Simulations are most likely to fail when exploring very reduced fO2

[planetmariana]

While discussing my projects with another PhD student who uses the PACMAN code, we noticed the same trend: simulations tend to fail more often under very reducing conditions (fO2<-2). In both models, runs in this low-fugacity regime are much less stable. In PROTEUS, these failures originate primarily from the interior module, where extreme redox states lead to steep gradients in melt fraction, viscosity, and convective flux that challenge the numerical stability of the solver. This convergence behavior suggests that the reduced end of the parameter space represents a physically and computationally stiff regime, reflecting the sensitivity of interior in this cases. Any thoughts about this?

[timlichtenberg]

Do you have potentially a plot or so of some simulations just before they fail? Or do they all fail immediately at the start? This would help narrow down the problem more.

[planetmariana]

They fail once they start to enter in the mixed-phase region. Here's two examples, take into account that this is varying fO2 in the melting curves.