Tune the hydrofrac example so that it at least runs

[paveltomin]

Trying to fix the reported issue #2632
Changes for the case to make it run without time step cuts:

1. Switch linear solver preconditioner to mgr instead of amg
2. Change meanPermCoefficient to 0.0
3. To make the simulation a bit faster and remove time steps cuts - limit max newton iterations to 20 and relax convergence tolerance to 1e-5 (was 1e-6 which is close to linear solver tolerance making convergence tricky)

After the changes, the case runs reasonably:

hydrofracture, number of time steps: 87
hydrofracture, number of successful nonlinear iterations: 764
hydrofracture, number of successful linear iterations: 20970
hydrofracture, number of time step cuts: 0
hydrofracture, number of discarded nonlinear iterations: 0
hydrofracture, number of discarded linear iterations: 0

Does not require rebaseline since the case is not included in the integatedTests.

[CusiniM]

I forgot, this means using the arithmetic average?

[paveltomin]

yes, this is the key change to make it run

[frankfeifan]

I have tested this example with @zhishuaizhang in the workshop. I think we can keep the harmonic average, i.e., meanPermCoefficient="1.0", and a better solution I believe is to set a nonzero penaltyStiffness (e.g., 1e12) as we have negative elementAperture here. We may also need more Newton Iteration allowance, but it can converge now with penaltyStiffness=1e12.

[paveltomin]

oh ok, interesting, thanks
i didn't try to change the penalty coefficient, can you share the output of that run?

[frankfeifan]

oh ok, interesting, thanks
i didn't try to change the penalty coefficient, can you share the output of that run?

here it is
log.txt

[rrsettgast]

should this be put back to 1.0?

[paveltomin]

yes, let me do that

[paveltomin]

@rrsettgast should be good to merge now

[zhishuaizhang]

I'm looking at the result of the modeling result after the fix and it seems to be different from the modeling result from the original documentation. Do with need to figure out why there is a difference? Or it is for another pull request?

[paveltomin]

@zhishuaizhang please post results with meanPermCoefficient=1

[CusiniM]

ah yeah, this explains the struggle.

[jhuang2601]

Great! Thanks for fixing it.