Session: Aspects of NLP solver implementations
Chair: Andreas Waechter
Cluster: Nonlinear Optimization
Talk 1: The JuliaSmoothOptimizers Ecosystem for Numerical Linear Algebra and Optimization in Julia
Speaker: Tangi Migot
Abstract: JuliaSmoothOptimizers (JSO) is an organization that provides infrastructure and solvers for smooth and nonsmooth optimization, and numerical linear algebra in the Julia programming language. Those packages are intended to work consistently together and exploit the structure present in problems. They offer modeling facilities, widely useful known solvers, either in the form of interfaces or pure Julia implementations, but also unique methods that are the product of active research. JSO provides building blocks to quickly prototype solvers in a high-level language and implement efficient large-scale solvers. We present an overview of the organization and show how its facilities address the needs of students, instructors, modelers, users of optimization and researchers.
Talk 2: The Journey of a nonlinear expression through the Gurobi Optimizer
Speaker: Robert Luce
Abstract: The Gurobi optimizer can solve nonlinear optimization problems to local or global optimality. Gurobi expects these problems being modeled equation-based, that is, as explicit expressions for all nonlinear constraints. In this talk we trace the journey of a nonlinear expression through our solution framework. It starts on our Python based modeling API gurobipy, where such expressions are naturally integrated with the greater modeling functionality. Once transferred to the Optimizer, these expressions undergo our presolving algorithms, which may simplify and homogenize the expressions. In this form the expression becomes part of the solution process: Our nonlinear interior point algorithm will evaluate expressions to obtain residuals as given points, as well as evaluate derivatives in order to obtain data on first order optimality.
Talk 3: Implemention of the Gurobi NLP solver
Speaker: Andreas Waechter
Abstract: We present details on the implementation of a local nonlinear optimization solver in Gurobi. The method is based on a primal-dual interior-point method with line search. Numerical experiments for a large set of test problems is presented.
