[msrinath80]

In OF 1.3, we used the following syntax in the fvSolution file:

p AMG 1e-06 0 25;

If I were to translate the same into OF 1.4, I would expect the following syntax:

p GAMG
{
tolerance 1e-06;
relTol 0;
nCellsInCoarsestLevel 25;

};

My question is whether we need to specify nPreSweeps, nPostSweeps and nFinestSweeps? Also, can someone explain the use of mergeLevels?

Also what smoothers are available?

Many thanks!

[eugene]

mergeLevels is the number of merge operations to do between grid levels. mergeLevel 1 is equivalent to what is hard coded in AMG and is what I normally use.

agglomerator faceAreaPair; seems to work best

smoother options are: GaussSeidel, DIC and DICGausSeidel. GausSeidel seems to work best, but DICGaussSeidel might be better for more difficult problems. It is more expensive though.

nPreSweeps, nPostSweeps, nFinestSweeps: I use 0,2,2 and this should work for most problems. Feel free to experiment though.

[msrinath80]

Many thanks for that clear explanation Eugene!

[KateEisenhower]

Originally Posted by eugene

mergeLevels is the number of merge operations to do between grid levels. mergeLevel 1 is equivalent to what is hard coded in AMG and is what I normally use.

agglomerator faceAreaPair; seems to work best

smoother options are: GaussSeidel, DIC and DICGausSeidel. GausSeidel seems to work best, but DICGaussSeidel might be better for more difficult problems. It is more expensive though.

nPreSweeps, nPostSweeps, nFinestSweeps: I use 0,2,2 and this should work for most problems. Feel free to experiment though.

Hello Eugene,

can you or someone else please explain what merging means in this context?

Kind regards,

Kate

[akidess]

AFAIK merging is the coarsening of the grids. On a structured 2D grid, a standard geometric merge would make one cell out of four. If you'd specify mergeLevel 2, the solver would skip that level and make one cell out of 16. While that saves some work, typically mergeLevel 1 should give you optimal convergence.