trilinos

Trilinos is an open-source collection of modular software libraries designed for solving large-scale scientific and engineering problems on high-performance computing platforms. It emphasizes scalable linear and nonlinear solvers, discretizations for partial differential equations, optimization, and uncertainty quantification in parallel environments.

Originating at Sandia National Laboratories in the early 2000s as a collaborative effort to share scalable numerical software, Trilinos is now maintained by Sandia and a broad community of academic and industry contributors. The project is organized as a suite of packages that can be used independently or together, often through templated C++ interfaces and MPI-based parallel execution. It also provides wrappers and interfaces for other languages, including Fortran and Python (via PyTrilinos).

Key components span several layers. Teuchos provides utilities, memory management, and parameter lists; Kokkos offers performance portability across many-core architectures; Tpetra delivers templated distributed linear algebra, while Epetra serves as a legacy linear algebra package. Belos and Ifpack2 supply iterative solvers and preconditioners, and Amesos2 covers direct solvers. For discretizations and finite elements, packages such as Intrepid2 and related tools are used, along with Sacado for automatic differentiation and Stokhos for stochastic methods. Multigrid and related techniques are supported by packages like MueLu.

Trilinos is widely used in high-performance computing research and industry for simulations in fluid dynamics, structural mechanics, subsurface flow, and other large-scale applications. It aims to maximize portability across architectures, including CPUs and GPUs, and to support extensibility and interoperability among its diverse libraries.