A new linearization technique for multi-quadratic 0-1 programming problems

We consider the reduction of multi-quadratic 0-1 programming problems to linear mixed 0-1 programming problems. In this reduction, the number of additional continuous variables is O(kn) (n is the number of initial 0-1 variables and k is the number of quadratic constraints). The number of 0-1 variables remains the same.     

An efficient linearization technique for mixed 0-1 polynomial problem

This paper addresses a new and efficient linearization technique to solve mixed 0-1 polynomial problems to achieve a global optimal solution. Given a mixed 0-1 polynomial term z=c_tx₁x₂…x_ny, where x₁,x₂,…,x_n are binary (0-1) variables and y is a continuous variable. Also, c_t can be either a positive or a negative parameter. We transform z into a set of auxiliary constraints which are linear and can be solved by exact methods such as branch and bound algorithms. For this purpose, we will introduce a method in which the number of additional constraints is decreased significantly rather than the previous methods proposed in the literature. As is known in any operations research problem decreasing the number of constraints leads to decreasing the mathematical computations, extensively. Thus, research on the reducing number of constraints in mathematical problems in complicated situations have high priority for decision makers. In this method, each n-auxiliary constraints proposed in the last method in the literature for the linearization problem will be replaced by only 3 novel constraints. In other words, previous methods were dependent on the number of 0-1 variables and therefore, one auxiliary constraint was considered per 0-1 variable, but this method is completely independent of the number of 0-1 variables and this illustrates the high performance of this method in computation considerations. The analysis of this method illustrates the efficiency of the proposed algorithm.     

An exact algorithm for the general quadratic assignment problem

We develop an algorithm that is based on the linearization and decomposition of a general Quadratic Assignment Problem of size n into n² Linear Assignment Problems of size (n − 1). The solutions to these subproblems are used to calculate a lower bound for the original problem, and this bound is then used in an exact branch and bound procedure. These subproblems are similar to the ‘minors’ defined by Lawler [16], but permit us to calculate tighter bounds. Computational experience is given for solution to optimality of general quadratic assignment problems is sizes up to n = 10.     

A coordinate gradient descent method for linearly constrained smooth optimization and support vector machines training

Support vector machines (SVMs) training may be posed as a large quadratic program (QP) with bound constraints and a single linear equality constraint. We propose a (block) coordinate gradient descent method for solving this problem and, more generally, linearly constrained smooth optimization. Our method is closely related to decomposition methods currently popular for SVM training. We establish global convergence and, under a local error bound assumption (which is satisfied by the SVM QP), linear rate of convergence for our method when the coordinate block is chosen by a Gauss-Southwell-type rule to ensure sufficient descent. We show that, for the SVM QP with n variables, this rule can be implemented in O(n) operations using Rockafellar’s notion of conformal realization. Thus, for SVM training, our method requires only O(n) operations per iteration and, in contrast to existing decomposition methods, achieves linear convergence without additional assumptions. We report our numerical experience with the method on some large SVM QP arising from two-class data classification. Our experience suggests that the method can be efficient for SVM training with nonlinear kernel.     

Analyzing quadratic unconstrained binary optimization problems via multicommodity flows

Quadratic Unconstrained Binary Optimization (QUBO) problems concern the minimization of quadratic polynomials in n{0,1}-valued variables. These problems are NP-complete, but prior work has identified a sequence of polynomial-time computable lower bounds on the minimum value, denoted by C₂,C₃,C₄,…. It is known that C₂ can be computed by solving a maximum flow problem, whereas the only previously known algorithms for computing require solving a linear program. In this paper we prove that C₃ can be computed by solving a maximum multicommodity flow problem in a graph constructed from the quadratic function. In addition to providing a lower bound on the minimum value of the quadratic function on {0,1}ⁿ, this multicommodity flow problem also provides some information about the coordinates of the point where this minimum is achieved. By looking at the edges that are never saturated in any maximum multicommodity flow, we can identify relational persistencies: pairs of variables that must have the same or different values in any minimizing assignment. We furthermore show that all of these persistencies can be detected by solving single-commodity flow problems in the same network.     

