[CIG-SHORT] Sign of shear stress changes from one fault node to another

[Brad Aagaard]

On 11/06/2015 02:37 AM, Romain Jolivet wrote:
> Yes, this is what I thought: These are the tolerances used for that
> run:

[pylithapp.timedependent.interfaces.fault]
zero_tolerance = 1e-13

[pylithapp.petsc]
ksp_rtol = 1.0e-50
ksp_atol = 1.0e-15
snes_rtol = 1.0e-50
snes_atol = 1.0e-10

These tolerances are tighter than what may be necessary. Looking back at 
your earlier email, the most likely explanation looks to be "there is 
one time step that diverged, although it does not corresponds to the 
time step when the first of the nodes concerned sees its friction 
dropping to negative values." Did it diverge BEFORE the funny behavior 
or AFTER? Once you divergence occurs, unless the divergence is extremely 
small, you pretty much have garbage.

We recommend that you also use:
ksp_error_if_not_converged = true
snes_error_if_not_converged = true

These will trigger an error and cause PyLith to abort with an error 
message when the solution doesn't converge.

Regards,
Brad