[PeterDoings]

Hello,
I'm trying to implement the Spalart-Allmaras model in my code, but I ran into some difficulties.
It seems the diffusion coefficient, which depends on the eddy viscosity, makes the solution process very unstable. Atm I'm using a structured grid with Finite Volume discretization and linear interpolation at the boundaries to calculate the fluxes, with GMRES to solve the system of eqs.
Will using a higher order scheme help? Is the use of flux limiters recommended? If so, which one is the best? I plan on applying the method to unstructured grids later on, by the way.

Thank you for your comments.

[cdegroot]

Higher order schemes are almost always more difficult to converge. Don't move on to higher order schemes until you can get a first order upwind scheme to converge. That being said, a flux limiter will be helpful when you are working with the higher order schemes.

Also it sounds like your problem is with diffusion so changing the advection discretization probably will not help. Maybe you need to apply under-relaxation to the eddy viscosity to smooth things out and keep things stable. That is to say, you should blend in some of the previous estimate to prevent the eddy viscosity from changing too abruptly between iterations.

[PeterDoings]

Well, I'm already using quite low values for the under-relaxation factor.
Another question: limiting the viscosity to positive values seems to help. Is this frowned upon or too unrealistic to be used in practice?

[SergeAS]

Take a look http://turbmodels.larc.nasa.gov/spalart.html especially notes.
For test your implementation take a look http://turbmodels.larc.nasa.gov/flatplate_sa.html
http://turbmodels.larc.nasa.gov/shear_sa.html http://turbmodels.larc.nasa.gov/bump_sa.html http://turbmodels.larc.nasa.gov/bump3d_sa.html