Solving linear least squares problems by Gram-Schmidt orthogonalization

If inner-products are accumulated in double precision then the errors in the computedx andr are less than the errors resulting from some simultaneous initial perturbation δA, δb such that the condition of the linear least squares problem is given.