Closed oscardssmith closed 1 year ago
So https://github.com/SciML/OrdinaryDiffEq.jl/pull/1763 introduces regressions?
It is slightly less accurate when the timespan is greater than 1, but the policy of picking an initial dt that is fixed makes a lot less sense than picking one proportional to the timespan.
Because of the nested structure (nesting ODE solvers inside ODE solvers) tiny regressions in OrdinaryDiffEq are amplified in DelayDiffEq. So I'm also worried about small regressions. Could one introduce a branch such that larger time spans are still handled in the same way?
This was actually nu-necessary. I only need this change for DAEs.
Needed for https://github.com/SciML/OrdinaryDiffEq.jl/pull/1763. The previous tolerance was overly strict.