Thank you for your response. I understand better the point.
Nevertheless, what I am exactly looking for is the numerical method used by Nastran for reducing the stiffness. In the manual, this is not very clear, they are talking about underintegrating some terms. Nevertheless, this is not true...