Finding Instability

I have been running a rather large model in Steady State and not achiving convergence. I am using the default convergence criterion (Euclidian error norm 1 x10-3).  My model has dramatic topography and dramatic water table topography ranging from 600m to 2200m.  At these elevations the error criterion seems especially large so I do not want to increase the criterion.

The model is 6 slices and has several drainages simulated with Transfer boundaries.  The mesh is highly refined around the boundaries.  There is not dramatic hydraulic conductivity zonation.

I have used both PCG and SAMG  both fail to coverge but SAMG gives bizzare head results whereas PCG produces realistic heads and has a very low mass balance error.

I have since run this model in quasi steady state (long term transient) with reference observation points.  During the quasi steady state run  i noticed a few of the reference observation points oscillated several meters during the simulation.    The oscillation i believe indicates the model has instability at that location. 

I am wondering if there is a FEFLOW running file that keeps track of nodes in the model that have the greatest error during each iteration.  Or some otherway to locate problem areas within the model.

Amelia,

I do agree with you, you should not increase the convergence criterion. Actually it seems allready a little bit to rough for your model.

What about the problem class you use? Is it a confined or unconfined model? Allthough most models are unconfined in nature, it maybe helpful first to calculate a confined model. Maybe the difference could be neclectable for a regional scale.

The solving of the problem should be more easy in a confined model. So you could try first to solve it for the confined case. Analysing that solution you maybe detect some problems in your model (e.g. some inconsistend BCs ...). Also this could give you better starting heads for the unconfined model.

In using an unconfined approach, I made some good experiences with the "phreatic" option. Here you can force more stability in the solution by increasing the parameter for the "residual water depth" in the "specific options" settings. This water depth should be somehow related to your vertical discretization.

Unfortunately FEFLOW gives no control over spatial distribution of occuring errors an convergance problems. But if you check your results carefully you will find places with strange hydraulics which indicates normally also numerical problems.

The most frequent error source in models as yours are the boundary conditions. You could try to eliminate them step by step and run the model to find out which one makes the most problems.


Zebra