This repository contains a collection of linear systems arising from interior-point methods for quadratic optimization in MatrixMarket format. The distinguishing features of the collection are that
- systems contain accompanying right-hand sides
- systems are supplied in the form of their blocks, allowing users to solve several equivalent formulations of the same systems
- sets of related systems are supplied, generated during the iterations of an interior-point method applied to the same optimization problem
A Matlab interface to the systems is provided. From the top-level folder,
[P, K, nz, rhs] = getK(@my_assembler, problem, iter, @my_preconditioner, args...)
returns the system generated from problem
problem at interior-point iteration
rhs. The assembler
@my_assembler assembles the system
from its blocks, as read by
read_blocks(). Example assemblers are provided in
my_preconditioner should return an adequate preconditioner
together with a measure of its "complexity" (e.g., its number of nonzeros) in
nz. Additional arguments
args... are passed unchanged to
my_preconditioner. If no preconditioner is supplied,
P is set to a sparse
See the technical report below for examples.
Citing this Collection
If you use this collection in your research, please cite the following sources
- Orban D., A Collection of Linear Systems Arising from Interior-Point Methods for Quadratic Optimization, Cahier du GERAD G-2015-00, GERAD, Montreal, Canada, 2015. BibTeX.
- Orban D., A Collection of Linear Systems Arising from Interior-Point Methods for Quadratic Optimization, online data set, 2015. BibTeX.