Hi all, I sent the following email about the bandwidth of FEM matrices to the mailing list on the 14th but see that it has not appeared in the archives. I presume this means that something went wrong and so am trying it again.
---------- Forwarded message ---------- From: Michael Rapson <[email protected]> Date: Fri, Oct 14, 2011 at 12:57 PM Subject: Bandwidth and scaling of FEM matrices To: dealii <[email protected]> Hi all, The example program step-22 provides a very interesting discussion of how matrix factorization scales for 2D and 3D models in the introduction. It mentions estimates for the bandwidth (O(N^1/2) and O(N^2/3) in 2D and 3D respectively) as well as computing sparse factors O(NB^2) where B is the bandwidth. I have looked at FEM books that I have on hand but these do not cover this type of detail. Chapter 7 in "Computational Science and Engineering" by Gilbert Strang does address these issues but uses finite difference examples. While the general approach he uses is the same, the specific figures he reaches are slightly different. Does anyone know where the estimates in step-22 came from, or alternatively, can anyone point me to a good reference for these type of calculations specific to FEM? The reviewers on a paper came back asking me to provide some profiling information or calculations to show scaling properties for the code I discuss. The general picture in Strang and the step-22 discussion agrees with what I am seeing but I am looking for some more specific references. Thanks for your help! Michael _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
