Providing working version of BBM

[pminak]

BBM gives NaNs on LKF Benchmark. Tuning parameters, eventually modification of tilde P.

[einola]

It seems the reason BBM gives NaNs is that damage becomes equal to one, causing time_viscous to become zero and multiplicator to become NaN.

The time evolution equations don't allow damage to become equal to one - only to approach it asymptotically. I guess what's happening is that there's a slight overshoot in the advection (as is to be expected) and that this is only capped to <= 1 and not < 1.

For now, I am testing this with a .min(1.-1e-12) call when creating d_gauss. But I guess this should be controlled in the advection?

See branch issue156_BBM_crash.

[ailefroide]

Could not we imagine advecting "1-d" instead of "d" in order to prevent the overshoot? Together with @brodeau, we did that for our implementation of BBM in the Eulerian SI3 code... and it seemed to fix the issue.

[einola]

Yes, that's a good point too @ailefroide! We'll have a look.

[einola]

Depends on PR #204