0

We give a derivation of the result for the rate of linear convergence in p. 4 of the paper. Consider an objective function f to be minimized using a search direction obtained from Bkpk = −∇f(xk) where Bk = B(xk) is a positive definite partial Hessian for k = 0, 1, 2,... Under the assumptions of theorem 3.1 in the paper, xk converges to a stationary point x∗. Assume that B(x) is Lipschitz continuous in the region of interest (that is, ∃L> 0: ‖B(x)−B(y) ‖ ≤ L ‖x − y ‖ for any two points x, y in the region) with bounded condition number, and that we take unit steps in the line search. Then: xk + pk − x ∗ = xk − x ∗ −B−1