AITEC Contract Research Projects in FY1998 : Software |
Principal Investigator : | John Darlington |
Imperial College |
This software distribution represents the revised prototype of solvers for integer programming (IP) problems from specification to solution. It includes three representative systems: TIP, a logic modelling tool for IP, MGBAS, an IP solver based on Minimised Geometric Buchberger Algorithm (MGBA), and SIPS, a stochastic IP solver based on MGBA. 1. TIP TIP (Transformation of IP) is a tool on Mathematica for modelling IP with logic. It is based on our research for modelling IP with logic: language and implementation. In this research, we propose a language L+ for IP, in which we write logical specification of an IP problem. L+ is a language based on the predicate logic, but is extended with meta predicates such as at_least(m,S), where m is a non-negative integer, meaning that at least m predicates in the set S of formulas hold. The meta predicates are introduced to facilitate reasoning about a model of an IP problem rigorously and logically. L+ is executable in the sense that formulas in L+ are mechanically translated into a set of mathematical formulas, called IP formulas, which most of existing IP solvers accept. We give a systematic method for translating formulas in L+ to IP formulas. The translation is rigorously defined, verified and implemented in Mathematica 3.0. TIP consists of the functions GenIP[spec_,decl_,opts___]}, RRC[spec_], RuleE[spec_], RuleT[spec_], RuleN[spec_], RuleF[spec_] and RuleR[spec_]. RRC[spec_] (Remove Redundant Constraints) checks whether a given constraint is redundant and outputs empty set {} if redundant or else a simplified constraint after substituting the value of the bounds. RuleE[spec_] to RuleR[spec_] define the transformation rules E to R respectively. GenIP[spec_,decl_,opts___] is the main function which is called with a logical specification spec, a list of propositional variables and integer variables with their bounds decl and an option opts. If user want to simplify the generated IP-formulas, they can input IPFormSimplify->True as an option, otherwise they input nothing. TIP first checks each variable in the logic specification spec whether it is declared in decl and outputs an error message if some variables are not declared or else outputs the generated IP-formulas. Please see the whole program of TIP and some examples in detail. 2. MGBAS MGBAS is a generic symbolic IP solver based on Minimised Geometric Buchberger Algorithm (MGBA). Algorithm MGBA is the result of the joint work with Yike and Prof. John Darlington. Recently, various algebraic IP solvers have been proposed based on the theory of Groebner bases. The key idea is to encode an IP problem IP{A,C} into a special ideal associated with the constraint matrix A and the cost (object) function C. An important property of such an encoding is that its Groebner basis corresponds directly to the test set of the IP problem. The main difficulty of these new methods is the size of the Groebner bases generated. In the proposed algorithms, large Groebner bases are caused by either introducing additional variables or by considering the generic IP problem IP{A,C}. Based on these analyses, we propose a new algebraic algorithm MGBA which combines various recently developed symbolic IP solvers such as Hosten and Sturmfel's method(GRIN) and Thomas's truncated GBA to achieve a very efficient and general IP solver. MGBAS includes two parts: one is to compute the reduced Groebner basis i.e. the test set for IP problem; the other is to find a feasible solution and derive an optimal solution by using the test set to reduce the feasible solution. For any standard formulation of IP problems like as Min {Cx : Ax=b, x in N^n} if you input the objective function Cx and constraints Ax=b, you can get an optimal solution for the IP problems directly by running MGBAS. Please see the whole program of MGBAS and some examples in details. 3. SIPS SIPS is a stochastic IP solver based on MGBA. A stochastic IP problem is an IP problem with uncertain parameters describing optimisation problems with uncertainty. The applications of IP range from future productivity in a production problem to a hydro-electric power station to demands at various nodes in a transportation network. We consider a class of stochastic IP called chance constrained IP (CIP) P0: Min h(x) subject to Prob{Tx<= $\xi$}>=$\gamma$ Ax=b x in N^n, $\xi$ is a vector of random variables.\\ Model P0 has two properties: (1) The probabilistic constraint is not separable and (2) integer variables are required to model setup decisions. The available techniques for stochastic programming do not provide a satisfactory solution for models that have integer variables and nonseparable chance (probabilistic) constraints arising from nonnormal distributions. Based on MGBA, SIPS solve the above model successfully. We also applied SIPS to a job scheduling problem. Users can input some The preliminary experiments indicate that this method provides a realistic solution for the challenge of stochastic IP.
Currently the only environment necessary for running the denoted software is the computer algebra system Mathematica(tm) 3.0 from Wolfram Research. Mathematica is available for all major Unix platforms including Linux, for Windows NT/95, and for MacOS. For further information on Mathematica, see http://www.wolfram.com
The package is currently 6.3 MB in size (gzipped tar file) and structured as follows: Readme-E ... document for explaining the structure of the package doct ... this file TIP/ ... Transformation of IP README-TIP TIP.nb MGBAS/ ... Minimised Geometric Buchberger Solver for IP README-MGBA Example-MGBA mgba.c mgba SIPS/ ... Stochastic IP Solver README-SIPS Example-SIPS sips.c sips
www-admin@icot.or.jp