Understanding Coupled Solvers Strategy

[xborras7]

Hej gents,

It might seem quite obvious for many but I could not find a clear explanation anywhere (Elmer Solvers Manual p.8 there is smth). Therefore, I decided to share my understanding on the Word file attached. Please correct me if I am wrong and I hope it helps.

It is for a model with 2 solvers coupled (LinearElastic + Reynolds):

Code: Select all

Equation 1 :: Active Solvers (1) = 1
Equation 2 :: Active Solvers (1) = 2

I am still unclear about:
1. Different levels of Coupling Systems. Is there any difference between:

Code: Select all

Equation 1 :: Active Solvers (1) = 1
Equation 2 :: Active Solvers (1) = 2

and

Code: Select all

Equation 1 :: Active Solvers (2) = 1 2

Highly coupled system? In COMSOL you have the option of "Fully Coupled", how you do that in Elmer? What is the difference in the solving scheme?

2. I noticed that the command "Linear System Abort Not Converged = Logical True/False" somes times does not work. So you need to check if the results are convergent manually.

RELC = relative change? I use this one.
NRM = ?
How do you do it?

It would be great to print a message at the end of the simulation with Convergence Yes/No.

[raback]

Hi Xavier,

The "Equation" or "Active Solvers" cannot really be used to make a set of equations fully coupled in the sense that they would be solved from the same equation. The "Equation" block is used to define the set of Solvers (i.e. physical equations) that certain body has. In your case you cannot have the two equations in the same body since Reynolds equation has a dimension one lower than the elasticity equation.

In principle the elasticity solver and Reynolds solver could be stronly coupled but the current solvers don't support this. The coupled system assembly can be done but this should be done with elemental subroutines using the block solver strategy. And there one should also write the cross terms. E.g. linearize the effect of displacement to pressure field in the Reynolds equation etc. So this requires a lot of coding and cannot be achieved by sif modifications alone.

The keyword "Linear system abort not converged" is by default True but note that it only affects the linear system. You can have divergence of nonlinear, or coupled system even if you can solve the linear system. There is actually also a keyword in Simulation section "Coupled system abort not converged" (I think) that is by default False but could be made True if you don't want to even look at diverged results.

A possible way to study convergence is to direct the simulation into output as in
> ElmerSolver case.sif > out.txt

And then study the various norms of the steady state system, e.g.
> grep NRM out.txt
or
> grep NRM out.txt | grep SS | grep reynolds

Also SaveScalars can be used to write convergence behaviour into a file.

-Peter

[xborras7]

Hej Peter!

Thanks a lot for you explanation. Crystal clear!

/Xavi