Thank you! Honestly, i am a beginner, in the sense i am an ungraduated student, but i have studied structural dynamics, familiar with both methods you mentioned! I am using Rayleight method, C=aM+bK, M=massmatrix, K, stiffness matrix.
I found a way solving my problem without using any matrix-matrix multiplication, very simple manipulation of the roths equation, ie i multiply the first equation in step-23 ( in semidiscretized form ) by (1+kC) then substitute in the second and small manipulation of the code i am done! I messed up my first manipulation by using (1+kC)^-1~1-kC into second, with the complication of doing matrix-matrix of sparse matrix. So the loss in computional sense is minor. After the space discretization i do the Rayleight approximation. Anyway thank you for the response. Best Regards Johan > Hi Johan, > > I do not konw how you are familiar with FEM, if you are beginear my > comment > may helps you. > > I think that your reported issue is quit know in transient FEM analysis. > To > my knowledge there are generally 2 methods to handdle mass matrix (matrix > C > in your case): > > 1. mass lumping (lumped mass) which is notihng else diagonilization of > mass > matrix (e.g., summation of off diagonal enties to diagonal one), it is > usually more stable but not always more accuracy. main advantage is > possiblity of fully explicit time integration > > 2. Consistent Mass Matrix: which is common result if we use interpolation > of > mass matrix like field. > > > if you search by these keywords you can find sufficient information and > idea. > > > hope this helps > > B regards > > RT > > > PS: in my case (couple of years ago): i solved nonlinear heat transfer, > for > it mass lumping works supperior. > > > > > > On Sun, Oct 12, 2008 at 9:32 PM, Johan Lorentzon > <[EMAIL PROTECTED]>wrote: > >> I was not clear in previous mail the issue is that by rayleight C >> non-diagonal i get matrix-matrix multilication thatbreaks the sparsity. >> I >> am not sure what kind of approximation i should use, approximate >> matrix-matrix, or go use diagonal c matrix or do anyone of you a better >> suggestion to involve damping into the dynamical system. >> >> Best Regards >> >> Johan >> >> >> _______________________________________________ >> > _______________________________________________
