Evaluating the performance of an Inexact Newton method with a preconditioner for dynamic building system simulation

Efficiently, robustly, and accurately solving systems of nonlinear differential algebraic equations for dynamic building system simulation is becoming more important due to the increasing need to simulate large-scale problems. The focus of this paper is an investigation of methods for solving the nonlinear algebraic equations encountered in dynamic building system simulations. Dynamic building system simulations employ a myriad of solution techniques, many of which use derivative information. As the problem size grows, so do the memory requirements, leading to challenges in solving large-scale problems. Newton-Krylov methods are a promising candidate for large-scale simulation. The performance of these methods is often improved, sometimes drastically, when employed with a preconditioning technique. Here, a cost-effective preconditioning technique is applied to a Newton-Krylov method and tested on a series of problems. To illustrate the benefits of the approach, comparison is made to the frequently employed Powell's Hybrid Method.