“…Since the left-hand sides of equations (12) and (13) are known (see above) and the components of r are also known (from the direction of the prediction load path), we have two equations in the eight unknown backward and forward estimates of @W c =@ M n Y , @W c =@ M n T , @W c =@ F n Y and @W c =@ F n T . However, since the gradient vector of a scalar function at a point is normal to the level surface of the function through the point and since, in the present case, the complementary work W c is a scalar function of four independent variables, it follows that the normal vector to the W c2 level surface is parallel to the gradient vector of W c , the components of which are @W c =@ M n Y , @W c =@ M n T , @W c =@ F n Y and @W c =@ F n T .…”