Simeon Nedelchev
A linear solve avoids explicitly forming
Diagonal division misses the coupling.
Recursive elimination computes tree accelerations.
The algorithm can avoid assembling dense M.
Constraint solvers need repeated inertia solves.
An engine may factor M rather than use bare ABA.
Kinematics and forces precede acceleration solving.
Integration advances the state.
Derived arrays need a consistent measurement point.
Compare the same physical query and model.
Small Python timings cannot establish asymptotic complexity.