Mathieu Tanneau | Design and implementation of the interior-point solver Tulip
jump-dev ·Mathieu Tanneau, a Ph.D. student at Polytechnique Montréal, gave his talk that was accepted for JuMP-dev 2020 on the design and implementation of the interior-point solver Tulip.jl during August 2020’s JuMP monthly developer call.
The talk material is available on his Github.
Abstract
Tulip.jl is an open-source linear optimization interior-point solver written in Julia. Its main features include the handling of arbitrary numerical precision, and its ability to exploit specialized routines for structured linear algebra. We present Tulip’s regularized homogeneous interior-point algorithm, and outline some of the design choices and challenges in its implementation. In particular, we will focus on linear algebra techniques, data structures, and interfaces.