Title: Suppressing variance in 21-cm signal simulations during reionization

Abstract: Current best limits on the 21-cm signal during reionization are provided at large scales ($\gtrsim$100 Mpc). To model these scales, enormous simulation volumes are required which are computationally expensive. We find that the primary source of uncertainty at these large scales is sample variance, which decides the minimum size of simulations required to analyse current and upcoming observations. In large-scale structure simulations, the method of `fixing' the initial conditions (ICs) to exactly follow the initial power spectrum and `pairing' two simulations with exactly out-of-phase ICs has been shown to significantly reduce sample variance. Here we apply this `fixing and pairing' (F\&P) approach to reionization simulations whose clustering signal originates from both density fluctuations and reionization bubbles. Using a semi-numerical code, we show that with the traditional method, simulation boxes of $L\simeq 500$ (300) Mpc are required to model the large-scale clustering signal at $k$=0.1 Mpc$^{-1}$ with a precision of 5 (10) per cent. Using F\&P, the simulation boxes can be reduced by a factor of 2 to obtain the same precision level. We conclude that the computing costs can be reduced by at least a factor of 4 when using the F\&P approach.