CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   grid independence (http://www.cfd-online.com/Forums/main/97345-grid-independence.html)

moniker February 15, 2012 07:19

grid independence
 
how do I conduct a grid independence test ,and what is the difference
between grid independence and grid convergence test?

siw February 15, 2012 09:00

All you have to do is run the same simulation on progressively finer grids (3 grids should be the minimum) by changing a global grid sizing parameter rather than local refinement, unless it takes too long to run/insufficient computer resources to see if the solution changes as you refine the grid. Hopefully, when the grid is fine enough the solution will not change (or change less than a tolerance you are happy with) so that the solution is 'grid independent'. This gives you an feel about reducing the discretization error inherent in CFD.

There's no difference between the two, it's just a name.

moniker February 15, 2012 09:09

when you say finer grids ,do you mean by changing coarse,medium and fine mesh settings?will I see the changes in the form of a graph in the results(chart viewer)
do grids and mesh mean the same ?

mettler February 15, 2012 09:16

When you do a grid-check you are making sure that the grid size is the correct value. Many times you can get a grid to converge but the answer is wrong. So, pick a parameter and double your grid size and record the parameter answer. Then, cut the grid in half and check the parameter answer. That will show you how your answer depends on grid size. This can also help in computational time, because you might find that you have a very fine mesh when one is not needed.

moniker February 15, 2012 09:26

right , will I view these changes in the form of a graph in the results ?

I set monitor points for certain parameters and viewed the results chart ,so there I tried choosing the number of nodes as my X axis ,which I could'nt as there were only mesh refinement factor ,expansion factor etc

mettler February 15, 2012 09:29

yes, as a graph. When I did this I changed the grid count until the answer changed. That shows the 'least' amount of refinement necessary for an adequate answer. So, your graph should be a horizontal line until it reaches the number of cells that are critical. With your parameter on the y-axis and grid number on the x-axis.

FMDenaro February 15, 2012 13:45

I think you can distinguish two aspects:
- a numerical solution is independent from the grid if after a refining in the grid the difference of the two solutions are (in some norm) lesser than a suitable tolerance.
- a grid independent solution can be also "wrong", that means that you did not get convergence towards the solution you really want. To define convergence is necessary that for vanishing local truncation errors, the discretization error vanishes too.


All times are GMT -4. The time now is 07:39.