Re: [eigen] question on mapping of fields to equation systems |

[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]

*To*: Eigen development <eigen@xxxxxxxxxxxxxxxxxxx>
*Subject*: Re: [eigen] question on mapping of fields to equation systems
*From*: Rhys Ulerich <rhys.ulerich@xxxxxxxxx>
*Date*: Wed, 19 Feb 2014 08:28:01 -0600
*Dkim-signature*: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :content-type:content-transfer-encoding; bh=sdy3tl6UgkQUUZCqAtOBLb9lwOMCTnaPk/b6LDH3DSI=; b=Opmpz3VmTSww06O9KPsw4FsCNU2Nf7SJbxGM/12yjYYBwyfgI0sjUasvPiBrSXwCad Uo6BEIbhzgl9n5zYxEOi6Y4hamv6aGwLQSqCFqEmhJZMLIUROhY8H3H9kyIghA+Sy8Yt znwCokbpVscKD5IOTiaHdxEd4qVRSn0LbOnHFOghnWGu4LFlYwtCuqyfr6tlrO30jGmI wgr1zmbhwo8xyP1VG4vKRk7+ovbXwtkWeFUuUbZI5SrN5h3K4zORlzoZjrsQ64TmQgNl DLJeAwvCnfy4w+is9nthFXI8bE3Y3u2pc3CHUlauFWsyOq1l+KaPHSiFPOQ2kOwqrCKT gMlw==

> I also tried having a
> separate vectors for each quantity, but this got really messy very quickly.
> This leads me to the problem, that for every step, I have to calculate the
> respective matrix entries from them, then solve and copy them back. While
> these operation is O(n) while solving is mostly O(n²) or even O(n³), it is
> often a ton actual coding effort required.
I am stuck with this on account of some parallel FFT infrastructure,
and I agree that the packing/unpacking to perform linear algebra is a
PITA.
> Are there any high-level c++ libraries which are doing just this abstraction
> level, but not directly are full-featured solvers? Which provide a
> convenient data structures to have a bunch of point/elements in space and
> allow you so simplify the mapping to the linear equation system?
I think, but am not 100% certain, that I have seen a PETSc
presentation discussing automated APIs for accomplishing such things.
I'm not good enough with PETSc to dig out the details from their
documentation.
- Rhys