Searcharxiv⌕ Search

arXiv subjects

Martin Řehoř

Publications and source records attributed to Martin Řehoř.

2 recordsLinked to original sources

FEniCSx-pctools: Tools for PETSc block linear algebra preconditioning in FEniCSx

Solving partial differential equations with the finite element method leads to large linear systems of equations that must be solved. When these systems have a natural block structure due to multiple field variables, using iterative solvers with carefully designed preconditioning strategies that exploit the underlying physical structure becomes necessary for an efficient and scalable solution process. FEniCSx Preconditioning Tools (FEniCSx-pctools) is a software package that eases the specification of PETSc (Portable, Extensible Toolkit for Scientific Computation) block preconditioning strategies on linear systems assembled using the DOLFINx finite element solver of the FEniCS Project. The package automatically attaches all necessary metadata so that preconditioning strategies can be applied via PETSc's standard options database to monolithic and block assembled systems. The documented examples include a simple mixed Poisson system and more complex pressure convection-diffusion approach to preconditioning the Navier-Stokes equations. We show weak parallel scaling on a fully coupled temperature-Navier-Stokes system up to 8192 MPI (Message Passing Interface) processes, demonstrating the applicability of the approach to large-scale problems.

cs.MS↗

Colombeau algebra as a mathematical tool for investigating step load and step deformation of systems of nonlinear springs and dashpots

The response of mechanical systems composed of springs and dashpots to a step input is of eminent interest in the applications. If the system is formed by linear elements, then its response is governed by a system of linear ordinary differential equations, and the mathematical method of choice for the analysis of the response of such systems is the classical theory of distributions. However, if the system contains nonlinear elements, then the classical theory of distributions is of no use, since it is strictly limited to the linear setting. Consequently, a question arises whether it is even possible or reasonable to study the response of nonlinear systems to step inputs. The answer is positive. A mathematical theory that can handle the challenge is the so-called Colombeau algebra. Building on the abstract result by (Průša & Rajagopal 2016, Int. J. Non-Linear Mech) we show how to use the theory in the analysis of response of a simple nonlinear mass--spring--dashpot system.

math-ph↗