Options for ILUT preconditioner implemented by Xiaodong Wang.
ilut_opts is used by Hom and SolverOpts.
Description of ILUT:
The ILU(p) fill methodology usually employed is based on stencil only and does not account for element size, though for highly-diagonally dominant systems, deep fill elements tend to be small and so this method is fast and effective for most CFD applications. ILUT stands for ILU(tau), where tau is a drop tolerance. This method is much slower than ILU(p) but according to Yousef Saad (1993), is far more effective than ILU(p) for indefinite systems or systems that are less diagonally dominant. Turning on pivoting is also generally recommended for performance, in which case the algorithm is called ILUTp. However, our implementation does not seem to work very well.
Click here to return to Jetstream Input Parameters main page.
ilut% droptol
Description: Drop tolerance - values smaller than this will be dropped. Parameter Type: real Range of Acceptable values: (0, inf) User Notes:
This parameter greatly affects performance, smaller drop tolerance will lead to higher fill levels (better conditioning, but more expensive and more data storage).
ilut% adpfator
Description: Adaptive tolerance for drop tolerance. Parameter Type: real Range of Acceptable values: (?)
ilut% lev_fil
Description: Fill level for ilut. Parameter Type: Integer Range of Acceptable values: (0, inf) User Notes:
This acts as an upper bound on the fill level.It is not the same as the fill level in iluk, it actually takes values 4-5 times larger.
ilut% bnorm
Description: Measurement for block dropping. Parameter Type: Integer Range of Acceptable values: (?)
ilut% pivoting
Description: Use pivoting? Parameter Type: logical
ilut% diagcomp
Description: ? Parameter Type: logical
ilut% rowave
Description: Use row-averaged value as a reference for ILUT? Parameter Type: Logical
ilut_opts is used by Hom and SolverOpts.
Description of ILUT:
The ILU(p) fill methodology usually employed is based on stencil only and does not account for element size, though for highly-diagonally dominant systems, deep fill elements tend to be small and so this method is fast and effective for most CFD applications. ILUT stands for ILU(tau), where tau is a drop tolerance. This method is much slower than ILU(p) but according to Yousef Saad (1993), is far more effective than ILU(p) for indefinite systems or systems that are less diagonally dominant. Turning on pivoting is also generally recommended for performance, in which case the algorithm is called ILUTp. However, our implementation does not seem to work very well.Click here to return to Jetstream Input Parameters main page.
ilut% droptol
Description: Drop tolerance - values smaller than this will be dropped.Parameter Type: real
Range of Acceptable values: (0, inf)
User Notes:
This parameter greatly affects performance, smaller drop tolerance will lead to higher fill levels (better conditioning, but more expensive and more data storage).
ilut% adpfator
Description: Adaptive tolerance for drop tolerance.Parameter Type: real
Range of Acceptable values: (?)
ilut% lev_fil
Description: Fill level for ilut.Parameter Type: Integer
Range of Acceptable values: (0, inf)
User Notes:
This acts as an upper bound on the fill level.It is not the same as the fill level in iluk, it actually takes values 4-5 times larger.
ilut% bnorm
Description: Measurement for block dropping.Parameter Type: Integer
Range of Acceptable values: (?)
ilut% pivoting
Description: Use pivoting?Parameter Type: logical
ilut% diagcomp
Description: ?Parameter Type: logical
ilut% rowave
Description: Use row-averaged value as a reference for ILUT?Parameter Type: Logical