SVD problem.?bug

[parisse]

Same condition number since it will diagonalize R^t R=A^t A, and cond(R^t R)=cond(A)^2 (for the Euclidean norm). This means that if say cond(R)=100, you could expect to loose 2 digits with a state of the art SVD and 4 with the quick and dirty method.
Fortunately the situation is not always that bad. I tried with hilbert(6), despite a high condition number 1.5e7, the SVD is accurate.