Abstract

The spectral/hp element method combines the geometric flexibility of traditional low-order finite element methods, with the performance advantages of pure spectral techniques. The price of these benefits is increased implementational complexity and challenges in consistently obtaining good performance across the richer space of discretisations and hardware architectures. Nektar++ is a framework designed with the goal of addressing these challenges and enabling the use of these otherwise complex methods by a range of application scientists and industrial users.

In this talk, motivated by a range of example problems, I will give an overview of the design of the framework and some of the implementation choices made. I will discuss some recent algorithmic advances in the library and our journey towards achieving scalable performance across a broad range of polynomial orders and element shapes. Finally, I will also touch on the growth of a community around the framework and our development practices which we depend on to make Nektar++ a successful software project.