You are here
Triangle reversion in FEM
Wed, 2007-11-28 12:24 - Alexander V. Sa...
Hello,
I am trying to solve a dynamic problem in nonlinear elasticity using FEM (the FreeFem++ software). The constitutive law is compressible neo-Hookean, with viscosity terms and a time-dependent body force. The computation successfully completes even for relatively fine meshes. However, as soon as I refine the mesh beyond a certain (large) number of boundary nodes, the program reports the "triangle reversed" error at a very early stage. And, if I change the linear solver from GMRES to LU, the computation terminates because of a poorly conditioned matrix. Can anyone suggest a way of testing whether this indicates a numerical problem, or whether the model itself is flawed?
Thanks, A.S.
»
- Alexander V. Sadovsky's blog
- Log in or register to post comments
- 5476 reads
Comments
CFL?
Are you using explicit time marching? In that case, you have to reduce the time step if you refine the mesh. I think this is due to violation of the CFL condition. Does the problem still persist if you use lower time step?
No, I march implicitly,
No, I march implicitly, using the Newmark scheme and Newton's method at each time step.
Try decreasing your load increment
If you have large number of boundary nodes or a fine mesh, you would need to decrease increment for your boundary conditions, otherwise as you observed, the nodes will cross over at the initial step itself. I would suggest repeatedly halving your intial boundary conditions (Dirichlet or Neumann) until you get the problem to converge.
Well, the boundary condition
Well, the boundary condition on the boundary with many nodes is homogeneous Neumann, i.e. zero traction...