I thought about Eigen because from what I understand it will be a replacement for uBLAS in FEniCS. But I agree that petsc4py might be more useful. For the kernel in the script, the assembled matrix is not sparse so SparsityPattern is probably not going to matter and I will get away with C = PETSc.Mat().createDense() and C.setValues().
Thanks for feedback. Miro > On 02 Mar 2015, at 19:29, Corrado Maurini <[email protected]> wrote: > > What about using petsc4py or Matrix.set() instead of numpy? Then I think it > will be much more useful. > > I was incidentally looking to do something very similar and the documentation > on manual and efficient matrix assembling is very ‘sparse'. > For example it is difficult to understand how to use Matrix.set(…) and > SparsityPattern settings. > > Thanks in any case for sharing the code. > > Corrado > > > >> On 02 Mar 2015, at 16:37, Miroslav Kuchta <[email protected]> wrote: >> >> >>> On 02 Mar 2015, at 16:15, Garth N. Wells <[email protected]> wrote: >>> >>> On Mon, Mar 2, 2015 at 11:21 AM, Miroslav Kuchta <[email protected]> wrote: >>>> Hi, >>>> >>>> I wrote a simple script which shows how FEniCS can be used to solve >>>> integral equations. >>>> At the moment FEniCS is combined with numpy to solve the problem, so I’d >>>> be grateful for >>>> pointers on how to solve the entire problem in FEniCS. Also, are there any >>>> objections to >>>> this becoming a demo at some point? Thanks. >>>> >>> >>> uBLAS is slated to removed at some point, so it would be better if it >>> didn't depend on uBLAS. >>> >> >> Yes, I know that the days of uBLAS are numbered, so I will try to make it >> work with Eigen. >> >>> My other concern is that it's maybe more a NumPy demo than a FEniCS >>> demo - the use of FEniCS in the demo is (understandably) pretty >>> limited. >>> >>> An option would be to paste it on the Q&A forum under a question 'How >>> do I solve . . . ‘. >> >> I started looking into this because of some Q&A question but I believe that >> if posted >> there or under a question ‘How do I solve’, the number of users that might >> benefit >> from it would be much smaller than if it were a demo (or perhaps some FEniCS >> recipe). I will make a pull request when the code is ready with Eigen and >> the cpp >> version and then I’ll let you decide. >> >> Miro >> >>> >>> Garth >>> >>>> Best regards, Miro >>>> >>>> >>>> >>>> >>>> >>>> _______________________________________________ >>>> fenics-support mailing list >>>> [email protected] >>>> http://fenicsproject.org/mailman/listinfo/fenics-support >>>> >> >> _______________________________________________ >> fenics mailing list >> [email protected] >> http://fenicsproject.org/mailman/listinfo/fenics > _______________________________________________ fenics mailing list [email protected] http://fenicsproject.org/mailman/listinfo/fenics
