ErRoR StUdY

January 7, 2003, 07:37

Can somebody advice me how to do an error study in a grid convergance study !!!

January 7, 2003, 09:50

The way I do a grid convergence study is to look at the parameters in which I'm interested, in the region I'm studying, for at least three different mesh densities. For example, at an inlet where I want to know velocity and/or velocity fluctuation, I look at the individual values at grid nodes and also the mean across the inlet region for each of the mesh densities.

The error can be as simple as |solution_1-solution_2| or you can look at the percent difference from one solution to the next: (|solution_1-solution_2|/solution_1)*100. You can then plot the error. If the curve flattens, i.e. the error remains constant (and low, like nearly 0), then you can qualitatively conclude grid convergence. As far as I know, this is really all you need to demonstrate whether or not you have grid convergence.

January 8, 2003, 10:09

thanks for repy, do you mean by this

|solution_1-solution_2|

the NORM !!! & under which crateria you obtain the results....

January 8, 2003, 14:08

as in a vector norm? no, this is just an absolute value.

the criteria for obtaining the results? not sure what you mean by that. the method of taking the absolute value of the difference between two results from different iterations (grid convergence iterations in this case, of which there only a few) is very common in numerical analysis.

January 7, 2003, 07:37	ErRoR StUdY	#1
su Guest Posts: n/a	Can somebody advice me how to do an error study in a grid convergance study !!!

January 7, 2003, 09:50	Re: grid convergence	#2
Aaron Guest Posts: n/a	The way I do a grid convergence study is to look at the parameters in which I'm interested, in the region I'm studying, for at least three different mesh densities. For example, at an inlet where I want to know velocity and/or velocity fluctuation, I look at the individual values at grid nodes and also the mean across the inlet region for each of the mesh densities. The error can be as simple as \|solution_1-solution_2\| or you can look at the percent difference from one solution to the next: (\|solution_1-solution_2\|/solution_1)*100. You can then plot the error. If the curve flattens, i.e. the error remains constant (and low, like nearly 0), then you can qualitatively conclude grid convergence. As far as I know, this is really all you need to demonstrate whether or not you have grid convergence.

January 8, 2003, 10:09	Re: grid convergence	#3
su Guest Posts: n/a	thanks for repy, do you mean by this \|solution_1-solution_2\| the NORM !!! & under which crateria you obtain the results....

January 8, 2003, 14:08	Re: grid convergence	#4
Aaron Guest Posts: n/a	as in a vector norm? no, this is just an absolute value. the criteria for obtaining the results? not sure what you mean by that. the method of taking the absolute value of the difference between two results from different iterations (grid convergence iterations in this case, of which there only a few) is very common in numerical analysis.