An algorithm for the quadratic assignment problem using Bender's decomposition

A well known method used for solving quadratic assignment problems proceeds by the construction of an equivalent much larger linear assignment problem with many side constraints. The disadvantage of this method lies in the weakness of the bounds obtained by solving the linear problem. An alternate linearization has been suggested using a general method of Glover. In this paper the mixed integer program obtained by Glover's method is discussed and a solution using Bender's decomposition is proposed.     

Integer programming formulation of combinatorial optimization problems

This paper considers in a somewhat general setting when a combinatorial optimization problem can be formulated as an (all-integer) integer programming (IP) problem. The main result is that any combinatorial optimization problem can be formulated as an IP problem if its feasible region S is finite but there are many rather sample problems that have no IP formulation if their S is infinite. The approach used for finite S usually gives a formulation with a relatively small number of additional variables for example, an integer polynomial of n 0?1 variables requires at most n + 1 additional variables by our approach, whereas 2ⁿ - (n + 1) additional variables at maximum are required by other existing methods. Finally, the decision problem of deciding whether an arbitrarily given combinatorial optimization problem has an IP formulation is considered and it is shown by an argument closely related to Hilbert's tenth problem (drophantine equations) that no such algorithm exists.     

Solving large single allocation p-hub problems with two or three hubs

In this paper we present an efficient approach for solving single allocation p-hub problems with two or three hubs. Two different variants of the problem are considered: the uncapacitated single allocation p-hub median problem and the p-hub allocation problem. We solve these problems using new mixed integer linear programming formulations that require fewer variables than those formerly used in the literature. The problems that we solve here are the largest single allocation problems ever solved. The numerical results presented here will demonstrate the superior performance of our mixed integer linear programs over traditional approaches for large problems. Finally we present the first mixed integer linear program for solving single allocation hub location problems that requires only O(n²) variables and O(n²) constraints that is valid for any number of hubs.     

Approximate methods for convex minimization problems with series–parallel structure

Consider a problem of minimizing a separable, strictly convex, monotone and differentiable function on a convex polyhedron generated by a system of m linear inequalities. The problem has a series–parallel structure, with the variables divided serially into n disjoint subsets, whose elements are considered in parallel. This special structure is exploited in two algorithms proposed here for the approximate solution of the problem. The first algorithm solves at most min{m, ν − n + 1} subproblems; each subproblem has exactly one equality constraint and at most n variables. The second algorithm solves a dynamically generated sequence of subproblems; each subproblem has at most ν − n + 1 equality constraints, where ν is the total number of variables. To solve these subproblems both algorithms use the authors’ Projected Newton Bracketing method for linearly constrained convex minimization, in conjunction with the steepest descent method. We report the results of numerical experiments for both algorithms.     

An Improved Formulation for the Job-Shop Scheduling Problem

This paper presents an extension of an earlier integer programming model developed by other authors to formulate a general n-job, m-machine job-shop problem. The new formulation involves substantially fewer functional constraints at the expense of an increase in the number of upper bound variables. This reduction of functional constraints, together with the imposition of upper and lower bounds on the objective value, significantly reduces the computation time for solving the integer model for the job-shop scheduling problem.     

A bicriteria flowshop scheduling with a learning effect

In many situations, a worker’s ability improves as a result of repeating the same or similar tasks; this phenomenon is known as the learning effect. In this paper the learning effect is considered in a two-machine flowshop. The objective is to find a sequence that minimizes a weighted sum of total completion time and makespan. Total completion time and makespan are widely used performance measures in scheduling literature. To solve this scheduling problem, an integer programming model with n² + 6n variables and 7n constraints where n is the number of jobs is formulated. Because of the lengthy computing time and high computing complexity of the integer programming model, the problem with up to 30 jobs can be solved. A heuristic algorithm and a tabu search based heuristic algorithm are presented to solve large size problems. Experimental results show that the proposed heuristic methods can solve this problem with up to 300 jobs rapidly. According to the best of our knowledge, no work exists on the bicriteria flowshop with a learning effect.     

