1: \begin{abstract}

2:   Accurate numerical solutions of the equations of hydrodynamics play

3:   an ever more important role in many fields of astrophysics.  In this

4:   work, we reinvestigate the accuracy of the moving-mesh code

5:   \textsc{Arepo} and show how its convergence order can be improved

6:   for general problems. In particular, we clarify that for certain

7:   problems \textsc{Arepo} only reaches first-order convergence for its

8:   original formulation. This can be rectified by simple modifications

9:   we propose to the time integration scheme and the spatial gradient

10:   estimates of the code, both improving the accuracy of the code. We

11:   demonstrate that the new implementation is indeed second-order

12:   accurate under the $L^1$ norm, and in particular substantially

13:   improves conservation of angular momentum. Interestingly, whereas

14:   these improvements can significantly change the results of smooth

15:   test problems, we also find that cosmological simulations of galaxy

16:   formation are unaffected, demonstrating that the numerical errors

17:   eliminated by the new formulation do not impact these

18:   simulations. In contrast, simulations of binary stars followed over

19:   a large number of orbital times are strongly affected, as here it is

20:   particularly crucial to avoid a long-term build up of errors in

21:   angular momentum conservation.

22: \end{abstract}

23: