Calculation of material property (permeability) as function of variable (temperature)

[RotmanC]

Hello,
I'm working on the evolution of supercritical conditions by a magmatic intrusion at depth (7km), so the permeability (k) changes with the evolution of this high temperature body (825K).

I implement an equation to calculate k as function of temperature following the equation of Hayba & Ingebritsen (1997). I include it using the permeability as PorousFlowPermeabilityTensorFromVar that call an AuxKernel and its AuxVariable.

The problem is that it doesn't converge. May be because the permeability has a big change (k = [1e-22, 1e-14]), which is at the middle bottom of the mesh in a very short volume, that is, around the intrusion of magma (825K, k = 1e-22). So the Jacobian has a wide range of values to deal.

I tried:

• Refine the mesh
• Smooth the BC of the intrusion of 825K
• Smooth the equation that computes the permeability
• Modify the Executer, Preconditions and time steps
• Include controls to start the BC of the intrusion late, after some timesteps
• Include Controls to start the calculation of k after some timesteps after the starting of the intrusion
• Include and Adaptivity, but I cannot use k as variable (and spend a lot of time running)

May be I can calculate k as a MultiApp, or can I calculate outside and call the results later (but I don't know how to do it and I didn't found anything in the discussions).

Another question that I have is why the size of the initial execution of this simulation is more than 2Gb (2368 MB) if only runs 3570 elements in 2D.

Here is the project with the mesh and the script with no Controls and Adaptivity

Any support and comment to improve it are more than welcome

[RotmanC]

Actually I tried to use execute_on = CUSTOM too, but I don't understand how to include the list of times to compute k (to avoid to create a custom execute flag). Is it posible?

[GiudGiud]

Hello

Does it converge if you set k to be a constant?
To troubleshoot convergence issues there are a list of techniques here:
https://mooseframework.inl.gov/moose/application_usage/failed_solves.html

Guillaume

[RotmanC]

Many thanks! Now it works :)

Also was helpful to check:

1. the Jacobian using FiniteDifferencePreconditioner
2. the convergence criteria
3. the numerical stabilization

Also, I have some questions left about the use of execute_on = CUSTOM. Isn't it possible to include a list of time steps or include a range of time steps to avoid using to calculate a variable (for example, calculate it every 10 time steps) without creating a flag?

Many thanks for your help

[GiudGiud]

Hello

Execute_on = custom is not for selecting a list of time steps currently.

What you can use for that is the Controls system

Guillaume