1: \begin{abstract}

2:   \noindent

3:   Augmentation methods for mixed-integer (linear) programs are a class of

4:   primal solution approaches in which a current iterate is augmented to a

5:   better solution or proved optimal. It is well known

6:   that the performance of these methods, i.e., number of iterations needed,

7:   can theoretically be improved by scaling methods. We extend these results

8:   by an improved and extended convergence analysis, which shows that bit

9:   scaling and geometric scaling theoretically perform similarly well

10:  in the worst case for 0/1 polytopes as well as show that in some

11:  cases geometric scaling can outperform bit scaling arbitrarily. 

12: 

13:  We also investigate the performance of implementations of these

14:  methods, where the augmentation directions are computed by a MIP

15:  solver. It turns out that the number of iterations is usually

16:  low. While scaling methods usually do not improve the performance for

17:  easier problems, in the case of hard mixed-integer optimization

18:  problems they allow to compute solutions of very good quality and are

19:  often superior.

20: \end{abstract}

21: