silviana amethyst
silviana amethyst
* all core tests are represented * built easily on my m3 mac just now as-is.
`p_` is the order of the method. i dislike this, too.
I think there's a mathematical error in this code, from explicit_predictors.hpp 741: ``` // TODO this random vector should not be made fresh every time. especiallyif the numeric type is...
No, this is actually correct. But doing that extra LU is potentially VERY expensive. do we really need this? Is there a better way?
I think that the ODE predictor information, the numbers for the Butcher tables, etc, should be written in at compile time. Refactor to use a typetraits system?
`EvalRHS()` overwrites the columns of K. It does not add to them. Is it pointless to set K to 0 before calling this function? In FullStep, we tell `EvalRHS` the...
Why do stages later than `0` NOT re-use the LU decomposition member variable? Are we saving the one for stage 0 for something special? ``` S.SetAndReset(space, time); Mat& dhdxtempref =...
Inside all stages, `EvalRHS` overwrites the stage-th column of K. They both have this line of code: ``` K.col(stage) = LUref.solve(-dhdtref); ```
See https://eigen.tuxfamily.org/dox/group__InplaceDecomposition.html, where they describe how to do in-place LU decompositions. This could really be nice. Todo: rewrite the LU stuff in `explicit_predictors.hpp` to use In-place. Warning: if there are...
in `LUPartialPivotDecompositionSuccessful`, why doesn't the (0,0) entry come first?