[petsc-users] Is it possible to have two decompositions in one system

[Danyang Su]

On 14-08-28 07:18 PM, Barry Smith wrote:
> On Aug 28, 2014, at 9:11 PM, Danyang Su <danyang.su at gmail.com> wrote:
>
>> Hi All,
>>
>> I have flow and reactive transport equations in one system.
>     What do you mean by “one system”? Are you solving the flow and reactive transport coupled or do you solve flow then transport, then flow etc.?
They are decoupled, solver flow then reactive transport then flow again...
>
>> The share the same domain decomposition. Usually solving flow equations takes much less time than reactive transport equations, say 10% vs 90%. For some extreme cases, the flow equations are not well scaled while reactive transport equations are well scaled. The performance for flow equations may get worse when use more processors. Is it possible to set different number of processors for flow and reactive transport equations? E.g, 16 processors for flow equations and 128 processors for reactive transport equations.
>    It is possible but the programming will get more complicated. I assume that you are solving them decoupled. This means that after you solve the flow you must distribute the solution over the large number of processes and then solve the transport. How are you managing the mesh? Are you using a PETSc DMDA or your own mesh management?
>
>     Barry
I use PETSc DMDA with structured mesh. I thought about solving flow 
first and distribute solution to the processors to solver reactive 
transport. This will introduce a lot of external coding work? I just 
wonder if there is any easier way to do it.
Thanks,
Danyang
>> Thanks,
>>
>> Danyang