[mavguy]

This might be a silly question, but I wrote a CFD code (compressible, unstructured mixed element in 2D and 3D) that I am having trouble converging in 3D for relatively low Mach numbers (~0.1).

I am aware that the N-S and Euler equations become stiff at low Mach numbers. How are these cases typically dealt with? Are solutions to complex 3D problems typically time stepped to a steady-state solution? This will certainly make the linear system easier to solve, but seems rather inefficient. Is seems that straight-up Newton-Krylov would be better.

I'm using GMRES to solve the linear system and have put in a lot of work to get a good preconditioner, to the point that I don't think I can do any better on the preconditioning side of things.

Any advice would be great!

[FMDenaro]

Originally Posted by mavguy

This might be a silly question, but I wrote a CFD code (compressible, unstructured mixed element in 2D and 3D) that I am having trouble converging in 3D for relatively low Mach numbers (~0.1).

I am aware that the N-S and Euler equations become stiff at low Mach numbers. How are these cases typically dealt with? Are solutions to complex 3D problems typically time stepped to a steady-state solution? This will certainly make the linear system easier to solve, but seems rather inefficient. Is seems that straight-up Newton-Krylov would be better.

I'm using GMRES to solve the linear system and have put in a lot of work to get a good preconditioner, to the point that I don't think I can do any better on the preconditioning side of things.

Any advice would be great!

at low Mach you can use pre-conditioning techniques or, if accettaple, using the incompressible model. If the steady state really exists, you can reach it by using a time-integration method untile the time derivatives become smaller (in some norm) than some tolerance

[mavguy]

Originally Posted by FMDenaro

If the steady state really exists, you can reach it by using a time-integration method untile the time derivatives become smaller (in some norm) than some tolerance

Shouldn't steady-state be reachable without time-stepping? In 2D, I use Newton's method and reach steady-state just fine. Is the initial guess for Newton typically just too far away from the solution in 3D for Newton to converge?

[FMDenaro]

Again, you can directly solve the steady-state solution. Sometimes, there is not a numerical solution to the discrete system and that can be a signal that the flow is unsteady.
I suppose that for converging, the Newton method require you start from an initial guess not very far from the solution