Achieving Convergence When Solving Nonlinear Equations
Nonlinear problems are often difficult to solve. In many cases, no unique solution exists. The COMSOL Multiphysics software uses a Newton-type iterative method to solve nonlinear systems of PDEs. This solution method can be sensitive to the initial estimate of the solution. If the initial conditions are too far from the desired solution, convergence might be impossible, even though convergence might be possible from another, more suitable starting value.
You can do several things to improve the chances for finding the relevant solutions to difficult nonlinear problems:
•
Provide the best possible initial values.
•
Solve sequentially and iterate between single-physics equations; finish by solving the fully coupled multiphysics problem when you have obtained better starting guesses.
•
Ensure that the boundary conditions are consistent with the initial solution and that neighboring boundaries have compatible conditions that do not create singularities.
•
Refine the mesh in regions of steep gradients.
•
For convection-type problems, introduce artificial diffusion to improve the numerical properties. Most physics interfaces for modeling of fluid flow and chemical species transport provide artificial diffusion as part of the default settings.
•
Scaling can be an issue when one solution component is zero. In those cases, automatic scaling might not work.
•
Turn a stationary nonlinear PDE into a time-dependent problem. Making the problem time-dependent generally results in smoother convergence. By making sure to solve the time-dependent problem for a time span long enough for the solution to reach a steady state, you solve the original stationary problem.
•
Use the parametric solver and vary a material property or a PDE coefficient starting from a value that makes the equations less nonlinear to the value at which you want to compute the solution. This way you solve a series of increasingly difficult nonlinear problems. The solution of a slightly nonlinear problem that is easy to solve serves as the initial value for a more difficult nonlinear problem.
•
The residual operator can provide insight into the location and development of the algebraic residual in models with convergence issues.
•
Stabilization Techniques
•
Convergence Plots
•
Introduction to Solvers and Studies