Use this URL to cite or link to this record in EThOS:
Title: Advances in interior point methods for large-scale linear programming
Author: Colombo, Marco
ISNI:       0000 0004 2726 3150
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2007
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This research studies two computational techniques that improve the practical performance of existing implementations of interior point methods for linear programming. Both are based on the concept of symmetric neighbourhood as the driving tool for the analysis of the good performance of some practical algorithms. The symmetric neighbourhood adds explicit upper bounds on the complementarity pairs, besides the lower bound already present in the common N−1 neighbourhood. This allows the algorithm to keep under control the spread among complementarity pairs and reduce it with the barrier parameter μ. We show that a long-step feasible algorithm based on this neighbourhood is globally convergent and converges in O(nL) iterations. The use of the symmetric neighbourhood and the recent theoretical under- standing of the behaviour of Mehrotra’s corrector direction motivate the introduction of a weighting mechanism that can be applied to any corrector direction, whether originating from Mehrotra’s predictor–corrector algorithm or as part of the multiple centrality correctors technique. Such modification in the way a correction is applied aims to ensure that any computed search direction can positively contribute to a successful iteration by increasing the overall stepsize, thus avoid- ing the case that a corrector is rejected. The usefulness of the weighting strategy is documented through complete numerical experiments on various sets of publicly available test problems. The implementation within the hopdm interior point code shows remarkable time savings for large-scale linear programming problems. The second technique develops an efficient way of constructing a starting point for structured large-scale stochastic linear programs. We generate a computation- ally viable warm-start point by solving to low accuracy a stochastic problem of much smaller dimension. The reduced problem is the deterministic equivalent program corresponding to an event tree composed of a restricted number of scenarios. The solution to the reduced problem is then expanded to the size of the problem instance, and used to initialise the interior point algorithm. We present theoretical conditions that the warm-start iterate has to satisfy in order to be successful. We implemented this technique in both the hopdm and the oops frameworks, and its performance is verified through a series of tests on problem instances coming from various stochastic programming sources.
Supervisor: Gondzio, Jacek. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Mathematics