Re: [Libmesh-users] NonlinearImplicitSystem Jacobian

2015-09-29 Thread Harshad Sahasrabudhe
> > How far non-local and how sparse are you talking? If i and j are on > neighboring elements we have some options for automatically expanding > the Jacobian that far. If i and j might be halfway across the mesh > then support is more limited and potentially much more expensive. I basically ha

Re: [Libmesh-users] NonlinearImplicitSystem Jacobian

2015-09-29 Thread Harshad Sahasrabudhe
> > How far non-local and how sparse are you talking? If i and j are on > neighboring elements we have some options for automatically expanding > the Jacobian that far. If i and j might be halfway across the mesh > then support is more limited and potentially much more expensive. I basically ha

Re: [Libmesh-users] NonlinearImplicitSystem Jacobian

2015-09-29 Thread Roy Stogner
On Tue, 29 Sep 2015, Harshad Sahasrabudhe wrote: > I'm solving a system of equations in which I need non-local contributions > to the Jacobian, as in Ke(i,j) where i and j are dof indices which might > not be on the same element. Can this be done in NonlinearImplicitSystem? If > yes, then how is

[Libmesh-users] NonlinearImplicitSystem Jacobian

2015-09-29 Thread Harshad Sahasrabudhe
Hi, I'm solving a system of equations in which I need non-local contributions to the Jacobian, as in Ke(i,j) where i and j are dof indices which might not be on the same element. Can this be done in NonlinearImplicitSystem? If yes, then how is the matrix parallelized (number of nz, oz etc...)? Th