On 02/01/14 13:19, Karsten Ahnert wrote:
On 02/01/2014 03:47 PM, Utsav Tiwary wrote:
Does it involve the use of parallel programming ? If not, is doing it a good idea ?
There was a GSOC last year about parallelization of odeint with MPI and OpenMP [1]. I think it was successful, although the code has not yet been merged into the odeint trunk. Parallelization with GPUs already works with the help of Thrust, VexCL or ViennaCL.
odeint is a library devoted to solve the initial value problem of ODEs. I think building shooting methods for boundary value problems on top of odeint would be a good idea. And we also would like to mentor a GSOC project about implicit methods in odeint. There are extremely important to solve PDEs.
Back in June, 2011, petros posted a question to the user's list about multi-array and pdes: http://thread.gmane.org/gmane.comp.lib.boost.user/68017 Later, he posted a question about rotating the arrays axes: http://article.gmane.org/gmane.comp.lib.boost.user/68718/match=multi_array+r... to help solve a pde using ADI (Alternating Direction Implicit): http://article.gmane.org/gmane.comp.lib.boost.user/71096 In response, I prototyped a method using ADI and mentioned it here: http://article.gmane.org/gmane.comp.lib.boost.user/71047 HTH. -regards, Larry