Roundoff error seems oddly high ~1e-13 in some cases.

[tristanmontoya]

Roundoff error seems oddly high ~1e-13 in some cases.

[tristanmontoya]

Update: After looking into this, it seems like the roundoff error is on the order of machine_epsilon*largest_field_variable or something like that (also likely depends on scaling of other values being added to or subtracted from those field variables), which is why I didn't see it on the advection tests with solutions where everything is of magnitude O(1).

We end up seeing similar things for linear advection when the initial condition is multiplied by or added to a constant, so I think the Euler flux calculation is not problematic. Entropy/conservation changes and residuals appear to stay on the order of machine zero when normalized e.g. by initial integrated values (but not necessarily when normalized by total domain volume). There may be a better way to deal with roundoff error and better handle scale issues, but for now I think I'll close this issue.