Hi all, I'm sorry I'm still confusing that if I have calculated the Jacobian elements, how can I go ahead to solve the linear system with dof>1 at each node? The example code ex19 solves the linear system with Guass-Seidel method, and now I do want to use Jacobian matrix because I've found the Jacobian elements. Any ideas? Many thanks!
On Thu, Oct 2, 2014 at 1:26 PM, Sharp Stone <[email protected]> wrote: > Hi Matt and Jed, > > Thank you very much for your replies. In source code ex19, the routine > SNESSetNGS() is used, and no other options or Jacobi routines are used. How > can we change the iteration method? The routine SNESSetFromOptions(snes) > could do this for the user? If I understand correctly, petsc would use > Gauss-Seidel or Jacobi method as designated with even the same source code? > > Thank you in advance! > > On Thu, Oct 2, 2014 at 1:12 PM, Jed Brown <[email protected]> wrote: > >> Sharp Stone <[email protected]> writes: >> >> > Matt, >> > >> > Thank you very much! >> > >> > The ex19 used the Gauss-Seidel method. But I'm a little bit wonder >> whether >> > we have any examples of Jacobi method to solve such linear systems >> "(mat)A >> > dot (vec)X = (vec)B" with dof>1? Thanks in advance! >> >> ex19 does whatever you want: -pc_type jacobi and the like. >> > > > > -- > Best regards, > > Feng > -- Best regards, Feng
