From h to p efficiently: optimal implementation strategies for explicit time‐dependent problems using the spectral/hp element method

Abstract

We investigate the relative performance of a second-order Adams–Bashforth scheme and second-order and fourth-order Runge–Kutta schemes when time stepping a 2D linear advection problem discretised using a spectral/hp element technique for a range of different mesh sizes and polynomial orders. Numerical experiments explore the effects of short (two wavelengths) and long (32 wavelengths) time integration for sets of uniform and non-uniform meshes. The choice of time-integration scheme and discretisation together fixes a CFL limit that imposes a restriction on the maximum time step, which can be taken to ensure numerical stability. The number of steps, together with the order of the scheme, affects not only the runtime but also the accuracy of the solution. Through numerical experiments, we systematically highlight the relative effects of spatial resolution and choice of time integration on performance and provide general guidelines on how best to achieve the minimal execution time in order to obtain a prescribed solution accuracy. The significant role played by higher polynomial orders in reducing CPU time while preserving accuracy becomes more evident, especially for uniform meshes, compared with what has been typically considered when studying this type of problem.© 2014. The Authors. International Journal for Numerical Methods in Fluids published by John Wiley & Sons, Ltd.

A benchmark study of numerical schemes for one‐dimensional arterial blood flow modelling

“…Time integration is performed by means of an explicit second‐order Adams–Bashforth method. The Courant–Friedrichs–Lewy (CFL) limit, which imposes a restriction on the maximum time step, is fixed by the time‐integration scheme and discretisation . A detailed description of this scheme is given in , and a description of the spectral/ hp element method in .…”

On Memory Traffic and Optimisations for Low-order Finite Element Assembly Algorithms on Multi-core CPUs

“…A lot of work on optimising inite element assembly has focused on reducing the number of loating point operations that are required to compute element matrices, for instance, using a technique called sum factorisation [Bolis et al 2014;Cantwell et al 2011;Kronbichler and Kormann 2012;Vos et al 2010], which exploits tensor-product structure that is particularly relevant for quadrilateral and hexahedral meshes and higher-order polynomial spaces. Sometimes, a fully assembled global matrix itself is not needed, but it is suicient to evaluate the product of such a matrix with a vector.…”

Spectral/hp element methods: Recent developments, applications, and perspectives

“…A number of studies [46,47,48,49] have concluded that for higher polynomial orders, one should instead scatter the global degrees of freedom back onto their local elemental representation and perform the action of the operator in an element-wise fashion using a matrix-vector operation. This enables a more compact representation of the operator in memory, greater cache-locality in applying the operator and a reduction in overall memory usage.…”