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.
Geometric shape order
Posted Jul 18, 2017, 7:37 a.m. EDT 6 Replies
Please login with a confirmed email address before reporting spam
Hello,
I am pretty new to comsol, so this question might be little bit basic. On the other hand, googling and using the help function didnt really help.
I tried meshing a sphere with different elements. There seem to be two options to influence the element order. One is the option "Geometry shape order", that appears when clicking on the component in the Model Builder.
The other option appears when clicking on the physical problem (in my case Heat Transfer in Solids).
Does this mean that the geometric shape function of an element is in Comsol independent from the function used to approximate the physical quantity?
Whats also confusion to me, is that when changing any of the two options, the mesh doesnt change. When changing from linear to quadratic shape, there should, in my opinion, appear mid nodes, probably located on the surface of the described solid?
Or am I wrong, and the geometric shape order is not related to the shape order of the finite element, but an option for the surface representation in comsol?
Thanks for your help!
Cheers,
Daniel
I am pretty new to comsol, so this question might be little bit basic. On the other hand, googling and using the help function didnt really help.
I tried meshing a sphere with different elements. There seem to be two options to influence the element order. One is the option "Geometry shape order", that appears when clicking on the component in the Model Builder.
The other option appears when clicking on the physical problem (in my case Heat Transfer in Solids).
Does this mean that the geometric shape function of an element is in Comsol independent from the function used to approximate the physical quantity?
Whats also confusion to me, is that when changing any of the two options, the mesh doesnt change. When changing from linear to quadratic shape, there should, in my opinion, appear mid nodes, probably located on the surface of the described solid?
Or am I wrong, and the geometric shape order is not related to the shape order of the finite element, but an option for the surface representation in comsol?
Thanks for your help!
Cheers,
Daniel
6 Replies Last Post Jul 19, 2017, 7:32 a.m. EDT
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Hello Daniel,
You are correct in your guess: in COMSOL, you can specify the orders for the geometry representation and for the solution representation ("discretization order") separately. In other words, elements are not necessarily isoparametric.
This has a bunch of benefits, including the fact that you can use different discretization orders for different physics in a single model.
When an order strictly greater than 1 is used for the solution, nodes that are not at the vertices of elements are not shown on screen, but they are there.
Best,
Jeff
You are correct in your guess: in COMSOL, you can specify the orders for the geometry representation and for the solution representation ("discretization order") separately. In other words, elements are not necessarily isoparametric.
This has a bunch of benefits, including the fact that you can use different discretization orders for different physics in a single model.
When an order strictly greater than 1 is used for the solution, nodes that are not at the vertices of elements are not shown on screen, but they are there.
Best,
Jeff
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Updated:
7 years ago
Hi,
In addition to Jeff's comment, I would like to stress that if you are going use a geometrical shape order other than 'Automatic', you have to know a bit of FE theory. Superparametric elements can cause interesting effects, like stresses under rigid body displacements in solid mechanics.
Also, with 'Automatic', there are some clever things going on under the hood in the case of multiphysics, where different shape function orders may be used for the DOFs in two or more physics interfaces. So for essentially all but research purposes, 'Automatic' is the recommended setting.
Regards,
Henrik
In addition to Jeff's comment, I would like to stress that if you are going use a geometrical shape order other than 'Automatic', you have to know a bit of FE theory. Superparametric elements can cause interesting effects, like stresses under rigid body displacements in solid mechanics.
Also, with 'Automatic', there are some clever things going on under the hood in the case of multiphysics, where different shape function orders may be used for the DOFs in two or more physics interfaces. So for essentially all but research purposes, 'Automatic' is the recommended setting.
Regards,
Henrik
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Hello,
thanks for the quick answers.
When using a geometric order greater than 1, are the nodes, that are not located at the vertices and thus not displayed located at the boundary of the discretized body?
Please find attached a picture of the illustration, that appears when I use the options "automatic" or "Quadratic" in the field "Geometry shape order" and the option "Quadratic" in the field Discretization => Temperature.
I am trying to discretize a hollow sphere. Shouldn't the elements next to the boundary be a little bit more round? Otherwise there would be, as far as I know, an unneccessary geometric discretization error. Or is the curvature just not displayed?
Thanks!
Greetings
Daniel
thanks for the quick answers.
When using a geometric order greater than 1, are the nodes, that are not located at the vertices and thus not displayed located at the boundary of the discretized body?
Please find attached a picture of the illustration, that appears when I use the options "automatic" or "Quadratic" in the field "Geometry shape order" and the option "Quadratic" in the field Discretization => Temperature.
I am trying to discretize a hollow sphere. Shouldn't the elements next to the boundary be a little bit more round? Otherwise there would be, as far as I know, an unneccessary geometric discretization error. Or is the curvature just not displayed?
Thanks!
Greetings
Daniel
Attachments:
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
The curvature is not displayed.
Jeff
Jeff
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Hi,
okay, thanks.
another small question related to this topic. Does COMSOL use isoparametric elements by default, when selecting "Automatic" as Geometric Shape Order?
Thanks :)
Greetings
Daniel
okay, thanks.
another small question related to this topic. Does COMSOL use isoparametric elements by default, when selecting "Automatic" as Geometric Shape Order?
Thanks :)
Greetings
Daniel
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
another small question related to this topic. Does COMSOL use isoparametric elements by default, when selecting "Automatic" as Geometric Shape Order?
The short answer is 'Yes'.
Here is a longer one:
As I indicated above, this may not be the whole truth. If, for example, you are doing a thermal stress analysis where you are using first order for the temperatures in the Heat Transfer in Solids interface, and second order for displacements in the Solid Mechanics interface, then both types of DOFs cannot be isoparametric given the same geometry approximation. Either temperatures would be isoparametric and displacements subparametric, or displacements isoparametric and temperatures superparametric. With 'Automatic', there is some logic making these decisions based on the type of equations being solved and the shape functions used.
Also, for some formulations it is not really meaningful to talk about 'isoparametric', since different fields within the same physics interface have different discretization orders (P2+P1 etc.).
Regards,
Henrik