Π-BEM: A flexible parallel implementation for adaptive, geometry aware, and high order boundary element methods. (July 2018)

• Record Type: Journal Article
• Authors: Giuliani, Nicola; Mola, Andrea; Heltai, Luca
• Is Part Of: Advances in engineering software. Volume 121(2018)
• Page Start: 39
• Page End: 58
• Publication Date: 2018-07
• Subjects: BEM -- Fast multiple method -- High order elements -- Local refinement -- MPI -- Multi-threaded -- OpenSOURCE -- CAD; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2018.03.008
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Flexible implementation for high order, locally refined Boundary Element Methods. Exact geometry awareness using CAD files to adaptively refine the computational mesh. Parallel Implementation using hybrid distributed-shared memory techniques. OpenSOURCE software developed using best practices in scientific software and available on github atpi-BEM . Abstract: Many physical phenomena can be modelled using boundary integral equations, and discretised using the boundary element method (BEM). Such models only require the discretisation of the boundary of the domain, making the setup of the simulation straightforward and lowering the number of degrees of freedom. However, while many parallel efficient libraries are available for the Finite Element Method (FEM), the implementation of scalable BEM solvers still poses many challenges. We present the open source framework π -BEM (where π stands for parallel ): a novel boundary element method solver, combining distributed and shared memory paradigms to achieve high scalability. π -BEM exploits high performance libraries and graph partitioning tools to deliver a parallel solver employing automatic domain decomposition, high order elements, local refinement capabilities, and exact geometry-adaptivity (using CAD files). A preliminary fast multipole accelerator is included in the implementation. Every aspect of the library is modular and easily extendible by the community. We discuss the internal structure of the code, and present some examples to demonstrate the reliability and scalability of our implementation.