Hi. I'm simulating laminar flow in a long and thin rectangular duct with two large 90 degree inlet and outlet: http://img99.imageshack.us/my.php?image=ductmz8.jpg

When I use the QUICK or MUSCL discretization (with SIMPLEC), I have trouble converging and there are also reversed flow at the outlet. These problems disappear if I use the first order upwind. Since the first order solution is probably not accurate, i wonder if those reversed flow are "real". If not, how can I deal with it. I have already lengthened the outlet and can't extend it further because of lack of computing power.

In addition, should I manipulate the under relaxation factors or multigrid for better convergence? I judged convergence from the area-weighted z velocity and static pressure at the outlet.

"Double vs single precision:

If your geometry has features of very disparate length scales (e.g., a very long, thin pipe), single-precision calculations may not be adequate to represent the node coordinates.

If your geometry involves multiple enclosures connected via small-diameter pipes (e.g., automotive manifolds), mean pressure levels in all but one of the zones can be quite large (since you can set only one global reference pressure location). Double-precision calculations may therefore be necessary to resolve the pressure differences that drive the flow, since these will typically be much smaller than the pressure levels."

Well, the simulation was done in double precision...

Hi

I have encountered such a problem before when combining QUICK with SIMPLEC. Try 2nd order instead, the results will be very close to QUICK. The reversed flow at the outlet depends on the velocity of the flow and the distance from the neck to the outlet (among others).