Modal Analysis->Iterative Method?

[Supituki]

Hi everyone,

I've done recently some linear static analyses of a model with 532301 tetrahedral elements of 1st order.

It took Elmer about 15minutes to finish each analysis using the following parameters in the solver:
Solver
Optimize Bandwidth = True
Steady State Convergence Tolerance = 1.0e-7
Linear System Solver = Iterative
Linear System Max Iterations = 10000
Linear System Iterative Method = BiCGstabl
BiCGstabl polynomial degree = 3
Linear System Convergence Tolerance = 5.0e-7
Linear System Preconditioning = ILUT
Linear System ILUT Tolerance = 1.0e-3
Linear System Abort Not Converged = False
Linear System Residual Output = 1
Linear System Precondition Recompute = 1
Solver Timing Cumulative = True
Linear System Timing Cumulative = True
End

Afterwards I wanted to perform a modal analysis of the first 5 eigenvalues, but it took ages to finish. I have tried different iterative methods with different preconditioners, but it always takes too long.

Could someone give me a hint of how could i reduce times in this analysis?
I could still reduce the size of the mesh but i would like to try as it is first. Besides I could reduce the convergence tolerance.

Thanks in advance

Supituki

[raback]

Hi

Have you tried parallel computation? Unfortunately there the problem might be that the problem does not converge at all. The ILU preconditioners are somewhat problematic since they are done only partitionwise and therefore are not as good in parallel. The preconditioner "none" (which is actually diagonal preconditioner when the scaling is active) is exactly the same in parallel, so you could first test this in serial and then go parallel if successfull.

There are also some developments in block preconditioner techniques that may be more robust, are exactly the same in parallel, but are maybe currently not yet ideal. You could still try, there is some example in tests/BlockSolver*, if I remember correctly.

And then there is MUMPS that is the last resort if nothing else works in parallel. It may require too much memory though. Depends on your machine.

-Peter