No abstract

Discrete Applied Mathematics 123 (2002) 75-102. doi:10.1016/S0166-218X(01)00338-92016-03-04T18:46:49

In this paper, we study inverse optimization problems defined as follows. Let S denote the set of feasible solutions of an optimization problem P, let c be a specified cost vector, and x 0 be a given feasible solution. The solution x 0 may or may not be an optimal solution of P with respect to the cost vector c. The inverse optimization problem is to perturb the cost vector c to d so that x 0 is an optimal solution of P with respect to d and d − c p is minimum, where d − c p is some selected L p norm. In this paper, we consider the inverse linear programming problem under L 1 norm (where d − c p = i∈J w j d j − c j , with J denoting the index set of variables x j and w j denoting the weight of the variable j) and under L norm (where d − c p = max j∈J w j d j − c j. We prove the following results: (i) If the problem P is a linear programming problem, then its inverse problem under the L 1 as well as L norm is also a linear programming problem. (ii) If the problem P is a shortest path, assignment or minimum cut problem, then its inverse problem under the L 1 norm and unit weights can be solved by solving a problem of the same kind. For the nonunit weight case, the inverse problem reduces to solving a minimum cost flow problem. (iii) If the problem P is a minimum cost flow problem, then its inverse problem under the L 1 norm and unit weights reduces to solving a unit-capacity minimum cost flow problem. For the nonunit weight case, the inverse problem reduces to solving a minimum cost flow problem. (iv) If the problem P is a minimum cost flow problem, then its inverse problem under the L norm and unit weights reduces to solving a minimum mean cycle problem. For the nonunit weight case, the inverse problem reduces to solving a minimum cost-to-time ratio cycle problem. (v) If the problem P is polynomially solvable for linear cost functions, then inverse versions of P under the L 1 and L norms are also polynomially solvable.

Efficient implementations of Dijkstra's shortest path algorithm are investigated. A new data structure, called the radix heap , is proposed for use in this algorithm. On a network with n vertices, m edges, and nonnegative integer arc costs bounded by C , a one-level form of radix heap gives a time bound for Dijkstra's algorithm of O ( m + n log C ). A two-level form of radix heap gives a bound of O ( m + n log C /log log C ). A combination of a radix heap and a previously known data structure called a Fibonacci heap gives a bound of O ( m + n a @@@@log C ). The best previously known bounds are O ( m + n log n ) using Fibonacci heaps alone and O ( m log log C ) using the priority queue structure of Van Emde Boas et al. [ 17].

This paper considers the problem of radiation therapy treatment planning for cancer patients. During radiation therapy, beams of radiation pass through a patient. This radiation kills both cancerous and normal cells, so the radiation therapy must be carefully planned to deliver a clinically prescribed dose to certain targets while sparing nearby organs and tissues. Currently, a technique called intensity modulated radiation therapy (IMRT) is considered to be the most effective radiation therapy for many forms of cancer. In IMRT, the patient is irradiated from several different directions. From each direction, one or more irregularly shaped radiation beams of uniform intensity are used to deliver the treatment. This paper deals with the problem of designing a treatment plan for IMRT that determines an optimal set of such shapes (called apertures) and their corresponding intensities. This is in contrast with established two-stage approaches where, in the first phase, each radiation beam is viewed as consisting of a set of individual beamlets, each with its own intensity. A second phase is then needed to approximate and decompose the optimal intensity profile into a set of apertures with corresponding intensities. The problem is formulated as a large-scale convex programming problem, and a column generation approach to deal with its dimensionality is developed. The associated pricing problem determines, in each iteration, one or more apertures to be added to our problem. Several variants of this pricing problem are discussed, each corresponding to a particular set of constraints that the apertures must satisfy in one or more of the currently available types of commercial IMRT equipment. Polynomial-time algorithms are presented for solving each of these variants of the pricing problem to optimality. Finally, the effectiveness of our approach is demonstrated on clinical data.