Convergence problem _ steady state flow only model

[b]Context:[/b]

I am modelling a saturated near-surface four-layer system (flow only).  Each of the four geological layers is discretised into at least two layers in FEFLOW.  The top slice is set as “free &movable” (all other slices are “fixed”) and is assigned a seepage face boundary condition.  The model gets recharge at the surface from rainfall, and the water table is expected to rise to the surface at certain locations.  I am also modelling some near-surface drains through the use of head BCs (h = z of base of drains).

[b]Problem:[/b]

I have previously run my model in steady state (unconfined) and not had any convergence issues.  I have since reduced the thickness of the top three formations and I am now having some problems with convergence.  Note that I am using the default PCG solver.

[b]Questions:[/b]

I have read about some of the convergence issues sent by users of this forum and I have a few questions.

• I could try to refine the mesh, set initial conditions closer to the results (from say a long transient run), and increase the residual water depth for phreatic elements.  Then, what else could I try?  I do not necessarily want to change the Ks of the formations.

• What does it mean if my model converges in quasi steady-state (long transient) still using the PCG solver?  That I have convergence problem?  I guess I could use the quasi steady-state final results for the steady-state run (i.e., initial heads closer to the results) and hope for convergence?

• Is it possible that the model’s complexity prevents it from achieving convergence when run in steady-state, and that the only way to successfully run the model would be in long transient mode?

• If my model converges using the SAMG solver, does it mean that the there are no instability issues (which are probably causing the model to refuse to converge using the PCG solver)?  Do I simply accept that the SAMG is the more robust solver for a complex problem and move on?

Thank you for helping with the above.

Dev,

some ideas to your questions:
[list]
[li] A practical method for obtaining better starting conditions could be to run the model in a confined mode. If this does not run, there is probably some major error in your system. In most practical cases the confined solution should not differ to much from the unconfined one. So this will give you good starting heads.[/li]
[li] If there is one thing about models: you can think of very complex and detailed models, but if they do not run they do not help anything. Especially the vertical discretization (together with a free and movable surface) has some limitations. In most cases I do work with the phreatic option. Maybe this could help you (If you have here convergance problems you could increase the residual water depth in the options settings)[/li]
[li] According to your question about solvers: I would just accept, that each solver has a limited convergance field. So some problems will be solved better with PCG and others better with SAMG. I am always lucky if I have found a running system. The rest of the problem is for mathematicians... [/li]
[/list]

Zebra

What is the limit up to which i can increase the residual water content?

Regards,
Manish Chopra

The larger the residual water content the more error.  In a complex model, a long transient run to equilibrium will almost always provide a better answer than a steady state simply because more the greater amount of "solver work" done and the damping effect of storage on convergence.  Running in confined mode will work as long as the model is mostly submerged; confined-mode is not designed to handle flow conditions where layers can dry out due to topography or due to water extraction.  In such cases, the best solution is to use variable saturation mode.