1: \begin{abstract}

2:   Numerical relativity is central to the investigation of

3:   astrophysical sources in the dynamical and strong-field gravity regime,

4:   such as binary black hole and neutron star coalescences.

5:   %

6:   Current challenges set by gravitational-wave and multi-messenger

7:   astronomy call for highly performant and scalable codes on modern

8:   massively-parallel architectures.

9:   We present \GRAthena{}, a general-relativistic,

10:   high-order, vertex-centered solver that extends the oct-tree, adaptive

11:   mesh refinement capabilities of the astrophysical (radiation) magnetohydrodynamics

12:   code \Athena{}.

13:   To simulate dynamical space-times

14:   \GRAthena{} uses the \z4c{}

15:   evolution scheme of

16:   numerical relativity

17:   coupled to the moving puncture gauge.

18:   %

19:   We demonstrate stable and accurate binary black hole merger evolutions

20:   via extensive convergence testing, cross-code validation, and

21:   verification against state-of-the-art effective-one-body waveforms.

22:   %

23:   \GRAthena{} leverages the task-based parallelism paradigm

24:   of \Athena{} to achieve excellent scalability.

25:   We measure strong scaling

26:   efficiencies above $95\%$ for up to $\sim 1.2\times10^4$ CPUs and

27:   excellent weak scaling is shown up to $\sim 10^5$ CPUs in a

28:   production binary black hole setup with adaptive mesh

29:   refinement.

30:   %

31:   \GRAthena{} thus allows for the robust simulation of

32:   compact binary coalescences and offers a viable path towards

33:   numerical relativity at exascale.

34: %

35: \end{abstract}

36: