High performance linear solvers for large scale problems in multibody dynamics

Dario Mangoni, Alessandro Tasora, Iryna Yasinskaya

University of Parma


The increasing complexity of multibody dynamic systems demands for efficient linear solvers that could take advantage not only of computational resources of sequential machine, but also of parallel and distributed systems. As high performance computing is becoming an affordable commodity for engineering applications, the development of efficient sparse parallel solvers in modern multibody simulation software is becoming a relevant topic, especially for large scale problems. Altough a wide range of iterative methods has been developed and efficiently implemented, a good number of cases - such as the ones that involve interior-point methods for unilateral contacts - still calls for a greater precision and robustness. To this end direct solvers are commonly addressed as the most effective. In this paper we would give a close-up insight into the most praised direct solvers currently available (e.g. MKL Pardiso, Eigen, MUMPS, PETSc) and their application in the Chrono::Engine multibody-dynamics library.

Benchmarks and real world application examples are provided, involving both rigid and flexible bodies, in order to assess the performance of these solvers.