Improved constraint consensus methods for seeking feasibility in nonlinear programs

Authors:
Laurence Smith;John Chinneck;Victor Aitken
Affiliations:
Systems and Computer Engineering, Carleton University, Ottawa, Canada;Systems and Computer Engineering, Carleton University, Ottawa, Canada;Systems and Computer Engineering, Carleton University, Ottawa, Canada
Venue:
Computational Optimization and Applications
Year:
2013

Citing 8
Cited 0

Solving systems of nonlinear equations by means of an accelerated successive orthogonal projections method

Journal of Computational and Applied Mathematics
Historical development of the Newton-Raphson method

SIAM Review
Component averaging: An efficient iterative parallel algorithm for large and sparse unstructured problems

Parallel Computing
Parallel Optimization: Theory, Algorithms and Applications

Parallel Optimization: Theory, Algorithms and Applications
The Constraint Consensus Method for Finding Approximately Feasible Points in Nonlinear Programs

INFORMS Journal on Computing
On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming

Mathematical Programming: Series A and B
Improving solver success in reaching feasibility for sets of nonlinear constraints

Computers and Operations Research
Feasibility and Infeasibility in Optimization: Algorithms and Computational Methods

Feasibility and Infeasibility in Optimization: Algorithms and Computational Methods

Quantified Score

Hi-index	0.00

Visualization

Abstract

The Constraint Consensus method moves quickly from an initial infeasible point to a point that is close to feasibility for a set of nonlinear constraints. It is a useful first step prior to launching an expensive local solver, improving the probability that the local solver will find a solution and the speed with which it is found. The two main ingredients are the method for calculating the feasibility vector for each violated constraint (the estimated vector to the closest point that satisfies the constraint), and the method of combining the feasibility vectors into a single consensus vector that updates the current point. We propose several improvements: (i) a simple new method for calculating the consensus vector, (ii) a predictor-corrector approach to adjusting the consensus vector, (iii) an improved way of selecting the output point, and (iv) ways of selecting subsets of the constraints to operate on at a given iteration. These techniques greatly improve the performance of barrier method local solvers. Quadratic feasibility vectors are also investigated. Empirical results are given for a large set of nonlinear and nonconvex models.