Please wait...
×

Error

Posted Tue, 22 Nov 2016 10:24:28 GMT by Hannes Räuschel PhD Student
Hello,
another question I have is: Which possibilities do I have to reduce the errors during a simulation. I know that you have to check your 'Boundary Conditions' for any conflicts and that 'Delauny criterion', 'Peclet & Courant number' and 'Condition number' are good hints as well as 'Max. interior angle of triangles'. But are there other ways to reduce it?
Thank you in advance!
Best regards,
HannesR
Posted Tue, 22 Nov 2016 23:17:02 GMT by adacovsk
Timestep size, timestepping growth, mesh refinement, the solver, the predictor-corrector scheme can make a difference with the error... There are lots of different ways to reduce the error! Probably the best thing to fix is the mesh refinement, especially in regions of high elevation gradients from my experience, and the type of boundary condition you use could also impact it.
Posted Sat, 26 Nov 2016 14:14:36 GMT by Björn Kaiser
I fully agree with all suggestions by adacovsk. Depending on the problem you are solving you could try to switch to the fully implict [b]first-order accurate (FE/BE)[/b] predictor corrector scheme. You may find this setting in [b]Simulation - Time Control[/b] of the [b]Problem Class[/b].

Apart from these suggestions you could also spatially identify where the largest error occur. The spatial distribution of the error is available as a nodal distribution. Go to the [b]Data Panel[/b] and right-click on [b]User Data[/b]. Choose [b]Add Predefined Distribution[/b].

Depending on your Problem Class one or more of the following distributions are available:
[list]
[li]error_norm_flow[/li]
[li]error_norm_mass#[b]i[/b][/li]
[li]error_norm_age#[b]i[/b][/li]
[li]error_norm_lte#[b]i[/b][/li]
[li]error_norm_epr#[b]i[/b][/li]
[li]error_norm_heat[/li]
[/list]
[b]i[/b] stands for the speciesID in case multiple species are available.

After you spatially identified the largest error, you could check local settings at this site with regard to the settings mentioned by adacovsk.
Posted Sat, 26 Nov 2016 14:17:18 GMT by Björn Kaiser
Please note that the error distributions need to be activated in the pre-processor (before you start the simulation).

You must be signed in to post in this forum.