Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

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.

Eliiminate inverted elements

Please login with a confirmed email address before reporting spam

Hi, I have a 3D problem with lots of terrible aspect ratios in the geometry, which seems to encourage COMSOL to create a few inverted elements.

When I run solver, it says it takes care of these by using linear elements, whatever that means.....

Could this be done at the meshing stage - to have some intermediate processing stage which checks the geometry for any mistakes before I then go and run solvers? E.g. at the moment I'm running Laplace equation, but then next it'll be advection, etc. Would be nice to get the mesh cleaned up before that.

Regards, john


9 Replies Last Post 30.07.2014, 18:13 GMT-4
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 21.02.2012, 05:59 GMT-5
Hi

it's worth to take a look at the do to better understand what it means, as the correctness of your results depend there-on, as a preléiminary summary: these warnings indicate that locally your mesh shape is so bad (so elongated) that COMSOL cannot use higher order shape elements (by default mostly quadratic elements) to represent them, A second order shape element is like if yxou add extra nodes to your element in the middle of the faces, this allows among other to extract first and possibly 2nd order derivatives too of your dependent variables, hence get a better precision on gradients that are mostly used for integration by us users ...

If you use ALE, you might get fully inverted elements, in which case the volume has turned inside out, that is worse and is to be considered as an error and severly limits your results, at least locally around these regions.

So how to avoid them: better mesh, i.e. 1) remove unvanted "samll" features from the geoemetry, then 2) use manual mesh and check the quality locally, or cut up your domains a few times to improve the overall mesh shape.

But note all cannot be done automatically. A mesh refinement can help, there are also options for doing it live during the solver, see the solver node Extension

the mesh is your "sampling" of your model, just as for digital signal sampling, if you do not have anough samples per oscillation / gradient slope, you get wrong results out, aliasing etc, even if they might look great and colorful your results are problably wrong.

But as all FEM users you do certainly a thoroughfull validation by hand calculations of your model so you should know where and wehn it's wrong, no ?

--
Good luck
Ivar
Hi it's worth to take a look at the do to better understand what it means, as the correctness of your results depend there-on, as a preléiminary summary: these warnings indicate that locally your mesh shape is so bad (so elongated) that COMSOL cannot use higher order shape elements (by default mostly quadratic elements) to represent them, A second order shape element is like if yxou add extra nodes to your element in the middle of the faces, this allows among other to extract first and possibly 2nd order derivatives too of your dependent variables, hence get a better precision on gradients that are mostly used for integration by us users ... If you use ALE, you might get fully inverted elements, in which case the volume has turned inside out, that is worse and is to be considered as an error and severly limits your results, at least locally around these regions. So how to avoid them: better mesh, i.e. 1) remove unvanted "samll" features from the geoemetry, then 2) use manual mesh and check the quality locally, or cut up your domains a few times to improve the overall mesh shape. But note all cannot be done automatically. A mesh refinement can help, there are also options for doing it live during the solver, see the solver node Extension the mesh is your "sampling" of your model, just as for digital signal sampling, if you do not have anough samples per oscillation / gradient slope, you get wrong results out, aliasing etc, even if they might look great and colorful your results are problably wrong. But as all FEM users you do certainly a thoroughfull validation by hand calculations of your model so you should know where and wehn it's wrong, no ? -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 04.03.2012, 10:33 GMT-5
Yes, this makes sense.

Thanks Ivar,
Yes, this makes sense. Thanks Ivar,

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 01.05.2012, 11:34 GMT-4
Is there a way to visualize inverted elements? That would go a long way to helping adjust geometry or meshes to eliminate them. I get coordinates on their location, but that does not always allow me to find them.
Is there a way to visualize inverted elements? That would go a long way to helping adjust geometry or meshes to eliminate them. I get coordinates on their location, but that does not always allow me to find them.

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 01.05.2012, 14:56 GMT-4
Hi

there is a variable to use to plot the mesh quality (forgotten its long name, it's somewhere in the doc ;) and inverted elements are mostly linked to poor quality elements. But I'm not aware of any other way to loclise exactly THE inverted element(s)

--
Good luck
Ivar
Hi there is a variable to use to plot the mesh quality (forgotten its long name, it's somewhere in the doc ;) and inverted elements are mostly linked to poor quality elements. But I'm not aware of any other way to loclise exactly THE inverted element(s) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 01.05.2012, 16:45 GMT-4
One thing that I have done to help "locate" items where the mesh is questionable is to write down the coordinates location when you get mesh errors, go back to the model geometry, and insert a small cylinder or similar using those coordinates. While cheesy, it does help a little with knowing where the problem is at. I would hope that maybe comsol would have a way to locate coordinates easier in future revisions but sometime you have to do it the hard way until that happens.
One thing that I have done to help "locate" items where the mesh is questionable is to write down the coordinates location when you get mesh errors, go back to the model geometry, and insert a small cylinder or similar using those coordinates. While cheesy, it does help a little with knowing where the problem is at. I would hope that maybe comsol would have a way to locate coordinates easier in future revisions but sometime you have to do it the hard way until that happens.

Jeff Hiller COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 01.05.2012, 17:05 GMT-4
See User's Guide manual, version 4.2a, pages 881-882.
See User's Guide manual, version 4.2a, pages 881-882.

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 01.05.2012, 17:30 GMT-4
Hi and thanks Jean Fraçois,

"reldetjacmin" is not that a common name ;)

I suppose it means something like "Relative_Determinant_Jacobian_minimum".
I havent used it for some time, but it worked fine back in 3.5 ;)

--
Good luck
Ivar
Hi and thanks Jean Fraçois, "reldetjacmin" is not that a common name ;) I suppose it means something like "Relative_Determinant_Jacobian_minimum". I havent used it for some time, but it worked fine back in 3.5 ;) -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 02.05.2012, 00:32 GMT-4
Thanks, Jean-Francois. I missed this completely.
Thanks, Jean-Francois. I missed this completely.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 30.07.2014, 18:13 GMT-4
Hello,

Although I have a BS in chemical engineering and did quite well in mathematics, it was a while ago. With that said, I am completely clueless with many COMSOL error messages.

Although I do not consider my geometry that "complex," it does have some fine detail[photo]. As an example, there is a small gap between the parabolic surfaces, which were created with parametric equations. Before I ask my question about 'inverted,' I would like briefly discuss some of my problems with this relatively simple geometry.

Currently, I have a support ticket regarding the strange fact that COMSOL only extrudes my center[green in photo] as a solid when I minimize the extension of the parabolic surfaces beyond the cylinder. If I increase the extension of the parabolic surfaces to a certain point, all domains in the cylinder are extruded as solids. Therefore, it appears that the extrusion seems to be dependent upon the range of my parametric variable.

As far as meshing is concerned, I am having so many problems. I have tried to fix minimum element size through custom manipulation only to be told that, as one example, the edge is locked by adjacent geometry. I have no interest making the minimum element size of the adjacent geometry near 0.014 cm. Now, "inverted" has become a problem even though it did not appear as a problem in the past. Remember, I am a newbie. Sorry.

How do I fix an "inverted" problem?
Hello, Although I have a BS in chemical engineering and did quite well in mathematics, it was a while ago. With that said, I am completely clueless with many COMSOL error messages. Although I do not consider my geometry that "complex," it does have some fine detail[photo]. As an example, there is a small gap between the parabolic surfaces, which were created with parametric equations. Before I ask my question about 'inverted,' I would like briefly discuss some of my problems with this relatively simple geometry. Currently, I have a support ticket regarding the strange fact that COMSOL only extrudes my center[green in photo] as a solid when I minimize the extension of the parabolic surfaces beyond the cylinder. If I increase the extension of the parabolic surfaces to a certain point, all domains in the cylinder are extruded as solids. Therefore, it appears that the extrusion seems to be dependent upon the range of my parametric variable. As far as meshing is concerned, I am having so many problems. I have tried to fix minimum element size through custom manipulation only to be told that, as one example, the edge is locked by adjacent geometry. I have no interest making the minimum element size of the adjacent geometry near 0.014 cm. Now, "inverted" has become a problem even though it did not appear as a problem in the past. Remember, I am a newbie. Sorry. How do I fix an "inverted" problem?

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.