UC San Diego

Large-scale simulations of time-dependent partial differential equations are, at present, largely reliant on massively parallel computers. As a result, the parallel scalability of numerical methods for partial differential equations is of crucial importance. In recent years, continuous space-time finite element methods have emerged as a promising technique for approximating these equations in a scalable, flexible way. In a space-time finite element method, the space and time variables of a time-dependent equation are treated as a single unified variable in higher-dimensional space. The higher-dimensional space-time domain is discretized into a collection of simplices and finite element methods may then be defined over this discretization. Parallelization is then achieved through domain decomposition techniques.

In this dissertation, we extend the theory of space-time finite element methods to a more general class of problems. We prove new theoretical results describing the stability of space-time methods applied to parabolic partial differential equations with nontrivial convection and reaction terms. In particular, we define a streamline-upwind scheme which upwinds in the direction of the space-time convection. The stabilized method is proved to be coercive with respect to an energy norm and asymptotic error bounds are derived.

This dissertation also proposes several operations for the construction and refinement of unstructured, conforming four-dimensional simplex meshes. We define a simple algorithm which takes as input any tetrahedral mesh and produces a corresponding four-dimensional, simplicial space-time mesh. Our algorithm always produces conforming triangulations and may be run concurrently for each spatial element. In addition, we describe how four-dimensional simplex elements can be bisected in order to achieve local spatiotemporal refinement.