1: \begin{abstract}

2: Trust-region algorithms can be applied to very abstract optimization problems

3: because they do not require a specific direction of descent or anti-gradient.

4: This has lead to recent interest in them, in particular in the area of integer

5: optimal control problems, where the infinite-dimensional problem

6: formulations do not assume vector space structure.

7: 

8: We analyze a trust-region algorithm in the abstract setting of a compact metric

9: space, a setting in which integer optimal control problems with total variation

10: regularization can be formulated in. Our analysis allows to avoid a reset of the 

11: trust-region radius upon acceptance of the iterates when proving convergence to stationary points. This reset has been present

12: in previous analyses of trust-region algorithms for integer optimal control

13: problems. Our computational benchmark shows that the runtime can be

14: considerably improved when avoiding this reset, which is now theoretically

15: justified.

16: \end{abstract}

17: