num::lu_invert

Defined in header <linear/factorization/lu.hpp>

template <std::floating_point T, class Index>
void lu_invert(T *inverse, const T *LU, const Index *piv, idx n, T *work) noexcept;

Inverse from packed LU factors: inverse <- A^{-1}, one column at a time.

Column j of the inverse is the solve A x = e_j, so this is n calls of lu_solve written without the per-column copy. Out of place: every column's solve reads the whole factor, so the result cannot overwrite it. work holds n elements.