code_saturne User's Forum

Dear Users,
I am actually working on validation of Code_Saturne 1.3.3 on ROCOM tests (RPV mock up).
The inlet boundary conditions (given by experiments) came from "wire mesh sensors" : 200 sensors on a pipe section, with results over several time steps.
My question is : How can I define my inlet boundary conditions considering those results ?
 
By talking with people using CFX, it is possible to give all sensors position on the pipe section "x,y", and all sensors results "R(x,y)", so that the code is able to find a polynomial F(x,y) = R to define parametric boundary conditions.
I really like to have the same approach with Code_Saturne, since the F(x,y) polynomial is easy to find (by resolving a system of linear equations). With 200 points the method should be quiet accurate (9 or 10 orders for F(x,y))
Are user's subroutine able to call the "Blas" library to resolve such a system ? Or should I write my own matrice inversion algorithm ?
 
 
If not, does anyone know a better strategy to define such a boundary condition ?
Many thanks for your help,
 
Bruno
EDF SEPTEN, Lyon
 

Hi Bruno,
After a quick look on the Blas API, it seems that this library does not propose an matrix inversion subroutine.  Nevertheless this url  http://www.dreamincode.net/code/snippet1272.htm proposes an algorithm that allows to inverse a square matrix. Perhaps could you adapt this program for your need. Of course, we can discuss if this feature should be in the next standard version.

Hi,
Thank you very much for this answer.
Considering that is not possible throught BLAS to resolve a system of equations (i'll make a check on your link), is there a way with Code_Saturne functions ?
I guess Code_Saturne has to solve system of equations (non-linear), may I use similar function in my usclim.F file for my linear system ? Can I reach these functions in user's files like usclim.F ?

I
do not think you can use the iterative solvers of Code_Saturne because they do not
reverse any of matrices, but matrices that have special properties
as they are built with the finite volume discretization. Therefore I think a better solution is to use a direct method for any matrix.
It is of course my point of view. Any agreements or disagreements are welcome. :)

Sounds clear to me, thanks for the answer :)