I have been trying to solve Stokes equation in 3D micro - geometrical meshed. I'm currently exploring the Fluid model>Creeping flow>stationary of COMSOL 4.4. Comsol failed to solve my problem. It continue showing me "Failed to find a solution. Divergence of the linear iterations. Returned solution is not converged". Going through discussion forum, someone suggested that unchecking "p" on "solver manager" in "solver paramters" would help solve the problem. Please, can anyone help me on how I can locate this solver manager in COMSOL 4.4? Or what is a suitable approach? Thank you all,
Otaru, A.J.
Hi Otaru,
Like any numerical problem, I would suggest trying to simplify the problem to obtain a "result" which you can judge. Then add more complexity as you go. I know this is quite basic advice, however it does tend to work (most of the time). Best wishes in your work.
Regards
Murat
The problem could be due to initial values. Can you solve the problem with a direct solver?
Simulate you problem with a different study group instead of stationary. For example, use frequency domain with a quite small frequency. COMSOL settings will be adjusted accordingly.
I will address the general solution divergence issue in Comsol. You can try following steps to improve the convergence.
Use structural meshes where you think you are getting the error (or use everywhere), Play with adaptive mesh refinement in solver or adapt option inside mesh node ~ Use this everytime to fine your meshes
Use the Generalized Alpha rather than BDF with strict or intermediate time stepping ~ this improves the convergence in many cases for me
Play with the tolerances (relative and absolute both) ~ sometimes if I change from factor to manual absolute tolerance (with enough time-discretization) it works.
Get the Jacobian update on every iteration than minimum ~ very important for non-linear solvers
Increase the number of maximum iterations ~ good option if you are using very fine tolerances
Change from fully coupled to segregated solvers ~ if you have many integrated physics
Use assembly in geometry and play with source to destination mesh refinements ~ when you are defining BCs or ICs inside the domains
Change scaled values of variables to manual and use values of their respective maximums ~ very helpful at the start of simulation fails.
Regards
Noman
- Badges
- Science topic