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.
Why do I get convergence error when I increase the mesh?
Posted May 30, 2016, 10:04 a.m. EDT Low-Frequency Electromagnetics, Mesh, Studies & Solvers, Structural Mechanics Version 5.2 3 Replies
Please login with a confirmed email address before reporting spam
I solved a model with low mesh like 11. The results were not satisfactory at all but no error. Then I increased the mesh to 6000, and it stuck in 4% and after 20 minutes I got the convergence error. However, when I chose a 160 mesh it took me four days to solve, and the results are satisfactory for the same model. I did not change anything here except mesh. Why the increased number of mesh to 6000 gave convergence error? It does not make sense to me. It should be opposite of that.
Here is some more information:
Model is Time dependent (0, 0.1,7)
Four different physics are running in my model.
I used Edged mesh with distribution properties of "fixed number of elements."
I sometimes get this warning too "Minimum element quality: 1.031e-011 Warning: Low minimum element quality."
or this warning "Warning: The number of allocated threads (4) exceeds the number of available physical cores (2)"
The reciprocal of size varied from 10^5 to 900 or maybe near zero.
The complete mesh consists of 2754 domain elements and 384 boundary elements.
Here is some more information:
Model is Time dependent (0, 0.1,7)
Four different physics are running in my model.
I used Edged mesh with distribution properties of "fixed number of elements."
I sometimes get this warning too "Minimum element quality: 1.031e-011 Warning: Low minimum element quality."
or this warning "Warning: The number of allocated threads (4) exceeds the number of available physical cores (2)"
The reciprocal of size varied from 10^5 to 900 or maybe near zero.
The complete mesh consists of 2754 domain elements and 384 boundary elements.
3 Replies Last Post May 30, 2016, 5:09 p.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi
there is no direct reply to your questions, we on the FORUM would need to have more info on which physics, which type of geometry etc. But I can say a few things that puzzles me in your list of questions and info given:
Mesh is the discretization of space, used to solve your PDE, as you are in the time domain you have time and space discretization a 4 dimensional model.
But you state also "many physics (4) which each have different PDEs hence different needs for discretization and initial conditions, perhaps one mesh is good for 1-2 physics but no good for the others ...
Also some physics link the spatial and the time stepping resolution via some material parameters (i.e. heat diffusivity and space versus time discretisaton).
Perhaps by starting with a stationary study to better establish coherent initial conditions could help?
Mesh quality of 1E-11 is BAD, for me this is a sign of a problems, if this appears in the beginning I would re-mesh differently (first one need to identify where that happens, why ...) normally the alert goes around 0.01 or somewhat below, all depends ;)
If such a warning arrive during solving, its probably because you are using ALE or other morphing mesh approaches, then a re-mesh should be done well before the quality goes under 1/1000 or there around.
Fixed meshing along edges is not a problem per say, but if the mesh is not respecting coherent rules w.r.t. the PDE you are solving then you must adapt the distribution of points (there are also arithmetic and geometric distribution options, not only evenly distributed point)s. Using "seed" mesh on edges is often a good technique for complex geometries.
I do not see how what to say more here
--
Good luck
Ivar
there is no direct reply to your questions, we on the FORUM would need to have more info on which physics, which type of geometry etc. But I can say a few things that puzzles me in your list of questions and info given:
Mesh is the discretization of space, used to solve your PDE, as you are in the time domain you have time and space discretization a 4 dimensional model.
But you state also "many physics (4) which each have different PDEs hence different needs for discretization and initial conditions, perhaps one mesh is good for 1-2 physics but no good for the others ...
Also some physics link the spatial and the time stepping resolution via some material parameters (i.e. heat diffusivity and space versus time discretisaton).
Perhaps by starting with a stationary study to better establish coherent initial conditions could help?
Mesh quality of 1E-11 is BAD, for me this is a sign of a problems, if this appears in the beginning I would re-mesh differently (first one need to identify where that happens, why ...) normally the alert goes around 0.01 or somewhat below, all depends ;)
If such a warning arrive during solving, its probably because you are using ALE or other morphing mesh approaches, then a re-mesh should be done well before the quality goes under 1/1000 or there around.
Fixed meshing along edges is not a problem per say, but if the mesh is not respecting coherent rules w.r.t. the PDE you are solving then you must adapt the distribution of points (there are also arithmetic and geometric distribution options, not only evenly distributed point)s. Using "seed" mesh on edges is often a good technique for complex geometries.
I do not see how what to say more here
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Thank you, Ivar, I understand that it is hard to give help without having the model but I am not allowed to share model here even though my model is very basic and simple.
My geometry consists of simple strips of the rectangular strip of metal and polymer over each other.
What are ALE(moving mesh?) and morphing mesh? Sorry, I am a new user, and it looks this terminology is for older versions maybe?
When I solved for Stationary, I got error!! The nature of this problem is very time depended, though.
What is "element quality"? I thought it was something related to the error. Is it possible you direct me to somewhere that either I read it or you briefly explain it to me?
My geometry consists of simple strips of the rectangular strip of metal and polymer over each other.
What are ALE(moving mesh?) and morphing mesh? Sorry, I am a new user, and it looks this terminology is for older versions maybe?
When I solved for Stationary, I got error!! The nature of this problem is very time depended, though.
What is "element quality"? I thought it was something related to the error. Is it possible you direct me to somewhere that either I read it or you briefly explain it to me?
Please login with a confirmed email address before reporting spam
Posted:
9 years ago
Hi
indeed COSMOL has many thousands buttons and starting with COMSOL and then adding in 4 physics is a major challenge :) The basic rule is start with one physics at the time, then combine 2 then add another ...
Mesh quality has to do with the width to height ratio of a mesh element, an "ideal regular mesh" element enters a sphere with all modes on the surface. So a mesh quality of < 0.1 means that the tetrahedral has a height 100 times lower than its base width i.e. its almost "flat" and very anisotropic shape, so 1E-11 its thinner than a molecule.
You say you have material layers, then probably you have "thin" layers or stacks of ... then often "surface physics" should be considered. Anyhow for such layered material (if I'm right) mostly one should use structured mesh and then the aspect ratio can fall well below 0.001 and still give good results, it all depends ...
ALE is the moving mesh, yes more on that in the doc
often its helping to get reasonable initial values to run a stationary solver first and then a time dependent directly thereafter (under the same study node) again often (very dependent on your physics) the stationary case only concerns a few physics of all involved, it all depends again ,,, ;)
--
Good luck
Ivar
indeed COSMOL has many thousands buttons and starting with COMSOL and then adding in 4 physics is a major challenge :) The basic rule is start with one physics at the time, then combine 2 then add another ...
Mesh quality has to do with the width to height ratio of a mesh element, an "ideal regular mesh" element enters a sphere with all modes on the surface. So a mesh quality of < 0.1 means that the tetrahedral has a height 100 times lower than its base width i.e. its almost "flat" and very anisotropic shape, so 1E-11 its thinner than a molecule.
You say you have material layers, then probably you have "thin" layers or stacks of ... then often "surface physics" should be considered. Anyhow for such layered material (if I'm right) mostly one should use structured mesh and then the aspect ratio can fall well below 0.001 and still give good results, it all depends ...
ALE is the moving mesh, yes more on that in the doc
often its helping to get reasonable initial values to run a stationary solver first and then a time dependent directly thereafter (under the same study node) again often (very dependent on your physics) the stationary case only concerns a few physics of all involved, it all depends again ,,, ;)
--
Good luck
Ivar