Zero densities being present in a model does not work well with interpolation on a logarithmic scale. Currently, some floating point math edge cases make it such that Magritte can crash on new MacOs systems (when running on models with some zero density points), while we do not see the same behavior on linux.
We could either:
- Interpolate between zero and non-zero densities using a fixed amount of interpolation points, which hides the issue
- Decide that zero densities are non-physical in our models, and require non-zero densities as input
I prefer the latter solution, given that it does not make sense to interpolate in log space between log(0) and any number. Though I will need to add a warning when reading in models with zero densities.
Zero densities being present in a model does not work well with interpolation on a logarithmic scale. Currently, some floating point math edge cases make it such that Magritte can crash on new MacOs systems (when running on models with some zero density points), while we do not see the same behavior on linux.
We could either:
I prefer the latter solution, given that it does not make sense to interpolate in log space between log(0) and any number. Though I will need to add a warning when reading in models with zero densities.