Can CFD problems be solved Parallely using Finite Element Method?
Can we divide and conquer a CFD problem using FEM and process it in a cluster/grid?
I'm planning to maybe implement it on BOINC. I'm from a computer science background and not very familiar with fluid dynamics (although I've good understanding of calculus and very basic fluid dynamics equations). I've heard it is difficult to parallel compute CFD in distributed computing circles. Why is it so? |
Its not difficult to parallel processing of CFD code. Many commercial code exists with distributed and parallel processing. Even pre and post processing.
|
To your first question:
As I understand it you can use any way of solving the equations on a grid. But: In CFD, much efford is put into getting conserving equations. That is an inherent feature of the finite volume method and the reason why everyone uses it. |
It is not just a feature of the FV method, but of all integral methods. The main reason FV is used is because it can be stabilized so easily and is can deal with unstructured grids. But by far not everyone uses FV :)
|
Could you elaborate on this, cfdnewbie? I just have sparse knowledge about FEM, but I don't see how the equations can intrinsically conserve mass, like FV. Do you have any good paper about that?
|
the answer is two-fold, I guess. First of all, you are starting from the integral form of the equations (like in FV), and then you are using a globally connected ansatz: elements share nodes at the faces, so like in FV, what goes in, must come out...you must take care of that in FV by ensuring the consistency of the fluxes at the interface.
|
I am not sure that I already got it... Do you know any good lecture notes?
|
no, not really, sorry. What is still bothering you about it? Any global method (take Fourier Galerkin for example) conserves the conservation properties...
|
All times are GMT -4. The time now is 20:40. |