I am trying to determine whether my model is grid convergent using relative error, where finer grids are compared with the previous grid using the following formula:

(Obs(new)-Obs(old))/Obs(old)*100.

My attempts at determining a grid convergence criteria is flawed by the fact that areas of low velocity and wall shear stress are extremely sensitive to minor changes, which yield a large relative error (as values approach zero). This happens particularly in areas of recirculation and reattachment. Graphically, these values look converged, however I don't not have an exact solution to have a better comparison.

My model is 2D with a steady state condition with a Re = 980, run using the CFD-ACE+ package. Each model that I have run converges U,V and P to 1E-4.

I am basically looking to see if there is another way to compare course and fine grids, or to determine a negligible difference in these areas. Any references will be greatly appreciated.

Hi dear Haugh,

Please take a look at follwoing site, it may be helpful for u.

best

http://www.grc.nasa.gov/WWW/wind/val.../spatconv.html

Hey, Thanks alot for your response.

The article mentions the need to normalize the data of the fine mesh (f1) should this value be extremely small, otherwise the estimated fractional error would be meaningless. In general, how should this be normalized?