implement matrix-free 3d no-normal flux

[tjhei]

No description provided.

[tjhei]

Solutions are the approximately the same with and without GMG. Without:

*** Timestep 3:  t=5.46897e+07 years
   Solving temperature system... 14 iterations.
   Solving Stokes system... 200+16 iterations.

   Postprocessing:
     Writing graphical output:           output-shell_simple_3d/solution/solution-00003
     RMS, max velocity:                  0.000928 m/year, 0.00339 m/year
     Temperature min/avg/max:            973 K, 1463 K, 1973 K
     Heat fluxes through boundary parts: -3.483e+12 W, 1.373e+13 W
     Writing depth average:              output-shell_simple_3d/depth_average.vtu

with

*** Timestep 3:  t=5.46904e+07 years
   Solving temperature system... 14 iterations.
   Solving Stokes system... 43+0 iterations.

   Postprocessing:
     Writing graphical output:           output-shell_simple_3d/solution/solution-00003
     RMS, max velocity:                  0.000928 m/year, 0.00339 m/year
     Temperature min/avg/max:            973 K, 1463 K, 1973 K
     Heat fluxes through boundary parts: -3.483e+12 W, 1.373e+13 W
     Writing depth average:              output-shell_simple_3d/depth_average.vtu

[tjhei]

@tcclevenger FYI

[tcclevenger]

This seems reasonable. Iterations are as I expect.

[gassmoeller]

Very nice! Looking forward to giving this a try. We can also think about converting some of our cookbooks and benchmarks to GMG now. They are often used as templates for new models, and starting from GMG should be faster and less memory-consuming.

Edit: Just realized this requires a new deal.II. Will have to wait with cookbooks and benchmarks then.

For future reference: The corresponding deal.II functionality that is used here is in: https://github.com/dealii/dealii/blob/master/include/deal.II/numerics/vector_tools.templates.h function add_constraint currently (commit 082d75bebda3ccb9f88e440162ff803d8536ba60) line 3833.

[tjhei]

Edit: Just realized this requires a new deal.II. Will have to wait with cookbooks and benchmarks then.

Yep, sadly. I am hoping we might be able to do 9.2 release at some point soonish and then move forward with ASPECT as well.