1: \begin{abstract}

2:   We address the slow convergence and poor stability of quasi-newton

3:   sequential quadratic programming (SQP) methods that is observed when

4:   solving experimental design problems, in particular when they are

5:   large.  Our findings suggest that this behavior is due to the fact

6:   that these problems often have bad absolute condition numbers.  To

7:   shed light onto the structure of the problem close to the solution,

8:   we formulate a model problem (based on the $A$-criterion), that is

9:   defined in terms of a given initial design that is to be improved.

10:   We prove that the absolute condition number of the model problem

11:   grows without bounds as the quality of the initial design improves.

12:   Additionally, we devise a preconditioner that ensures that the

13:   condition number will instead stay uniformly bounded.  Using

14:   numerical experiments, we study the effect of this reformulation on

15:   relevant cases of the general problem, and find that it leads to

16:   significant improvements in stability and convergence behavior.

17: 

18: \keywords{Sequential Quadratic Programming \and Preconditioning \and

19:   Design of Experiments}

20: \subclass{90C30 \and 90C55 \and 62K99}

21: \end{abstract}

22: