Backward transport problem

Hi everyone,

I'm running a large model with steady flow and transient transport. For the transport, I'm using a reverse flow field and I'm simulating transport in backward from 7 pumping wells in order to estimate the contributing area of each well.

Feflow allow an easy convergence of flow but the transport result is an horror. There's many log messages during the run indicating error with BICGSTABD (default) over than 1 e60. Results are incoherent with positive values and negatives (-1e60; 1e60).

I'm using a consistent mass and convective form of transport.

What can cause this problem ? Good idea is welcome because I'm trying different scenarios and it takes over than 12 h for each run.

Thanks, Michael

Hi Michael,

have you tried to modify the type of upwinding (in temporal and control data)?

This is a switch I once pulled when I was facing a similar problem...

Also, there may be something in the Iterative solver settings menu, but I've never dared changing anything in this menu.

Hopefully, I'm not pointing wrong directions...

Good luck: transport convergence for big models can be painful...

Thomas

Wou ou,    :o

It works pretty well, and the number of iteration required to reach the convergence is smaller. Can you explain to me the reason ?

Michael

I could, but you wouldn't understand!

Hahahhahaha! Just kidding!

The numerical implementation of flow and transport equations is not something I'm familiar with. All I can do is refer you to help (isn't it a fantastic answer?)

I've tried this as I was desperate to see convergence of a mass transport problem in a sub regional scale model...

Take care

Thomas

The reason is that all the upwind techniques increase dispersivity, therefore leading to a smoother distribution of the contaminant and less steep concentration gradients. This makes the solution much easier. But be careful - some of the methods might add a lot of additional dispersivity, therefore spreading your contaminant much wider than realistic.

Ok I see, but can we ''play'' with lateral and longitudinal dispersion in order to eliminate this effect ?

I have 2 tracer tests to compare the simulated and measured tracer BTC.

Michael

Michael,

instead of using an upwinding scheme, you can also enlarge your dispersion. This should also work and maybe you have more controll over your computation.

The theory of numerical computations would tell you that your simulation should be stable if you use mesh Peclet numbers smaller or equal to 1. (The mesh Peclet number is defined: Pe = ds / alpha, where ds is the length of your element edges and alpha your logitudinal dispersivity).

To my experience you can accept Peclet numbers up to 2, but beyond that numerical problems may begin.

So there are 2 things you can do: assume a higher longitudinal dispersivity (if you do this, you should also enlarge the transversal dispersivity; ratios od longitudinal/tranversal 5:1 - 10:1 should be ok) or refine the mesh.

Nomally the most critical parts of the simulation are arround your transport boundaries where you start the simulation with high concentration gradients. So maybe a mesh refinement arround your wells could help.

Zebra