If the estimation error has zero mean and has Gaussian distribution it can be shown that Eq. (A27) reduces to [./P-] (A28)The variance of g is the mean square minus the squared mean Var[0] =h r p-h + rr[Jp-Jp-]/2 (A29) The numerator in Eq. (A21) becomes -tr[Jp-]/2)] = P~h Thus, the optimum gain vector isThe covariance of the estimation error after incorporating the measurement was given in Eq. (A9). Recall b = 0. Using Eqs.(A19, 20, 29, 30) the covariance is P + -(/-kh r )p-(/-kh r ) r + k(r + rr[Jp-Jp-]/2)k r (A32)Note if the nonlinearity J is zero, then Eq. (A31) and Eq. (A32) are equivalent to the extended (linearized) Kalman measurement incorporation equations.To summarize, the measurement incorporation equations arewhere g 0 , h, and J were defined in Eqs. (A23-25). These equations are equivalent to those proposed previously byThe rate of convergence of the finite element method is determined by the ability of the approximate strains to assume arbitrary polynomial states. A systematic way to generate shape functions assuring a certain accuracy is to start, then, with assumed polynomial strains and integrate for the displacements. In curved structures, this leads to a coupled system of shape functions for the normal and tangential displacements, which are not necessarily polynomials. Some further simplifications reduce these to coupled polynomial shape functions which might prove more efficient than uncoupled shape functions assumed independently for the different displacements. The spectral condition number of the global stiffness matrix for the curved beam increases with (r/t) 2 -the ratio of radius of curvature to thickness squared. The extensional energy portion of the elastic energy is proportional to (t/r) 2 . For a coarse mesh this might be smaller than the discretization error, not warranting the introduction of the exact small t/r into the extensional energy. By relating r/t to the number of elements so as to balance the extensional energy and the energy discretization error, r/t is eliminated from the extensional energy and consequently from the condition number. The resulting element converges with its full rate to the inextensional solution but without undue ill-conditioning in the stiffness matrix.