Convergence in (diffn only)coupled problems(FIDAP)
I am a chemical engineer trying to model a set of coupled equations. To ensure that I am on the safe side of the computation, I had devised some simple test problems for which I obtained the analytical solutions myself and wanted to check with the FIDAP results. TO my surprise, I found that the solutions varied as widely as the acceleration factors I have used. (a point to be noted is that the "Fully coupled solver" failed completely and diverged. I use the USRSRC subroutine and hence the segregated solver)
IS there a disciplined way to approach this problem ? The test problem is simple and since I know the analytical solution I am trying to verify the FIDAP results against the analytical solution. But, how will I be convinced that the bigger version of the test problem, which I will have to encounter soon, will have dependable results ? If anybody can thrwo some light on the usage of "SOLUTION" command in FIDAP and the "velconv" and "resconv" arguements, I will be more than grateful. Thx 
Re: Convergence in (diffn only)coupled problems(FI
(1). The only reason why the forum is here is because some users have some problems. (2). Nobody guarantee a solution when you are using a commercial code. They all provide no warranty page somewhere. (3). One way to do is to ask the vendor to provide good sample cases, and see whether you can repeat those cases. (4). The chances of getting a good solution would be higher if both of you and the vendor can obtain the same solution to a sample problem. (5). It is useless to design your test cases to check out the code. Unless the vendor happened to include your cases in the code in the code development phase. (6). There have been some discussions on the proper ways to use a commercial code in this forum. Use search to locate the old messages.

Re: Convergence in (diffn only)coupled problems(FI
The Segregated solver is less accurate than the fully coupled but it is much more stable and very sensitive to the RELAXATION FACTORS and number of GRIDS.To get a converged solution from the fully coupled solver (N.R) you should start with (S.S) solver for 5 iterations or so and then switch to N.R or Q.N.

Re: Convergence in (diffn only)coupled problems(FI
Ahmed:
thx a lot for your suggestion. I shall implement your suggestion and 'll discuss the outcome here. 
Re: Convergence in (diffn only)coupled problems(FI
There is also one more thing. You should put the VELC and RESC at least 0.0001 to get the converged solution (Segregated solver) at higher # of iterations (at least 500 or so) for the solution to be more accurate.

All times are GMT 4. The time now is 05:51. 