Should current_cohesion and current_friction_angle visu output take viscosity averaging type into account?

[anne-glerum]

The Additional Output of the visco-plastic material model gives the cohesion and friction angle, possibly strain weakened. It therefore averages the compositional contributions arithmetically:

https://github.com/geodynamics/aspect/blob/ec6fba1d7addccaa757e816faf5fb3a6facfc853/source/material_model/rheology/visco_plastic.cc#L730

When using for example a maximum composition viscosity averaging scheme, the resulting current_cohesion in paraview can be quite surprising.

For example, for the brick extension benchmark, with an inclusion with a cohesion of 1e20 Pa and background cohesion of 4e7, in the vicinity of the inclusion, cohesion will be ~1e17 Pa even though the volumetric contribution of the inclusion is only 0.005. With the maximum composition averaging of the viscosity, I would expect the cohesion to be 4e7 outside the inclusion.

So should we use the same averaging as for the viscosity for the averaging of the plastic additional output?

NB this only applies to the visualization of the cohesion and the friction angle, not to the values used in the computation of the viscosity.

[jdannberg]

I think that makes sense since the plastic viscosity directly scales with the yield stress. And it's definitely better than averaging these values arithmetically.

[anne-glerum]

Alright, I'll make a PR.

[anne-glerum]

There are certain combinations though where using the viscosity averaging would not work, for example when the friction angles are zero and the viscosity averaging is harmonic. This will trigger an Assert. Perhaps using maximum composition averaging would be better?