Wm. Hartmann et Re. Hartwig, COMPUTING THE MOORE-PENROSE INVERSE FOR THE COVARIANCE-MATRIX IN CONSTRAINED NONLINEAR ESTIMATION, SIAM journal on optimization, 6(3), 1996, pp. 727-747
A new algorithm is developed to compute the Moore-Penrose inverse of t
he Lagrangian matrix which is used to compute the covariance matrix of
parameter estimates in constrained nonlinear optimization. The algori
thm takes into account the bordered structure of the Lagrangian matrix
and. that the projected Hessian is available at no cost at the end of
the optimisation process. For many applications and especially for an
increasing number of active constraints at the optimum, the new algor
ithm will be considerably more efficient than the traditional one.