In this paper, a new update criterion is proposed to improve the Nocedal-Ov
erton update criterion for reduced Hessian successive quadratic programming
(SQP). Global and R-linear convergence is proved for the new criterion and
the Nocedal-Overton criterion using nonorthogonal basis matrices, which al
low efficient implementations of the reduced Hessian SQP for solving large-
scale equality constrained problems.