Implicit Solvers

Implicit Solvers with Parallel Version Compatibility Table

  Option Radioss SMP Radioss MPP SPMD
Implicit Resolution /IMPL/SOLVER PCG
Buckling Modes /IMPL/BUCKL/1

with Isolv=1

Eigen Modes /EIG (Starter) SuperLU MUMPS
Iterative Preconditioned Conjugate Gradient
Boeing Solver
Massively Parallel Multi-Frontal Solver

Linear Solvers

Direct, Iterative and Mixed

Linear solver will be used in both Linear and Nonlinear Analyses, so it is very important to choose an appropriate solver for your application.

The PCG (Preconditioned Conjugate Gradient) iterative solver has been available from the first version of Radioss Implicit. Direct solvers are also available. The default solver is the PCG with Factored Approximate Inverse preconditioning method.

Choosing the appropriate solver depends on the application model. In general, an iterative solver is suited for well-conditioned models with homogeneous stiffness (for example: solid element models); but computationally more expensive for an ill-conditioned model with heterogeneous stiffness. On the other hand, direct solvers provide more accurate results and are less sensitive to matrix quality but require more memory storage. When out-of-core memory is used, the performance of direct solvers could be greatly reduced.

If you are not sure as to which solver to use for a particular application, it is recommended to try a direct solver first, provided that memory is not an issue. For large simulations, such as Full-Vehicle Analysis, where memory might be an issue, the PCG method with higher quality preconditioner (this is set using /IMPL/PREPAT/n, for example: n=2) could be used instead.

For a Nonlinear Analysis, it is worth comparing the two methods on your model by running a simple Linear Analysis before launching the actual analysis.

Mixed solvers may provide better performance for simulations with contact where the contact stiffness DOF is much lower than the total DOF.

Nonlinear Solvers

Modified Newton and Quasi-Newton Methods

Once again, the choice of solver depends on the type of analysis. Generally, the Quasi-Newton method is more suitable for an analysis with a high degree of nonlinearity, but it requires more memory and costs more per iteration.