Dear all,

I finally had time to go back to my code. Quick problem summary: I am 
currently working on a semi-Lagrangian method for an advection-diffusion 
equation with a multiscale method that I designed for advection dominated 
processes. During runtime I must evaluate the (known) multiscale 
FE-solution object at a previous time step. 

Problem: The mesh is distributed and the object that will give me the 
evaluated data at a point p is possibly owned on another processor. The 
processor that owes the object is the same as the one owing the cell that 
surrounds p. To find the correct cell on the same processor that needs the 
information I am doing things similarly as the FEFieldFunction class to 
find the cell. But what if p is in a cell that is not locally owned? It 
could be in a ghost cell or an artificial one (also ghosts no not owe the 
data I need).

Now, I have have to manually MPI communicate such points to find out the 
MPI rank that owes the cell surrounding p as I found out with your help. 
Any way of doing this efficiently? I would like to avoid the brute force 
version: communicating all points to all ranks and search the owned cells. 
(Hint: Mostly (but not always) p will be in a ghost layer. )

Any way to get the actual MPI rank of the owner?

Any help would be much appreciated.

Best,
Konrad

On Thursday, May 14, 2020 at 4:10:33 PM UTC+2 Konrad Simon wrote:

> Thank you, Bruno. :-)
>
>
> On Tuesday, May 12, 2020 at 2:50:35 PM UTC+2, Bruno Turcksin wrote:
>
>> Konrad, 
>>
>> There is nothing out of the box. However, deal.II uses p4est which can 
>> use more that one layer of ghost cells. So you should take a look 
>> there to see how hard it is to change the code. If that's the way you 
>> want to go, I am sure that someone will be able to provide you with 
>> some guidance. 
>>
>> Best, 
>>
>> Bruno 
>>
>> Le mar. 12 mai 2020 à 06:04, Konrad Simon <[email protected]> a écrit : 
>> > 
>> > Thank you, Bruno. What is sometimes done for semi-Lagrangian methods is 
>> that one defines a halo region around the locally owned cells (like ghost 
>> cells) that contains information from previous time steps and then limits 
>> the time step so that one never leaves the halo region when asking for 
>> information from previous time steps. Is there a way in Deal.ii to control 
>> the size of the halo region. In my case I have pretty uniform grids if it 
>> helps. 
>> > 
>> > Best, 
>> > Konrad 
>> > 
>> > On Monday, May 11, 2020 at 2:49:21 PM UTC+2, Bruno Turcksin wrote: 
>> >> 
>> >> Konrad, 
>> >> 
>> >> Unfortunately, you will need to do the communication yourself. You can 
>> only evaluate the solution on cells that locally owned or on ghost cells. 
>> >> 
>> >> Best, 
>> >> 
>> >> Bruno 
>> > 
>> > -- 
>> > The deal.II project is located at http://www.dealii.org/ 
>> > For mailing list/forum options, see 
>> https://groups.google.com/d/forum/dealii?hl=en 
>> > --- 
>> > You received this message because you are subscribed to a topic in the 
>> Google Groups "deal.II User Group" group. 
>> > To unsubscribe from this topic, visit 
>> https://groups.google.com/d/topic/dealii/9li9twVxrGE/unsubscribe. 
>>
> > To unsubscribe from this group and all its topics, send an email to 
>> [email protected]. 
>>
> > To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/dealii/bfc6dd81-abb3-421c-8f6c-539c234b624f%40googlegroups.com.
>>  
>>
>>
>

-- 
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see 
https://groups.google.com/d/forum/dealii?hl=en
--- 
You received this message because you are subscribed to the Google Groups 
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/dealii/1b60b7a8-361a-4f2f-9216-50190ab7e950n%40googlegroups.com.

Reply via email to