The class of POPs (Polynomial Optimization Problems) over cones covers a wide range of optimization problems such as 0-1 integer linear and quadratic programs, nonconvex quadratic programs and bilinear matrix inequalities. This paper presents a new framework for convex relaxation of POPs over cones in terms of linear optimization problems over cones. It provides a unified treatment of many existing convex relaxation methods based on the lift-and-project linear programming procedure, the reformulation-linearization technique and the semidefinite programming relaxation for a variety of problems. It also extends the theory of convex relaxation methods, and thereby brings flexibility and richness in practical use of the theory.
An improved surrogate constraints method for solving separable nonlinear integer programming problems with multiple constraints is presented. The surrogate constraints method is very effective in solving problems with multiple constraints. The method solves a succession of surrogate constraints problems having a single constraint instead of the original multiple constraint problem. A surrogate problem with an optimal multiplier vector solves the original problem exactly if there is no duality gap. However, the surrogate constraints method often has a duality gap, that is it fails to find an exact solution to the original problem. The modification proposed closes the surrogate duality gap. The modification solves a succession of target problems that enumerates all solutions hitting a particular target. The target problems are produced by using an optimal surrogate multiplier vector. The computational results show that the modification is very effective at closing the surrogate gap of multiple constraint problems.
We propose a new relaxation scheme for the MAX-CUT problem using second-order cone programming. We construct relaxation problems to reflect the structure of the original graph. Numerical experiments show that our relaxation gives better bounds than those based on the spectral decomposition proposed by Kim and Kojima [16], and that the efficiency of the branch-and-bound method using our relaxation is comparable to that using semidefinite relaxation in some cases.
We propose a 0.935-approximation algorithm for MAX 2SAT and a 0.863-approximation algorithm for MAX DICUT. The approximation ratios improve upon the recent results of Zwick, which are equal to 0.93109 and 0.8596434254 respectively. Also proposed are derandomized versions of the same approximation ratios. We note that these approximation ratios are obtained by numerical computation rather than theoretical proof. The algorithms are based on the SDP relaxation proposed by Goemans and Williamson but do not use the 'rotation' technique proposed by Feige and Goemans. The improvements in the approximation ratios are obtained by the technique of 'hyperplane separation with skewed distribution function on the sphere.'
In this paper, by using Data Envelopment Analysis (DEA) technique a method is proposed to find efficient solutions of 0-1 Multiple Objective Linear Programming (MOLP) problem. In this method from a feasible solution of 0-1 MOLP problem, a Decision Making Unit (DMU) without input vector is constructed in which output vector for DMU is the values of objective functions. The method consists of a two-stage algorithm. In the first stage, some efficient solutions are generated. In the second stage, the DMUs corresponding to the generated efficient solutions in the first stage together with the generated DMUs in the previous iterations are evaluated by using the additive model without input.
Well-known goods do not sell well these days. However goods are required to be easily recognizable. This paper treats an optimal allocation of marketing effort just after new items have been put on the market. The optimal allocation minimizes the number of people who cannot recognize the new items. We consider two types of means of communication. The first type includes television and newspapers that reach great numbers of people. The second type includes standing signboards and neon signs that reach small numbers of people. Formulating our allocation of marketing effort into a non-linear program, we develop an efficient algorithm for it. We give an illustrative example and show the result solved by the proposed algorithm. In addition, we show the relation between unrecognition and marketing effort amount.
This paper considers a FIFO single-server queue with independent and homogeneous sources. Each source generates exactly one message every interval of fixed length. Each message is then divided into a constant number of fixed-size cells and these cells arrive to the queue back to back as if they form a train of cells. We call this arrival process clustered periodic arrivals. The queue with clustered periodic arrivals is an obvious generalization of �� D/D/1 queue which corresponds to the case that each message consists of only one cell. This paper derives the stationary probability distributions of sojourn times of respective cells in a message. An interesting feature of these sojourn time distributions is that they are not continuous functions of time, but they have masses at multiples of the cell transmission time. This paper also derives the joint probability distribution of differences between sojourn times of successive cells in a message and the mean waiting times of respective cells in a message. At last, the overall mean waiting time in the queue with clustered periodic arrivals is compared with those in the corresponding queues with dispersed periodic arrivals and periodic batch arrivals, and the efficiency of dispersing cells is quantitatively shown by simple formulas.