Studies & Solvers

Faster ODE Solvers and Explicit Time Solver Dormand-Prince 5 (Runge-Kutta 4/5)

The time stepping algorithms have been optimized for solving ODEs and initial value problems more efficiently. Certain problem classes can be an order of magnitude faster to solve. The new Dormand-Prince 5 solver is an explicit time-stepper with adaptive time steps, suitable for non-stiff systems of ordinary differential equations and initial value problems. It is similar to the existing Runge-Kutta 4 solver, but with a new feature that decides and adapts the size of the time step automatically. The solver has a stiffness detection algorithm and, if the problem is considered to be stiff, the solver stops and reports this to use user who can then change to a more suitable solver.

The Lorenz Attractor tutorial solves many times faster with the optimized time-stepping algorithms and the new Dormand-Prince 5 solver. The Lorenz Attractor tutorial solves many times faster with the optimized time-stepping algorithms and the new Dormand-Prince 5 solver.

The Lorenz Attractor tutorial solves many times faster with the optimized time-stepping algorithms and the new Dormand-Prince 5 solver.

Matrix-Free Domain-Decomposition Solver

A new solver option allows computing without explicitly forming the global system matrix. This can dramatically reduce memory requirements for large simulations. The new option Recompute and clear subdomain data is available for the domain decomposition solver and can be combined with virtually any of the sparse linear solvers.

Using the matrix-free domain-decomposition solver is particularly useful in simulations where a direct solver is the only option due to the structure of the problem. The domain decomposition solver works with both shared memory and distributed memory computations. For cluster computations (distributed memory), the matrix-free option is not needed since each compute node stores only the matrices for a subset of the domains. For a shared memory computer, such as a conventional workstation computer, the new matrix-free solver enables much larger simulations for the given memory with a direct solver.

The Domain Decomposition solver with its new matrix-free option. The Domain Decomposition solver with its new matrix-free option.

The Domain Decomposition solver with its new matrix-free option.

Goal-Oriented Error Estimate

For stationary and frequency studies, an accuracy tool called Goal-oriented error estimation is now available. The accuracy tool implements the dual-weighted residual method where an error estimate with respect to a given goal functional is computed. The error estimate is calculated as the sum of contributions from individual mesh elements. For each mesh element, the contribution is split over equations and is a product of a residual and dual weights. The error contributions can be visualized. The global error estimate and component-wise sums of error estimates are also available.

Enhanced Progress Monitoring

The Progress View in the COMSOL Desktop taskbar now shows the progress for all computations. For example, when a solver sequence with several study steps is run, the progress for the whole sequence of operations is shown. This enhancement also applies to geometry, mesh, and postprocessing operations. By hovering over the progress bar, a tooltip shows the current operation.