## Fluids

#### Grid convergence index

• akban
Subscriber
Hi,nI am trying to find a suitable grid size for my simulation.nI have continued grid independence test and calculated GCI for grid-1, 2, 3.nI have performed Richardson extrapolation of the results. I find that we are in the asymptotic range -- (GCI_23/r^p GCI_12) ~1nNow, how to decide which grid size to go with ? Grid-1, 2 or 3 ?nPlease suggest.nnRegards,nAyann
• Rob
Ansys Employee
Which was the coarsest mesh that gave the same answer as the finest one? Also, how much longer did the fine mesh take to run over the mesh from my first question?.
• akban
Subscriber
f1 is the finest mesh. Now, the center line velocity come out to be-- nf1-41.2874 cm, f2- 41.2403 cm, f3- 41.1684cm, and Richardson extrapolation, F1-41.3771 cm/s.nf1 provides very big matrix, so MATLAB processing becomes very slow, f2 also gives big matrix. f3 is decent.nIt is 2D simulation, so simulation run time is not an issue. But f1 takes few hours longer.nn
• Rob
Ansys Employee
I'd run another finer mesh than f1 to confirm the change is minimal and I'd run something between f1 and f2 if you need to report results. As I don't use Matlab I'd run f1 or the next finer mesh as I can use more parallel for Fluent, you need to decide how accurate you need to be. n
• akban
Subscriber
Could you please tell me why do you suggest to run finer mesh than f1 ?nI thought f2 is good enough. Even I thought f3 could be used as well. Moreover, all the values from f1, f2. f3 seems to be closed by to the Richardson extrapolation, F1. I plotted velocities at 2 different points using 4 mesh - f1, f2, f3 and f4. f4 is the coarsest. Please, values for f1, f2, f3 are very close and deviation comes in terms of 0.05 mm/s, which is very small compared to velocity in the problem (~30-50 cm/s). Please clarify.nnAlso please let me know how do I decide based on GCI values? GCI_12=0.0027 and GCI_23=0.0041nn
• DrAmine
Ansys Employee
You go the grid where the error is below certain tolerance where the result starts not deviating much. Richard Extrapolation is intended to get an extrapolated value (case of spacing=0) based on some analysis done on at least two arbitrary meshes:nnExamining Spatial (Grid) Convergence (nasa.gov)
• DrAmine
Ansys Employee
The link does also provide more info regarding GCI and the level of grid required:nnExamining Spatial (Grid) Convergence (nasa.gov)n
• akban
Subscriber
I studied this link. But could not understand how to decide which grid to go after I see that i am in the asymptotic range, my values are very close to the extrapolated value (case of spacing=0) and GCI values are very small.n
• DrAmine
Ansys Employee
Then please check with your Professor about more details here. You can rely on Required Grid Resolution to have an idea which accuracy level relates to which grid level but I still think here as engineer and if the results are starting not to deviate much (below certain tolerance) one can start using the first grid where the behavior is getting asymptotic. That is what we are doing when carrying out a grid sensitivity analysis. n
• DrAmine
Ansys Employee
Another tip: have a look into the CFX Reference Guide Chapter 6.2.1: You will love it! Consider Equation 6-27. n
• Rob
Ansys Employee
The extra couple of runs are because you have 3 data points, 2 near enough the same and one a bit different. Can you conform that f1 and f2 are actually the fully grid independent answer? n
• akban
Subscriber
Yes, I think f1, f2 are grid independent. If you look at the velocity plot for f1, f2, f3 & f4-- f1, f2 even f3 are quite close. n