Change preconditioning

Change preconditioning when *_linear_iterations rises sharply, a retained preconditioner changes, or an inner linear solve repeatedly fails.

Match the preconditioner to the equation

Use the unstable equation as the starting point:

These are starting points rather than guarantees. Mesh shape, material jumps, and the current flow field change the matrix as the solve progresses.

Supply ordered fallbacks

Prefer a fallback list when no single method is reliable throughout the run. For a difficult energy solve, for example:

"numerics": {
  "stable": {
    "linear_solvers": {
      "energy": {
        "right_preconditioner": ["mg32", "mg64", "ilu55"]
      }
    }
  }
}

Vanellus tries the list from left to right and retains the first successful solve. Combining both multigrid precisions with a strong ILU alternative gives the equation different recovery paths rather than repeatedly retrying one poorly suited method.

For momentum or turbulence, start with a cheaper ILU and place a stronger ILU later in the list. Override only the equation whose diagnostics indicate a problem.

After rerunning, inspect *_right_preconditioner to see which method was retained and *_linear_iterations to judge its work. A fallback that is selected frequently may deserve to move earlier in the list, but compare total runtime rather than iteration count alone.

See also

• Preconditioners and fallbacks explains every method and the fallback state machine.
• Residuals and linear-solver convergence explains the diagnostic columns.