Hello,

I'm a graduate student at the American University of Beirut.

I am trying to implement a coupled solver similar to the CFX solver for a course project . The speed up in solution is noticeable as well as the scalability with the refinement of the mesh. However for high reynolds number problem the convergence is very slow due to the convection term dominance.

The use of the false transient term relaxation technique does not solve the problem. I had to under-relax using the old technique and thus the scalability is lost.

I tried using CFX some high reynolds number problems and it seems fine.. What do u think i should do to solve my problem...

I would appreciate your reply, Thanks

Hi Ihab,

The coupled solver on it's own will only get you so far. The strenght of the CFX solver comes from solving the coupled equations with an algebraic multigrid solver.

Regards, Robin

Hello Robin,

Well u r right, but i am using algebraic multigrid solver. The problem is with high reynolds number ( for example 2000 in a pipe flow which i am considering laminar flow ) along with fine mesh (30,000 and up) only.

For smaller meshes it is fine even with the high reynolds number........

THanks in advance

It could be unsteadiness due to laminar flow. A turbulence model will add turbulent viscosity, which would stabilize the solution.

-Robin

The problem is either that the equations you are passing the linear solver are too heavily under relaxed or the the linear solver is not solving the equations you pass to it.

Just saying you are using an algebriac multigrid solver does not help. How are you forming the coarse grid equations for example? What kind of cycle do you use, how do you transfer corrections back and forth from the coarse grids to the find grid??.... The devil is in the details.