Lagrangean relaxation for a lower bound to a set partitioning problem with side constraints: properties and algorithms

The problem (P) addressed here is a special set partitioning problem with two additional non-trivial constraints. A Lagrangean Relaxation (LR_u) is proposed to provide a lower bound to the optimal solution to this problem. This Lagrangean relaxation is accomplished by a subgradient optimization procedure which solves at each iteration a special 0–1 knapsack problem (KP-k). We give two procedures to solve (KP-k), namely an implicity enumeration algorithm and a column generation method. The approach is promising for it provides feasible integer solutions to the side constraints that will hopefully be optimal to most of the instances of the problem (P). Properties of the feasible solutions to (KP-k) are highlighted and it is shown that the linear programming relaxation to this problem has a worst case time bound of order O(n³).     

Linear programming for the 0–1 quadratic knapsack problem

In this paper we consider the quadratic knapsack problem which consists in maximizing a positive quadratic pseudo-Boolean function subject to a linear capacity constraint. We propose a new method for computing an upper bound. This method is based on the solution of a continuous linear program constructed by adding to a classical linearization of the problem some constraints rebundant in 0–1 variables but nonredundant in continuous variables. The obtained upper bound is better than the bounds given by other known methods. We also propose an algorithm for computing a good feasible solution. This algorithm is an elaboration of the heuristic methods proposed by Chaillou, Hansen and Mahieu and by Gallo, Hammer and Simeone. The relative error between this feasible solution and the optimum solution is generally less than 1%. We show how these upper and lower bounds can be efficiently used to determine the values of some variables at the optimum. Finally we propose a branch-and-bound algorithm for solving the quadratic knapsack problem and report extensive computational tests.     

A linearization framework for unconstrained quadratic (0-1) problems

In this paper, we are interested in linearization techniques for the exact solution of the Unconstrained Quadratic (0-1) Problem. Our purpose is to propose “economical” linear formulations. We first extend current techniques in a general linearization framework containing many other schemes and propose a new linear formulation. Numerical results comparing classical, Glover’s and the new linearization are reported.     

The balanced linear programming problem

The Balanced Linear Programming Problem (BLPP) arises in situations which require equitable distribution of a scarce resource. The BLPP can be transformed to the standard form of the linear programming problem by introducing 2∥N∥ + 2 additional variables and 2∥N∥ additional constraints. This transformation is not desirable from the computational point of view for larger values of ∥N∥ as it increases the problem size substantially. It is also undesirable from a theoretical perspective as it might affect the special structure of the constraint matrix. In this paper, we develop an algorithm for the BLPP which does not require problem enlargement. The algorithm is based on the relationship between the BLPP and the minimax linear programming problem, and solving the latter problem parametrically. Our algorithm, in essence, performs steps that are similar to those performed in the parametric simplex method with parametric right hand side. We then adapt our algorithm for the network flow problem and this specialized algorithm can be applied on the network directly without maintaining the simplex tableau.     

A Decomposition Method for Global and Local Quadratic Minimization

We present a decomposition method for indefinite quadratic programming problems having n variables and m linear constraints. The given problem is decomposed into at most m QP subproblems each having m linear constraints and n-1 variables. All global minima, all isolated local minima and some of the non-isolated local minima for the given problem are obtained from those of the lower dimensional subproblems. One way to continue solving the given problem is to apply the decomposition method again to the subproblems and repeatedly doing so until subproblems of dimension 1 are produced and these can be solved directly. A technique to reduce the potentially large number of subproblems is formulated.     

Intrinsic ordering,combinatorial numbers and reliability engineering

A new algorithm for evaluating the top event probability of large fault trees (FTs) is presented. This algorithm does not require any previous qualitative analysis of the FT. Indeed, its efficiency is independent of the FT logic, and it only depends on the number n of basic system components and on their failure probabilities. Our method provides exact lower and upper bounds on the top event probability by using new properties of the intrinsic order relation between binary strings. The intrinsic order enables one to select binary n  -tuples with large occurrence probabilities without necessity to evaluate them. This drastically reduces the complexity of the problem from exponential (2ⁿ$2^n$ binary n-tuples) to linear (n Boolean variables). Our algorithm is mainly based on a recursive formula for rapidly computing the sum of the occurrence probabilities of all binary n-tuples with weight m whose 1s are placed among the k right-most positions. This formula, as well as the balance between accuracy and computational cost, is closely related to the famous Pascal’s triangle.