Scalable Strategies for Large-scale Structured Nonlinear Optimization
A robust and efficient parallel nonlinear solver is alway a challenge for large-scale stochastic NLP problems. We discuss the details of our implementation PIPS-NLP, which is based a parallel nonlinear interior-point solver. The parallel strategy is designed by decomposing the problem structure and building the Schur Complement.
Our nonlinear interior-point algorithm adopts a filter-based line search method. In order to guarantee the global convergence of the filter algorithm, we apply a curvature test instead of the inertia test. We present the global convergence analysis and also some numerical results based on CUTEr test problems and real energy applications.