nCellsInCoarsestLevel

ChrisPro · January 12, 2012, 13:08

Hey All!

While running a calculation on 48processors on 8 machines i get the following error when exectuting potentialFoam:

Code:

Calculating potential flow
[0] 
[0] 
[0] --> FOAM FATAL ERROR: 
[0] No coarse levels created, either matrix too small for GAMG or nCellsInCoarsestLevel too large.
    Either choose another solver of reduce nCellsInCoarsestLevel.
[0] 
[0]     From function GAMGSolver::GAMGSolver(const word& fieldName,const lduMatrix& matrix,const FieldField<Field, scalar>& interfaceBouCoeffs,const FieldField<Field, scalar>& interfaceIntCoeffs,const lduInterfaceFieldPtrsList& interfaces,const dictionary& solverControls)
[0]     in file matrices/lduMatrix/solvers/GAMG/GAMGSolver.C at line 116.
[0] 
FOAM parallel run exiting
[0]

what is nCellsInCoarsestLevel?

i set it to 1 instead of 10 and now at least potentialFoam works. ...but simpleFoam does not konvergate. It even aborts after 3 timesteps. The case runs perfectly when calculated with only 4 processors.

but what did i change with that?

greets
Christian

akidess · January 13, 2012, 02:14

Read about multigrid methods. During the solution the grid is coarsened, and this parameter tells the algorithm when to stop. 10 is already a very low value, so I have the feeling you are wasting resources running this problem on 48 CPUs. Does your domain have less than 100 000 cells?

ChrisPro · January 21, 2012, 04:52

Oh Ok!

Thanks!! that was the problem!

Nucleophobe · October 9, 2012, 13:45

Quote:

Originally Posted by ChrisPro

Oh Ok!

Thanks!! that was the problem!

Sorry to bring up an old thread, but what did you do to resolve this? I am getting the same error. My case has some processor directories which contain no mesh due to the complexity of its shape. Is this an issue?

Edit: I realized that having empty processor directories was the issue, and remedied this by changing my how my mesh was decomposed (in decomposeParDict).

ChrisPro · October 10, 2012, 05:21

So have you solved your problem?

Either take less processors or use a finer mesh

greets
Christian

January 13, 2012, 02:14		#2
akidess Senior Member Anton Kidess Join Date: May 2009 Location: Germany Posts: 1,377 Rep Power: 29	Read about multigrid methods. During the solution the grid is coarsened, and this parameter tells the algorithm when to stop. 10 is already a very low value, so I have the feeling you are wasting resources running this problem on 48 CPUs. Does your domain have less than 100 000 cells? Zaffar likes this. __________________ On twitter @akidTwit Spend as much time formulating your questions as you expect people to spend on their answer.

January 21, 2012, 04:52		#3
ChrisPro Member Join Date: Jun 2011 Posts: 38 Rep Power: 14	Oh Ok! Thanks!! that was the problem!

October 10, 2012, 05:21		#5
ChrisPro Member Join Date: Jun 2011 Posts: 38 Rep Power: 14	So have you solved your problem? Either take less processors or use a finer mesh greets Christian