Hi guys,

I'm running the following simulation: starting from an impinging turbulent jet (Re=3000), the flow later goes in a radial diffuser increasing its static pressure (Re=300). The numerical model has got an y+ around 7. By using the first order discretization scheme, and adopting the k-e High Reynolds Number with standard wall functions I've obtained a good convergence and good results. By using the first order discretization scheme, and adopting the k-e Lowh Reynolds Number with hybrid wall functions turbulence model I've obtained better convergence and better results.

To improve the numerical accuracy I've tried to use a second order discretization scheme, but the residuals reach asymptotic values and remain very high (most of all k and e). I've tried LUD, QUICK, CD and MARS with different blending factors, but things don't change. I have also tried the k-w Low Reynolds Number turbulence model and obtained the same bad results.

Furthermore, I have refined the mesh in order to have the y+ close to unity, but nothing change.

What I could do?

This is just a rough advice. It is probably more important to apply second order discretisation to velocities than using it for k-e equations. There are a few reasons for this: 1. all transport equations depends on velocity field. 2. for k-e equations, specially k, usually the dominant terms are production and dissipation, that is convection is smaller than these terms. 3. Higher order schemes are usually unbounded and can result in negative values for k and e, which due to the enforcement of positiveness, can cause convergence difficulty.