search

CovingtonResearchGroup / champ

Model for bedrock channel evolution including self-adjusting cross-sectional shape.
0 stars 0 forks source link

solver gets stuck with 'stable' knickpoints #25

Closed speleophysics closed 2 years ago

speleophysics commented 2 years ago

Some simulations create stable high slope zones after passing of weak rock layer.

speleophysics commented 2 years ago

This seems to be related to the secant root solver failing without providing any notification that it has not converged. Its measure of convergence passes, but the solution is far from zero. Replaced this with a bracket search using brenth. This seems to remove the stable knicks. However, the solver does still regularly fail and fallback to the minimization solver. Should create some independent check of convergence and perhaps specify atol and rtol for the solvers. They are probably more aggressive than necessary.

speleophysics commented 2 years ago

Knickpoints are not stable with brenth root solver. However, the flow solver does eventually become unstable in the example case as the giant knickpoint migrates to near the RHS of the domain. Solver fixed in 7bcb65269769ef61d80fef7c3a120d53afb16db4