximize the distance to the fixed point over the polytope are the same points which maximize the distance to the fixed point over the constructed intersection of balls. For the latter problem we give an original result which allows the characterization of the optimum points as follows: with the centers of the balls and the said fixed point we form a function whom sublevel sets are polytopes. As such we obtain a uni-parameter, family of polytopes. We show that by increasing this parameter from 0 to a finite maximum value (which is computed through a convex optimization problem), the polytopes in the family evolve from initially containing the intersection of balls towards three possible outcomes: a) included in the interior of the intersection of balls b) included in the interior of the complementary of the intersection of balls c) the border of the intersection of balls and the polytope of maximum parameter share at least a point. Then we show that the maximum distance to the fixed point over the intersection of balls is given by a) the smallest parameter for which the polytopes enter the intersection of balls, b) the largest parameter for which the polytopes still share points with the intersection of balls c) the maximum value of the parameter. ...