1: \begin{abstract}

2: Future large-scale structure surveys of the Universe will aim to

3: constrain the cosmological model and the true nature of dark energy

4: with unprecedented accuracy. In order for these surveys to achieve

5: their designed goals, they will require predictions for the nonlinear

6: matter power spectrum to sub-percent accuracy. Through the use of a

7: large ensemble of cosmological $N$-body simulations, we demonstrate

8: that if we do not understand the uncertainties associated with

9: simulating structure formation, i.e. knowledge of the `optimal'

10: simulation parameters, and simply seek to marginalize over them, then

11: the constraining power of such future surveys can be significantly

12: reduced. However, for the parameters $\{n_s,h,\Omega_b,\Omega_m\}$,

13: this effect can be largely mitigated by adding the information from a

14: CMB experiment, like Planck. In contrast, for the amplitude of

15: fluctuations $\sigma_8$ and the time-evolving equation of state of

16: dark energy $\{w_0,w_a\}$, the mitigation is mild. On marginalizing

17: over the simulation parameters, we find that the dark-energy figure of

18: merit can be degraded by $\sim2$. This is likely an optimistic

19: assessment, since we do not take into account other important

20: simulation parameters. A caveat is our assumption that the Hessian of

21: the likelihood function does not vary significantly when moving from

22: our adopted to the {\em optimal} simulation parameter set. This paper

23: therefore provides strong motivation for rigorous convergence testing

24: of $N$-body codes to meet the future challenges of precision

25: cosmology.

26: \end{abstract}

27: