Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Stopping the problem before finishing
Posted Oct 27, 2010, 6:25 p.m. EDT 3 Replies
Please login with a confirmed email address before reporting spam
Stopping the problem before finishing
Hello everybody, I have several simulations in comsol 3.5a and after several iteration it goes to a distorting curve the same as the figure I added. The solver goes to 99pecent but return back to 80-90 per and this may repeat and repeat for more than 20-30 hours even with a computer with 191GB RAM and 16 cores CPU. Will the answers be wrong if I stop the process when it is very close to the relative error? (E.g. relative error is 1e-5 but it goes to 1.5e-5 and then returns to e-4 and so on).
Hello everybody, I have several simulations in comsol 3.5a and after several iteration it goes to a distorting curve the same as the figure I added. The solver goes to 99pecent but return back to 80-90 per and this may repeat and repeat for more than 20-30 hours even with a computer with 191GB RAM and 16 cores CPU. Will the answers be wrong if I stop the process when it is very close to the relative error? (E.g. relative error is 1e-5 but it goes to 1.5e-5 and then returns to e-4 and so on).
Attachments:
3 Replies Last Post Oct 29, 2010, 1:31 a.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
obviously your model is set-up such that you cannot reach the relative precision desired. Normally if you STOP it (not cancel) youre last results are stored (you have also options to store intermediate results, the easiest is to use a parametric sweep) but you must check that the one active with your stop is really one with error estimates at the "grass level" as the solvers sometimes make some jumps around and the last might not be the "best of the last solutions"
It's probably linked to the mesh density, perhaps in some critical region that has a great influence on the results, you might try a mesh refinement, all depends on your model
--
Good luck
Ivar
obviously your model is set-up such that you cannot reach the relative precision desired. Normally if you STOP it (not cancel) youre last results are stored (you have also options to store intermediate results, the easiest is to use a parametric sweep) but you must check that the one active with your stop is really one with error estimates at the "grass level" as the solvers sometimes make some jumps around and the last might not be the "best of the last solutions"
It's probably linked to the mesh density, perhaps in some critical region that has a great influence on the results, you might try a mesh refinement, all depends on your model
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
obviously your model is set-up such that you cannot reach the relative precision desired. Normally if you STOP it (not cancel) youre last results are stored (you have also options to store intermediate results, the easiest is to use a parametric sweep) but you must check that the one active with your stop is really one with error estimates at the "grass level" as the solvers sometimes make some jumps around and the last might not be the "best of the last solutions"
It's probably linked to the mesh density, perhaps in some critical region that has a great influence on the results, you might try a mesh refinement, all depends on your model
--
Good luck
Ivar
Very good and helpful. I have two more question but since they are from the similar simulation, I want to ask in this thread:
firstly, which method do you suggest for solving nonlinear problems? secondly, what is the best method for computing accurate flux? I heard that weak constraint and reaction force operator are the two related methods but I tried for the first one and it didn't work and the second one get just a digit but I want accurate flowing of flux in the core.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
I usually use the default solver of COMSOL (apart from structural in 4.0a (not 4.1) I prefere direct, if I have enough RAM.
One thing thoug, to get a good starting point, you should clearly define as many initial values as possible and not leave them at default "0". One way is to do a static linear analysis first and to "restart" from there.
For fluxes its the weak method that is the best, for postprocessing (only ? to be checked) you have the reaction forces or reacf() operators to help you. In any case be sure your mesh is dense AND regular around the loop you use for integration
--
Good luck
Ivar
I usually use the default solver of COMSOL (apart from structural in 4.0a (not 4.1) I prefere direct, if I have enough RAM.
One thing thoug, to get a good starting point, you should clearly define as many initial values as possible and not leave them at default "0". One way is to do a static linear analysis first and to "restart" from there.
For fluxes its the weak method that is the best, for postprocessing (only ? to be checked) you have the reaction forces or reacf() operators to help you. In any case be sure your mesh is dense AND regular around the loop you use for integration
--
Good luck
Ivar