A scalable hp-adaptive finite element software with applications in fiber optics
In this dissertation, we present a scalable parallel version of hp3D—a finite element (FE) software for analysis and discretization of complex three-dimensional multiphysics applications. The developed software supports hybrid MPI/OpenMP parallelism for large-scale computation on modern manycore architectures. The focus of the effort lies on the development and optimization of the parallel software infrastructure underlying all distributed computation. We discuss the challenges of designing efficient data structures for isotropic and anisotropic hp-adaptive meshes with tetrahedral, hexahedral, prismatic, and pyramidal elements supporting discretization of the exact sequence energy spaces. While the code supports standard Galerkin methods, special emphasis is given to systems arising from discretization with the discontinuous Petrov–Galerkin (DPG) method. The method guarantees discrete stability by employing locally optimal test functions, and it has a built-in error indicator which we exploit to guide mesh adaptivity. In addition to interfacing with third-party packages for various tasks, we have developed our own tools including a parallel nested dissection solver suitable for scalable FE computation of waveguide geometries. We present weak-scaling results with up to 24576 CPU cores and numerical simulations with more than one billion degrees of freedom.
The new software capabilities enable solution of challenging wave propagation problems with important applications in acoustics, elastodynamics, and electromagnetics. These applications are difficult to solve in the high-frequency regime because the FE discretization suffers from significant numerical pollution errors that increase with the wavenumber. It is critical to control these errors to obtain a stable and accurate method. We study the pollution effect for waveguide problems with more than 8000 wavelengths in the context of robust DPG FE discretizations for the time-harmonic Maxwell equations. We also discuss adaptive refinement strategies for multi-mode fiber waveguides where the propagating transverse modes must be resolved sufficiently. Our study shows the applicability of the DPG error indicator to this class of problems.
Finally, we present both modeling and computational advancements to a unique three-dimensional DPG FE model for the simulation of laser amplification in a fiber amplifier. Fiber laser amplifiers are of interest in communication technology, medical applications, military defense capabilities, and various other fields. Silica fiber amplifiers can achieve high-power operation with great efficiency. At high optical intensities, multi-mode amplifiers suffer from undesired thermal coupling effects which pose a major obstacle in power-scaling of such devices. Our nonlinear 3D vectorial model is based on the time-harmonic Maxwell equations, and it incorporates both amplification via an active dopant and thermal effects via coupling with the heat equation. The model supports co-, counter-, and bi-directional pumping configurations, as well as inhomogeneous and anisotropic material properties. The high-fidelity simulation comes at the cost of a high-order FE discretization with many degrees of freedom per wavelength. To make the computation more feasible, we have developed a novel longitudinal model rescaling, using artificial material parameters with the goal of preserving certain quantities of interest. Numerical tests demonstrate the applicability and utility of this scaled model in the simulation of an ytterbium-doped, step-index fiber amplifier that experiences laser amplification and heating.