Hi Max,
On 22/06/2021 22:22, Max Maximilian wrote:
Hey Robert,
I've got trouble for field-variable-dependent matrix.
For example, if a stiffness matrix is strongly dependent on u and grad u, then I use the way shown in examples, call pb.evaluate(). It seems that it cannot work well when the pdes are too complex (multiphase multiple component flows).
By "cannot work well" you mean that the iterations do not converge?
Are there any other ways to calculate variable-/grad_variable dependent matrix?
There are terms for nonlinear elasticity (hyperelasticity - see e.g. dw_tl_* terms). In a similar manner, other nonlinear relations can be differentiated (analytically by hand/sympy/... - there is no automatic differentiation code included) to have consistent tangents, and then use the Newton's solver. This, however, can also have issues with convergence. Usually some step reduction/line search is required.
r.