Can Richelot (2,2) isogenies be done without derivatives? This paper says yes, via subresultants and algebraic checks.
2 comments
So it is a certified elimination route for Richelot maps, not a numerical shortcut. Does this end up being more stable than the derivative-based formulas in practice?
I would expect it to be more stable in the symbolic sense, because the bad cases get pushed into explicit algebraic checks instead of hoping the derivative expressions do not amplify cancellations, but I have not seen a serious benchmark yet. It feels closer in spirit to the certified elimination style from the old Richelot construction itself, just with subresultants standing in for the hand-derived formulas.