***linear_solver dd_mpfeti

Description

This keyword specifies that the AMPFETI domain decomposition method is used to solve the linear system of equations involved in the global step of the Newton-Raphson algorithm. AMPFETI stands for Adaptive Multi Preconditioned FETI. This family of solver was previously associated with the keyword dd_sfeti. Both keywords are accepted but dd_mpfeti has to be preferred.

Note

This solver is distributed in a separate plugin that is the property of ONERA and SAFRAN.

The use of this solver in conjunction with one of these iterative solver : ddcgb, ddcgb_adaptive_global, ddcgb_adaptive_local may be used under authorization of ONERA and SAFRAN.

The use of this solver in conjunction with one of these iterative solver : ddbicgb, dd_orthob, may be used under authorization of ONERA.

Syntax

The following options are common with the dd_feti solver. Please refer to page  for a detailed documentation.

[ **precond type ] [ **scaling type ] [ **projector_schur type ] [ **projector_scaling type ] [ **solver_config predefined_config ] [ **local_solver local_solver ] [ **acceleration acceleration ]

**solver_config

activates some predefined configuration for preconditionning and corse space projector.

**precond

specifies the operator used during preconditioning.

**scaling

specifies the scaling used during preconditioning.

**projector_schur

specifies the operator used to build the coarse space projector.

**projector_scaling

specifies the scaling used to build the projector.

**local_solver

specifies the solver used to solve problems defined in each subdomain.

**acceleration

allows to reuse search space across Newton steps and/or iterations, it may be useful for ill conditioned nonlinear simulations. Please note that this feature is still experimental. Available acceleration names and commands are detailed in the **acceleration section.

The following command has special argument in the context of dd_mpfeti solver.

[ **iterative_solver iterative_solver ]

**iterative_solver

specifies the iterative solver used. For symmetric positive definite problems, you may consider these three values ddcgb, ddcgb_adaptive_global, ddcgb_adaptive_local. All these methods pertain to the multi preconditioned conjugate gradient family. Adaptive variants allow better performance in terms of computational time, such that it is recommended to avoid the use of ddcgb. The local adaptive method (ddcgb_adaptive_local) is particularly useful when the computation complexity (in terms of condition number) is localized and not spread over all subdomains. Available iterative solvers names and commands are detailed in the **iterative_solver section on . Default value is ddcgb_adaptive_global.

Example

***linear_solver dd_mpfeti
 **solver_config moderate
 **iterative_solver ddcgb_adaptive_local
  *adaptivity_threshold 0.01
  *precision 1.e-8
  *output_every_iter 1
  *max_iteration 100
  *collinearity_treatment lu_rank_revealing
  *double_check_collinearity
**local_solver mumps no_option