Galaxy formation has been studied using idealized numerical simulations of isolated disk galaxies and galaxy mergers for decades, but most simulations performed to date have suffered from two potentially significant limitations: First, when comparing simulations with observations, physical quantities -- rather than observables -- from the simulations are used. Second, the most-commonly used techniques, smoothed-particle hydrodynamics (SPH) and adaptive mesh refinement, suffer from numerical inaccuracies that can potentially jeopardize the results of simulations performed with those techniques.
I will discuss methods for solving both of these limitations. I address the first limitation by performing 3-D dust radiative transfer on hydrodynamical simulations to calculate spatially resolved UV-mm spectral energy distributions of simulated galaxies. I will present an
application to submillimeter galaxies, for which a realistic comparison with observables yields results that are qualitatively different from those of more naive comparisons. I address the second limitation by using the more-accurate moving-mesh hydrodynamics code Arepo. I will
discuss how merger simulations performed with the moving-mesh technique differ from otherwise identical simulations performed using SPH.
Finally, based on this comparison and other work, I will outline the types of galaxy-formation simulations for which the traditional formulation of SPH is sufficiently accurate and describe when and why this is not the case.