Is It Possible to Make Step-46 Run in Parallel?

[Lex Lee]

Hello all,

I developed my FSI code based on the FENothing framework used in step-46. Step-46 shows how to identify the interface between the fluid and solid domain. Also it shows how to set constraints on variables and how to assemble terms on the interface. I just used the same logic in my FSI code.

In step-46, the interface handling part gets access to the cells, cell->neighbours, neighbour of neighbour and global index of the face grids on the interface. However, things are much different and difficult in parallel mode, as a cell is locally owned but its neighbour may be not (ghost cells). In a ghost cell, values are only readable. Therefore, it seems impossible to set constraints / write values on index of neighbour cells. 

I have tried many ways to fix this problem but failed. Do you guys have the similiar experience on this problem? If yes, how to fix it? Any suggestion? 

I appreciate any feedback/help from you. 

Lex

[Wolfgang Bangerth]

Lex:

> I developed my FSI code based on the FENothing framework used in step-46.
> Step-46 shows how to identify the interface between the fluid and solid
> domain. Also it shows how to set constraints on variables and how to assemble
> terms on the interface. I just used the same logic in my FSI code.
>
> In step-46, the interface handling part gets access to the cells,
> cell->neighbours, neighbour of neighbour and global index of the face grids on
> the interface. However, things are much different and difficult in parallel
> mode, as a cell is locally owned but its neighbour may be not (ghost cells).
> In a ghost cell, values are only readable. Therefore, it seems impossible to
> set constraints / write values on index of neighbour cells.

No, that is not true. You record constraints in an AffineConstraints object,
and that continues to be the work in parallel as well -- see step-40, for example.

> I have tried many ways to fix this problem but failed. Do you guys have the
> similiar experience on this problem? If yes, how to fix it? Any suggestion?

You need to be specific about how you tried to fix the problem, and what the
error message (or other way of failure) is.

Best
W.

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/

[Lex Lee]

"You record constraints in an AffineConstraints object,
and that continues to be the work in parallel as well" 

Thanks a lot, Prof. Bangerth.  I may misunderstand the error messages. I will try to debug on other parts and keep this topic updated.      

Best,

Lex