***linear_solver dissection
#
Description#
This keyword specifies that the dissection solver is used to solve the linear system of equations involved in the global step of the Newton-Raphson algorithm. All parameters below are optional. Default values are already set and satisfactory in most cases.
Note
The use of this tool requires the Dissection interface to be compiled,
which can be checked by the presence of libZDissection
in your
library path.
Syntax#
[ **scaling
scaling_type ]
[ **ordering
ordering_type ]
[ **pivoting_threshold
threshold ]
[ **kernel_detection_all
]
[ **number_iterations
int ]
[ **minimal_nodes_per_leaf
int ]
[ **dim_aug_kern
dim for kernel detectiontype ]
[ **check_solution
]
[ **dump_operator
]
Where
**scaling
specifies the scaling used to increase the accuracy of the solution. scaling can take these three values
none
(i.e. scaling disabled),diagonal
(i.e. scaling using diagonal coefficients of the matrix) orkkt
(i.e. scaling based on infinite norm of rows and columns). Default value isdiagonal
.**ordering
specifies the renumbering used to reduce fill in. ordering can take these two values
tridiag
(i.e. Cuthill-McKee algorithm),scotch
(i.e. ordering using SCOTCH software). Default value isscotch
.**pivoting_threshold
specifies the threshold used to determine null pivots. Default value is set to \(10^{-2}\).
**kernel_detection_all
activate the full detection of kernel. Not needed for semi-definite matrices. Must be enabled for indefinite matrices. Disabled by default.
**number_iterations
specifies the number of nested dissections. It is by default set to \(-1\) (i.e. automatic choice).
**minimal_nodes_per_leaf
specifies the minimal size of dissected parts. It is by default set to \(128\).
**dim_aug_kern
specifies the size of Schur complements involved for null pivots detection. Default value is \(4\) (i.e \(2 \times 2\) Schur complements).
**check_solution
checks the residual. Only for debugging purpose.
**dump_operator
dumps the matrix. Only for debugging purpose.
Example#
***linear_solver dissection
**scaling kkt
**ordering scotch
**pivoting_threshold 1.0e-2