num::seq::svd

Defined in header <linear/svd/svd.hpp>

svd_result svd(const mat<real> &A_in, real tol, idx max_sweeps);

One-sided Jacobi (Hestenes) SVD on the transpose.

It rotates rows of \(A^T\), which are contiguous, and carries the norms through each sweep. It is slower than bidiagonalization but accurate for small singular values.