Elmer Discussion Forum

Hi

Some more information would be needed. I think that indeed Comsol uses a 2nd order elements even though when saving the .mphtxt file it looks like a linear mesh. Are we sure that it stops there. Basically when using p-elements the element order could be further increased.

If the polynomial order is the same there can be some additional numerical tricks. In Elmer we have stabilization with various methods and experimental flux corrected transport. Comsol has also something, perhaps better tricks. Maybe you can share your findings in more detail and dig into the Comsol settings what could the settings be.

-Peter

Hi,

we had a COMSOL demo version to test some time ago. AFAIR Comsol can generate a report with all the settings listed. So that could be of some help here.
Just a suggestion.

Matthias

I tried to re-run this case and learn to use Elmer for acoustic cases, but when runing the solver, the below errors occurs:

SingleSolver: Attempting to call solver
SingleSolver: Solver Equation string is: helmholtz equation

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0 ffffffffffffffff

Does anyone know what is the reason for this?

It is the error that occurs when you point to the wrong body. I downloaded your mesh and sif and the mesh was body 1 not body 2

Body 1
Target Bodies(1) = 2
Name = "Body 1"
Equation = 1
Material = 1
End

Changed to
Target Bodies(1) = 1

and it ran.

Thanks, kevinarden! It works.

I continue to recieve the message that "WARNING:: IterSolve: Numerical Error: Too many iterations were needed. ", is there anything to worry about this?

It means it has not converged to the criteria specified. It may be a good solution, but the math is not converged to the tolerances requested. It could also be a solution but not a correct one.

you specified

Linear System Max Iterations = 500
Linear System Convergence Tolerance = 1.0e-10

which is the default. You con increase the iterations and/or decrease the convergence tolerance.

Convergence can also be improved by a finer mesh and/or higher order elements.

It takes too much time to converge or is it just me_

A key factor to obtain fast convergence is the preconditioning strategy. Do you see any improvement if you set for example

Linear System Preconditioning = ILUn

with n in the keyword value replaced by a larger integer (i.e., ILU0, ILU1, ILU2, ...)?

-- Mika