1: \begin{abstract}

2:   

3: %    By mapping the large-scale thermal and ionization stucture of the intergalactic medium (IGM),

4:   Cosmic 21cm interferometry is set to revolutionize our understanding of the Epoch of Reionization (EoR) and the Cosmic Dawn (CD).  However, the signal has structure on a huge range of scales, requiring large simulation boxes to statistically capture the relevant fields. In this work we quantify the minimum box size for simulating the power spectrum (PS) of the cosmic 21cm signal.  We perform multiple realizations of the initial conditions, for a range of box sizes.  We quantify convergence with respect to a mock observation of box length $1.1$ Gpc, with thermal noise computed for a 1000h observation with SKA1-low. We find that simulations of box lengths $ L \sim 200$--300 Mpc underestimate the large-scale power during the CD by $\sim$ 7--9 \% on average. % The absolute error on the PS from small-box realizations goes up to $\sim 7$ mK$^2$, or analogously up to $\sim$ 4 $\sigma$ of the total noise.%As expected, this bias and scatter between different realizations decreases with increasing box size. 

5: We conclude that box lengths of $L \gsim$ 250 Mpc are needed to converge at the level of $\lsim$ 1 $\sigma$ of the total noise.  %This corresponds to simulation volumes $\gsim$ 10 times larger than state-of-the-art radiative transfer simulations that resolve atomic cooling galaxies.

6: \end{abstract}

7: