Hello, I'm using Plaxis 2D to perform fully coupled flow-deformation analyses. The preliminary step (which plaxis does automatically) to the fully coupled analyses is a simple steady state groundwater flow analysis to generate the steady state pore pressures. In my relatively simple models very often these preliminary analyses do not converge. As a result, I tried doing flow-only steady state analyses separately (with the same parameters), and they indeed do not converge, with global errors of the order of magnitude from 10 to 100. Now I'm breaking my head about what can cause non-convergence in simple steady state analyses. Before answering, check all that I've done to find a solution:

1. The flow is either defined by boundary conditions, or by global water levels, but both behave equally. This means that it should not be boundary condition assignment problem.
2. I have tried with different mesh sizes, with relative element size factor from 2 to 0.17 (lower than very fine), with local refinements, but no success.
3. I have also exported the geometry and copied the materials to try in different software (only steady state analysis in Geostudio), and convergence is achieved.
4. I have increased the flow tolerated error from the default 0.005 to 0.05.
5. I have performed transient flow-only analyses with the "problematic" parameters and they work (converge) just fine.

So the problem is only in the steady state flow analyses, either conducted separately, or as part of the fully coupled flow-def. analyses. I found that by changing the permeabilities of the berms, which are the most permeable material here, to much lower values, the analyses converge. But this is not an actual solution to the problem as the low permeabilites are not the realistically expected ones, and it doesn't make sense that the differences in the original permeabilities should cause convergence issues (the parameters are listed below).

What else might be causing convergence issues in simple steady state analyses in general, and in Plaxis?

The materials' permeabilities for this specific case are (for other cases they vary a little):
Berms: ky = 1e-3 m/s, kx = 1,7e-3 m/s (does not work even if the anisotropy is removed, works when both are changed to a maximum of 1e-6 m/s)
Levee body: ky = 6.3e-5 m/s, kx = 2.5e-4 m/s
Core: ky = 2.2e-9 m/s, kx = 8.6e-9 m/s
Topsoil layer (purple): ky = 4.6e-8 m/s, kx = 1.4e-7 m/s
Foundation soil (green): ky = 2.2e-5 m/s, kx = 6.5e-5 m/s
All the materials are set to use the Hypres data set parameters for the Van Genuchten model, but this should not be relevant for steady state flow.

Example of such model for this specific case is shown below. Other models vary a little by berm width, slope angles, etc. For reference, in this model the levee (blue material) is 35 m wide and 6 m high.