Accelerating VEP convergence with new damping approach

[luraess]

Stokes2D_simpleVEP/newdamp/Stokes2D_vep_reg_IU_LR.jl

Line 153 in ed6d619

@. η_vep = τii / 2.0 / Eii * 19.3 # nx, ny = 63, 63

Using η_vep (instead of η_ve) in the definition of dτ_ρ reduces the global number of iterations in this example by nearly a factor 2 (using the switch on l.12).

Code	# of tiers
Previous damping approach reference code	67000
New damping approach η_ve code	24000
New damping approach η_vep code	13996

However, one needs to "adjust" the magnitude of η_vep as function of number of grid points. Optimal convergence is found with

• η_vep * 19.3 for nx = ny = 63
• η_vep * 19.3 * 1.99 for nx = ny = 127

I did not yet understand all the bits there. Maybe @utkinis have an idea.