Hi
I know it was written for educational purposes only but can it deal with the sort of FEM I read about in appendices E, F and G of the Elmer Solver manual ? I mean, without actually ignoring keywords in myfile.sif that would be related to those topics.
Regards, Marc
Can ModelPDE.F90 handle higher order finite elements ?
-
- Site Admin
- Posts: 4827
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Can ModelPDE.F90 handle higher order finite elements ?
Sure, use "Element = p:n" for n:th order polynomials. Values n=2 or 3 make sense, but you could go to 10.
-Peter
-Peter
Re: Can ModelPDE.F90 handle higher order finite elements ?
Thank you but what about those keywords Stabilize and Bubbles which are described under some models in the models manual ?
Yours truly
Yours truly
-
- Site Admin
- Posts: 4827
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Can ModelPDE.F90 handle higher order finite elements ?
Hi
"Stabilize" is not available. Bubble stabilization is available. You can order additional dofs and eliminate them. Maybe something like
The number of bubbles needed varies between element types. Not really my specialty...
-Peter
"Stabilize" is not available. Bubble stabilization is available. You can order additional dofs and eliminate them. Maybe something like
Code: Select all
Element = n:1 b:3
Bubbles in Global System = False (or True)
-Peter
Re: Can ModelPDE.F90 handle higher order finite elements ?
I think there is a table expressing the DOF's with regards to the different polynomial degrees in the documentation (https://www.nic.funet.fi/pub/sci/physic ... Manual.pdf) under chapter E1.
Re: Can ModelPDE.F90 handle higher order finite elements ?
Belated thanks
What triggered my interest in ModelPDE was the message
error system diverges over maximum tolerance
in either HeatSolver or StatElecSolver whether I use mpi or not (and the faster the flow, the quicker the divergence happens).
So, using ModelPDE as a template, I was thinking of solving one linear system encompassing all plasma physics equations. It could be though that it wiil take an inordinate amount of time to solve a system of such a huge size. But would it cure my divergence problem or does it have nothing to do with it ?
Best, Marc
What triggered my interest in ModelPDE was the message
error system diverges over maximum tolerance
in either HeatSolver or StatElecSolver whether I use mpi or not (and the faster the flow, the quicker the divergence happens).
So, using ModelPDE as a template, I was thinking of solving one linear system encompassing all plasma physics equations. It could be though that it wiil take an inordinate amount of time to solve a system of such a huge size. But would it cure my divergence problem or does it have nothing to do with it ?
Best, Marc
-
- Site Admin
- Posts: 4827
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Can ModelPDE.F90 handle higher order finite elements ?
Hi Marc,
The question is where your problem arises.
Making a monolithic solver for several PDEs makes most sense when you have a strongly coupled system that does not converge easily using loosely coupled schemes. An example here could be thermoelectric phenomena where electric potential and temperature are strongly coupled but the equation itself is linear.
Monolithic approach does not make the linear systems easier to solve. Often the other way around. For that reason often for monolithic problems a good strategy for the linear system level is to use block preconditioning which basically means using block gauss-seidel as the preconditioner for the monolithic system. For example, for monolithic (Navier-)Stokes equation the block-preconditioner is a good strategy.
Also, there may be nonlinearity of the equations themselves. It would not help to resolve the nonlinearities related to Navier-Stokes that you would add some other equation to the same matrix. Still the nonlinearities need to be treated, or maybe even modeled using turbulence modeling. Generally for Navier-Stokes the monolithic approach is preferable on the elliptic regime while at higher Reynolds number various pressure or velocity correction splitting schemes dominate. Elmer solves usually the Navier-Stokes equations as monolithic system which is also why the leading application area of Elmer in flow problems is glaceology.
-Peter
The question is where your problem arises.
Making a monolithic solver for several PDEs makes most sense when you have a strongly coupled system that does not converge easily using loosely coupled schemes. An example here could be thermoelectric phenomena where electric potential and temperature are strongly coupled but the equation itself is linear.
Monolithic approach does not make the linear systems easier to solve. Often the other way around. For that reason often for monolithic problems a good strategy for the linear system level is to use block preconditioning which basically means using block gauss-seidel as the preconditioner for the monolithic system. For example, for monolithic (Navier-)Stokes equation the block-preconditioner is a good strategy.
Also, there may be nonlinearity of the equations themselves. It would not help to resolve the nonlinearities related to Navier-Stokes that you would add some other equation to the same matrix. Still the nonlinearities need to be treated, or maybe even modeled using turbulence modeling. Generally for Navier-Stokes the monolithic approach is preferable on the elliptic regime while at higher Reynolds number various pressure or velocity correction splitting schemes dominate. Elmer solves usually the Navier-Stokes equations as monolithic system which is also why the leading application area of Elmer in flow problems is glaceology.
-Peter