Hello All,

I am writing my masters in Geophysics and am investigating whether Julia is 
ready to perform the kind of task my thesis will require.  After reading 
the section on parallel computations, I am a bit weary to dive in and 
attempt a Julia implementation.

Currently I am trying to learn PETSc to perform domain decomposition.  I 
know this know this route will work, but I think the learning curve will be 
great, and in the end the implementation will be too black box.

What I would like is to be able to:

 1) Initialize and fill a distributed array consisting of the left hand 
side of a discretized PDE, and a distributed vector consisting of the right 
hand side of the same PDE 
 2) solve in parallel every other sub-domain with dirichlet conditions 
using direct method ( or others, I would like to explore options here)
 3) solve in parallel remaining sub-domains using Neumann conditions on the 
shared boundary using results from 2).
 4) Retrieve solution into global array, plot, etc...

I would love to hear if anyone has tried anything similar, or if they think 
this should be achievable based on their experience.  My main concerns are 
- flexibility of the distributed array decomposition (only over a single 
dimension?), ability to share results between distributed arrays (step 3). 
 Whether linear algebra functions are available to distributed arrays?

Thank You,

Ben


Reply via email to