Residual velocities in Chunk Geometry model

[gassmoeller]

There are a number of superseded pull requests and issues that describe this problem, but since a part of it is addressed I closed the other issues and opened a new clean discussion about the remaining problem.

To summarize the problem and our findings: Tangential velocity boundaries of the Chunk geometry model show residual velocities even without driving forces that are higher than what we expect (~1cm/year for 8 cell resolution in depth). This only occurs on curved boundaries (i.e. not the longitude boundaries), and converges to 0 with increased resolution. They also:

• scale with radius (for a nondimensional radius of 1 they are 1e-16)
• rapidly decay with increased finite element degree
• do not vanish if we internally scale the radius of the chart coordinate system of the manifold (i.e. they are created by something in the real space, not by the manifold)
• scale with gravity, therefore they are likely related to the absolute pressure in the model

Related to:

822

1016

1018

1361

1533

File to reproduce result: chunk_3d.prm.txt

[bangerth]

@gassmoeller and @anne-glerum : Can you attach an input file here for easy reproducibility?

[gassmoeller]

Added the file.

• Also the residual velocities do not scale with linear solver tolerence