Recent progress in the massively parallel solution of implicit problems

L. Margetts, J.D. Arregui, Paul Mummery, Evans Ll, A. Shterenlikht, F. Levrero, P. Pankaj

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


    Explicit strategies for the solution of finite element problems, such as crash simulation, scale well on computing platforms with thousands of cores. In contrast, it is a widely held view that implicit analysis is not suited to parallel computing. The purpose of this paper is to challenge that belief. The open source parallel finite element software, ParaFEM, uses an “element by element” approach with iterative solvers to solve problems in an implicit way; such as static equilibrium, material nonlinearity, fluid flow, and free and forced vibrations. This iterative strategy leads to a program structure similar to explicit algorithms and the consequence is excellent scalability of implicit analysis on parallel platforms. The software is modularised in such a way that the parallel code is hidden away in a library of subroutines, enabling engineers and researchers who have no training in parallel programming to adapt the software for their own needs. In this paper, we report on recent progress in community-based extensions to ParaFEM including support for stochastic random fields, large strain plasticity, transient thermal analysis and the multiscale modelling of polycrystalline structures. We will report on how well each of those types of problem, analysed using massively parallel implicit solvers, scale on hardware platforms with tens of thousands of cores.
    Original languageEnglish
    Title of host publicationProceedings of the NAFEMS World Congress 2015
    Publication statusPublished - 21 Jun 2015
    EventNAFEMS World Congress - San Diego, USA
    Duration: 21 Jun 201524 Jun 2015


    ConferenceNAFEMS World Congress
    CitySan Diego, USA


    • Implicit
    • Finite element method
    • Parallel computing
    • High performance computing


    Dive into the research topics of 'Recent progress in the massively parallel solution of implicit problems'. Together they form a unique fingerprint.

    Cite this