1: \begin{abstract}

2: Binary trust-region steepest descent (BTR) and

3: combinatorial integral approximation (CIA) 

4: are two recently investigated approaches for the solution of

5: optimization problems with distributed binary-/discrete-valued

6: variables (control functions). We show improved convergence results for

7: BTR by imposing a compactness assumption that is similar to the

8: convergence theory of CIA. As a corollary we conclude that BTR also 

9: constitutes a descent algorithm on the continuous relaxation

10: and its iterates converge weakly-$^*$ to stationary points of

11: the latter. We provide computational results that validate our

12: findings. In addition, we observe a regularizing effect of

13: BTR, which we explore by means of a hybridization of

14: CIA and BTR.\\[2ex]

15: \noindent

16: \emph{Keywords:}

17: mixed-integer optimal control, trust-region methods, relaxation-based methods

18: \\[2ex]

19: \noindent

20: \emph{MSC (2020):} 

21:  49J45, 49M05, 90C30

22: \end{abstract}

23: